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 | 308 | |
7c941438 | 309 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 310 | |
6aa645ea IM |
311 | /* CFS-related fields in a runqueue */ |
312 | struct cfs_rq { | |
313 | struct load_weight load; | |
314 | unsigned long nr_running; | |
315 | ||
6aa645ea | 316 | u64 exec_clock; |
e9acbff6 | 317 | u64 min_vruntime; |
6aa645ea IM |
318 | |
319 | struct rb_root tasks_timeline; | |
320 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
321 | |
322 | struct list_head tasks; | |
323 | struct list_head *balance_iterator; | |
324 | ||
325 | /* | |
326 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
327 | * It is set to NULL otherwise (i.e when none are currently running). |
328 | */ | |
4793241b | 329 | struct sched_entity *curr, *next, *last; |
ddc97297 | 330 | |
5ac5c4d6 | 331 | unsigned int nr_spread_over; |
ddc97297 | 332 | |
62160e3f | 333 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
334 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
335 | ||
41a2d6cf IM |
336 | /* |
337 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
338 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
339 | * (like users, containers etc.) | |
340 | * | |
341 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
342 | * list is used during load balance. | |
343 | */ | |
41a2d6cf IM |
344 | struct list_head leaf_cfs_rq_list; |
345 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
346 | |
347 | #ifdef CONFIG_SMP | |
c09595f6 | 348 | /* |
c8cba857 | 349 | * the part of load.weight contributed by tasks |
c09595f6 | 350 | */ |
c8cba857 | 351 | unsigned long task_weight; |
c09595f6 | 352 | |
c8cba857 PZ |
353 | /* |
354 | * h_load = weight * f(tg) | |
355 | * | |
356 | * Where f(tg) is the recursive weight fraction assigned to | |
357 | * this group. | |
358 | */ | |
359 | unsigned long h_load; | |
c09595f6 | 360 | |
c8cba857 PZ |
361 | /* |
362 | * this cpu's part of tg->shares | |
363 | */ | |
364 | unsigned long shares; | |
f1d239f7 PZ |
365 | |
366 | /* | |
367 | * load.weight at the time we set shares | |
368 | */ | |
369 | unsigned long rq_weight; | |
c09595f6 | 370 | #endif |
6aa645ea IM |
371 | #endif |
372 | }; | |
1da177e4 | 373 | |
6aa645ea IM |
374 | /* Real-Time classes' related field in a runqueue: */ |
375 | struct rt_rq { | |
376 | struct rt_prio_array active; | |
63489e45 | 377 | unsigned long rt_nr_running; |
052f1dc7 | 378 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
379 | struct { |
380 | int curr; /* highest queued rt task prio */ | |
398a153b | 381 | #ifdef CONFIG_SMP |
e864c499 | 382 | int next; /* next highest */ |
398a153b | 383 | #endif |
e864c499 | 384 | } highest_prio; |
6f505b16 | 385 | #endif |
fa85ae24 | 386 | #ifdef CONFIG_SMP |
73fe6aae | 387 | unsigned long rt_nr_migratory; |
a1ba4d8b | 388 | unsigned long rt_nr_total; |
a22d7fc1 | 389 | int overloaded; |
917b627d | 390 | struct plist_head pushable_tasks; |
fa85ae24 | 391 | #endif |
6f505b16 | 392 | int rt_throttled; |
fa85ae24 | 393 | u64 rt_time; |
ac086bc2 | 394 | u64 rt_runtime; |
ea736ed5 | 395 | /* Nests inside the rq lock: */ |
0986b11b | 396 | raw_spinlock_t rt_runtime_lock; |
6f505b16 | 397 | |
052f1dc7 | 398 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
399 | unsigned long rt_nr_boosted; |
400 | ||
6f505b16 PZ |
401 | struct rq *rq; |
402 | struct list_head leaf_rt_rq_list; | |
403 | struct task_group *tg; | |
6f505b16 | 404 | #endif |
6aa645ea IM |
405 | }; |
406 | ||
57d885fe GH |
407 | #ifdef CONFIG_SMP |
408 | ||
409 | /* | |
410 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
411 | * variables. Each exclusive cpuset essentially defines an island domain by |
412 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
413 | * exclusive cpuset is created, we also create and attach a new root-domain |
414 | * object. | |
415 | * | |
57d885fe GH |
416 | */ |
417 | struct root_domain { | |
418 | atomic_t refcount; | |
c6c4927b RR |
419 | cpumask_var_t span; |
420 | cpumask_var_t online; | |
637f5085 | 421 | |
0eab9146 | 422 | /* |
637f5085 GH |
423 | * The "RT overload" flag: it gets set if a CPU has more than |
424 | * one runnable RT task. | |
425 | */ | |
c6c4927b | 426 | cpumask_var_t rto_mask; |
0eab9146 | 427 | atomic_t rto_count; |
6e0534f2 GH |
428 | #ifdef CONFIG_SMP |
429 | struct cpupri cpupri; | |
430 | #endif | |
57d885fe GH |
431 | }; |
432 | ||
dc938520 GH |
433 | /* |
434 | * By default the system creates a single root-domain with all cpus as | |
435 | * members (mimicking the global state we have today). | |
436 | */ | |
57d885fe GH |
437 | static struct root_domain def_root_domain; |
438 | ||
439 | #endif | |
440 | ||
1da177e4 LT |
441 | /* |
442 | * This is the main, per-CPU runqueue data structure. | |
443 | * | |
444 | * Locking rule: those places that want to lock multiple runqueues | |
445 | * (such as the load balancing or the thread migration code), lock | |
446 | * acquire operations must be ordered by ascending &runqueue. | |
447 | */ | |
70b97a7f | 448 | struct rq { |
d8016491 | 449 | /* runqueue lock: */ |
05fa785c | 450 | raw_spinlock_t lock; |
1da177e4 LT |
451 | |
452 | /* | |
453 | * nr_running and cpu_load should be in the same cacheline because | |
454 | * remote CPUs use both these fields when doing load calculation. | |
455 | */ | |
456 | unsigned long nr_running; | |
6aa645ea IM |
457 | #define CPU_LOAD_IDX_MAX 5 |
458 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
46cb4b7c | 459 | #ifdef CONFIG_NO_HZ |
39c0cbe2 | 460 | u64 nohz_stamp; |
46cb4b7c SS |
461 | unsigned char in_nohz_recently; |
462 | #endif | |
a64692a3 MG |
463 | unsigned int skip_clock_update; |
464 | ||
d8016491 IM |
465 | /* capture load from *all* tasks on this cpu: */ |
466 | struct load_weight load; | |
6aa645ea IM |
467 | unsigned long nr_load_updates; |
468 | u64 nr_switches; | |
469 | ||
470 | struct cfs_rq cfs; | |
6f505b16 | 471 | struct rt_rq rt; |
6f505b16 | 472 | |
6aa645ea | 473 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
474 | /* list of leaf cfs_rq on this cpu: */ |
475 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
476 | #endif |
477 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 478 | struct list_head leaf_rt_rq_list; |
1da177e4 | 479 | #endif |
1da177e4 LT |
480 | |
481 | /* | |
482 | * This is part of a global counter where only the total sum | |
483 | * over all CPUs matters. A task can increase this counter on | |
484 | * one CPU and if it got migrated afterwards it may decrease | |
485 | * it on another CPU. Always updated under the runqueue lock: | |
486 | */ | |
487 | unsigned long nr_uninterruptible; | |
488 | ||
36c8b586 | 489 | struct task_struct *curr, *idle; |
c9819f45 | 490 | unsigned long next_balance; |
1da177e4 | 491 | struct mm_struct *prev_mm; |
6aa645ea | 492 | |
3e51f33f | 493 | u64 clock; |
6aa645ea | 494 | |
1da177e4 LT |
495 | atomic_t nr_iowait; |
496 | ||
497 | #ifdef CONFIG_SMP | |
0eab9146 | 498 | struct root_domain *rd; |
1da177e4 LT |
499 | struct sched_domain *sd; |
500 | ||
e51fd5e2 PZ |
501 | unsigned long cpu_power; |
502 | ||
a0a522ce | 503 | unsigned char idle_at_tick; |
1da177e4 | 504 | /* For active balancing */ |
3f029d3c | 505 | int post_schedule; |
1da177e4 LT |
506 | int active_balance; |
507 | int push_cpu; | |
969c7921 | 508 | struct cpu_stop_work active_balance_work; |
d8016491 IM |
509 | /* cpu of this runqueue: */ |
510 | int cpu; | |
1f11eb6a | 511 | int online; |
1da177e4 | 512 | |
a8a51d5e | 513 | unsigned long avg_load_per_task; |
1da177e4 | 514 | |
e9e9250b PZ |
515 | u64 rt_avg; |
516 | u64 age_stamp; | |
1b9508f6 MG |
517 | u64 idle_stamp; |
518 | u64 avg_idle; | |
1da177e4 LT |
519 | #endif |
520 | ||
dce48a84 TG |
521 | /* calc_load related fields */ |
522 | unsigned long calc_load_update; | |
523 | long calc_load_active; | |
524 | ||
8f4d37ec | 525 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
526 | #ifdef CONFIG_SMP |
527 | int hrtick_csd_pending; | |
528 | struct call_single_data hrtick_csd; | |
529 | #endif | |
8f4d37ec PZ |
530 | struct hrtimer hrtick_timer; |
531 | #endif | |
532 | ||
1da177e4 LT |
533 | #ifdef CONFIG_SCHEDSTATS |
534 | /* latency stats */ | |
535 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
536 | unsigned long long rq_cpu_time; |
537 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
538 | |
539 | /* sys_sched_yield() stats */ | |
480b9434 | 540 | unsigned int yld_count; |
1da177e4 LT |
541 | |
542 | /* schedule() stats */ | |
480b9434 KC |
543 | unsigned int sched_switch; |
544 | unsigned int sched_count; | |
545 | unsigned int sched_goidle; | |
1da177e4 LT |
546 | |
547 | /* try_to_wake_up() stats */ | |
480b9434 KC |
548 | unsigned int ttwu_count; |
549 | unsigned int ttwu_local; | |
b8efb561 IM |
550 | |
551 | /* BKL stats */ | |
480b9434 | 552 | unsigned int bkl_count; |
1da177e4 LT |
553 | #endif |
554 | }; | |
555 | ||
f34e3b61 | 556 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 557 | |
7d478721 PZ |
558 | static inline |
559 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) | |
dd41f596 | 560 | { |
7d478721 | 561 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); |
a64692a3 MG |
562 | |
563 | /* | |
564 | * A queue event has occurred, and we're going to schedule. In | |
565 | * this case, we can save a useless back to back clock update. | |
566 | */ | |
567 | if (test_tsk_need_resched(p)) | |
568 | rq->skip_clock_update = 1; | |
dd41f596 IM |
569 | } |
570 | ||
0a2966b4 CL |
571 | static inline int cpu_of(struct rq *rq) |
572 | { | |
573 | #ifdef CONFIG_SMP | |
574 | return rq->cpu; | |
575 | #else | |
576 | return 0; | |
577 | #endif | |
578 | } | |
579 | ||
497f0ab3 | 580 | #define rcu_dereference_check_sched_domain(p) \ |
d11c563d PM |
581 | rcu_dereference_check((p), \ |
582 | rcu_read_lock_sched_held() || \ | |
583 | lockdep_is_held(&sched_domains_mutex)) | |
584 | ||
674311d5 NP |
585 | /* |
586 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 587 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
588 | * |
589 | * The domain tree of any CPU may only be accessed from within | |
590 | * preempt-disabled sections. | |
591 | */ | |
48f24c4d | 592 | #define for_each_domain(cpu, __sd) \ |
497f0ab3 | 593 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) |
1da177e4 LT |
594 | |
595 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
596 | #define this_rq() (&__get_cpu_var(runqueues)) | |
597 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
598 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 599 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 600 | |
dc61b1d6 PZ |
601 | #ifdef CONFIG_CGROUP_SCHED |
602 | ||
603 | /* | |
604 | * Return the group to which this tasks belongs. | |
605 | * | |
606 | * We use task_subsys_state_check() and extend the RCU verification | |
607 | * with lockdep_is_held(&task_rq(p)->lock) because cpu_cgroup_attach() | |
608 | * holds that lock for each task it moves into the cgroup. Therefore | |
609 | * by holding that lock, we pin the task to the current cgroup. | |
610 | */ | |
611 | static inline struct task_group *task_group(struct task_struct *p) | |
612 | { | |
613 | struct cgroup_subsys_state *css; | |
614 | ||
615 | css = task_subsys_state_check(p, cpu_cgroup_subsys_id, | |
616 | lockdep_is_held(&task_rq(p)->lock)); | |
617 | return container_of(css, struct task_group, css); | |
618 | } | |
619 | ||
620 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
621 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
622 | { | |
623 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
624 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; | |
625 | p->se.parent = task_group(p)->se[cpu]; | |
626 | #endif | |
627 | ||
628 | #ifdef CONFIG_RT_GROUP_SCHED | |
629 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; | |
630 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
631 | #endif | |
632 | } | |
633 | ||
634 | #else /* CONFIG_CGROUP_SCHED */ | |
635 | ||
636 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
637 | static inline struct task_group *task_group(struct task_struct *p) | |
638 | { | |
639 | return NULL; | |
640 | } | |
641 | ||
642 | #endif /* CONFIG_CGROUP_SCHED */ | |
643 | ||
aa9c4c0f | 644 | inline void update_rq_clock(struct rq *rq) |
3e51f33f | 645 | { |
a64692a3 MG |
646 | if (!rq->skip_clock_update) |
647 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
3e51f33f PZ |
648 | } |
649 | ||
bf5c91ba IM |
650 | /* |
651 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
652 | */ | |
653 | #ifdef CONFIG_SCHED_DEBUG | |
654 | # define const_debug __read_mostly | |
655 | #else | |
656 | # define const_debug static const | |
657 | #endif | |
658 | ||
017730c1 IM |
659 | /** |
660 | * runqueue_is_locked | |
e17b38bf | 661 | * @cpu: the processor in question. |
017730c1 IM |
662 | * |
663 | * Returns true if the current cpu runqueue is locked. | |
664 | * This interface allows printk to be called with the runqueue lock | |
665 | * held and know whether or not it is OK to wake up the klogd. | |
666 | */ | |
89f19f04 | 667 | int runqueue_is_locked(int cpu) |
017730c1 | 668 | { |
05fa785c | 669 | return raw_spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
670 | } |
671 | ||
bf5c91ba IM |
672 | /* |
673 | * Debugging: various feature bits | |
674 | */ | |
f00b45c1 PZ |
675 | |
676 | #define SCHED_FEAT(name, enabled) \ | |
677 | __SCHED_FEAT_##name , | |
678 | ||
bf5c91ba | 679 | enum { |
f00b45c1 | 680 | #include "sched_features.h" |
bf5c91ba IM |
681 | }; |
682 | ||
f00b45c1 PZ |
683 | #undef SCHED_FEAT |
684 | ||
685 | #define SCHED_FEAT(name, enabled) \ | |
686 | (1UL << __SCHED_FEAT_##name) * enabled | | |
687 | ||
bf5c91ba | 688 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
689 | #include "sched_features.h" |
690 | 0; | |
691 | ||
692 | #undef SCHED_FEAT | |
693 | ||
694 | #ifdef CONFIG_SCHED_DEBUG | |
695 | #define SCHED_FEAT(name, enabled) \ | |
696 | #name , | |
697 | ||
983ed7a6 | 698 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
699 | #include "sched_features.h" |
700 | NULL | |
701 | }; | |
702 | ||
703 | #undef SCHED_FEAT | |
704 | ||
34f3a814 | 705 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 706 | { |
f00b45c1 PZ |
707 | int i; |
708 | ||
709 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
710 | if (!(sysctl_sched_features & (1UL << i))) |
711 | seq_puts(m, "NO_"); | |
712 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 713 | } |
34f3a814 | 714 | seq_puts(m, "\n"); |
f00b45c1 | 715 | |
34f3a814 | 716 | return 0; |
f00b45c1 PZ |
717 | } |
718 | ||
719 | static ssize_t | |
720 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
721 | size_t cnt, loff_t *ppos) | |
722 | { | |
723 | char buf[64]; | |
724 | char *cmp = buf; | |
725 | int neg = 0; | |
726 | int i; | |
727 | ||
728 | if (cnt > 63) | |
729 | cnt = 63; | |
730 | ||
731 | if (copy_from_user(&buf, ubuf, cnt)) | |
732 | return -EFAULT; | |
733 | ||
734 | buf[cnt] = 0; | |
735 | ||
c24b7c52 | 736 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
737 | neg = 1; |
738 | cmp += 3; | |
739 | } | |
740 | ||
741 | for (i = 0; sched_feat_names[i]; i++) { | |
742 | int len = strlen(sched_feat_names[i]); | |
743 | ||
744 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
745 | if (neg) | |
746 | sysctl_sched_features &= ~(1UL << i); | |
747 | else | |
748 | sysctl_sched_features |= (1UL << i); | |
749 | break; | |
750 | } | |
751 | } | |
752 | ||
753 | if (!sched_feat_names[i]) | |
754 | return -EINVAL; | |
755 | ||
42994724 | 756 | *ppos += cnt; |
f00b45c1 PZ |
757 | |
758 | return cnt; | |
759 | } | |
760 | ||
34f3a814 LZ |
761 | static int sched_feat_open(struct inode *inode, struct file *filp) |
762 | { | |
763 | return single_open(filp, sched_feat_show, NULL); | |
764 | } | |
765 | ||
828c0950 | 766 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
767 | .open = sched_feat_open, |
768 | .write = sched_feat_write, | |
769 | .read = seq_read, | |
770 | .llseek = seq_lseek, | |
771 | .release = single_release, | |
f00b45c1 PZ |
772 | }; |
773 | ||
774 | static __init int sched_init_debug(void) | |
775 | { | |
f00b45c1 PZ |
776 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
777 | &sched_feat_fops); | |
778 | ||
779 | return 0; | |
780 | } | |
781 | late_initcall(sched_init_debug); | |
782 | ||
783 | #endif | |
784 | ||
785 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 786 | |
b82d9fdd PZ |
787 | /* |
788 | * Number of tasks to iterate in a single balance run. | |
789 | * Limited because this is done with IRQs disabled. | |
790 | */ | |
791 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
792 | ||
2398f2c6 PZ |
793 | /* |
794 | * ratelimit for updating the group shares. | |
55cd5340 | 795 | * default: 0.25ms |
2398f2c6 | 796 | */ |
55cd5340 | 797 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
0bcdcf28 | 798 | unsigned int normalized_sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 799 | |
ffda12a1 PZ |
800 | /* |
801 | * Inject some fuzzyness into changing the per-cpu group shares | |
802 | * this avoids remote rq-locks at the expense of fairness. | |
803 | * default: 4 | |
804 | */ | |
805 | unsigned int sysctl_sched_shares_thresh = 4; | |
806 | ||
e9e9250b PZ |
807 | /* |
808 | * period over which we average the RT time consumption, measured | |
809 | * in ms. | |
810 | * | |
811 | * default: 1s | |
812 | */ | |
813 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
814 | ||
fa85ae24 | 815 | /* |
9f0c1e56 | 816 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
817 | * default: 1s |
818 | */ | |
9f0c1e56 | 819 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 820 | |
6892b75e IM |
821 | static __read_mostly int scheduler_running; |
822 | ||
9f0c1e56 PZ |
823 | /* |
824 | * part of the period that we allow rt tasks to run in us. | |
825 | * default: 0.95s | |
826 | */ | |
827 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 828 | |
d0b27fa7 PZ |
829 | static inline u64 global_rt_period(void) |
830 | { | |
831 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
832 | } | |
833 | ||
834 | static inline u64 global_rt_runtime(void) | |
835 | { | |
e26873bb | 836 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
837 | return RUNTIME_INF; |
838 | ||
839 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
840 | } | |
fa85ae24 | 841 | |
1da177e4 | 842 | #ifndef prepare_arch_switch |
4866cde0 NP |
843 | # define prepare_arch_switch(next) do { } while (0) |
844 | #endif | |
845 | #ifndef finish_arch_switch | |
846 | # define finish_arch_switch(prev) do { } while (0) | |
847 | #endif | |
848 | ||
051a1d1a DA |
849 | static inline int task_current(struct rq *rq, struct task_struct *p) |
850 | { | |
851 | return rq->curr == p; | |
852 | } | |
853 | ||
4866cde0 | 854 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 855 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 856 | { |
051a1d1a | 857 | return task_current(rq, p); |
4866cde0 NP |
858 | } |
859 | ||
70b97a7f | 860 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
861 | { |
862 | } | |
863 | ||
70b97a7f | 864 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 865 | { |
da04c035 IM |
866 | #ifdef CONFIG_DEBUG_SPINLOCK |
867 | /* this is a valid case when another task releases the spinlock */ | |
868 | rq->lock.owner = current; | |
869 | #endif | |
8a25d5de IM |
870 | /* |
871 | * If we are tracking spinlock dependencies then we have to | |
872 | * fix up the runqueue lock - which gets 'carried over' from | |
873 | * prev into current: | |
874 | */ | |
875 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
876 | ||
05fa785c | 877 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 NP |
878 | } |
879 | ||
880 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 881 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
882 | { |
883 | #ifdef CONFIG_SMP | |
884 | return p->oncpu; | |
885 | #else | |
051a1d1a | 886 | return task_current(rq, p); |
4866cde0 NP |
887 | #endif |
888 | } | |
889 | ||
70b97a7f | 890 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
891 | { |
892 | #ifdef CONFIG_SMP | |
893 | /* | |
894 | * We can optimise this out completely for !SMP, because the | |
895 | * SMP rebalancing from interrupt is the only thing that cares | |
896 | * here. | |
897 | */ | |
898 | next->oncpu = 1; | |
899 | #endif | |
900 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
05fa785c | 901 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 | 902 | #else |
05fa785c | 903 | raw_spin_unlock(&rq->lock); |
4866cde0 NP |
904 | #endif |
905 | } | |
906 | ||
70b97a7f | 907 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
908 | { |
909 | #ifdef CONFIG_SMP | |
910 | /* | |
911 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
912 | * We must ensure this doesn't happen until the switch is completely | |
913 | * finished. | |
914 | */ | |
915 | smp_wmb(); | |
916 | prev->oncpu = 0; | |
917 | #endif | |
918 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
919 | local_irq_enable(); | |
1da177e4 | 920 | #endif |
4866cde0 NP |
921 | } |
922 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 923 | |
0970d299 | 924 | /* |
65cc8e48 PZ |
925 | * Check whether the task is waking, we use this to synchronize ->cpus_allowed |
926 | * against ttwu(). | |
0970d299 PZ |
927 | */ |
928 | static inline int task_is_waking(struct task_struct *p) | |
929 | { | |
0017d735 | 930 | return unlikely(p->state == TASK_WAKING); |
0970d299 PZ |
931 | } |
932 | ||
b29739f9 IM |
933 | /* |
934 | * __task_rq_lock - lock the runqueue a given task resides on. | |
935 | * Must be called interrupts disabled. | |
936 | */ | |
70b97a7f | 937 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
938 | __acquires(rq->lock) |
939 | { | |
0970d299 PZ |
940 | struct rq *rq; |
941 | ||
3a5c359a | 942 | for (;;) { |
0970d299 | 943 | rq = task_rq(p); |
