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> | |
b5aadf7f | 58 | #include <linux/proc_fs.h> |
1da177e4 | 59 | #include <linux/seq_file.h> |
969c7921 | 60 | #include <linux/stop_machine.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> |
5a0e3ad6 | 74 | #include <linux/slab.h> |
1da177e4 | 75 | |
5517d86b | 76 | #include <asm/tlb.h> |
838225b4 | 77 | #include <asm/irq_regs.h> |
1da177e4 | 78 | |
6e0534f2 GH |
79 | #include "sched_cpupri.h" |
80 | ||
a8d154b0 | 81 | #define CREATE_TRACE_POINTS |
ad8d75ff | 82 | #include <trace/events/sched.h> |
a8d154b0 | 83 | |
1da177e4 LT |
84 | /* |
85 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
86 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
87 | * and back. | |
88 | */ | |
89 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
90 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
91 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
92 | ||
93 | /* | |
94 | * 'User priority' is the nice value converted to something we | |
95 | * can work with better when scaling various scheduler parameters, | |
96 | * it's a [ 0 ... 39 ] range. | |
97 | */ | |
98 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
99 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
100 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
101 | ||
102 | /* | |
d7876a08 | 103 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 104 | */ |
d6322faf | 105 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 106 | |
6aa645ea IM |
107 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
108 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
109 | ||
1da177e4 LT |
110 | /* |
111 | * These are the 'tuning knobs' of the scheduler: | |
112 | * | |
a4ec24b4 | 113 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
114 | * Timeslices get refilled after they expire. |
115 | */ | |
1da177e4 | 116 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 117 | |
d0b27fa7 PZ |
118 | /* |
119 | * single value that denotes runtime == period, ie unlimited time. | |
120 | */ | |
121 | #define RUNTIME_INF ((u64)~0ULL) | |
122 | ||
e05606d3 IM |
123 | static inline int rt_policy(int policy) |
124 | { | |
3f33a7ce | 125 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
126 | return 1; |
127 | return 0; | |
128 | } | |
129 | ||
130 | static inline int task_has_rt_policy(struct task_struct *p) | |
131 | { | |
132 | return rt_policy(p->policy); | |
133 | } | |
134 | ||
1da177e4 | 135 | /* |
6aa645ea | 136 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 137 | */ |
6aa645ea IM |
138 | struct rt_prio_array { |
139 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
140 | struct list_head queue[MAX_RT_PRIO]; | |
141 | }; | |
142 | ||
d0b27fa7 | 143 | struct rt_bandwidth { |
ea736ed5 | 144 | /* nests inside the rq lock: */ |
0986b11b | 145 | raw_spinlock_t rt_runtime_lock; |
ea736ed5 IM |
146 | ktime_t rt_period; |
147 | u64 rt_runtime; | |
148 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
149 | }; |
150 | ||
151 | static struct rt_bandwidth def_rt_bandwidth; | |
152 | ||
153 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
154 | ||
155 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
156 | { | |
157 | struct rt_bandwidth *rt_b = | |
158 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
159 | ktime_t now; | |
160 | int overrun; | |
161 | int idle = 0; | |
162 | ||
163 | for (;;) { | |
164 | now = hrtimer_cb_get_time(timer); | |
165 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
166 | ||
167 | if (!overrun) | |
168 | break; | |
169 | ||
170 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
171 | } | |
172 | ||
173 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
174 | } | |
175 | ||
176 | static | |
177 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
178 | { | |
179 | rt_b->rt_period = ns_to_ktime(period); | |
180 | rt_b->rt_runtime = runtime; | |
181 | ||
0986b11b | 182 | raw_spin_lock_init(&rt_b->rt_runtime_lock); |
ac086bc2 | 183 | |
d0b27fa7 PZ |
184 | hrtimer_init(&rt_b->rt_period_timer, |
185 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
186 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
187 | } |
188 | ||
c8bfff6d KH |
189 | static inline int rt_bandwidth_enabled(void) |
190 | { | |
191 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
192 | } |
193 | ||
194 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
195 | { | |
196 | ktime_t now; | |
197 | ||
cac64d00 | 198 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
199 | return; |
200 | ||
201 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
202 | return; | |
203 | ||
0986b11b | 204 | raw_spin_lock(&rt_b->rt_runtime_lock); |
d0b27fa7 | 205 | for (;;) { |
7f1e2ca9 PZ |
206 | unsigned long delta; |
207 | ktime_t soft, hard; | |
208 | ||
d0b27fa7 PZ |
209 | if (hrtimer_active(&rt_b->rt_period_timer)) |
210 | break; | |
211 | ||
212 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
213 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
214 | |
215 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
216 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
217 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
218 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
5c333864 | 219 | HRTIMER_MODE_ABS_PINNED, 0); |
d0b27fa7 | 220 | } |
0986b11b | 221 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
d0b27fa7 PZ |
222 | } |
223 | ||
224 | #ifdef CONFIG_RT_GROUP_SCHED | |
225 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
226 | { | |
227 | hrtimer_cancel(&rt_b->rt_period_timer); | |
228 | } | |
229 | #endif | |
230 | ||
712555ee HC |
231 | /* |
232 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
233 | * detach_destroy_domains and partition_sched_domains. | |
234 | */ | |
235 | static DEFINE_MUTEX(sched_domains_mutex); | |
236 | ||
7c941438 | 237 | #ifdef CONFIG_CGROUP_SCHED |
29f59db3 | 238 | |
68318b8e SV |
239 | #include <linux/cgroup.h> |
240 | ||
29f59db3 SV |
241 | struct cfs_rq; |
242 | ||
6f505b16 PZ |
243 | static LIST_HEAD(task_groups); |
244 | ||
29f59db3 | 245 | /* task group related information */ |
4cf86d77 | 246 | struct task_group { |
68318b8e | 247 | struct cgroup_subsys_state css; |
6c415b92 | 248 | |
052f1dc7 | 249 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
250 | /* schedulable entities of this group on each cpu */ |
251 | struct sched_entity **se; | |
252 | /* runqueue "owned" by this group on each cpu */ | |
253 | struct cfs_rq **cfs_rq; | |
254 | unsigned long shares; | |
052f1dc7 PZ |
255 | #endif |
256 | ||
257 | #ifdef CONFIG_RT_GROUP_SCHED | |
258 | struct sched_rt_entity **rt_se; | |
259 | struct rt_rq **rt_rq; | |
260 | ||
d0b27fa7 | 261 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 262 | #endif |
6b2d7700 | 263 | |
ae8393e5 | 264 | struct rcu_head rcu; |
6f505b16 | 265 | struct list_head list; |
f473aa5e PZ |
266 | |
267 | struct task_group *parent; | |
268 | struct list_head siblings; | |
269 | struct list_head children; | |
29f59db3 SV |
270 | }; |
271 | ||
eff766a6 | 272 | #define root_task_group init_task_group |
6f505b16 | 273 | |
8ed36996 | 274 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
275 | * a task group's cpu shares. |
276 | */ | |
8ed36996 | 277 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 278 | |
e9036b36 CG |
279 | #ifdef CONFIG_FAIR_GROUP_SCHED |
280 | ||
57310a98 PZ |
281 | #ifdef CONFIG_SMP |
282 | static int root_task_group_empty(void) | |
283 | { | |
284 | return list_empty(&root_task_group.children); | |
285 | } | |
286 | #endif | |
287 | ||
052f1dc7 | 288 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
052f1dc7 | 289 | |
cb4ad1ff | 290 | /* |
2e084786 LJ |
291 | * A weight of 0 or 1 can cause arithmetics problems. |
292 | * A weight of a cfs_rq is the sum of weights of which entities | |
293 | * are queued on this cfs_rq, so a weight of a entity should not be | |
294 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
295 | * (The default weight is 1024 - so there's no practical |
296 | * limitation from this.) | |
297 | */ | |
18d95a28 | 298 | #define MIN_SHARES 2 |
2e084786 | 299 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 300 | |
052f1dc7 PZ |
301 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
302 | #endif | |
303 | ||
29f59db3 | 304 | /* Default task group. |
3a252015 | 305 | * Every task in system belong to this group at bootup. |
29f59db3 | 306 | */ |
434d53b0 | 307 | struct task_group init_task_group; |
29f59db3 SV |
308 | |
309 | /* return group to which a task belongs */ | |
4cf86d77 | 310 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 311 | { |
4cf86d77 | 312 | struct task_group *tg; |
9b5b7751 | 313 | |
7c941438 | 314 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
315 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
316 | struct task_group, css); | |
24e377a8 | 317 | #else |
41a2d6cf | 318 | tg = &init_task_group; |
24e377a8 | 319 | #endif |
9b5b7751 | 320 | return tg; |
29f59db3 SV |
321 | } |
322 | ||
323 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 324 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 325 | { |
8b08ca52 PZ |
326 | /* |
327 | * Strictly speaking this rcu_read_lock() is not needed since the | |
328 | * task_group is tied to the cgroup, which in turn can never go away | |
329 | * as long as there are tasks attached to it. | |
330 | * | |
331 | * However since task_group() uses task_subsys_state() which is an | |
332 | * rcu_dereference() user, this quiets CONFIG_PROVE_RCU. | |
333 | */ | |
334 | rcu_read_lock(); | |
052f1dc7 | 335 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
336 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
337 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 338 | #endif |
6f505b16 | 339 | |
052f1dc7 | 340 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
341 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
342 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 343 | #endif |
8b08ca52 | 344 | rcu_read_unlock(); |
29f59db3 SV |
345 | } |
346 | ||
347 | #else | |
348 | ||
6f505b16 | 349 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
350 | static inline struct task_group *task_group(struct task_struct *p) |
351 | { | |
352 | return NULL; | |
353 | } | |
29f59db3 | 354 | |
7c941438 | 355 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 356 | |
6aa645ea IM |
357 | /* CFS-related fields in a runqueue */ |
358 | struct cfs_rq { | |
359 | struct load_weight load; | |
360 | unsigned long nr_running; | |
361 | ||
6aa645ea | 362 | u64 exec_clock; |
e9acbff6 | 363 | u64 min_vruntime; |
6aa645ea IM |
364 | |
365 | struct rb_root tasks_timeline; | |
366 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
367 | |
368 | struct list_head tasks; | |
369 | struct list_head *balance_iterator; | |
370 | ||
371 | /* | |
372 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
373 | * It is set to NULL otherwise (i.e when none are currently running). |
374 | */ | |
4793241b | 375 | struct sched_entity *curr, *next, *last; |
ddc97297 | 376 | |
5ac5c4d6 | 377 | unsigned int nr_spread_over; |
ddc97297 | 378 | |
62160e3f | 379 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
380 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
381 | ||
41a2d6cf IM |
382 | /* |
383 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
384 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
385 | * (like users, containers etc.) | |
386 | * | |
387 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
388 | * list is used during load balance. | |
389 | */ | |
41a2d6cf IM |
390 | struct list_head leaf_cfs_rq_list; |
391 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
392 | |
393 | #ifdef CONFIG_SMP | |
c09595f6 | 394 | /* |
c8cba857 | 395 | * the part of load.weight contributed by tasks |
c09595f6 | 396 | */ |
c8cba857 | 397 | unsigned long task_weight; |
c09595f6 | 398 | |
c8cba857 PZ |
399 | /* |
400 | * h_load = weight * f(tg) | |
401 | * | |
402 | * Where f(tg) is the recursive weight fraction assigned to | |
403 | * this group. | |
404 | */ | |
405 | unsigned long h_load; | |
c09595f6 | 406 | |
c8cba857 PZ |
407 | /* |
408 | * this cpu's part of tg->shares | |
409 | */ | |
410 | unsigned long shares; | |
f1d239f7 PZ |
411 | |
412 | /* | |
413 | * load.weight at the time we set shares | |
414 | */ | |
415 | unsigned long rq_weight; | |
c09595f6 | 416 | #endif |
6aa645ea IM |
417 | #endif |
418 | }; | |
1da177e4 | 419 | |
6aa645ea IM |
420 | /* Real-Time classes' related field in a runqueue: */ |
421 | struct rt_rq { | |
422 | struct rt_prio_array active; | |
63489e45 | 423 | unsigned long rt_nr_running; |
052f1dc7 | 424 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
425 | struct { |
426 | int curr; /* highest queued rt task prio */ | |
398a153b | 427 | #ifdef CONFIG_SMP |
e864c499 | 428 | int next; /* next highest */ |
398a153b | 429 | #endif |
e864c499 | 430 | } highest_prio; |
6f505b16 | 431 | #endif |
fa85ae24 | 432 | #ifdef CONFIG_SMP |
73fe6aae | 433 | unsigned long rt_nr_migratory; |
a1ba4d8b | 434 | unsigned long rt_nr_total; |
a22d7fc1 | 435 | int overloaded; |
917b627d | 436 | struct plist_head pushable_tasks; |
fa85ae24 | 437 | #endif |
6f505b16 | 438 | int rt_throttled; |
fa85ae24 | 439 | u64 rt_time; |
ac086bc2 | 440 | u64 rt_runtime; |
ea736ed5 | 441 | /* Nests inside the rq lock: */ |
0986b11b | 442 | raw_spinlock_t rt_runtime_lock; |
6f505b16 | 443 | |
052f1dc7 | 444 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
445 | unsigned long rt_nr_boosted; |
446 | ||
6f505b16 PZ |
447 | struct rq *rq; |
448 | struct list_head leaf_rt_rq_list; | |
449 | struct task_group *tg; | |
6f505b16 | 450 | #endif |
6aa645ea IM |
451 | }; |
452 | ||
57d885fe GH |
453 | #ifdef CONFIG_SMP |
454 | ||
455 | /* | |
456 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
457 | * variables. Each exclusive cpuset essentially defines an island domain by |
458 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
459 | * exclusive cpuset is created, we also create and attach a new root-domain |
460 | * object. | |
461 | * | |
57d885fe GH |
462 | */ |
463 | struct root_domain { | |
464 | atomic_t refcount; | |
c6c4927b RR |
465 | cpumask_var_t span; |
466 | cpumask_var_t online; | |
637f5085 | 467 | |
0eab9146 | 468 | /* |
637f5085 GH |
469 | * The "RT overload" flag: it gets set if a CPU has more than |
470 | * one runnable RT task. | |
471 | */ | |
c6c4927b | 472 | cpumask_var_t rto_mask; |
0eab9146 | 473 | atomic_t rto_count; |
6e0534f2 GH |
474 | #ifdef CONFIG_SMP |
475 | struct cpupri cpupri; | |
476 | #endif | |
57d885fe GH |
477 | }; |
478 | ||
dc938520 GH |
479 | /* |
480 | * By default the system creates a single root-domain with all cpus as | |
481 | * members (mimicking the global state we have today). | |
482 | */ | |
57d885fe GH |
483 | static struct root_domain def_root_domain; |
484 | ||
485 | #endif | |
486 | ||
1da177e4 LT |
487 | /* |
488 | * This is the main, per-CPU runqueue data structure. | |
489 | * | |
490 | * Locking rule: those places that want to lock multiple runqueues | |
491 | * (such as the load balancing or the thread migration code), lock | |
492 | * acquire operations must be ordered by ascending &runqueue. | |
493 | */ | |
70b97a7f | 494 | struct rq { |
d8016491 | 495 | /* runqueue lock: */ |
05fa785c | 496 | raw_spinlock_t lock; |
1da177e4 LT |
497 | |
498 | /* | |
499 | * nr_running and cpu_load should be in the same cacheline because | |
500 | * remote CPUs use both these fields when doing load calculation. | |
501 | */ | |
502 | unsigned long nr_running; | |
6aa645ea IM |
503 | #define CPU_LOAD_IDX_MAX 5 |
504 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
46cb4b7c | 505 | #ifdef CONFIG_NO_HZ |
39c0cbe2 | 506 | u64 nohz_stamp; |
46cb4b7c SS |
507 | unsigned char in_nohz_recently; |
508 | #endif | |
a64692a3 MG |
509 | unsigned int skip_clock_update; |
510 | ||
d8016491 IM |
511 | /* capture load from *all* tasks on this cpu: */ |
512 | struct load_weight load; | |
6aa645ea IM |
513 | unsigned long nr_load_updates; |
514 | u64 nr_switches; | |
515 | ||
516 | struct cfs_rq cfs; | |
6f505b16 | 517 | struct rt_rq rt; |
6f505b16 | 518 | |
6aa645ea | 519 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
520 | /* list of leaf cfs_rq on this cpu: */ |
521 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
522 | #endif |
523 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 524 | struct list_head leaf_rt_rq_list; |
1da177e4 | 525 | #endif |
1da177e4 LT |
526 | |
527 | /* | |
528 | * This is part of a global counter where only the total sum | |
529 | * over all CPUs matters. A task can increase this counter on | |
530 | * one CPU and if it got migrated afterwards it may decrease | |
531 | * it on another CPU. Always updated under the runqueue lock: | |
532 | */ | |
533 | unsigned long nr_uninterruptible; | |
534 | ||
36c8b586 | 535 | struct task_struct *curr, *idle; |
c9819f45 | 536 | unsigned long next_balance; |
1da177e4 | 537 | struct mm_struct *prev_mm; |
6aa645ea | 538 | |
3e51f33f | 539 | u64 clock; |
6aa645ea | 540 | |
1da177e4 LT |
541 | atomic_t nr_iowait; |
542 | ||
543 | #ifdef CONFIG_SMP | |
0eab9146 | 544 | struct root_domain *rd; |
1da177e4 LT |
545 | struct sched_domain *sd; |
546 | ||
a0a522ce | 547 | unsigned char idle_at_tick; |
1da177e4 | 548 | /* For active balancing */ |
3f029d3c | 549 | int post_schedule; |
1da177e4 LT |
550 | int active_balance; |
551 | int push_cpu; | |
969c7921 | 552 | struct cpu_stop_work active_balance_work; |
d8016491 IM |
553 | /* cpu of this runqueue: */ |
554 | int cpu; | |
1f11eb6a | 555 | int online; |
1da177e4 | 556 | |
a8a51d5e | 557 | unsigned long avg_load_per_task; |
1da177e4 | 558 | |
e9e9250b PZ |
559 | u64 rt_avg; |
560 | u64 age_stamp; | |
1b9508f6 MG |
561 | u64 idle_stamp; |
562 | u64 avg_idle; | |
1da177e4 LT |
563 | #endif |
564 | ||
dce48a84 TG |
565 | /* calc_load related fields */ |
566 | unsigned long calc_load_update; | |
567 | long calc_load_active; | |
568 | ||
8f4d37ec | 569 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
570 | #ifdef CONFIG_SMP |
571 | int hrtick_csd_pending; | |
572 | struct call_single_data hrtick_csd; | |
573 | #endif | |
8f4d37ec PZ |
574 | struct hrtimer hrtick_timer; |
575 | #endif | |
576 | ||
1da177e4 LT |
577 | #ifdef CONFIG_SCHEDSTATS |
578 | /* latency stats */ | |
579 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
580 | unsigned long long rq_cpu_time; |
581 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
582 | |
583 | /* sys_sched_yield() stats */ | |
480b9434 | 584 | unsigned int yld_count; |
1da177e4 LT |
585 | |
586 | /* schedule() stats */ | |
480b9434 KC |
587 | unsigned int sched_switch; |
588 | unsigned int sched_count; | |
589 | unsigned int sched_goidle; | |
1da177e4 LT |
590 | |
591 | /* try_to_wake_up() stats */ | |
480b9434 KC |
592 | unsigned int ttwu_count; |
593 | unsigned int ttwu_local; | |
b8efb561 IM |
594 | |
595 | /* BKL stats */ | |
480b9434 | 596 | unsigned int bkl_count; |
1da177e4 LT |
597 | #endif |
598 | }; | |
599 | ||
f34e3b61 | 600 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 601 | |
7d478721 PZ |
602 | static inline |
603 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) | |
dd41f596 | 604 | { |
7d478721 | 605 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); |
a64692a3 MG |
606 | |
607 | /* | |
608 | * A queue event has occurred, and we're going to schedule. In | |
609 | * this case, we can save a useless back to back clock update. | |
610 | */ | |
611 | if (test_tsk_need_resched(p)) | |
612 | rq->skip_clock_update = 1; | |
dd41f596 IM |
613 | } |
614 | ||
0a2966b4 CL |
615 | static inline int cpu_of(struct rq *rq) |
616 | { | |
617 | #ifdef CONFIG_SMP | |
618 | return rq->cpu; | |
619 | #else | |
620 | return 0; | |
621 | #endif | |
622 | } | |
623 | ||
497f0ab3 | 624 | #define rcu_dereference_check_sched_domain(p) \ |
d11c563d PM |
625 | rcu_dereference_check((p), \ |
626 | rcu_read_lock_sched_held() || \ | |
627 | lockdep_is_held(&sched_domains_mutex)) | |
628 | ||
674311d5 NP |
629 | /* |
630 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 631 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
632 | * |
633 | * The domain tree of any CPU may only be accessed from within | |
634 | * preempt-disabled sections. | |
635 | */ | |
48f24c4d | 636 | #define for_each_domain(cpu, __sd) \ |
497f0ab3 | 637 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) |
1da177e4 LT |
638 | |
639 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
640 | #define this_rq() (&__get_cpu_var(runqueues)) | |
641 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
642 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 643 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 644 | |
aa9c4c0f | 645 | inline void update_rq_clock(struct rq *rq) |
3e51f33f | 646 | { |
a64692a3 MG |
647 | if (!rq->skip_clock_update) |
648 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
3e51f33f PZ |
649 | } |
650 | ||
bf5c91ba IM |
651 | /* |
652 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
653 | */ | |
654 | #ifdef CONFIG_SCHED_DEBUG | |
655 | # define const_debug __read_mostly | |
656 | #else | |
657 | # define const_debug static const | |
658 | #endif | |
659 | ||
017730c1 IM |
660 | /** |
661 | * runqueue_is_locked | |
e17b38bf | 662 | * @cpu: the processor in question. |
017730c1 IM |
663 | * |
664 | * Returns true if the current cpu runqueue is locked. | |
665 | * This interface allows printk to be called with the runqueue lock | |
666 | * held and know whether or not it is OK to wake up the klogd. | |
667 | */ | |
89f19f04 | 668 | int runqueue_is_locked(int cpu) |
017730c1 | 669 | { |
05fa785c | 670 | return raw_spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
671 | } |
672 | ||
bf5c91ba IM |
673 | /* |
674 | * Debugging: various feature bits | |
675 | */ | |
f00b45c1 PZ |
676 | |
677 | #define SCHED_FEAT(name, enabled) \ | |
678 | __SCHED_FEAT_##name , | |
679 | ||
bf5c91ba | 680 | enum { |
f00b45c1 | 681 | #include "sched_features.h" |
bf5c91ba IM |
682 | }; |
683 | ||
f00b45c1 PZ |
684 | #undef SCHED_FEAT |
685 | ||
686 | #define SCHED_FEAT(name, enabled) \ | |
687 | (1UL << __SCHED_FEAT_##name) * enabled | | |
688 | ||
bf5c91ba | 689 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
690 | #include "sched_features.h" |
691 | 0; | |
692 | ||
693 | #undef SCHED_FEAT | |
694 | ||
695 | #ifdef CONFIG_SCHED_DEBUG | |
696 | #define SCHED_FEAT(name, enabled) \ | |
697 | #name , | |
698 | ||
983ed7a6 | 699 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
700 | #include "sched_features.h" |
701 | NULL | |
702 | }; | |
703 | ||
704 | #undef SCHED_FEAT | |
705 | ||
34f3a814 | 706 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 707 | { |
f00b45c1 PZ |
708 | int i; |
709 | ||
710 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
711 | if (!(sysctl_sched_features & (1UL << i))) |
712 | seq_puts(m, "NO_"); | |
713 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 714 | } |
34f3a814 | 715 | seq_puts(m, "\n"); |
f00b45c1 | 716 | |
34f3a814 | 717 | return 0; |
f00b45c1 PZ |
718 | } |
719 | ||
720 | static ssize_t | |
721 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
722 | size_t cnt, loff_t *ppos) | |
723 | { | |
724 | char buf[64]; | |
725 | char *cmp = buf; | |
726 | int neg = 0; | |
727 | int i; | |
728 | ||
729 | if (cnt > 63) | |
730 | cnt = 63; | |
731 | ||
732 | if (copy_from_user(&buf, ubuf, cnt)) | |
733 | return -EFAULT; | |
734 | ||
735 | buf[cnt] = 0; | |
736 | ||
c24b7c52 | 737 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
738 | neg = 1; |
739 | cmp += 3; | |
740 | } | |
741 | ||
742 | for (i = 0; sched_feat_names[i]; i++) { | |
743 | int len = strlen(sched_feat_names[i]); | |
744 | ||
745 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
746 | if (neg) | |
747 | sysctl_sched_features &= ~(1UL << i); | |
748 | else | |
749 | sysctl_sched_features |= (1UL << i); | |
750 | break; | |
751 | } | |
752 | } | |
753 | ||
754 | if (!sched_feat_names[i]) | |
755 | return -EINVAL; | |
756 | ||
42994724 | 757 | *ppos += cnt; |
f00b45c1 PZ |
758 | |
759 | return cnt; | |
760 | } | |
761 | ||
34f3a814 LZ |
762 | static int sched_feat_open(struct inode *inode, struct file *filp) |
763 | { | |
764 | return single_open(filp, sched_feat_show, NULL); | |
765 | } | |
766 | ||
828c0950 | 767 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
768 | .open = sched_feat_open, |
769 | .write = sched_feat_write, | |
770 | .read = seq_read, | |
771 | .llseek = seq_lseek, | |
772 | .release = single_release, | |
f00b45c1 PZ |
773 | }; |
774 | ||
775 | static __init int sched_init_debug(void) | |
776 | { | |
f00b45c1 PZ |
777 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
778 | &sched_feat_fops); | |
779 | ||
780 | return 0; | |
781 | } | |
782 | late_initcall(sched_init_debug); | |
783 | ||
784 | #endif | |
785 | ||
786 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 787 | |
b82d9fdd PZ |
788 | /* |
789 | * Number of tasks to iterate in a single balance run. | |
790 | * Limited because this is done with IRQs disabled. | |
791 | */ | |
792 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
793 | ||
2398f2c6 PZ |
794 | /* |
795 | * ratelimit for updating the group shares. | |
55cd5340 | 796 | * default: 0.25ms |
2398f2c6 | 797 | */ |
55cd5340 | 798 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
0bcdcf28 | 799 | unsigned int normalized_sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 800 | |
ffda12a1 PZ |
801 | /* |
802 | * Inject some fuzzyness into changing the per-cpu group shares | |
803 | * this avoids remote rq-locks at the expense of fairness. | |
804 | * default: 4 | |
805 | */ | |
806 | unsigned int sysctl_sched_shares_thresh = 4; | |
807 | ||
e9e9250b PZ |
808 | /* |
809 | * period over which we average the RT time consumption, measured | |
810 | * in ms. | |
811 | * | |
812 | * default: 1s | |
813 | */ | |
814 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
815 | ||
fa85ae24 | 816 | /* |
9f0c1e56 | 817 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
818 | * default: 1s |
819 | */ | |
9f0c1e56 | 820 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 821 | |
6892b75e IM |
822 | static __read_mostly int scheduler_running; |
823 | ||
9f0c1e56 PZ |
824 | /* |
825 | * part of the period that we allow rt tasks to run in us. | |
826 | * default: 0.95s | |
827 | */ | |
828 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 829 | |
d0b27fa7 PZ |
830 | static inline u64 global_rt_period(void) |
831 | { | |
832 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
833 | } | |
834 | ||
835 | static inline u64 global_rt_runtime(void) | |
836 | { | |
e26873bb | 837 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
838 | return RUNTIME_INF; |
839 | ||
840 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
841 | } | |
fa85ae24 | 842 | |
1da177e4 | 843 | #ifndef prepare_arch_switch |
4866cde0 NP |
844 | # define prepare_arch_switch(next) do { } while (0) |
845 | #endif | |
846 | #ifndef finish_arch_switch | |
847 | # define finish_arch_switch(prev) do { } while (0) | |
848 | #endif | |
849 | ||
051a1d1a DA |
850 | static inline int task_current(struct rq *rq, struct task_struct *p) |
851 | { | |
852 | return rq->curr == p; | |
853 | } | |
854 | ||
4866cde0 | 855 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 856 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 857 | { |
051a1d1a | 858 | return task_current(rq, p); |
4866cde0 NP |
859 | } |
860 | ||
70b97a7f | 861 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
862 | { |
863 | } | |
864 | ||
70b97a7f | 865 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 866 | { |
da04c035 IM |
867 | #ifdef CONFIG_DEBUG_SPINLOCK |
868 | /* this is a valid case when another task releases the spinlock */ | |
869 | rq->lock.owner = current; | |
870 | #endif | |
8a25d5de IM |
871 | /* |
872 | * If we are tracking spinlock dependencies then we have to | |
873 | * fix up the runqueue lock - which gets 'carried over' from | |
874 | * prev into current: | |
875 | */ | |
876 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
877 | ||
05fa785c | 878 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 NP |
879 | } |
880 | ||
881 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 882 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
883 | { |
884 | #ifdef CONFIG_SMP | |
885 | return p->oncpu; | |
886 | #else | |
051a1d1a | 887 | return task_current(rq, p); |
4866cde0 NP |
888 | #endif |
889 | } | |
890 | ||
70b97a7f | 891 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
892 | { |
893 | #ifdef CONFIG_SMP | |
894 | /* | |
895 | * We can optimise this out completely for !SMP, because the | |
896 | * SMP rebalancing from interrupt is the only thing that cares | |
897 | * here. | |
898 | */ | |
899 | next->oncpu = 1; | |
900 | #endif | |
901 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
05fa785c | 902 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 | 903 | #else |
05fa785c | 904 | raw_spin_unlock(&rq->lock); |
4866cde0 NP |
905 | #endif |
906 | } | |
907 | ||
70b97a7f | 908 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
909 | { |
910 | #ifdef CONFIG_SMP | |
911 | /* | |
912 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
913 | * We must ensure this doesn't happen until the switch is completely | |
914 | * finished. | |
915 | */ | |
916 | smp_wmb(); | |
917 | prev->oncpu = 0; | |
918 | #endif | |
919 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
920 | local_irq_enable(); | |
1da177e4 | 921 | #endif |
4866cde0 NP |
922 | } |
923 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 924 | |
0970d299 | 925 | /* |
65cc8e48 PZ |
926 | * Check whether the task is waking, we use this to synchronize ->cpus_allowed |
927 | * against ttwu(). | |
0970d299 PZ |
928 | */ |
929 | static inline int task_is_waking(struct task_struct *p) | |
930 | { | |
0017d735 | 931 | return unlikely(p->state == TASK_WAKING); |
0970d299 PZ |
932 | } |
933 | ||
b29739f9 IM |
934 | /* |
935 | * __task_rq_lock - lock the runqueue a given task resides on. | |
936 | * Must be called interrupts disabled. | |
937 | */ | |
70b97a7f | 938 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
939 | __acquires(rq->lock) |
940 | { | |
0970d299 PZ |
941 | struct rq *rq; |
942 | ||
3a5c359a | 943 | for (;;) { |
0970d299 | 944 | rq = task_rq(p); |
05fa785c | 945 | raw_spin_lock(&rq->lock); |
65cc8e48 | 946 | if (likely(rq == task_rq(p))) |
3a5c359a | 947 | return rq; |
05fa785c | 948 | raw_spin_unlock(&rq->lock); |
b29739f9 | 949 | } |
b29739f9 IM |
950 | } |
951 | ||
1da177e4 LT |
952 | /* |
953 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 954 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
955 | * explicitly disabling preemption. |
956 | */ | |
70b97a7f | 957 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
958 | __acquires(rq->lock) |
959 | { | |
70b97a7f | 960 | struct rq *rq; |
1da177e4 | 961 | |
3a5c359a AK |
962 | for (;;) { |
963 | local_irq_save(*flags); | |
964 | rq = task_rq(p); | |
05fa785c | 965 | raw_spin_lock(&rq->lock); |
65cc8e48 | 966 | if (likely(rq == task_rq(p))) |
3a5c359a | 967 | return rq; |
05fa785c | 968 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 969 | } |
1da177e4 LT |
970 | } |
971 | ||
a9957449 | 972 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
973 | __releases(rq->lock) |
974 | { | |
05fa785c | 975 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
976 | } |
977 | ||
70b97a7f | 978 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
979 | __releases(rq->lock) |
980 | { | |
05fa785c | 981 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 LT |
982 | } |
983 | ||
1da177e4 | 984 | /* |
cc2a73b5 | 985 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 986 | */ |
a9957449 | 987 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
988 | __acquires(rq->lock) |
989 | { | |
70b97a7f | 990 | struct rq *rq; |
1da177e4 LT |
991 | |
992 | local_irq_disable(); | |
993 | rq = this_rq(); | |
05fa785c | 994 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
995 | |
996 | return rq; | |
997 | } | |
998 | ||
8f4d37ec PZ |
999 | #ifdef CONFIG_SCHED_HRTICK |
1000 | /* | |
1001 | * Use HR-timers to deliver accurate preemption points. | |
1002 | * | |
1003 | * Its all a bit involved since we cannot program an hrt while holding the | |
1004 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1005 | * reschedule event. | |
1006 | * | |
1007 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1008 | * rq->lock. | |
1009 | */ | |
8f4d37ec PZ |
1010 | |
1011 | /* | |
1012 | * Use hrtick when: | |
1013 | * - enabled by features | |
1014 | * - hrtimer is actually high res | |
1015 | */ | |
1016 | static inline int hrtick_enabled(struct rq *rq) | |
1017 | { | |
1018 | if (!sched_feat(HRTICK)) | |
1019 | return 0; | |
ba42059f | 1020 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1021 | return 0; |
8f4d37ec PZ |
1022 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1023 | } | |
1024 | ||
8f4d37ec PZ |
1025 | static void hrtick_clear(struct rq *rq) |
1026 | { | |
1027 | if (hrtimer_active(&rq->hrtick_timer)) | |
1028 | hrtimer_cancel(&rq->hrtick_timer); | |
1029 | } | |
1030 | ||
8f4d37ec PZ |
1031 | /* |
1032 | * High-resolution timer tick. | |
1033 | * Runs from hardirq context with interrupts disabled. | |
1034 | */ | |
1035 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1036 | { | |
1037 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1038 | ||
1039 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1040 | ||
05fa785c | 1041 | raw_spin_lock(&rq->lock); |
3e51f33f | 1042 | update_rq_clock(rq); |
8f4d37ec | 1043 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 1044 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
1045 | |
1046 | return HRTIMER_NORESTART; | |
1047 | } | |
1048 | ||
95e904c7 | 1049 | #ifdef CONFIG_SMP |
31656519 PZ |
1050 | /* |
1051 | * called from hardirq (IPI) context | |
1052 | */ | |
1053 | static void __hrtick_start(void *arg) | |
b328ca18 | 1054 | { |
31656519 | 1055 | struct rq *rq = arg; |
b328ca18 | 1056 | |
05fa785c | 1057 | raw_spin_lock(&rq->lock); |
31656519 PZ |
1058 | hrtimer_restart(&rq->hrtick_timer); |
1059 | rq->hrtick_csd_pending = 0; | |
05fa785c | 1060 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
1061 | } |
1062 | ||
31656519 PZ |
1063 | /* |
1064 | * Called to set the hrtick timer state. | |
1065 | * | |
1066 | * called with rq->lock held and irqs disabled | |
1067 | */ | |
1068 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1069 | { |
31656519 PZ |
1070 | struct hrtimer *timer = &rq->hrtick_timer; |
1071 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1072 | |
cc584b21 | 1073 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1074 | |
1075 | if (rq == this_rq()) { | |
1076 | hrtimer_restart(timer); | |
1077 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1078 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1079 | rq->hrtick_csd_pending = 1; |
1080 | } | |
b328ca18 PZ |
1081 | } |
1082 | ||
1083 | static int | |
1084 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1085 | { | |
1086 | int cpu = (int)(long)hcpu; | |
1087 | ||
1088 | switch (action) { | |
1089 | case CPU_UP_CANCELED: | |
1090 | case CPU_UP_CANCELED_FROZEN: | |
1091 | case CPU_DOWN_PREPARE: | |
1092 | case CPU_DOWN_PREPARE_FROZEN: | |
1093 | case CPU_DEAD: | |
1094 | case CPU_DEAD_FROZEN: | |
31656519 | 1095 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1096 | return NOTIFY_OK; |
1097 | } | |
1098 | ||
1099 | return NOTIFY_DONE; | |
1100 | } | |
1101 | ||
fa748203 | 1102 | static __init void init_hrtick(void) |
b328ca18 PZ |
1103 | { |
1104 | hotcpu_notifier(hotplug_hrtick, 0); | |
1105 | } | |
31656519 PZ |
1106 | #else |
1107 | /* | |
1108 | * Called to set the hrtick timer state. | |
1109 | * | |
1110 | * called with rq->lock held and irqs disabled | |
1111 | */ | |
1112 | static void hrtick_start(struct rq *rq, u64 delay) | |
1113 | { | |
7f1e2ca9 | 1114 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1115 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1116 | } |
b328ca18 | 1117 | |
006c75f1 | 1118 | static inline void init_hrtick(void) |
8f4d37ec | 1119 | { |
8f4d37ec | 1120 | } |
31656519 | 1121 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1122 | |
31656519 | 1123 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1124 | { |
31656519 PZ |
1125 | #ifdef CONFIG_SMP |
1126 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1127 | |
31656519 PZ |
1128 | rq->hrtick_csd.flags = 0; |
1129 | rq->hrtick_csd.func = __hrtick_start; | |
1130 | rq->hrtick_csd.info = rq; | |
1131 | #endif | |
8f4d37ec | 1132 | |
31656519 PZ |
1133 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1134 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1135 | } |
006c75f1 | 1136 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1137 | static inline void hrtick_clear(struct rq *rq) |
1138 | { | |
1139 | } | |
1140 | ||
8f4d37ec PZ |
1141 | static inline void init_rq_hrtick(struct rq *rq) |
1142 | { | |
1143 | } | |
1144 | ||
b328ca18 PZ |
1145 | static inline void init_hrtick(void) |
1146 | { | |
1147 | } | |
006c75f1 | 1148 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1149 | |
c24d20db IM |
1150 | /* |
1151 | * resched_task - mark a task 'to be rescheduled now'. | |
1152 | * | |
1153 | * On UP this means the setting of the need_resched flag, on SMP it | |
1154 | * might also involve a cross-CPU call to trigger the scheduler on | |
1155 | * the target CPU. | |
1156 | */ | |
1157 | #ifdef CONFIG_SMP | |
1158 | ||
1159 | #ifndef tsk_is_polling | |
1160 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1161 | #endif | |
1162 | ||
31656519 | 1163 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1164 | { |
1165 | int cpu; | |
1166 | ||
05fa785c | 1167 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 1168 | |
5ed0cec0 | 1169 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1170 | return; |
1171 | ||
5ed0cec0 | 1172 | set_tsk_need_resched(p); |
c24d20db IM |
1173 | |
1174 | cpu = task_cpu(p); | |
1175 | if (cpu == smp_processor_id()) | |
1176 | return; | |
1177 | ||
1178 | /* NEED_RESCHED must be visible before we test polling */ | |
1179 | smp_mb(); | |
1180 | if (!tsk_is_polling(p)) | |
1181 | smp_send_reschedule(cpu); | |
1182 | } | |
1183 | ||
1184 | static void resched_cpu(int cpu) | |
1185 | { | |
1186 | struct rq *rq = cpu_rq(cpu); | |
1187 | unsigned long flags; | |
1188 | ||
05fa785c | 1189 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
1190 | return; |
1191 | resched_task(cpu_curr(cpu)); | |
05fa785c | 1192 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 1193 | } |
06d8308c TG |
1194 | |
1195 | #ifdef CONFIG_NO_HZ | |
1196 | /* | |
1197 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1198 | * idle CPU then this timer might expire before the next timer event | |
1199 | * which is scheduled to wake up that CPU. In case of a completely | |
1200 | * idle system the next event might even be infinite time into the | |
1201 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1202 | * leaves the inner idle loop so the newly added timer is taken into | |
1203 | * account when the CPU goes back to idle and evaluates the timer | |
1204 | * wheel for the next timer event. | |
1205 | */ | |
1206 | void wake_up_idle_cpu(int cpu) | |
1207 | { | |
1208 | struct rq *rq = cpu_rq(cpu); | |
1209 | ||
1210 | if (cpu == smp_processor_id()) | |
1211 | return; | |
1212 | ||
1213 | /* | |
1214 | * This is safe, as this function is called with the timer | |
1215 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1216 | * to idle and has not yet set rq->curr to idle then it will | |
1217 | * be serialized on the timer wheel base lock and take the new | |
1218 | * timer into account automatically. | |
1219 | */ | |
1220 | if (rq->curr != rq->idle) | |
1221 | return; | |
1222 | ||
1223 | /* | |
1224 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1225 | * lockless. The worst case is that the other CPU runs the | |
1226 | * idle task through an additional NOOP schedule() | |
1227 | */ | |
5ed0cec0 | 1228 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1229 | |
1230 | /* NEED_RESCHED must be visible before we test polling */ | |
1231 | smp_mb(); | |
1232 | if (!tsk_is_polling(rq->idle)) | |
1233 | smp_send_reschedule(cpu); | |
1234 | } | |
39c0cbe2 MG |
1235 | |
1236 | int nohz_ratelimit(int cpu) | |
1237 | { | |