05fa785c | 944 | raw_spin_lock(&rq->lock); |
65cc8e48 | 945 | if (likely(rq == task_rq(p))) |
3a5c359a | 946 | return rq; |
05fa785c | 947 | raw_spin_unlock(&rq->lock); |
b29739f9 | 948 | } |
b29739f9 IM |
949 | } |
950 | ||
1da177e4 LT |
951 | /* |
952 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 953 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
954 | * explicitly disabling preemption. |
955 | */ | |
70b97a7f | 956 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
957 | __acquires(rq->lock) |
958 | { | |
70b97a7f | 959 | struct rq *rq; |
1da177e4 | 960 | |
3a5c359a AK |
961 | for (;;) { |
962 | local_irq_save(*flags); | |
963 | rq = task_rq(p); | |
05fa785c | 964 | raw_spin_lock(&rq->lock); |
65cc8e48 | 965 | if (likely(rq == task_rq(p))) |
3a5c359a | 966 | return rq; |
05fa785c | 967 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 968 | } |
1da177e4 LT |
969 | } |
970 | ||
a9957449 | 971 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
972 | __releases(rq->lock) |
973 | { | |
05fa785c | 974 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
975 | } |
976 | ||
70b97a7f | 977 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
978 | __releases(rq->lock) |
979 | { | |
05fa785c | 980 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 LT |
981 | } |
982 | ||
1da177e4 | 983 | /* |
cc2a73b5 | 984 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 985 | */ |
a9957449 | 986 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
987 | __acquires(rq->lock) |
988 | { | |
70b97a7f | 989 | struct rq *rq; |
1da177e4 LT |
990 | |
991 | local_irq_disable(); | |
992 | rq = this_rq(); | |
05fa785c | 993 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
994 | |
995 | return rq; | |
996 | } | |
997 | ||
8f4d37ec PZ |
998 | #ifdef CONFIG_SCHED_HRTICK |
999 | /* | |
1000 | * Use HR-timers to deliver accurate preemption points. | |
1001 | * | |
1002 | * Its all a bit involved since we cannot program an hrt while holding the | |
1003 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1004 | * reschedule event. | |
1005 | * | |
1006 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1007 | * rq->lock. | |
1008 | */ | |
8f4d37ec PZ |
1009 | |
1010 | /* | |
1011 | * Use hrtick when: | |
1012 | * - enabled by features | |
1013 | * - hrtimer is actually high res | |
1014 | */ | |
1015 | static inline int hrtick_enabled(struct rq *rq) | |
1016 | { | |
1017 | if (!sched_feat(HRTICK)) | |
1018 | return 0; | |
ba42059f | 1019 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1020 | return 0; |
8f4d37ec PZ |
1021 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1022 | } | |
1023 | ||
8f4d37ec PZ |
1024 | static void hrtick_clear(struct rq *rq) |
1025 | { | |
1026 | if (hrtimer_active(&rq->hrtick_timer)) | |
1027 | hrtimer_cancel(&rq->hrtick_timer); | |
1028 | } | |
1029 | ||
8f4d37ec PZ |
1030 | /* |
1031 | * High-resolution timer tick. | |
1032 | * Runs from hardirq context with interrupts disabled. | |
1033 | */ | |
1034 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1035 | { | |
1036 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1037 | ||
1038 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1039 | ||
05fa785c | 1040 | raw_spin_lock(&rq->lock); |
3e51f33f | 1041 | update_rq_clock(rq); |
8f4d37ec | 1042 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 1043 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
1044 | |
1045 | return HRTIMER_NORESTART; | |
1046 | } | |
1047 | ||
95e904c7 | 1048 | #ifdef CONFIG_SMP |
31656519 PZ |
1049 | /* |
1050 | * called from hardirq (IPI) context | |
1051 | */ | |
1052 | static void __hrtick_start(void *arg) | |
b328ca18 | 1053 | { |
31656519 | 1054 | struct rq *rq = arg; |
b328ca18 | 1055 | |
05fa785c | 1056 | raw_spin_lock(&rq->lock); |
31656519 PZ |
1057 | hrtimer_restart(&rq->hrtick_timer); |
1058 | rq->hrtick_csd_pending = 0; | |
05fa785c | 1059 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
1060 | } |
1061 | ||
31656519 PZ |
1062 | /* |
1063 | * Called to set the hrtick timer state. | |
1064 | * | |
1065 | * called with rq->lock held and irqs disabled | |
1066 | */ | |
1067 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1068 | { |
31656519 PZ |
1069 | struct hrtimer *timer = &rq->hrtick_timer; |
1070 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1071 | |
cc584b21 | 1072 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1073 | |
1074 | if (rq == this_rq()) { | |
1075 | hrtimer_restart(timer); | |
1076 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1077 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1078 | rq->hrtick_csd_pending = 1; |
1079 | } | |
b328ca18 PZ |
1080 | } |
1081 | ||
1082 | static int | |
1083 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1084 | { | |
1085 | int cpu = (int)(long)hcpu; | |
1086 | ||
1087 | switch (action) { | |
1088 | case CPU_UP_CANCELED: | |
1089 | case CPU_UP_CANCELED_FROZEN: | |
1090 | case CPU_DOWN_PREPARE: | |
1091 | case CPU_DOWN_PREPARE_FROZEN: | |
1092 | case CPU_DEAD: | |
1093 | case CPU_DEAD_FROZEN: | |
31656519 | 1094 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1095 | return NOTIFY_OK; |
1096 | } | |
1097 | ||
1098 | return NOTIFY_DONE; | |
1099 | } | |
1100 | ||
fa748203 | 1101 | static __init void init_hrtick(void) |
b328ca18 PZ |
1102 | { |
1103 | hotcpu_notifier(hotplug_hrtick, 0); | |
1104 | } | |
31656519 PZ |
1105 | #else |
1106 | /* | |
1107 | * Called to set the hrtick timer state. | |
1108 | * | |
1109 | * called with rq->lock held and irqs disabled | |
1110 | */ | |
1111 | static void hrtick_start(struct rq *rq, u64 delay) | |
1112 | { | |
7f1e2ca9 | 1113 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1114 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1115 | } |
b328ca18 | 1116 | |
006c75f1 | 1117 | static inline void init_hrtick(void) |
8f4d37ec | 1118 | { |
8f4d37ec | 1119 | } |
31656519 | 1120 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1121 | |
31656519 | 1122 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1123 | { |
31656519 PZ |
1124 | #ifdef CONFIG_SMP |
1125 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1126 | |
31656519 PZ |
1127 | rq->hrtick_csd.flags = 0; |
1128 | rq->hrtick_csd.func = __hrtick_start; | |
1129 | rq->hrtick_csd.info = rq; | |
1130 | #endif | |
8f4d37ec | 1131 | |
31656519 PZ |
1132 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1133 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1134 | } |
006c75f1 | 1135 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1136 | static inline void hrtick_clear(struct rq *rq) |
1137 | { | |
1138 | } | |
1139 | ||
8f4d37ec PZ |
1140 | static inline void init_rq_hrtick(struct rq *rq) |
1141 | { | |
1142 | } | |
1143 | ||
b328ca18 PZ |
1144 | static inline void init_hrtick(void) |
1145 | { | |
1146 | } | |
006c75f1 | 1147 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1148 | |
c24d20db IM |
1149 | /* |
1150 | * resched_task - mark a task 'to be rescheduled now'. | |
1151 | * | |
1152 | * On UP this means the setting of the need_resched flag, on SMP it | |
1153 | * might also involve a cross-CPU call to trigger the scheduler on | |
1154 | * the target CPU. | |
1155 | */ | |
1156 | #ifdef CONFIG_SMP | |
1157 | ||
1158 | #ifndef tsk_is_polling | |
1159 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1160 | #endif | |
1161 | ||
31656519 | 1162 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1163 | { |
1164 | int cpu; | |
1165 | ||
05fa785c | 1166 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 1167 | |
5ed0cec0 | 1168 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1169 | return; |
1170 | ||
5ed0cec0 | 1171 | set_tsk_need_resched(p); |
c24d20db IM |
1172 | |
1173 | cpu = task_cpu(p); | |
1174 | if (cpu == smp_processor_id()) | |
1175 | return; | |
1176 | ||
1177 | /* NEED_RESCHED must be visible before we test polling */ | |
1178 | smp_mb(); | |
1179 | if (!tsk_is_polling(p)) | |
1180 | smp_send_reschedule(cpu); | |
1181 | } | |
1182 | ||
1183 | static void resched_cpu(int cpu) | |
1184 | { | |
1185 | struct rq *rq = cpu_rq(cpu); | |
1186 | unsigned long flags; | |
1187 | ||
05fa785c | 1188 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
1189 | return; |
1190 | resched_task(cpu_curr(cpu)); | |
05fa785c | 1191 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 1192 | } |
06d8308c TG |
1193 | |
1194 | #ifdef CONFIG_NO_HZ | |
1195 | /* | |
1196 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1197 | * idle CPU then this timer might expire before the next timer event | |
1198 | * which is scheduled to wake up that CPU. In case of a completely | |
1199 | * idle system the next event might even be infinite time into the | |
1200 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1201 | * leaves the inner idle loop so the newly added timer is taken into | |
1202 | * account when the CPU goes back to idle and evaluates the timer | |
1203 | * wheel for the next timer event. | |
1204 | */ | |
1205 | void wake_up_idle_cpu(int cpu) | |
1206 | { | |
1207 | struct rq *rq = cpu_rq(cpu); | |
1208 | ||
1209 | if (cpu == smp_processor_id()) | |
1210 | return; | |
1211 | ||
1212 | /* | |
1213 | * This is safe, as this function is called with the timer | |
1214 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1215 | * to idle and has not yet set rq->curr to idle then it will | |
1216 | * be serialized on the timer wheel base lock and take the new | |
1217 | * timer into account automatically. | |
1218 | */ | |
1219 | if (rq->curr != rq->idle) | |
1220 | return; | |
1221 | ||
1222 | /* | |
1223 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1224 | * lockless. The worst case is that the other CPU runs the | |
1225 | * idle task through an additional NOOP schedule() | |
1226 | */ | |
5ed0cec0 | 1227 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1228 | |
1229 | /* NEED_RESCHED must be visible before we test polling */ | |
1230 | smp_mb(); | |
1231 | if (!tsk_is_polling(rq->idle)) | |
1232 | smp_send_reschedule(cpu); | |
1233 | } | |
39c0cbe2 MG |
1234 | |
1235 | int nohz_ratelimit(int cpu) | |
1236 | { | |
1237 | struct rq *rq = cpu_rq(cpu); | |
1238 | u64 diff = rq->clock - rq->nohz_stamp; | |
1239 | ||
1240 | rq->nohz_stamp = rq->clock; | |
1241 | ||
1242 | return diff < (NSEC_PER_SEC / HZ) >> 1; | |
1243 | } | |
1244 | ||
6d6bc0ad | 1245 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1246 | |
e9e9250b PZ |
1247 | static u64 sched_avg_period(void) |
1248 | { | |
1249 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1250 | } | |
1251 | ||
1252 | static void sched_avg_update(struct rq *rq) | |
1253 | { | |
1254 | s64 period = sched_avg_period(); | |
1255 | ||
1256 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
1257 | /* |
1258 | * Inline assembly required to prevent the compiler | |
1259 | * optimising this loop into a divmod call. | |
1260 | * See __iter_div_u64_rem() for another example of this. | |
1261 | */ | |
1262 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
1263 | rq->age_stamp += period; |
1264 | rq->rt_avg /= 2; | |
1265 | } | |
1266 | } | |
1267 | ||
1268 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1269 | { | |
1270 | rq->rt_avg += rt_delta; | |
1271 | sched_avg_update(rq); | |
1272 | } | |
1273 | ||
6d6bc0ad | 1274 | #else /* !CONFIG_SMP */ |
31656519 | 1275 | static void resched_task(struct task_struct *p) |
c24d20db | 1276 | { |
05fa785c | 1277 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 1278 | set_tsk_need_resched(p); |
c24d20db | 1279 | } |
e9e9250b PZ |
1280 | |
1281 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1282 | { | |
1283 | } | |
6d6bc0ad | 1284 | #endif /* CONFIG_SMP */ |
c24d20db | 1285 | |
45bf76df IM |
1286 | #if BITS_PER_LONG == 32 |
1287 | # define WMULT_CONST (~0UL) | |
1288 | #else | |
1289 | # define WMULT_CONST (1UL << 32) | |
1290 | #endif | |
1291 | ||
1292 | #define WMULT_SHIFT 32 | |
1293 | ||
194081eb IM |
1294 | /* |
1295 | * Shift right and round: | |
1296 | */ | |
cf2ab469 | 1297 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1298 | |
a7be37ac PZ |
1299 | /* |
1300 | * delta *= weight / lw | |
1301 | */ | |
cb1c4fc9 | 1302 | static unsigned long |
45bf76df IM |
1303 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1304 | struct load_weight *lw) | |
1305 | { | |
1306 | u64 tmp; | |
1307 | ||
7a232e03 LJ |
1308 | if (!lw->inv_weight) { |
1309 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1310 | lw->inv_weight = 1; | |
1311 | else | |
1312 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1313 | / (lw->weight+1); | |
1314 | } | |
45bf76df IM |
1315 | |
1316 | tmp = (u64)delta_exec * weight; | |
1317 | /* | |
1318 | * Check whether we'd overflow the 64-bit multiplication: | |
1319 | */ | |
194081eb | 1320 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1321 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1322 | WMULT_SHIFT/2); |
1323 | else | |
cf2ab469 | 1324 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1325 | |
ecf691da | 1326 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1327 | } |
1328 | ||
1091985b | 1329 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1330 | { |
1331 | lw->weight += inc; | |
e89996ae | 1332 | lw->inv_weight = 0; |
45bf76df IM |
1333 | } |
1334 | ||
1091985b | 1335 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1336 | { |
1337 | lw->weight -= dec; | |
e89996ae | 1338 | lw->inv_weight = 0; |
45bf76df IM |
1339 | } |
1340 | ||
2dd73a4f PW |
1341 | /* |
1342 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1343 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1344 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1345 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1346 | * scaled version of the new time slice allocation that they receive on time |
1347 | * slice expiry etc. | |
1348 | */ | |
1349 | ||
cce7ade8 PZ |
1350 | #define WEIGHT_IDLEPRIO 3 |
1351 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1352 | |
1353 | /* | |
1354 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1355 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1356 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1357 | * that remained on nice 0. | |
1358 | * | |
1359 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1360 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1361 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1362 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1363 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1364 | */ |
1365 | static const int prio_to_weight[40] = { | |
254753dc IM |
1366 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1367 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1368 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1369 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1370 | /* 0 */ 1024, 820, 655, 526, 423, | |
1371 | /* 5 */ 335, 272, 215, 172, 137, | |
1372 | /* 10 */ 110, 87, 70, 56, 45, | |
1373 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1374 | }; |
1375 | ||
5714d2de IM |
1376 | /* |
1377 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1378 | * | |
1379 | * In cases where the weight does not change often, we can use the | |
1380 | * precalculated inverse to speed up arithmetics by turning divisions | |
1381 | * into multiplications: | |
1382 | */ | |
dd41f596 | 1383 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1384 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1385 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1386 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1387 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1388 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1389 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1390 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1391 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1392 | }; |
2dd73a4f | 1393 | |
ef12fefa BR |
1394 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1395 | enum cpuacct_stat_index { | |
1396 | CPUACCT_STAT_USER, /* ... user mode */ | |
1397 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1398 | ||
1399 | CPUACCT_STAT_NSTATS, | |
1400 | }; | |
1401 | ||
d842de87 SV |
1402 | #ifdef CONFIG_CGROUP_CPUACCT |
1403 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1404 | static void cpuacct_update_stats(struct task_struct *tsk, |
1405 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1406 | #else |
1407 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1408 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1409 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1410 | #endif |
1411 | ||
18d95a28 PZ |
1412 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1413 | { | |
1414 | update_load_add(&rq->load, load); | |
1415 | } | |
1416 | ||
1417 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1418 | { | |
1419 | update_load_sub(&rq->load, load); | |
1420 | } | |
1421 | ||
7940ca36 | 1422 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1423 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1424 | |
1425 | /* | |
1426 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1427 | * leaving it for the final time. | |
1428 | */ | |
eb755805 | 1429 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1430 | { |
1431 | struct task_group *parent, *child; | |
eb755805 | 1432 | int ret; |
c09595f6 PZ |
1433 | |
1434 | rcu_read_lock(); | |
1435 | parent = &root_task_group; | |
1436 | down: | |
eb755805 PZ |
1437 | ret = (*down)(parent, data); |
1438 | if (ret) | |
1439 | goto out_unlock; | |
c09595f6 PZ |
1440 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1441 | parent = child; | |
1442 | goto down; | |
1443 | ||
1444 | up: | |
1445 | continue; | |
1446 | } | |
eb755805 PZ |
1447 | ret = (*up)(parent, data); |
1448 | if (ret) | |
1449 | goto out_unlock; | |
c09595f6 PZ |
1450 | |
1451 | child = parent; | |
1452 | parent = parent->parent; | |
1453 | if (parent) | |
1454 | goto up; | |
eb755805 | 1455 | out_unlock: |
c09595f6 | 1456 | rcu_read_unlock(); |
eb755805 PZ |
1457 | |
1458 | return ret; | |
c09595f6 PZ |
1459 | } |
1460 | ||
eb755805 PZ |
1461 | static int tg_nop(struct task_group *tg, void *data) |
1462 | { | |
1463 | return 0; | |
c09595f6 | 1464 | } |
eb755805 PZ |
1465 | #endif |
1466 | ||
1467 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1468 | /* Used instead of source_load when we know the type == 0 */ |
1469 | static unsigned long weighted_cpuload(const int cpu) | |
1470 | { | |
1471 | return cpu_rq(cpu)->load.weight; | |
1472 | } | |
1473 | ||
1474 | /* | |
1475 | * Return a low guess at the load of a migration-source cpu weighted | |
1476 | * according to the scheduling class and "nice" value. | |
1477 | * | |
1478 | * We want to under-estimate the load of migration sources, to | |
1479 | * balance conservatively. | |
1480 | */ | |
1481 | static unsigned long source_load(int cpu, int type) | |
1482 | { | |
1483 | struct rq *rq = cpu_rq(cpu); | |
1484 | unsigned long total = weighted_cpuload(cpu); | |
1485 | ||
1486 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1487 | return total; | |
1488 | ||
1489 | return min(rq->cpu_load[type-1], total); | |
1490 | } | |
1491 | ||
1492 | /* | |
1493 | * Return a high guess at the load of a migration-target cpu weighted | |
1494 | * according to the scheduling class and "nice" value. | |
1495 | */ | |
1496 | static unsigned long target_load(int cpu, int type) | |
1497 | { | |
1498 | struct rq *rq = cpu_rq(cpu); | |
1499 | unsigned long total = weighted_cpuload(cpu); | |
1500 | ||
1501 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1502 | return total; | |
1503 | ||
1504 | return max(rq->cpu_load[type-1], total); | |
1505 | } | |
1506 | ||
ae154be1 PZ |
1507 | static unsigned long power_of(int cpu) |
1508 | { | |
e51fd5e2 | 1509 | return cpu_rq(cpu)->cpu_power; |
ae154be1 PZ |
1510 | } |
1511 | ||
eb755805 PZ |
1512 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1513 | ||
1514 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1515 | { | |
1516 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1517 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1518 | |
4cd42620 SR |
1519 | if (nr_running) |
1520 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1521 | else |
1522 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1523 | |
1524 | return rq->avg_load_per_task; | |
1525 | } | |
1526 | ||
1527 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1528 | |
43cf38eb | 1529 | static __read_mostly unsigned long __percpu *update_shares_data; |
34d76c41 | 1530 | |
c09595f6 PZ |
1531 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1532 | ||
1533 | /* | |
1534 | * Calculate and set the cpu's group shares. | |
1535 | */ | |
34d76c41 PZ |
1536 | static void update_group_shares_cpu(struct task_group *tg, int cpu, |
1537 | unsigned long sd_shares, | |
1538 | unsigned long sd_rq_weight, | |
4a6cc4bd | 1539 | unsigned long *usd_rq_weight) |
18d95a28 | 1540 | { |
34d76c41 | 1541 | unsigned long shares, rq_weight; |
a5004278 | 1542 | int boost = 0; |
c09595f6 | 1543 | |
4a6cc4bd | 1544 | rq_weight = usd_rq_weight[cpu]; |
a5004278 PZ |
1545 | if (!rq_weight) { |
1546 | boost = 1; | |
1547 | rq_weight = NICE_0_LOAD; | |
1548 | } | |
c8cba857 | 1549 | |
c09595f6 | 1550 | /* |
a8af7246 PZ |
1551 | * \Sum_j shares_j * rq_weight_i |
1552 | * shares_i = ----------------------------- | |
1553 | * \Sum_j rq_weight_j | |
c09595f6 | 1554 | */ |
ec4e0e2f | 1555 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1556 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1557 | |
ffda12a1 PZ |
1558 | if (abs(shares - tg->se[cpu]->load.weight) > |
1559 | sysctl_sched_shares_thresh) { | |
1560 | struct rq *rq = cpu_rq(cpu); | |
1561 | unsigned long flags; | |
c09595f6 | 1562 | |
05fa785c | 1563 | raw_spin_lock_irqsave(&rq->lock, flags); |
34d76c41 | 1564 | tg->cfs_rq[cpu]->rq_weight = boost ? 0 : rq_weight; |
a5004278 | 1565 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
ffda12a1 | 1566 | __set_se_shares(tg->se[cpu], shares); |
05fa785c | 1567 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
ffda12a1 | 1568 | } |
18d95a28 | 1569 | } |
c09595f6 PZ |
1570 | |
1571 | /* | |
c8cba857 PZ |
1572 | * Re-compute the task group their per cpu shares over the given domain. |
1573 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1574 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1575 | */ |
eb755805 | 1576 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1577 | { |
cd8ad40d | 1578 | unsigned long weight, rq_weight = 0, sum_weight = 0, shares = 0; |
4a6cc4bd | 1579 | unsigned long *usd_rq_weight; |
eb755805 | 1580 | struct sched_domain *sd = data; |
34d76c41 | 1581 | unsigned long flags; |
c8cba857 | 1582 | int i; |
c09595f6 | 1583 | |
34d76c41 PZ |
1584 | if (!tg->se[0]) |
1585 | return 0; | |
1586 | ||
1587 | local_irq_save(flags); | |
4a6cc4bd | 1588 | usd_rq_weight = per_cpu_ptr(update_shares_data, smp_processor_id()); |
34d76c41 | 1589 | |
758b2cdc | 1590 | for_each_cpu(i, sched_domain_span(sd)) { |
34d76c41 | 1591 | weight = tg->cfs_rq[i]->load.weight; |
4a6cc4bd | 1592 | usd_rq_weight[i] = weight; |
34d76c41 | 1593 | |
cd8ad40d | 1594 | rq_weight += weight; |
ec4e0e2f KC |
1595 | /* |
1596 | * If there are currently no tasks on the cpu pretend there | |
1597 | * is one of average load so that when a new task gets to | |
1598 | * run here it will not get delayed by group starvation. | |
1599 | */ | |
ec4e0e2f KC |
1600 | if (!weight) |
1601 | weight = NICE_0_LOAD; | |
1602 | ||
cd8ad40d | 1603 | sum_weight += weight; |
c8cba857 | 1604 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1605 | } |
c09595f6 | 1606 | |
cd8ad40d PZ |
1607 | if (!rq_weight) |
1608 | rq_weight = sum_weight; | |
1609 | ||
c8cba857 PZ |
1610 | if ((!shares && rq_weight) || shares > tg->shares) |
1611 | shares = tg->shares; | |
1612 | ||
1613 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1614 | shares = tg->shares; | |
c09595f6 | 1615 | |
758b2cdc | 1616 | for_each_cpu(i, sched_domain_span(sd)) |
4a6cc4bd | 1617 | update_group_shares_cpu(tg, i, shares, rq_weight, usd_rq_weight); |
34d76c41 PZ |
1618 | |
1619 | local_irq_restore(flags); | |
eb755805 PZ |
1620 | |
1621 | return 0; | |
c09595f6 PZ |
1622 | } |
1623 | ||
1624 | /* | |
c8cba857 PZ |
1625 | * Compute the cpu's hierarchical load factor for each task group. |
1626 | * This needs to be done in a top-down fashion because the load of a child | |
1627 | * group is a fraction of its parents load. | |
c09595f6 | 1628 | */ |
eb755805 | 1629 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1630 | { |
c8cba857 | 1631 | unsigned long load; |
eb755805 | 1632 | long cpu = (long)data; |
c09595f6 | 1633 | |
c8cba857 PZ |
1634 | if (!tg->parent) { |
1635 | load = cpu_rq(cpu)->load.weight; | |
1636 | } else { | |
1637 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1638 | load *= tg->cfs_rq[cpu]->shares; | |
1639 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1640 | } | |
c09595f6 | 1641 | |
c8cba857 | 1642 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1643 | |
eb755805 | 1644 | return 0; |
c09595f6 PZ |
1645 | } |
1646 | ||
c8cba857 | 1647 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1648 | { |
e7097159 PZ |
1649 | s64 elapsed; |
1650 | u64 now; | |
1651 | ||
1652 | if (root_task_group_empty()) | |
1653 | return; | |
1654 | ||
1655 | now = cpu_clock(raw_smp_processor_id()); | |
1656 | elapsed = now - sd->last_update; | |
2398f2c6 PZ |
1657 | |
1658 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1659 | sd->last_update = now; | |
eb755805 | 1660 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1661 | } |
4d8d595d PZ |
1662 | } |
1663 | ||
eb755805 | 1664 | static void update_h_load(long cpu) |
c09595f6 | 1665 | { |
eb755805 | 1666 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1667 | } |
1668 | ||
c09595f6 PZ |
1669 | #else |
1670 | ||
c8cba857 | 1671 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1672 | { |
1673 | } | |
1674 | ||
18d95a28 PZ |
1675 | #endif |
1676 | ||
8f45e2b5 GH |
1677 | #ifdef CONFIG_PREEMPT |
1678 | ||
b78bb868 PZ |
1679 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1680 | ||
70574a99 | 1681 | /* |
8f45e2b5 GH |
1682 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1683 | * way at the expense of forcing extra atomic operations in all | |
1684 | * invocations. This assures that the double_lock is acquired using the | |
1685 | * same underlying policy as the spinlock_t on this architecture, which | |
1686 | * reduces latency compared to the unfair variant below. However, it | |
1687 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1688 | */ |
8f45e2b5 GH |
1689 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1690 | __releases(this_rq->lock) | |
1691 | __acquires(busiest->lock) | |
1692 | __acquires(this_rq->lock) | |
1693 | { | |
05fa785c | 1694 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1695 | double_rq_lock(this_rq, busiest); |
1696 | ||
1697 | return 1; | |
1698 | } | |
1699 | ||
1700 | #else | |
1701 | /* | |
1702 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1703 | * latency by eliminating extra atomic operations when the locks are | |
1704 | * already in proper order on entry. This favors lower cpu-ids and will | |
1705 | * grant the double lock to lower cpus over higher ids under contention, | |
1706 | * regardless of entry order into the function. | |
1707 | */ | |
1708 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1709 | __releases(this_rq->lock) |
1710 | __acquires(busiest->lock) | |
1711 | __acquires(this_rq->lock) | |
1712 | { | |
1713 | int ret = 0; | |
1714 | ||
05fa785c | 1715 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1716 | if (busiest < this_rq) { |
05fa785c TG |
1717 | raw_spin_unlock(&this_rq->lock); |
1718 | raw_spin_lock(&busiest->lock); | |
1719 | raw_spin_lock_nested(&this_rq->lock, | |
1720 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1721 | ret = 1; |
1722 | } else | |
05fa785c TG |
1723 | raw_spin_lock_nested(&busiest->lock, |
1724 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1725 | } |
1726 | return ret; | |
1727 | } | |
1728 | ||
8f45e2b5 GH |
1729 | #endif /* CONFIG_PREEMPT */ |
1730 | ||
1731 | /* | |
1732 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1733 | */ | |
1734 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1735 | { | |
1736 | if (unlikely(!irqs_disabled())) { | |
1737 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1738 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1739 | BUG_ON(1); |
1740 | } | |
1741 | ||
1742 | return _double_lock_balance(this_rq, busiest); | |
1743 | } | |
1744 | ||
70574a99 AD |
1745 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1746 | __releases(busiest->lock) | |
1747 | { | |
05fa785c | 1748 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1749 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1750 | } | |
1e3c88bd PZ |
1751 | |
1752 | /* | |
1753 | * double_rq_lock - safely lock two runqueues | |
1754 | * | |
1755 | * Note this does not disable interrupts like task_rq_lock, | |
1756 | * you need to do so manually before calling. | |
1757 | */ | |
1758 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1759 | __acquires(rq1->lock) | |
1760 | __acquires(rq2->lock) | |
1761 | { | |
1762 | BUG_ON(!irqs_disabled()); | |
1763 | if (rq1 == rq2) { | |
1764 | raw_spin_lock(&rq1->lock); | |
1765 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1766 | } else { | |
1767 | if (rq1 < rq2) { | |
1768 | raw_spin_lock(&rq1->lock); | |
1769 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1770 | } else { | |
1771 | raw_spin_lock(&rq2->lock); | |
1772 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1773 | } | |
1774 | } | |
1e3c88bd PZ |
1775 | } |
1776 | ||
1777 | /* | |
1778 | * double_rq_unlock - safely unlock two runqueues | |
1779 | * | |
1780 | * Note this does not restore interrupts like task_rq_unlock, | |
1781 | * you need to do so manually after calling. | |
1782 | */ | |
1783 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1784 | __releases(rq1->lock) | |
1785 | __releases(rq2->lock) | |
1786 | { | |
1787 | raw_spin_unlock(&rq1->lock); | |
1788 | if (rq1 != rq2) | |
1789 | raw_spin_unlock(&rq2->lock); | |
1790 | else | |
1791 | __release(rq2->lock); | |
1792 | } | |
1793 | ||
18d95a28 PZ |
1794 | #endif |
1795 | ||
30432094 | 1796 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1797 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1798 | { | |
30432094 | 1799 | #ifdef CONFIG_SMP |
34e83e85 IM |
1800 | cfs_rq->shares = shares; |
1801 | #endif | |
1802 | } | |
30432094 | 1803 | #endif |
e7693a36 | 1804 | |
74f5187a | 1805 | static void calc_load_account_idle(struct rq *this_rq); |
0bcdcf28 | 1806 | static void update_sysctl(void); |
acb4a848 | 1807 | static int get_update_sysctl_factor(void); |
dce48a84 | 1808 | |
cd29fe6f PZ |
1809 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1810 | { | |
1811 | set_task_rq(p, cpu); | |
1812 | #ifdef CONFIG_SMP | |
1813 | /* | |
1814 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1815 | * successfuly executed on another CPU. We must ensure that updates of | |
1816 | * per-task data have been completed by this moment. | |
1817 | */ | |
1818 | smp_wmb(); | |
1819 | task_thread_info(p)->cpu = cpu; | |
1820 | #endif | |
1821 | } | |
dce48a84 | 1822 | |
1e3c88bd | 1823 | static const struct sched_class rt_sched_class; |
dd41f596 IM |
1824 | |
1825 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1826 | #define for_each_class(class) \ |
1827 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1828 | |
1e3c88bd PZ |
1829 | #include "sched_stats.h" |
1830 | ||
c09595f6 | 1831 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1832 | { |
1833 | rq->nr_running++; | |
9c217245 IM |
1834 | } |
1835 | ||
c09595f6 | 1836 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1837 | { |
1838 | rq->nr_running--; | |
9c217245 IM |
1839 | } |
1840 | ||
45bf76df IM |
1841 | static void set_load_weight(struct task_struct *p) |
1842 | { | |
1843 | if (task_has_rt_policy(p)) { | |
e51fd5e2 PZ |
1844 | p->se.load.weight = 0; |
1845 | p->se.load.inv_weight = WMULT_CONST; | |
dd41f596 IM |
1846 | return; |
1847 | } | |
45bf76df | 1848 | |
dd41f596 IM |
1849 | /* |
1850 | * SCHED_IDLE tasks get minimal weight: | |
1851 | */ | |
1852 | if (p->policy == SCHED_IDLE) { | |
1853 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1854 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1855 | return; | |
1856 | } | |
71f8bd46 | 1857 | |
dd41f596 IM |
1858 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1859 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1860 | } |
1861 | ||
371fd7e7 | 1862 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1863 | { |
a64692a3 | 1864 | update_rq_clock(rq); |
dd41f596 | 1865 | sched_info_queued(p); |
371fd7e7 | 1866 | p->sched_class->enqueue_task(rq, p, flags); |
dd41f596 | 1867 | p->se.on_rq = 1; |
71f8bd46 IM |
1868 | } |
1869 | ||
371fd7e7 | 1870 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1871 | { |
a64692a3 | 1872 | update_rq_clock(rq); |
46ac22ba | 1873 | sched_info_dequeued(p); |
371fd7e7 | 1874 | p->sched_class->dequeue_task(rq, p, flags); |
dd41f596 | 1875 | p->se.on_rq = 0; |
71f8bd46 IM |
1876 | } |
1877 | ||
1e3c88bd PZ |
1878 | /* |
1879 | * activate_task - move a task to the runqueue. | |
1880 | */ | |
371fd7e7 | 1881 | static void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1882 | { |
1883 | if (task_contributes_to_load(p)) | |
1884 | rq->nr_uninterruptible--; | |
1885 | ||
371fd7e7 | 1886 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
1887 | inc_nr_running(rq); |
1888 | } | |
1889 | ||
1890 | /* | |
1891 | * deactivate_task - remove a task from the runqueue. | |
1892 | */ | |
371fd7e7 | 1893 | static void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1894 | { |
1895 | if (task_contributes_to_load(p)) | |
1896 | rq->nr_uninterruptible++; | |
1897 | ||
371fd7e7 | 1898 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1899 | dec_nr_running(rq); |
1900 | } | |
1901 | ||
1902 | #include "sched_idletask.c" | |
1903 | #include "sched_fair.c" | |
1904 | #include "sched_rt.c" | |
1905 | #ifdef CONFIG_SCHED_DEBUG | |
1906 | # include "sched_debug.c" | |
1907 | #endif | |
1908 | ||
14531189 | 1909 | /* |
dd41f596 | 1910 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1911 | */ |
14531189 IM |
1912 | static inline int __normal_prio(struct task_struct *p) |
1913 | { | |
dd41f596 | 1914 | return p->static_prio; |
14531189 IM |
1915 | } |
1916 | ||
b29739f9 IM |
1917 | /* |
1918 | * Calculate the expected normal priority: i.e. priority | |
1919 | * without taking RT-inheritance into account. Might be | |
1920 | * boosted by interactivity modifiers. Changes upon fork, | |
1921 | * setprio syscalls, and whenever the interactivity | |
1922 | * estimator recalculates. | |
1923 | */ | |
36c8b586 | 1924 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1925 | { |
1926 | int prio; | |
1927 | ||
e05606d3 | 1928 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1929 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1930 | else | |
1931 | prio = __normal_prio(p); | |
1932 | return prio; | |
1933 | } | |
1934 | ||
1935 | /* | |
1936 | * Calculate the current priority, i.e. the priority | |
1937 | * taken into account by the scheduler. This value might | |
1938 | * be boosted by RT tasks, or might be boosted by | |
1939 | * interactivity modifiers. Will be RT if the task got | |
1940 | * RT-boosted. If not then it returns p->normal_prio. | |
1941 | */ | |
36c8b586 | 1942 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1943 | { |
1944 | p->normal_prio = normal_prio(p); | |
1945 | /* | |
1946 | * If we are RT tasks or we were boosted to RT priority, | |
1947 | * keep the priority unchanged. Otherwise, update priority | |
1948 | * to the normal priority: | |
1949 | */ | |
1950 | if (!rt_prio(p->prio)) | |
1951 | return p->normal_prio; | |
1952 | return p->prio; | |
1953 | } | |
1954 | ||
1da177e4 LT |
1955 | /** |
1956 | * task_curr - is this task currently executing on a CPU? | |
1957 | * @p: the task in question. | |
1958 | */ | |
36c8b586 | 1959 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1960 | { |
1961 | return cpu_curr(task_cpu(p)) == p; | |
1962 | } | |
1963 | ||
cb469845 SR |
1964 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1965 | const struct sched_class *prev_class, | |
1966 | int oldprio, int running) | |
1967 | { | |
1968 | if (prev_class != p->sched_class) { | |
1969 | if (prev_class->switched_from) | |
1970 | prev_class->switched_from(rq, p, running); | |
1971 | p->sched_class->switched_to(rq, p, running); | |
1972 | } else | |
1973 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1974 | } | |
1975 | ||
1da177e4 | 1976 | #ifdef CONFIG_SMP |
cc367732 IM |
1977 | /* |
1978 | * Is this task likely cache-hot: | |
1979 | */ | |
e7693a36 | 1980 | static int |
cc367732 IM |
1981 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1982 | { | |
1983 | s64 delta; | |
1984 | ||
e6c8fba7 PZ |
1985 | if (p->sched_class != &fair_sched_class) |
1986 | return 0; | |
1987 | ||
f540a608 IM |
1988 | /* |
1989 | * Buddy candidates are cache hot: | |
1990 | */ | |
f685ceac | 1991 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
1992 | (&p->se == cfs_rq_of(&p->se)->next || |
1993 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
1994 | return 1; |
1995 | ||
6bc1665b IM |
1996 | if (sysctl_sched_migration_cost == -1) |
1997 | return 1; | |
1998 | if (sysctl_sched_migration_cost == 0) | |
1999 | return 0; | |
2000 | ||
cc367732 IM |
2001 | delta = now - p->se.exec_start; |
2002 | ||
2003 | return delta < (s64)sysctl_sched_migration_cost; | |
2004 | } | |
2005 | ||
dd41f596 | 2006 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2007 | { |
e2912009 PZ |
2008 | #ifdef CONFIG_SCHED_DEBUG |
2009 | /* | |
2010 | * We should never call set_task_cpu() on a blocked task, | |
2011 | * ttwu() will sort out the placement. | |
2012 | */ | |
077614ee PZ |
2013 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
2014 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
e2912009 PZ |
2015 | #endif |
2016 | ||
de1d7286 | 2017 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2018 | |
0c69774e PZ |
2019 | if (task_cpu(p) != new_cpu) { |
2020 | p->se.nr_migrations++; | |
2021 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); | |
2022 | } | |
dd41f596 IM |
2023 | |
2024 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2025 | } |
2026 | ||
969c7921 | 2027 | struct migration_arg { |
36c8b586 | 2028 | struct task_struct *task; |
1da177e4 | 2029 | int dest_cpu; |
70b97a7f | 2030 | }; |
1da177e4 | 2031 | |
969c7921 TH |
2032 | static int migration_cpu_stop(void *data); |
2033 | ||
1da177e4 LT |
2034 | /* |
2035 | * The task's runqueue lock must be held. | |
2036 | * Returns true if you have to wait for migration thread. | |
2037 | */ | |
969c7921 | 2038 | static bool migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2039 | { |
70b97a7f | 2040 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2041 | |
2042 | /* | |
2043 | * If the task is not on a runqueue (and not running), then | |
e2912009 | 2044 | * the next wake-up will properly place the task. |
1da177e4 | 2045 | */ |
969c7921 | 2046 | return p->se.on_rq || task_running(rq, p); |
1da177e4 LT |
2047 | } |
2048 | ||
2049 | /* | |
2050 | * wait_task_inactive - wait for a thread to unschedule. | |
2051 | * | |
85ba2d86 RM |
2052 | * If @match_state is nonzero, it's the @p->state value just checked and |
2053 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2054 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2055 | * we return a positive number (its total switch count). If a second call | |
2056 | * a short while later returns the same number, the caller can be sure that | |
2057 | * @p has remained unscheduled the whole time. | |
2058 | * | |
1da177e4 LT |
2059 | * The caller must ensure that the task *will* unschedule sometime soon, |
2060 | * else this function might spin for a *long* time. This function can't | |
2061 | * be called with interrupts off, or it may introduce deadlock with | |
2062 | * smp_call_function() if an IPI is sent by the same process we are | |
2063 | * waiting to become inactive. | |
2064 | */ | |
85ba2d86 | 2065 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2066 | { |
2067 | unsigned long flags; | |
dd41f596 | 2068 | int running, on_rq; |
85ba2d86 | 2069 | unsigned long ncsw; |
70b97a7f | 2070 | struct rq *rq; |
1da177e4 | 2071 | |
3a5c359a AK |
2072 | for (;;) { |
2073 | /* | |
2074 | * We do the initial early heuristics without holding | |
2075 | * any task-queue locks at all. We'll only try to get | |
2076 | * the runqueue lock when things look like they will | |
2077 | * work out! | |
2078 | */ | |
2079 | rq = task_rq(p); | |
fa490cfd | 2080 | |
3a5c359a AK |
2081 | /* |
2082 | * If the task is actively running on another CPU | |
2083 | * still, just relax and busy-wait without holding | |
2084 | * any locks. | |
2085 | * | |
2086 | * NOTE! Since we don't hold any locks, it's not | |
2087 | * even sure that "rq" stays as the right runqueue! | |
2088 | * But we don't care, since "task_running()" will | |
2089 | * return false if the runqueue has changed and p | |
2090 | * is actually now running somewhere else! | |
2091 | */ | |
85ba2d86 RM |
2092 | while (task_running(rq, p)) { |
2093 | if (match_state && unlikely(p->state != match_state)) | |
2094 | return 0; | |
3a5c359a | 2095 | cpu_relax(); |
85ba2d86 | 2096 | } |
fa490cfd | 2097 | |
3a5c359a AK |
2098 | /* |
2099 | * Ok, time to look more closely! We need the rq | |
2100 | * lock now, to be *sure*. If we're wrong, we'll | |
2101 | * just go back and repeat. | |
2102 | */ | |
2103 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 2104 | trace_sched_wait_task(p); |
3a5c359a AK |
2105 | running = task_running(rq, p); |
2106 | on_rq = p->se.on_rq; | |
85ba2d86 | 2107 | ncsw = 0; |
f31e11d8 | 2108 | if (!match_state || p->state == match_state) |
93dcf55f | 2109 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2110 | task_rq_unlock(rq, &flags); |
fa490cfd | 2111 | |
85ba2d86 RM |
2112 | /* |
2113 | * If it changed from the expected state, bail out now. | |
2114 | */ | |
2115 | if (unlikely(!ncsw)) | |
2116 | break; | |
2117 | ||
3a5c359a AK |
2118 | /* |
2119 | * Was it really running after all now that we | |
2120 | * checked with the proper locks actually held? | |
2121 | * | |
2122 | * Oops. Go back and try again.. | |
2123 | */ | |
2124 | if (unlikely(running)) { | |
2125 | cpu_relax(); | |
2126 | continue; | |
2127 | } | |
fa490cfd | 2128 | |
3a5c359a AK |
2129 | /* |
2130 | * It's not enough that it's not actively running, | |
2131 | * it must be off the runqueue _entirely_, and not | |
2132 | * preempted! | |
2133 | * | |
80dd99b3 | 2134 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2135 | * running right now), it's preempted, and we should |
2136 | * yield - it could be a while. | |
2137 | */ | |
2138 | if (unlikely(on_rq)) { | |
2139 | schedule_timeout_uninterruptible(1); | |
2140 | continue; | |
2141 | } | |
fa490cfd | 2142 | |
3a5c359a AK |
2143 | /* |
2144 | * Ahh, all good. It wasn't running, and it wasn't | |
2145 | * runnable, which means that it will never become | |
2146 | * running in the future either. We're all done! | |
2147 | */ | |
2148 | break; | |
2149 | } | |
85ba2d86 RM |
2150 | |
2151 | return ncsw; | |
1da177e4 LT |
2152 | } |
2153 | ||
2154 | /*** | |
2155 | * kick_process - kick a running thread to enter/exit the kernel | |
2156 | * @p: the to-be-kicked thread | |
2157 | * | |