1238 | struct rq *rq = cpu_rq(cpu); | |
1239 | u64 diff = rq->clock - rq->nohz_stamp; | |
1240 | ||
1241 | rq->nohz_stamp = rq->clock; | |
1242 | ||
1243 | return diff < (NSEC_PER_SEC / HZ) >> 1; | |
1244 | } | |
1245 | ||
6d6bc0ad | 1246 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1247 | |
e9e9250b PZ |
1248 | static u64 sched_avg_period(void) |
1249 | { | |
1250 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1251 | } | |
1252 | ||
1253 | static void sched_avg_update(struct rq *rq) | |
1254 | { | |
1255 | s64 period = sched_avg_period(); | |
1256 | ||
1257 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
1258 | rq->age_stamp += period; | |
1259 | rq->rt_avg /= 2; | |
1260 | } | |
1261 | } | |
1262 | ||
1263 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1264 | { | |
1265 | rq->rt_avg += rt_delta; | |
1266 | sched_avg_update(rq); | |
1267 | } | |
1268 | ||
6d6bc0ad | 1269 | #else /* !CONFIG_SMP */ |
31656519 | 1270 | static void resched_task(struct task_struct *p) |
c24d20db | 1271 | { |
05fa785c | 1272 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 1273 | set_tsk_need_resched(p); |
c24d20db | 1274 | } |
e9e9250b PZ |
1275 | |
1276 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1277 | { | |
1278 | } | |
6d6bc0ad | 1279 | #endif /* CONFIG_SMP */ |
c24d20db | 1280 | |
45bf76df IM |
1281 | #if BITS_PER_LONG == 32 |
1282 | # define WMULT_CONST (~0UL) | |
1283 | #else | |
1284 | # define WMULT_CONST (1UL << 32) | |
1285 | #endif | |
1286 | ||
1287 | #define WMULT_SHIFT 32 | |
1288 | ||
194081eb IM |
1289 | /* |
1290 | * Shift right and round: | |
1291 | */ | |
cf2ab469 | 1292 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1293 | |
a7be37ac PZ |
1294 | /* |
1295 | * delta *= weight / lw | |
1296 | */ | |
cb1c4fc9 | 1297 | static unsigned long |
45bf76df IM |
1298 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1299 | struct load_weight *lw) | |
1300 | { | |
1301 | u64 tmp; | |
1302 | ||
7a232e03 LJ |
1303 | if (!lw->inv_weight) { |
1304 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1305 | lw->inv_weight = 1; | |
1306 | else | |
1307 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1308 | / (lw->weight+1); | |
1309 | } | |
45bf76df IM |
1310 | |
1311 | tmp = (u64)delta_exec * weight; | |
1312 | /* | |
1313 | * Check whether we'd overflow the 64-bit multiplication: | |
1314 | */ | |
194081eb | 1315 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1316 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1317 | WMULT_SHIFT/2); |
1318 | else | |
cf2ab469 | 1319 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1320 | |
ecf691da | 1321 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1322 | } |
1323 | ||
1091985b | 1324 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1325 | { |
1326 | lw->weight += inc; | |
e89996ae | 1327 | lw->inv_weight = 0; |
45bf76df IM |
1328 | } |
1329 | ||
1091985b | 1330 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1331 | { |
1332 | lw->weight -= dec; | |
e89996ae | 1333 | lw->inv_weight = 0; |
45bf76df IM |
1334 | } |
1335 | ||
2dd73a4f PW |
1336 | /* |
1337 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1338 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1339 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1340 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1341 | * scaled version of the new time slice allocation that they receive on time |
1342 | * slice expiry etc. | |
1343 | */ | |
1344 | ||
cce7ade8 PZ |
1345 | #define WEIGHT_IDLEPRIO 3 |
1346 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1347 | |
1348 | /* | |
1349 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1350 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1351 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1352 | * that remained on nice 0. | |
1353 | * | |
1354 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1355 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1356 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1357 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1358 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1359 | */ |
1360 | static const int prio_to_weight[40] = { | |
254753dc IM |
1361 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1362 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1363 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1364 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1365 | /* 0 */ 1024, 820, 655, 526, 423, | |
1366 | /* 5 */ 335, 272, 215, 172, 137, | |
1367 | /* 10 */ 110, 87, 70, 56, 45, | |
1368 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1369 | }; |
1370 | ||
5714d2de IM |
1371 | /* |
1372 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1373 | * | |
1374 | * In cases where the weight does not change often, we can use the | |
1375 | * precalculated inverse to speed up arithmetics by turning divisions | |
1376 | * into multiplications: | |
1377 | */ | |
dd41f596 | 1378 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1379 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1380 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1381 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1382 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1383 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1384 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1385 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1386 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1387 | }; |
2dd73a4f | 1388 | |
ef12fefa BR |
1389 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1390 | enum cpuacct_stat_index { | |
1391 | CPUACCT_STAT_USER, /* ... user mode */ | |
1392 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1393 | ||
1394 | CPUACCT_STAT_NSTATS, | |
1395 | }; | |
1396 | ||
d842de87 SV |
1397 | #ifdef CONFIG_CGROUP_CPUACCT |
1398 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1399 | static void cpuacct_update_stats(struct task_struct *tsk, |
1400 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1401 | #else |
1402 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1403 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1404 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1405 | #endif |
1406 | ||
18d95a28 PZ |
1407 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1408 | { | |
1409 | update_load_add(&rq->load, load); | |
1410 | } | |
1411 | ||
1412 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1413 | { | |
1414 | update_load_sub(&rq->load, load); | |
1415 | } | |
1416 | ||
7940ca36 | 1417 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1418 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1419 | |
1420 | /* | |
1421 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1422 | * leaving it for the final time. | |
1423 | */ | |
eb755805 | 1424 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1425 | { |
1426 | struct task_group *parent, *child; | |
eb755805 | 1427 | int ret; |
c09595f6 PZ |
1428 | |
1429 | rcu_read_lock(); | |
1430 | parent = &root_task_group; | |
1431 | down: | |
eb755805 PZ |
1432 | ret = (*down)(parent, data); |
1433 | if (ret) | |
1434 | goto out_unlock; | |
c09595f6 PZ |
1435 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1436 | parent = child; | |
1437 | goto down; | |
1438 | ||
1439 | up: | |
1440 | continue; | |
1441 | } | |
eb755805 PZ |
1442 | ret = (*up)(parent, data); |
1443 | if (ret) | |
1444 | goto out_unlock; | |
c09595f6 PZ |
1445 | |
1446 | child = parent; | |
1447 | parent = parent->parent; | |
1448 | if (parent) | |
1449 | goto up; | |
eb755805 | 1450 | out_unlock: |
c09595f6 | 1451 | rcu_read_unlock(); |
eb755805 PZ |
1452 | |
1453 | return ret; | |
c09595f6 PZ |
1454 | } |
1455 | ||
eb755805 PZ |
1456 | static int tg_nop(struct task_group *tg, void *data) |
1457 | { | |
1458 | return 0; | |
c09595f6 | 1459 | } |
eb755805 PZ |
1460 | #endif |
1461 | ||
1462 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1463 | /* Used instead of source_load when we know the type == 0 */ |
1464 | static unsigned long weighted_cpuload(const int cpu) | |
1465 | { | |
1466 | return cpu_rq(cpu)->load.weight; | |
1467 | } | |
1468 | ||
1469 | /* | |
1470 | * Return a low guess at the load of a migration-source cpu weighted | |
1471 | * according to the scheduling class and "nice" value. | |
1472 | * | |
1473 | * We want to under-estimate the load of migration sources, to | |
1474 | * balance conservatively. | |
1475 | */ | |
1476 | static unsigned long source_load(int cpu, int type) | |
1477 | { | |
1478 | struct rq *rq = cpu_rq(cpu); | |
1479 | unsigned long total = weighted_cpuload(cpu); | |
1480 | ||
1481 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1482 | return total; | |
1483 | ||
1484 | return min(rq->cpu_load[type-1], total); | |
1485 | } | |
1486 | ||
1487 | /* | |
1488 | * Return a high guess at the load of a migration-target cpu weighted | |
1489 | * according to the scheduling class and "nice" value. | |
1490 | */ | |
1491 | static unsigned long target_load(int cpu, int type) | |
1492 | { | |
1493 | struct rq *rq = cpu_rq(cpu); | |
1494 | unsigned long total = weighted_cpuload(cpu); | |
1495 | ||
1496 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1497 | return total; | |
1498 | ||
1499 | return max(rq->cpu_load[type-1], total); | |
1500 | } | |
1501 | ||
ae154be1 PZ |
1502 | static struct sched_group *group_of(int cpu) |
1503 | { | |
d11c563d | 1504 | struct sched_domain *sd = rcu_dereference_sched(cpu_rq(cpu)->sd); |
ae154be1 PZ |
1505 | |
1506 | if (!sd) | |
1507 | return NULL; | |
1508 | ||
1509 | return sd->groups; | |
1510 | } | |
1511 | ||
1512 | static unsigned long power_of(int cpu) | |
1513 | { | |
1514 | struct sched_group *group = group_of(cpu); | |
1515 | ||
1516 | if (!group) | |
1517 | return SCHED_LOAD_SCALE; | |
1518 | ||
1519 | return group->cpu_power; | |
1520 | } | |
1521 | ||
eb755805 PZ |
1522 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1523 | ||
1524 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1525 | { | |
1526 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1527 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1528 | |
4cd42620 SR |
1529 | if (nr_running) |
1530 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1531 | else |
1532 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1533 | |
1534 | return rq->avg_load_per_task; | |
1535 | } | |
1536 | ||
1537 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1538 | |
43cf38eb | 1539 | static __read_mostly unsigned long __percpu *update_shares_data; |
34d76c41 | 1540 | |
c09595f6 PZ |
1541 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1542 | ||
1543 | /* | |
1544 | * Calculate and set the cpu's group shares. | |
1545 | */ | |
34d76c41 PZ |
1546 | static void update_group_shares_cpu(struct task_group *tg, int cpu, |
1547 | unsigned long sd_shares, | |
1548 | unsigned long sd_rq_weight, | |
4a6cc4bd | 1549 | unsigned long *usd_rq_weight) |
18d95a28 | 1550 | { |
34d76c41 | 1551 | unsigned long shares, rq_weight; |
a5004278 | 1552 | int boost = 0; |
c09595f6 | 1553 | |
4a6cc4bd | 1554 | rq_weight = usd_rq_weight[cpu]; |
a5004278 PZ |
1555 | if (!rq_weight) { |
1556 | boost = 1; | |
1557 | rq_weight = NICE_0_LOAD; | |
1558 | } | |
c8cba857 | 1559 | |
c09595f6 | 1560 | /* |
a8af7246 PZ |
1561 | * \Sum_j shares_j * rq_weight_i |
1562 | * shares_i = ----------------------------- | |
1563 | * \Sum_j rq_weight_j | |
c09595f6 | 1564 | */ |
ec4e0e2f | 1565 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1566 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1567 | |
ffda12a1 PZ |
1568 | if (abs(shares - tg->se[cpu]->load.weight) > |
1569 | sysctl_sched_shares_thresh) { | |
1570 | struct rq *rq = cpu_rq(cpu); | |
1571 | unsigned long flags; | |
c09595f6 | 1572 | |
05fa785c | 1573 | raw_spin_lock_irqsave(&rq->lock, flags); |
34d76c41 | 1574 | tg->cfs_rq[cpu]->rq_weight = boost ? 0 : rq_weight; |
a5004278 | 1575 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
ffda12a1 | 1576 | __set_se_shares(tg->se[cpu], shares); |
05fa785c | 1577 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
ffda12a1 | 1578 | } |
18d95a28 | 1579 | } |
c09595f6 PZ |
1580 | |
1581 | /* | |
c8cba857 PZ |
1582 | * Re-compute the task group their per cpu shares over the given domain. |
1583 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1584 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1585 | */ |
eb755805 | 1586 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1587 | { |
cd8ad40d | 1588 | unsigned long weight, rq_weight = 0, sum_weight = 0, shares = 0; |
4a6cc4bd | 1589 | unsigned long *usd_rq_weight; |
eb755805 | 1590 | struct sched_domain *sd = data; |
34d76c41 | 1591 | unsigned long flags; |
c8cba857 | 1592 | int i; |
c09595f6 | 1593 | |
34d76c41 PZ |
1594 | if (!tg->se[0]) |
1595 | return 0; | |
1596 | ||
1597 | local_irq_save(flags); | |
4a6cc4bd | 1598 | usd_rq_weight = per_cpu_ptr(update_shares_data, smp_processor_id()); |
34d76c41 | 1599 | |
758b2cdc | 1600 | for_each_cpu(i, sched_domain_span(sd)) { |
34d76c41 | 1601 | weight = tg->cfs_rq[i]->load.weight; |
4a6cc4bd | 1602 | usd_rq_weight[i] = weight; |
34d76c41 | 1603 | |
cd8ad40d | 1604 | rq_weight += weight; |
ec4e0e2f KC |
1605 | /* |
1606 | * If there are currently no tasks on the cpu pretend there | |
1607 | * is one of average load so that when a new task gets to | |
1608 | * run here it will not get delayed by group starvation. | |
1609 | */ | |
ec4e0e2f KC |
1610 | if (!weight) |
1611 | weight = NICE_0_LOAD; | |
1612 | ||
cd8ad40d | 1613 | sum_weight += weight; |
c8cba857 | 1614 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1615 | } |
c09595f6 | 1616 | |
cd8ad40d PZ |
1617 | if (!rq_weight) |
1618 | rq_weight = sum_weight; | |
1619 | ||
c8cba857 PZ |
1620 | if ((!shares && rq_weight) || shares > tg->shares) |
1621 | shares = tg->shares; | |
1622 | ||
1623 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1624 | shares = tg->shares; | |
c09595f6 | 1625 | |
758b2cdc | 1626 | for_each_cpu(i, sched_domain_span(sd)) |
4a6cc4bd | 1627 | update_group_shares_cpu(tg, i, shares, rq_weight, usd_rq_weight); |
34d76c41 PZ |
1628 | |
1629 | local_irq_restore(flags); | |
eb755805 PZ |
1630 | |
1631 | return 0; | |
c09595f6 PZ |
1632 | } |
1633 | ||
1634 | /* | |
c8cba857 PZ |
1635 | * Compute the cpu's hierarchical load factor for each task group. |
1636 | * This needs to be done in a top-down fashion because the load of a child | |
1637 | * group is a fraction of its parents load. | |
c09595f6 | 1638 | */ |
eb755805 | 1639 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1640 | { |
c8cba857 | 1641 | unsigned long load; |
eb755805 | 1642 | long cpu = (long)data; |
c09595f6 | 1643 | |
c8cba857 PZ |
1644 | if (!tg->parent) { |
1645 | load = cpu_rq(cpu)->load.weight; | |
1646 | } else { | |
1647 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1648 | load *= tg->cfs_rq[cpu]->shares; | |
1649 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1650 | } | |
c09595f6 | 1651 | |
c8cba857 | 1652 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1653 | |
eb755805 | 1654 | return 0; |
c09595f6 PZ |
1655 | } |
1656 | ||
c8cba857 | 1657 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1658 | { |
e7097159 PZ |
1659 | s64 elapsed; |
1660 | u64 now; | |
1661 | ||
1662 | if (root_task_group_empty()) | |
1663 | return; | |
1664 | ||
1665 | now = cpu_clock(raw_smp_processor_id()); | |
1666 | elapsed = now - sd->last_update; | |
2398f2c6 PZ |
1667 | |
1668 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1669 | sd->last_update = now; | |
eb755805 | 1670 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1671 | } |
4d8d595d PZ |
1672 | } |
1673 | ||
eb755805 | 1674 | static void update_h_load(long cpu) |
c09595f6 | 1675 | { |
e7097159 PZ |
1676 | if (root_task_group_empty()) |
1677 | return; | |
1678 | ||
eb755805 | 1679 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1680 | } |
1681 | ||
c09595f6 PZ |
1682 | #else |
1683 | ||
c8cba857 | 1684 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1685 | { |
1686 | } | |
1687 | ||
18d95a28 PZ |
1688 | #endif |
1689 | ||
8f45e2b5 GH |
1690 | #ifdef CONFIG_PREEMPT |
1691 | ||
b78bb868 PZ |
1692 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1693 | ||
70574a99 | 1694 | /* |
8f45e2b5 GH |
1695 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1696 | * way at the expense of forcing extra atomic operations in all | |
1697 | * invocations. This assures that the double_lock is acquired using the | |
1698 | * same underlying policy as the spinlock_t on this architecture, which | |
1699 | * reduces latency compared to the unfair variant below. However, it | |
1700 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1701 | */ |
8f45e2b5 GH |
1702 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1703 | __releases(this_rq->lock) | |
1704 | __acquires(busiest->lock) | |
1705 | __acquires(this_rq->lock) | |
1706 | { | |
05fa785c | 1707 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1708 | double_rq_lock(this_rq, busiest); |
1709 | ||
1710 | return 1; | |
1711 | } | |
1712 | ||
1713 | #else | |
1714 | /* | |
1715 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1716 | * latency by eliminating extra atomic operations when the locks are | |
1717 | * already in proper order on entry. This favors lower cpu-ids and will | |
1718 | * grant the double lock to lower cpus over higher ids under contention, | |
1719 | * regardless of entry order into the function. | |
1720 | */ | |
1721 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1722 | __releases(this_rq->lock) |
1723 | __acquires(busiest->lock) | |
1724 | __acquires(this_rq->lock) | |
1725 | { | |
1726 | int ret = 0; | |
1727 | ||
05fa785c | 1728 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1729 | if (busiest < this_rq) { |
05fa785c TG |
1730 | raw_spin_unlock(&this_rq->lock); |
1731 | raw_spin_lock(&busiest->lock); | |
1732 | raw_spin_lock_nested(&this_rq->lock, | |
1733 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1734 | ret = 1; |
1735 | } else | |
05fa785c TG |
1736 | raw_spin_lock_nested(&busiest->lock, |
1737 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1738 | } |
1739 | return ret; | |
1740 | } | |
1741 | ||
8f45e2b5 GH |
1742 | #endif /* CONFIG_PREEMPT */ |
1743 | ||
1744 | /* | |
1745 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1746 | */ | |
1747 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1748 | { | |
1749 | if (unlikely(!irqs_disabled())) { | |
1750 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1751 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1752 | BUG_ON(1); |
1753 | } | |
1754 | ||
1755 | return _double_lock_balance(this_rq, busiest); | |
1756 | } | |
1757 | ||
70574a99 AD |
1758 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1759 | __releases(busiest->lock) | |
1760 | { | |
05fa785c | 1761 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1762 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1763 | } | |
1e3c88bd PZ |
1764 | |
1765 | /* | |
1766 | * double_rq_lock - safely lock two runqueues | |
1767 | * | |
1768 | * Note this does not disable interrupts like task_rq_lock, | |
1769 | * you need to do so manually before calling. | |
1770 | */ | |
1771 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1772 | __acquires(rq1->lock) | |
1773 | __acquires(rq2->lock) | |
1774 | { | |
1775 | BUG_ON(!irqs_disabled()); | |
1776 | if (rq1 == rq2) { | |
1777 | raw_spin_lock(&rq1->lock); | |
1778 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1779 | } else { | |
1780 | if (rq1 < rq2) { | |
1781 | raw_spin_lock(&rq1->lock); | |
1782 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1783 | } else { | |
1784 | raw_spin_lock(&rq2->lock); | |
1785 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1786 | } | |
1787 | } | |
1e3c88bd PZ |
1788 | } |
1789 | ||
1790 | /* | |
1791 | * double_rq_unlock - safely unlock two runqueues | |
1792 | * | |
1793 | * Note this does not restore interrupts like task_rq_unlock, | |
1794 | * you need to do so manually after calling. | |
1795 | */ | |
1796 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1797 | __releases(rq1->lock) | |
1798 | __releases(rq2->lock) | |
1799 | { | |
1800 | raw_spin_unlock(&rq1->lock); | |
1801 | if (rq1 != rq2) | |
1802 | raw_spin_unlock(&rq2->lock); | |
1803 | else | |
1804 | __release(rq2->lock); | |
1805 | } | |
1806 | ||
18d95a28 PZ |
1807 | #endif |
1808 | ||
30432094 | 1809 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1810 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1811 | { | |
30432094 | 1812 | #ifdef CONFIG_SMP |
34e83e85 IM |
1813 | cfs_rq->shares = shares; |
1814 | #endif | |
1815 | } | |
30432094 | 1816 | #endif |
e7693a36 | 1817 | |
74f5187a | 1818 | static void calc_load_account_idle(struct rq *this_rq); |
0bcdcf28 | 1819 | static void update_sysctl(void); |
acb4a848 | 1820 | static int get_update_sysctl_factor(void); |
dce48a84 | 1821 | |
cd29fe6f PZ |
1822 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1823 | { | |
1824 | set_task_rq(p, cpu); | |
1825 | #ifdef CONFIG_SMP | |
1826 | /* | |
1827 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1828 | * successfuly executed on another CPU. We must ensure that updates of | |
1829 | * per-task data have been completed by this moment. | |
1830 | */ | |
1831 | smp_wmb(); | |
1832 | task_thread_info(p)->cpu = cpu; | |
1833 | #endif | |
1834 | } | |
dce48a84 | 1835 | |
1e3c88bd | 1836 | static const struct sched_class rt_sched_class; |
dd41f596 IM |
1837 | |
1838 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1839 | #define for_each_class(class) \ |
1840 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1841 | |
1e3c88bd PZ |
1842 | #include "sched_stats.h" |
1843 | ||
c09595f6 | 1844 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1845 | { |
1846 | rq->nr_running++; | |
9c217245 IM |
1847 | } |
1848 | ||
c09595f6 | 1849 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1850 | { |
1851 | rq->nr_running--; | |
9c217245 IM |
1852 | } |
1853 | ||
45bf76df IM |
1854 | static void set_load_weight(struct task_struct *p) |
1855 | { | |
1856 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1857 | p->se.load.weight = prio_to_weight[0] * 2; |
1858 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1859 | return; | |
1860 | } | |
45bf76df | 1861 | |
dd41f596 IM |
1862 | /* |
1863 | * SCHED_IDLE tasks get minimal weight: | |
1864 | */ | |
1865 | if (p->policy == SCHED_IDLE) { | |
1866 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1867 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1868 | return; | |
1869 | } | |
71f8bd46 | 1870 | |
dd41f596 IM |
1871 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1872 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1873 | } |
1874 | ||
371fd7e7 | 1875 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1876 | { |
a64692a3 | 1877 | update_rq_clock(rq); |
dd41f596 | 1878 | sched_info_queued(p); |
371fd7e7 | 1879 | p->sched_class->enqueue_task(rq, p, flags); |
dd41f596 | 1880 | p->se.on_rq = 1; |
71f8bd46 IM |
1881 | } |
1882 | ||
371fd7e7 | 1883 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1884 | { |
a64692a3 | 1885 | update_rq_clock(rq); |
46ac22ba | 1886 | sched_info_dequeued(p); |
371fd7e7 | 1887 | p->sched_class->dequeue_task(rq, p, flags); |
dd41f596 | 1888 | p->se.on_rq = 0; |
71f8bd46 IM |
1889 | } |
1890 | ||
1e3c88bd PZ |
1891 | /* |
1892 | * activate_task - move a task to the runqueue. | |
1893 | */ | |
371fd7e7 | 1894 | static void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1895 | { |
1896 | if (task_contributes_to_load(p)) | |
1897 | rq->nr_uninterruptible--; | |
1898 | ||
371fd7e7 | 1899 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
1900 | inc_nr_running(rq); |
1901 | } | |
1902 | ||
1903 | /* | |
1904 | * deactivate_task - remove a task from the runqueue. | |
1905 | */ | |
371fd7e7 | 1906 | static void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1907 | { |
1908 | if (task_contributes_to_load(p)) | |
1909 | rq->nr_uninterruptible++; | |
1910 | ||
371fd7e7 | 1911 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1912 | dec_nr_running(rq); |
1913 | } | |
1914 | ||
1915 | #include "sched_idletask.c" | |
1916 | #include "sched_fair.c" | |
1917 | #include "sched_rt.c" | |
1918 | #ifdef CONFIG_SCHED_DEBUG | |
1919 | # include "sched_debug.c" | |
1920 | #endif | |
1921 | ||
14531189 | 1922 | /* |
dd41f596 | 1923 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1924 | */ |
14531189 IM |
1925 | static inline int __normal_prio(struct task_struct *p) |
1926 | { | |
dd41f596 | 1927 | return p->static_prio; |
14531189 IM |
1928 | } |
1929 | ||
b29739f9 IM |
1930 | /* |
1931 | * Calculate the expected normal priority: i.e. priority | |
1932 | * without taking RT-inheritance into account. Might be | |
1933 | * boosted by interactivity modifiers. Changes upon fork, | |
1934 | * setprio syscalls, and whenever the interactivity | |
1935 | * estimator recalculates. | |
1936 | */ | |
36c8b586 | 1937 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1938 | { |
1939 | int prio; | |
1940 | ||
e05606d3 | 1941 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1942 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1943 | else | |
1944 | prio = __normal_prio(p); | |
1945 | return prio; | |
1946 | } | |
1947 | ||
1948 | /* | |
1949 | * Calculate the current priority, i.e. the priority | |
1950 | * taken into account by the scheduler. This value might | |
1951 | * be boosted by RT tasks, or might be boosted by | |
1952 | * interactivity modifiers. Will be RT if the task got | |
1953 | * RT-boosted. If not then it returns p->normal_prio. | |
1954 | */ | |
36c8b586 | 1955 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1956 | { |
1957 | p->normal_prio = normal_prio(p); | |
1958 | /* | |
1959 | * If we are RT tasks or we were boosted to RT priority, | |
1960 | * keep the priority unchanged. Otherwise, update priority | |
1961 | * to the normal priority: | |
1962 | */ | |
1963 | if (!rt_prio(p->prio)) | |
1964 | return p->normal_prio; | |
1965 | return p->prio; | |
1966 | } | |
1967 | ||
1da177e4 LT |
1968 | /** |
1969 | * task_curr - is this task currently executing on a CPU? | |
1970 | * @p: the task in question. | |
1971 | */ | |
36c8b586 | 1972 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1973 | { |
1974 | return cpu_curr(task_cpu(p)) == p; | |
1975 | } | |
1976 | ||
cb469845 SR |
1977 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1978 | const struct sched_class *prev_class, | |
1979 | int oldprio, int running) | |
1980 | { | |
1981 | if (prev_class != p->sched_class) { | |
1982 | if (prev_class->switched_from) | |
1983 | prev_class->switched_from(rq, p, running); | |
1984 | p->sched_class->switched_to(rq, p, running); | |
1985 | } else | |
1986 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1987 | } | |
1988 | ||
1da177e4 | 1989 | #ifdef CONFIG_SMP |
cc367732 IM |
1990 | /* |
1991 | * Is this task likely cache-hot: | |
1992 | */ | |
e7693a36 | 1993 | static int |
cc367732 IM |
1994 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1995 | { | |
1996 | s64 delta; | |
1997 | ||
e6c8fba7 PZ |
1998 | if (p->sched_class != &fair_sched_class) |
1999 | return 0; | |
2000 | ||
f540a608 IM |
2001 | /* |
2002 | * Buddy candidates are cache hot: | |
2003 | */ | |
f685ceac | 2004 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
2005 | (&p->se == cfs_rq_of(&p->se)->next || |
2006 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2007 | return 1; |
2008 | ||
6bc1665b IM |
2009 | if (sysctl_sched_migration_cost == -1) |
2010 | return 1; | |
2011 | if (sysctl_sched_migration_cost == 0) | |
2012 | return 0; | |
2013 | ||
cc367732 IM |
2014 | delta = now - p->se.exec_start; |
2015 | ||
2016 | return delta < (s64)sysctl_sched_migration_cost; | |
2017 | } | |
2018 | ||
dd41f596 | 2019 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2020 | { |
e2912009 PZ |
2021 | #ifdef CONFIG_SCHED_DEBUG |
2022 | /* | |
2023 | * We should never call set_task_cpu() on a blocked task, | |
2024 | * ttwu() will sort out the placement. | |
2025 | */ | |
077614ee PZ |
2026 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
2027 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
e2912009 PZ |
2028 | #endif |
2029 | ||
de1d7286 | 2030 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2031 | |
0c69774e PZ |
2032 | if (task_cpu(p) != new_cpu) { |
2033 | p->se.nr_migrations++; | |
2034 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); | |
2035 | } | |
dd41f596 IM |
2036 | |
2037 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2038 | } |
2039 | ||
969c7921 | 2040 | struct migration_arg { |
36c8b586 | 2041 | struct task_struct *task; |
1da177e4 | 2042 | int dest_cpu; |
70b97a7f | 2043 | }; |
1da177e4 | 2044 | |
969c7921 TH |
2045 | static int migration_cpu_stop(void *data); |
2046 | ||
1da177e4 LT |
2047 | /* |
2048 | * The task's runqueue lock must be held. | |
2049 | * Returns true if you have to wait for migration thread. | |
2050 | */ | |
969c7921 | 2051 | static bool migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2052 | { |
70b97a7f | 2053 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2054 | |
2055 | /* | |
2056 | * If the task is not on a runqueue (and not running), then | |
e2912009 | 2057 | * the next wake-up will properly place the task. |
1da177e4 | 2058 | */ |
969c7921 | 2059 | return p->se.on_rq || task_running(rq, p); |
1da177e4 LT |
2060 | } |
2061 | ||
2062 | /* | |
2063 | * wait_task_inactive - wait for a thread to unschedule. | |
2064 | * | |
85ba2d86 RM |
2065 | * If @match_state is nonzero, it's the @p->state value just checked and |
2066 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2067 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2068 | * we return a positive number (its total switch count). If a second call | |
2069 | * a short while later returns the same number, the caller can be sure that | |
2070 | * @p has remained unscheduled the whole time. | |
2071 | * | |
1da177e4 LT |
2072 | * The caller must ensure that the task *will* unschedule sometime soon, |
2073 | * else this function might spin for a *long* time. This function can't | |
2074 | * be called with interrupts off, or it may introduce deadlock with | |
2075 | * smp_call_function() if an IPI is sent by the same process we are | |
2076 | * waiting to become inactive. | |
2077 | */ | |
85ba2d86 | 2078 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2079 | { |
2080 | unsigned long flags; | |
dd41f596 | 2081 | int running, on_rq; |
85ba2d86 | 2082 | unsigned long ncsw; |
70b97a7f | 2083 | struct rq *rq; |
1da177e4 | 2084 | |
3a5c359a AK |
2085 | for (;;) { |
2086 | /* | |
2087 | * We do the initial early heuristics without holding | |
2088 | * any task-queue locks at all. We'll only try to get | |
2089 | * the runqueue lock when things look like they will | |
2090 | * work out! | |
2091 | */ | |
2092 | rq = task_rq(p); | |
fa490cfd | 2093 | |
3a5c359a AK |
2094 | /* |
2095 | * If the task is actively running on another CPU | |
2096 | * still, just relax and busy-wait without holding | |
2097 | * any locks. | |
2098 | * | |
2099 | * NOTE! Since we don't hold any locks, it's not | |
2100 | * even sure that "rq" stays as the right runqueue! | |
2101 | * But we don't care, since "task_running()" will | |
2102 | * return false if the runqueue has changed and p | |
2103 | * is actually now running somewhere else! | |
2104 | */ | |
85ba2d86 RM |
2105 | while (task_running(rq, p)) { |
2106 | if (match_state && unlikely(p->state != match_state)) | |
2107 | return 0; | |
3a5c359a | 2108 | cpu_relax(); |
85ba2d86 | 2109 | } |
fa490cfd | 2110 | |
3a5c359a AK |
2111 | /* |
2112 | * Ok, time to look more closely! We need the rq | |
2113 | * lock now, to be *sure*. If we're wrong, we'll | |
2114 | * just go back and repeat. | |
2115 | */ | |
2116 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 2117 | trace_sched_wait_task(p); |
3a5c359a AK |
2118 | running = task_running(rq, p); |
2119 | on_rq = p->se.on_rq; | |
85ba2d86 | 2120 | ncsw = 0; |
f31e11d8 | 2121 | if (!match_state || p->state == match_state) |
93dcf55f | 2122 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2123 | task_rq_unlock(rq, &flags); |
fa490cfd | 2124 | |
85ba2d86 RM |
2125 | /* |
2126 | * If it changed from the expected state, bail out now. | |
2127 | */ | |
2128 | if (unlikely(!ncsw)) | |
2129 | break; | |
2130 | ||
3a5c359a AK |
2131 | /* |
2132 | * Was it really running after all now that we | |
2133 | * checked with the proper locks actually held? | |
2134 | * | |
2135 | * Oops. Go back and try again.. | |
2136 | */ | |
2137 | if (unlikely(running)) { | |
2138 | cpu_relax(); | |
2139 | continue; | |
2140 | } | |
fa490cfd | 2141 | |
3a5c359a AK |
2142 | /* |
2143 | * It's not enough that it's not actively running, | |
2144 | * it must be off the runqueue _entirely_, and not | |
2145 | * preempted! | |
2146 | * | |
80dd99b3 | 2147 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2148 | * running right now), it's preempted, and we should |
2149 | * yield - it could be a while. | |
2150 | */ | |
2151 | if (unlikely(on_rq)) { | |
2152 | schedule_timeout_uninterruptible(1); | |
2153 | continue; | |
2154 | } | |
fa490cfd | 2155 | |
3a5c359a AK |
2156 | /* |
2157 | * Ahh, all good. It wasn't running, and it wasn't | |
2158 | * runnable, which means that it will never become | |
2159 | * running in the future either. We're all done! | |
2160 | */ | |
2161 | break; | |
2162 | } | |
85ba2d86 RM |
2163 | |
2164 | return ncsw; | |
1da177e4 LT |
2165 | } |
2166 | ||
2167 | /*** | |
2168 | * kick_process - kick a running thread to enter/exit the kernel | |
2169 | * @p: the to-be-kicked thread | |
2170 | * | |
2171 | * Cause a process which is running on another CPU to enter | |
2172 | * kernel-mode, without any delay. (to get signals handled.) | |
2173 | * | |
2174 | * NOTE: this function doesnt have to take the runqueue lock, | |
2175 | * because all it wants to ensure is that the remote task enters | |
2176 | * the kernel. If the IPI races and the task has been migrated | |
2177 | * to another CPU then no harm is done and the purpose has been | |
2178 | * achieved as well. | |
2179 | */ | |
36c8b586 | 2180 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2181 | { |
2182 | int cpu; | |
2183 | ||
2184 | preempt_disable(); | |
2185 | cpu = task_cpu(p); | |
2186 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2187 | smp_send_reschedule(cpu); | |
2188 | preempt_enable(); | |
2189 | } | |
b43e3521 | 2190 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2191 | #endif /* CONFIG_SMP */ |
1da177e4 | 2192 | |
0793a61d TG |
2193 | /** |
2194 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2195 | * @p: the task to evaluate | |
2196 | * @func: the function to be called | |
2197 | * @info: the function call argument | |
2198 | * | |
2199 | * Calls the function @func when the task is currently running. This might | |
2200 | * be on the current CPU, which just calls the function directly | |
2201 | */ | |
2202 | void task_oncpu_function_call(struct task_struct *p, | |
2203 | void (*func) (void *info), void *info) | |
2204 | { | |
2205 | int cpu; | |
2206 | ||
2207 | preempt_disable(); | |
2208 | cpu = task_cpu(p); | |
2209 | if (task_curr(p)) | |
2210 | smp_call_function_single(cpu, func, info, 1); | |
2211 | preempt_enable(); | |
2212 | } | |
2213 | ||
970b13ba | 2214 | #ifdef CONFIG_SMP |
30da688e ON |
2215 | /* |
2216 | * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held. | |
2217 | */ | |
5da9a0fb PZ |
2218 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2219 | { | |
2220 | int dest_cpu; | |
2221 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
2222 | ||
2223 | /* Look for allowed, online CPU in same node. */ | |
2224 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
2225 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
2226 | return dest_cpu; | |
2227 | ||
2228 | /* Any allowed, online CPU? */ | |
2229 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); | |
2230 | if (dest_cpu < nr_cpu_ids) | |
2231 | return dest_cpu; | |
2232 | ||
2233 | /* No more Mr. Nice Guy. */ | |
897f0b3c | 2234 | if (unlikely(dest_cpu >= nr_cpu_ids)) { |
9084bb82 | 2235 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
5da9a0fb PZ |
2236 | /* |
2237 | * Don't tell them about moving exiting tasks or | |
2238 | * kernel threads (both mm NULL), since they never | |
2239 | * leave kernel. | |
2240 | */ | |
2241 | if (p->mm && printk_ratelimit()) { | |
2242 | printk(KERN_INFO "process %d (%s) no " | |
2243 | "longer affine to cpu%d\n", | |
2244 | task_pid_nr(p), p->comm, cpu); | |
2245 | } | |
2246 | } | |
2247 | ||
2248 | return dest_cpu; | |
2249 | } | |
2250 | ||
e2912009 | 2251 | /* |
30da688e | 2252 | * The caller (fork, wakeup) owns TASK_WAKING, ->cpus_allowed is stable. |
e2912009 | 2253 | */ |
970b13ba | 2254 | static inline |
0017d735 | 2255 | int select_task_rq(struct rq *rq, struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 2256 | { |
0017d735 | 2257 | int cpu = p->sched_class->select_task_rq(rq, p, sd_flags, wake_flags); |
e2912009 PZ |
2258 | |
2259 | /* | |
2260 | * In order not to call set_task_cpu() on a blocking task we need | |
2261 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2262 | * cpu. | |
2263 | * | |
2264 | * Since this is common to all placement strategies, this lives here. | |
2265 | * | |
2266 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2267 | * not worry about this generic constraint ] | |
2268 | */ | |
2269 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
70f11205 | 2270 | !cpu_online(cpu))) |
5da9a0fb | 2271 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2272 | |
2273 | return cpu; | |
970b13ba | 2274 | } |
09a40af5 MG |
2275 | |
2276 | static void update_avg(u64 *avg, u64 sample) | |
2277 | { | |
2278 | s64 diff = sample - *avg; | |
2279 | *avg += diff >> 3; | |
2280 | } | |
970b13ba PZ |
2281 | #endif |
2282 | ||
1da177e4 LT |
2283 | /*** |
2284 | * try_to_wake_up - wake up a thread | |
2285 | * @p: the to-be-woken-up thread | |
2286 | * @state: the mask of task states that can be woken | |
2287 | * @sync: do a synchronous wakeup? | |
2288 | * | |
2289 | * Put it on the run-queue if it's not already there. The "current" | |
2290 | * thread is always on the run-queue (except when the actual | |
2291 | * re-schedule is in progress), and as such you're allowed to do | |
2292 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2293 | * runnable without the overhead of this. | |
2294 | * | |
2295 | * returns failure only if the task is already active. | |
2296 | */ | |
7d478721 PZ |
2297 | static int try_to_wake_up(struct task_struct *p, unsigned int state, |
2298 | int wake_flags) | |
1da177e4 | 2299 | { |
cc367732 | 2300 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 | 2301 | unsigned long flags; |
371fd7e7 | 2302 | unsigned long en_flags = ENQUEUE_WAKEUP; |
ab3b3aa5 | 2303 | struct rq *rq; |
1da177e4 | 2304 | |
e9c84311 | 2305 | this_cpu = get_cpu(); |
2398f2c6 | 2306 | |
04e2f174 | 2307 | smp_wmb(); |
ab3b3aa5 | 2308 | rq = task_rq_lock(p, &flags); |
e9c84311 | 2309 | if (!(p->state & state)) |
1da177e4 LT |
2310 | goto out; |
2311 | ||
dd41f596 | 2312 | if (p->se.on_rq) |
1da177e4 LT |
2313 | goto out_running; |
2314 | ||
2315 | cpu = task_cpu(p); | |
cc367732 | 2316 | orig_cpu = cpu; |
1da177e4 LT |
2317 | |
2318 | #ifdef CONFIG_SMP | |
2319 | if (unlikely(task_running(rq, p))) | |
2320 | goto out_activate; | |
2321 | ||
e9c84311 PZ |
2322 | /* |
2323 | * In order to handle concurrent wakeups and release the rq->lock | |
2324 | * we put the task in TASK_WAKING state. | |
eb24073b IM |
2325 | * |
2326 | * First fix up the nr_uninterruptible count: | |
e9c84311 | 2327 | */ |
cc87f76a PZ |
2328 | if (task_contributes_to_load(p)) { |
2329 | if (likely(cpu_online(orig_cpu))) | |
2330 | rq->nr_uninterruptible--; | |
2331 | else | |
2332 | this_rq()->nr_uninterruptible--; | |
2333 | } | |
e9c84311 | 2334 | p->state = TASK_WAKING; |
efbbd05a | 2335 | |
371fd7e7 | 2336 | if (p->sched_class->task_waking) { |
efbbd05a | 2337 | p->sched_class->task_waking(rq, p); |
371fd7e7 PZ |
2338 | en_flags |= ENQUEUE_WAKING; |
2339 | } | |
efbbd05a | 2340 | |
0017d735 PZ |
2341 | cpu = select_task_rq(rq, p, SD_BALANCE_WAKE, wake_flags); |
2342 | if (cpu != orig_cpu) | |
5d2f5a61 | 2343 | set_task_cpu(p, cpu); |