2158 | * Cause a process which is running on another CPU to enter | |
2159 | * kernel-mode, without any delay. (to get signals handled.) | |
2160 | * | |
2161 | * NOTE: this function doesnt have to take the runqueue lock, | |
2162 | * because all it wants to ensure is that the remote task enters | |
2163 | * the kernel. If the IPI races and the task has been migrated | |
2164 | * to another CPU then no harm is done and the purpose has been | |
2165 | * achieved as well. | |
2166 | */ | |
36c8b586 | 2167 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2168 | { |
2169 | int cpu; | |
2170 | ||
2171 | preempt_disable(); | |
2172 | cpu = task_cpu(p); | |
2173 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2174 | smp_send_reschedule(cpu); | |
2175 | preempt_enable(); | |
2176 | } | |
b43e3521 | 2177 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2178 | #endif /* CONFIG_SMP */ |
1da177e4 | 2179 | |
0793a61d TG |
2180 | /** |
2181 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2182 | * @p: the task to evaluate | |
2183 | * @func: the function to be called | |
2184 | * @info: the function call argument | |
2185 | * | |
2186 | * Calls the function @func when the task is currently running. This might | |
2187 | * be on the current CPU, which just calls the function directly | |
2188 | */ | |
2189 | void task_oncpu_function_call(struct task_struct *p, | |
2190 | void (*func) (void *info), void *info) | |
2191 | { | |
2192 | int cpu; | |
2193 | ||
2194 | preempt_disable(); | |
2195 | cpu = task_cpu(p); | |
2196 | if (task_curr(p)) | |
2197 | smp_call_function_single(cpu, func, info, 1); | |
2198 | preempt_enable(); | |
2199 | } | |
2200 | ||
970b13ba | 2201 | #ifdef CONFIG_SMP |
30da688e ON |
2202 | /* |
2203 | * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held. | |
2204 | */ | |
5da9a0fb PZ |
2205 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2206 | { | |
2207 | int dest_cpu; | |
2208 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
2209 | ||
2210 | /* Look for allowed, online CPU in same node. */ | |
2211 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
2212 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
2213 | return dest_cpu; | |
2214 | ||
2215 | /* Any allowed, online CPU? */ | |
2216 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); | |
2217 | if (dest_cpu < nr_cpu_ids) | |
2218 | return dest_cpu; | |
2219 | ||
2220 | /* No more Mr. Nice Guy. */ | |
897f0b3c | 2221 | if (unlikely(dest_cpu >= nr_cpu_ids)) { |
9084bb82 | 2222 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
5da9a0fb PZ |
2223 | /* |
2224 | * Don't tell them about moving exiting tasks or | |
2225 | * kernel threads (both mm NULL), since they never | |
2226 | * leave kernel. | |
2227 | */ | |
2228 | if (p->mm && printk_ratelimit()) { | |
2229 | printk(KERN_INFO "process %d (%s) no " | |
2230 | "longer affine to cpu%d\n", | |
2231 | task_pid_nr(p), p->comm, cpu); | |
2232 | } | |
2233 | } | |
2234 | ||
2235 | return dest_cpu; | |
2236 | } | |
2237 | ||
e2912009 | 2238 | /* |
30da688e | 2239 | * The caller (fork, wakeup) owns TASK_WAKING, ->cpus_allowed is stable. |
e2912009 | 2240 | */ |
970b13ba | 2241 | static inline |
0017d735 | 2242 | int select_task_rq(struct rq *rq, struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 2243 | { |
0017d735 | 2244 | int cpu = p->sched_class->select_task_rq(rq, p, sd_flags, wake_flags); |
e2912009 PZ |
2245 | |
2246 | /* | |
2247 | * In order not to call set_task_cpu() on a blocking task we need | |
2248 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2249 | * cpu. | |
2250 | * | |
2251 | * Since this is common to all placement strategies, this lives here. | |
2252 | * | |
2253 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2254 | * not worry about this generic constraint ] | |
2255 | */ | |
2256 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
70f11205 | 2257 | !cpu_online(cpu))) |
5da9a0fb | 2258 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2259 | |
2260 | return cpu; | |
970b13ba | 2261 | } |
09a40af5 MG |
2262 | |
2263 | static void update_avg(u64 *avg, u64 sample) | |
2264 | { | |
2265 | s64 diff = sample - *avg; | |
2266 | *avg += diff >> 3; | |
2267 | } | |
970b13ba PZ |
2268 | #endif |
2269 | ||
1da177e4 LT |
2270 | /*** |
2271 | * try_to_wake_up - wake up a thread | |
2272 | * @p: the to-be-woken-up thread | |
2273 | * @state: the mask of task states that can be woken | |
2274 | * @sync: do a synchronous wakeup? | |
2275 | * | |
2276 | * Put it on the run-queue if it's not already there. The "current" | |
2277 | * thread is always on the run-queue (except when the actual | |
2278 | * re-schedule is in progress), and as such you're allowed to do | |
2279 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2280 | * runnable without the overhead of this. | |
2281 | * | |
2282 | * returns failure only if the task is already active. | |
2283 | */ | |
7d478721 PZ |
2284 | static int try_to_wake_up(struct task_struct *p, unsigned int state, |
2285 | int wake_flags) | |
1da177e4 | 2286 | { |
cc367732 | 2287 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 | 2288 | unsigned long flags; |
371fd7e7 | 2289 | unsigned long en_flags = ENQUEUE_WAKEUP; |
ab3b3aa5 | 2290 | struct rq *rq; |
1da177e4 | 2291 | |
e9c84311 | 2292 | this_cpu = get_cpu(); |
2398f2c6 | 2293 | |
04e2f174 | 2294 | smp_wmb(); |
ab3b3aa5 | 2295 | rq = task_rq_lock(p, &flags); |
e9c84311 | 2296 | if (!(p->state & state)) |
1da177e4 LT |
2297 | goto out; |
2298 | ||
dd41f596 | 2299 | if (p->se.on_rq) |
1da177e4 LT |
2300 | goto out_running; |
2301 | ||
2302 | cpu = task_cpu(p); | |
cc367732 | 2303 | orig_cpu = cpu; |
1da177e4 LT |
2304 | |
2305 | #ifdef CONFIG_SMP | |
2306 | if (unlikely(task_running(rq, p))) | |
2307 | goto out_activate; | |
2308 | ||
e9c84311 PZ |
2309 | /* |
2310 | * In order to handle concurrent wakeups and release the rq->lock | |
2311 | * we put the task in TASK_WAKING state. | |
eb24073b IM |
2312 | * |
2313 | * First fix up the nr_uninterruptible count: | |
e9c84311 | 2314 | */ |
cc87f76a PZ |
2315 | if (task_contributes_to_load(p)) { |
2316 | if (likely(cpu_online(orig_cpu))) | |
2317 | rq->nr_uninterruptible--; | |
2318 | else | |
2319 | this_rq()->nr_uninterruptible--; | |
2320 | } | |
e9c84311 | 2321 | p->state = TASK_WAKING; |
efbbd05a | 2322 | |
371fd7e7 | 2323 | if (p->sched_class->task_waking) { |
efbbd05a | 2324 | p->sched_class->task_waking(rq, p); |
371fd7e7 PZ |
2325 | en_flags |= ENQUEUE_WAKING; |
2326 | } | |
efbbd05a | 2327 | |
0017d735 PZ |
2328 | cpu = select_task_rq(rq, p, SD_BALANCE_WAKE, wake_flags); |
2329 | if (cpu != orig_cpu) | |
5d2f5a61 | 2330 | set_task_cpu(p, cpu); |
0017d735 | 2331 | __task_rq_unlock(rq); |
ab19cb23 | 2332 | |
0970d299 PZ |
2333 | rq = cpu_rq(cpu); |
2334 | raw_spin_lock(&rq->lock); | |
f5dc3753 | 2335 | |
0970d299 PZ |
2336 | /* |
2337 | * We migrated the task without holding either rq->lock, however | |
2338 | * since the task is not on the task list itself, nobody else | |
2339 | * will try and migrate the task, hence the rq should match the | |
2340 | * cpu we just moved it to. | |
2341 | */ | |
2342 | WARN_ON(task_cpu(p) != cpu); | |
e9c84311 | 2343 | WARN_ON(p->state != TASK_WAKING); |
1da177e4 | 2344 | |
e7693a36 GH |
2345 | #ifdef CONFIG_SCHEDSTATS |
2346 | schedstat_inc(rq, ttwu_count); | |
2347 | if (cpu == this_cpu) | |
2348 | schedstat_inc(rq, ttwu_local); | |
2349 | else { | |
2350 | struct sched_domain *sd; | |
2351 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2352 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2353 | schedstat_inc(sd, ttwu_wake_remote); |
2354 | break; | |
2355 | } | |
2356 | } | |
2357 | } | |
6d6bc0ad | 2358 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2359 | |
1da177e4 LT |
2360 | out_activate: |
2361 | #endif /* CONFIG_SMP */ | |
41acab88 | 2362 | schedstat_inc(p, se.statistics.nr_wakeups); |
7d478721 | 2363 | if (wake_flags & WF_SYNC) |
41acab88 | 2364 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
cc367732 | 2365 | if (orig_cpu != cpu) |
41acab88 | 2366 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); |
cc367732 | 2367 | if (cpu == this_cpu) |
41acab88 | 2368 | schedstat_inc(p, se.statistics.nr_wakeups_local); |
cc367732 | 2369 | else |
41acab88 | 2370 | schedstat_inc(p, se.statistics.nr_wakeups_remote); |
371fd7e7 | 2371 | activate_task(rq, p, en_flags); |
1da177e4 LT |
2372 | success = 1; |
2373 | ||
2374 | out_running: | |
27a9da65 | 2375 | trace_sched_wakeup(p, success); |
7d478721 | 2376 | check_preempt_curr(rq, p, wake_flags); |
4ae7d5ce | 2377 | |
1da177e4 | 2378 | p->state = TASK_RUNNING; |
9a897c5a | 2379 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2380 | if (p->sched_class->task_woken) |
2381 | p->sched_class->task_woken(rq, p); | |
eae0c9df MG |
2382 | |
2383 | if (unlikely(rq->idle_stamp)) { | |
2384 | u64 delta = rq->clock - rq->idle_stamp; | |
2385 | u64 max = 2*sysctl_sched_migration_cost; | |
2386 | ||
2387 | if (delta > max) | |
2388 | rq->avg_idle = max; | |
2389 | else | |
2390 | update_avg(&rq->avg_idle, delta); | |
2391 | rq->idle_stamp = 0; | |
2392 | } | |
9a897c5a | 2393 | #endif |
1da177e4 LT |
2394 | out: |
2395 | task_rq_unlock(rq, &flags); | |
e9c84311 | 2396 | put_cpu(); |
1da177e4 LT |
2397 | |
2398 | return success; | |
2399 | } | |
2400 | ||
50fa610a DH |
2401 | /** |
2402 | * wake_up_process - Wake up a specific process | |
2403 | * @p: The process to be woken up. | |
2404 | * | |
2405 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2406 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2407 | * running. | |
2408 | * | |
2409 | * It may be assumed that this function implies a write memory barrier before | |
2410 | * changing the task state if and only if any tasks are woken up. | |
2411 | */ | |
7ad5b3a5 | 2412 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2413 | { |
d9514f6c | 2414 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2415 | } |
1da177e4 LT |
2416 | EXPORT_SYMBOL(wake_up_process); |
2417 | ||
7ad5b3a5 | 2418 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2419 | { |
2420 | return try_to_wake_up(p, state, 0); | |
2421 | } | |
2422 | ||
1da177e4 LT |
2423 | /* |
2424 | * Perform scheduler related setup for a newly forked process p. | |
2425 | * p is forked by current. | |
dd41f596 IM |
2426 | * |
2427 | * __sched_fork() is basic setup used by init_idle() too: | |
2428 | */ | |
2429 | static void __sched_fork(struct task_struct *p) | |
2430 | { | |
dd41f596 IM |
2431 | p->se.exec_start = 0; |
2432 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2433 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2434 | p->se.nr_migrations = 0; |
6cfb0d5d IM |
2435 | |
2436 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 2437 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2438 | #endif |
476d139c | 2439 | |
fa717060 | 2440 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2441 | p->se.on_rq = 0; |
4a55bd5e | 2442 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2443 | |
e107be36 AK |
2444 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2445 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2446 | #endif | |
dd41f596 IM |
2447 | } |
2448 | ||
2449 | /* | |
2450 | * fork()/clone()-time setup: | |
2451 | */ | |
2452 | void sched_fork(struct task_struct *p, int clone_flags) | |
2453 | { | |
2454 | int cpu = get_cpu(); | |
2455 | ||
2456 | __sched_fork(p); | |
06b83b5f | 2457 | /* |
0017d735 | 2458 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2459 | * nobody will actually run it, and a signal or other external |
2460 | * event cannot wake it up and insert it on the runqueue either. | |
2461 | */ | |
0017d735 | 2462 | p->state = TASK_RUNNING; |
dd41f596 | 2463 | |
b9dc29e7 MG |
2464 | /* |
2465 | * Revert to default priority/policy on fork if requested. | |
2466 | */ | |
2467 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2468 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2469 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2470 | p->normal_prio = p->static_prio; |
2471 | } | |
b9dc29e7 | 2472 | |
6c697bdf MG |
2473 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2474 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2475 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2476 | set_load_weight(p); |
2477 | } | |
2478 | ||
b9dc29e7 MG |
2479 | /* |
2480 | * We don't need the reset flag anymore after the fork. It has | |
2481 | * fulfilled its duty: | |
2482 | */ | |
2483 | p->sched_reset_on_fork = 0; | |
2484 | } | |
ca94c442 | 2485 | |
f83f9ac2 PW |
2486 | /* |
2487 | * Make sure we do not leak PI boosting priority to the child. | |
2488 | */ | |
2489 | p->prio = current->normal_prio; | |
2490 | ||
2ddbf952 HS |
2491 | if (!rt_prio(p->prio)) |
2492 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2493 | |
cd29fe6f PZ |
2494 | if (p->sched_class->task_fork) |
2495 | p->sched_class->task_fork(p); | |
2496 | ||
86951599 PZ |
2497 | /* |
2498 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2499 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2500 | * is ran before sched_fork(). | |
2501 | * | |
2502 | * Silence PROVE_RCU. | |
2503 | */ | |
2504 | rcu_read_lock(); | |
5f3edc1b | 2505 | set_task_cpu(p, cpu); |
86951599 | 2506 | rcu_read_unlock(); |
5f3edc1b | 2507 | |
52f17b6c | 2508 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2509 | if (likely(sched_info_on())) |
52f17b6c | 2510 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2511 | #endif |
d6077cb8 | 2512 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2513 | p->oncpu = 0; |
2514 | #endif | |
1da177e4 | 2515 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2516 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2517 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2518 | #endif |
917b627d GH |
2519 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2520 | ||
476d139c | 2521 | put_cpu(); |
1da177e4 LT |
2522 | } |
2523 | ||
2524 | /* | |
2525 | * wake_up_new_task - wake up a newly created task for the first time. | |
2526 | * | |
2527 | * This function will do some initial scheduler statistics housekeeping | |
2528 | * that must be done for every newly created context, then puts the task | |
2529 | * on the runqueue and wakes it. | |
2530 | */ | |
7ad5b3a5 | 2531 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2532 | { |
2533 | unsigned long flags; | |
dd41f596 | 2534 | struct rq *rq; |
c890692b | 2535 | int cpu __maybe_unused = get_cpu(); |
fabf318e PZ |
2536 | |
2537 | #ifdef CONFIG_SMP | |
0017d735 PZ |
2538 | rq = task_rq_lock(p, &flags); |
2539 | p->state = TASK_WAKING; | |
2540 | ||
fabf318e PZ |
2541 | /* |
2542 | * Fork balancing, do it here and not earlier because: | |
2543 | * - cpus_allowed can change in the fork path | |
2544 | * - any previously selected cpu might disappear through hotplug | |
2545 | * | |
0017d735 PZ |
2546 | * We set TASK_WAKING so that select_task_rq() can drop rq->lock |
2547 | * without people poking at ->cpus_allowed. | |
fabf318e | 2548 | */ |
0017d735 | 2549 | cpu = select_task_rq(rq, p, SD_BALANCE_FORK, 0); |
fabf318e | 2550 | set_task_cpu(p, cpu); |
1da177e4 | 2551 | |
06b83b5f | 2552 | p->state = TASK_RUNNING; |
0017d735 PZ |
2553 | task_rq_unlock(rq, &flags); |
2554 | #endif | |
2555 | ||
2556 | rq = task_rq_lock(p, &flags); | |
cd29fe6f | 2557 | activate_task(rq, p, 0); |
27a9da65 | 2558 | trace_sched_wakeup_new(p, 1); |
a7558e01 | 2559 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2560 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2561 | if (p->sched_class->task_woken) |
2562 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2563 | #endif |
dd41f596 | 2564 | task_rq_unlock(rq, &flags); |
fabf318e | 2565 | put_cpu(); |
1da177e4 LT |
2566 | } |
2567 | ||
e107be36 AK |
2568 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2569 | ||
2570 | /** | |
80dd99b3 | 2571 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2572 | * @notifier: notifier struct to register |
e107be36 AK |
2573 | */ |
2574 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2575 | { | |
2576 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2577 | } | |
2578 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2579 | ||
2580 | /** | |
2581 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2582 | * @notifier: notifier struct to unregister |
e107be36 AK |
2583 | * |
2584 | * This is safe to call from within a preemption notifier. | |
2585 | */ | |
2586 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2587 | { | |
2588 | hlist_del(¬ifier->link); | |
2589 | } | |
2590 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2591 | ||
2592 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2593 | { | |
2594 | struct preempt_notifier *notifier; | |
2595 | struct hlist_node *node; | |
2596 | ||
2597 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2598 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2599 | } | |
2600 | ||
2601 | static void | |
2602 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2603 | struct task_struct *next) | |
2604 | { | |
2605 | struct preempt_notifier *notifier; | |
2606 | struct hlist_node *node; | |
2607 | ||
2608 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2609 | notifier->ops->sched_out(notifier, next); | |
2610 | } | |
2611 | ||
6d6bc0ad | 2612 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2613 | |
2614 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2615 | { | |
2616 | } | |
2617 | ||
2618 | static void | |
2619 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2620 | struct task_struct *next) | |
2621 | { | |
2622 | } | |
2623 | ||
6d6bc0ad | 2624 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2625 | |
4866cde0 NP |
2626 | /** |
2627 | * prepare_task_switch - prepare to switch tasks | |
2628 | * @rq: the runqueue preparing to switch | |
421cee29 | 2629 | * @prev: the current task that is being switched out |
4866cde0 NP |
2630 | * @next: the task we are going to switch to. |
2631 | * | |
2632 | * This is called with the rq lock held and interrupts off. It must | |
2633 | * be paired with a subsequent finish_task_switch after the context | |
2634 | * switch. | |
2635 | * | |
2636 | * prepare_task_switch sets up locking and calls architecture specific | |
2637 | * hooks. | |
2638 | */ | |
e107be36 AK |
2639 | static inline void |
2640 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2641 | struct task_struct *next) | |
4866cde0 | 2642 | { |
e107be36 | 2643 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2644 | prepare_lock_switch(rq, next); |
2645 | prepare_arch_switch(next); | |
2646 | } | |
2647 | ||
1da177e4 LT |
2648 | /** |
2649 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2650 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2651 | * @prev: the thread we just switched away from. |
2652 | * | |
4866cde0 NP |
2653 | * finish_task_switch must be called after the context switch, paired |
2654 | * with a prepare_task_switch call before the context switch. | |
2655 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2656 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2657 | * |
2658 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2659 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2660 | * with the lock held can cause deadlocks; see schedule() for |
2661 | * details.) | |
2662 | */ | |
a9957449 | 2663 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2664 | __releases(rq->lock) |
2665 | { | |
1da177e4 | 2666 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2667 | long prev_state; |
1da177e4 LT |
2668 | |
2669 | rq->prev_mm = NULL; | |
2670 | ||
2671 | /* | |
2672 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2673 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2674 | * schedule one last time. The schedule call will never return, and |
2675 | * the scheduled task must drop that reference. | |
c394cc9f | 2676 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2677 | * still held, otherwise prev could be scheduled on another cpu, die |
2678 | * there before we look at prev->state, and then the reference would | |
2679 | * be dropped twice. | |
2680 | * Manfred Spraul <manfred@colorfullife.com> | |
2681 | */ | |
55a101f8 | 2682 | prev_state = prev->state; |
4866cde0 | 2683 | finish_arch_switch(prev); |
8381f65d JI |
2684 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2685 | local_irq_disable(); | |
2686 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
49f47433 | 2687 | perf_event_task_sched_in(current); |
8381f65d JI |
2688 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2689 | local_irq_enable(); | |
2690 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 2691 | finish_lock_switch(rq, prev); |
e8fa1362 | 2692 | |
e107be36 | 2693 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2694 | if (mm) |
2695 | mmdrop(mm); | |
c394cc9f | 2696 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2697 | /* |
2698 | * Remove function-return probe instances associated with this | |
2699 | * task and put them back on the free list. | |
9761eea8 | 2700 | */ |
c6fd91f0 | 2701 | kprobe_flush_task(prev); |
1da177e4 | 2702 | put_task_struct(prev); |
c6fd91f0 | 2703 | } |
1da177e4 LT |
2704 | } |
2705 | ||
3f029d3c GH |
2706 | #ifdef CONFIG_SMP |
2707 | ||
2708 | /* assumes rq->lock is held */ | |
2709 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2710 | { | |
2711 | if (prev->sched_class->pre_schedule) | |
2712 | prev->sched_class->pre_schedule(rq, prev); | |
2713 | } | |
2714 | ||
2715 | /* rq->lock is NOT held, but preemption is disabled */ | |
2716 | static inline void post_schedule(struct rq *rq) | |
2717 | { | |
2718 | if (rq->post_schedule) { | |
2719 | unsigned long flags; | |
2720 | ||
05fa785c | 2721 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2722 | if (rq->curr->sched_class->post_schedule) |
2723 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2724 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2725 | |
2726 | rq->post_schedule = 0; | |
2727 | } | |
2728 | } | |
2729 | ||
2730 | #else | |
da19ab51 | 2731 | |
3f029d3c GH |
2732 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2733 | { | |
2734 | } | |
2735 | ||
2736 | static inline void post_schedule(struct rq *rq) | |
2737 | { | |
1da177e4 LT |
2738 | } |
2739 | ||
3f029d3c GH |
2740 | #endif |
2741 | ||
1da177e4 LT |
2742 | /** |
2743 | * schedule_tail - first thing a freshly forked thread must call. | |
2744 | * @prev: the thread we just switched away from. | |
2745 | */ | |
36c8b586 | 2746 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2747 | __releases(rq->lock) |
2748 | { | |
70b97a7f IM |
2749 | struct rq *rq = this_rq(); |
2750 | ||
4866cde0 | 2751 | finish_task_switch(rq, prev); |
da19ab51 | 2752 | |
3f029d3c GH |
2753 | /* |
2754 | * FIXME: do we need to worry about rq being invalidated by the | |
2755 | * task_switch? | |
2756 | */ | |
2757 | post_schedule(rq); | |
70b97a7f | 2758 | |
4866cde0 NP |
2759 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2760 | /* In this case, finish_task_switch does not reenable preemption */ | |
2761 | preempt_enable(); | |
2762 | #endif | |
1da177e4 | 2763 | if (current->set_child_tid) |
b488893a | 2764 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2765 | } |
2766 | ||
2767 | /* | |
2768 | * context_switch - switch to the new MM and the new | |
2769 | * thread's register state. | |
2770 | */ | |
dd41f596 | 2771 | static inline void |
70b97a7f | 2772 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2773 | struct task_struct *next) |
1da177e4 | 2774 | { |
dd41f596 | 2775 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2776 | |
e107be36 | 2777 | prepare_task_switch(rq, prev, next); |
27a9da65 | 2778 | trace_sched_switch(prev, next); |
dd41f596 IM |
2779 | mm = next->mm; |
2780 | oldmm = prev->active_mm; | |
9226d125 ZA |
2781 | /* |
2782 | * For paravirt, this is coupled with an exit in switch_to to | |
2783 | * combine the page table reload and the switch backend into | |
2784 | * one hypercall. | |
2785 | */ | |
224101ed | 2786 | arch_start_context_switch(prev); |
9226d125 | 2787 | |
710390d9 | 2788 | if (likely(!mm)) { |
1da177e4 LT |
2789 | next->active_mm = oldmm; |
2790 | atomic_inc(&oldmm->mm_count); | |
2791 | enter_lazy_tlb(oldmm, next); | |
2792 | } else | |
2793 | switch_mm(oldmm, mm, next); | |
2794 | ||
710390d9 | 2795 | if (likely(!prev->mm)) { |
1da177e4 | 2796 | prev->active_mm = NULL; |
1da177e4 LT |
2797 | rq->prev_mm = oldmm; |
2798 | } | |
3a5f5e48 IM |
2799 | /* |
2800 | * Since the runqueue lock will be released by the next | |
2801 | * task (which is an invalid locking op but in the case | |
2802 | * of the scheduler it's an obvious special-case), so we | |
2803 | * do an early lockdep release here: | |
2804 | */ | |
2805 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2806 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2807 | #endif |
1da177e4 LT |
2808 | |
2809 | /* Here we just switch the register state and the stack. */ | |
2810 | switch_to(prev, next, prev); | |
2811 | ||
dd41f596 IM |
2812 | barrier(); |
2813 | /* | |
2814 | * this_rq must be evaluated again because prev may have moved | |
2815 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2816 | * frame will be invalid. | |
2817 | */ | |
2818 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2819 | } |
2820 | ||
2821 | /* | |
2822 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2823 | * | |
2824 | * externally visible scheduler statistics: current number of runnable | |
2825 | * threads, current number of uninterruptible-sleeping threads, total | |
2826 | * number of context switches performed since bootup. | |
2827 | */ | |
2828 | unsigned long nr_running(void) | |
2829 | { | |
2830 | unsigned long i, sum = 0; | |
2831 | ||
2832 | for_each_online_cpu(i) | |
2833 | sum += cpu_rq(i)->nr_running; | |
2834 | ||
2835 | return sum; | |
f711f609 | 2836 | } |
1da177e4 LT |
2837 | |
2838 | unsigned long nr_uninterruptible(void) | |
f711f609 | 2839 | { |
1da177e4 | 2840 | unsigned long i, sum = 0; |
f711f609 | 2841 | |
0a945022 | 2842 | for_each_possible_cpu(i) |
1da177e4 | 2843 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
2844 | |
2845 | /* | |
1da177e4 LT |
2846 | * Since we read the counters lockless, it might be slightly |
2847 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 2848 | */ |
1da177e4 LT |
2849 | if (unlikely((long)sum < 0)) |
2850 | sum = 0; | |
f711f609 | 2851 | |
1da177e4 | 2852 | return sum; |
f711f609 | 2853 | } |
f711f609 | 2854 | |
1da177e4 | 2855 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2856 | { |
cc94abfc SR |
2857 | int i; |
2858 | unsigned long long sum = 0; | |
46cb4b7c | 2859 | |
0a945022 | 2860 | for_each_possible_cpu(i) |
1da177e4 | 2861 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2862 | |
1da177e4 LT |
2863 | return sum; |
2864 | } | |
483b4ee6 | 2865 | |
1da177e4 LT |
2866 | unsigned long nr_iowait(void) |
2867 | { | |
2868 | unsigned long i, sum = 0; | |
483b4ee6 | 2869 | |
0a945022 | 2870 | for_each_possible_cpu(i) |
1da177e4 | 2871 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2872 | |
1da177e4 LT |
2873 | return sum; |
2874 | } | |
483b4ee6 | 2875 | |
8c215bd3 | 2876 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2877 | { |
8c215bd3 | 2878 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2879 | return atomic_read(&this->nr_iowait); |
2880 | } | |
46cb4b7c | 2881 | |
69d25870 AV |
2882 | unsigned long this_cpu_load(void) |
2883 | { | |
2884 | struct rq *this = this_rq(); | |
2885 | return this->cpu_load[0]; | |
2886 | } | |
e790fb0b | 2887 | |
46cb4b7c | 2888 | |
dce48a84 TG |
2889 | /* Variables and functions for calc_load */ |
2890 | static atomic_long_t calc_load_tasks; | |
2891 | static unsigned long calc_load_update; | |
2892 | unsigned long avenrun[3]; | |
2893 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 2894 | |
74f5187a PZ |
2895 | static long calc_load_fold_active(struct rq *this_rq) |
2896 | { | |
2897 | long nr_active, delta = 0; | |
2898 | ||
2899 | nr_active = this_rq->nr_running; | |
2900 | nr_active += (long) this_rq->nr_uninterruptible; | |
2901 | ||
2902 | if (nr_active != this_rq->calc_load_active) { | |
2903 | delta = nr_active - this_rq->calc_load_active; | |
2904 | this_rq->calc_load_active = nr_active; | |
2905 | } | |
2906 | ||
2907 | return delta; | |
2908 | } | |
2909 | ||
2910 | #ifdef CONFIG_NO_HZ | |
2911 | /* | |
2912 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | |
2913 | * | |
2914 | * When making the ILB scale, we should try to pull this in as well. | |
2915 | */ | |
2916 | static atomic_long_t calc_load_tasks_idle; | |
2917 | ||
2918 | static void calc_load_account_idle(struct rq *this_rq) | |
2919 | { | |
2920 | long delta; | |
2921 | ||
2922 | delta = calc_load_fold_active(this_rq); | |
2923 | if (delta) | |
2924 | atomic_long_add(delta, &calc_load_tasks_idle); | |
2925 | } | |
2926 | ||
2927 | static long calc_load_fold_idle(void) | |
2928 | { | |
2929 | long delta = 0; | |
2930 | ||
2931 | /* | |
2932 | * Its got a race, we don't care... | |
2933 | */ | |
2934 | if (atomic_long_read(&calc_load_tasks_idle)) | |
2935 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | |
2936 | ||
2937 | return delta; | |
2938 | } | |
2939 | #else | |
2940 | static void calc_load_account_idle(struct rq *this_rq) | |
2941 | { | |
2942 | } | |
2943 | ||
2944 | static inline long calc_load_fold_idle(void) | |
2945 | { | |
2946 | return 0; | |
2947 | } | |
2948 | #endif | |
2949 | ||
2d02494f TG |
2950 | /** |
2951 | * get_avenrun - get the load average array | |
2952 | * @loads: pointer to dest load array | |
2953 | * @offset: offset to add | |
2954 | * @shift: shift count to shift the result left | |
2955 | * | |
2956 | * These values are estimates at best, so no need for locking. | |
2957 | */ | |
2958 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2959 | { | |
2960 | loads[0] = (avenrun[0] + offset) << shift; | |
2961 | loads[1] = (avenrun[1] + offset) << shift; | |
2962 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 2963 | } |
46cb4b7c | 2964 | |
dce48a84 TG |
2965 | static unsigned long |
2966 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
db1b1fef | 2967 | { |
dce48a84 TG |
2968 | load *= exp; |
2969 | load += active * (FIXED_1 - exp); | |
2970 | return load >> FSHIFT; | |
2971 | } | |
46cb4b7c SS |
2972 | |
2973 | /* | |
dce48a84 TG |
2974 | * calc_load - update the avenrun load estimates 10 ticks after the |
2975 | * CPUs have updated calc_load_tasks. | |
7835b98b | 2976 | */ |
dce48a84 | 2977 | void calc_global_load(void) |
7835b98b | 2978 | { |
dce48a84 TG |
2979 | unsigned long upd = calc_load_update + 10; |
2980 | long active; | |
1da177e4 | 2981 | |
dce48a84 TG |
2982 | if (time_before(jiffies, upd)) |
2983 | return; | |
1da177e4 | 2984 | |
dce48a84 TG |
2985 | active = atomic_long_read(&calc_load_tasks); |
2986 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 2987 | |
dce48a84 TG |
2988 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
2989 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
2990 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 2991 | |
dce48a84 TG |
2992 | calc_load_update += LOAD_FREQ; |
2993 | } | |
1da177e4 | 2994 | |
dce48a84 | 2995 | /* |
74f5187a PZ |
2996 | * Called from update_cpu_load() to periodically update this CPU's |
2997 | * active count. | |
dce48a84 TG |
2998 | */ |
2999 | static void calc_load_account_active(struct rq *this_rq) | |
3000 | { | |
74f5187a | 3001 | long delta; |
08c183f3 | 3002 | |
74f5187a PZ |
3003 | if (time_before(jiffies, this_rq->calc_load_update)) |
3004 | return; | |
783609c6 | 3005 | |
74f5187a PZ |
3006 | delta = calc_load_fold_active(this_rq); |
3007 | delta += calc_load_fold_idle(); | |
3008 | if (delta) | |
dce48a84 | 3009 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
3010 | |
3011 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
3012 | } |
3013 | ||
3014 | /* | |
dd41f596 IM |
3015 | * Update rq->cpu_load[] statistics. This function is usually called every |
3016 | * scheduler tick (TICK_NSEC). | |
46cb4b7c | 3017 | */ |
dd41f596 | 3018 | static void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 3019 | { |
495eca49 | 3020 | unsigned long this_load = this_rq->load.weight; |
dd41f596 | 3021 | int i, scale; |
46cb4b7c | 3022 | |
dd41f596 | 3023 | this_rq->nr_load_updates++; |
46cb4b7c | 3024 | |
dd41f596 IM |
3025 | /* Update our load: */ |
3026 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
3027 | unsigned long old_load, new_load; | |
7d1e6a9b | 3028 | |
dd41f596 | 3029 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 3030 | |
dd41f596 IM |
3031 | old_load = this_rq->cpu_load[i]; |
3032 | new_load = this_load; | |
a25707f3 IM |
3033 | /* |
3034 | * Round up the averaging division if load is increasing. This | |
3035 | * prevents us from getting stuck on 9 if the load is 10, for | |
3036 | * example. | |
3037 | */ | |
3038 | if (new_load > old_load) | |
3039 | new_load += scale-1; | |
dd41f596 IM |
3040 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
3041 | } | |
46cb4b7c | 3042 | |
74f5187a | 3043 | calc_load_account_active(this_rq); |
46cb4b7c SS |
3044 | } |
3045 | ||
dd41f596 | 3046 | #ifdef CONFIG_SMP |
8a0be9ef | 3047 | |
46cb4b7c | 3048 | /* |
38022906 PZ |
3049 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3050 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3051 | */ |
38022906 | 3052 | void sched_exec(void) |
46cb4b7c | 3053 | { |
38022906 | 3054 | struct task_struct *p = current; |
1da177e4 | 3055 | unsigned long flags; |
70b97a7f | 3056 | struct rq *rq; |
0017d735 | 3057 | int dest_cpu; |
46cb4b7c | 3058 | |
1da177e4 | 3059 | rq = task_rq_lock(p, &flags); |
0017d735 PZ |
3060 | dest_cpu = p->sched_class->select_task_rq(rq, p, SD_BALANCE_EXEC, 0); |
3061 | if (dest_cpu == smp_processor_id()) | |
3062 | goto unlock; | |
38022906 | 3063 | |
46cb4b7c | 3064 | /* |
38022906 | 3065 | * select_task_rq() can race against ->cpus_allowed |
46cb4b7c | 3066 | */ |
30da688e | 3067 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed) && |
969c7921 TH |
3068 | likely(cpu_active(dest_cpu)) && migrate_task(p, dest_cpu)) { |
3069 | struct migration_arg arg = { p, dest_cpu }; | |
46cb4b7c | 3070 | |
1da177e4 | 3071 | task_rq_unlock(rq, &flags); |
969c7921 | 3072 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
3073 | return; |
3074 | } | |
0017d735 | 3075 | unlock: |
1da177e4 | 3076 | task_rq_unlock(rq, &flags); |
1da177e4 | 3077 | } |
dd41f596 | 3078 | |
1da177e4 LT |
3079 | #endif |
3080 | ||
1da177e4 LT |
3081 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3082 | ||
3083 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3084 | ||
3085 | /* | |
c5f8d995 | 3086 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 3087 | * @p in case that task is currently running. |
c5f8d995 HS |
3088 | * |
3089 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 3090 | */ |
c5f8d995 HS |
3091 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
3092 | { | |
3093 | u64 ns = 0; | |
3094 | ||
3095 | if (task_current(rq, p)) { | |
3096 | update_rq_clock(rq); | |
3097 | ns = rq->clock - p->se.exec_start; | |
3098 | if ((s64)ns < 0) | |
3099 | ns = 0; | |
3100 | } | |
3101 | ||
3102 | return ns; | |
3103 | } | |
3104 | ||
bb34d92f | 3105 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 3106 | { |
1da177e4 | 3107 | unsigned long flags; |
41b86e9c | 3108 | struct rq *rq; |
bb34d92f | 3109 | u64 ns = 0; |
48f24c4d | 3110 | |
41b86e9c | 3111 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
3112 | ns = do_task_delta_exec(p, rq); |
3113 | task_rq_unlock(rq, &flags); | |
1508487e | 3114 | |
c5f8d995 HS |
3115 | return ns; |
3116 | } | |
f06febc9 | 3117 | |
c5f8d995 HS |
3118 | /* |
3119 | * Return accounted runtime for the task. | |
3120 | * In case the task is currently running, return the runtime plus current's | |
3121 | * pending runtime that have not been accounted yet. | |
3122 | */ | |
3123 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3124 | { | |
3125 | unsigned long flags; | |
3126 | struct rq *rq; | |
3127 | u64 ns = 0; | |
3128 | ||
3129 | rq = task_rq_lock(p, &flags); | |
3130 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
3131 | task_rq_unlock(rq, &flags); | |
3132 | ||
3133 | return ns; | |
3134 | } | |
48f24c4d | 3135 | |
c5f8d995 HS |
3136 | /* |
3137 | * Return sum_exec_runtime for the thread group. | |
3138 | * In case the task is currently running, return the sum plus current's | |
3139 | * pending runtime that have not been accounted yet. | |
3140 | * | |
3141 | * Note that the thread group might have other running tasks as well, | |
3142 | * so the return value not includes other pending runtime that other | |
3143 | * running tasks might have. | |
3144 | */ | |
3145 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
3146 | { | |
3147 | struct task_cputime totals; | |
3148 | unsigned long flags; | |
3149 | struct rq *rq; | |
3150 | u64 ns; | |
3151 | ||
3152 | rq = task_rq_lock(p, &flags); | |
3153 | thread_group_cputime(p, &totals); | |
3154 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 3155 | task_rq_unlock(rq, &flags); |
48f24c4d | 3156 | |
1da177e4 LT |
3157 | return ns; |
3158 | } | |
3159 | ||
1da177e4 LT |
3160 | /* |
3161 | * Account user cpu time to a process. | |
3162 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 3163 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 3164 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 3165 | */ |
457533a7 MS |
3166 | void account_user_time(struct task_struct *p, cputime_t cputime, |
3167 | cputime_t cputime_scaled) | |
1da177e4 LT |
3168 | { |
3169 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3170 | cputime64_t tmp; | |
3171 | ||
457533a7 | 3172 | /* Add user time to process. */ |
1da177e4 | 3173 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3174 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3175 | account_group_user_time(p, cputime); |
1da177e4 LT |
3176 | |
3177 | /* Add user time to cpustat. */ | |
3178 | tmp = cputime_to_cputime64(cputime); | |
3179 | if (TASK_NICE(p) > 0) | |
3180 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3181 | else | |
3182 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
3183 | |
3184 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
3185 | /* Account for user time used */ |
3186 | acct_update_integrals(p); | |
1da177e4 LT |
3187 | } |
3188 | ||
94886b84 LV |
3189 | /* |
3190 | * Account guest cpu time to a process. | |
3191 | * @p: the process that the cpu time gets accounted to | |
3192 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 3193 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 3194 | */ |
457533a7 MS |
3195 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
3196 | cputime_t cputime_scaled) | |
94886b84 LV |
3197 | { |
3198 | cputime64_t tmp; | |
3199 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3200 | ||
3201 | tmp = cputime_to_cputime64(cputime); | |
3202 | ||
457533a7 | 3203 | /* Add guest time to process. */ |
94886b84 | 3204 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3205 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3206 | account_group_user_time(p, cputime); |
94886b84 LV |
3207 | p->gtime = cputime_add(p->gtime, cputime); |
3208 | ||
457533a7 | 3209 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
3210 | if (TASK_NICE(p) > 0) { |
3211 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3212 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
3213 | } else { | |
3214 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3215 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3216 | } | |
94886b84 LV |
3217 | } |
3218 | ||
1da177e4 LT |
3219 | /* |
3220 | * Account system cpu time to a process. | |
3221 | * @p: the process that the cpu time gets accounted to | |
3222 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3223 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 3224 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
3225 | */ |
3226 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 3227 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
3228 | { |
3229 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
3230 | cputime64_t tmp; |
3231 | ||
983ed7a6 | 3232 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 3233 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
3234 | return; |
3235 | } | |
94886b84 | 3236 | |
457533a7 | 3237 | /* Add system time to process. */ |
1da177e4 | 3238 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 3239 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 3240 | account_group_system_time(p, cputime); |
1da177e4 LT |
3241 | |
3242 | /* Add system time to cpustat. */ | |
3243 | tmp = cputime_to_cputime64(cputime); | |
3244 | if (hardirq_count() - hardirq_offset) | |
3245 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3246 | else if (softirq_count()) | |
3247 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 3248 | else |
79741dd3 MS |
3249 | cpustat->system = cputime64_add(cpustat->system, tmp); |
3250 | ||
ef12fefa BR |
3251 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
3252 | ||
1da177e4 LT |
3253 | /* Account for system time used */ |
3254 | acct_update_integrals(p); | |
1da177e4 LT |
3255 | } |
3256 | ||
c66f08be | 3257 | /* |
1da177e4 | 3258 | * Account for involuntary wait time. |
1da177e4 | 3259 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 3260 | */ |
79741dd3 | 3261 | void account_steal_time(cputime_t cputime) |
c66f08be | 3262 | { |
79741dd3 MS |
3263 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
3264 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
3265 | ||
3266 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
3267 | } |
3268 | ||
1da177e4 | 3269 | /* |
79741dd3 MS |
3270 | * Account for idle time. |
3271 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 3272 | */ |
79741dd3 | 3273 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
3274 | { |
3275 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 3276 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 3277 | struct rq *rq = this_rq(); |
1da177e4 | 3278 | |
79741dd3 MS |
3279 | if (atomic_read(&rq->nr_iowait) > 0) |
3280 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
3281 | else | |
3282 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
3283 | } |
3284 | ||
79741dd3 MS |
3285 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
3286 | ||
3287 | /* | |
3288 | * Account a single tick of cpu time. | |
3289 | * @p: the process that the cpu time gets accounted to | |
3290 | * @user_tick: indicates if the tick is a user or a system tick | |
3291 | */ | |
3292 | void account_process_tick(struct task_struct *p, int user_tick) | |
3293 | { | |
a42548a1 | 3294 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
3295 | struct rq *rq = this_rq(); |
3296 | ||
3297 | if (user_tick) | |
a42548a1 | 3298 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 3299 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 3300 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
3301 | one_jiffy_scaled); |
3302 | else | |
a42548a1 | 3303 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
3304 | } |
3305 | ||
3306 | /* | |
3307 | * Account multiple ticks of steal time. | |
3308 | * @p: the process from which the cpu time has been stolen | |
3309 | * @ticks: number of stolen ticks | |
3310 | */ | |
3311 | void account_steal_ticks(unsigned long ticks) | |
3312 | { | |
3313 | account_steal_time(jiffies_to_cputime(ticks)); | |
3314 | } | |
3315 | ||
3316 | /* | |
3317 | * Account multiple ticks of idle time. | |