0017d735 | 2344 | __task_rq_unlock(rq); |
ab19cb23 | 2345 | |
0970d299 PZ |
2346 | rq = cpu_rq(cpu); |
2347 | raw_spin_lock(&rq->lock); | |
f5dc3753 | 2348 | |
0970d299 PZ |
2349 | /* |
2350 | * We migrated the task without holding either rq->lock, however | |
2351 | * since the task is not on the task list itself, nobody else | |
2352 | * will try and migrate the task, hence the rq should match the | |
2353 | * cpu we just moved it to. | |
2354 | */ | |
2355 | WARN_ON(task_cpu(p) != cpu); | |
e9c84311 | 2356 | WARN_ON(p->state != TASK_WAKING); |
1da177e4 | 2357 | |
e7693a36 GH |
2358 | #ifdef CONFIG_SCHEDSTATS |
2359 | schedstat_inc(rq, ttwu_count); | |
2360 | if (cpu == this_cpu) | |
2361 | schedstat_inc(rq, ttwu_local); | |
2362 | else { | |
2363 | struct sched_domain *sd; | |
2364 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2365 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2366 | schedstat_inc(sd, ttwu_wake_remote); |
2367 | break; | |
2368 | } | |
2369 | } | |
2370 | } | |
6d6bc0ad | 2371 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2372 | |
1da177e4 LT |
2373 | out_activate: |
2374 | #endif /* CONFIG_SMP */ | |
41acab88 | 2375 | schedstat_inc(p, se.statistics.nr_wakeups); |
7d478721 | 2376 | if (wake_flags & WF_SYNC) |
41acab88 | 2377 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
cc367732 | 2378 | if (orig_cpu != cpu) |
41acab88 | 2379 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); |
cc367732 | 2380 | if (cpu == this_cpu) |
41acab88 | 2381 | schedstat_inc(p, se.statistics.nr_wakeups_local); |
cc367732 | 2382 | else |
41acab88 | 2383 | schedstat_inc(p, se.statistics.nr_wakeups_remote); |
371fd7e7 | 2384 | activate_task(rq, p, en_flags); |
1da177e4 LT |
2385 | success = 1; |
2386 | ||
2387 | out_running: | |
27a9da65 | 2388 | trace_sched_wakeup(p, success); |
7d478721 | 2389 | check_preempt_curr(rq, p, wake_flags); |
4ae7d5ce | 2390 | |
1da177e4 | 2391 | p->state = TASK_RUNNING; |
9a897c5a | 2392 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2393 | if (p->sched_class->task_woken) |
2394 | p->sched_class->task_woken(rq, p); | |
eae0c9df MG |
2395 | |
2396 | if (unlikely(rq->idle_stamp)) { | |
2397 | u64 delta = rq->clock - rq->idle_stamp; | |
2398 | u64 max = 2*sysctl_sched_migration_cost; | |
2399 | ||
2400 | if (delta > max) | |
2401 | rq->avg_idle = max; | |
2402 | else | |
2403 | update_avg(&rq->avg_idle, delta); | |
2404 | rq->idle_stamp = 0; | |
2405 | } | |
9a897c5a | 2406 | #endif |
1da177e4 LT |
2407 | out: |
2408 | task_rq_unlock(rq, &flags); | |
e9c84311 | 2409 | put_cpu(); |
1da177e4 LT |
2410 | |
2411 | return success; | |
2412 | } | |
2413 | ||
50fa610a DH |
2414 | /** |
2415 | * wake_up_process - Wake up a specific process | |
2416 | * @p: The process to be woken up. | |
2417 | * | |
2418 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2419 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2420 | * running. | |
2421 | * | |
2422 | * It may be assumed that this function implies a write memory barrier before | |
2423 | * changing the task state if and only if any tasks are woken up. | |
2424 | */ | |
7ad5b3a5 | 2425 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2426 | { |
d9514f6c | 2427 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2428 | } |
1da177e4 LT |
2429 | EXPORT_SYMBOL(wake_up_process); |
2430 | ||
7ad5b3a5 | 2431 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2432 | { |
2433 | return try_to_wake_up(p, state, 0); | |
2434 | } | |
2435 | ||
1da177e4 LT |
2436 | /* |
2437 | * Perform scheduler related setup for a newly forked process p. | |
2438 | * p is forked by current. | |
dd41f596 IM |
2439 | * |
2440 | * __sched_fork() is basic setup used by init_idle() too: | |
2441 | */ | |
2442 | static void __sched_fork(struct task_struct *p) | |
2443 | { | |
dd41f596 IM |
2444 | p->se.exec_start = 0; |
2445 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2446 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2447 | p->se.nr_migrations = 0; |
6cfb0d5d IM |
2448 | |
2449 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 2450 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2451 | #endif |
476d139c | 2452 | |
fa717060 | 2453 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2454 | p->se.on_rq = 0; |
4a55bd5e | 2455 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2456 | |
e107be36 AK |
2457 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2458 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2459 | #endif | |
dd41f596 IM |
2460 | } |
2461 | ||
2462 | /* | |
2463 | * fork()/clone()-time setup: | |
2464 | */ | |
2465 | void sched_fork(struct task_struct *p, int clone_flags) | |
2466 | { | |
2467 | int cpu = get_cpu(); | |
2468 | ||
2469 | __sched_fork(p); | |
06b83b5f | 2470 | /* |
0017d735 | 2471 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2472 | * nobody will actually run it, and a signal or other external |
2473 | * event cannot wake it up and insert it on the runqueue either. | |
2474 | */ | |
0017d735 | 2475 | p->state = TASK_RUNNING; |
dd41f596 | 2476 | |
b9dc29e7 MG |
2477 | /* |
2478 | * Revert to default priority/policy on fork if requested. | |
2479 | */ | |
2480 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2481 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2482 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2483 | p->normal_prio = p->static_prio; |
2484 | } | |
b9dc29e7 | 2485 | |
6c697bdf MG |
2486 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2487 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2488 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2489 | set_load_weight(p); |
2490 | } | |
2491 | ||
b9dc29e7 MG |
2492 | /* |
2493 | * We don't need the reset flag anymore after the fork. It has | |
2494 | * fulfilled its duty: | |
2495 | */ | |
2496 | p->sched_reset_on_fork = 0; | |
2497 | } | |
ca94c442 | 2498 | |
f83f9ac2 PW |
2499 | /* |
2500 | * Make sure we do not leak PI boosting priority to the child. | |
2501 | */ | |
2502 | p->prio = current->normal_prio; | |
2503 | ||
2ddbf952 HS |
2504 | if (!rt_prio(p->prio)) |
2505 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2506 | |
cd29fe6f PZ |
2507 | if (p->sched_class->task_fork) |
2508 | p->sched_class->task_fork(p); | |
2509 | ||
5f3edc1b PZ |
2510 | set_task_cpu(p, cpu); |
2511 | ||
52f17b6c | 2512 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2513 | if (likely(sched_info_on())) |
52f17b6c | 2514 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2515 | #endif |
d6077cb8 | 2516 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2517 | p->oncpu = 0; |
2518 | #endif | |
1da177e4 | 2519 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2520 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2521 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2522 | #endif |
917b627d GH |
2523 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2524 | ||
476d139c | 2525 | put_cpu(); |
1da177e4 LT |
2526 | } |
2527 | ||
2528 | /* | |
2529 | * wake_up_new_task - wake up a newly created task for the first time. | |
2530 | * | |
2531 | * This function will do some initial scheduler statistics housekeeping | |
2532 | * that must be done for every newly created context, then puts the task | |
2533 | * on the runqueue and wakes it. | |
2534 | */ | |
7ad5b3a5 | 2535 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2536 | { |
2537 | unsigned long flags; | |
dd41f596 | 2538 | struct rq *rq; |
c890692b | 2539 | int cpu __maybe_unused = get_cpu(); |
fabf318e PZ |
2540 | |
2541 | #ifdef CONFIG_SMP | |
0017d735 PZ |
2542 | rq = task_rq_lock(p, &flags); |
2543 | p->state = TASK_WAKING; | |
2544 | ||
fabf318e PZ |
2545 | /* |
2546 | * Fork balancing, do it here and not earlier because: | |
2547 | * - cpus_allowed can change in the fork path | |
2548 | * - any previously selected cpu might disappear through hotplug | |
2549 | * | |
0017d735 PZ |
2550 | * We set TASK_WAKING so that select_task_rq() can drop rq->lock |
2551 | * without people poking at ->cpus_allowed. | |
fabf318e | 2552 | */ |
0017d735 | 2553 | cpu = select_task_rq(rq, p, SD_BALANCE_FORK, 0); |
fabf318e | 2554 | set_task_cpu(p, cpu); |
1da177e4 | 2555 | |
06b83b5f | 2556 | p->state = TASK_RUNNING; |
0017d735 PZ |
2557 | task_rq_unlock(rq, &flags); |
2558 | #endif | |
2559 | ||
2560 | rq = task_rq_lock(p, &flags); | |
cd29fe6f | 2561 | activate_task(rq, p, 0); |
27a9da65 | 2562 | trace_sched_wakeup_new(p, 1); |
a7558e01 | 2563 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2564 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2565 | if (p->sched_class->task_woken) |
2566 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2567 | #endif |
dd41f596 | 2568 | task_rq_unlock(rq, &flags); |
fabf318e | 2569 | put_cpu(); |
1da177e4 LT |
2570 | } |
2571 | ||
e107be36 AK |
2572 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2573 | ||
2574 | /** | |
80dd99b3 | 2575 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2576 | * @notifier: notifier struct to register |
e107be36 AK |
2577 | */ |
2578 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2579 | { | |
2580 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2581 | } | |
2582 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2583 | ||
2584 | /** | |
2585 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2586 | * @notifier: notifier struct to unregister |
e107be36 AK |
2587 | * |
2588 | * This is safe to call from within a preemption notifier. | |
2589 | */ | |
2590 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2591 | { | |
2592 | hlist_del(¬ifier->link); | |
2593 | } | |
2594 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2595 | ||
2596 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2597 | { | |
2598 | struct preempt_notifier *notifier; | |
2599 | struct hlist_node *node; | |
2600 | ||
2601 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2602 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2603 | } | |
2604 | ||
2605 | static void | |
2606 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2607 | struct task_struct *next) | |
2608 | { | |
2609 | struct preempt_notifier *notifier; | |
2610 | struct hlist_node *node; | |
2611 | ||
2612 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2613 | notifier->ops->sched_out(notifier, next); | |
2614 | } | |
2615 | ||
6d6bc0ad | 2616 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2617 | |
2618 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2619 | { | |
2620 | } | |
2621 | ||
2622 | static void | |
2623 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2624 | struct task_struct *next) | |
2625 | { | |
2626 | } | |
2627 | ||
6d6bc0ad | 2628 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2629 | |
4866cde0 NP |
2630 | /** |
2631 | * prepare_task_switch - prepare to switch tasks | |
2632 | * @rq: the runqueue preparing to switch | |
421cee29 | 2633 | * @prev: the current task that is being switched out |
4866cde0 NP |
2634 | * @next: the task we are going to switch to. |
2635 | * | |
2636 | * This is called with the rq lock held and interrupts off. It must | |
2637 | * be paired with a subsequent finish_task_switch after the context | |
2638 | * switch. | |
2639 | * | |
2640 | * prepare_task_switch sets up locking and calls architecture specific | |
2641 | * hooks. | |
2642 | */ | |
e107be36 AK |
2643 | static inline void |
2644 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2645 | struct task_struct *next) | |
4866cde0 | 2646 | { |
e107be36 | 2647 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2648 | prepare_lock_switch(rq, next); |
2649 | prepare_arch_switch(next); | |
2650 | } | |
2651 | ||
1da177e4 LT |
2652 | /** |
2653 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2654 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2655 | * @prev: the thread we just switched away from. |
2656 | * | |
4866cde0 NP |
2657 | * finish_task_switch must be called after the context switch, paired |
2658 | * with a prepare_task_switch call before the context switch. | |
2659 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2660 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2661 | * |
2662 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2663 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2664 | * with the lock held can cause deadlocks; see schedule() for |
2665 | * details.) | |
2666 | */ | |
a9957449 | 2667 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2668 | __releases(rq->lock) |
2669 | { | |
1da177e4 | 2670 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2671 | long prev_state; |
1da177e4 LT |
2672 | |
2673 | rq->prev_mm = NULL; | |
2674 | ||
2675 | /* | |
2676 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2677 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2678 | * schedule one last time. The schedule call will never return, and |
2679 | * the scheduled task must drop that reference. | |
c394cc9f | 2680 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2681 | * still held, otherwise prev could be scheduled on another cpu, die |
2682 | * there before we look at prev->state, and then the reference would | |
2683 | * be dropped twice. | |
2684 | * Manfred Spraul <manfred@colorfullife.com> | |
2685 | */ | |
55a101f8 | 2686 | prev_state = prev->state; |
4866cde0 | 2687 | finish_arch_switch(prev); |
8381f65d JI |
2688 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2689 | local_irq_disable(); | |
2690 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
49f47433 | 2691 | perf_event_task_sched_in(current); |
8381f65d JI |
2692 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2693 | local_irq_enable(); | |
2694 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 2695 | finish_lock_switch(rq, prev); |
e8fa1362 | 2696 | |
e107be36 | 2697 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2698 | if (mm) |
2699 | mmdrop(mm); | |
c394cc9f | 2700 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2701 | /* |
2702 | * Remove function-return probe instances associated with this | |
2703 | * task and put them back on the free list. | |
9761eea8 | 2704 | */ |
c6fd91f0 | 2705 | kprobe_flush_task(prev); |
1da177e4 | 2706 | put_task_struct(prev); |
c6fd91f0 | 2707 | } |
1da177e4 LT |
2708 | } |
2709 | ||
3f029d3c GH |
2710 | #ifdef CONFIG_SMP |
2711 | ||
2712 | /* assumes rq->lock is held */ | |
2713 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2714 | { | |
2715 | if (prev->sched_class->pre_schedule) | |
2716 | prev->sched_class->pre_schedule(rq, prev); | |
2717 | } | |
2718 | ||
2719 | /* rq->lock is NOT held, but preemption is disabled */ | |
2720 | static inline void post_schedule(struct rq *rq) | |
2721 | { | |
2722 | if (rq->post_schedule) { | |
2723 | unsigned long flags; | |
2724 | ||
05fa785c | 2725 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2726 | if (rq->curr->sched_class->post_schedule) |
2727 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2728 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2729 | |
2730 | rq->post_schedule = 0; | |
2731 | } | |
2732 | } | |
2733 | ||
2734 | #else | |
da19ab51 | 2735 | |
3f029d3c GH |
2736 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2737 | { | |
2738 | } | |
2739 | ||
2740 | static inline void post_schedule(struct rq *rq) | |
2741 | { | |
1da177e4 LT |
2742 | } |
2743 | ||
3f029d3c GH |
2744 | #endif |
2745 | ||
1da177e4 LT |
2746 | /** |
2747 | * schedule_tail - first thing a freshly forked thread must call. | |
2748 | * @prev: the thread we just switched away from. | |
2749 | */ | |
36c8b586 | 2750 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2751 | __releases(rq->lock) |
2752 | { | |
70b97a7f IM |
2753 | struct rq *rq = this_rq(); |
2754 | ||
4866cde0 | 2755 | finish_task_switch(rq, prev); |
da19ab51 | 2756 | |
3f029d3c GH |
2757 | /* |
2758 | * FIXME: do we need to worry about rq being invalidated by the | |
2759 | * task_switch? | |
2760 | */ | |
2761 | post_schedule(rq); | |
70b97a7f | 2762 | |
4866cde0 NP |
2763 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2764 | /* In this case, finish_task_switch does not reenable preemption */ | |
2765 | preempt_enable(); | |
2766 | #endif | |
1da177e4 | 2767 | if (current->set_child_tid) |
b488893a | 2768 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2769 | } |
2770 | ||
2771 | /* | |
2772 | * context_switch - switch to the new MM and the new | |
2773 | * thread's register state. | |
2774 | */ | |
dd41f596 | 2775 | static inline void |
70b97a7f | 2776 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2777 | struct task_struct *next) |
1da177e4 | 2778 | { |
dd41f596 | 2779 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2780 | |
e107be36 | 2781 | prepare_task_switch(rq, prev, next); |
27a9da65 | 2782 | trace_sched_switch(prev, next); |
dd41f596 IM |
2783 | mm = next->mm; |
2784 | oldmm = prev->active_mm; | |
9226d125 ZA |
2785 | /* |
2786 | * For paravirt, this is coupled with an exit in switch_to to | |
2787 | * combine the page table reload and the switch backend into | |
2788 | * one hypercall. | |
2789 | */ | |
224101ed | 2790 | arch_start_context_switch(prev); |
9226d125 | 2791 | |
710390d9 | 2792 | if (likely(!mm)) { |
1da177e4 LT |
2793 | next->active_mm = oldmm; |
2794 | atomic_inc(&oldmm->mm_count); | |
2795 | enter_lazy_tlb(oldmm, next); | |
2796 | } else | |
2797 | switch_mm(oldmm, mm, next); | |
2798 | ||
710390d9 | 2799 | if (likely(!prev->mm)) { |
1da177e4 | 2800 | prev->active_mm = NULL; |
1da177e4 LT |
2801 | rq->prev_mm = oldmm; |
2802 | } | |
3a5f5e48 IM |
2803 | /* |
2804 | * Since the runqueue lock will be released by the next | |
2805 | * task (which is an invalid locking op but in the case | |
2806 | * of the scheduler it's an obvious special-case), so we | |
2807 | * do an early lockdep release here: | |
2808 | */ | |
2809 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2810 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2811 | #endif |
1da177e4 LT |
2812 | |
2813 | /* Here we just switch the register state and the stack. */ | |
2814 | switch_to(prev, next, prev); | |
2815 | ||
dd41f596 IM |
2816 | barrier(); |
2817 | /* | |
2818 | * this_rq must be evaluated again because prev may have moved | |
2819 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2820 | * frame will be invalid. | |
2821 | */ | |
2822 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2823 | } |
2824 | ||
2825 | /* | |
2826 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2827 | * | |
2828 | * externally visible scheduler statistics: current number of runnable | |
2829 | * threads, current number of uninterruptible-sleeping threads, total | |
2830 | * number of context switches performed since bootup. | |
2831 | */ | |
2832 | unsigned long nr_running(void) | |
2833 | { | |
2834 | unsigned long i, sum = 0; | |
2835 | ||
2836 | for_each_online_cpu(i) | |
2837 | sum += cpu_rq(i)->nr_running; | |
2838 | ||
2839 | return sum; | |
f711f609 | 2840 | } |
1da177e4 LT |
2841 | |
2842 | unsigned long nr_uninterruptible(void) | |
f711f609 | 2843 | { |
1da177e4 | 2844 | unsigned long i, sum = 0; |
f711f609 | 2845 | |
0a945022 | 2846 | for_each_possible_cpu(i) |
1da177e4 | 2847 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
2848 | |
2849 | /* | |
1da177e4 LT |
2850 | * Since we read the counters lockless, it might be slightly |
2851 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 2852 | */ |
1da177e4 LT |
2853 | if (unlikely((long)sum < 0)) |
2854 | sum = 0; | |
f711f609 | 2855 | |
1da177e4 | 2856 | return sum; |
f711f609 | 2857 | } |
f711f609 | 2858 | |
1da177e4 | 2859 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2860 | { |
cc94abfc SR |
2861 | int i; |
2862 | unsigned long long sum = 0; | |
46cb4b7c | 2863 | |
0a945022 | 2864 | for_each_possible_cpu(i) |
1da177e4 | 2865 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2866 | |
1da177e4 LT |
2867 | return sum; |
2868 | } | |
483b4ee6 | 2869 | |
1da177e4 LT |
2870 | unsigned long nr_iowait(void) |
2871 | { | |
2872 | unsigned long i, sum = 0; | |
483b4ee6 | 2873 | |
0a945022 | 2874 | for_each_possible_cpu(i) |
1da177e4 | 2875 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2876 | |
1da177e4 LT |
2877 | return sum; |
2878 | } | |
483b4ee6 | 2879 | |
69d25870 AV |
2880 | unsigned long nr_iowait_cpu(void) |
2881 | { | |
2882 | struct rq *this = this_rq(); | |
2883 | return atomic_read(&this->nr_iowait); | |
2884 | } | |
46cb4b7c | 2885 | |
69d25870 AV |
2886 | unsigned long this_cpu_load(void) |
2887 | { | |
2888 | struct rq *this = this_rq(); | |
2889 | return this->cpu_load[0]; | |
2890 | } | |
e790fb0b | 2891 | |
46cb4b7c | 2892 | |
dce48a84 TG |
2893 | /* Variables and functions for calc_load */ |
2894 | static atomic_long_t calc_load_tasks; | |
2895 | static unsigned long calc_load_update; | |
2896 | unsigned long avenrun[3]; | |
2897 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 2898 | |
74f5187a PZ |
2899 | static long calc_load_fold_active(struct rq *this_rq) |
2900 | { | |
2901 | long nr_active, delta = 0; | |
2902 | ||
2903 | nr_active = this_rq->nr_running; | |
2904 | nr_active += (long) this_rq->nr_uninterruptible; | |
2905 | ||
2906 | if (nr_active != this_rq->calc_load_active) { | |
2907 | delta = nr_active - this_rq->calc_load_active; | |
2908 | this_rq->calc_load_active = nr_active; | |
2909 | } | |
2910 | ||
2911 | return delta; | |
2912 | } | |
2913 | ||
2914 | #ifdef CONFIG_NO_HZ | |
2915 | /* | |
2916 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | |
2917 | * | |
2918 | * When making the ILB scale, we should try to pull this in as well. | |
2919 | */ | |
2920 | static atomic_long_t calc_load_tasks_idle; | |
2921 | ||
2922 | static void calc_load_account_idle(struct rq *this_rq) | |
2923 | { | |
2924 | long delta; | |
2925 | ||
2926 | delta = calc_load_fold_active(this_rq); | |
2927 | if (delta) | |
2928 | atomic_long_add(delta, &calc_load_tasks_idle); | |
2929 | } | |
2930 | ||
2931 | static long calc_load_fold_idle(void) | |
2932 | { | |
2933 | long delta = 0; | |
2934 | ||
2935 | /* | |
2936 | * Its got a race, we don't care... | |
2937 | */ | |
2938 | if (atomic_long_read(&calc_load_tasks_idle)) | |
2939 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | |
2940 | ||
2941 | return delta; | |
2942 | } | |
2943 | #else | |
2944 | static void calc_load_account_idle(struct rq *this_rq) | |
2945 | { | |
2946 | } | |
2947 | ||
2948 | static inline long calc_load_fold_idle(void) | |
2949 | { | |
2950 | return 0; | |
2951 | } | |
2952 | #endif | |
2953 | ||
2d02494f TG |
2954 | /** |
2955 | * get_avenrun - get the load average array | |
2956 | * @loads: pointer to dest load array | |
2957 | * @offset: offset to add | |
2958 | * @shift: shift count to shift the result left | |
2959 | * | |
2960 | * These values are estimates at best, so no need for locking. | |
2961 | */ | |
2962 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2963 | { | |
2964 | loads[0] = (avenrun[0] + offset) << shift; | |
2965 | loads[1] = (avenrun[1] + offset) << shift; | |
2966 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 2967 | } |
46cb4b7c | 2968 | |
dce48a84 TG |
2969 | static unsigned long |
2970 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
db1b1fef | 2971 | { |
dce48a84 TG |
2972 | load *= exp; |
2973 | load += active * (FIXED_1 - exp); | |
2974 | return load >> FSHIFT; | |
2975 | } | |
46cb4b7c SS |
2976 | |
2977 | /* | |
dce48a84 TG |
2978 | * calc_load - update the avenrun load estimates 10 ticks after the |
2979 | * CPUs have updated calc_load_tasks. | |
7835b98b | 2980 | */ |
dce48a84 | 2981 | void calc_global_load(void) |
7835b98b | 2982 | { |
dce48a84 TG |
2983 | unsigned long upd = calc_load_update + 10; |
2984 | long active; | |
1da177e4 | 2985 | |
dce48a84 TG |
2986 | if (time_before(jiffies, upd)) |
2987 | return; | |
1da177e4 | 2988 | |
dce48a84 TG |
2989 | active = atomic_long_read(&calc_load_tasks); |
2990 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 2991 | |
dce48a84 TG |
2992 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
2993 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
2994 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 2995 | |
dce48a84 TG |
2996 | calc_load_update += LOAD_FREQ; |
2997 | } | |
1da177e4 | 2998 | |
dce48a84 | 2999 | /* |
74f5187a PZ |
3000 | * Called from update_cpu_load() to periodically update this CPU's |
3001 | * active count. | |
dce48a84 TG |
3002 | */ |
3003 | static void calc_load_account_active(struct rq *this_rq) | |
3004 | { | |
74f5187a | 3005 | long delta; |
08c183f3 | 3006 | |
74f5187a PZ |
3007 | if (time_before(jiffies, this_rq->calc_load_update)) |
3008 | return; | |
783609c6 | 3009 | |
74f5187a PZ |
3010 | delta = calc_load_fold_active(this_rq); |
3011 | delta += calc_load_fold_idle(); | |
3012 | if (delta) | |
dce48a84 | 3013 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
3014 | |
3015 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
3016 | } |
3017 | ||
3018 | /* | |
dd41f596 IM |
3019 | * Update rq->cpu_load[] statistics. This function is usually called every |
3020 | * scheduler tick (TICK_NSEC). | |
46cb4b7c | 3021 | */ |
dd41f596 | 3022 | static void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 3023 | { |
495eca49 | 3024 | unsigned long this_load = this_rq->load.weight; |
dd41f596 | 3025 | int i, scale; |
46cb4b7c | 3026 | |
dd41f596 | 3027 | this_rq->nr_load_updates++; |
46cb4b7c | 3028 | |
dd41f596 IM |
3029 | /* Update our load: */ |
3030 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
3031 | unsigned long old_load, new_load; | |
7d1e6a9b | 3032 | |
dd41f596 | 3033 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 3034 | |
dd41f596 IM |
3035 | old_load = this_rq->cpu_load[i]; |
3036 | new_load = this_load; | |
a25707f3 IM |
3037 | /* |
3038 | * Round up the averaging division if load is increasing. This | |
3039 | * prevents us from getting stuck on 9 if the load is 10, for | |
3040 | * example. | |
3041 | */ | |
3042 | if (new_load > old_load) | |
3043 | new_load += scale-1; | |
dd41f596 IM |
3044 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
3045 | } | |
46cb4b7c | 3046 | |
74f5187a | 3047 | calc_load_account_active(this_rq); |
46cb4b7c SS |
3048 | } |
3049 | ||
dd41f596 | 3050 | #ifdef CONFIG_SMP |
8a0be9ef | 3051 | |
46cb4b7c | 3052 | /* |
38022906 PZ |
3053 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3054 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3055 | */ |
38022906 | 3056 | void sched_exec(void) |
46cb4b7c | 3057 | { |
38022906 | 3058 | struct task_struct *p = current; |
1da177e4 | 3059 | unsigned long flags; |
70b97a7f | 3060 | struct rq *rq; |
0017d735 | 3061 | int dest_cpu; |
46cb4b7c | 3062 | |
1da177e4 | 3063 | rq = task_rq_lock(p, &flags); |
0017d735 PZ |
3064 | dest_cpu = p->sched_class->select_task_rq(rq, p, SD_BALANCE_EXEC, 0); |
3065 | if (dest_cpu == smp_processor_id()) | |
3066 | goto unlock; | |
38022906 | 3067 | |
46cb4b7c | 3068 | /* |
38022906 | 3069 | * select_task_rq() can race against ->cpus_allowed |
46cb4b7c | 3070 | */ |
30da688e | 3071 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed) && |
969c7921 TH |
3072 | likely(cpu_active(dest_cpu)) && migrate_task(p, dest_cpu)) { |
3073 | struct migration_arg arg = { p, dest_cpu }; | |
46cb4b7c | 3074 | |
1da177e4 | 3075 | task_rq_unlock(rq, &flags); |
969c7921 | 3076 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
3077 | return; |
3078 | } | |
0017d735 | 3079 | unlock: |
1da177e4 | 3080 | task_rq_unlock(rq, &flags); |
1da177e4 | 3081 | } |
dd41f596 | 3082 | |
1da177e4 LT |
3083 | #endif |
3084 | ||
1da177e4 LT |
3085 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3086 | ||
3087 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3088 | ||
3089 | /* | |
c5f8d995 | 3090 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 3091 | * @p in case that task is currently running. |
c5f8d995 HS |
3092 | * |
3093 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 3094 | */ |
c5f8d995 HS |
3095 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
3096 | { | |
3097 | u64 ns = 0; | |
3098 | ||
3099 | if (task_current(rq, p)) { | |
3100 | update_rq_clock(rq); | |
3101 | ns = rq->clock - p->se.exec_start; | |
3102 | if ((s64)ns < 0) | |
3103 | ns = 0; | |
3104 | } | |
3105 | ||
3106 | return ns; | |
3107 | } | |
3108 | ||
bb34d92f | 3109 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 3110 | { |
1da177e4 | 3111 | unsigned long flags; |
41b86e9c | 3112 | struct rq *rq; |
bb34d92f | 3113 | u64 ns = 0; |
48f24c4d | 3114 | |
41b86e9c | 3115 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
3116 | ns = do_task_delta_exec(p, rq); |
3117 | task_rq_unlock(rq, &flags); | |
1508487e | 3118 | |
c5f8d995 HS |
3119 | return ns; |
3120 | } | |
f06febc9 | 3121 | |
c5f8d995 HS |
3122 | /* |
3123 | * Return accounted runtime for the task. | |
3124 | * In case the task is currently running, return the runtime plus current's | |
3125 | * pending runtime that have not been accounted yet. | |
3126 | */ | |
3127 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3128 | { | |
3129 | unsigned long flags; | |
3130 | struct rq *rq; | |
3131 | u64 ns = 0; | |
3132 | ||
3133 | rq = task_rq_lock(p, &flags); | |
3134 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
3135 | task_rq_unlock(rq, &flags); | |
3136 | ||
3137 | return ns; | |
3138 | } | |
48f24c4d | 3139 | |
c5f8d995 HS |
3140 | /* |
3141 | * Return sum_exec_runtime for the thread group. | |
3142 | * In case the task is currently running, return the sum plus current's | |
3143 | * pending runtime that have not been accounted yet. | |
3144 | * | |
3145 | * Note that the thread group might have other running tasks as well, | |
3146 | * so the return value not includes other pending runtime that other | |
3147 | * running tasks might have. | |
3148 | */ | |
3149 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
3150 | { | |
3151 | struct task_cputime totals; | |
3152 | unsigned long flags; | |
3153 | struct rq *rq; | |
3154 | u64 ns; | |
3155 | ||
3156 | rq = task_rq_lock(p, &flags); | |
3157 | thread_group_cputime(p, &totals); | |
3158 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 3159 | task_rq_unlock(rq, &flags); |
48f24c4d | 3160 | |
1da177e4 LT |
3161 | return ns; |
3162 | } | |
3163 | ||
1da177e4 LT |
3164 | /* |
3165 | * Account user cpu time to a process. | |
3166 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 3167 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 3168 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 3169 | */ |
457533a7 MS |
3170 | void account_user_time(struct task_struct *p, cputime_t cputime, |
3171 | cputime_t cputime_scaled) | |
1da177e4 LT |
3172 | { |
3173 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3174 | cputime64_t tmp; | |
3175 | ||
457533a7 | 3176 | /* Add user time to process. */ |
1da177e4 | 3177 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3178 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3179 | account_group_user_time(p, cputime); |
1da177e4 LT |
3180 | |
3181 | /* Add user time to cpustat. */ | |
3182 | tmp = cputime_to_cputime64(cputime); | |
3183 | if (TASK_NICE(p) > 0) | |
3184 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3185 | else | |
3186 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
3187 | |
3188 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
3189 | /* Account for user time used */ |
3190 | acct_update_integrals(p); | |
1da177e4 LT |
3191 | } |
3192 | ||
94886b84 LV |
3193 | /* |
3194 | * Account guest cpu time to a process. | |
3195 | * @p: the process that the cpu time gets accounted to | |
3196 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 3197 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 3198 | */ |
457533a7 MS |
3199 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
3200 | cputime_t cputime_scaled) | |
94886b84 LV |
3201 | { |
3202 | cputime64_t tmp; | |
3203 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3204 | ||
3205 | tmp = cputime_to_cputime64(cputime); | |
3206 | ||
457533a7 | 3207 | /* Add guest time to process. */ |
94886b84 | 3208 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3209 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3210 | account_group_user_time(p, cputime); |
94886b84 LV |
3211 | p->gtime = cputime_add(p->gtime, cputime); |
3212 | ||
457533a7 | 3213 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
3214 | if (TASK_NICE(p) > 0) { |
3215 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3216 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
3217 | } else { | |
3218 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3219 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3220 | } | |
94886b84 LV |
3221 | } |
3222 | ||
1da177e4 LT |
3223 | /* |
3224 | * Account system cpu time to a process. | |
3225 | * @p: the process that the cpu time gets accounted to | |
3226 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3227 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 3228 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
3229 | */ |
3230 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 3231 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
3232 | { |
3233 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
3234 | cputime64_t tmp; |
3235 | ||
983ed7a6 | 3236 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 3237 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
3238 | return; |
3239 | } | |
94886b84 | 3240 | |
457533a7 | 3241 | /* Add system time to process. */ |
1da177e4 | 3242 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 3243 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 3244 | account_group_system_time(p, cputime); |
1da177e4 LT |
3245 | |
3246 | /* Add system time to cpustat. */ | |
3247 | tmp = cputime_to_cputime64(cputime); | |
3248 | if (hardirq_count() - hardirq_offset) | |
3249 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3250 | else if (softirq_count()) | |
3251 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 3252 | else |
79741dd3 MS |
3253 | cpustat->system = cputime64_add(cpustat->system, tmp); |
3254 | ||
ef12fefa BR |
3255 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
3256 | ||
1da177e4 LT |
3257 | /* Account for system time used */ |
3258 | acct_update_integrals(p); | |
1da177e4 LT |
3259 | } |
3260 | ||
c66f08be | 3261 | /* |
1da177e4 | 3262 | * Account for involuntary wait time. |
1da177e4 | 3263 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 3264 | */ |
79741dd3 | 3265 | void account_steal_time(cputime_t cputime) |
c66f08be | 3266 | { |
79741dd3 MS |
3267 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
3268 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
3269 | ||
3270 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
3271 | } |
3272 | ||
1da177e4 | 3273 | /* |
79741dd3 MS |
3274 | * Account for idle time. |
3275 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 3276 | */ |
79741dd3 | 3277 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
3278 | { |
3279 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 3280 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 3281 | struct rq *rq = this_rq(); |
1da177e4 | 3282 | |
79741dd3 MS |
3283 | if (atomic_read(&rq->nr_iowait) > 0) |
3284 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
3285 | else | |
3286 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
3287 | } |
3288 | ||
79741dd3 MS |
3289 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
3290 | ||
3291 | /* | |
3292 | * Account a single tick of cpu time. | |
3293 | * @p: the process that the cpu time gets accounted to | |
3294 | * @user_tick: indicates if the tick is a user or a system tick | |
3295 | */ | |
3296 | void account_process_tick(struct task_struct *p, int user_tick) | |
3297 | { | |
a42548a1 | 3298 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
3299 | struct rq *rq = this_rq(); |
3300 | ||
3301 | if (user_tick) | |
a42548a1 | 3302 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 3303 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 3304 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
3305 | one_jiffy_scaled); |
3306 | else | |
a42548a1 | 3307 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
3308 | } |
3309 | ||
3310 | /* | |
3311 | * Account multiple ticks of steal time. | |
3312 | * @p: the process from which the cpu time has been stolen | |
3313 | * @ticks: number of stolen ticks | |
3314 | */ | |
3315 | void account_steal_ticks(unsigned long ticks) | |
3316 | { | |
3317 | account_steal_time(jiffies_to_cputime(ticks)); | |
3318 | } | |
3319 | ||
3320 | /* | |
3321 | * Account multiple ticks of idle time. | |
3322 | * @ticks: number of stolen ticks | |
3323 | */ | |
3324 | void account_idle_ticks(unsigned long ticks) | |
3325 | { | |
3326 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
3327 | } |
3328 | ||
79741dd3 MS |
3329 | #endif |
3330 | ||
49048622 BS |
3331 | /* |
3332 | * Use precise platform statistics if available: | |
3333 | */ | |
3334 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 3335 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3336 | { |
d99ca3b9 HS |
3337 | *ut = p->utime; |
3338 | *st = p->stime; | |
49048622 BS |
3339 | } |
3340 | ||
0cf55e1e | 3341 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3342 | { |
0cf55e1e HS |
3343 | struct task_cputime cputime; |
3344 | ||
3345 | thread_group_cputime(p, &cputime); | |
3346 | ||
3347 | *ut = cputime.utime; | |
3348 | *st = cputime.stime; | |
49048622 BS |
3349 | } |
3350 | #else | |
761b1d26 HS |
3351 | |
3352 | #ifndef nsecs_to_cputime | |
b7b20df9 | 3353 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
3354 | #endif |
3355 | ||
d180c5bc | 3356 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3357 | { |
d99ca3b9 | 3358 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
3359 | |
3360 | /* | |
3361 | * Use CFS's precise accounting: | |
3362 | */ | |
d180c5bc | 3363 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
3364 | |
3365 | if (total) { | |
d180c5bc HS |
3366 | u64 temp; |
3367 | ||
3368 | temp = (u64)(rtime * utime); | |
49048622 | 3369 | do_div(temp, total); |
d180c5bc HS |
3370 | utime = (cputime_t)temp; |
3371 | } else | |
3372 | utime = rtime; | |
49048622 | 3373 | |
d180c5bc HS |
3374 | /* |
3375 | * Compare with previous values, to keep monotonicity: | |
3376 | */ | |
761b1d26 | 3377 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 3378 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 3379 | |
d99ca3b9 HS |
3380 | *ut = p->prev_utime; |
3381 | *st = p->prev_stime; | |
49048622 BS |
3382 | } |
3383 | ||
0cf55e1e HS |
3384 | /* |
3385 | * Must be called with siglock held. | |
3386 | */ | |
3387 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 3388 | { |
0cf55e1e HS |
3389 | struct signal_struct *sig = p->signal; |
3390 | struct task_cputime cputime; | |
3391 | cputime_t rtime, utime, total; | |
49048622 | 3392 | |
0cf55e1e | 3393 | thread_group_cputime(p, &cputime); |
49048622 | 3394 | |
0cf55e1e HS |
3395 | total = cputime_add(cputime.utime, cputime.stime); |
3396 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 3397 | |
0cf55e1e HS |
3398 | if (total) { |
3399 | u64 temp; | |
49048622 | 3400 | |
0cf55e1e HS |
3401 | temp = (u64)(rtime * cputime.utime); |
3402 | do_div(temp, total); | |
3403 | utime = (cputime_t)temp; | |
3404 | } else | |
3405 | utime = rtime; | |
3406 | ||
3407 | sig->prev_utime = max(sig->prev_utime, utime); | |
3408 | sig->prev_stime = max(sig->prev_stime, | |
3409 | cputime_sub(rtime, sig->prev_utime)); | |
3410 | ||
3411 | *ut = sig->prev_utime; | |
3412 | *st = sig->prev_stime; | |
49048622 | 3413 | } |
49048622 | 3414 | #endif |
49048622 | 3415 | |
7835b98b CL |
3416 | /* |
3417 | * This function gets called by the timer code, with HZ frequency. | |
3418 | * We call it with interrupts disabled. | |
3419 | * | |
3420 | * It also gets called by the fork code, when changing the parent's | |
3421 | * timeslices. | |
3422 | */ | |
3423 | void scheduler_tick(void) | |
3424 | { | |
7835b98b CL |
3425 | int cpu = smp_processor_id(); |
3426 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3427 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3428 | |
3429 | sched_clock_tick(); | |
dd41f596 | 3430 | |
05fa785c | 3431 | raw_spin_lock(&rq->lock); |
3e51f33f | 3432 | update_rq_clock(rq); |
f1a438d8 | 3433 | update_cpu_load(rq); |
fa85ae24 | 3434 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 3435 | raw_spin_unlock(&rq->lock); |
7835b98b | 3436 | |
49f47433 | 3437 | perf_event_task_tick(curr); |
e220d2dc | 3438 | |
e418e1c2 | 3439 | #ifdef CONFIG_SMP |
dd41f596 IM |
3440 | rq->idle_at_tick = idle_cpu(cpu); |
3441 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3442 | #endif |
1da177e4 LT |
3443 | } |
3444 | ||
132380a0 | 3445 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
3446 | { |
3447 | if (in_lock_functions(addr)) { | |
3448 | addr = CALLER_ADDR2; | |
3449 | if (in_lock_functions(addr)) | |
3450 | addr = CALLER_ADDR3; | |
3451 | } | |
3452 | return addr; | |
3453 | } | |
1da177e4 | 3454 | |
7e49fcce SR |
3455 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3456 | defined(CONFIG_PREEMPT_TRACER)) | |
3457 | ||
43627582 | 3458 | void __kprobes add_preempt_count(int val) |
1da177e4 | 3459 | { |
6cd8a4bb | 3460 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3461 | /* |
3462 | * Underflow? | |
3463 | */ | |
9a11b49a IM |
3464 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3465 | return; | |
6cd8a4bb | 3466 | #endif |
1da177e4 | 3467 | preempt_count() += val; |
6cd8a4bb | 3468 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3469 | /* |
3470 | * Spinlock count overflowing soon? | |
3471 | */ | |
33859f7f MOS |
3472 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3473 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
3474 | #endif |
3475 | if (preempt_count() == val) | |
3476 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3477 | } |
3478 | EXPORT_SYMBOL(add_preempt_count); | |
3479 | ||
43627582 | 3480 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 3481 | { |
6cd8a4bb | 3482 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3483 | /* |
3484 | * Underflow? | |
3485 | */ | |
01e3eb82 | 3486 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3487 | return; |
1da177e4 LT |
3488 | /* |
3489 | * Is the spinlock portion underflowing? | |
3490 | */ | |
9a11b49a IM |
3491 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3492 | !(preempt_count() & PREEMPT_MASK))) | |
3493 | return; | |
6cd8a4bb | 3494 | #endif |
9a11b49a | 3495 | |
6cd8a4bb SR |
3496 | if (preempt_count() == val) |
3497 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3498 | preempt_count() -= val; |
3499 | } | |
3500 | EXPORT_SYMBOL(sub_preempt_count); | |
3501 | ||
3502 | #endif | |
3503 | ||
3504 | /* | |
dd41f596 | 3505 | * Print scheduling while atomic bug: |
1da177e4 | 3506 | */ |
dd41f596 | 3507 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3508 | { |
838225b4 SS |
3509 | struct pt_regs *regs = get_irq_regs(); |
3510 | ||
3df0fc5b PZ |
3511 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3512 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3513 | |
dd41f596 | 3514 | debug_show_held_locks(prev); |
e21f5b15 | 3515 | print_modules(); |
dd41f596 IM |
3516 | if (irqs_disabled()) |
3517 | print_irqtrace_events(prev); | |
838225b4 SS |
3518 | |
3519 | if (regs) | |
3520 | show_regs(regs); | |
3521 | else | |
3522 | dump_stack(); | |
dd41f596 | 3523 | } |
1da177e4 | 3524 | |
dd41f596 IM |
3525 | /* |
3526 | * Various schedule()-time debugging checks and statistics: | |
3527 | */ | |
3528 | static inline void schedule_debug(struct task_struct *prev) | |
3529 | { | |
1da177e4 | 3530 | /* |
41a2d6cf | 3531 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3532 | * schedule() atomically, we ignore that path for now. |
3533 | * Otherwise, whine if we are scheduling when we should not be. | |
3534 | */ | |
3f33a7ce | 3535 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
3536 | __schedule_bug(prev); |
3537 | ||
1da177e4 LT |
3538 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3539 | ||
2d72376b | 3540 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3541 | #ifdef CONFIG_SCHEDSTATS |
3542 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3543 | schedstat_inc(this_rq(), bkl_count); |
3544 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3545 | } |
3546 | #endif | |
dd41f596 IM |
3547 | } |
3548 | ||
6cecd084 | 3549 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 3550 | { |
a64692a3 MG |
3551 | if (prev->se.on_rq) |
3552 | update_rq_clock(rq); | |
3553 | rq->skip_clock_update = 0; | |
6cecd084 | 3554 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
3555 | } |
3556 | ||
dd41f596 IM |
3557 | /* |
3558 | * Pick up the highest-prio task: | |
3559 | */ | |
3560 | static inline struct task_struct * | |
b67802ea | 3561 | pick_next_task(struct rq *rq) |
dd41f596 | 3562 | { |
5522d5d5 | 3563 | const struct sched_class *class; |
dd41f596 | 3564 | struct task_struct *p; |
1da177e4 LT |
3565 | |
3566 | /* | |
dd41f596 IM |
3567 | * Optimization: we know that if all tasks are in |
3568 | * the fair class we can call that function directly: | |
1da177e4 | 3569 | */ |
dd41f596 | 3570 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3571 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3572 | if (likely(p)) |
3573 | return p; | |
1da177e4 LT |
3574 | } |
3575 | ||
dd41f596 IM |
3576 | class = sched_class_highest; |
3577 | for ( ; ; ) { | |
fb8d4724 | 3578 | p = class->pick_next_task(rq); |
dd41f596 IM |
3579 | if (p) |
3580 | return p; | |
3581 | /* | |
3582 | * Will never be NULL as the idle class always | |
3583 | * returns a non-NULL p: | |
3584 | */ | |
3585 | class = class->next; | |
3586 | } | |
3587 | } | |
1da177e4 | 3588 | |
dd41f596 IM |
3589 | /* |
3590 | * schedule() is the main scheduler function. | |
3591 | */ | |
ff743345 | 3592 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
3593 | { |
3594 | struct task_struct *prev, *next; | |
67ca7bde | 3595 | unsigned long *switch_count; |
dd41f596 | 3596 | struct rq *rq; |
31656519 | 3597 | int cpu; |
dd41f596 | 3598 | |
ff743345 PZ |
3599 | need_resched: |
3600 | preempt_disable(); | |
dd41f596 IM |
3601 | cpu = smp_processor_id(); |
3602 | rq = cpu_rq(cpu); | |
25502a6c | 3603 | rcu_note_context_switch(cpu); |
dd41f596 IM |
3604 | prev = rq->curr; |
3605 | switch_count = &prev->nivcsw; | |
3606 | ||
3607 | release_kernel_lock(prev); | |
3608 | need_resched_nonpreemptible: | |
3609 | ||
3610 | schedule_debug(prev); | |
1da177e4 | 3611 | |
31656519 | 3612 | if (sched_feat(HRTICK)) |
f333fdc9 | 3613 | hrtick_clear(rq); |
8f4d37ec | 3614 | |
05fa785c | 3615 | raw_spin_lock_irq(&rq->lock); |
1e819950 | 3616 | clear_tsk_need_resched(prev); |
1da177e4 | 3617 | |
1da177e4 | 3618 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 3619 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 3620 | prev->state = TASK_RUNNING; |
16882c1e | 3621 | else |
371fd7e7 | 3622 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
dd41f596 | 3623 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3624 | } |
3625 | ||
3f029d3c | 3626 | pre_schedule(rq, prev); |
f65eda4f | 3627 | |
dd41f596 | 3628 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3629 | idle_balance(cpu, rq); |
1da177e4 | 3630 | |
df1c99d4 | 3631 | put_prev_task(rq, prev); |
b67802ea | 3632 | next = pick_next_task(rq); |
1da177e4 | 3633 | |
1da177e4 | 3634 | if (likely(prev != next)) { |
673a90a1 | 3635 | sched_info_switch(prev, next); |
49f47433 | 3636 | perf_event_task_sched_out(prev, next); |
673a90a1 | 3637 | |
1da177e4 LT |
3638 | rq->nr_switches++; |
3639 | rq->curr = next; | |
3640 | ++*switch_count; | |
3641 | ||
dd41f596 | 3642 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
3643 | /* |
3644 | * the context switch might have flipped the stack from under | |
3645 | * us, hence refresh the local variables. | |
3646 | */ | |
3647 | cpu = smp_processor_id(); | |
3648 | rq = cpu_rq(cpu); | |
1da177e4 | 3649 | } else |
05fa785c | 3650 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 3651 | |
3f029d3c | 3652 | post_schedule(rq); |
1da177e4 | 3653 | |
6d558c3a YZ |
3654 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
3655 | prev = rq->curr; | |
3656 | switch_count = &prev->nivcsw; | |
1da177e4 | 3657 | goto need_resched_nonpreemptible; |
6d558c3a | 3658 | } |
8f4d37ec | 3659 | |
1da177e4 | 3660 | preempt_enable_no_resched(); |
ff743345 | 3661 | if (need_resched()) |
1da177e4 LT |
3662 | goto need_resched; |
3663 | } | |
1da177e4 LT |
3664 | EXPORT_SYMBOL(schedule); |
3665 | ||
c08f7829 | 3666 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d PZ |
3667 | /* |
3668 | * Look out! "owner" is an entirely speculative pointer | |
3669 | * access and not reliable. | |
3670 | */ | |
3671 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
3672 | { | |
3673 | unsigned int cpu; | |
3674 | struct rq *rq; | |
3675 | ||
3676 | if (!sched_feat(OWNER_SPIN)) | |
3677 | return 0; | |
3678 | ||
3679 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3680 | /* | |
3681 | * Need to access the cpu field knowing that | |
3682 | * DEBUG_PAGEALLOC could have unmapped it if | |
3683 | * the mutex owner just released it and exited. | |
3684 | */ | |
3685 | if (probe_kernel_address(&owner->cpu, cpu)) | |
4b402210 | 3686 | return 0; |
0d66bf6d PZ |
3687 | #else |
3688 | cpu = owner->cpu; | |
3689 | #endif | |
3690 | ||
3691 | /* | |
3692 | * Even if the access succeeded (likely case), | |
3693 | * the cpu field may no longer be valid. | |
3694 | */ | |
3695 | if (cpu >= nr_cpumask_bits) | |
4b402210 | 3696 | return 0; |
0d66bf6d PZ |
3697 | |
3698 | /* | |
3699 | * We need to validate that we can do a | |
3700 | * get_cpu() and that we have the percpu area. | |
3701 | */ | |
3702 | if (!cpu_online(cpu)) | |
4b402210 | 3703 | return 0; |
0d66bf6d PZ |
3704 | |
3705 | rq = cpu_rq(cpu); | |
3706 | ||
3707 | for (;;) { | |
3708 | /* | |
3709 | * Owner changed, break to re-assess state. | |
3710 | */ | |
3711 | if (lock->owner != owner) | |
3712 | break; | |
3713 | ||
3714 | /* | |
3715 | * Is that owner really running on that cpu? | |
3716 | */ | |
3717 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
3718 | return 0; | |
3719 | ||
3720 | cpu_relax(); | |
3721 | } | |
4b402210 | 3722 | |
0d66bf6d PZ |
3723 | return 1; |
3724 | } | |
3725 | #endif | |
3726 | ||
1da177e4 LT |
3727 | #ifdef CONFIG_PREEMPT |
3728 | /* | |
2ed6e34f | 3729 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3730 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3731 | * occur there and call schedule directly. |
3732 | */ | |
3733 | asmlinkage void __sched preempt_schedule(void) | |
3734 | { | |
3735 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3736 | |
1da177e4 LT |
3737 | /* |
3738 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3739 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3740 | */ |
beed33a8 | 3741 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3742 | return; |
3743 | ||
3a5c359a AK |
3744 | do { |
3745 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 3746 | schedule(); |
3a5c359a | 3747 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3748 | |
3a5c359a AK |
3749 | /* |
3750 | * Check again in case we missed a preemption opportunity | |
3751 | * between schedule and now. | |
3752 | */ | |
3753 | barrier(); | |
5ed0cec0 | 3754 | } while (need_resched()); |
1da177e4 | 3755 | } |
1da177e4 LT |
3756 | EXPORT_SYMBOL(preempt_schedule); |
3757 | ||
3758 | /* | |
2ed6e34f | 3759 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3760 | * off of irq context. |
3761 | * Note, that this is called and return with irqs disabled. This will | |
3762 | * protect us against recursive calling from irq. | |
3763 | */ | |
3764 | asmlinkage void __sched preempt_schedule_irq(void) | |
3765 | { | |
3766 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3767 | |
2ed6e34f | 3768 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3769 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3770 | ||
3a5c359a AK |
3771 | do { |
3772 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
3773 | local_irq_enable(); |
3774 | schedule(); | |
3775 | local_irq_disable(); | |
3a5c359a | 3776 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3777 | |
3a5c359a AK |
3778 | /* |
3779 | * Check again in case we missed a preemption opportunity | |
3780 | * between schedule and now. | |
3781 | */ | |
3782 | barrier(); | |
5ed0cec0 | 3783 | } while (need_resched()); |
1da177e4 LT |
3784 | } |
3785 | ||
3786 | #endif /* CONFIG_PREEMPT */ | |
3787 | ||
63859d4f | 3788 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3789 | void *key) |
1da177e4 | 3790 | { |
63859d4f | 3791 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3792 | } |
1da177e4 LT |
3793 | EXPORT_SYMBOL(default_wake_function); |
3794 | ||
3795 | /* | |
41a2d6cf IM |
3796 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3797 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3798 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3799 | * | |
3800 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3801 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3802 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3803 | */ | |
78ddb08f | 3804 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 3805 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 3806 | { |
2e45874c | 3807 | wait_queue_t *curr, *next; |
1da177e4 | 3808 | |
2e45874c | 3809 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3810 | unsigned flags = curr->flags; |
3811 | ||
63859d4f | 3812 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 3813 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3814 | break; |
3815 | } | |
3816 | } | |
3817 | ||
3818 | /** | |
3819 | * __wake_up - wake up threads blocked on a waitqueue. | |
3820 | * @q: the waitqueue | |
3821 | * @mode: which threads | |
3822 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3823 | * @key: is directly passed to the wakeup function |
50fa610a DH |
3824 | * |
3825 | * It may be assumed that this function implies a write memory barrier before | |
3826 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3827 | */ |
7ad5b3a5 | 3828 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 3829 | int nr_exclusive, void *key) |
1da177e4 LT |
3830 | { |
3831 | unsigned long flags; | |
3832 | ||
3833 | spin_lock_irqsave(&q->lock, flags); | |
3834 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3835 | spin_unlock_irqrestore(&q->lock, flags); | |
3836 | } | |
1da177e4 LT |
3837 | EXPORT_SYMBOL(__wake_up); |
3838 | ||
3839 | /* | |
3840 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3841 | */ | |
7ad5b3a5 | 3842 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
3843 | { |
3844 | __wake_up_common(q, mode, 1, 0, NULL); | |
3845 | } | |
22c43c81 | 3846 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 3847 | |
4ede816a DL |
3848 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
3849 | { | |
3850 | __wake_up_common(q, mode, 1, 0, key); | |
3851 | } | |
3852 | ||
1da177e4 | 3853 | /** |
4ede816a | 3854 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3855 | * @q: the waitqueue |
3856 | * @mode: which threads | |
3857 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 3858 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
3859 | * |
3860 | * The sync wakeup differs that the waker knows that it will schedule | |
3861 | * away soon, so while the target thread will be woken up, it will not | |
3862 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3863 | * with each other. This can prevent needless bouncing between CPUs. | |
3864 | * | |
3865 | * On UP it can prevent extra preemption. | |
50fa610a DH |
3866 | * |
3867 | * It may be assumed that this function implies a write memory barrier before | |
3868 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3869 | */ |
4ede816a DL |
3870 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
3871 | int nr_exclusive, void *key) | |
1da177e4 LT |
3872 | { |
3873 | unsigned long flags; | |
7d478721 | 3874 | int wake_flags = WF_SYNC; |
1da177e4 LT |
3875 | |
3876 | if (unlikely(!q)) | |
3877 | return; | |
3878 | ||
3879 | if (unlikely(!nr_exclusive)) | |
7d478721 | 3880 | wake_flags = 0; |
1da177e4 LT |
3881 | |
3882 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 3883 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
3884 | spin_unlock_irqrestore(&q->lock, flags); |
3885 | } | |
4ede816a DL |
3886 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
3887 | ||
3888 | /* | |
3889 | * __wake_up_sync - see __wake_up_sync_key() | |
3890 | */ | |
3891 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
3892 | { | |
3893 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
3894 | } | |
1da177e4 LT |
3895 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
3896 | ||
65eb3dc6 KD |
3897 | /** |
3898 | * complete: - signals a single thread waiting on this completion | |
3899 | * @x: holds the state of this particular completion | |
3900 | * | |
3901 | * This will wake up a single thread waiting on this completion. Threads will be | |
3902 | * awakened in the same order in which they were queued. | |
3903 | * | |
3904 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
3905 | * |
3906 | * It may be assumed that this function implies a write memory barrier before | |
3907 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3908 | */ |
b15136e9 | 3909 | void complete(struct completion *x) |
1da177e4 LT |
3910 | { |
3911 | unsigned long flags; | |
3912 | ||
3913 | spin_lock_irqsave(&x->wait.lock, flags); | |
3914 | x->done++; | |
d9514f6c | 3915 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
3916 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3917 | } | |
3918 | EXPORT_SYMBOL(complete); | |
3919 | ||
65eb3dc6 KD |
3920 | /** |
3921 | * complete_all: - signals all threads waiting on this completion | |
3922 | * @x: holds the state of this particular completion | |
3923 | * | |
3924 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
3925 | * |
3926 | * It may be assumed that this function implies a write memory barrier before | |
3927 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3928 | */ |
b15136e9 | 3929 | void complete_all(struct completion *x) |
1da177e4 LT |
3930 | { |
3931 | unsigned long flags; | |
3932 | ||
3933 | spin_lock_irqsave(&x->wait.lock, flags); | |
3934 | x->done += UINT_MAX/2; | |
d9514f6c | 3935 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
3936 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3937 | } | |
3938 | EXPORT_SYMBOL(complete_all); | |
3939 | ||
8cbbe86d AK |
3940 | static inline long __sched |
3941 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3942 | { |
1da177e4 LT |
3943 | if (!x->done) { |
3944 | DECLARE_WAITQUEUE(wait, current); | |
3945 | ||
a93d2f17 | 3946 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 3947 | do { |
94d3d824 | 3948 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
3949 | timeout = -ERESTARTSYS; |
3950 | break; | |
8cbbe86d AK |
3951 | } |
3952 | __set_current_state(state); | |
1da177e4 LT |
3953 | spin_unlock_irq(&x->wait.lock); |
3954 | timeout = schedule_timeout(timeout); | |
3955 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 3956 | } while (!x->done && timeout); |
1da177e4 | 3957 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
3958 | if (!x->done) |
3959 | return timeout; | |
1da177e4 LT |
3960 | } |
3961 | x->done--; | |
ea71a546 | 3962 | return timeout ?: 1; |
1da177e4 | 3963 | } |
1da177e4 | 3964 | |
8cbbe86d AK |
3965 | static long __sched |
3966 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3967 | { |
1da177e4 LT |
3968 | might_sleep(); |
3969 | ||
3970 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 3971 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 3972 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3973 | return timeout; |
3974 | } | |
1da177e4 | 3975 | |
65eb3dc6 KD |
3976 | /** |
3977 | * wait_for_completion: - waits for completion of a task | |
3978 | * @x: holds the state of this particular completion | |
3979 | * | |
3980 | * This waits to be signaled for completion of a specific task. It is NOT | |
3981 | * interruptible and there is no timeout. | |
3982 | * | |
3983 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
3984 | * and interrupt capability. Also see complete(). | |
3985 | */ | |
b15136e9 | 3986 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
3987 | { |
3988 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 3989 | } |
8cbbe86d | 3990 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 3991 | |
65eb3dc6 KD |
3992 | /** |
3993 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
3994 | * @x: holds the state of this particular completion | |
3995 | * @timeout: timeout value in jiffies | |
3996 | * | |
3997 | * This waits for either a completion of a specific task to be signaled or for a | |
3998 | * specified timeout to expire. The timeout is in jiffies. It is not | |
3999 | * interruptible. | |
4000 | */ | |
b15136e9 | 4001 | unsigned long __sched |
8cbbe86d | 4002 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4003 | { |
8cbbe86d | 4004 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4005 | } |
8cbbe86d | 4006 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4007 | |
65eb3dc6 KD |
4008 | /** |
4009 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4010 | * @x: holds the state of this particular completion | |
4011 | * | |
4012 | * This waits for completion of a specific task to be signaled. It is | |
4013 | * interruptible. | |
4014 | */ | |
8cbbe86d | 4015 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4016 | { |
51e97990 AK |
4017 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4018 | if (t == -ERESTARTSYS) | |
4019 | return t; | |
4020 | return 0; | |
0fec171c | 4021 | } |
8cbbe86d | 4022 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4023 | |
65eb3dc6 KD |
4024 | /** |
4025 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4026 | * @x: holds the state of this particular completion | |
4027 | * @timeout: timeout value in jiffies | |
4028 | * | |
4029 | * This waits for either a completion of a specific task to be signaled or for a | |
4030 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4031 | */ | |
b15136e9 | 4032 | unsigned long __sched |
8cbbe86d AK |
4033 | wait_for_completion_interruptible_timeout(struct completion *x, |
4034 | unsigned long timeout) | |
0fec171c | 4035 | { |
8cbbe86d | 4036 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4037 | } |
8cbbe86d | 4038 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4039 | |
65eb3dc6 KD |
4040 | /** |
4041 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4042 | * @x: holds the state of this particular completion | |
4043 | * | |
4044 | * This waits to be signaled for completion of a specific task. It can be | |
4045 | * interrupted by a kill signal. | |
4046 | */ | |
009e577e MW |
4047 | int __sched wait_for_completion_killable(struct completion *x) |
4048 | { | |
4049 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4050 | if (t == -ERESTARTSYS) | |
4051 | return t; | |
4052 | return 0; | |
4053 | } | |
4054 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4055 | ||
0aa12fb4 SW |
4056 | /** |
4057 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
4058 | * @x: holds the state of this particular completion | |
4059 | * @timeout: timeout value in jiffies | |
4060 | * | |
4061 | * This waits for either a completion of a specific task to be | |
4062 | * signaled or for a specified timeout to expire. It can be | |
4063 | * interrupted by a kill signal. The timeout is in jiffies. | |
4064 | */ | |
4065 | unsigned long __sched | |
4066 | wait_for_completion_killable_timeout(struct completion *x, | |
4067 | unsigned long timeout) | |
4068 | { | |
4069 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
4070 | } | |
4071 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
4072 | ||
be4de352 DC |
4073 | /** |
4074 | * try_wait_for_completion - try to decrement a completion without blocking | |
4075 | * @x: completion structure | |
4076 | * | |
4077 | * Returns: 0 if a decrement cannot be done without blocking | |
4078 | * 1 if a decrement succeeded. | |
4079 | * | |
4080 | * If a completion is being used as a counting completion, | |
4081 | * attempt to decrement the counter without blocking. This | |
4082 | * enables us to avoid waiting if the resource the completion | |
4083 | * is protecting is not available. | |
4084 | */ | |
4085 | bool try_wait_for_completion(struct completion *x) | |
4086 | { | |
7539a3b3 | 4087 | unsigned long flags; |
be4de352 DC |
4088 | int ret = 1; |
4089 | ||
7539a3b3 | 4090 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4091 | if (!x->done) |
4092 | ret = 0; | |
4093 | else | |
4094 | x->done--; | |
7539a3b3 | 4095 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4096 | return ret; |
4097 | } | |
4098 | EXPORT_SYMBOL(try_wait_for_completion); | |
4099 | ||
4100 | /** | |
4101 | * completion_done - Test to see if a completion has any waiters | |
4102 | * @x: completion structure | |
4103 | * | |
4104 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4105 | * 1 if there are no waiters. | |
4106 | * | |
4107 | */ | |
4108 | bool completion_done(struct completion *x) | |
4109 | { | |
7539a3b3 | 4110 | unsigned long flags; |
be4de352 DC |
4111 | int ret = 1; |
4112 | ||
7539a3b3 | 4113 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4114 | if (!x->done) |
4115 | ret = 0; | |
7539a3b3 | 4116 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4117 | return ret; |
4118 | } | |
4119 | EXPORT_SYMBOL(completion_done); | |
4120 | ||
8cbbe86d AK |
4121 | static long __sched |
4122 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4123 | { |
0fec171c IM |
4124 | unsigned long flags; |
4125 | wait_queue_t wait; | |
4126 | ||
4127 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4128 | |
8cbbe86d | 4129 | __set_current_state(state); |
1da177e4 | 4130 | |
8cbbe86d AK |
4131 | spin_lock_irqsave(&q->lock, flags); |
4132 | __add_wait_queue(q, &wait); | |
4133 | spin_unlock(&q->lock); | |
4134 | timeout = schedule_timeout(timeout); | |
4135 | spin_lock_irq(&q->lock); | |
4136 | __remove_wait_queue(q, &wait); | |
4137 | spin_unlock_irqrestore(&q->lock, flags); | |
4138 | ||
4139 | return timeout; | |
4140 | } | |
4141 | ||
4142 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4143 | { | |
4144 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4145 | } |
1da177e4 LT |
4146 | EXPORT_SYMBOL(interruptible_sleep_on); |
4147 | ||
0fec171c | 4148 | long __sched |
95cdf3b7 | 4149 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4150 | { |
8cbbe86d | 4151 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4152 | } |
1da177e4 LT |
4153 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4154 | ||
0fec171c | 4155 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4156 | { |
8cbbe86d | 4157 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4158 | } |
1da177e4 LT |
4159 | EXPORT_SYMBOL(sleep_on); |
4160 | ||
0fec171c | 4161 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4162 | { |
8cbbe86d | 4163 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4164 | } |
1da177e4 LT |
4165 | EXPORT_SYMBOL(sleep_on_timeout); |
4166 | ||
b29739f9 IM |
4167 | #ifdef CONFIG_RT_MUTEXES |
4168 | ||
4169 | /* | |
4170 | * rt_mutex_setprio - set the current priority of a task | |
4171 | * @p: task | |
4172 | * @prio: prio value (kernel-internal form) | |
4173 | * | |
4174 | * This function changes the 'effective' priority of a task. It does | |
4175 | * not touch ->normal_prio like __setscheduler(). | |
4176 | * | |
4177 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4178 | */ | |
36c8b586 | 4179 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4180 | { |
4181 | unsigned long flags; | |
83b699ed | 4182 | int oldprio, on_rq, running; |
70b97a7f | 4183 | struct rq *rq; |
83ab0aa0 | 4184 | const struct sched_class *prev_class; |
b29739f9 IM |
4185 | |
4186 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4187 | ||
4188 | rq = task_rq_lock(p, &flags); | |
4189 | ||
d5f9f942 | 4190 | oldprio = p->prio; |
83ab0aa0 | 4191 | prev_class = p->sched_class; |
dd41f596 | 4192 | on_rq = p->se.on_rq; |
051a1d1a | 4193 | running = task_current(rq, p); |
0e1f3483 | 4194 | if (on_rq) |
69be72c1 | 4195 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4196 | if (running) |
4197 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4198 | |
4199 | if (rt_prio(prio)) | |
4200 | p->sched_class = &rt_sched_class; | |
4201 | else | |
4202 | p->sched_class = &fair_sched_class; | |
4203 | ||
b29739f9 IM |
4204 | p->prio = prio; |
4205 | ||
0e1f3483 HS |
4206 | if (running) |
4207 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4208 | if (on_rq) { |
371fd7e7 | 4209 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 SR |
4210 | |
4211 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4212 | } |
4213 | task_rq_unlock(rq, &flags); | |
4214 | } | |
4215 | ||
4216 | #endif | |
4217 | ||
36c8b586 | 4218 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4219 | { |
dd41f596 | 4220 | int old_prio, delta, on_rq; |
1da177e4 | 4221 | unsigned long flags; |
70b97a7f | 4222 | struct rq *rq; |
1da177e4 LT |
4223 | |
4224 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4225 | return; | |
4226 | /* | |
4227 | * We have to be careful, if called from sys_setpriority(), | |
4228 | * the task might be in the middle of scheduling on another CPU. | |
4229 | */ | |
4230 | rq = task_rq_lock(p, &flags); | |
4231 | /* | |
4232 | * The RT priorities are set via sched_setscheduler(), but we still | |
4233 | * allow the 'normal' nice value to be set - but as expected | |
4234 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4235 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4236 | */ |
e05606d3 | 4237 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4238 | p->static_prio = NICE_TO_PRIO(nice); |
4239 | goto out_unlock; | |
4240 | } | |
dd41f596 | 4241 | on_rq = p->se.on_rq; |
c09595f6 | 4242 | if (on_rq) |
69be72c1 | 4243 | dequeue_task(rq, p, 0); |
1da177e4 | 4244 | |
1da177e4 | 4245 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4246 | set_load_weight(p); |
b29739f9 IM |
4247 | old_prio = p->prio; |
4248 | p->prio = effective_prio(p); | |
4249 | delta = p->prio - old_prio; | |
1da177e4 | 4250 | |
dd41f596 | 4251 | if (on_rq) { |
371fd7e7 | 4252 | enqueue_task(rq, p, 0); |
1da177e4 | 4253 | /* |
d5f9f942 AM |
4254 | * If the task increased its priority or is running and |
4255 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4256 | */ |
d5f9f942 | 4257 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4258 | resched_task(rq->curr); |
4259 | } | |
4260 | out_unlock: | |
4261 | task_rq_unlock(rq, &flags); | |
4262 | } | |
1da177e4 LT |
4263 | EXPORT_SYMBOL(set_user_nice); |
4264 | ||
e43379f1 MM |
4265 | /* |
4266 | * can_nice - check if a task can reduce its nice value | |
4267 | * @p: task | |
4268 | * @nice: nice value | |
4269 | */ | |
36c8b586 | 4270 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4271 | { |
024f4747 MM |
4272 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4273 | int nice_rlim = 20 - nice; | |
48f24c4d | 4274 | |
78d7d407 | 4275 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
4276 | capable(CAP_SYS_NICE)); |
4277 | } | |
4278 | ||
1da177e4 LT |
4279 | #ifdef __ARCH_WANT_SYS_NICE |
4280 | ||
4281 | /* | |
4282 | * sys_nice - change the priority of the current process. | |
4283 | * @increment: priority increment | |
4284 | * | |
4285 | * sys_setpriority is a more generic, but much slower function that | |
4286 | * does similar things. | |
4287 | */ | |
5add95d4 | 4288 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 4289 | { |
48f24c4d | 4290 | long nice, retval; |
1da177e4 LT |
4291 | |
4292 | /* | |
4293 | * Setpriority might change our priority at the same moment. | |
4294 | * We don't have to worry. Conceptually one call occurs first | |
4295 | * and we have a single winner. | |
4296 | */ | |
e43379f1 MM |
4297 | if (increment < -40) |
4298 | increment = -40; | |
1da177e4 LT |
4299 | if (increment > 40) |
4300 | increment = 40; | |
4301 | ||
2b8f836f | 4302 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
4303 | if (nice < -20) |
4304 | nice = -20; | |
4305 | if (nice > 19) | |
4306 | nice = 19; | |
4307 | ||
e43379f1 MM |
4308 | if (increment < 0 && !can_nice(current, nice)) |
4309 | return -EPERM; | |
4310 | ||
1da177e4 LT |
4311 | retval = security_task_setnice(current, nice); |
4312 | if (retval) | |
4313 | return retval; | |
4314 | ||
4315 | set_user_nice(current, nice); | |
4316 | return 0; | |
4317 | } | |
4318 | ||
4319 | #endif | |
4320 | ||
4321 | /** | |
4322 | * task_prio - return the priority value of a given task. | |
4323 | * @p: the task in question. | |
4324 | * | |
4325 | * This is the priority value as seen by users in /proc. | |
4326 | * RT tasks are offset by -200. Normal tasks are centered | |
4327 | * around 0, value goes from -16 to +15. | |
4328 | */ | |
36c8b586 | 4329 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4330 | { |
4331 | return p->prio - MAX_RT_PRIO; | |
4332 | } | |
4333 | ||
4334 | /** | |
4335 | * task_nice - return the nice value of a given task. | |
4336 | * @p: the task in question. | |
4337 | */ | |
36c8b586 | 4338 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4339 | { |
4340 | return TASK_NICE(p); | |
4341 | } | |
150d8bed | 4342 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4343 | |
4344 | /** | |
4345 | * idle_cpu - is a given cpu idle currently? | |
4346 | * @cpu: the processor in question. | |
4347 | */ | |
4348 | int idle_cpu(int cpu) | |
4349 | { | |
4350 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4351 | } | |
4352 | ||
1da177e4 LT |
4353 | /** |
4354 | * idle_task - return the idle task for a given cpu. | |
4355 | * @cpu: the processor in question. | |
4356 | */ | |
36c8b586 | 4357 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4358 | { |
4359 | return cpu_rq(cpu)->idle; | |
4360 | } | |
4361 | ||
4362 | /** | |
4363 | * find_process_by_pid - find a process with a matching PID value. | |
4364 | * @pid: the pid in question. | |
4365 | */ | |
a9957449 | 4366 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4367 | { |
228ebcbe | 4368 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4369 | } |
4370 | ||
4371 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4372 | static void |
4373 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4374 | { |
dd41f596 | 4375 | BUG_ON(p->se.on_rq); |
48f24c4d | 4376 | |
1da177e4 LT |
4377 | p->policy = policy; |
4378 | p->rt_priority = prio; | |
b29739f9 IM |
4379 | p->normal_prio = normal_prio(p); |
4380 | /* we are holding p->pi_lock already */ | |
4381 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4382 | if (rt_prio(p->prio)) |
4383 | p->sched_class = &rt_sched_class; | |
4384 | else | |
4385 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 4386 | set_load_weight(p); |
1da177e4 LT |
4387 | } |
4388 | ||
c69e8d9c DH |
4389 | /* |
4390 | * check the target process has a UID that matches the current process's | |
4391 | */ | |
4392 | static bool check_same_owner(struct task_struct *p) | |
4393 | { | |
4394 | const struct cred *cred = current_cred(), *pcred; | |
4395 | bool match; | |
4396 | ||
4397 | rcu_read_lock(); | |
4398 | pcred = __task_cred(p); | |
4399 | match = (cred->euid == pcred->euid || | |
4400 | cred->euid == pcred->uid); | |
4401 | rcu_read_unlock(); | |
4402 | return match; | |
4403 | } | |
4404 | ||
961ccddd RR |
4405 | static int __sched_setscheduler(struct task_struct *p, int policy, |
4406 | struct sched_param *param, bool user) | |
1da177e4 | 4407 | { |
83b699ed | 4408 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4409 | unsigned long flags; |
83ab0aa0 | 4410 | const struct sched_class *prev_class; |
70b97a7f | 4411 | struct rq *rq; |
ca94c442 | 4412 | int reset_on_fork; |
1da177e4 | 4413 | |
66e5393a SR |
4414 | /* may grab non-irq protected spin_locks */ |
4415 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4416 | recheck: |
4417 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4418 | if (policy < 0) { |
4419 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4420 | policy = oldpolicy = p->policy; |
ca94c442 LP |
4421 | } else { |
4422 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
4423 | policy &= ~SCHED_RESET_ON_FORK; | |
4424 | ||
4425 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
4426 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
4427 | policy != SCHED_IDLE) | |
4428 | return -EINVAL; | |
4429 | } | |
4430 | ||
1da177e4 LT |
4431 | /* |
4432 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4433 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4434 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4435 | */ |
4436 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4437 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4438 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4439 | return -EINVAL; |
e05606d3 | 4440 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4441 | return -EINVAL; |
4442 | ||
37e4ab3f OC |
4443 | /* |
4444 | * Allow unprivileged RT tasks to decrease priority: | |
4445 | */ | |
961ccddd | 4446 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4447 | if (rt_policy(policy)) { |
8dc3e909 | 4448 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4449 | |
4450 | if (!lock_task_sighand(p, &flags)) | |
4451 | return -ESRCH; | |
78d7d407 | 4452 | rlim_rtprio = task_rlimit(p, RLIMIT_RTPRIO); |
8dc3e909 ON |
4453 | unlock_task_sighand(p, &flags); |
4454 | ||
4455 | /* can't set/change the rt policy */ | |
4456 | if (policy != p->policy && !rlim_rtprio) | |
4457 | return -EPERM; | |
4458 | ||
4459 | /* can't increase priority */ | |
4460 | if (param->sched_priority > p->rt_priority && | |
4461 | param->sched_priority > rlim_rtprio) | |
4462 | return -EPERM; | |
4463 | } | |
dd41f596 IM |
4464 | /* |
4465 | * Like positive nice levels, dont allow tasks to | |
4466 | * move out of SCHED_IDLE either: | |
4467 | */ | |
4468 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4469 | return -EPERM; | |
5fe1d75f | 4470 | |
37e4ab3f | 4471 | /* can't change other user's priorities */ |
c69e8d9c | 4472 | if (!check_same_owner(p)) |
37e4ab3f | 4473 | return -EPERM; |
ca94c442 LP |
4474 | |
4475 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4476 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4477 | return -EPERM; | |
37e4ab3f | 4478 | } |
1da177e4 | 4479 | |
725aad24 | 4480 | if (user) { |
b68aa230 | 4481 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
4482 | /* |
4483 | * Do not allow realtime tasks into groups that have no runtime | |
4484 | * assigned. | |
4485 | */ | |
9a7e0b18 PZ |
4486 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
4487 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 4488 | return -EPERM; |
b68aa230 PZ |
4489 | #endif |
4490 | ||
725aad24 JF |
4491 | retval = security_task_setscheduler(p, policy, param); |
4492 | if (retval) | |
4493 | return retval; | |
4494 | } | |
4495 | ||
b29739f9 IM |
4496 | /* |
4497 | * make sure no PI-waiters arrive (or leave) while we are | |
4498 | * changing the priority of the task: | |
4499 | */ | |
1d615482 | 4500 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
1da177e4 LT |
4501 | /* |
4502 | * To be able to change p->policy safely, the apropriate | |
4503 | * runqueue lock must be held. | |
4504 | */ | |
b29739f9 | 4505 | rq = __task_rq_lock(p); |
1da177e4 LT |
4506 | /* recheck policy now with rq lock held */ |
4507 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4508 | policy = oldpolicy = -1; | |
b29739f9 | 4509 | __task_rq_unlock(rq); |
1d615482 | 4510 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4511 | goto recheck; |
4512 | } | |
dd41f596 | 4513 | on_rq = p->se.on_rq; |
051a1d1a | 4514 | running = task_current(rq, p); |
0e1f3483 | 4515 | if (on_rq) |
2e1cb74a | 4516 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
4517 | if (running) |
4518 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4519 | |
ca94c442 LP |
4520 | p->sched_reset_on_fork = reset_on_fork; |
4521 | ||
1da177e4 | 4522 | oldprio = p->prio; |
83ab0aa0 | 4523 | prev_class = p->sched_class; |
dd41f596 | 4524 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4525 | |
0e1f3483 HS |
4526 | if (running) |
4527 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
4528 | if (on_rq) { |
4529 | activate_task(rq, p, 0); | |
cb469845 SR |
4530 | |
4531 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 4532 | } |
b29739f9 | 4533 | __task_rq_unlock(rq); |
1d615482 | 4534 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
b29739f9 | 4535 | |
95e02ca9 TG |
4536 | rt_mutex_adjust_pi(p); |