3318 | * @ticks: number of stolen ticks | |
3319 | */ | |
3320 | void account_idle_ticks(unsigned long ticks) | |
3321 | { | |
3322 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
3323 | } |
3324 | ||
79741dd3 MS |
3325 | #endif |
3326 | ||
49048622 BS |
3327 | /* |
3328 | * Use precise platform statistics if available: | |
3329 | */ | |
3330 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 3331 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3332 | { |
d99ca3b9 HS |
3333 | *ut = p->utime; |
3334 | *st = p->stime; | |
49048622 BS |
3335 | } |
3336 | ||
0cf55e1e | 3337 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3338 | { |
0cf55e1e HS |
3339 | struct task_cputime cputime; |
3340 | ||
3341 | thread_group_cputime(p, &cputime); | |
3342 | ||
3343 | *ut = cputime.utime; | |
3344 | *st = cputime.stime; | |
49048622 BS |
3345 | } |
3346 | #else | |
761b1d26 HS |
3347 | |
3348 | #ifndef nsecs_to_cputime | |
b7b20df9 | 3349 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
3350 | #endif |
3351 | ||
d180c5bc | 3352 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3353 | { |
d99ca3b9 | 3354 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
3355 | |
3356 | /* | |
3357 | * Use CFS's precise accounting: | |
3358 | */ | |
d180c5bc | 3359 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
3360 | |
3361 | if (total) { | |
d180c5bc HS |
3362 | u64 temp; |
3363 | ||
3364 | temp = (u64)(rtime * utime); | |
49048622 | 3365 | do_div(temp, total); |
d180c5bc HS |
3366 | utime = (cputime_t)temp; |
3367 | } else | |
3368 | utime = rtime; | |
49048622 | 3369 | |
d180c5bc HS |
3370 | /* |
3371 | * Compare with previous values, to keep monotonicity: | |
3372 | */ | |
761b1d26 | 3373 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 3374 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 3375 | |
d99ca3b9 HS |
3376 | *ut = p->prev_utime; |
3377 | *st = p->prev_stime; | |
49048622 BS |
3378 | } |
3379 | ||
0cf55e1e HS |
3380 | /* |
3381 | * Must be called with siglock held. | |
3382 | */ | |
3383 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 3384 | { |
0cf55e1e HS |
3385 | struct signal_struct *sig = p->signal; |
3386 | struct task_cputime cputime; | |
3387 | cputime_t rtime, utime, total; | |
49048622 | 3388 | |
0cf55e1e | 3389 | thread_group_cputime(p, &cputime); |
49048622 | 3390 | |
0cf55e1e HS |
3391 | total = cputime_add(cputime.utime, cputime.stime); |
3392 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 3393 | |
0cf55e1e HS |
3394 | if (total) { |
3395 | u64 temp; | |
49048622 | 3396 | |
0cf55e1e HS |
3397 | temp = (u64)(rtime * cputime.utime); |
3398 | do_div(temp, total); | |
3399 | utime = (cputime_t)temp; | |
3400 | } else | |
3401 | utime = rtime; | |
3402 | ||
3403 | sig->prev_utime = max(sig->prev_utime, utime); | |
3404 | sig->prev_stime = max(sig->prev_stime, | |
3405 | cputime_sub(rtime, sig->prev_utime)); | |
3406 | ||
3407 | *ut = sig->prev_utime; | |
3408 | *st = sig->prev_stime; | |
49048622 | 3409 | } |
49048622 | 3410 | #endif |
49048622 | 3411 | |
7835b98b CL |
3412 | /* |
3413 | * This function gets called by the timer code, with HZ frequency. | |
3414 | * We call it with interrupts disabled. | |
3415 | * | |
3416 | * It also gets called by the fork code, when changing the parent's | |
3417 | * timeslices. | |
3418 | */ | |
3419 | void scheduler_tick(void) | |
3420 | { | |
7835b98b CL |
3421 | int cpu = smp_processor_id(); |
3422 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3423 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3424 | |
3425 | sched_clock_tick(); | |
dd41f596 | 3426 | |
05fa785c | 3427 | raw_spin_lock(&rq->lock); |
3e51f33f | 3428 | update_rq_clock(rq); |
f1a438d8 | 3429 | update_cpu_load(rq); |
fa85ae24 | 3430 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 3431 | raw_spin_unlock(&rq->lock); |
7835b98b | 3432 | |
49f47433 | 3433 | perf_event_task_tick(curr); |
e220d2dc | 3434 | |
e418e1c2 | 3435 | #ifdef CONFIG_SMP |
dd41f596 IM |
3436 | rq->idle_at_tick = idle_cpu(cpu); |
3437 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3438 | #endif |
1da177e4 LT |
3439 | } |
3440 | ||
132380a0 | 3441 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
3442 | { |
3443 | if (in_lock_functions(addr)) { | |
3444 | addr = CALLER_ADDR2; | |
3445 | if (in_lock_functions(addr)) | |
3446 | addr = CALLER_ADDR3; | |
3447 | } | |
3448 | return addr; | |
3449 | } | |
1da177e4 | 3450 | |
7e49fcce SR |
3451 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3452 | defined(CONFIG_PREEMPT_TRACER)) | |
3453 | ||
43627582 | 3454 | void __kprobes add_preempt_count(int val) |
1da177e4 | 3455 | { |
6cd8a4bb | 3456 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3457 | /* |
3458 | * Underflow? | |
3459 | */ | |
9a11b49a IM |
3460 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3461 | return; | |
6cd8a4bb | 3462 | #endif |
1da177e4 | 3463 | preempt_count() += val; |
6cd8a4bb | 3464 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3465 | /* |
3466 | * Spinlock count overflowing soon? | |
3467 | */ | |
33859f7f MOS |
3468 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3469 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
3470 | #endif |
3471 | if (preempt_count() == val) | |
3472 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3473 | } |
3474 | EXPORT_SYMBOL(add_preempt_count); | |
3475 | ||
43627582 | 3476 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 3477 | { |
6cd8a4bb | 3478 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3479 | /* |
3480 | * Underflow? | |
3481 | */ | |
01e3eb82 | 3482 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3483 | return; |
1da177e4 LT |
3484 | /* |
3485 | * Is the spinlock portion underflowing? | |
3486 | */ | |
9a11b49a IM |
3487 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3488 | !(preempt_count() & PREEMPT_MASK))) | |
3489 | return; | |
6cd8a4bb | 3490 | #endif |
9a11b49a | 3491 | |
6cd8a4bb SR |
3492 | if (preempt_count() == val) |
3493 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3494 | preempt_count() -= val; |
3495 | } | |
3496 | EXPORT_SYMBOL(sub_preempt_count); | |
3497 | ||
3498 | #endif | |
3499 | ||
3500 | /* | |
dd41f596 | 3501 | * Print scheduling while atomic bug: |
1da177e4 | 3502 | */ |
dd41f596 | 3503 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3504 | { |
838225b4 SS |
3505 | struct pt_regs *regs = get_irq_regs(); |
3506 | ||
3df0fc5b PZ |
3507 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3508 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3509 | |
dd41f596 | 3510 | debug_show_held_locks(prev); |
e21f5b15 | 3511 | print_modules(); |
dd41f596 IM |
3512 | if (irqs_disabled()) |
3513 | print_irqtrace_events(prev); | |
838225b4 SS |
3514 | |
3515 | if (regs) | |
3516 | show_regs(regs); | |
3517 | else | |
3518 | dump_stack(); | |
dd41f596 | 3519 | } |
1da177e4 | 3520 | |
dd41f596 IM |
3521 | /* |
3522 | * Various schedule()-time debugging checks and statistics: | |
3523 | */ | |
3524 | static inline void schedule_debug(struct task_struct *prev) | |
3525 | { | |
1da177e4 | 3526 | /* |
41a2d6cf | 3527 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3528 | * schedule() atomically, we ignore that path for now. |
3529 | * Otherwise, whine if we are scheduling when we should not be. | |
3530 | */ | |
3f33a7ce | 3531 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
3532 | __schedule_bug(prev); |
3533 | ||
1da177e4 LT |
3534 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3535 | ||
2d72376b | 3536 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3537 | #ifdef CONFIG_SCHEDSTATS |
3538 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3539 | schedstat_inc(this_rq(), bkl_count); |
3540 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3541 | } |
3542 | #endif | |
dd41f596 IM |
3543 | } |
3544 | ||
6cecd084 | 3545 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 3546 | { |
a64692a3 MG |
3547 | if (prev->se.on_rq) |
3548 | update_rq_clock(rq); | |
3549 | rq->skip_clock_update = 0; | |
6cecd084 | 3550 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
3551 | } |
3552 | ||
dd41f596 IM |
3553 | /* |
3554 | * Pick up the highest-prio task: | |
3555 | */ | |
3556 | static inline struct task_struct * | |
b67802ea | 3557 | pick_next_task(struct rq *rq) |
dd41f596 | 3558 | { |
5522d5d5 | 3559 | const struct sched_class *class; |
dd41f596 | 3560 | struct task_struct *p; |
1da177e4 LT |
3561 | |
3562 | /* | |
dd41f596 IM |
3563 | * Optimization: we know that if all tasks are in |
3564 | * the fair class we can call that function directly: | |
1da177e4 | 3565 | */ |
dd41f596 | 3566 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3567 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3568 | if (likely(p)) |
3569 | return p; | |
1da177e4 LT |
3570 | } |
3571 | ||
dd41f596 IM |
3572 | class = sched_class_highest; |
3573 | for ( ; ; ) { | |
fb8d4724 | 3574 | p = class->pick_next_task(rq); |
dd41f596 IM |
3575 | if (p) |
3576 | return p; | |
3577 | /* | |
3578 | * Will never be NULL as the idle class always | |
3579 | * returns a non-NULL p: | |
3580 | */ | |
3581 | class = class->next; | |
3582 | } | |
3583 | } | |
1da177e4 | 3584 | |
dd41f596 IM |
3585 | /* |
3586 | * schedule() is the main scheduler function. | |
3587 | */ | |
ff743345 | 3588 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
3589 | { |
3590 | struct task_struct *prev, *next; | |
67ca7bde | 3591 | unsigned long *switch_count; |
dd41f596 | 3592 | struct rq *rq; |
31656519 | 3593 | int cpu; |
dd41f596 | 3594 | |
ff743345 PZ |
3595 | need_resched: |
3596 | preempt_disable(); | |
dd41f596 IM |
3597 | cpu = smp_processor_id(); |
3598 | rq = cpu_rq(cpu); | |
25502a6c | 3599 | rcu_note_context_switch(cpu); |
dd41f596 IM |
3600 | prev = rq->curr; |
3601 | switch_count = &prev->nivcsw; | |
3602 | ||
3603 | release_kernel_lock(prev); | |
3604 | need_resched_nonpreemptible: | |
3605 | ||
3606 | schedule_debug(prev); | |
1da177e4 | 3607 | |
31656519 | 3608 | if (sched_feat(HRTICK)) |
f333fdc9 | 3609 | hrtick_clear(rq); |
8f4d37ec | 3610 | |
05fa785c | 3611 | raw_spin_lock_irq(&rq->lock); |
1e819950 | 3612 | clear_tsk_need_resched(prev); |
1da177e4 | 3613 | |
1da177e4 | 3614 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 3615 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 3616 | prev->state = TASK_RUNNING; |
16882c1e | 3617 | else |
371fd7e7 | 3618 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
dd41f596 | 3619 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3620 | } |
3621 | ||
3f029d3c | 3622 | pre_schedule(rq, prev); |
f65eda4f | 3623 | |
dd41f596 | 3624 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3625 | idle_balance(cpu, rq); |
1da177e4 | 3626 | |
df1c99d4 | 3627 | put_prev_task(rq, prev); |
b67802ea | 3628 | next = pick_next_task(rq); |
1da177e4 | 3629 | |
1da177e4 | 3630 | if (likely(prev != next)) { |
673a90a1 | 3631 | sched_info_switch(prev, next); |
49f47433 | 3632 | perf_event_task_sched_out(prev, next); |
673a90a1 | 3633 | |
1da177e4 LT |
3634 | rq->nr_switches++; |
3635 | rq->curr = next; | |
3636 | ++*switch_count; | |
3637 | ||
dd41f596 | 3638 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
3639 | /* |
3640 | * the context switch might have flipped the stack from under | |
3641 | * us, hence refresh the local variables. | |
3642 | */ | |
3643 | cpu = smp_processor_id(); | |
3644 | rq = cpu_rq(cpu); | |
1da177e4 | 3645 | } else |
05fa785c | 3646 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 3647 | |
3f029d3c | 3648 | post_schedule(rq); |
1da177e4 | 3649 | |
6d558c3a YZ |
3650 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
3651 | prev = rq->curr; | |
3652 | switch_count = &prev->nivcsw; | |
1da177e4 | 3653 | goto need_resched_nonpreemptible; |
6d558c3a | 3654 | } |
8f4d37ec | 3655 | |
1da177e4 | 3656 | preempt_enable_no_resched(); |
ff743345 | 3657 | if (need_resched()) |
1da177e4 LT |
3658 | goto need_resched; |
3659 | } | |
1da177e4 LT |
3660 | EXPORT_SYMBOL(schedule); |
3661 | ||
c08f7829 | 3662 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d PZ |
3663 | /* |
3664 | * Look out! "owner" is an entirely speculative pointer | |
3665 | * access and not reliable. | |
3666 | */ | |
3667 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
3668 | { | |
3669 | unsigned int cpu; | |
3670 | struct rq *rq; | |
3671 | ||
3672 | if (!sched_feat(OWNER_SPIN)) | |
3673 | return 0; | |
3674 | ||
3675 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3676 | /* | |
3677 | * Need to access the cpu field knowing that | |
3678 | * DEBUG_PAGEALLOC could have unmapped it if | |
3679 | * the mutex owner just released it and exited. | |
3680 | */ | |
3681 | if (probe_kernel_address(&owner->cpu, cpu)) | |
4b402210 | 3682 | return 0; |
0d66bf6d PZ |
3683 | #else |
3684 | cpu = owner->cpu; | |
3685 | #endif | |
3686 | ||
3687 | /* | |
3688 | * Even if the access succeeded (likely case), | |
3689 | * the cpu field may no longer be valid. | |
3690 | */ | |
3691 | if (cpu >= nr_cpumask_bits) | |
4b402210 | 3692 | return 0; |
0d66bf6d PZ |
3693 | |
3694 | /* | |
3695 | * We need to validate that we can do a | |
3696 | * get_cpu() and that we have the percpu area. | |
3697 | */ | |
3698 | if (!cpu_online(cpu)) | |
4b402210 | 3699 | return 0; |
0d66bf6d PZ |
3700 | |
3701 | rq = cpu_rq(cpu); | |
3702 | ||
3703 | for (;;) { | |
3704 | /* | |
3705 | * Owner changed, break to re-assess state. | |
3706 | */ | |
3707 | if (lock->owner != owner) | |
3708 | break; | |
3709 | ||
3710 | /* | |
3711 | * Is that owner really running on that cpu? | |
3712 | */ | |
3713 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
3714 | return 0; | |
3715 | ||
3716 | cpu_relax(); | |
3717 | } | |
4b402210 | 3718 | |
0d66bf6d PZ |
3719 | return 1; |
3720 | } | |
3721 | #endif | |
3722 | ||
1da177e4 LT |
3723 | #ifdef CONFIG_PREEMPT |
3724 | /* | |
2ed6e34f | 3725 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3726 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3727 | * occur there and call schedule directly. |
3728 | */ | |
3729 | asmlinkage void __sched preempt_schedule(void) | |
3730 | { | |
3731 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3732 | |
1da177e4 LT |
3733 | /* |
3734 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3735 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3736 | */ |
beed33a8 | 3737 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3738 | return; |
3739 | ||
3a5c359a AK |
3740 | do { |
3741 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 3742 | schedule(); |
3a5c359a | 3743 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3744 | |
3a5c359a AK |
3745 | /* |
3746 | * Check again in case we missed a preemption opportunity | |
3747 | * between schedule and now. | |
3748 | */ | |
3749 | barrier(); | |
5ed0cec0 | 3750 | } while (need_resched()); |
1da177e4 | 3751 | } |
1da177e4 LT |
3752 | EXPORT_SYMBOL(preempt_schedule); |
3753 | ||
3754 | /* | |
2ed6e34f | 3755 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3756 | * off of irq context. |
3757 | * Note, that this is called and return with irqs disabled. This will | |
3758 | * protect us against recursive calling from irq. | |
3759 | */ | |
3760 | asmlinkage void __sched preempt_schedule_irq(void) | |
3761 | { | |
3762 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3763 | |
2ed6e34f | 3764 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3765 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3766 | ||
3a5c359a AK |
3767 | do { |
3768 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
3769 | local_irq_enable(); |
3770 | schedule(); | |
3771 | local_irq_disable(); | |
3a5c359a | 3772 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3773 | |
3a5c359a AK |
3774 | /* |
3775 | * Check again in case we missed a preemption opportunity | |
3776 | * between schedule and now. | |
3777 | */ | |
3778 | barrier(); | |
5ed0cec0 | 3779 | } while (need_resched()); |
1da177e4 LT |
3780 | } |
3781 | ||
3782 | #endif /* CONFIG_PREEMPT */ | |
3783 | ||
63859d4f | 3784 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3785 | void *key) |
1da177e4 | 3786 | { |
63859d4f | 3787 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3788 | } |
1da177e4 LT |
3789 | EXPORT_SYMBOL(default_wake_function); |
3790 | ||
3791 | /* | |
41a2d6cf IM |
3792 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3793 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3794 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3795 | * | |
3796 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3797 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3798 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3799 | */ | |
78ddb08f | 3800 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 3801 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 3802 | { |
2e45874c | 3803 | wait_queue_t *curr, *next; |
1da177e4 | 3804 | |
2e45874c | 3805 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3806 | unsigned flags = curr->flags; |
3807 | ||
63859d4f | 3808 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 3809 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3810 | break; |
3811 | } | |
3812 | } | |
3813 | ||
3814 | /** | |
3815 | * __wake_up - wake up threads blocked on a waitqueue. | |
3816 | * @q: the waitqueue | |
3817 | * @mode: which threads | |
3818 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3819 | * @key: is directly passed to the wakeup function |
50fa610a DH |
3820 | * |
3821 | * It may be assumed that this function implies a write memory barrier before | |
3822 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3823 | */ |
7ad5b3a5 | 3824 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 3825 | int nr_exclusive, void *key) |
1da177e4 LT |
3826 | { |
3827 | unsigned long flags; | |
3828 | ||
3829 | spin_lock_irqsave(&q->lock, flags); | |
3830 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3831 | spin_unlock_irqrestore(&q->lock, flags); | |
3832 | } | |
1da177e4 LT |
3833 | EXPORT_SYMBOL(__wake_up); |
3834 | ||
3835 | /* | |
3836 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3837 | */ | |
7ad5b3a5 | 3838 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
3839 | { |
3840 | __wake_up_common(q, mode, 1, 0, NULL); | |
3841 | } | |
22c43c81 | 3842 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 3843 | |
4ede816a DL |
3844 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
3845 | { | |
3846 | __wake_up_common(q, mode, 1, 0, key); | |
3847 | } | |
3848 | ||
1da177e4 | 3849 | /** |
4ede816a | 3850 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3851 | * @q: the waitqueue |
3852 | * @mode: which threads | |
3853 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 3854 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
3855 | * |
3856 | * The sync wakeup differs that the waker knows that it will schedule | |
3857 | * away soon, so while the target thread will be woken up, it will not | |
3858 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3859 | * with each other. This can prevent needless bouncing between CPUs. | |
3860 | * | |
3861 | * On UP it can prevent extra preemption. | |
50fa610a DH |
3862 | * |
3863 | * It may be assumed that this function implies a write memory barrier before | |
3864 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3865 | */ |
4ede816a DL |
3866 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
3867 | int nr_exclusive, void *key) | |
1da177e4 LT |
3868 | { |
3869 | unsigned long flags; | |
7d478721 | 3870 | int wake_flags = WF_SYNC; |
1da177e4 LT |
3871 | |
3872 | if (unlikely(!q)) | |
3873 | return; | |
3874 | ||
3875 | if (unlikely(!nr_exclusive)) | |
7d478721 | 3876 | wake_flags = 0; |
1da177e4 LT |
3877 | |
3878 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 3879 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
3880 | spin_unlock_irqrestore(&q->lock, flags); |
3881 | } | |
4ede816a DL |
3882 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
3883 | ||
3884 | /* | |
3885 | * __wake_up_sync - see __wake_up_sync_key() | |
3886 | */ | |
3887 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
3888 | { | |
3889 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
3890 | } | |
1da177e4 LT |
3891 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
3892 | ||
65eb3dc6 KD |
3893 | /** |
3894 | * complete: - signals a single thread waiting on this completion | |
3895 | * @x: holds the state of this particular completion | |
3896 | * | |
3897 | * This will wake up a single thread waiting on this completion. Threads will be | |
3898 | * awakened in the same order in which they were queued. | |
3899 | * | |
3900 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
3901 | * |
3902 | * It may be assumed that this function implies a write memory barrier before | |
3903 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3904 | */ |
b15136e9 | 3905 | void complete(struct completion *x) |
1da177e4 LT |
3906 | { |
3907 | unsigned long flags; | |
3908 | ||
3909 | spin_lock_irqsave(&x->wait.lock, flags); | |
3910 | x->done++; | |
d9514f6c | 3911 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
3912 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3913 | } | |
3914 | EXPORT_SYMBOL(complete); | |
3915 | ||
65eb3dc6 KD |
3916 | /** |
3917 | * complete_all: - signals all threads waiting on this completion | |
3918 | * @x: holds the state of this particular completion | |
3919 | * | |
3920 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
3921 | * |
3922 | * It may be assumed that this function implies a write memory barrier before | |
3923 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3924 | */ |
b15136e9 | 3925 | void complete_all(struct completion *x) |
1da177e4 LT |
3926 | { |
3927 | unsigned long flags; | |
3928 | ||