4537 | ||
1da177e4 LT |
4538 | return 0; |
4539 | } | |
961ccddd RR |
4540 | |
4541 | /** | |
4542 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4543 | * @p: the task in question. | |
4544 | * @policy: new policy. | |
4545 | * @param: structure containing the new RT priority. | |
4546 | * | |
4547 | * NOTE that the task may be already dead. | |
4548 | */ | |
4549 | int sched_setscheduler(struct task_struct *p, int policy, | |
4550 | struct sched_param *param) | |
4551 | { | |
4552 | return __sched_setscheduler(p, policy, param, true); | |
4553 | } | |
1da177e4 LT |
4554 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4555 | ||
961ccddd RR |
4556 | /** |
4557 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4558 | * @p: the task in question. | |
4559 | * @policy: new policy. | |
4560 | * @param: structure containing the new RT priority. | |
4561 | * | |
4562 | * Just like sched_setscheduler, only don't bother checking if the | |
4563 | * current context has permission. For example, this is needed in | |
4564 | * stop_machine(): we create temporary high priority worker threads, | |
4565 | * but our caller might not have that capability. | |
4566 | */ | |
4567 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
4568 | struct sched_param *param) | |
4569 | { | |
4570 | return __sched_setscheduler(p, policy, param, false); | |
4571 | } | |
4572 | ||
95cdf3b7 IM |
4573 | static int |
4574 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4575 | { |
1da177e4 LT |
4576 | struct sched_param lparam; |
4577 | struct task_struct *p; | |
36c8b586 | 4578 | int retval; |
1da177e4 LT |
4579 | |
4580 | if (!param || pid < 0) | |
4581 | return -EINVAL; | |
4582 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4583 | return -EFAULT; | |
5fe1d75f ON |
4584 | |
4585 | rcu_read_lock(); | |
4586 | retval = -ESRCH; | |
1da177e4 | 4587 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4588 | if (p != NULL) |
4589 | retval = sched_setscheduler(p, policy, &lparam); | |
4590 | rcu_read_unlock(); | |
36c8b586 | 4591 | |
1da177e4 LT |
4592 | return retval; |
4593 | } | |
4594 | ||
4595 | /** | |
4596 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4597 | * @pid: the pid in question. | |
4598 | * @policy: new policy. | |
4599 | * @param: structure containing the new RT priority. | |
4600 | */ | |
5add95d4 HC |
4601 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4602 | struct sched_param __user *, param) | |
1da177e4 | 4603 | { |
c21761f1 JB |
4604 | /* negative values for policy are not valid */ |
4605 | if (policy < 0) | |
4606 | return -EINVAL; | |
4607 | ||
1da177e4 LT |
4608 | return do_sched_setscheduler(pid, policy, param); |
4609 | } | |
4610 | ||
4611 | /** | |
4612 | * sys_sched_setparam - set/change the RT priority of a thread | |
4613 | * @pid: the pid in question. | |
4614 | * @param: structure containing the new RT priority. | |
4615 | */ | |
5add95d4 | 4616 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4617 | { |
4618 | return do_sched_setscheduler(pid, -1, param); | |
4619 | } | |
4620 | ||
4621 | /** | |
4622 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4623 | * @pid: the pid in question. | |
4624 | */ | |
5add95d4 | 4625 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4626 | { |
36c8b586 | 4627 | struct task_struct *p; |
3a5c359a | 4628 | int retval; |
1da177e4 LT |
4629 | |
4630 | if (pid < 0) | |
3a5c359a | 4631 | return -EINVAL; |
1da177e4 LT |
4632 | |
4633 | retval = -ESRCH; | |
5fe85be0 | 4634 | rcu_read_lock(); |
1da177e4 LT |
4635 | p = find_process_by_pid(pid); |
4636 | if (p) { | |
4637 | retval = security_task_getscheduler(p); | |
4638 | if (!retval) | |
ca94c442 LP |
4639 | retval = p->policy |
4640 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4641 | } |
5fe85be0 | 4642 | rcu_read_unlock(); |
1da177e4 LT |
4643 | return retval; |
4644 | } | |
4645 | ||
4646 | /** | |
ca94c442 | 4647 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4648 | * @pid: the pid in question. |
4649 | * @param: structure containing the RT priority. | |
4650 | */ | |
5add95d4 | 4651 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4652 | { |
4653 | struct sched_param lp; | |
36c8b586 | 4654 | struct task_struct *p; |
3a5c359a | 4655 | int retval; |
1da177e4 LT |
4656 | |
4657 | if (!param || pid < 0) | |
3a5c359a | 4658 | return -EINVAL; |
1da177e4 | 4659 | |
5fe85be0 | 4660 | rcu_read_lock(); |
1da177e4 LT |
4661 | p = find_process_by_pid(pid); |
4662 | retval = -ESRCH; | |
4663 | if (!p) | |
4664 | goto out_unlock; | |
4665 | ||
4666 | retval = security_task_getscheduler(p); | |
4667 | if (retval) | |
4668 | goto out_unlock; | |
4669 | ||
4670 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4671 | rcu_read_unlock(); |
1da177e4 LT |
4672 | |
4673 | /* | |
4674 | * This one might sleep, we cannot do it with a spinlock held ... | |
4675 | */ | |
4676 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4677 | ||
1da177e4 LT |
4678 | return retval; |
4679 | ||
4680 | out_unlock: | |
5fe85be0 | 4681 | rcu_read_unlock(); |
1da177e4 LT |
4682 | return retval; |
4683 | } | |
4684 | ||
96f874e2 | 4685 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4686 | { |
5a16f3d3 | 4687 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4688 | struct task_struct *p; |
4689 | int retval; | |
1da177e4 | 4690 | |
95402b38 | 4691 | get_online_cpus(); |
23f5d142 | 4692 | rcu_read_lock(); |
1da177e4 LT |
4693 | |
4694 | p = find_process_by_pid(pid); | |
4695 | if (!p) { | |
23f5d142 | 4696 | rcu_read_unlock(); |
95402b38 | 4697 | put_online_cpus(); |
1da177e4 LT |
4698 | return -ESRCH; |
4699 | } | |
4700 | ||
23f5d142 | 4701 | /* Prevent p going away */ |
1da177e4 | 4702 | get_task_struct(p); |
23f5d142 | 4703 | rcu_read_unlock(); |
1da177e4 | 4704 | |
5a16f3d3 RR |
4705 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4706 | retval = -ENOMEM; | |
4707 | goto out_put_task; | |
4708 | } | |
4709 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4710 | retval = -ENOMEM; | |
4711 | goto out_free_cpus_allowed; | |
4712 | } | |
1da177e4 | 4713 | retval = -EPERM; |
c69e8d9c | 4714 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
4715 | goto out_unlock; |
4716 | ||
e7834f8f DQ |
4717 | retval = security_task_setscheduler(p, 0, NULL); |
4718 | if (retval) | |
4719 | goto out_unlock; | |
4720 | ||
5a16f3d3 RR |
4721 | cpuset_cpus_allowed(p, cpus_allowed); |
4722 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 4723 | again: |
5a16f3d3 | 4724 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4725 | |
8707d8b8 | 4726 | if (!retval) { |
5a16f3d3 RR |
4727 | cpuset_cpus_allowed(p, cpus_allowed); |
4728 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4729 | /* |
4730 | * We must have raced with a concurrent cpuset | |
4731 | * update. Just reset the cpus_allowed to the | |
4732 | * cpuset's cpus_allowed | |
4733 | */ | |
5a16f3d3 | 4734 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4735 | goto again; |
4736 | } | |
4737 | } | |
1da177e4 | 4738 | out_unlock: |
5a16f3d3 RR |
4739 | free_cpumask_var(new_mask); |
4740 | out_free_cpus_allowed: | |
4741 | free_cpumask_var(cpus_allowed); | |
4742 | out_put_task: | |
1da177e4 | 4743 | put_task_struct(p); |
95402b38 | 4744 | put_online_cpus(); |
1da177e4 LT |
4745 | return retval; |
4746 | } | |
4747 | ||
4748 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4749 | struct cpumask *new_mask) |
1da177e4 | 4750 | { |
96f874e2 RR |
4751 | if (len < cpumask_size()) |
4752 | cpumask_clear(new_mask); | |
4753 | else if (len > cpumask_size()) | |
4754 | len = cpumask_size(); | |
4755 | ||
1da177e4 LT |
4756 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4757 | } | |
4758 | ||
4759 | /** | |
4760 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4761 | * @pid: pid of the process | |
4762 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4763 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4764 | */ | |
5add95d4 HC |
4765 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4766 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4767 | { |
5a16f3d3 | 4768 | cpumask_var_t new_mask; |
1da177e4 LT |
4769 | int retval; |
4770 | ||
5a16f3d3 RR |
4771 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4772 | return -ENOMEM; | |
1da177e4 | 4773 | |
5a16f3d3 RR |
4774 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4775 | if (retval == 0) | |
4776 | retval = sched_setaffinity(pid, new_mask); | |
4777 | free_cpumask_var(new_mask); | |
4778 | return retval; | |
1da177e4 LT |
4779 | } |
4780 | ||
96f874e2 | 4781 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4782 | { |
36c8b586 | 4783 | struct task_struct *p; |
31605683 TG |
4784 | unsigned long flags; |
4785 | struct rq *rq; | |
1da177e4 | 4786 | int retval; |
1da177e4 | 4787 | |
95402b38 | 4788 | get_online_cpus(); |
23f5d142 | 4789 | rcu_read_lock(); |
1da177e4 LT |
4790 | |
4791 | retval = -ESRCH; | |
4792 | p = find_process_by_pid(pid); | |
4793 | if (!p) | |
4794 | goto out_unlock; | |
4795 | ||
e7834f8f DQ |
4796 | retval = security_task_getscheduler(p); |
4797 | if (retval) | |
4798 | goto out_unlock; | |
4799 | ||
31605683 | 4800 | rq = task_rq_lock(p, &flags); |
96f874e2 | 4801 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
31605683 | 4802 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
4803 | |
4804 | out_unlock: | |
23f5d142 | 4805 | rcu_read_unlock(); |
95402b38 | 4806 | put_online_cpus(); |
1da177e4 | 4807 | |
9531b62f | 4808 | return retval; |
1da177e4 LT |
4809 | } |
4810 | ||
4811 | /** | |
4812 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4813 | * @pid: pid of the process | |
4814 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4815 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4816 | */ | |
5add95d4 HC |
4817 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4818 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4819 | { |
4820 | int ret; | |
f17c8607 | 4821 | cpumask_var_t mask; |
1da177e4 | 4822 | |
84fba5ec | 4823 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4824 | return -EINVAL; |
4825 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4826 | return -EINVAL; |
4827 | ||
f17c8607 RR |
4828 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4829 | return -ENOMEM; | |
1da177e4 | 4830 | |
f17c8607 RR |
4831 | ret = sched_getaffinity(pid, mask); |
4832 | if (ret == 0) { | |
8bc037fb | 4833 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4834 | |
4835 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4836 | ret = -EFAULT; |
4837 | else | |
cd3d8031 | 4838 | ret = retlen; |
f17c8607 RR |
4839 | } |
4840 | free_cpumask_var(mask); | |
1da177e4 | 4841 | |
f17c8607 | 4842 | return ret; |
1da177e4 LT |
4843 | } |
4844 | ||
4845 | /** | |
4846 | * sys_sched_yield - yield the current processor to other threads. | |
4847 | * | |
dd41f596 IM |
4848 | * This function yields the current CPU to other tasks. If there are no |
4849 | * other threads running on this CPU then this function will return. | |
1da177e4 | 4850 | */ |
5add95d4 | 4851 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4852 | { |
70b97a7f | 4853 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4854 | |
2d72376b | 4855 | schedstat_inc(rq, yld_count); |
4530d7ab | 4856 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4857 | |
4858 | /* | |
4859 | * Since we are going to call schedule() anyway, there's | |
4860 | * no need to preempt or enable interrupts: | |
4861 | */ | |
4862 | __release(rq->lock); | |
8a25d5de | 4863 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4864 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
4865 | preempt_enable_no_resched(); |
4866 | ||
4867 | schedule(); | |
4868 | ||
4869 | return 0; | |
4870 | } | |
4871 | ||
d86ee480 PZ |
4872 | static inline int should_resched(void) |
4873 | { | |
4874 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
4875 | } | |
4876 | ||
e7b38404 | 4877 | static void __cond_resched(void) |
1da177e4 | 4878 | { |
e7aaaa69 FW |
4879 | add_preempt_count(PREEMPT_ACTIVE); |
4880 | schedule(); | |
4881 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
4882 | } |
4883 | ||
02b67cc3 | 4884 | int __sched _cond_resched(void) |
1da177e4 | 4885 | { |
d86ee480 | 4886 | if (should_resched()) { |
1da177e4 LT |
4887 | __cond_resched(); |
4888 | return 1; | |
4889 | } | |
4890 | return 0; | |
4891 | } | |
02b67cc3 | 4892 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4893 | |
4894 | /* | |
613afbf8 | 4895 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4896 | * call schedule, and on return reacquire the lock. |
4897 | * | |
41a2d6cf | 4898 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4899 | * operations here to prevent schedule() from being called twice (once via |
4900 | * spin_unlock(), once by hand). | |
4901 | */ | |
613afbf8 | 4902 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4903 | { |
d86ee480 | 4904 | int resched = should_resched(); |
6df3cecb JK |
4905 | int ret = 0; |
4906 | ||
f607c668 PZ |
4907 | lockdep_assert_held(lock); |
4908 | ||
95c354fe | 4909 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4910 | spin_unlock(lock); |
d86ee480 | 4911 | if (resched) |
95c354fe NP |
4912 | __cond_resched(); |
4913 | else | |
4914 | cpu_relax(); | |
6df3cecb | 4915 | ret = 1; |
1da177e4 | 4916 | spin_lock(lock); |
1da177e4 | 4917 | } |
6df3cecb | 4918 | return ret; |
1da177e4 | 4919 | } |
613afbf8 | 4920 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4921 | |
613afbf8 | 4922 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4923 | { |
4924 | BUG_ON(!in_softirq()); | |
4925 | ||
d86ee480 | 4926 | if (should_resched()) { |
98d82567 | 4927 | local_bh_enable(); |
1da177e4 LT |
4928 | __cond_resched(); |
4929 | local_bh_disable(); | |
4930 | return 1; | |
4931 | } | |
4932 | return 0; | |
4933 | } | |
613afbf8 | 4934 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4935 | |
1da177e4 LT |
4936 | /** |
4937 | * yield - yield the current processor to other threads. | |
4938 | * | |
72fd4a35 | 4939 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4940 | * thread runnable and calls sys_sched_yield(). |
4941 | */ | |
4942 | void __sched yield(void) | |
4943 | { | |
4944 | set_current_state(TASK_RUNNING); | |
4945 | sys_sched_yield(); | |
4946 | } | |
1da177e4 LT |
4947 | EXPORT_SYMBOL(yield); |
4948 | ||
4949 | /* | |
41a2d6cf | 4950 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4951 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4952 | */ |
4953 | void __sched io_schedule(void) | |
4954 | { | |
54d35f29 | 4955 | struct rq *rq = raw_rq(); |
1da177e4 | 4956 | |
0ff92245 | 4957 | delayacct_blkio_start(); |
1da177e4 | 4958 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 4959 | current->in_iowait = 1; |
1da177e4 | 4960 | schedule(); |
8f0dfc34 | 4961 | current->in_iowait = 0; |
1da177e4 | 4962 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4963 | delayacct_blkio_end(); |
1da177e4 | 4964 | } |
1da177e4 LT |
4965 | EXPORT_SYMBOL(io_schedule); |
4966 | ||
4967 | long __sched io_schedule_timeout(long timeout) | |
4968 | { | |
54d35f29 | 4969 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4970 | long ret; |
4971 | ||
0ff92245 | 4972 | delayacct_blkio_start(); |
1da177e4 | 4973 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 4974 | current->in_iowait = 1; |
1da177e4 | 4975 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4976 | current->in_iowait = 0; |
1da177e4 | 4977 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4978 | delayacct_blkio_end(); |
1da177e4 LT |
4979 | return ret; |
4980 | } | |
4981 | ||
4982 | /** | |
4983 | * sys_sched_get_priority_max - return maximum RT priority. | |
4984 | * @policy: scheduling class. | |
4985 | * | |
4986 | * this syscall returns the maximum rt_priority that can be used | |
4987 | * by a given scheduling class. | |
4988 | */ | |
5add95d4 | 4989 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4990 | { |
4991 | int ret = -EINVAL; | |
4992 | ||
4993 | switch (policy) { | |
4994 | case SCHED_FIFO: | |
4995 | case SCHED_RR: | |
4996 | ret = MAX_USER_RT_PRIO-1; | |
4997 | break; | |
4998 | case SCHED_NORMAL: | |
b0a9499c | 4999 | case SCHED_BATCH: |
dd41f596 | 5000 | case SCHED_IDLE: |
1da177e4 LT |
5001 | ret = 0; |
5002 | break; | |
5003 | } | |
5004 | return ret; | |
5005 | } | |
5006 | ||
5007 | /** | |
5008 | * sys_sched_get_priority_min - return minimum RT priority. | |
5009 | * @policy: scheduling class. | |
5010 | * | |
5011 | * this syscall returns the minimum rt_priority that can be used | |
5012 | * by a given scheduling class. | |
5013 | */ | |
5add95d4 | 5014 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5015 | { |
5016 | int ret = -EINVAL; | |
5017 | ||
5018 | switch (policy) { | |
5019 | case SCHED_FIFO: | |
5020 | case SCHED_RR: | |
5021 | ret = 1; | |
5022 | break; | |
5023 | case SCHED_NORMAL: | |
b0a9499c | 5024 | case SCHED_BATCH: |
dd41f596 | 5025 | case SCHED_IDLE: |
1da177e4 LT |
5026 | ret = 0; |
5027 | } | |
5028 | return ret; | |
5029 | } | |
5030 | ||
5031 | /** | |
5032 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5033 | * @pid: pid of the process. | |
5034 | * @interval: userspace pointer to the timeslice value. | |
5035 | * | |
5036 | * this syscall writes the default timeslice value of a given process | |
5037 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5038 | */ | |
17da2bd9 | 5039 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5040 | struct timespec __user *, interval) |
1da177e4 | 5041 | { |
36c8b586 | 5042 | struct task_struct *p; |
a4ec24b4 | 5043 | unsigned int time_slice; |
dba091b9 TG |
5044 | unsigned long flags; |
5045 | struct rq *rq; | |
3a5c359a | 5046 | int retval; |
1da177e4 | 5047 | struct timespec t; |
1da177e4 LT |
5048 | |
5049 | if (pid < 0) | |
3a5c359a | 5050 | return -EINVAL; |
1da177e4 LT |
5051 | |
5052 | retval = -ESRCH; | |
1a551ae7 | 5053 | rcu_read_lock(); |
1da177e4 LT |
5054 | p = find_process_by_pid(pid); |
5055 | if (!p) | |
5056 | goto out_unlock; | |
5057 | ||
5058 | retval = security_task_getscheduler(p); | |
5059 | if (retval) | |
5060 | goto out_unlock; | |
5061 | ||
dba091b9 TG |
5062 | rq = task_rq_lock(p, &flags); |
5063 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
5064 | task_rq_unlock(rq, &flags); | |
a4ec24b4 | 5065 | |
1a551ae7 | 5066 | rcu_read_unlock(); |
a4ec24b4 | 5067 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5068 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5069 | return retval; |
3a5c359a | 5070 | |
1da177e4 | 5071 | out_unlock: |
1a551ae7 | 5072 | rcu_read_unlock(); |
1da177e4 LT |
5073 | return retval; |
5074 | } | |
5075 | ||
7c731e0a | 5076 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5077 | |
82a1fcb9 | 5078 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5079 | { |
1da177e4 | 5080 | unsigned long free = 0; |
36c8b586 | 5081 | unsigned state; |
1da177e4 | 5082 | |
1da177e4 | 5083 | state = p->state ? __ffs(p->state) + 1 : 0; |
3df0fc5b | 5084 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5085 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5086 | #if BITS_PER_LONG == 32 |
1da177e4 | 5087 | if (state == TASK_RUNNING) |
3df0fc5b | 5088 | printk(KERN_CONT " running "); |
1da177e4 | 5089 | else |
3df0fc5b | 5090 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5091 | #else |
5092 | if (state == TASK_RUNNING) | |
3df0fc5b | 5093 | printk(KERN_CONT " running task "); |
1da177e4 | 5094 | else |
3df0fc5b | 5095 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5096 | #endif |
5097 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5098 | free = stack_not_used(p); |
1da177e4 | 5099 | #endif |
3df0fc5b | 5100 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
5101 | task_pid_nr(p), task_pid_nr(p->real_parent), |
5102 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 5103 | |
5fb5e6de | 5104 | show_stack(p, NULL); |
1da177e4 LT |
5105 | } |
5106 | ||
e59e2ae2 | 5107 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5108 | { |
36c8b586 | 5109 | struct task_struct *g, *p; |
1da177e4 | 5110 | |
4bd77321 | 5111 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5112 | printk(KERN_INFO |
5113 | " task PC stack pid father\n"); | |
1da177e4 | 5114 | #else |
3df0fc5b PZ |
5115 | printk(KERN_INFO |
5116 | " task PC stack pid father\n"); | |
1da177e4 LT |
5117 | #endif |
5118 | read_lock(&tasklist_lock); | |
5119 | do_each_thread(g, p) { | |
5120 | /* | |
5121 | * reset the NMI-timeout, listing all files on a slow | |
5122 | * console might take alot of time: | |
5123 | */ | |
5124 | touch_nmi_watchdog(); | |
39bc89fd | 5125 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5126 | sched_show_task(p); |
1da177e4 LT |
5127 | } while_each_thread(g, p); |
5128 | ||
04c9167f JF |
5129 | touch_all_softlockup_watchdogs(); |
5130 | ||
dd41f596 IM |
5131 | #ifdef CONFIG_SCHED_DEBUG |
5132 | sysrq_sched_debug_show(); | |
5133 | #endif | |
1da177e4 | 5134 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5135 | /* |
5136 | * Only show locks if all tasks are dumped: | |
5137 | */ | |
93335a21 | 5138 | if (!state_filter) |
e59e2ae2 | 5139 | debug_show_all_locks(); |
1da177e4 LT |
5140 | } |
5141 | ||
1df21055 IM |
5142 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5143 | { | |
dd41f596 | 5144 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5145 | } |
5146 | ||
f340c0d1 IM |
5147 | /** |
5148 | * init_idle - set up an idle thread for a given CPU | |
5149 | * @idle: task in question | |
5150 | * @cpu: cpu the idle task belongs to | |
5151 | * | |
5152 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5153 | * flag, to make booting more robust. | |
5154 | */ | |
5c1e1767 | 5155 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5156 | { |
70b97a7f | 5157 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5158 | unsigned long flags; |
5159 | ||
05fa785c | 5160 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 5161 | |
dd41f596 | 5162 | __sched_fork(idle); |
06b83b5f | 5163 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5164 | idle->se.exec_start = sched_clock(); |
5165 | ||
96f874e2 | 5166 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 5167 | __set_task_cpu(idle, cpu); |
1da177e4 | 5168 | |
1da177e4 | 5169 | rq->curr = rq->idle = idle; |
4866cde0 NP |
5170 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5171 | idle->oncpu = 1; | |
5172 | #endif | |
05fa785c | 5173 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5174 | |
5175 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5176 | #if defined(CONFIG_PREEMPT) |
5177 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5178 | #else | |
a1261f54 | 5179 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5180 | #endif |
dd41f596 IM |
5181 | /* |
5182 | * The idle tasks have their own, simple scheduling class: | |
5183 | */ | |
5184 | idle->sched_class = &idle_sched_class; | |
fb52607a | 5185 | ftrace_graph_init_task(idle); |
1da177e4 LT |
5186 | } |
5187 | ||
5188 | /* | |
5189 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5190 | * indicates which cpus entered this state. This is used | |
5191 | * in the rcu update to wait only for active cpus. For system | |
5192 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 5193 | * always be CPU_BITS_NONE. |
1da177e4 | 5194 | */ |
6a7b3dc3 | 5195 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 5196 | |
19978ca6 IM |
5197 | /* |
5198 | * Increase the granularity value when there are more CPUs, | |
5199 | * because with more CPUs the 'effective latency' as visible | |
5200 | * to users decreases. But the relationship is not linear, | |
5201 | * so pick a second-best guess by going with the log2 of the | |
5202 | * number of CPUs. | |
5203 | * | |
5204 | * This idea comes from the SD scheduler of Con Kolivas: | |
5205 | */ | |
acb4a848 | 5206 | static int get_update_sysctl_factor(void) |
19978ca6 | 5207 | { |
4ca3ef71 | 5208 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
5209 | unsigned int factor; |
5210 | ||
5211 | switch (sysctl_sched_tunable_scaling) { | |
5212 | case SCHED_TUNABLESCALING_NONE: | |
5213 | factor = 1; | |
5214 | break; | |
5215 | case SCHED_TUNABLESCALING_LINEAR: | |
5216 | factor = cpus; | |
5217 | break; | |
5218 | case SCHED_TUNABLESCALING_LOG: | |
5219 | default: | |
5220 | factor = 1 + ilog2(cpus); | |
5221 | break; | |
5222 | } | |
19978ca6 | 5223 | |
acb4a848 CE |
5224 | return factor; |
5225 | } | |
19978ca6 | 5226 | |
acb4a848 CE |
5227 | static void update_sysctl(void) |
5228 | { | |
5229 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 5230 | |
0bcdcf28 CE |
5231 | #define SET_SYSCTL(name) \ |
5232 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
5233 | SET_SYSCTL(sched_min_granularity); | |
5234 | SET_SYSCTL(sched_latency); | |
5235 | SET_SYSCTL(sched_wakeup_granularity); | |
5236 | SET_SYSCTL(sched_shares_ratelimit); | |
5237 | #undef SET_SYSCTL | |
5238 | } | |
55cd5340 | 5239 | |
0bcdcf28 CE |
5240 | static inline void sched_init_granularity(void) |
5241 | { | |
5242 | update_sysctl(); | |
19978ca6 IM |
5243 | } |
5244 | ||
1da177e4 LT |
5245 | #ifdef CONFIG_SMP |
5246 | /* | |
5247 | * This is how migration works: | |
5248 | * | |
969c7921 TH |
5249 | * 1) we invoke migration_cpu_stop() on the target CPU using |
5250 | * stop_one_cpu(). | |
5251 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
5252 | * off the CPU) | |
5253 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
5254 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 5255 | * it and puts it into the right queue. |
969c7921 TH |
5256 | * 5) stopper completes and stop_one_cpu() returns and the migration |
5257 | * is done. | |
1da177e4 LT |
5258 | */ |
5259 | ||
5260 | /* | |
5261 | * Change a given task's CPU affinity. Migrate the thread to a | |
5262 | * proper CPU and schedule it away if the CPU it's executing on | |
5263 | * is removed from the allowed bitmask. | |
5264 | * | |
5265 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5266 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5267 | * call is not atomic; no spinlocks may be held. |
5268 | */ | |
96f874e2 | 5269 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
5270 | { |
5271 | unsigned long flags; | |
70b97a7f | 5272 | struct rq *rq; |
969c7921 | 5273 | unsigned int dest_cpu; |
48f24c4d | 5274 | int ret = 0; |
1da177e4 | 5275 | |
65cc8e48 PZ |
5276 | /* |
5277 | * Serialize against TASK_WAKING so that ttwu() and wunt() can | |
5278 | * drop the rq->lock and still rely on ->cpus_allowed. | |
5279 | */ | |
5280 | again: | |
5281 | while (task_is_waking(p)) | |
5282 | cpu_relax(); | |
1da177e4 | 5283 | rq = task_rq_lock(p, &flags); |
65cc8e48 PZ |
5284 | if (task_is_waking(p)) { |
5285 | task_rq_unlock(rq, &flags); | |
5286 | goto again; | |
5287 | } | |
e2912009 | 5288 | |
6ad4c188 | 5289 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
5290 | ret = -EINVAL; |
5291 | goto out; | |
5292 | } | |
5293 | ||
9985b0ba | 5294 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 5295 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
5296 | ret = -EINVAL; |
5297 | goto out; | |
5298 | } | |
5299 | ||
73fe6aae | 5300 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5301 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5302 | else { |
96f874e2 RR |
5303 | cpumask_copy(&p->cpus_allowed, new_mask); |
5304 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
5305 | } |
5306 | ||
1da177e4 | 5307 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 5308 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
5309 | goto out; |
5310 | ||
969c7921 TH |
5311 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
5312 | if (migrate_task(p, dest_cpu)) { | |
5313 | struct migration_arg arg = { p, dest_cpu }; | |
1da177e4 LT |
5314 | /* Need help from migration thread: drop lock and wait. */ |
5315 | task_rq_unlock(rq, &flags); | |
969c7921 | 5316 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
5317 | tlb_migrate_finish(p->mm); |
5318 | return 0; | |
5319 | } | |
5320 | out: | |
5321 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5322 | |
1da177e4 LT |
5323 | return ret; |
5324 | } | |
cd8ba7cd | 5325 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5326 | |
5327 | /* | |
41a2d6cf | 5328 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5329 | * this because either it can't run here any more (set_cpus_allowed() |
5330 | * away from this CPU, or CPU going down), or because we're | |
5331 | * attempting to rebalance this task on exec (sched_exec). | |
5332 | * | |
5333 | * So we race with normal scheduler movements, but that's OK, as long | |
5334 | * as the task is no longer on this CPU. | |
efc30814 KK |
5335 | * |
5336 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5337 | */ |
efc30814 | 5338 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5339 | { |
70b97a7f | 5340 | struct rq *rq_dest, *rq_src; |
e2912009 | 5341 | int ret = 0; |
1da177e4 | 5342 | |
e761b772 | 5343 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5344 | return ret; |
1da177e4 LT |
5345 | |
5346 | rq_src = cpu_rq(src_cpu); | |
5347 | rq_dest = cpu_rq(dest_cpu); | |
5348 | ||
5349 | double_rq_lock(rq_src, rq_dest); | |
5350 | /* Already moved. */ | |
5351 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5352 | goto done; |
1da177e4 | 5353 | /* Affinity changed (again). */ |
96f874e2 | 5354 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 5355 | goto fail; |
1da177e4 | 5356 | |
e2912009 PZ |
5357 | /* |
5358 | * If we're not on a rq, the next wake-up will ensure we're | |
5359 | * placed properly. | |
5360 | */ | |
5361 | if (p->se.on_rq) { | |
2e1cb74a | 5362 | deactivate_task(rq_src, p, 0); |
e2912009 | 5363 | set_task_cpu(p, dest_cpu); |
dd41f596 | 5364 | activate_task(rq_dest, p, 0); |
15afe09b | 5365 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 5366 | } |
b1e38734 | 5367 | done: |
efc30814 | 5368 | ret = 1; |
b1e38734 | 5369 | fail: |
1da177e4 | 5370 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 5371 | return ret; |
1da177e4 LT |
5372 | } |
5373 | ||
5374 | /* | |
969c7921 TH |
5375 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
5376 | * and performs thread migration by bumping thread off CPU then | |
5377 | * 'pushing' onto another runqueue. | |
1da177e4 | 5378 | */ |
969c7921 | 5379 | static int migration_cpu_stop(void *data) |
1da177e4 | 5380 | { |
969c7921 | 5381 | struct migration_arg *arg = data; |
f7b4cddc | 5382 | |
969c7921 TH |
5383 | /* |
5384 | * The original target cpu might have gone down and we might | |
5385 | * be on another cpu but it doesn't matter. | |
5386 | */ | |
f7b4cddc | 5387 | local_irq_disable(); |
969c7921 | 5388 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 5389 | local_irq_enable(); |
1da177e4 | 5390 | return 0; |
f7b4cddc ON |
5391 | } |
5392 | ||
1da177e4 | 5393 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 5394 | /* |
3a4fa0a2 | 5395 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 5396 | */ |
6a1bdc1b | 5397 | void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5398 | { |
1445c08d ON |
5399 | struct rq *rq = cpu_rq(dead_cpu); |
5400 | int needs_cpu, uninitialized_var(dest_cpu); | |
5401 | unsigned long flags; | |
e76bd8d9 | 5402 | |
1445c08d | 5403 | local_irq_save(flags); |
e76bd8d9 | 5404 | |
1445c08d ON |
5405 | raw_spin_lock(&rq->lock); |
5406 | needs_cpu = (task_cpu(p) == dead_cpu) && (p->state != TASK_WAKING); | |
5407 | if (needs_cpu) | |
5408 | dest_cpu = select_fallback_rq(dead_cpu, p); | |
5409 | raw_spin_unlock(&rq->lock); | |
c1804d54 ON |
5410 | /* |
5411 | * It can only fail if we race with set_cpus_allowed(), | |
5412 | * in the racer should migrate the task anyway. | |
5413 | */ | |
1445c08d | 5414 | if (needs_cpu) |
c1804d54 | 5415 | __migrate_task(p, dead_cpu, dest_cpu); |
1445c08d | 5416 | local_irq_restore(flags); |
1da177e4 LT |
5417 | } |
5418 | ||
5419 | /* | |
5420 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5421 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5422 | * for performance reasons the counter is not stricly tracking tasks to | |
5423 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5424 | * to keep the global sum constant after CPU-down: | |
5425 | */ | |
70b97a7f | 5426 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5427 | { |
6ad4c188 | 5428 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 LT |
5429 | unsigned long flags; |
5430 | ||
5431 | local_irq_save(flags); | |
5432 | double_rq_lock(rq_src, rq_dest); | |
5433 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5434 | rq_src->nr_uninterruptible = 0; | |
5435 | double_rq_unlock(rq_src, rq_dest); | |
5436 | local_irq_restore(flags); | |
5437 | } | |
5438 | ||
5439 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5440 | static void migrate_live_tasks(int src_cpu) | |
5441 | { | |
48f24c4d | 5442 | struct task_struct *p, *t; |
1da177e4 | 5443 | |
f7b4cddc | 5444 | read_lock(&tasklist_lock); |
1da177e4 | 5445 | |
48f24c4d IM |
5446 | do_each_thread(t, p) { |
5447 | if (p == current) | |
1da177e4 LT |
5448 | continue; |
5449 | ||
48f24c4d IM |
5450 | if (task_cpu(p) == src_cpu) |
5451 | move_task_off_dead_cpu(src_cpu, p); | |
5452 | } while_each_thread(t, p); | |
1da177e4 | 5453 | |
f7b4cddc | 5454 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5455 | } |
5456 | ||
dd41f596 IM |
5457 | /* |
5458 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5459 | * It does so by boosting its priority to highest possible. |
5460 | * Used by CPU offline code. | |
1da177e4 LT |
5461 | */ |
5462 | void sched_idle_next(void) | |
5463 | { | |
48f24c4d | 5464 | int this_cpu = smp_processor_id(); |
70b97a7f | 5465 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5466 | struct task_struct *p = rq->idle; |
5467 | unsigned long flags; | |
5468 | ||
5469 | /* cpu has to be offline */ | |
48f24c4d | 5470 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5471 | |
48f24c4d IM |
5472 | /* |
5473 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5474 | * and interrupts disabled on the current cpu. | |
1da177e4 | 5475 | */ |
05fa785c | 5476 | raw_spin_lock_irqsave(&rq->lock, flags); |
1da177e4 | 5477 | |
dd41f596 | 5478 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 5479 | |
94bc9a7b | 5480 | activate_task(rq, p, 0); |
1da177e4 | 5481 | |
05fa785c | 5482 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5483 | } |
5484 | ||
48f24c4d IM |
5485 | /* |
5486 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5487 | * offline. |
5488 | */ | |
5489 | void idle_task_exit(void) | |
5490 | { | |
5491 | struct mm_struct *mm = current->active_mm; | |
5492 | ||
5493 | BUG_ON(cpu_online(smp_processor_id())); | |
5494 | ||
5495 | if (mm != &init_mm) | |
5496 | switch_mm(mm, &init_mm, current); | |
5497 | mmdrop(mm); | |
5498 | } | |
5499 | ||
054b9108 | 5500 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5501 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5502 | { |
70b97a7f | 5503 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5504 | |
5505 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5506 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5507 | |
5508 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5509 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5510 | |
48f24c4d | 5511 | get_task_struct(p); |
1da177e4 LT |
5512 | |
5513 | /* | |
5514 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 5515 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
5516 | * fine. |
5517 | */ | |
05fa785c | 5518 | raw_spin_unlock_irq(&rq->lock); |
48f24c4d | 5519 | move_task_off_dead_cpu(dead_cpu, p); |
05fa785c | 5520 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 5521 | |
48f24c4d | 5522 | put_task_struct(p); |
1da177e4 LT |
5523 | } |
5524 | ||
5525 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5526 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5527 | { | |
70b97a7f | 5528 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5529 | struct task_struct *next; |
48f24c4d | 5530 | |
dd41f596 IM |
5531 | for ( ; ; ) { |
5532 | if (!rq->nr_running) | |
5533 | break; | |
b67802ea | 5534 | next = pick_next_task(rq); |
dd41f596 IM |
5535 | if (!next) |
5536 | break; | |
79c53799 | 5537 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 5538 | migrate_dead(dead_cpu, next); |
e692ab53 | 5539 | |
1da177e4 LT |
5540 | } |
5541 | } | |
dce48a84 TG |
5542 | |
5543 | /* | |
5544 | * remove the tasks which were accounted by rq from calc_load_tasks. | |
5545 | */ | |
5546 | static void calc_global_load_remove(struct rq *rq) | |
5547 | { | |
5548 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | |
a468d389 | 5549 | rq->calc_load_active = 0; |
dce48a84 | 5550 | } |
1da177e4 LT |
5551 | #endif /* CONFIG_HOTPLUG_CPU */ |
5552 | ||
e692ab53 NP |
5553 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5554 | ||
5555 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5556 | { |
5557 | .procname = "sched_domain", | |
c57baf1e | 5558 | .mode = 0555, |
e0361851 | 5559 | }, |
56992309 | 5560 | {} |
e692ab53 NP |
5561 | }; |
5562 | ||
5563 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5564 | { |
5565 | .procname = "kernel", | |
c57baf1e | 5566 | .mode = 0555, |
e0361851 AD |
5567 | .child = sd_ctl_dir, |
5568 | }, | |
56992309 | 5569 | {} |
e692ab53 NP |
5570 | }; |
5571 | ||
5572 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5573 | { | |
5574 | struct ctl_table *entry = | |
5cf9f062 | 5575 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5576 | |
e692ab53 NP |
5577 | return entry; |
5578 | } | |
5579 | ||
6382bc90 MM |
5580 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5581 | { | |
cd790076 | 5582 | struct ctl_table *entry; |
6382bc90 | 5583 | |
cd790076 MM |
5584 | /* |
5585 | * In the intermediate directories, both the child directory and | |
5586 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5587 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5588 | * static strings and all have proc handlers. |
5589 | */ | |
5590 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5591 | if (entry->child) |
5592 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5593 | if (entry->proc_handler == NULL) |
5594 | kfree(entry->procname); | |
5595 | } | |
6382bc90 MM |
5596 | |
5597 | kfree(*tablep); | |
5598 | *tablep = NULL; | |
5599 | } | |
5600 | ||
e692ab53 | 5601 | static void |
e0361851 | 5602 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5603 | const char *procname, void *data, int maxlen, |
5604 | mode_t mode, proc_handler *proc_handler) | |
5605 | { | |
e692ab53 NP |
5606 | entry->procname = procname; |
5607 | entry->data = data; | |
5608 | entry->maxlen = maxlen; | |
5609 | entry->mode = mode; | |
5610 | entry->proc_handler = proc_handler; | |
5611 | } | |
5612 | ||
5613 | static struct ctl_table * | |
5614 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5615 | { | |
a5d8c348 | 5616 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 5617 | |
ad1cdc1d MM |
5618 | if (table == NULL) |
5619 | return NULL; | |
5620 | ||
e0361851 | 5621 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5622 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5623 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5624 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5625 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5626 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5627 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5628 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5629 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5630 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5631 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5632 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5633 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5634 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5635 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5636 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5637 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5638 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5639 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5640 | &sd->cache_nice_tries, |
5641 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5642 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5643 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
5644 | set_table_entry(&table[11], "name", sd->name, |
5645 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
5646 | /* &table[12] is terminator */ | |
e692ab53 NP |
5647 | |
5648 | return table; | |
5649 | } | |
5650 | ||
9a4e7159 | 5651 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5652 | { |
5653 | struct ctl_table *entry, *table; | |
5654 | struct sched_domain *sd; | |
5655 | int domain_num = 0, i; | |
5656 | char buf[32]; | |
5657 | ||
5658 | for_each_domain(cpu, sd) | |
5659 | domain_num++; | |
5660 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5661 | if (table == NULL) |
5662 | return NULL; | |
e692ab53 NP |
5663 | |
5664 | i = 0; | |
5665 | for_each_domain(cpu, sd) { | |
5666 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5667 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5668 | entry->mode = 0555; |
e692ab53 NP |
5669 | entry->child = sd_alloc_ctl_domain_table(sd); |
5670 | entry++; | |
5671 | i++; | |
5672 | } | |
5673 | return table; | |
5674 | } | |
5675 | ||
5676 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5677 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5678 | { |
6ad4c188 | 5679 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5680 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5681 | char buf[32]; | |
5682 | ||
7378547f MM |
5683 | WARN_ON(sd_ctl_dir[0].child); |
5684 | sd_ctl_dir[0].child = entry; | |
5685 | ||
ad1cdc1d MM |
5686 | if (entry == NULL) |
5687 | return; | |
5688 | ||
6ad4c188 | 5689 | for_each_possible_cpu(i) { |
e692ab53 | 5690 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5691 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5692 | entry->mode = 0555; |
e692ab53 | 5693 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5694 | entry++; |
e692ab53 | 5695 | } |
7378547f MM |
5696 | |
5697 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5698 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5699 | } | |
6382bc90 | 5700 | |
7378547f | 5701 | /* may be called multiple times per register */ |
6382bc90 MM |
5702 | static void unregister_sched_domain_sysctl(void) |
5703 | { | |
7378547f MM |
5704 | if (sd_sysctl_header) |
5705 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5706 | sd_sysctl_header = NULL; |
7378547f MM |
5707 | if (sd_ctl_dir[0].child) |
5708 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5709 | } |
e692ab53 | 5710 | #else |
6382bc90 MM |
5711 | static void register_sched_domain_sysctl(void) |
5712 | { | |
5713 | } | |
5714 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5715 | { |
5716 | } | |
5717 | #endif | |
5718 | ||
1f11eb6a GH |
5719 | static void set_rq_online(struct rq *rq) |
5720 | { | |
5721 | if (!rq->online) { | |
5722 | const struct sched_class *class; | |
5723 | ||
c6c4927b | 5724 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5725 | rq->online = 1; |
5726 | ||
5727 | for_each_class(class) { | |
5728 | if (class->rq_online) | |
5729 | class->rq_online(rq); | |
5730 | } | |
5731 | } | |
5732 | } | |
5733 | ||
5734 | static void set_rq_offline(struct rq *rq) | |
5735 | { | |
5736 | if (rq->online) { | |
5737 | const struct sched_class *class; | |
5738 | ||
5739 | for_each_class(class) { | |
5740 | if (class->rq_offline) | |
5741 | class->rq_offline(rq); | |
5742 | } | |
5743 | ||
c6c4927b | 5744 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5745 | rq->online = 0; |
5746 | } | |
5747 | } | |
5748 | ||
1da177e4 LT |
5749 | /* |
5750 | * migration_call - callback that gets triggered when a CPU is added. | |
5751 | * Here we can start up the necessary migration thread for the new CPU. | |
5752 | */ | |
48f24c4d IM |
5753 | static int __cpuinit |
5754 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5755 | { |
48f24c4d | 5756 | int cpu = (long)hcpu; |
1da177e4 | 5757 | unsigned long flags; |
969c7921 | 5758 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5759 | |
5760 | switch (action) { | |
5be9361c | 5761 | |
1da177e4 | 5762 | case CPU_UP_PREPARE: |
8bb78442 | 5763 | case CPU_UP_PREPARE_FROZEN: |
a468d389 | 5764 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5765 | break; |
48f24c4d | 5766 | |
1da177e4 | 5767 | case CPU_ONLINE: |
8bb78442 | 5768 | case CPU_ONLINE_FROZEN: |
1f94ef59 | 5769 | /* Update our root-domain */ |
05fa785c | 5770 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5771 | if (rq->rd) { |
c6c4927b | 5772 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5773 | |
5774 | set_rq_online(rq); | |
1f94ef59 | 5775 | } |
05fa785c | 5776 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5777 | break; |
48f24c4d | 5778 | |
1da177e4 | 5779 | #ifdef CONFIG_HOTPLUG_CPU |
1da177e4 | 5780 | case CPU_DEAD: |
8bb78442 | 5781 | case CPU_DEAD_FROZEN: |
1da177e4 | 5782 | migrate_live_tasks(cpu); |
1da177e4 | 5783 | /* Idle task back to normal (off runqueue, low prio) */ |
05fa785c | 5784 | raw_spin_lock_irq(&rq->lock); |
2e1cb74a | 5785 | deactivate_task(rq, rq->idle, 0); |
dd41f596 IM |
5786 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5787 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 5788 | migrate_dead_tasks(cpu); |
05fa785c | 5789 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 LT |
5790 | migrate_nr_uninterruptible(rq); |
5791 | BUG_ON(rq->nr_running != 0); | |
dce48a84 | 5792 | calc_global_load_remove(rq); |
1da177e4 | 5793 | break; |
57d885fe | 5794 | |
08f503b0 GH |
5795 | case CPU_DYING: |
5796 | case CPU_DYING_FROZEN: | |
57d885fe | 5797 | /* Update our root-domain */ |
05fa785c | 5798 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5799 | if (rq->rd) { |
c6c4927b | 5800 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5801 | set_rq_offline(rq); |
57d885fe | 5802 | } |
05fa785c | 5803 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
57d885fe | 5804 | break; |
1da177e4 LT |
5805 | #endif |
5806 | } | |
5807 | return NOTIFY_OK; | |
5808 | } | |
5809 | ||
f38b0820 PM |
5810 | /* |
5811 | * Register at high priority so that task migration (migrate_all_tasks) | |
5812 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5813 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5814 | */ |
26c2143b | 5815 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5816 | .notifier_call = migration_call, |
5817 | .priority = 10 | |
5818 | }; | |
5819 | ||
7babe8db | 5820 | static int __init migration_init(void) |
1da177e4 LT |
5821 | { |
5822 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5823 | int err; |
48f24c4d IM |
5824 | |
5825 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5826 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5827 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5828 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5829 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5830 | |
a004cd42 | 5831 | return 0; |
1da177e4 | 5832 | } |
7babe8db | 5833 | early_initcall(migration_init); |
1da177e4 LT |
5834 | #endif |
5835 | ||
5836 | #ifdef CONFIG_SMP | |
476f3534 | 5837 | |
3e9830dc | 5838 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5839 | |
f6630114 MT |
5840 | static __read_mostly int sched_domain_debug_enabled; |
5841 | ||
5842 | static int __init sched_domain_debug_setup(char *str) | |
5843 | { | |
5844 | sched_domain_debug_enabled = 1; | |
5845 | ||
5846 | return 0; | |
5847 | } | |
5848 | early_param("sched_debug", sched_domain_debug_setup); | |
5849 | ||
7c16ec58 | 5850 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5851 | struct cpumask *groupmask) |
1da177e4 | 5852 | { |
4dcf6aff | 5853 | struct sched_group *group = sd->groups; |
434d53b0 | 5854 | char str[256]; |
1da177e4 | 5855 | |
968ea6d8 | 5856 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5857 | cpumask_clear(groupmask); |
4dcf6aff IM |
5858 | |
5859 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5860 | ||
5861 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5862 | printk("does not load-balance\n"); |
4dcf6aff | 5863 | if (sd->parent) |
3df0fc5b PZ |
5864 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5865 | " has parent"); | |
4dcf6aff | 5866 | return -1; |
41c7ce9a NP |
5867 | } |
5868 | ||
3df0fc5b | 5869 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5870 | |
758b2cdc | 5871 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5872 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5873 | "CPU%d\n", cpu); | |
4dcf6aff | 5874 | } |
758b2cdc | 5875 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5876 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5877 | " CPU%d\n", cpu); | |
4dcf6aff | 5878 | } |
1da177e4 | 5879 | |
4dcf6aff | 5880 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5881 | do { |
4dcf6aff | 5882 | if (!group) { |
3df0fc5b PZ |
5883 | printk("\n"); |
5884 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5885 | break; |
5886 | } | |
5887 | ||
18a3885f | 5888 | if (!group->cpu_power) { |
3df0fc5b PZ |
5889 | printk(KERN_CONT "\n"); |
5890 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5891 | "set\n"); | |
4dcf6aff IM |
5892 | break; |
5893 | } | |
1da177e4 | 5894 | |
758b2cdc | 5895 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5896 | printk(KERN_CONT "\n"); |
5897 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5898 | break; |
5899 | } | |
1da177e4 | 5900 | |
758b2cdc | 5901 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
5902 | printk(KERN_CONT "\n"); |
5903 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5904 | break; |
5905 | } | |
1da177e4 | 5906 | |
758b2cdc | 5907 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5908 | |
968ea6d8 | 5909 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 5910 | |
3df0fc5b | 5911 | printk(KERN_CONT " %s", str); |
18a3885f | 5912 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
3df0fc5b PZ |
5913 | printk(KERN_CONT " (cpu_power = %d)", |
5914 | group->cpu_power); | |
381512cf | 5915 | } |
1da177e4 | 5916 | |
4dcf6aff IM |
5917 | group = group->next; |
5918 | } while (group != sd->groups); | |
3df0fc5b | 5919 | printk(KERN_CONT "\n"); |
1da177e4 | 5920 | |
758b2cdc | 5921 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5922 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5923 | |
758b2cdc RR |
5924 | if (sd->parent && |
5925 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5926 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5927 | "of domain->span\n"); | |
4dcf6aff IM |
5928 | return 0; |
5929 | } | |
1da177e4 | 5930 | |
4dcf6aff IM |
5931 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5932 | { | |
d5dd3db1 | 5933 | cpumask_var_t groupmask; |
4dcf6aff | 5934 | int level = 0; |
1da177e4 | 5935 | |
f6630114 MT |
5936 | if (!sched_domain_debug_enabled) |
5937 | return; | |
5938 | ||
4dcf6aff IM |
5939 | if (!sd) { |
5940 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5941 | return; | |
5942 | } | |
1da177e4 | 5943 | |
4dcf6aff IM |
5944 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5945 | ||
d5dd3db1 | 5946 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
5947 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
5948 | return; | |
5949 | } | |
5950 | ||
4dcf6aff | 5951 | for (;;) { |
7c16ec58 | 5952 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 5953 | break; |
1da177e4 LT |
5954 | level++; |
5955 | sd = sd->parent; | |
33859f7f | 5956 | if (!sd) |
4dcf6aff IM |
5957 | break; |
5958 | } | |
d5dd3db1 | 5959 | free_cpumask_var(groupmask); |
1da177e4 | 5960 | } |
6d6bc0ad | 5961 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5962 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 5963 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5964 | |
1a20ff27 | 5965 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5966 | { |
758b2cdc | 5967 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5968 | return 1; |
5969 | ||
5970 | /* Following flags need at least 2 groups */ | |
5971 | if (sd->flags & (SD_LOAD_BALANCE | | |
5972 | SD_BALANCE_NEWIDLE | | |
5973 | SD_BALANCE_FORK | | |
89c4710e SS |
5974 | SD_BALANCE_EXEC | |
5975 | SD_SHARE_CPUPOWER | | |
5976 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5977 | if (sd->groups != sd->groups->next) |
5978 | return 0; | |
5979 | } | |
5980 | ||
5981 | /* Following flags don't use groups */ | |
c88d5910 | 5982 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5983 | return 0; |
5984 | ||
5985 | return 1; | |
5986 | } | |
5987 | ||
48f24c4d IM |
5988 | static int |
5989 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5990 | { |
5991 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5992 | ||
5993 | if (sd_degenerate(parent)) | |
5994 | return 1; | |
5995 | ||
758b2cdc | 5996 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5997 | return 0; |
5998 | ||
245af2c7 SS |
5999 | /* Flags needing groups don't count if only 1 group in parent */ |
6000 | if (parent->groups == parent->groups->next) { | |
6001 | pflags &= ~(SD_LOAD_BALANCE | | |
6002 | SD_BALANCE_NEWIDLE | | |
6003 | SD_BALANCE_FORK | | |
89c4710e SS |
6004 | SD_BALANCE_EXEC | |
6005 | SD_SHARE_CPUPOWER | | |
6006 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6007 | if (nr_node_ids == 1) |
6008 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6009 | } |
6010 | if (~cflags & pflags) | |
6011 | return 0; | |
6012 | ||
6013 | return 1; | |
6014 | } | |
6015 | ||
c6c4927b RR |
6016 | static void free_rootdomain(struct root_domain *rd) |
6017 | { | |
047106ad PZ |
6018 | synchronize_sched(); |
6019 | ||
68e74568 RR |
6020 | cpupri_cleanup(&rd->cpupri); |
6021 | ||
c6c4927b RR |
6022 | free_cpumask_var(rd->rto_mask); |
6023 | free_cpumask_var(rd->online); | |
6024 | free_cpumask_var(rd->span); | |
6025 | kfree(rd); | |
6026 | } | |
6027 | ||
57d885fe GH |
6028 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6029 | { | |
a0490fa3 | 6030 | struct root_domain *old_rd = NULL; |
57d885fe | 6031 | unsigned long flags; |
57d885fe | 6032 | |
05fa785c | 6033 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
6034 | |
6035 | if (rq->rd) { | |
a0490fa3 | 6036 | old_rd = rq->rd; |
57d885fe | 6037 | |
c6c4927b | 6038 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6039 | set_rq_offline(rq); |
57d885fe | 6040 | |
c6c4927b | 6041 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6042 | |
a0490fa3 IM |
6043 | /* |
6044 | * If we dont want to free the old_rt yet then | |
6045 | * set old_rd to NULL to skip the freeing later | |
6046 | * in this function: | |
6047 | */ | |
6048 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
6049 | old_rd = NULL; | |
57d885fe GH |
6050 | } |
6051 | ||
6052 | atomic_inc(&rd->refcount); | |
6053 | rq->rd = rd; | |
6054 | ||
c6c4927b | 6055 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 6056 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 6057 | set_rq_online(rq); |
57d885fe | 6058 | |
05fa785c | 6059 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
6060 | |
6061 | if (old_rd) | |
6062 | free_rootdomain(old_rd); | |
57d885fe GH |
6063 | } |
6064 | ||
fd5e1b5d | 6065 | static int init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe | 6066 | { |
36b7b6d4 PE |
6067 | gfp_t gfp = GFP_KERNEL; |
6068 | ||
57d885fe GH |
6069 | memset(rd, 0, sizeof(*rd)); |
6070 | ||
36b7b6d4 PE |
6071 | if (bootmem) |
6072 | gfp = GFP_NOWAIT; | |
c6c4927b | 6073 | |
36b7b6d4 | 6074 | if (!alloc_cpumask_var(&rd->span, gfp)) |
0c910d28 | 6075 | goto out; |
36b7b6d4 | 6076 | if (!alloc_cpumask_var(&rd->online, gfp)) |
c6c4927b | 6077 | goto free_span; |
36b7b6d4 | 6078 | if (!alloc_cpumask_var(&rd->rto_mask, gfp)) |
c6c4927b | 6079 | goto free_online; |
6e0534f2 | 6080 | |
0fb53029 | 6081 | if (cpupri_init(&rd->cpupri, bootmem) != 0) |
68e74568 | 6082 | goto free_rto_mask; |
c6c4927b | 6083 | return 0; |
6e0534f2 | 6084 | |
68e74568 RR |
6085 | free_rto_mask: |
6086 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6087 | free_online: |
6088 | free_cpumask_var(rd->online); | |
6089 | free_span: | |
6090 | free_cpumask_var(rd->span); | |
0c910d28 | 6091 | out: |
c6c4927b | 6092 | return -ENOMEM; |
57d885fe GH |
6093 | } |
6094 | ||
6095 | static void init_defrootdomain(void) | |
6096 | { | |
c6c4927b RR |
6097 | init_rootdomain(&def_root_domain, true); |
6098 | ||
57d885fe GH |
6099 | atomic_set(&def_root_domain.refcount, 1); |
6100 | } | |
6101 | ||
dc938520 | 6102 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6103 | { |
6104 | struct root_domain *rd; | |
6105 | ||
6106 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6107 | if (!rd) | |
6108 | return NULL; | |
6109 | ||
c6c4927b RR |
6110 | if (init_rootdomain(rd, false) != 0) { |
6111 | kfree(rd); | |
6112 | return NULL; | |
6113 | } | |
57d885fe GH |
6114 | |
6115 | return rd; | |
6116 | } | |
6117 | ||
1da177e4 | 6118 | /* |
0eab9146 | 6119 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6120 | * hold the hotplug lock. |
6121 | */ | |
0eab9146 IM |
6122 | static void |
6123 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6124 | { |
70b97a7f | 6125 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6126 | struct sched_domain *tmp; |
6127 | ||
669c55e9 PZ |
6128 | for (tmp = sd; tmp; tmp = tmp->parent) |
6129 | tmp->span_weight = cpumask_weight(sched_domain_span(tmp)); | |
6130 | ||
245af2c7 | 6131 | /* Remove the sched domains which do not contribute to scheduling. */ |
f29c9b1c | 6132 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6133 | struct sched_domain *parent = tmp->parent; |
6134 | if (!parent) | |
6135 | break; | |
f29c9b1c | 6136 | |
1a848870 | 6137 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6138 | tmp->parent = parent->parent; |
1a848870 SS |
6139 | if (parent->parent) |
6140 | parent->parent->child = tmp; | |
f29c9b1c LZ |
6141 | } else |
6142 | tmp = tmp->parent; | |
245af2c7 SS |
6143 | } |
6144 | ||
1a848870 | 6145 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6146 | sd = sd->parent; |
1a848870 SS |
6147 | if (sd) |
6148 | sd->child = NULL; | |
6149 | } | |
1da177e4 LT |
6150 | |
6151 | sched_domain_debug(sd, cpu); | |
6152 | ||
57d885fe | 6153 | rq_attach_root(rq, rd); |
674311d5 | 6154 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6155 | } |
6156 | ||
6157 | /* cpus with isolated domains */ | |
dcc30a35 | 6158 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
6159 | |
6160 | /* Setup the mask of cpus configured for isolated domains */ | |
6161 | static int __init isolated_cpu_setup(char *str) | |
6162 | { | |
bdddd296 | 6163 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 6164 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
6165 | return 1; |
6166 | } | |
6167 | ||
8927f494 | 6168 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6169 | |
6170 | /* | |
6711cab4 SS |
6171 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6172 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
6173 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
6174 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
6175 | * |
6176 | * init_sched_build_groups will build a circular linked list of the groups | |
6177 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6178 | * and ->cpu_power to 0. | |
6179 | */ | |
a616058b | 6180 | static void |
96f874e2 RR |
6181 | init_sched_build_groups(const struct cpumask *span, |
6182 | const struct cpumask *cpu_map, | |
6183 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 6184 | struct sched_group **sg, |
96f874e2 RR |
6185 | struct cpumask *tmpmask), |
6186 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
6187 | { |
6188 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6189 | int i; |
6190 | ||
96f874e2 | 6191 | cpumask_clear(covered); |
7c16ec58 | 6192 | |
abcd083a | 6193 | for_each_cpu(i, span) { |
6711cab4 | 6194 | struct sched_group *sg; |
7c16ec58 | 6195 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6196 | int j; |
6197 | ||
758b2cdc | 6198 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
6199 | continue; |
6200 | ||
758b2cdc | 6201 | cpumask_clear(sched_group_cpus(sg)); |
18a3885f | 6202 | sg->cpu_power = 0; |
1da177e4 | 6203 | |
abcd083a | 6204 | for_each_cpu(j, span) { |
7c16ec58 | 6205 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6206 | continue; |
6207 | ||
96f874e2 | 6208 | cpumask_set_cpu(j, covered); |
758b2cdc | 6209 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
6210 | } |
6211 | if (!first) | |
6212 | first = sg; | |
6213 | if (last) | |
6214 | last->next = sg; | |
6215 | last = sg; | |
6216 | } | |
6217 | last->next = first; | |
6218 | } | |
6219 | ||
9c1cfda2 | 6220 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6221 | |
9c1cfda2 | 6222 | #ifdef CONFIG_NUMA |
198e2f18 | 6223 | |
9c1cfda2 JH |
6224 | /** |
6225 | * find_next_best_node - find the next node to include in a sched_domain | |
6226 | * @node: node whose sched_domain we're building | |
6227 | * @used_nodes: nodes already in the sched_domain | |
6228 | * | |
41a2d6cf | 6229 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6230 | * finds the closest node not already in the @used_nodes map. |
6231 | * | |
6232 | * Should use nodemask_t. | |
6233 | */ | |
c5f59f08 | 6234 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6235 | { |
6236 | int i, n, val, min_val, best_node = 0; | |
6237 | ||
6238 | min_val = INT_MAX; | |
6239 | ||
076ac2af | 6240 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6241 | /* Start at @node */ |
076ac2af | 6242 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6243 | |
6244 | if (!nr_cpus_node(n)) | |
6245 | continue; | |
6246 | ||
6247 | /* Skip already used nodes */ | |
c5f59f08 | 6248 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6249 | continue; |
6250 | ||
6251 | /* Simple min distance search */ | |
6252 | val = node_distance(node, n); | |
6253 | ||
6254 | if (val < min_val) { | |
6255 | min_val = val; | |
6256 | best_node = n; | |
6257 | } | |
6258 | } | |
6259 | ||
c5f59f08 | 6260 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6261 | return best_node; |
6262 | } | |
6263 | ||
6264 | /** | |
6265 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6266 | * @node: node whose cpumask we're constructing | |
73486722 | 6267 | * @span: resulting cpumask |
9c1cfda2 | 6268 | * |
41a2d6cf | 6269 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6270 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6271 | * out optimally. | |
6272 | */ | |
96f874e2 | 6273 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 6274 | { |
c5f59f08 | 6275 | nodemask_t used_nodes; |
48f24c4d | 6276 | int i; |
9c1cfda2 | 6277 | |
6ca09dfc | 6278 | cpumask_clear(span); |
c5f59f08 | 6279 | nodes_clear(used_nodes); |
9c1cfda2 | 6280 | |
6ca09dfc | 6281 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 6282 | node_set(node, used_nodes); |
9c1cfda2 JH |
6283 | |
6284 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6285 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6286 | |
6ca09dfc | 6287 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 6288 | } |
9c1cfda2 | 6289 | } |
6d6bc0ad | 6290 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6291 | |
5c45bf27 | 6292 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6293 | |
6c99e9ad RR |
6294 | /* |
6295 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
6296 | * |
6297 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
6298 | * and struct sched_domain. ) | |
6c99e9ad RR |
6299 | */ |
6300 | struct static_sched_group { | |
6301 | struct sched_group sg; | |
6302 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
6303 | }; | |
6304 | ||
6305 | struct static_sched_domain { | |
6306 | struct sched_domain sd; | |
6307 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
6308 | }; | |
6309 | ||
49a02c51 AH |
6310 | struct s_data { |
6311 | #ifdef CONFIG_NUMA | |
6312 | int sd_allnodes; | |
6313 | cpumask_var_t domainspan; | |
6314 | cpumask_var_t covered; | |
6315 | cpumask_var_t notcovered; | |
6316 | #endif | |
6317 | cpumask_var_t nodemask; | |
6318 | cpumask_var_t this_sibling_map; | |
6319 | cpumask_var_t this_core_map; | |
6320 | cpumask_var_t send_covered; | |
6321 | cpumask_var_t tmpmask; | |
6322 | struct sched_group **sched_group_nodes; | |
6323 | struct root_domain *rd; | |
6324 | }; | |
6325 | ||
2109b99e AH |
6326 | enum s_alloc { |
6327 | sa_sched_groups = 0, | |
6328 | sa_rootdomain, | |
6329 | sa_tmpmask, | |
6330 | sa_send_covered, | |
6331 | sa_this_core_map, | |
6332 | sa_this_sibling_map, | |
6333 | sa_nodemask, | |
6334 | sa_sched_group_nodes, | |
6335 | #ifdef CONFIG_NUMA | |
6336 | sa_notcovered, | |
6337 | sa_covered, | |
6338 | sa_domainspan, | |
6339 | #endif | |
6340 | sa_none, | |
6341 | }; | |
6342 | ||
9c1cfda2 | 6343 | /* |
48f24c4d | 6344 | * SMT sched-domains: |
9c1cfda2 | 6345 | */ |
1da177e4 | 6346 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 6347 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
1871e52c | 6348 | static DEFINE_PER_CPU(struct static_sched_group, sched_groups); |
48f24c4d | 6349 | |
41a2d6cf | 6350 | static int |
96f874e2 RR |
6351 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
6352 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 6353 | { |
6711cab4 | 6354 | if (sg) |
1871e52c | 6355 | *sg = &per_cpu(sched_groups, cpu).sg; |
1da177e4 LT |
6356 | return cpu; |
6357 | } | |
6d6bc0ad | 6358 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6359 | |
48f24c4d IM |
6360 | /* |
6361 | * multi-core sched-domains: | |
6362 | */ | |
1e9f28fa | 6363 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
6364 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
6365 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 6366 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
6367 | |
6368 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 6369 | static int |
96f874e2 RR |
6370 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6371 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6372 | { |
6711cab4 | 6373 | int group; |
7c16ec58 | 6374 | |
c69fc56d | 6375 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6376 | group = cpumask_first(mask); |
6711cab4 | 6377 | if (sg) |
6c99e9ad | 6378 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 6379 | return group; |
1e9f28fa SS |
6380 | } |
6381 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 6382 | static int |
96f874e2 RR |
6383 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6384 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 6385 | { |
6711cab4 | 6386 | if (sg) |
6c99e9ad | 6387 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
6388 | return cpu; |
6389 | } | |
6390 | #endif | |
6391 | ||
6c99e9ad RR |
6392 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
6393 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 6394 | |
41a2d6cf | 6395 | static int |
96f874e2 RR |
6396 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
6397 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 6398 | { |
6711cab4 | 6399 | int group; |
48f24c4d | 6400 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 6401 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 6402 | group = cpumask_first(mask); |
1e9f28fa | 6403 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 6404 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6405 | group = cpumask_first(mask); |
1da177e4 | 6406 | #else |
6711cab4 | 6407 | group = cpu; |
1da177e4 | 6408 | #endif |
6711cab4 | 6409 | if (sg) |
6c99e9ad | 6410 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 6411 | return group; |
1da177e4 LT |
6412 | } |
6413 | ||
6414 | #ifdef CONFIG_NUMA | |
1da177e4 | 6415 | /* |
9c1cfda2 JH |
6416 | * The init_sched_build_groups can't handle what we want to do with node |
6417 | * groups, so roll our own. Now each node has its own list of groups which | |
6418 | * gets dynamically allocated. | |
1da177e4 | 6419 | */ |
62ea9ceb | 6420 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 6421 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6422 | |
62ea9ceb | 6423 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 6424 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 6425 | |
96f874e2 RR |
6426 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
6427 | struct sched_group **sg, | |
6428 | struct cpumask *nodemask) | |
9c1cfda2 | 6429 | { |
6711cab4 SS |
6430 | int group; |
6431 | ||
6ca09dfc | 6432 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 6433 | group = cpumask_first(nodemask); |
6711cab4 SS |
6434 | |
6435 | if (sg) | |
6c99e9ad | 6436 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 6437 | return group; |
1da177e4 | 6438 | } |
6711cab4 | 6439 | |
08069033 SS |
6440 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6441 | { | |
6442 | struct sched_group *sg = group_head; | |
6443 | int j; | |
6444 | ||
6445 | if (!sg) | |
6446 | return; | |
3a5c359a | 6447 | do { |
758b2cdc | 6448 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 6449 | struct sched_domain *sd; |
08069033 | 6450 | |
6c99e9ad | 6451 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 6452 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
6453 | /* |
6454 | * Only add "power" once for each | |
6455 | * physical package. | |
6456 | */ | |
6457 | continue; | |
6458 | } | |
08069033 | 6459 | |
18a3885f | 6460 | sg->cpu_power += sd->groups->cpu_power; |
3a5c359a AK |
6461 | } |
6462 | sg = sg->next; | |
6463 | } while (sg != group_head); | |
08069033 | 6464 | } |
0601a88d AH |
6465 | |
6466 | static int build_numa_sched_groups(struct s_data *d, | |
6467 | const struct cpumask *cpu_map, int num) | |
6468 | { | |
6469 | struct sched_domain *sd; | |
6470 | struct sched_group *sg, *prev; | |
6471 | int n, j; | |
6472 | ||
6473 | cpumask_clear(d->covered); | |
6474 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
6475 | if (cpumask_empty(d->nodemask)) { | |
6476 | d->sched_group_nodes[num] = NULL; | |
6477 | goto out; | |
6478 | } | |
6479 | ||
6480 | sched_domain_node_span(num, d->domainspan); | |
6481 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
6482 | ||
6483 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6484 | GFP_KERNEL, num); | |
6485 | if (!sg) { | |
3df0fc5b PZ |
6486 | printk(KERN_WARNING "Can not alloc domain group for node %d\n", |
6487 | num); | |
0601a88d AH |
6488 | return -ENOMEM; |
6489 | } | |
6490 | d->sched_group_nodes[num] = sg; | |
6491 | ||
6492 | for_each_cpu(j, d->nodemask) { | |
6493 | sd = &per_cpu(node_domains, j).sd; | |
6494 | sd->groups = sg; | |
6495 | } | |
6496 | ||
18a3885f | 6497 | sg->cpu_power = 0; |
0601a88d AH |
6498 | cpumask_copy(sched_group_cpus(sg), d->nodemask); |
6499 | sg->next = sg; | |
6500 | cpumask_or(d->covered, d->covered, d->nodemask); | |
6501 | ||
6502 | prev = sg; | |
6503 | for (j = 0; j < nr_node_ids; j++) { | |
6504 | n = (num + j) % nr_node_ids; | |
6505 | cpumask_complement(d->notcovered, d->covered); | |
6506 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
6507 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
6508 | if (cpumask_empty(d->tmpmask)) | |
6509 | break; | |
6510 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
6511 | if (cpumask_empty(d->tmpmask)) | |
6512 | continue; | |
6513 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6514 | GFP_KERNEL, num); | |
6515 | if (!sg) { | |
3df0fc5b PZ |
6516 | printk(KERN_WARNING |
6517 | "Can not alloc domain group for node %d\n", j); | |
0601a88d AH |
6518 | return -ENOMEM; |
6519 | } | |
18a3885f | 6520 | sg->cpu_power = 0; |
0601a88d AH |
6521 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); |
6522 | sg->next = prev->next; | |
6523 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
6524 | prev->next = sg; | |
6525 | prev = sg; | |
6526 | } | |
6527 | out: | |
6528 | return 0; | |
6529 | } | |
6d6bc0ad | 6530 | #endif /* CONFIG_NUMA */ |
1da177e4 | 6531 | |
a616058b | 6532 | #ifdef CONFIG_NUMA |
51888ca2 | 6533 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
6534 | static void free_sched_groups(const struct cpumask *cpu_map, |
6535 | struct cpumask *nodemask) | |
51888ca2 | 6536 | { |
a616058b | 6537 | int cpu, i; |
51888ca2 | 6538 | |
abcd083a | 6539 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
6540 | struct sched_group **sched_group_nodes |
6541 | = sched_group_nodes_bycpu[cpu]; | |
6542 | ||
51888ca2 SV |
6543 | if (!sched_group_nodes) |
6544 | continue; | |
6545 | ||
076ac2af | 6546 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
6547 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
6548 | ||
6ca09dfc | 6549 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 6550 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
6551 | continue; |
6552 | ||
6553 | if (sg == NULL) | |
6554 | continue; | |
6555 | sg = sg->next; | |
6556 | next_sg: | |
6557 | oldsg = sg; | |
6558 | sg = sg->next; | |
6559 | kfree(oldsg); | |
6560 | if (oldsg != sched_group_nodes[i]) | |
6561 | goto next_sg; | |
6562 | } | |
6563 | kfree(sched_group_nodes); | |
6564 | sched_group_nodes_bycpu[cpu] = NULL; | |
6565 | } | |
51888ca2 | 6566 | } |
6d6bc0ad | 6567 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
6568 | static void free_sched_groups(const struct cpumask *cpu_map, |
6569 | struct cpumask *nodemask) | |
a616058b SS |
6570 | { |
6571 | } | |
6d6bc0ad | 6572 | #endif /* CONFIG_NUMA */ |
51888ca2 | 6573 | |
89c4710e SS |
6574 | /* |
6575 | * Initialize sched groups cpu_power. | |
6576 | * | |
6577 | * cpu_power indicates the capacity of sched group, which is used while | |
6578 | * distributing the load between different sched groups in a sched domain. | |
6579 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6580 | * there are asymmetries in the topology. If there are asymmetries, group | |
6581 | * having more cpu_power will pickup more load compared to the group having | |
6582 | * less cpu_power. | |
89c4710e SS |
6583 | */ |
6584 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6585 | { | |
6586 | struct sched_domain *child; | |
6587 | struct sched_group *group; | |
f93e65c1 PZ |
6588 | long power; |
6589 | int weight; | |
89c4710e SS |
6590 | |
6591 | WARN_ON(!sd || !sd->groups); | |
6592 | ||
13318a71 | 6593 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
6594 | return; |
6595 | ||
6596 | child = sd->child; | |
6597 | ||
18a3885f | 6598 | sd->groups->cpu_power = 0; |
5517d86b | 6599 | |
f93e65c1 PZ |
6600 | if (!child) { |
6601 | power = SCHED_LOAD_SCALE; | |
6602 | weight = cpumask_weight(sched_domain_span(sd)); | |
6603 | /* | |
6604 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
6605 | * Usually multiple threads get a better yield out of |
6606 | * that one core than a single thread would have, | |
6607 | * reflect that in sd->smt_gain. | |
f93e65c1 | 6608 | */ |
a52bfd73 PZ |
6609 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
6610 | power *= sd->smt_gain; | |
f93e65c1 | 6611 | power /= weight; |
a52bfd73 PZ |
6612 | power >>= SCHED_LOAD_SHIFT; |
6613 | } | |
18a3885f | 6614 | sd->groups->cpu_power += power; |
89c4710e SS |
6615 | return; |
6616 | } | |
6617 | ||
89c4710e | 6618 | /* |
f93e65c1 | 6619 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
6620 | */ |
6621 | group = child->groups; | |
6622 | do { | |
18a3885f | 6623 | sd->groups->cpu_power += group->cpu_power; |
89c4710e SS |
6624 | group = group->next; |
6625 | } while (group != child->groups); | |
6626 | } | |
6627 | ||
7c16ec58 MT |
6628 | /* |
6629 | * Initializers for schedule domains | |
6630 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6631 | */ | |
6632 | ||
a5d8c348 IM |
6633 | #ifdef CONFIG_SCHED_DEBUG |
6634 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
6635 | #else | |
6636 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
6637 | #endif | |
6638 | ||
7c16ec58 | 6639 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 6640 | |
7c16ec58 MT |
6641 | #define SD_INIT_FUNC(type) \ |
6642 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
6643 | { \ | |
6644 | memset(sd, 0, sizeof(*sd)); \ | |
6645 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 6646 | sd->level = SD_LV_##type; \ |
a5d8c348 | 6647 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
6648 | } |
6649 | ||
6650 | SD_INIT_FUNC(CPU) | |
6651 | #ifdef CONFIG_NUMA | |
6652 | SD_INIT_FUNC(ALLNODES) | |
6653 | SD_INIT_FUNC(NODE) | |
6654 | #endif | |
6655 | #ifdef CONFIG_SCHED_SMT | |
6656 | SD_INIT_FUNC(SIBLING) | |
6657 | #endif | |
6658 | #ifdef CONFIG_SCHED_MC | |
6659 | SD_INIT_FUNC(MC) | |
6660 | #endif | |
6661 | ||
1d3504fc HS |
6662 | static int default_relax_domain_level = -1; |
6663 | ||
6664 | static int __init setup_relax_domain_level(char *str) | |
6665 | { | |
30e0e178 LZ |
6666 | unsigned long val; |
6667 | ||
6668 | val = simple_strtoul(str, NULL, 0); | |
6669 | if (val < SD_LV_MAX) | |
6670 | default_relax_domain_level = val; | |
6671 | ||
1d3504fc HS |
6672 | return 1; |
6673 | } | |
6674 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6675 | ||
6676 | static void set_domain_attribute(struct sched_domain *sd, | |
6677 | struct sched_domain_attr *attr) | |
6678 | { | |
6679 | int request; | |
6680 | ||
6681 | if (!attr || attr->relax_domain_level < 0) { | |
6682 | if (default_relax_domain_level < 0) | |
6683 | return; | |
6684 | else | |
6685 | request = default_relax_domain_level; | |
6686 | } else | |
6687 | request = attr->relax_domain_level; | |
6688 | if (request < sd->level) { | |
6689 | /* turn off idle balance on this domain */ | |
c88d5910 | 6690 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6691 | } else { |
6692 | /* turn on idle balance on this domain */ | |
c88d5910 | 6693 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6694 | } |
6695 | } | |
6696 | ||
2109b99e AH |
6697 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6698 | const struct cpumask *cpu_map) | |
6699 | { | |
6700 | switch (what) { | |
6701 | case sa_sched_groups: | |
6702 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
6703 | d->sched_group_nodes = NULL; | |
6704 | case sa_rootdomain: | |
6705 | free_rootdomain(d->rd); /* fall through */ | |
6706 | case sa_tmpmask: | |
6707 | free_cpumask_var(d->tmpmask); /* fall through */ | |
6708 | case sa_send_covered: | |
6709 | free_cpumask_var(d->send_covered); /* fall through */ | |
6710 | case sa_this_core_map: | |
6711 | free_cpumask_var(d->this_core_map); /* fall through */ | |
6712 | case sa_this_sibling_map: | |
6713 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
6714 | case sa_nodemask: | |
6715 | free_cpumask_var(d->nodemask); /* fall through */ | |
6716 | case sa_sched_group_nodes: | |
d1b55138 | 6717 | #ifdef CONFIG_NUMA |
2109b99e AH |
6718 | kfree(d->sched_group_nodes); /* fall through */ |
6719 | case sa_notcovered: | |
6720 | free_cpumask_var(d->notcovered); /* fall through */ | |
6721 | case sa_covered: | |
6722 | free_cpumask_var(d->covered); /* fall through */ | |
6723 | case sa_domainspan: | |
6724 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 6725 | #endif |
2109b99e AH |
6726 | case sa_none: |
6727 | break; | |
6728 | } | |
6729 | } | |
3404c8d9 | 6730 | |
2109b99e AH |
6731 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6732 | const struct cpumask *cpu_map) | |
6733 | { | |
3404c8d9 | 6734 | #ifdef CONFIG_NUMA |
2109b99e AH |
6735 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
6736 | return sa_none; | |
6737 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
6738 | return sa_domainspan; | |
6739 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
6740 | return sa_covered; | |
6741 | /* Allocate the per-node list of sched groups */ | |
6742 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
6743 | sizeof(struct sched_group *), GFP_KERNEL); | |
6744 | if (!d->sched_group_nodes) { | |
3df0fc5b | 6745 | printk(KERN_WARNING "Can not alloc sched group node list\n"); |
2109b99e | 6746 | return sa_notcovered; |
d1b55138 | 6747 | } |
2109b99e | 6748 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 6749 | #endif |
2109b99e AH |
6750 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
6751 | return sa_sched_group_nodes; | |
6752 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
6753 | return sa_nodemask; | |
6754 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
6755 | return sa_this_sibling_map; | |
6756 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) | |
6757 | return sa_this_core_map; | |
6758 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) | |
6759 | return sa_send_covered; | |
6760 | d->rd = alloc_rootdomain(); | |
6761 | if (!d->rd) { | |
3df0fc5b | 6762 | printk(KERN_WARNING "Cannot alloc root domain\n"); |
2109b99e | 6763 | return sa_tmpmask; |
57d885fe | 6764 | } |
2109b99e AH |
6765 | return sa_rootdomain; |
6766 | } | |
57d885fe | 6767 | |
7f4588f3 AH |
6768 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