3929 | spin_lock_irqsave(&x->wait.lock, flags); | |
3930 | x->done += UINT_MAX/2; | |
d9514f6c | 3931 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
3932 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3933 | } | |
3934 | EXPORT_SYMBOL(complete_all); | |
3935 | ||
8cbbe86d AK |
3936 | static inline long __sched |
3937 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3938 | { |
1da177e4 LT |
3939 | if (!x->done) { |
3940 | DECLARE_WAITQUEUE(wait, current); | |
3941 | ||
a93d2f17 | 3942 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 3943 | do { |
94d3d824 | 3944 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
3945 | timeout = -ERESTARTSYS; |
3946 | break; | |
8cbbe86d AK |
3947 | } |
3948 | __set_current_state(state); | |
1da177e4 LT |
3949 | spin_unlock_irq(&x->wait.lock); |
3950 | timeout = schedule_timeout(timeout); | |
3951 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 3952 | } while (!x->done && timeout); |
1da177e4 | 3953 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
3954 | if (!x->done) |
3955 | return timeout; | |
1da177e4 LT |
3956 | } |
3957 | x->done--; | |
ea71a546 | 3958 | return timeout ?: 1; |
1da177e4 | 3959 | } |
1da177e4 | 3960 | |
8cbbe86d AK |
3961 | static long __sched |
3962 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3963 | { |
1da177e4 LT |
3964 | might_sleep(); |
3965 | ||
3966 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 3967 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 3968 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3969 | return timeout; |
3970 | } | |
1da177e4 | 3971 | |
65eb3dc6 KD |
3972 | /** |
3973 | * wait_for_completion: - waits for completion of a task | |
3974 | * @x: holds the state of this particular completion | |
3975 | * | |
3976 | * This waits to be signaled for completion of a specific task. It is NOT | |
3977 | * interruptible and there is no timeout. | |
3978 | * | |
3979 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
3980 | * and interrupt capability. Also see complete(). | |
3981 | */ | |
b15136e9 | 3982 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
3983 | { |
3984 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 3985 | } |
8cbbe86d | 3986 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 3987 | |
65eb3dc6 KD |
3988 | /** |
3989 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
3990 | * @x: holds the state of this particular completion | |
3991 | * @timeout: timeout value in jiffies | |
3992 | * | |
3993 | * This waits for either a completion of a specific task to be signaled or for a | |
3994 | * specified timeout to expire. The timeout is in jiffies. It is not | |
3995 | * interruptible. | |
3996 | */ | |
b15136e9 | 3997 | unsigned long __sched |
8cbbe86d | 3998 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 3999 | { |
8cbbe86d | 4000 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4001 | } |
8cbbe86d | 4002 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4003 | |
65eb3dc6 KD |
4004 | /** |
4005 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4006 | * @x: holds the state of this particular completion | |
4007 | * | |
4008 | * This waits for completion of a specific task to be signaled. It is | |
4009 | * interruptible. | |
4010 | */ | |
8cbbe86d | 4011 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4012 | { |
51e97990 AK |
4013 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4014 | if (t == -ERESTARTSYS) | |
4015 | return t; | |
4016 | return 0; | |
0fec171c | 4017 | } |
8cbbe86d | 4018 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4019 | |
65eb3dc6 KD |
4020 | /** |
4021 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4022 | * @x: holds the state of this particular completion | |
4023 | * @timeout: timeout value in jiffies | |
4024 | * | |
4025 | * This waits for either a completion of a specific task to be signaled or for a | |
4026 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4027 | */ | |
b15136e9 | 4028 | unsigned long __sched |
8cbbe86d AK |
4029 | wait_for_completion_interruptible_timeout(struct completion *x, |
4030 | unsigned long timeout) | |
0fec171c | 4031 | { |
8cbbe86d | 4032 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4033 | } |
8cbbe86d | 4034 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4035 | |
65eb3dc6 KD |
4036 | /** |
4037 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4038 | * @x: holds the state of this particular completion | |
4039 | * | |
4040 | * This waits to be signaled for completion of a specific task. It can be | |
4041 | * interrupted by a kill signal. | |
4042 | */ | |
009e577e MW |
4043 | int __sched wait_for_completion_killable(struct completion *x) |
4044 | { | |
4045 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4046 | if (t == -ERESTARTSYS) | |
4047 | return t; | |
4048 | return 0; | |
4049 | } | |
4050 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4051 | ||
0aa12fb4 SW |
4052 | /** |
4053 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
4054 | * @x: holds the state of this particular completion | |
4055 | * @timeout: timeout value in jiffies | |
4056 | * | |
4057 | * This waits for either a completion of a specific task to be | |
4058 | * signaled or for a specified timeout to expire. It can be | |
4059 | * interrupted by a kill signal. The timeout is in jiffies. | |
4060 | */ | |
4061 | unsigned long __sched | |
4062 | wait_for_completion_killable_timeout(struct completion *x, | |
4063 | unsigned long timeout) | |
4064 | { | |
4065 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
4066 | } | |
4067 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
4068 | ||
be4de352 DC |
4069 | /** |
4070 | * try_wait_for_completion - try to decrement a completion without blocking | |
4071 | * @x: completion structure | |
4072 | * | |
4073 | * Returns: 0 if a decrement cannot be done without blocking | |
4074 | * 1 if a decrement succeeded. | |
4075 | * | |
4076 | * If a completion is being used as a counting completion, | |
4077 | * attempt to decrement the counter without blocking. This | |
4078 | * enables us to avoid waiting if the resource the completion | |
4079 | * is protecting is not available. | |
4080 | */ | |
4081 | bool try_wait_for_completion(struct completion *x) | |
4082 | { | |
7539a3b3 | 4083 | unsigned long flags; |
be4de352 DC |
4084 | int ret = 1; |
4085 | ||
7539a3b3 | 4086 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4087 | if (!x->done) |
4088 | ret = 0; | |
4089 | else | |
4090 | x->done--; | |
7539a3b3 | 4091 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4092 | return ret; |
4093 | } | |
4094 | EXPORT_SYMBOL(try_wait_for_completion); | |
4095 | ||
4096 | /** | |
4097 | * completion_done - Test to see if a completion has any waiters | |
4098 | * @x: completion structure | |
4099 | * | |
4100 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4101 | * 1 if there are no waiters. | |
4102 | * | |
4103 | */ | |
4104 | bool completion_done(struct completion *x) | |
4105 | { | |
7539a3b3 | 4106 | unsigned long flags; |
be4de352 DC |
4107 | int ret = 1; |
4108 | ||
7539a3b3 | 4109 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4110 | if (!x->done) |
4111 | ret = 0; | |
7539a3b3 | 4112 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4113 | return ret; |
4114 | } | |
4115 | EXPORT_SYMBOL(completion_done); | |
4116 | ||
8cbbe86d AK |
4117 | static long __sched |
4118 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4119 | { |
0fec171c IM |
4120 | unsigned long flags; |
4121 | wait_queue_t wait; | |
4122 | ||
4123 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4124 | |
8cbbe86d | 4125 | __set_current_state(state); |
1da177e4 | 4126 | |
8cbbe86d AK |
4127 | spin_lock_irqsave(&q->lock, flags); |
4128 | __add_wait_queue(q, &wait); | |
4129 | spin_unlock(&q->lock); | |
4130 | timeout = schedule_timeout(timeout); | |
4131 | spin_lock_irq(&q->lock); | |
4132 | __remove_wait_queue(q, &wait); | |
4133 | spin_unlock_irqrestore(&q->lock, flags); | |
4134 | ||
4135 | return timeout; | |
4136 | } | |
4137 | ||
4138 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4139 | { | |
4140 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4141 | } |
1da177e4 LT |
4142 | EXPORT_SYMBOL(interruptible_sleep_on); |
4143 | ||
0fec171c | 4144 | long __sched |
95cdf3b7 | 4145 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4146 | { |
8cbbe86d | 4147 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4148 | } |
1da177e4 LT |
4149 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4150 | ||
0fec171c | 4151 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4152 | { |
8cbbe86d | 4153 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4154 | } |
1da177e4 LT |
4155 | EXPORT_SYMBOL(sleep_on); |
4156 | ||
0fec171c | 4157 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4158 | { |
8cbbe86d | 4159 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4160 | } |
1da177e4 LT |
4161 | EXPORT_SYMBOL(sleep_on_timeout); |
4162 | ||
b29739f9 IM |
4163 | #ifdef CONFIG_RT_MUTEXES |
4164 | ||
4165 | /* | |
4166 | * rt_mutex_setprio - set the current priority of a task | |
4167 | * @p: task | |
4168 | * @prio: prio value (kernel-internal form) | |
4169 | * | |
4170 | * This function changes the 'effective' priority of a task. It does | |
4171 | * not touch ->normal_prio like __setscheduler(). | |
4172 | * | |
4173 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4174 | */ | |
36c8b586 | 4175 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4176 | { |
4177 | unsigned long flags; | |
83b699ed | 4178 | int oldprio, on_rq, running; |
70b97a7f | 4179 | struct rq *rq; |
83ab0aa0 | 4180 | const struct sched_class *prev_class; |
b29739f9 IM |
4181 | |
4182 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4183 | ||
4184 | rq = task_rq_lock(p, &flags); | |
4185 | ||
d5f9f942 | 4186 | oldprio = p->prio; |
83ab0aa0 | 4187 | prev_class = p->sched_class; |
dd41f596 | 4188 | on_rq = p->se.on_rq; |
051a1d1a | 4189 | running = task_current(rq, p); |
0e1f3483 | 4190 | if (on_rq) |
69be72c1 | 4191 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4192 | if (running) |
4193 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4194 | |
4195 | if (rt_prio(prio)) | |
4196 | p->sched_class = &rt_sched_class; | |
4197 | else | |
4198 | p->sched_class = &fair_sched_class; | |
4199 | ||
b29739f9 IM |
4200 | p->prio = prio; |
4201 | ||
0e1f3483 HS |
4202 | if (running) |
4203 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4204 | if (on_rq) { |
371fd7e7 | 4205 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 SR |
4206 | |
4207 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4208 | } |
4209 | task_rq_unlock(rq, &flags); | |
4210 | } | |
4211 | ||
4212 | #endif | |
4213 | ||
36c8b586 | 4214 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4215 | { |
dd41f596 | 4216 | int old_prio, delta, on_rq; |
1da177e4 | 4217 | unsigned long flags; |
70b97a7f | 4218 | struct rq *rq; |
1da177e4 LT |
4219 | |
4220 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4221 | return; | |
4222 | /* | |
4223 | * We have to be careful, if called from sys_setpriority(), | |
4224 | * the task might be in the middle of scheduling on another CPU. | |
4225 | */ | |
4226 | rq = task_rq_lock(p, &flags); | |
4227 | /* | |
4228 | * The RT priorities are set via sched_setscheduler(), but we still | |
4229 | * allow the 'normal' nice value to be set - but as expected | |
4230 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4231 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4232 | */ |
e05606d3 | 4233 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4234 | p->static_prio = NICE_TO_PRIO(nice); |
4235 | goto out_unlock; | |
4236 | } | |
dd41f596 | 4237 | on_rq = p->se.on_rq; |
c09595f6 | 4238 | if (on_rq) |
69be72c1 | 4239 | dequeue_task(rq, p, 0); |
1da177e4 | 4240 | |
1da177e4 | 4241 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4242 | set_load_weight(p); |
b29739f9 IM |
4243 | old_prio = p->prio; |
4244 | p->prio = effective_prio(p); | |
4245 | delta = p->prio - old_prio; | |
1da177e4 | 4246 | |
dd41f596 | 4247 | if (on_rq) { |
371fd7e7 | 4248 | enqueue_task(rq, p, 0); |
1da177e4 | 4249 | /* |
d5f9f942 AM |
4250 | * If the task increased its priority or is running and |
4251 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4252 | */ |
d5f9f942 | 4253 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4254 | resched_task(rq->curr); |
4255 | } | |
4256 | out_unlock: | |
4257 | task_rq_unlock(rq, &flags); | |
4258 | } | |
1da177e4 LT |
4259 | EXPORT_SYMBOL(set_user_nice); |
4260 | ||
e43379f1 MM |
4261 | /* |
4262 | * can_nice - check if a task can reduce its nice value | |
4263 | * @p: task | |
4264 | * @nice: nice value | |
4265 | */ | |
36c8b586 | 4266 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4267 | { |
024f4747 MM |
4268 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4269 | int nice_rlim = 20 - nice; | |
48f24c4d | 4270 | |
78d7d407 | 4271 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
4272 | capable(CAP_SYS_NICE)); |
4273 | } | |
4274 | ||
1da177e4 LT |
4275 | #ifdef __ARCH_WANT_SYS_NICE |
4276 | ||
4277 | /* | |
4278 | * sys_nice - change the priority of the current process. | |
4279 | * @increment: priority increment | |
4280 | * | |
4281 | * sys_setpriority is a more generic, but much slower function that | |
4282 | * does similar things. | |
4283 | */ | |
5add95d4 | 4284 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 4285 | { |
48f24c4d | 4286 | long nice, retval; |
1da177e4 LT |
4287 | |
4288 | /* | |
4289 | * Setpriority might change our priority at the same moment. | |
4290 | * We don't have to worry. Conceptually one call occurs first | |
4291 | * and we have a single winner. | |
4292 | */ | |
e43379f1 MM |
4293 | if (increment < -40) |
4294 | increment = -40; | |
1da177e4 LT |
4295 | if (increment > 40) |
4296 | increment = 40; | |
4297 | ||
2b8f836f | 4298 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
4299 | if (nice < -20) |
4300 | nice = -20; | |
4301 | if (nice > 19) | |
4302 | nice = 19; | |
4303 | ||
e43379f1 MM |
4304 | if (increment < 0 && !can_nice(current, nice)) |
4305 | return -EPERM; | |
4306 | ||
1da177e4 LT |
4307 | retval = security_task_setnice(current, nice); |
4308 | if (retval) | |
4309 | return retval; | |
4310 | ||
4311 | set_user_nice(current, nice); | |
4312 | return 0; | |
4313 | } | |
4314 | ||
4315 | #endif | |
4316 | ||
4317 | /** | |
4318 | * task_prio - return the priority value of a given task. | |
4319 | * @p: the task in question. | |
4320 | * | |
4321 | * This is the priority value as seen by users in /proc. | |
4322 | * RT tasks are offset by -200. Normal tasks are centered | |
4323 | * around 0, value goes from -16 to +15. | |
4324 | */ | |
36c8b586 | 4325 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4326 | { |
4327 | return p->prio - MAX_RT_PRIO; | |
4328 | } | |
4329 | ||
4330 | /** | |
4331 | * task_nice - return the nice value of a given task. | |
4332 | * @p: the task in question. | |
4333 | */ | |
36c8b586 | 4334 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4335 | { |
4336 | return TASK_NICE(p); | |
4337 | } | |
150d8bed | 4338 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4339 | |
4340 | /** | |
4341 | * idle_cpu - is a given cpu idle currently? | |
4342 | * @cpu: the processor in question. | |
4343 | */ | |
4344 | int idle_cpu(int cpu) | |
4345 | { | |
4346 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4347 | } | |
4348 | ||
1da177e4 LT |
4349 | /** |
4350 | * idle_task - return the idle task for a given cpu. | |
4351 | * @cpu: the processor in question. | |
4352 | */ | |
36c8b586 | 4353 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4354 | { |
4355 | return cpu_rq(cpu)->idle; | |
4356 | } | |
4357 | ||
4358 | /** | |
4359 | * find_process_by_pid - find a process with a matching PID value. | |
4360 | * @pid: the pid in question. | |
4361 | */ | |
a9957449 | 4362 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4363 | { |
228ebcbe | 4364 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4365 | } |
4366 | ||
4367 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4368 | static void |
4369 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4370 | { |
dd41f596 | 4371 | BUG_ON(p->se.on_rq); |
48f24c4d | 4372 | |
1da177e4 LT |
4373 | p->policy = policy; |
4374 | p->rt_priority = prio; | |
b29739f9 IM |
4375 | p->normal_prio = normal_prio(p); |
4376 | /* we are holding p->pi_lock already */ | |
4377 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4378 | if (rt_prio(p->prio)) |
4379 | p->sched_class = &rt_sched_class; | |
4380 | else | |
4381 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 4382 | set_load_weight(p); |
1da177e4 LT |
4383 | } |
4384 | ||
c69e8d9c DH |
4385 | /* |
4386 | * check the target process has a UID that matches the current process's | |
4387 | */ | |
4388 | static bool check_same_owner(struct task_struct *p) | |
4389 | { | |
4390 | const struct cred *cred = current_cred(), *pcred; | |
4391 | bool match; | |
4392 | ||
4393 | rcu_read_lock(); | |
4394 | pcred = __task_cred(p); | |
4395 | match = (cred->euid == pcred->euid || | |
4396 | cred->euid == pcred->uid); | |
4397 | rcu_read_unlock(); | |
4398 | return match; | |
4399 | } | |
4400 | ||
961ccddd RR |
4401 | static int __sched_setscheduler(struct task_struct *p, int policy, |
4402 | struct sched_param *param, bool user) | |
1da177e4 | 4403 | { |
83b699ed | 4404 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4405 | unsigned long flags; |
83ab0aa0 | 4406 | const struct sched_class *prev_class; |
70b97a7f | 4407 | struct rq *rq; |
ca94c442 | 4408 | int reset_on_fork; |
1da177e4 | 4409 | |
66e5393a SR |
4410 | /* may grab non-irq protected spin_locks */ |
4411 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4412 | recheck: |
4413 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4414 | if (policy < 0) { |
4415 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4416 | policy = oldpolicy = p->policy; |
ca94c442 LP |
4417 | } else { |
4418 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
4419 | policy &= ~SCHED_RESET_ON_FORK; | |
4420 | ||
4421 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
4422 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
4423 | policy != SCHED_IDLE) | |
4424 | return -EINVAL; | |
4425 | } | |
4426 | ||
1da177e4 LT |
4427 | /* |
4428 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4429 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4430 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4431 | */ |
4432 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4433 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4434 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4435 | return -EINVAL; |
e05606d3 | 4436 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4437 | return -EINVAL; |
4438 | ||
37e4ab3f OC |
4439 | /* |
4440 | * Allow unprivileged RT tasks to decrease priority: | |
4441 | */ | |
961ccddd | 4442 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4443 | if (rt_policy(policy)) { |
8dc3e909 | 4444 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4445 | |
4446 | if (!lock_task_sighand(p, &flags)) | |
4447 | return -ESRCH; | |
78d7d407 | 4448 | rlim_rtprio = task_rlimit(p, RLIMIT_RTPRIO); |
8dc3e909 ON |
4449 | unlock_task_sighand(p, &flags); |
4450 | ||
4451 | /* can't set/change the rt policy */ | |
4452 | if (policy != p->policy && !rlim_rtprio) | |
4453 | return -EPERM; | |
4454 | ||
4455 | /* can't increase priority */ | |
4456 | if (param->sched_priority > p->rt_priority && | |
4457 | param->sched_priority > rlim_rtprio) | |
4458 | return -EPERM; | |
4459 | } | |
dd41f596 IM |
4460 | /* |
4461 | * Like positive nice levels, dont allow tasks to | |
4462 | * move out of SCHED_IDLE either: | |
4463 | */ | |
4464 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4465 | return -EPERM; | |
5fe1d75f | 4466 | |
37e4ab3f | 4467 | /* can't change other user's priorities */ |
c69e8d9c | 4468 | if (!check_same_owner(p)) |
37e4ab3f | 4469 | return -EPERM; |
ca94c442 LP |
4470 | |
4471 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4472 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4473 | return -EPERM; | |
37e4ab3f | 4474 | } |
1da177e4 | 4475 | |
725aad24 | 4476 | if (user) { |
725aad24 JF |
4477 | retval = security_task_setscheduler(p, policy, param); |
4478 | if (retval) | |
4479 | return retval; | |
4480 | } | |
4481 | ||
b29739f9 IM |
4482 | /* |
4483 | * make sure no PI-waiters arrive (or leave) while we are | |
4484 | * changing the priority of the task: | |
4485 | */ | |
1d615482 | 4486 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
1da177e4 LT |
4487 | /* |
4488 | * To be able to change p->policy safely, the apropriate | |
4489 | * runqueue lock must be held. | |
4490 | */ | |
b29739f9 | 4491 | rq = __task_rq_lock(p); |
dc61b1d6 PZ |
4492 | |
4493 | #ifdef CONFIG_RT_GROUP_SCHED | |
4494 | if (user) { | |
4495 | /* | |
4496 | * Do not allow realtime tasks into groups that have no runtime | |
4497 | * assigned. | |
4498 | */ | |
4499 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
4500 | task_group(p)->rt_bandwidth.rt_runtime == 0) { | |
4501 | __task_rq_unlock(rq); | |
4502 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
4503 | return -EPERM; | |
4504 | } | |
4505 | } | |
4506 | #endif | |
4507 | ||
1da177e4 LT |
4508 | /* recheck policy now with rq lock held */ |
4509 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4510 | policy = oldpolicy = -1; | |
b29739f9 | 4511 | __task_rq_unlock(rq); |
1d615482 | 4512 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4513 | goto recheck; |
4514 | } | |
dd41f596 | 4515 | on_rq = p->se.on_rq; |
051a1d1a | 4516 | running = task_current(rq, p); |
0e1f3483 | 4517 | if (on_rq) |