6769 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
6770 | { | |
6771 | struct sched_domain *sd = NULL; | |
7c16ec58 | 6772 | #ifdef CONFIG_NUMA |
7f4588f3 | 6773 | struct sched_domain *parent; |
1da177e4 | 6774 | |
7f4588f3 AH |
6775 | d->sd_allnodes = 0; |
6776 | if (cpumask_weight(cpu_map) > | |
6777 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
6778 | sd = &per_cpu(allnodes_domains, i).sd; | |
6779 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 6780 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
6781 | cpumask_copy(sched_domain_span(sd), cpu_map); |
6782 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
6783 | d->sd_allnodes = 1; | |
6784 | } | |
6785 | parent = sd; | |
6786 | ||
6787 | sd = &per_cpu(node_domains, i).sd; | |
6788 | SD_INIT(sd, NODE); | |
6789 | set_domain_attribute(sd, attr); | |
6790 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
6791 | sd->parent = parent; | |
6792 | if (parent) | |
6793 | parent->child = sd; | |
6794 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 6795 | #endif |
7f4588f3 AH |
6796 | return sd; |
6797 | } | |
1da177e4 | 6798 | |
87cce662 AH |
6799 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
6800 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6801 | struct sched_domain *parent, int i) | |
6802 | { | |
6803 | struct sched_domain *sd; | |
6804 | sd = &per_cpu(phys_domains, i).sd; | |
6805 | SD_INIT(sd, CPU); | |
6806 | set_domain_attribute(sd, attr); | |
6807 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
6808 | sd->parent = parent; | |
6809 | if (parent) | |
6810 | parent->child = sd; | |
6811 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
6812 | return sd; | |
6813 | } | |
1da177e4 | 6814 | |
410c4081 AH |
6815 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
6816 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6817 | struct sched_domain *parent, int i) | |
6818 | { | |
6819 | struct sched_domain *sd = parent; | |
1e9f28fa | 6820 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
6821 | sd = &per_cpu(core_domains, i).sd; |
6822 | SD_INIT(sd, MC); | |
6823 | set_domain_attribute(sd, attr); | |
6824 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
6825 | sd->parent = parent; | |
6826 | parent->child = sd; | |
6827 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 6828 | #endif |
410c4081 AH |
6829 | return sd; |
6830 | } | |
1e9f28fa | 6831 | |
d8173535 AH |
6832 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
6833 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6834 | struct sched_domain *parent, int i) | |
6835 | { | |
6836 | struct sched_domain *sd = parent; | |
1da177e4 | 6837 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
6838 | sd = &per_cpu(cpu_domains, i).sd; |
6839 | SD_INIT(sd, SIBLING); | |
6840 | set_domain_attribute(sd, attr); | |
6841 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
6842 | sd->parent = parent; | |
6843 | parent->child = sd; | |
6844 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 6845 | #endif |
d8173535 AH |
6846 | return sd; |
6847 | } | |
1da177e4 | 6848 | |
0e8e85c9 AH |
6849 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
6850 | const struct cpumask *cpu_map, int cpu) | |
6851 | { | |
6852 | switch (l) { | |
1da177e4 | 6853 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
6854 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
6855 | cpumask_and(d->this_sibling_map, cpu_map, | |
6856 | topology_thread_cpumask(cpu)); | |
6857 | if (cpu == cpumask_first(d->this_sibling_map)) | |
6858 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
6859 | &cpu_to_cpu_group, | |
6860 | d->send_covered, d->tmpmask); | |
6861 | break; | |
1da177e4 | 6862 | #endif |
1e9f28fa | 6863 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
6864 | case SD_LV_MC: /* set up multi-core groups */ |
6865 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
6866 | if (cpu == cpumask_first(d->this_core_map)) | |
6867 | init_sched_build_groups(d->this_core_map, cpu_map, | |
6868 | &cpu_to_core_group, | |
6869 | d->send_covered, d->tmpmask); | |
6870 | break; | |
1e9f28fa | 6871 | #endif |
86548096 AH |
6872 | case SD_LV_CPU: /* set up physical groups */ |
6873 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
6874 | if (!cpumask_empty(d->nodemask)) | |
6875 | init_sched_build_groups(d->nodemask, cpu_map, | |
6876 | &cpu_to_phys_group, | |
6877 | d->send_covered, d->tmpmask); | |
6878 | break; | |
1da177e4 | 6879 | #ifdef CONFIG_NUMA |
de616e36 AH |
6880 | case SD_LV_ALLNODES: |
6881 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
6882 | d->send_covered, d->tmpmask); | |
6883 | break; | |
6884 | #endif | |
0e8e85c9 AH |
6885 | default: |
6886 | break; | |
7c16ec58 | 6887 | } |
0e8e85c9 | 6888 | } |
9c1cfda2 | 6889 | |
2109b99e AH |
6890 | /* |
6891 | * Build sched domains for a given set of cpus and attach the sched domains | |
6892 | * to the individual cpus | |
6893 | */ | |
6894 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
6895 | struct sched_domain_attr *attr) | |
6896 | { | |
6897 | enum s_alloc alloc_state = sa_none; | |
6898 | struct s_data d; | |
294b0c96 | 6899 | struct sched_domain *sd; |
2109b99e | 6900 | int i; |
7c16ec58 | 6901 | #ifdef CONFIG_NUMA |
2109b99e | 6902 | d.sd_allnodes = 0; |
7c16ec58 | 6903 | #endif |
9c1cfda2 | 6904 | |
2109b99e AH |
6905 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6906 | if (alloc_state != sa_rootdomain) | |
6907 | goto error; | |
6908 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 6909 | |
1da177e4 | 6910 | /* |
1a20ff27 | 6911 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 6912 | */ |
abcd083a | 6913 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
6914 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
6915 | cpu_map); | |
9761eea8 | 6916 | |
7f4588f3 | 6917 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 6918 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 6919 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 6920 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 6921 | } |
9c1cfda2 | 6922 | |
abcd083a | 6923 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 6924 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
a2af04cd | 6925 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 6926 | } |
9c1cfda2 | 6927 | |
1da177e4 | 6928 | /* Set up physical groups */ |
86548096 AH |
6929 | for (i = 0; i < nr_node_ids; i++) |
6930 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 6931 | |
1da177e4 LT |
6932 | #ifdef CONFIG_NUMA |
6933 | /* Set up node groups */ | |
de616e36 AH |
6934 | if (d.sd_allnodes) |
6935 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 6936 | |
0601a88d AH |
6937 | for (i = 0; i < nr_node_ids; i++) |
6938 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 6939 | goto error; |
1da177e4 LT |
6940 | #endif |
6941 | ||
6942 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 6943 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 6944 | for_each_cpu(i, cpu_map) { |
294b0c96 | 6945 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 6946 | init_sched_groups_power(i, sd); |
5c45bf27 | 6947 | } |
1da177e4 | 6948 | #endif |
1e9f28fa | 6949 | #ifdef CONFIG_SCHED_MC |
abcd083a | 6950 | for_each_cpu(i, cpu_map) { |
294b0c96 | 6951 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 6952 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
6953 | } |
6954 | #endif | |
1e9f28fa | 6955 | |
abcd083a | 6956 | for_each_cpu(i, cpu_map) { |
294b0c96 | 6957 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 6958 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6959 | } |
6960 | ||
9c1cfda2 | 6961 | #ifdef CONFIG_NUMA |
076ac2af | 6962 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 6963 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 6964 | |
49a02c51 | 6965 | if (d.sd_allnodes) { |
6711cab4 | 6966 | struct sched_group *sg; |
f712c0c7 | 6967 | |
96f874e2 | 6968 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 6969 | d.tmpmask); |
f712c0c7 SS |
6970 | init_numa_sched_groups_power(sg); |
6971 | } | |
9c1cfda2 JH |
6972 | #endif |
6973 | ||
1da177e4 | 6974 | /* Attach the domains */ |
abcd083a | 6975 | for_each_cpu(i, cpu_map) { |
1da177e4 | 6976 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 6977 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 6978 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 6979 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 6980 | #else |
6c99e9ad | 6981 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 6982 | #endif |
49a02c51 | 6983 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6984 | } |
51888ca2 | 6985 | |
2109b99e AH |
6986 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
6987 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
6988 | return 0; | |
51888ca2 | 6989 | |
51888ca2 | 6990 | error: |
2109b99e AH |
6991 | __free_domain_allocs(&d, alloc_state, cpu_map); |
6992 | return -ENOMEM; | |
1da177e4 | 6993 | } |
029190c5 | 6994 | |
96f874e2 | 6995 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
6996 | { |
6997 | return __build_sched_domains(cpu_map, NULL); | |
6998 | } | |
6999 | ||
acc3f5d7 | 7000 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7001 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7002 | static struct sched_domain_attr *dattr_cur; |
7003 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7004 | |
7005 | /* | |
7006 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7007 | * cpumask) fails, then fallback to a single sched domain, |
7008 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7009 | */ |
4212823f | 7010 | static cpumask_var_t fallback_doms; |
029190c5 | 7011 | |
ee79d1bd HC |
7012 | /* |
7013 | * arch_update_cpu_topology lets virtualized architectures update the | |
7014 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7015 | * or 0 if it stayed the same. | |
7016 | */ | |
7017 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7018 | { |
ee79d1bd | 7019 | return 0; |
22e52b07 HC |
7020 | } |
7021 | ||
acc3f5d7 RR |
7022 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7023 | { | |
7024 | int i; | |
7025 | cpumask_var_t *doms; | |
7026 | ||
7027 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7028 | if (!doms) | |
7029 | return NULL; | |
7030 | for (i = 0; i < ndoms; i++) { | |
7031 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7032 | free_sched_domains(doms, i); | |
7033 | return NULL; | |
7034 | } | |
7035 | } | |
7036 | return doms; | |
7037 | } | |
7038 | ||
7039 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7040 | { | |
7041 | unsigned int i; | |
7042 | for (i = 0; i < ndoms; i++) | |
7043 | free_cpumask_var(doms[i]); | |
7044 | kfree(doms); | |
7045 | } | |
7046 | ||
1a20ff27 | 7047 | /* |
41a2d6cf | 7048 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7049 | * For now this just excludes isolated cpus, but could be used to |
7050 | * exclude other special cases in the future. | |
1a20ff27 | 7051 | */ |
96f874e2 | 7052 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7053 | { |
7378547f MM |
7054 | int err; |
7055 | ||
22e52b07 | 7056 | arch_update_cpu_topology(); |
029190c5 | 7057 | ndoms_cur = 1; |
acc3f5d7 | 7058 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7059 | if (!doms_cur) |
acc3f5d7 RR |
7060 | doms_cur = &fallback_doms; |
7061 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 7062 | dattr_cur = NULL; |
acc3f5d7 | 7063 | err = build_sched_domains(doms_cur[0]); |
6382bc90 | 7064 | register_sched_domain_sysctl(); |
7378547f MM |
7065 | |
7066 | return err; | |
1a20ff27 DG |
7067 | } |
7068 | ||
96f874e2 RR |
7069 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
7070 | struct cpumask *tmpmask) | |
1da177e4 | 7071 | { |
7c16ec58 | 7072 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7073 | } |
1da177e4 | 7074 | |
1a20ff27 DG |
7075 | /* |
7076 | * Detach sched domains from a group of cpus specified in cpu_map | |
7077 | * These cpus will now be attached to the NULL domain | |
7078 | */ | |
96f874e2 | 7079 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7080 | { |
96f874e2 RR |
7081 | /* Save because hotplug lock held. */ |
7082 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
7083 | int i; |
7084 | ||
abcd083a | 7085 | for_each_cpu(i, cpu_map) |
57d885fe | 7086 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7087 | synchronize_sched(); |
96f874e2 | 7088 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
7089 | } |
7090 | ||
1d3504fc HS |
7091 | /* handle null as "default" */ |
7092 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7093 | struct sched_domain_attr *new, int idx_new) | |
7094 | { | |
7095 | struct sched_domain_attr tmp; | |
7096 | ||
7097 | /* fast path */ | |
7098 | if (!new && !cur) | |
7099 | return 1; | |
7100 | ||
7101 | tmp = SD_ATTR_INIT; | |
7102 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7103 | new ? (new + idx_new) : &tmp, | |
7104 | sizeof(struct sched_domain_attr)); | |
7105 | } | |
7106 | ||
029190c5 PJ |
7107 | /* |
7108 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7109 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7110 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7111 | * It destroys each deleted domain and builds each new domain. | |
7112 | * | |
acc3f5d7 | 7113 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7114 | * The masks don't intersect (don't overlap.) We should setup one |
7115 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7116 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7117 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7118 | * it as it is. | |
7119 | * | |
acc3f5d7 RR |
7120 | * The passed in 'doms_new' should be allocated using |
7121 | * alloc_sched_domains. This routine takes ownership of it and will | |
7122 | * free_sched_domains it when done with it. If the caller failed the | |
7123 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7124 | * and partition_sched_domains() will fallback to the single partition | |
7125 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7126 | * |
96f874e2 | 7127 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7128 | * ndoms_new == 0 is a special case for destroying existing domains, |
7129 | * and it will not create the default domain. | |
dfb512ec | 7130 | * |
029190c5 PJ |
7131 | * Call with hotplug lock held |
7132 | */ | |
acc3f5d7 | 7133 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7134 | struct sched_domain_attr *dattr_new) |
029190c5 | 7135 | { |
dfb512ec | 7136 | int i, j, n; |
d65bd5ec | 7137 | int new_topology; |
029190c5 | 7138 | |
712555ee | 7139 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7140 | |
7378547f MM |
7141 | /* always unregister in case we don't destroy any domains */ |
7142 | unregister_sched_domain_sysctl(); | |
7143 | ||
d65bd5ec HC |
7144 | /* Let architecture update cpu core mappings. */ |
7145 | new_topology = arch_update_cpu_topology(); | |
7146 | ||
dfb512ec | 7147 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7148 | |
7149 | /* Destroy deleted domains */ | |
7150 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7151 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7152 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7153 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7154 | goto match1; |
7155 | } | |
7156 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7157 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7158 | match1: |
7159 | ; | |
7160 | } | |
7161 | ||
e761b772 MK |
7162 | if (doms_new == NULL) { |
7163 | ndoms_cur = 0; | |
acc3f5d7 | 7164 | doms_new = &fallback_doms; |
6ad4c188 | 7165 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7166 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7167 | } |
7168 | ||
029190c5 PJ |
7169 | /* Build new domains */ |
7170 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7171 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7172 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7173 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7174 | goto match2; |
7175 | } | |
7176 | /* no match - add a new doms_new */ | |
acc3f5d7 | 7177 | __build_sched_domains(doms_new[i], |
1d3504fc | 7178 | dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7179 | match2: |
7180 | ; | |
7181 | } | |
7182 | ||
7183 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7184 | if (doms_cur != &fallback_doms) |
7185 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7186 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7187 | doms_cur = doms_new; |
1d3504fc | 7188 | dattr_cur = dattr_new; |
029190c5 | 7189 | ndoms_cur = ndoms_new; |
7378547f MM |
7190 | |
7191 | register_sched_domain_sysctl(); | |
a1835615 | 7192 | |
712555ee | 7193 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7194 | } |
7195 | ||
5c45bf27 | 7196 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 7197 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 7198 | { |
95402b38 | 7199 | get_online_cpus(); |
dfb512ec MK |
7200 | |
7201 | /* Destroy domains first to force the rebuild */ | |
7202 | partition_sched_domains(0, NULL, NULL); | |
7203 | ||
e761b772 | 7204 | rebuild_sched_domains(); |
95402b38 | 7205 | put_online_cpus(); |
5c45bf27 SS |
7206 | } |
7207 | ||
7208 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7209 | { | |
afb8a9b7 | 7210 | unsigned int level = 0; |
5c45bf27 | 7211 | |
afb8a9b7 GS |
7212 | if (sscanf(buf, "%u", &level) != 1) |
7213 | return -EINVAL; | |
7214 | ||
7215 | /* | |
7216 | * level is always be positive so don't check for | |
7217 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7218 | * What happens on 0 or 1 byte write, | |
7219 | * need to check for count as well? | |
7220 | */ | |
7221 | ||
7222 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7223 | return -EINVAL; |
7224 | ||
7225 | if (smt) | |
afb8a9b7 | 7226 | sched_smt_power_savings = level; |
5c45bf27 | 7227 | else |
afb8a9b7 | 7228 | sched_mc_power_savings = level; |
5c45bf27 | 7229 | |
c70f22d2 | 7230 | arch_reinit_sched_domains(); |
5c45bf27 | 7231 | |
c70f22d2 | 7232 | return count; |
5c45bf27 SS |
7233 | } |
7234 | ||
5c45bf27 | 7235 | #ifdef CONFIG_SCHED_MC |
f718cd4a | 7236 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
c9be0a36 | 7237 | struct sysdev_class_attribute *attr, |
f718cd4a | 7238 | char *page) |
5c45bf27 SS |
7239 | { |
7240 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7241 | } | |
f718cd4a | 7242 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
c9be0a36 | 7243 | struct sysdev_class_attribute *attr, |
48f24c4d | 7244 | const char *buf, size_t count) |
5c45bf27 SS |
7245 | { |
7246 | return sched_power_savings_store(buf, count, 0); | |
7247 | } | |
f718cd4a AK |
7248 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7249 | sched_mc_power_savings_show, | |
7250 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7251 | #endif |
7252 | ||
7253 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a | 7254 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
c9be0a36 | 7255 | struct sysdev_class_attribute *attr, |
f718cd4a | 7256 | char *page) |
5c45bf27 SS |
7257 | { |
7258 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7259 | } | |
f718cd4a | 7260 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
c9be0a36 | 7261 | struct sysdev_class_attribute *attr, |
48f24c4d | 7262 | const char *buf, size_t count) |
5c45bf27 SS |
7263 | { |
7264 | return sched_power_savings_store(buf, count, 1); | |
7265 | } | |
f718cd4a AK |
7266 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7267 | sched_smt_power_savings_show, | |
6707de00 AB |
7268 | sched_smt_power_savings_store); |
7269 | #endif | |
7270 | ||
39aac648 | 7271 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
7272 | { |
7273 | int err = 0; | |
7274 | ||
7275 | #ifdef CONFIG_SCHED_SMT | |
7276 | if (smt_capable()) | |
7277 | err = sysfs_create_file(&cls->kset.kobj, | |
7278 | &attr_sched_smt_power_savings.attr); | |
7279 | #endif | |
7280 | #ifdef CONFIG_SCHED_MC | |
7281 | if (!err && mc_capable()) | |
7282 | err = sysfs_create_file(&cls->kset.kobj, | |
7283 | &attr_sched_mc_power_savings.attr); | |
7284 | #endif | |
7285 | return err; | |
7286 | } | |
6d6bc0ad | 7287 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7288 | |
e761b772 | 7289 | #ifndef CONFIG_CPUSETS |
1da177e4 | 7290 | /* |
e761b772 MK |
7291 | * Add online and remove offline CPUs from the scheduler domains. |
7292 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
7293 | */ |
7294 | static int update_sched_domains(struct notifier_block *nfb, | |
7295 | unsigned long action, void *hcpu) | |
e761b772 MK |
7296 | { |
7297 | switch (action) { | |
7298 | case CPU_ONLINE: | |
7299 | case CPU_ONLINE_FROZEN: | |
6ad4c188 PZ |
7300 | case CPU_DOWN_PREPARE: |
7301 | case CPU_DOWN_PREPARE_FROZEN: | |
7302 | case CPU_DOWN_FAILED: | |
7303 | case CPU_DOWN_FAILED_FROZEN: | |
dfb512ec | 7304 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
7305 | return NOTIFY_OK; |
7306 | ||
7307 | default: | |
7308 | return NOTIFY_DONE; | |
7309 | } | |
7310 | } | |
7311 | #endif | |
7312 | ||
7313 | static int update_runtime(struct notifier_block *nfb, | |
7314 | unsigned long action, void *hcpu) | |
1da177e4 | 7315 | { |
7def2be1 PZ |
7316 | int cpu = (int)(long)hcpu; |
7317 | ||
1da177e4 | 7318 | switch (action) { |
1da177e4 | 7319 | case CPU_DOWN_PREPARE: |
8bb78442 | 7320 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7321 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7322 | return NOTIFY_OK; |
7323 | ||
1da177e4 | 7324 | case CPU_DOWN_FAILED: |
8bb78442 | 7325 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7326 | case CPU_ONLINE: |
8bb78442 | 7327 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7328 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7329 | return NOTIFY_OK; |
7330 | ||
1da177e4 LT |
7331 | default: |
7332 | return NOTIFY_DONE; | |
7333 | } | |
1da177e4 | 7334 | } |
1da177e4 LT |
7335 | |
7336 | void __init sched_init_smp(void) | |
7337 | { | |
dcc30a35 RR |
7338 | cpumask_var_t non_isolated_cpus; |
7339 | ||
7340 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7341 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7342 | |
434d53b0 MT |
7343 | #if defined(CONFIG_NUMA) |
7344 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7345 | GFP_KERNEL); | |
7346 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7347 | #endif | |
95402b38 | 7348 | get_online_cpus(); |
712555ee | 7349 | mutex_lock(&sched_domains_mutex); |
6ad4c188 | 7350 | arch_init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7351 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7352 | if (cpumask_empty(non_isolated_cpus)) | |
7353 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7354 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7355 | put_online_cpus(); |
e761b772 MK |
7356 | |
7357 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
7358 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7359 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
7360 | #endif |
7361 | ||
7362 | /* RT runtime code needs to handle some hotplug events */ | |
7363 | hotcpu_notifier(update_runtime, 0); | |
7364 | ||
b328ca18 | 7365 | init_hrtick(); |
5c1e1767 NP |
7366 | |
7367 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7368 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7369 | BUG(); |
19978ca6 | 7370 | sched_init_granularity(); |
dcc30a35 | 7371 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7372 | |
0e3900e6 | 7373 | init_sched_rt_class(); |
1da177e4 LT |
7374 | } |
7375 | #else | |
7376 | void __init sched_init_smp(void) | |
7377 | { | |
19978ca6 | 7378 | sched_init_granularity(); |
1da177e4 LT |
7379 | } |
7380 | #endif /* CONFIG_SMP */ | |
7381 | ||
cd1bb94b AB |
7382 | const_debug unsigned int sysctl_timer_migration = 1; |
7383 | ||
1da177e4 LT |
7384 | int in_sched_functions(unsigned long addr) |
7385 | { | |
1da177e4 LT |
7386 | return in_lock_functions(addr) || |
7387 | (addr >= (unsigned long)__sched_text_start | |
7388 | && addr < (unsigned long)__sched_text_end); | |
7389 | } | |
7390 | ||
a9957449 | 7391 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7392 | { |
7393 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7394 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7395 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7396 | cfs_rq->rq = rq; | |
7397 | #endif | |
67e9fb2a | 7398 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7399 | } |
7400 | ||
fa85ae24 PZ |
7401 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7402 | { | |
7403 | struct rt_prio_array *array; | |
7404 | int i; | |
7405 | ||
7406 | array = &rt_rq->active; | |
7407 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7408 | INIT_LIST_HEAD(array->queue + i); | |
7409 | __clear_bit(i, array->bitmap); | |
7410 | } | |
7411 | /* delimiter for bitsearch: */ | |
7412 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7413 | ||
052f1dc7 | 7414 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 7415 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 7416 | #ifdef CONFIG_SMP |
e864c499 | 7417 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 7418 | #endif |
48d5e258 | 7419 | #endif |
fa85ae24 PZ |
7420 | #ifdef CONFIG_SMP |
7421 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 7422 | rt_rq->overloaded = 0; |
05fa785c | 7423 | plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
7424 | #endif |
7425 | ||
7426 | rt_rq->rt_time = 0; | |
7427 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 7428 | rt_rq->rt_runtime = 0; |
0986b11b | 7429 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
6f505b16 | 7430 | |
052f1dc7 | 7431 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7432 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7433 | rt_rq->rq = rq; |
7434 | #endif | |
fa85ae24 PZ |
7435 | } |
7436 | ||
6f505b16 | 7437 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
7438 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
7439 | struct sched_entity *se, int cpu, int add, | |
7440 | struct sched_entity *parent) | |
6f505b16 | 7441 | { |
ec7dc8ac | 7442 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7443 | tg->cfs_rq[cpu] = cfs_rq; |
7444 | init_cfs_rq(cfs_rq, rq); | |
7445 | cfs_rq->tg = tg; | |
7446 | if (add) | |
7447 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7448 | ||
7449 | tg->se[cpu] = se; | |
354d60c2 DG |
7450 | /* se could be NULL for init_task_group */ |
7451 | if (!se) | |
7452 | return; | |
7453 | ||
ec7dc8ac DG |
7454 | if (!parent) |
7455 | se->cfs_rq = &rq->cfs; | |
7456 | else | |
7457 | se->cfs_rq = parent->my_q; | |
7458 | ||
6f505b16 PZ |
7459 | se->my_q = cfs_rq; |
7460 | se->load.weight = tg->shares; | |
e05510d0 | 7461 | se->load.inv_weight = 0; |
ec7dc8ac | 7462 | se->parent = parent; |
6f505b16 | 7463 | } |
052f1dc7 | 7464 | #endif |
6f505b16 | 7465 | |
052f1dc7 | 7466 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
7467 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
7468 | struct sched_rt_entity *rt_se, int cpu, int add, | |
7469 | struct sched_rt_entity *parent) | |
6f505b16 | 7470 | { |
ec7dc8ac DG |
7471 | struct rq *rq = cpu_rq(cpu); |
7472 | ||
6f505b16 PZ |
7473 | tg->rt_rq[cpu] = rt_rq; |
7474 | init_rt_rq(rt_rq, rq); | |
7475 | rt_rq->tg = tg; | |
ac086bc2 | 7476 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7477 | if (add) |
7478 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
7479 | ||
7480 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7481 | if (!rt_se) |
7482 | return; | |
7483 | ||
ec7dc8ac DG |
7484 | if (!parent) |
7485 | rt_se->rt_rq = &rq->rt; | |
7486 | else | |
7487 | rt_se->rt_rq = parent->my_q; | |
7488 | ||
6f505b16 | 7489 | rt_se->my_q = rt_rq; |
ec7dc8ac | 7490 | rt_se->parent = parent; |
6f505b16 PZ |
7491 | INIT_LIST_HEAD(&rt_se->run_list); |
7492 | } | |
7493 | #endif | |
7494 | ||
1da177e4 LT |
7495 | void __init sched_init(void) |
7496 | { | |
dd41f596 | 7497 | int i, j; |
434d53b0 MT |
7498 | unsigned long alloc_size = 0, ptr; |
7499 | ||
7500 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7501 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7502 | #endif | |
7503 | #ifdef CONFIG_RT_GROUP_SCHED | |
7504 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7505 | #endif |
df7c8e84 | 7506 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 7507 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 7508 | #endif |
434d53b0 | 7509 | if (alloc_size) { |
36b7b6d4 | 7510 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7511 | |
7512 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7513 | init_task_group.se = (struct sched_entity **)ptr; | |
7514 | ptr += nr_cpu_ids * sizeof(void **); | |
7515 | ||
7516 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7517 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7518 | |
6d6bc0ad | 7519 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 MT |
7520 | #ifdef CONFIG_RT_GROUP_SCHED |
7521 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7522 | ptr += nr_cpu_ids * sizeof(void **); | |
7523 | ||
7524 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
7525 | ptr += nr_cpu_ids * sizeof(void **); |
7526 | ||
6d6bc0ad | 7527 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
7528 | #ifdef CONFIG_CPUMASK_OFFSTACK |
7529 | for_each_possible_cpu(i) { | |
7530 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
7531 | ptr += cpumask_size(); | |
7532 | } | |
7533 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 7534 | } |
dd41f596 | 7535 | |
57d885fe GH |
7536 | #ifdef CONFIG_SMP |
7537 | init_defrootdomain(); | |
7538 | #endif | |
7539 | ||
d0b27fa7 PZ |
7540 | init_rt_bandwidth(&def_rt_bandwidth, |
7541 | global_rt_period(), global_rt_runtime()); | |
7542 | ||
7543 | #ifdef CONFIG_RT_GROUP_SCHED | |
7544 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
7545 | global_rt_period(), global_rt_runtime()); | |
6d6bc0ad | 7546 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7547 | |
7c941438 | 7548 | #ifdef CONFIG_CGROUP_SCHED |
6f505b16 | 7549 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
7550 | INIT_LIST_HEAD(&init_task_group.children); |
7551 | ||
7c941438 | 7552 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7553 | |
4a6cc4bd JK |
7554 | #if defined CONFIG_FAIR_GROUP_SCHED && defined CONFIG_SMP |
7555 | update_shares_data = __alloc_percpu(nr_cpu_ids * sizeof(unsigned long), | |
7556 | __alignof__(unsigned long)); | |
7557 | #endif | |
0a945022 | 7558 | for_each_possible_cpu(i) { |
70b97a7f | 7559 | struct rq *rq; |
1da177e4 LT |
7560 | |
7561 | rq = cpu_rq(i); | |
05fa785c | 7562 | raw_spin_lock_init(&rq->lock); |
7897986b | 7563 | rq->nr_running = 0; |
dce48a84 TG |
7564 | rq->calc_load_active = 0; |
7565 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 7566 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7567 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7568 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 7569 | init_task_group.shares = init_task_group_load; |
6f505b16 | 7570 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
7571 | #ifdef CONFIG_CGROUP_SCHED |
7572 | /* | |
7573 | * How much cpu bandwidth does init_task_group get? | |
7574 | * | |
7575 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7576 | * gets 100% of the cpu resources in the system. This overall | |
7577 | * system cpu resource is divided among the tasks of | |
7578 | * init_task_group and its child task-groups in a fair manner, | |
7579 | * based on each entity's (task or task-group's) weight | |
7580 | * (se->load.weight). | |
7581 | * | |
7582 | * In other words, if init_task_group has 10 tasks of weight | |
7583 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
7584 | * then A0's share of the cpu resource is: | |
7585 | * | |
0d905bca | 7586 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
7587 | * |
7588 | * We achieve this by letting init_task_group's tasks sit | |
7589 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
7590 | */ | |
ec7dc8ac | 7591 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
052f1dc7 | 7592 | #endif |
354d60c2 DG |
7593 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7594 | ||
7595 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7596 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7597 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 7598 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 7599 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 7600 | #endif |
dd41f596 | 7601 | #endif |
1da177e4 | 7602 | |
dd41f596 IM |
7603 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7604 | rq->cpu_load[j] = 0; | |
1da177e4 | 7605 | #ifdef CONFIG_SMP |
41c7ce9a | 7606 | rq->sd = NULL; |
57d885fe | 7607 | rq->rd = NULL; |
3f029d3c | 7608 | rq->post_schedule = 0; |
1da177e4 | 7609 | rq->active_balance = 0; |
dd41f596 | 7610 | rq->next_balance = jiffies; |
1da177e4 | 7611 | rq->push_cpu = 0; |
0a2966b4 | 7612 | rq->cpu = i; |
1f11eb6a | 7613 | rq->online = 0; |
eae0c9df MG |
7614 | rq->idle_stamp = 0; |
7615 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
dc938520 | 7616 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 7617 | #endif |
8f4d37ec | 7618 | init_rq_hrtick(rq); |
1da177e4 | 7619 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7620 | } |
7621 | ||
2dd73a4f | 7622 | set_load_weight(&init_task); |
b50f60ce | 7623 | |
e107be36 AK |
7624 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7625 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7626 | #endif | |
7627 | ||
c9819f45 | 7628 | #ifdef CONFIG_SMP |
962cf36c | 7629 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
7630 | #endif |
7631 | ||
b50f60ce | 7632 | #ifdef CONFIG_RT_MUTEXES |
1d615482 | 7633 | plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock); |
b50f60ce HC |
7634 | #endif |
7635 | ||
1da177e4 LT |
7636 | /* |
7637 | * The boot idle thread does lazy MMU switching as well: | |
7638 | */ | |
7639 | atomic_inc(&init_mm.mm_count); | |
7640 | enter_lazy_tlb(&init_mm, current); | |
7641 | ||
7642 | /* | |
7643 | * Make us the idle thread. Technically, schedule() should not be | |
7644 | * called from this thread, however somewhere below it might be, | |
7645 | * but because we are the idle thread, we just pick up running again | |
7646 | * when this runqueue becomes "idle". | |
7647 | */ | |
7648 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7649 | |
7650 | calc_load_update = jiffies + LOAD_FREQ; | |
7651 | ||
dd41f596 IM |
7652 | /* |
7653 | * During early bootup we pretend to be a normal task: | |
7654 | */ | |
7655 | current->sched_class = &fair_sched_class; | |
6892b75e | 7656 | |
6a7b3dc3 | 7657 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 7658 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 7659 | #ifdef CONFIG_SMP |
7d1e6a9b | 7660 | #ifdef CONFIG_NO_HZ |
49557e62 | 7661 | zalloc_cpumask_var(&nohz.cpu_mask, GFP_NOWAIT); |
4bdddf8f | 7662 | alloc_cpumask_var(&nohz.ilb_grp_nohz_mask, GFP_NOWAIT); |
7d1e6a9b | 7663 | #endif |
bdddd296 RR |
7664 | /* May be allocated at isolcpus cmdline parse time */ |
7665 | if (cpu_isolated_map == NULL) | |
7666 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 7667 | #endif /* SMP */ |
6a7b3dc3 | 7668 | |
cdd6c482 | 7669 | perf_event_init(); |
0d905bca | 7670 | |
6892b75e | 7671 | scheduler_running = 1; |
1da177e4 LT |
7672 | } |
7673 | ||
7674 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
7675 | static inline int preempt_count_equals(int preempt_offset) |
7676 | { | |
234da7bc | 7677 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 FW |
7678 | |
7679 | return (nested == PREEMPT_INATOMIC_BASE + preempt_offset); | |
7680 | } | |
7681 | ||
d894837f | 7682 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7683 | { |
48f24c4d | 7684 | #ifdef in_atomic |
1da177e4 LT |
7685 | static unsigned long prev_jiffy; /* ratelimiting */ |
7686 | ||
e4aafea2 FW |
7687 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
7688 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
7689 | return; |
7690 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7691 | return; | |
7692 | prev_jiffy = jiffies; | |
7693 | ||
3df0fc5b PZ |
7694 | printk(KERN_ERR |
7695 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7696 | file, line); | |
7697 | printk(KERN_ERR | |
7698 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7699 | in_atomic(), irqs_disabled(), | |
7700 | current->pid, current->comm); | |
aef745fc IM |
7701 | |
7702 | debug_show_held_locks(current); | |
7703 | if (irqs_disabled()) | |
7704 | print_irqtrace_events(current); | |
7705 | dump_stack(); | |
1da177e4 LT |
7706 | #endif |
7707 | } | |
7708 | EXPORT_SYMBOL(__might_sleep); | |
7709 | #endif | |
7710 | ||
7711 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7712 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7713 | { | |
7714 | int on_rq; | |
3e51f33f | 7715 | |
3a5e4dc1 AK |
7716 | on_rq = p->se.on_rq; |
7717 | if (on_rq) | |
7718 | deactivate_task(rq, p, 0); | |
7719 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
7720 | if (on_rq) { | |
7721 | activate_task(rq, p, 0); | |
7722 | resched_task(rq->curr); | |
7723 | } | |
7724 | } | |
7725 | ||
1da177e4 LT |
7726 | void normalize_rt_tasks(void) |
7727 | { | |
a0f98a1c | 7728 | struct task_struct *g, *p; |
1da177e4 | 7729 | unsigned long flags; |
70b97a7f | 7730 | struct rq *rq; |
1da177e4 | 7731 | |
4cf5d77a | 7732 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7733 | do_each_thread(g, p) { |
178be793 IM |
7734 | /* |
7735 | * Only normalize user tasks: | |
7736 | */ | |
7737 | if (!p->mm) | |
7738 | continue; | |
7739 | ||
6cfb0d5d | 7740 | p->se.exec_start = 0; |
6cfb0d5d | 7741 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7742 | p->se.statistics.wait_start = 0; |
7743 | p->se.statistics.sleep_start = 0; | |
7744 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7745 | #endif |
dd41f596 IM |
7746 | |
7747 | if (!rt_task(p)) { | |
7748 | /* | |
7749 | * Renice negative nice level userspace | |
7750 | * tasks back to 0: | |
7751 | */ | |
7752 | if (TASK_NICE(p) < 0 && p->mm) | |
7753 | set_user_nice(p, 0); | |
1da177e4 | 7754 | continue; |
dd41f596 | 7755 | } |
1da177e4 | 7756 | |
1d615482 | 7757 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7758 | rq = __task_rq_lock(p); |
1da177e4 | 7759 | |
178be793 | 7760 | normalize_task(rq, p); |
3a5e4dc1 | 7761 | |
b29739f9 | 7762 | __task_rq_unlock(rq); |
1d615482 | 7763 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7764 | } while_each_thread(g, p); |
7765 | ||
4cf5d77a | 7766 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7767 | } |
7768 | ||
7769 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7770 | |
67fc4e0c | 7771 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7772 | /* |
67fc4e0c | 7773 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7774 | * |
7775 | * They can only be called when the whole system has been | |
7776 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7777 | * activity can take place. Using them for anything else would | |
7778 | * be a serious bug, and as a result, they aren't even visible | |
7779 | * under any other configuration. | |
7780 | */ | |
7781 | ||
7782 | /** | |
7783 | * curr_task - return the current task for a given cpu. | |
7784 | * @cpu: the processor in question. | |
7785 | * | |
7786 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7787 | */ | |
36c8b586 | 7788 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7789 | { |
7790 | return cpu_curr(cpu); | |
7791 | } | |
7792 | ||
67fc4e0c JW |
7793 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7794 | ||
7795 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7796 | /** |
7797 | * set_curr_task - set the current task for a given cpu. | |
7798 | * @cpu: the processor in question. | |
7799 | * @p: the task pointer to set. | |
7800 | * | |
7801 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7802 | * are serviced on a separate stack. It allows the architecture to switch the |
7803 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7804 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7805 | * and caller must save the original value of the current task (see | |
7806 | * curr_task() above) and restore that value before reenabling interrupts and | |
7807 | * re-starting the system. | |
7808 | * | |
7809 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7810 | */ | |
36c8b586 | 7811 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7812 | { |
7813 | cpu_curr(cpu) = p; | |
7814 | } | |
7815 | ||
7816 | #endif | |
29f59db3 | 7817 | |
bccbe08a PZ |
7818 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7819 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
7820 | { |
7821 | int i; | |
7822 | ||
7823 | for_each_possible_cpu(i) { | |
7824 | if (tg->cfs_rq) | |
7825 | kfree(tg->cfs_rq[i]); | |
7826 | if (tg->se) | |
7827 | kfree(tg->se[i]); | |
6f505b16 PZ |
7828 | } |
7829 | ||
7830 | kfree(tg->cfs_rq); | |
7831 | kfree(tg->se); | |
6f505b16 PZ |
7832 | } |
7833 | ||
ec7dc8ac DG |
7834 | static |
7835 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 7836 | { |
29f59db3 | 7837 | struct cfs_rq *cfs_rq; |
eab17229 | 7838 | struct sched_entity *se; |
9b5b7751 | 7839 | struct rq *rq; |
29f59db3 SV |
7840 | int i; |
7841 | ||
434d53b0 | 7842 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
7843 | if (!tg->cfs_rq) |
7844 | goto err; | |
434d53b0 | 7845 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
7846 | if (!tg->se) |
7847 | goto err; | |
052f1dc7 PZ |
7848 | |
7849 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
7850 | |
7851 | for_each_possible_cpu(i) { | |
9b5b7751 | 7852 | rq = cpu_rq(i); |
29f59db3 | 7853 | |
eab17229 LZ |
7854 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
7855 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
7856 | if (!cfs_rq) |
7857 | goto err; | |
7858 | ||
eab17229 LZ |
7859 | se = kzalloc_node(sizeof(struct sched_entity), |
7860 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 7861 | if (!se) |
dfc12eb2 | 7862 | goto err_free_rq; |
29f59db3 | 7863 | |
eab17229 | 7864 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
7865 | } |
7866 | ||
7867 | return 1; | |
7868 | ||
dfc12eb2 PC |
7869 | err_free_rq: |
7870 | kfree(cfs_rq); | |
bccbe08a PZ |
7871 | err: |
7872 | return 0; | |
7873 | } | |
7874 | ||
7875 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
7876 | { | |
7877 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
7878 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
7879 | } | |
7880 | ||
7881 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
7882 | { | |
7883 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
7884 | } | |
6d6bc0ad | 7885 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
7886 | static inline void free_fair_sched_group(struct task_group *tg) |
7887 | { | |
7888 | } | |
7889 | ||
ec7dc8ac DG |
7890 | static inline |
7891 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
7892 | { |
7893 | return 1; | |
7894 | } | |
7895 | ||
7896 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
7897 | { | |
7898 | } | |
7899 | ||
7900 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
7901 | { | |
7902 | } | |
6d6bc0ad | 7903 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
7904 | |
7905 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
7906 | static void free_rt_sched_group(struct task_group *tg) |
7907 | { | |
7908 | int i; | |
7909 | ||
d0b27fa7 PZ |
7910 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
7911 | ||
bccbe08a PZ |
7912 | for_each_possible_cpu(i) { |
7913 | if (tg->rt_rq) | |
7914 | kfree(tg->rt_rq[i]); | |
7915 | if (tg->rt_se) | |
7916 | kfree(tg->rt_se[i]); | |
7917 | } | |
7918 | ||
7919 | kfree(tg->rt_rq); | |
7920 | kfree(tg->rt_se); | |
7921 | } | |
7922 | ||
ec7dc8ac DG |
7923 | static |
7924 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
7925 | { |
7926 | struct rt_rq *rt_rq; | |
eab17229 | 7927 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
7928 | struct rq *rq; |
7929 | int i; | |
7930 | ||
434d53b0 | 7931 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
7932 | if (!tg->rt_rq) |
7933 | goto err; | |
434d53b0 | 7934 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
7935 | if (!tg->rt_se) |
7936 | goto err; | |
7937 | ||
d0b27fa7 PZ |
7938 | init_rt_bandwidth(&tg->rt_bandwidth, |
7939 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
7940 | |
7941 | for_each_possible_cpu(i) { | |
7942 | rq = cpu_rq(i); | |
7943 | ||
eab17229 LZ |
7944 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
7945 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
7946 | if (!rt_rq) |
7947 | goto err; | |
29f59db3 | 7948 | |
eab17229 LZ |
7949 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
7950 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 7951 | if (!rt_se) |
dfc12eb2 | 7952 | goto err_free_rq; |
29f59db3 | 7953 | |
eab17229 | 7954 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
7955 | } |
7956 | ||
bccbe08a PZ |
7957 | return 1; |
7958 | ||
dfc12eb2 PC |
7959 | err_free_rq: |
7960 | kfree(rt_rq); | |
bccbe08a PZ |
7961 | err: |
7962 | return 0; | |
7963 | } | |
7964 | ||
7965 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
7966 | { | |
7967 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
7968 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
7969 | } | |
7970 | ||
7971 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
7972 | { | |
7973 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
7974 | } | |
6d6bc0ad | 7975 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
7976 | static inline void free_rt_sched_group(struct task_group *tg) |
7977 | { | |
7978 | } | |
7979 | ||
ec7dc8ac DG |
7980 | static inline |
7981 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
7982 | { |
7983 | return 1; | |
7984 | } | |
7985 | ||
7986 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
7987 | { | |
7988 | } | |
7989 | ||
7990 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
7991 | { | |
7992 | } | |
6d6bc0ad | 7993 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 7994 | |
7c941438 | 7995 | #ifdef CONFIG_CGROUP_SCHED |
bccbe08a PZ |
7996 | static void free_sched_group(struct task_group *tg) |
7997 | { | |
7998 | free_fair_sched_group(tg); | |
7999 | free_rt_sched_group(tg); | |
8000 | kfree(tg); | |
8001 | } | |
8002 | ||
8003 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8004 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8005 | { |
8006 | struct task_group *tg; | |
8007 | unsigned long flags; | |
8008 | int i; | |
8009 | ||
8010 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8011 | if (!tg) | |
8012 | return ERR_PTR(-ENOMEM); | |
8013 | ||
ec7dc8ac | 8014 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8015 | goto err; |
8016 | ||
ec7dc8ac | 8017 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8018 | goto err; |
8019 | ||
8ed36996 | 8020 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8021 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8022 | register_fair_sched_group(tg, i); |
8023 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8024 | } |
6f505b16 | 8025 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8026 | |
8027 | WARN_ON(!parent); /* root should already exist */ | |
8028 | ||
8029 | tg->parent = parent; | |
f473aa5e | 8030 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8031 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8032 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8033 | |
9b5b7751 | 8034 | return tg; |
29f59db3 SV |
8035 | |
8036 | err: | |
6f505b16 | 8037 | free_sched_group(tg); |
29f59db3 SV |
8038 | return ERR_PTR(-ENOMEM); |
8039 | } | |
8040 | ||
9b5b7751 | 8041 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8042 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8043 | { |
29f59db3 | 8044 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8045 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8046 | } |
8047 | ||
9b5b7751 | 8048 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8049 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8050 | { |
8ed36996 | 8051 | unsigned long flags; |
9b5b7751 | 8052 | int i; |
29f59db3 | 8053 | |
8ed36996 | 8054 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8055 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8056 | unregister_fair_sched_group(tg, i); |
8057 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8058 | } |
6f505b16 | 8059 | list_del_rcu(&tg->list); |
f473aa5e | 8060 | list_del_rcu(&tg->siblings); |
8ed36996 | 8061 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8062 | |
9b5b7751 | 8063 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8064 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8065 | } |
8066 | ||
9b5b7751 | 8067 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8068 | * The caller of this function should have put the task in its new group |
8069 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8070 | * reflect its new group. | |
9b5b7751 SV |
8071 | */ |
8072 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8073 | { |
8074 | int on_rq, running; | |
8075 | unsigned long flags; | |
8076 | struct rq *rq; | |
8077 | ||
8078 | rq = task_rq_lock(tsk, &flags); | |
8079 | ||
051a1d1a | 8080 | running = task_current(rq, tsk); |
29f59db3 SV |
8081 | on_rq = tsk->se.on_rq; |
8082 | ||
0e1f3483 | 8083 | if (on_rq) |
29f59db3 | 8084 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8085 | if (unlikely(running)) |
8086 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8087 | |
6f505b16 | 8088 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8089 | |
810b3817 PZ |
8090 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8091 | if (tsk->sched_class->moved_group) | |
88ec22d3 | 8092 | tsk->sched_class->moved_group(tsk, on_rq); |
810b3817 PZ |
8093 | #endif |
8094 | ||
0e1f3483 HS |
8095 | if (unlikely(running)) |
8096 | tsk->sched_class->set_curr_task(rq); | |
8097 | if (on_rq) | |
371fd7e7 | 8098 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8099 | |
29f59db3 SV |
8100 | task_rq_unlock(rq, &flags); |
8101 | } | |
7c941438 | 8102 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8103 | |
052f1dc7 | 8104 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8105 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8106 | { |
8107 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8108 | int on_rq; |
8109 | ||
29f59db3 | 8110 | on_rq = se->on_rq; |
62fb1851 | 8111 | if (on_rq) |
29f59db3 SV |
8112 | dequeue_entity(cfs_rq, se, 0); |
8113 | ||
8114 | se->load.weight = shares; | |
e05510d0 | 8115 | se->load.inv_weight = 0; |
29f59db3 | 8116 | |
62fb1851 | 8117 | if (on_rq) |
29f59db3 | 8118 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8119 | } |
62fb1851 | 8120 | |
c09595f6 PZ |
8121 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8122 | { | |
8123 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8124 | struct rq *rq = cfs_rq->rq; | |
8125 | unsigned long flags; | |
8126 | ||
05fa785c | 8127 | raw_spin_lock_irqsave(&rq->lock, flags); |
c09595f6 | 8128 | __set_se_shares(se, shares); |
05fa785c | 8129 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
29f59db3 SV |
8130 | } |
8131 | ||
8ed36996 PZ |
8132 | static DEFINE_MUTEX(shares_mutex); |
8133 | ||
4cf86d77 | 8134 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8135 | { |
8136 | int i; | |
8ed36996 | 8137 | unsigned long flags; |
c61935fd | 8138 | |
ec7dc8ac DG |
8139 | /* |
8140 | * We can't change the weight of the root cgroup. | |
8141 | */ | |
8142 | if (!tg->se[0]) | |
8143 | return -EINVAL; | |
8144 | ||
18d95a28 PZ |
8145 | if (shares < MIN_SHARES) |
8146 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8147 | else if (shares > MAX_SHARES) |
8148 | shares = MAX_SHARES; | |
62fb1851 | 8149 | |
8ed36996 | 8150 | mutex_lock(&shares_mutex); |
9b5b7751 | 8151 | if (tg->shares == shares) |
5cb350ba | 8152 | goto done; |
29f59db3 | 8153 | |
8ed36996 | 8154 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8155 | for_each_possible_cpu(i) |
8156 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8157 | list_del_rcu(&tg->siblings); |
8ed36996 | 8158 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8159 | |
8160 | /* wait for any ongoing reference to this group to finish */ | |
8161 | synchronize_sched(); | |
8162 | ||
8163 | /* | |
8164 | * Now we are free to modify the group's share on each cpu | |
8165 | * w/o tripping rebalance_share or load_balance_fair. | |
8166 | */ | |
9b5b7751 | 8167 | tg->shares = shares; |
c09595f6 PZ |
8168 | for_each_possible_cpu(i) { |
8169 | /* | |
8170 | * force a rebalance | |
8171 | */ | |
8172 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8173 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8174 | } |
29f59db3 | 8175 | |
6b2d7700 SV |
8176 | /* |
8177 | * Enable load balance activity on this group, by inserting it back on | |
8178 | * each cpu's rq->leaf_cfs_rq_list. | |
8179 | */ | |
8ed36996 | 8180 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8181 | for_each_possible_cpu(i) |
8182 | register_fair_sched_group(tg, i); | |
f473aa5e | 8183 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8184 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8185 | done: |
8ed36996 | 8186 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8187 | return 0; |
29f59db3 SV |
8188 | } |
8189 | ||
5cb350ba DG |
8190 | unsigned long sched_group_shares(struct task_group *tg) |
8191 | { | |
8192 | return tg->shares; | |
8193 | } | |
052f1dc7 | 8194 | #endif |
5cb350ba | 8195 | |
052f1dc7 | 8196 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8197 | /* |
9f0c1e56 | 8198 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8199 | */ |
9f0c1e56 PZ |
8200 | static DEFINE_MUTEX(rt_constraints_mutex); |
8201 | ||
8202 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8203 | { | |
8204 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8205 | return 1ULL << 20; |
9f0c1e56 | 8206 | |
9a7e0b18 | 8207 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8208 | } |
8209 | ||
9a7e0b18 PZ |
8210 | /* Must be called with tasklist_lock held */ |
8211 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8212 | { |
9a7e0b18 | 8213 | struct task_struct *g, *p; |
b40b2e8e | 8214 | |
9a7e0b18 PZ |
8215 | do_each_thread(g, p) { |
8216 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8217 | return 1; | |
8218 | } while_each_thread(g, p); | |
b40b2e8e | 8219 | |
9a7e0b18 PZ |
8220 | return 0; |
8221 | } | |
b40b2e8e | 8222 | |
9a7e0b18 PZ |
8223 | struct rt_schedulable_data { |
8224 | struct task_group *tg; | |
8225 | u64 rt_period; | |
8226 | u64 rt_runtime; | |
8227 | }; | |
b40b2e8e | 8228 | |
9a7e0b18 PZ |
8229 | static int tg_schedulable(struct task_group *tg, void *data) |
8230 | { | |
8231 | struct rt_schedulable_data *d = data; | |
8232 | struct task_group *child; | |
8233 | unsigned long total, sum = 0; | |
8234 | u64 period, runtime; | |
b40b2e8e | 8235 | |
9a7e0b18 PZ |
8236 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8237 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8238 | |
9a7e0b18 PZ |
8239 | if (tg == d->tg) { |
8240 | period = d->rt_period; | |
8241 | runtime = d->rt_runtime; | |
b40b2e8e | 8242 | } |
b40b2e8e | 8243 | |
4653f803 PZ |
8244 | /* |
8245 | * Cannot have more runtime than the period. | |
8246 | */ | |
8247 | if (runtime > period && runtime != RUNTIME_INF) | |
8248 | return -EINVAL; | |
6f505b16 | 8249 | |
4653f803 PZ |
8250 | /* |
8251 | * Ensure we don't starve existing RT tasks. | |
8252 | */ | |
9a7e0b18 PZ |
8253 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8254 | return -EBUSY; | |
6f505b16 | 8255 | |
9a7e0b18 | 8256 | total = to_ratio(period, runtime); |
6f505b16 | 8257 | |
4653f803 PZ |
8258 | /* |
8259 | * Nobody can have more than the global setting allows. | |
8260 | */ | |
8261 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8262 | return -EINVAL; | |
6f505b16 | 8263 | |
4653f803 PZ |
8264 | /* |
8265 | * The sum of our children's runtime should not exceed our own. | |
8266 | */ | |
9a7e0b18 PZ |
8267 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8268 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8269 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8270 | |
9a7e0b18 PZ |
8271 | if (child == d->tg) { |
8272 | period = d->rt_period; | |
8273 | runtime = d->rt_runtime; | |
8274 | } | |
6f505b16 | 8275 | |
9a7e0b18 | 8276 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8277 | } |
6f505b16 | 8278 | |
9a7e0b18 PZ |
8279 | if (sum > total) |
8280 | return -EINVAL; | |
8281 | ||
8282 | return 0; | |
6f505b16 PZ |
8283 | } |
8284 | ||
9a7e0b18 | 8285 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8286 | { |
9a7e0b18 PZ |
8287 | struct rt_schedulable_data data = { |
8288 | .tg = tg, | |
8289 | .rt_period = period, | |
8290 | .rt_runtime = runtime, | |
8291 | }; | |
8292 | ||
8293 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8294 | } |
8295 | ||
d0b27fa7 PZ |
8296 | static int tg_set_bandwidth(struct task_group *tg, |
8297 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8298 | { |
ac086bc2 | 8299 | int i, err = 0; |
9f0c1e56 | 8300 | |
9f0c1e56 | 8301 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8302 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8303 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8304 | if (err) | |
9f0c1e56 | 8305 | goto unlock; |
ac086bc2 | 8306 | |
0986b11b | 8307 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8308 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8309 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8310 | |
8311 | for_each_possible_cpu(i) { | |
8312 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8313 | ||
0986b11b | 8314 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8315 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8316 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8317 | } |
0986b11b | 8318 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
9f0c1e56 | 8319 | unlock: |
521f1a24 | 8320 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8321 | mutex_unlock(&rt_constraints_mutex); |
8322 | ||
8323 | return err; | |
6f505b16 PZ |
8324 | } |
8325 | ||
d0b27fa7 PZ |
8326 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8327 | { | |
8328 | u64 rt_runtime, rt_period; | |
8329 | ||
8330 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8331 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8332 | if (rt_runtime_us < 0) | |
8333 | rt_runtime = RUNTIME_INF; | |
8334 | ||
8335 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8336 | } | |
8337 | ||
9f0c1e56 PZ |
8338 | long sched_group_rt_runtime(struct task_group *tg) |
8339 | { | |
8340 | u64 rt_runtime_us; | |
8341 | ||
d0b27fa7 | 8342 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8343 | return -1; |
8344 | ||
d0b27fa7 | 8345 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8346 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8347 | return rt_runtime_us; | |
8348 | } | |
d0b27fa7 PZ |
8349 | |
8350 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8351 | { | |
8352 | u64 rt_runtime, rt_period; | |
8353 | ||
8354 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8355 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8356 | ||
619b0488 R |
8357 | if (rt_period == 0) |
8358 | return -EINVAL; | |
8359 | ||
d0b27fa7 PZ |
8360 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8361 | } | |
8362 | ||
8363 | long sched_group_rt_period(struct task_group *tg) | |
8364 | { | |
8365 | u64 rt_period_us; | |
8366 | ||
8367 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8368 | do_div(rt_period_us, NSEC_PER_USEC); | |
8369 | return rt_period_us; | |
8370 | } | |
8371 | ||
8372 | static int sched_rt_global_constraints(void) | |
8373 | { | |
4653f803 | 8374 | u64 runtime, period; |
d0b27fa7 PZ |
8375 | int ret = 0; |
8376 | ||
ec5d4989 HS |
8377 | if (sysctl_sched_rt_period <= 0) |
8378 | return -EINVAL; | |
8379 | ||
4653f803 PZ |
8380 | runtime = global_rt_runtime(); |
8381 | period = global_rt_period(); | |
8382 | ||
8383 | /* | |
8384 | * Sanity check on the sysctl variables. | |
8385 | */ | |
8386 | if (runtime > period && runtime != RUNTIME_INF) | |
8387 | return -EINVAL; | |
10b612f4 | 8388 | |
d0b27fa7 | 8389 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 8390 | read_lock(&tasklist_lock); |
4653f803 | 8391 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 8392 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
8393 | mutex_unlock(&rt_constraints_mutex); |
8394 | ||
8395 | return ret; | |
8396 | } | |
54e99124 DG |
8397 | |
8398 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
8399 | { | |
8400 | /* Don't accept realtime tasks when there is no way for them to run */ | |
8401 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
8402 | return 0; | |
8403 | ||
8404 | return 1; | |
8405 | } | |
8406 | ||
6d6bc0ad | 8407 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8408 | static int sched_rt_global_constraints(void) |
8409 | { | |
ac086bc2 PZ |
8410 | unsigned long flags; |
8411 | int i; | |
8412 | ||
ec5d4989 HS |
8413 | if (sysctl_sched_rt_period <= 0) |
8414 | return -EINVAL; | |
8415 | ||
60aa605d PZ |
8416 | /* |
8417 | * There's always some RT tasks in the root group | |
8418 | * -- migration, kstopmachine etc.. | |
8419 | */ | |
8420 | if (sysctl_sched_rt_runtime == 0) | |
8421 | return -EBUSY; | |
8422 | ||
0986b11b | 8423 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
8424 | for_each_possible_cpu(i) { |
8425 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8426 | ||
0986b11b | 8427 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8428 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 8429 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8430 | } |
0986b11b | 8431 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 8432 | |
d0b27fa7 PZ |
8433 | return 0; |
8434 | } | |
6d6bc0ad | 8435 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8436 | |
8437 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 8438 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8439 | loff_t *ppos) |
8440 | { | |
8441 | int ret; | |
8442 | int old_period, old_runtime; | |
8443 | static DEFINE_MUTEX(mutex); | |
8444 | ||
8445 | mutex_lock(&mutex); | |
8446 | old_period = sysctl_sched_rt_period; | |
8447 | old_runtime = sysctl_sched_rt_runtime; | |
8448 | ||
8d65af78 | 8449 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
8450 | |
8451 | if (!ret && write) { | |
8452 | ret = sched_rt_global_constraints(); | |
8453 | if (ret) { | |
8454 | sysctl_sched_rt_period = old_period; | |
8455 | sysctl_sched_rt_runtime = old_runtime; | |
8456 | } else { | |
8457 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8458 | def_rt_bandwidth.rt_period = | |
8459 | ns_to_ktime(global_rt_period()); | |
8460 | } | |
8461 | } | |
8462 | mutex_unlock(&mutex); | |
8463 | ||
8464 | return ret; | |
8465 | } | |
68318b8e | 8466 | |
052f1dc7 | 8467 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8468 | |
8469 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8470 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8471 | { |
2b01dfe3 PM |
8472 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8473 | struct task_group, css); | |
68318b8e SV |
8474 | } |
8475 | ||
8476 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8477 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8478 | { |
ec7dc8ac | 8479 | struct task_group *tg, *parent; |
68318b8e | 8480 | |
2b01dfe3 | 8481 | if (!cgrp->parent) { |
68318b8e | 8482 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
8483 | return &init_task_group.css; |
8484 | } | |
8485 | ||
ec7dc8ac DG |
8486 | parent = cgroup_tg(cgrp->parent); |
8487 | tg = sched_create_group(parent); | |
68318b8e SV |
8488 | if (IS_ERR(tg)) |
8489 | return ERR_PTR(-ENOMEM); | |
8490 | ||
68318b8e SV |
8491 | return &tg->css; |
8492 | } | |
8493 | ||
41a2d6cf IM |
8494 | static void |
8495 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8496 | { |
2b01dfe3 | 8497 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8498 | |
8499 | sched_destroy_group(tg); | |
8500 | } | |
8501 | ||
41a2d6cf | 8502 | static int |
be367d09 | 8503 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 8504 | { |
b68aa230 | 8505 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 8506 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
8507 | return -EINVAL; |
8508 | #else | |
68318b8e SV |
8509 | /* We don't support RT-tasks being in separate groups */ |
8510 | if (tsk->sched_class != &fair_sched_class) | |
8511 | return -EINVAL; | |
b68aa230 | 8512 | #endif |
be367d09 BB |
8513 | return 0; |
8514 | } | |
68318b8e | 8515 | |
be367d09 BB |
8516 | static int |
8517 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8518 | struct task_struct *tsk, bool threadgroup) | |
8519 | { | |
8520 | int retval = cpu_cgroup_can_attach_task(cgrp, tsk); | |
8521 | if (retval) | |
8522 | return retval; | |
8523 | if (threadgroup) { | |
8524 | struct task_struct *c; | |
8525 | rcu_read_lock(); | |
8526 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8527 | retval = cpu_cgroup_can_attach_task(cgrp, c); | |
8528 | if (retval) { | |
8529 | rcu_read_unlock(); | |
8530 | return retval; | |
8531 | } | |
8532 | } | |
8533 | rcu_read_unlock(); | |
8534 | } | |
68318b8e SV |
8535 | return 0; |
8536 | } | |
8537 | ||
8538 | static void | |
2b01dfe3 | 8539 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
be367d09 BB |
8540 | struct cgroup *old_cont, struct task_struct *tsk, |
8541 | bool threadgroup) | |
68318b8e SV |
8542 | { |
8543 | sched_move_task(tsk); | |
be367d09 BB |
8544 | if (threadgroup) { |
8545 | struct task_struct *c; | |
8546 | rcu_read_lock(); | |
8547 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8548 | sched_move_task(c); | |
8549 | } | |
8550 | rcu_read_unlock(); | |
8551 | } | |
68318b8e SV |
8552 | } |
8553 | ||
052f1dc7 | 8554 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8555 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8556 | u64 shareval) |
68318b8e | 8557 | { |
2b01dfe3 | 8558 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8559 | } |
8560 | ||
f4c753b7 | 8561 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8562 | { |
2b01dfe3 | 8563 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8564 | |
8565 | return (u64) tg->shares; | |
8566 | } | |
6d6bc0ad | 8567 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8568 | |
052f1dc7 | 8569 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8570 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8571 | s64 val) |
6f505b16 | 8572 | { |
06ecb27c | 8573 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8574 | } |
8575 | ||
06ecb27c | 8576 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8577 | { |
06ecb27c | 8578 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8579 | } |
d0b27fa7 PZ |
8580 | |
8581 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8582 | u64 rt_period_us) | |
8583 | { | |
8584 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8585 | } | |
8586 | ||
8587 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8588 | { | |
8589 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8590 | } | |
6d6bc0ad | 8591 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8592 | |
fe5c7cc2 | 8593 | static struct cftype cpu_files[] = { |
052f1dc7 | 8594 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8595 | { |
8596 | .name = "shares", | |
f4c753b7 PM |
8597 | .read_u64 = cpu_shares_read_u64, |
8598 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8599 | }, |
052f1dc7 PZ |
8600 | #endif |
8601 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8602 | { |
9f0c1e56 | 8603 | .name = "rt_runtime_us", |
06ecb27c PM |
8604 | .read_s64 = cpu_rt_runtime_read, |
8605 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8606 | }, |
d0b27fa7 PZ |
8607 | { |
8608 | .name = "rt_period_us", | |
f4c753b7 PM |
8609 | .read_u64 = cpu_rt_period_read_uint, |
8610 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8611 | }, |
052f1dc7 | 8612 | #endif |
68318b8e SV |
8613 | }; |
8614 | ||
8615 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8616 | { | |
fe5c7cc2 | 8617 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8618 | } |
8619 | ||
8620 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8621 | .name = "cpu", |
8622 | .create = cpu_cgroup_create, | |
8623 | .destroy = cpu_cgroup_destroy, | |
8624 | .can_attach = cpu_cgroup_can_attach, | |
8625 | .attach = cpu_cgroup_attach, | |
8626 | .populate = cpu_cgroup_populate, | |
8627 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8628 | .early_init = 1, |
8629 | }; | |
8630 | ||
052f1dc7 | 8631 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8632 | |
8633 | #ifdef CONFIG_CGROUP_CPUACCT | |
8634 | ||
8635 | /* | |
8636 | * CPU accounting code for task groups. | |
8637 | * | |
8638 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8639 | * (balbir@in.ibm.com). | |
8640 | */ | |
8641 | ||
934352f2 | 8642 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
8643 | struct cpuacct { |
8644 | struct cgroup_subsys_state css; | |
8645 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43cf38eb | 8646 | u64 __percpu *cpuusage; |
ef12fefa | 8647 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 8648 | struct cpuacct *parent; |
d842de87 SV |
8649 | }; |
8650 | ||
8651 | struct cgroup_subsys cpuacct_subsys; | |
8652 | ||
8653 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8654 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8655 | { |
32cd756a | 8656 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8657 | struct cpuacct, css); |
8658 | } | |
8659 | ||
8660 | /* return cpu accounting group to which this task belongs */ | |
8661 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8662 | { | |
8663 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8664 | struct cpuacct, css); | |
8665 | } | |
8666 | ||
8667 | /* create a new cpu accounting group */ | |
8668 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8669 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
8670 | { |
8671 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 8672 | int i; |
d842de87 SV |
8673 | |
8674 | if (!ca) | |
ef12fefa | 8675 | goto out; |
d842de87 SV |
8676 | |
8677 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
8678 | if (!ca->cpuusage) |
8679 | goto out_free_ca; | |
8680 | ||
8681 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
8682 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
8683 | goto out_free_counters; | |
d842de87 | 8684 | |
934352f2 BR |
8685 | if (cgrp->parent) |
8686 | ca->parent = cgroup_ca(cgrp->parent); | |
8687 | ||
d842de87 | 8688 | return &ca->css; |
ef12fefa BR |
8689 | |
8690 | out_free_counters: | |
8691 | while (--i >= 0) | |
8692 | percpu_counter_destroy(&ca->cpustat[i]); | |
8693 | free_percpu(ca->cpuusage); | |
8694 | out_free_ca: | |
8695 | kfree(ca); | |
8696 | out: | |
8697 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
8698 | } |
8699 | ||
8700 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 8701 | static void |
32cd756a | 8702 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8703 | { |
32cd756a | 8704 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 8705 | int i; |
d842de87 | 8706 | |
ef12fefa BR |
8707 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
8708 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
8709 | free_percpu(ca->cpuusage); |
8710 | kfree(ca); | |
8711 | } | |
8712 | ||
720f5498 KC |
8713 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
8714 | { | |
b36128c8 | 8715 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8716 | u64 data; |
8717 | ||
8718 | #ifndef CONFIG_64BIT | |
8719 | /* | |
8720 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
8721 | */ | |
05fa785c | 8722 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8723 | data = *cpuusage; |
05fa785c | 8724 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8725 | #else |
8726 | data = *cpuusage; | |
8727 | #endif | |
8728 | ||
8729 | return data; | |
8730 | } | |
8731 | ||
8732 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
8733 | { | |
b36128c8 | 8734 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8735 | |
8736 | #ifndef CONFIG_64BIT | |
8737 | /* | |
8738 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
8739 | */ | |
05fa785c | 8740 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8741 | *cpuusage = val; |
05fa785c | 8742 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8743 | #else |
8744 | *cpuusage = val; | |
8745 | #endif | |
8746 | } | |
8747 | ||
d842de87 | 8748 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 8749 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 8750 | { |
32cd756a | 8751 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8752 | u64 totalcpuusage = 0; |
8753 | int i; | |
8754 | ||
720f5498 KC |
8755 | for_each_present_cpu(i) |
8756 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
8757 | |
8758 | return totalcpuusage; | |
8759 | } | |
8760 | ||
0297b803 DG |
8761 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
8762 | u64 reset) | |
8763 | { | |
8764 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8765 | int err = 0; | |
8766 | int i; | |
8767 | ||
8768 | if (reset) { | |
8769 | err = -EINVAL; | |
8770 | goto out; | |
8771 | } | |
8772 | ||
720f5498 KC |
8773 | for_each_present_cpu(i) |
8774 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 8775 | |
0297b803 DG |
8776 | out: |
8777 | return err; | |
8778 | } | |
8779 | ||
e9515c3c KC |
8780 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
8781 | struct seq_file *m) | |
8782 | { | |
8783 | struct cpuacct *ca = cgroup_ca(cgroup); | |
8784 | u64 percpu; | |
8785 | int i; | |
8786 | ||
8787 | for_each_present_cpu(i) { | |
8788 | percpu = cpuacct_cpuusage_read(ca, i); | |
8789 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
8790 | } | |
8791 | seq_printf(m, "\n"); | |
8792 | return 0; | |
8793 | } | |
8794 | ||
ef12fefa BR |
8795 | static const char *cpuacct_stat_desc[] = { |
8796 | [CPUACCT_STAT_USER] = "user", | |
8797 | [CPUACCT_STAT_SYSTEM] = "system", | |
8798 | }; | |
8799 | ||
8800 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
8801 | struct cgroup_map_cb *cb) | |
8802 | { | |
8803 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8804 | int i; | |
8805 | ||
8806 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
8807 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
8808 | val = cputime64_to_clock_t(val); | |
8809 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
8810 | } | |
8811 | return 0; | |
8812 | } | |
8813 | ||
d842de87 SV |
8814 | static struct cftype files[] = { |
8815 | { | |
8816 | .name = "usage", | |
f4c753b7 PM |
8817 | .read_u64 = cpuusage_read, |
8818 | .write_u64 = cpuusage_write, | |
d842de87 | 8819 | }, |
e9515c3c KC |
8820 | { |
8821 | .name = "usage_percpu", | |
8822 | .read_seq_string = cpuacct_percpu_seq_read, | |
8823 | }, | |
ef12fefa BR |
8824 | { |
8825 | .name = "stat", | |
8826 | .read_map = cpuacct_stats_show, | |
8827 | }, | |
d842de87 SV |
8828 | }; |
8829 | ||
32cd756a | 8830 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8831 | { |
32cd756a | 8832 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
8833 | } |
8834 | ||
8835 | /* | |
8836 | * charge this task's execution time to its accounting group. | |
8837 | * | |
8838 | * called with rq->lock held. | |
8839 | */ | |
8840 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
8841 | { | |
8842 | struct cpuacct *ca; | |
934352f2 | 8843 | int cpu; |
d842de87 | 8844 | |
c40c6f85 | 8845 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
8846 | return; |
8847 | ||
934352f2 | 8848 | cpu = task_cpu(tsk); |
a18b83b7 BR |
8849 | |
8850 | rcu_read_lock(); | |
8851 | ||
d842de87 | 8852 | ca = task_ca(tsk); |
d842de87 | 8853 | |
934352f2 | 8854 | for (; ca; ca = ca->parent) { |
b36128c8 | 8855 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
8856 | *cpuusage += cputime; |
8857 | } | |
a18b83b7 BR |
8858 | |
8859 | rcu_read_unlock(); | |
d842de87 SV |
8860 | } |
8861 | ||
fa535a77 AB |
8862 | /* |
8863 | * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large | |
8864 | * in cputime_t units. As a result, cpuacct_update_stats calls | |
8865 | * percpu_counter_add with values large enough to always overflow the | |
8866 | * per cpu batch limit causing bad SMP scalability. | |
8867 | * | |
8868 | * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we | |
8869 | * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled | |
8870 | * and enabled. We cap it at INT_MAX which is the largest allowed batch value. | |
8871 | */ | |
8872 | #ifdef CONFIG_SMP | |
8873 | #define CPUACCT_BATCH \ | |
8874 | min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX) | |
8875 | #else | |
8876 | #define CPUACCT_BATCH 0 | |
8877 | #endif | |
8878 | ||
ef12fefa BR |
8879 | /* |
8880 | * Charge the system/user time to the task's accounting group. | |
8881 | */ | |
8882 | static void cpuacct_update_stats(struct task_struct *tsk, | |
8883 | enum cpuacct_stat_index idx, cputime_t val) | |
8884 | { | |
8885 | struct cpuacct *ca; | |
fa535a77 | 8886 | int batch = CPUACCT_BATCH; |
ef12fefa BR |
8887 | |
8888 | if (unlikely(!cpuacct_subsys.active)) | |
8889 | return; | |
8890 | ||
8891 | rcu_read_lock(); | |
8892 | ca = task_ca(tsk); | |
8893 | ||
8894 | do { | |
fa535a77 | 8895 | __percpu_counter_add(&ca->cpustat[idx], val, batch); |
ef12fefa BR |
8896 | ca = ca->parent; |
8897 | } while (ca); | |
8898 | rcu_read_unlock(); | |
8899 | } | |
8900 | ||
d842de87 SV |
8901 | struct cgroup_subsys cpuacct_subsys = { |
8902 | .name = "cpuacct", | |
8903 | .create = cpuacct_create, | |
8904 | .destroy = cpuacct_destroy, | |
8905 | .populate = cpuacct_populate, | |
8906 | .subsys_id = cpuacct_subsys_id, | |
8907 | }; | |
8908 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf PM |
8909 | |
8910 | #ifndef CONFIG_SMP | |
8911 | ||
03b042bf PM |
8912 | void synchronize_sched_expedited(void) |
8913 | { | |
fc390cde | 8914 | barrier(); |
03b042bf PM |
8915 | } |
8916 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
8917 | ||
8918 | #else /* #ifndef CONFIG_SMP */ | |
8919 | ||
cc631fb7 | 8920 | static atomic_t synchronize_sched_expedited_count = ATOMIC_INIT(0); |
03b042bf | 8921 | |
cc631fb7 | 8922 | static int synchronize_sched_expedited_cpu_stop(void *data) |
03b042bf | 8923 | { |
969c7921 TH |
8924 | /* |
8925 | * There must be a full memory barrier on each affected CPU | |
8926 | * between the time that try_stop_cpus() is called and the | |
8927 | * time that it returns. | |
8928 | * | |
8929 | * In the current initial implementation of cpu_stop, the | |
8930 | * above condition is already met when the control reaches | |
8931 | * this point and the following smp_mb() is not strictly | |
8932 | * necessary. Do smp_mb() anyway for documentation and | |
8933 | * robustness against future implementation changes. | |
8934 | */ | |
cc631fb7 | 8935 | smp_mb(); /* See above comment block. */ |
969c7921 | 8936 | return 0; |
03b042bf | 8937 | } |
03b042bf PM |
8938 | |
8939 | /* | |
8940 | * Wait for an rcu-sched grace period to elapse, but use "big hammer" | |
8941 | * approach to force grace period to end quickly. This consumes | |
8942 | * significant time on all CPUs, and is thus not recommended for | |
8943 | * any sort of common-case code. | |
8944 | * | |
8945 | * Note that it is illegal to call this function while holding any | |
8946 | * lock that is acquired by a CPU-hotplug notifier. Failing to | |
8947 | * observe this restriction will result in deadlock. | |
8948 | */ | |
8949 | void synchronize_sched_expedited(void) | |
8950 | { | |
969c7921 | 8951 | int snap, trycount = 0; |
03b042bf PM |
8952 | |
8953 | smp_mb(); /* ensure prior mod happens before capturing snap. */ | |
969c7921 | 8954 | snap = atomic_read(&synchronize_sched_expedited_count) + 1; |
03b042bf | 8955 | get_online_cpus(); |
969c7921 TH |
8956 | while (try_stop_cpus(cpu_online_mask, |
8957 | synchronize_sched_expedited_cpu_stop, | |
94458d5e | 8958 | NULL) == -EAGAIN) { |
03b042bf PM |
8959 | put_online_cpus(); |
8960 | if (trycount++ < 10) | |
8961 | udelay(trycount * num_online_cpus()); | |
8962 | else { | |
8963 | synchronize_sched(); | |
8964 | return; | |
8965 | } | |
969c7921 | 8966 | if (atomic_read(&synchronize_sched_expedited_count) - snap > 0) { |
03b042bf PM |
8967 | smp_mb(); /* ensure test happens before caller kfree */ |
8968 | return; | |
8969 | } | |
8970 | get_online_cpus(); | |
8971 | } | |
969c7921 | 8972 | atomic_inc(&synchronize_sched_expedited_count); |
cc631fb7 | 8973 | smp_mb__after_atomic_inc(); /* ensure post-GP actions seen after GP. */ |
03b042bf | 8974 | put_online_cpus(); |
03b042bf PM |
8975 | } |
8976 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
8977 | ||
8978 | #endif /* #else #ifndef CONFIG_SMP */ |