2e1cb74a | 4518 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
4519 | if (running) |
4520 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4521 | |
ca94c442 LP |
4522 | p->sched_reset_on_fork = reset_on_fork; |
4523 | ||
1da177e4 | 4524 | oldprio = p->prio; |
83ab0aa0 | 4525 | prev_class = p->sched_class; |
dd41f596 | 4526 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4527 | |
0e1f3483 HS |
4528 | if (running) |
4529 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
4530 | if (on_rq) { |
4531 | activate_task(rq, p, 0); | |
cb469845 SR |
4532 | |
4533 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 4534 | } |
b29739f9 | 4535 | __task_rq_unlock(rq); |
1d615482 | 4536 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
b29739f9 | 4537 | |
95e02ca9 TG |
4538 | rt_mutex_adjust_pi(p); |
4539 | ||
1da177e4 LT |
4540 | return 0; |
4541 | } | |
961ccddd RR |
4542 | |
4543 | /** | |
4544 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4545 | * @p: the task in question. | |
4546 | * @policy: new policy. | |
4547 | * @param: structure containing the new RT priority. | |
4548 | * | |
4549 | * NOTE that the task may be already dead. | |
4550 | */ | |
4551 | int sched_setscheduler(struct task_struct *p, int policy, | |
4552 | struct sched_param *param) | |
4553 | { | |
4554 | return __sched_setscheduler(p, policy, param, true); | |
4555 | } | |
1da177e4 LT |
4556 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4557 | ||
961ccddd RR |
4558 | /** |
4559 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4560 | * @p: the task in question. | |
4561 | * @policy: new policy. | |
4562 | * @param: structure containing the new RT priority. | |
4563 | * | |
4564 | * Just like sched_setscheduler, only don't bother checking if the | |
4565 | * current context has permission. For example, this is needed in | |
4566 | * stop_machine(): we create temporary high priority worker threads, | |
4567 | * but our caller might not have that capability. | |
4568 | */ | |
4569 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
4570 | struct sched_param *param) | |
4571 | { | |
4572 | return __sched_setscheduler(p, policy, param, false); | |
4573 | } | |
4574 | ||
95cdf3b7 IM |
4575 | static int |
4576 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4577 | { |
1da177e4 LT |
4578 | struct sched_param lparam; |
4579 | struct task_struct *p; | |
36c8b586 | 4580 | int retval; |
1da177e4 LT |
4581 | |
4582 | if (!param || pid < 0) | |
4583 | return -EINVAL; | |
4584 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4585 | return -EFAULT; | |
5fe1d75f ON |
4586 | |
4587 | rcu_read_lock(); | |
4588 | retval = -ESRCH; | |
1da177e4 | 4589 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4590 | if (p != NULL) |
4591 | retval = sched_setscheduler(p, policy, &lparam); | |
4592 | rcu_read_unlock(); | |
36c8b586 | 4593 | |
1da177e4 LT |
4594 | return retval; |
4595 | } | |
4596 | ||
4597 | /** | |
4598 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4599 | * @pid: the pid in question. | |
4600 | * @policy: new policy. | |
4601 | * @param: structure containing the new RT priority. | |
4602 | */ | |
5add95d4 HC |
4603 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4604 | struct sched_param __user *, param) | |
1da177e4 | 4605 | { |
c21761f1 JB |
4606 | /* negative values for policy are not valid */ |
4607 | if (policy < 0) | |
4608 | return -EINVAL; | |
4609 | ||
1da177e4 LT |
4610 | return do_sched_setscheduler(pid, policy, param); |
4611 | } | |
4612 | ||
4613 | /** | |
4614 | * sys_sched_setparam - set/change the RT priority of a thread | |
4615 | * @pid: the pid in question. | |
4616 | * @param: structure containing the new RT priority. | |
4617 | */ | |
5add95d4 | 4618 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4619 | { |
4620 | return do_sched_setscheduler(pid, -1, param); | |
4621 | } | |
4622 | ||
4623 | /** | |
4624 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4625 | * @pid: the pid in question. | |
4626 | */ | |
5add95d4 | 4627 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4628 | { |
36c8b586 | 4629 | struct task_struct *p; |
3a5c359a | 4630 | int retval; |
1da177e4 LT |
4631 | |
4632 | if (pid < 0) | |
3a5c359a | 4633 | return -EINVAL; |
1da177e4 LT |
4634 | |
4635 | retval = -ESRCH; | |
5fe85be0 | 4636 | rcu_read_lock(); |
1da177e4 LT |
4637 | p = find_process_by_pid(pid); |
4638 | if (p) { | |
4639 | retval = security_task_getscheduler(p); | |
4640 | if (!retval) | |
ca94c442 LP |
4641 | retval = p->policy |
4642 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4643 | } |
5fe85be0 | 4644 | rcu_read_unlock(); |
1da177e4 LT |
4645 | return retval; |
4646 | } | |
4647 | ||
4648 | /** | |
ca94c442 | 4649 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4650 | * @pid: the pid in question. |
4651 | * @param: structure containing the RT priority. | |
4652 | */ | |
5add95d4 | 4653 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4654 | { |
4655 | struct sched_param lp; | |
36c8b586 | 4656 | struct task_struct *p; |
3a5c359a | 4657 | int retval; |
1da177e4 LT |
4658 | |
4659 | if (!param || pid < 0) | |
3a5c359a | 4660 | return -EINVAL; |
1da177e4 | 4661 | |
5fe85be0 | 4662 | rcu_read_lock(); |
1da177e4 LT |
4663 | p = find_process_by_pid(pid); |
4664 | retval = -ESRCH; | |
4665 | if (!p) | |
4666 | goto out_unlock; | |
4667 | ||
4668 | retval = security_task_getscheduler(p); | |
4669 | if (retval) | |
4670 | goto out_unlock; | |
4671 | ||
4672 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4673 | rcu_read_unlock(); |
1da177e4 LT |
4674 | |
4675 | /* | |
4676 | * This one might sleep, we cannot do it with a spinlock held ... | |
4677 | */ | |
4678 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4679 | ||
1da177e4 LT |
4680 | return retval; |
4681 | ||
4682 | out_unlock: | |
5fe85be0 | 4683 | rcu_read_unlock(); |
1da177e4 LT |
4684 | return retval; |
4685 | } | |
4686 | ||
96f874e2 | 4687 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4688 | { |
5a16f3d3 | 4689 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4690 | struct task_struct *p; |
4691 | int retval; | |
1da177e4 | 4692 | |
95402b38 | 4693 | get_online_cpus(); |
23f5d142 | 4694 | rcu_read_lock(); |
1da177e4 LT |
4695 | |
4696 | p = find_process_by_pid(pid); | |
4697 | if (!p) { | |
23f5d142 | 4698 | rcu_read_unlock(); |
95402b38 | 4699 | put_online_cpus(); |
1da177e4 LT |
4700 | return -ESRCH; |
4701 | } | |
4702 | ||
23f5d142 | 4703 | /* Prevent p going away */ |
1da177e4 | 4704 | get_task_struct(p); |
23f5d142 | 4705 | rcu_read_unlock(); |
1da177e4 | 4706 | |
5a16f3d3 RR |
4707 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4708 | retval = -ENOMEM; | |
4709 | goto out_put_task; | |
4710 | } | |
4711 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4712 | retval = -ENOMEM; | |
4713 | goto out_free_cpus_allowed; | |
4714 | } | |
1da177e4 | 4715 | retval = -EPERM; |
c69e8d9c | 4716 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
4717 | goto out_unlock; |
4718 | ||
e7834f8f DQ |
4719 | retval = security_task_setscheduler(p, 0, NULL); |
4720 | if (retval) | |
4721 | goto out_unlock; | |
4722 | ||
5a16f3d3 RR |
4723 | cpuset_cpus_allowed(p, cpus_allowed); |
4724 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 4725 | again: |
5a16f3d3 | 4726 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4727 | |
8707d8b8 | 4728 | if (!retval) { |
5a16f3d3 RR |
4729 | cpuset_cpus_allowed(p, cpus_allowed); |
4730 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4731 | /* |
4732 | * We must have raced with a concurrent cpuset | |
4733 | * update. Just reset the cpus_allowed to the | |
4734 | * cpuset's cpus_allowed | |
4735 | */ | |
5a16f3d3 | 4736 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4737 | goto again; |
4738 | } | |
4739 | } | |
1da177e4 | 4740 | out_unlock: |
5a16f3d3 RR |
4741 | free_cpumask_var(new_mask); |
4742 | out_free_cpus_allowed: | |
4743 | free_cpumask_var(cpus_allowed); | |
4744 | out_put_task: | |
1da177e4 | 4745 | put_task_struct(p); |
95402b38 | 4746 | put_online_cpus(); |
1da177e4 LT |
4747 | return retval; |
4748 | } | |
4749 | ||
4750 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4751 | struct cpumask *new_mask) |
1da177e4 | 4752 | { |
96f874e2 RR |
4753 | if (len < cpumask_size()) |
4754 | cpumask_clear(new_mask); | |
4755 | else if (len > cpumask_size()) | |
4756 | len = cpumask_size(); | |
4757 | ||
1da177e4 LT |
4758 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4759 | } | |
4760 | ||
4761 | /** | |
4762 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4763 | * @pid: pid of the process | |
4764 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4765 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4766 | */ | |
5add95d4 HC |
4767 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4768 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4769 | { |
5a16f3d3 | 4770 | cpumask_var_t new_mask; |
1da177e4 LT |
4771 | int retval; |
4772 | ||
5a16f3d3 RR |
4773 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4774 | return -ENOMEM; | |
1da177e4 | 4775 | |
5a16f3d3 RR |
4776 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4777 | if (retval == 0) | |
4778 | retval = sched_setaffinity(pid, new_mask); | |
4779 | free_cpumask_var(new_mask); | |
4780 | return retval; | |
1da177e4 LT |
4781 | } |
4782 | ||
96f874e2 | 4783 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4784 | { |
36c8b586 | 4785 | struct task_struct *p; |
31605683 TG |
4786 | unsigned long flags; |
4787 | struct rq *rq; | |
1da177e4 | 4788 | int retval; |
1da177e4 | 4789 | |
95402b38 | 4790 | get_online_cpus(); |
23f5d142 | 4791 | rcu_read_lock(); |
1da177e4 LT |
4792 | |
4793 | retval = -ESRCH; | |
4794 | p = find_process_by_pid(pid); | |
4795 | if (!p) | |
4796 | goto out_unlock; | |
4797 | ||
e7834f8f DQ |
4798 | retval = security_task_getscheduler(p); |
4799 | if (retval) | |
4800 | goto out_unlock; | |
4801 | ||
31605683 | 4802 | rq = task_rq_lock(p, &flags); |
96f874e2 | 4803 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
31605683 | 4804 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
4805 | |
4806 | out_unlock: | |
23f5d142 | 4807 | rcu_read_unlock(); |
95402b38 | 4808 | put_online_cpus(); |
1da177e4 | 4809 | |
9531b62f | 4810 | return retval; |
1da177e4 LT |
4811 | } |
4812 | ||
4813 | /** | |
4814 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4815 | * @pid: pid of the process | |
4816 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4817 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4818 | */ | |
5add95d4 HC |
4819 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4820 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4821 | { |
4822 | int ret; | |
f17c8607 | 4823 | cpumask_var_t mask; |
1da177e4 | 4824 | |
84fba5ec | 4825 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4826 | return -EINVAL; |
4827 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4828 | return -EINVAL; |
4829 | ||
f17c8607 RR |
4830 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4831 | return -ENOMEM; | |
1da177e4 | 4832 | |
f17c8607 RR |
4833 | ret = sched_getaffinity(pid, mask); |
4834 | if (ret == 0) { | |
8bc037fb | 4835 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4836 | |
4837 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4838 | ret = -EFAULT; |
4839 | else | |
cd3d8031 | 4840 | ret = retlen; |
f17c8607 RR |
4841 | } |
4842 | free_cpumask_var(mask); | |
1da177e4 | 4843 | |
f17c8607 | 4844 | return ret; |
1da177e4 LT |
4845 | } |
4846 | ||
4847 | /** | |
4848 | * sys_sched_yield - yield the current processor to other threads. | |
4849 | * | |
dd41f596 IM |
4850 | * This function yields the current CPU to other tasks. If there are no |
4851 | * other threads running on this CPU then this function will return. | |
1da177e4 | 4852 | */ |
5add95d4 | 4853 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4854 | { |
70b97a7f | 4855 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4856 | |
2d72376b | 4857 | schedstat_inc(rq, yld_count); |
4530d7ab | 4858 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4859 | |
4860 | /* | |
4861 | * Since we are going to call schedule() anyway, there's | |
4862 | * no need to preempt or enable interrupts: | |
4863 | */ | |
4864 | __release(rq->lock); | |
8a25d5de | 4865 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4866 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
4867 | preempt_enable_no_resched(); |
4868 | ||
4869 | schedule(); | |
4870 | ||
4871 | return 0; | |
4872 | } | |
4873 | ||
d86ee480 PZ |
4874 | static inline int should_resched(void) |
4875 | { | |
4876 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
4877 | } | |
4878 | ||
e7b38404 | 4879 | static void __cond_resched(void) |
1da177e4 | 4880 | { |
e7aaaa69 FW |
4881 | add_preempt_count(PREEMPT_ACTIVE); |
4882 | schedule(); | |
4883 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
4884 | } |
4885 | ||
02b67cc3 | 4886 | int __sched _cond_resched(void) |
1da177e4 | 4887 | { |
d86ee480 | 4888 | if (should_resched()) { |
1da177e4 LT |
4889 | __cond_resched(); |
4890 | return 1; | |
4891 | } | |
4892 | return 0; | |
4893 | } | |
02b67cc3 | 4894 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4895 | |
4896 | /* | |
613afbf8 | 4897 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4898 | * call schedule, and on return reacquire the lock. |
4899 | * | |
41a2d6cf | 4900 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4901 | * operations here to prevent schedule() from being called twice (once via |
4902 | * spin_unlock(), once by hand). | |
4903 | */ | |
613afbf8 | 4904 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4905 | { |
d86ee480 | 4906 | int resched = should_resched(); |
6df3cecb JK |
4907 | int ret = 0; |
4908 | ||
f607c668 PZ |
4909 | lockdep_assert_held(lock); |
4910 | ||
95c354fe | 4911 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4912 | spin_unlock(lock); |
d86ee480 | 4913 | if (resched) |
95c354fe NP |
4914 | __cond_resched(); |
4915 | else | |
4916 | cpu_relax(); | |
6df3cecb | 4917 | ret = 1; |
1da177e4 | 4918 | spin_lock(lock); |
1da177e4 | 4919 | } |
6df3cecb | 4920 | return ret; |
1da177e4 | 4921 | } |
613afbf8 | 4922 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4923 | |
613afbf8 | 4924 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4925 | { |
4926 | BUG_ON(!in_softirq()); | |
4927 | ||
d86ee480 | 4928 | if (should_resched()) { |
98d82567 | 4929 | local_bh_enable(); |
1da177e4 LT |
4930 | __cond_resched(); |
4931 | local_bh_disable(); | |
4932 | return 1; | |
4933 | } | |
4934 | return 0; | |
4935 | } | |
613afbf8 | 4936 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4937 | |
1da177e4 LT |
4938 | /** |
4939 | * yield - yield the current processor to other threads. | |
4940 | * | |
72fd4a35 | 4941 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4942 | * thread runnable and calls sys_sched_yield(). |
4943 | */ | |
4944 | void __sched yield(void) | |
4945 | { | |
4946 | set_current_state(TASK_RUNNING); | |
4947 | sys_sched_yield(); | |
4948 | } | |
1da177e4 LT |
4949 | EXPORT_SYMBOL(yield); |
4950 | ||
4951 | /* | |
41a2d6cf | 4952 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4953 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4954 | */ |
4955 | void __sched io_schedule(void) | |
4956 | { | |
54d35f29 | 4957 | struct rq *rq = raw_rq(); |
1da177e4 | 4958 | |
0ff92245 | 4959 | delayacct_blkio_start(); |
1da177e4 | 4960 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 4961 | current->in_iowait = 1; |
1da177e4 | 4962 | schedule(); |
8f0dfc34 | 4963 | current->in_iowait = 0; |
1da177e4 | 4964 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4965 | delayacct_blkio_end(); |
1da177e4 | 4966 | } |
1da177e4 LT |
4967 | EXPORT_SYMBOL(io_schedule); |
4968 | ||
4969 | long __sched io_schedule_timeout(long timeout) | |
4970 | { | |
54d35f29 | 4971 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4972 | long ret; |
4973 | ||
0ff92245 | 4974 | delayacct_blkio_start(); |
1da177e4 | 4975 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 4976 | current->in_iowait = 1; |
1da177e4 | 4977 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4978 | current->in_iowait = 0; |
1da177e4 | 4979 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4980 | delayacct_blkio_end(); |
1da177e4 LT |
4981 | return ret; |
4982 | } | |
4983 | ||
4984 | /** | |
4985 | * sys_sched_get_priority_max - return maximum RT priority. | |
4986 | * @policy: scheduling class. | |
4987 | * | |
4988 | * this syscall returns the maximum rt_priority that can be used | |
4989 | * by a given scheduling class. | |
4990 | */ | |
5add95d4 | 4991 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4992 | { |
4993 | int ret = -EINVAL; | |
4994 | ||
4995 | switch (policy) { | |
4996 | case SCHED_FIFO: | |
4997 | case SCHED_RR: | |
4998 | ret = MAX_USER_RT_PRIO-1; | |
4999 | break; | |
5000 | case SCHED_NORMAL: | |
b0a9499c | 5001 | case SCHED_BATCH: |
dd41f596 | 5002 | case SCHED_IDLE: |
1da177e4 LT |
5003 | ret = 0; |
5004 | break; | |
5005 | } | |
5006 | return ret; | |
5007 | } | |
5008 | ||
5009 | /** | |
5010 | * sys_sched_get_priority_min - return minimum RT priority. | |
5011 | * @policy: scheduling class. | |
5012 | * | |
5013 | * this syscall returns the minimum rt_priority that can be used | |
5014 | * by a given scheduling class. | |
5015 | */ | |
5add95d4 | 5016 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5017 | { |
5018 | int ret = -EINVAL; | |
5019 | ||
5020 | switch (policy) { | |
5021 | case SCHED_FIFO: | |
5022 | case SCHED_RR: | |
5023 | ret = 1; | |
5024 | break; | |
5025 | case SCHED_NORMAL: | |
b0a9499c | 5026 | case SCHED_BATCH: |
dd41f596 | 5027 | case SCHED_IDLE: |
1da177e4 LT |
5028 | ret = 0; |
5029 | } | |
5030 | return ret; | |
5031 | } | |
5032 | ||
5033 | /** | |
5034 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5035 | * @pid: pid of the process. | |
5036 | * @interval: userspace pointer to the timeslice value. | |
5037 | * | |
5038 | * this syscall writes the default timeslice value of a given process | |
5039 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5040 | */ | |
17da2bd9 | 5041 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5042 | struct timespec __user *, interval) |
1da177e4 | 5043 | { |
36c8b586 | 5044 | struct task_struct *p; |
a4ec24b4 | 5045 | unsigned int time_slice; |
dba091b9 TG |
5046 | unsigned long flags; |
5047 | struct rq *rq; | |
3a5c359a | 5048 | int retval; |
1da177e4 | 5049 | struct timespec t; |
1da177e4 LT |
5050 | |
5051 | if (pid < 0) | |
3a5c359a | 5052 | return -EINVAL; |
1da177e4 LT |
5053 | |
5054 | retval = -ESRCH; | |
1a551ae7 | 5055 | rcu_read_lock(); |
1da177e4 LT |
5056 | p = find_process_by_pid(pid); |
5057 | if (!p) | |
5058 | goto out_unlock; | |
5059 | ||
5060 | retval = security_task_getscheduler(p); | |
5061 | if (retval) | |
5062 | goto out_unlock; | |
5063 | ||
dba091b9 TG |
5064 | rq = task_rq_lock(p, &flags); |
5065 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
5066 | task_rq_unlock(rq, &flags); | |
a4ec24b4 | 5067 | |
1a551ae7 | 5068 | rcu_read_unlock(); |
a4ec24b4 | 5069 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5070 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5071 | return retval; |
3a5c359a | 5072 | |
1da177e4 | 5073 | out_unlock: |
1a551ae7 | 5074 | rcu_read_unlock(); |
1da177e4 LT |
5075 | return retval; |
5076 | } | |
5077 | ||
7c731e0a | 5078 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5079 | |
82a1fcb9 | 5080 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5081 | { |
1da177e4 | 5082 | unsigned long free = 0; |
36c8b586 | 5083 | unsigned state; |
1da177e4 | 5084 | |
1da177e4 | 5085 | state = p->state ? __ffs(p->state) + 1 : 0; |
3df0fc5b | 5086 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5087 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5088 | #if BITS_PER_LONG == 32 |
1da177e4 | 5089 | if (state == TASK_RUNNING) |
3df0fc5b | 5090 | printk(KERN_CONT " running "); |
1da177e4 | 5091 | else |
3df0fc5b | 5092 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5093 | #else |
5094 | if (state == TASK_RUNNING) | |
3df0fc5b | 5095 | printk(KERN_CONT " running task "); |
1da177e4 | 5096 | else |
3df0fc5b | 5097 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5098 | #endif |
5099 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5100 | free = stack_not_used(p); |
1da177e4 | 5101 | #endif |
3df0fc5b | 5102 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
5103 | task_pid_nr(p), task_pid_nr(p->real_parent), |
5104 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 5105 | |
5fb5e6de | 5106 | show_stack(p, NULL); |
1da177e4 LT |
5107 | } |
5108 | ||
e59e2ae2 | 5109 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5110 | { |
36c8b586 | 5111 | struct task_struct *g, *p; |
1da177e4 | 5112 | |
4bd77321 | 5113 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5114 | printk(KERN_INFO |
5115 | " task PC stack pid father\n"); | |
1da177e4 | 5116 | #else |
3df0fc5b PZ |
5117 | printk(KERN_INFO |
5118 | " task PC stack pid father\n"); | |
1da177e4 LT |
5119 | #endif |
5120 | read_lock(&tasklist_lock); | |
5121 | do_each_thread(g, p) { | |
5122 | /* | |
5123 | * reset the NMI-timeout, listing all files on a slow | |
5124 | * console might take alot of time: | |
5125 | */ | |
5126 | touch_nmi_watchdog(); | |
39bc89fd | 5127 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5128 | sched_show_task(p); |
1da177e4 LT |
5129 | } while_each_thread(g, p); |
5130 | ||
04c9167f JF |
5131 | touch_all_softlockup_watchdogs(); |
5132 | ||
dd41f596 IM |
5133 | #ifdef CONFIG_SCHED_DEBUG |
5134 | sysrq_sched_debug_show(); | |
5135 | #endif | |
1da177e4 | 5136 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5137 | /* |
5138 | * Only show locks if all tasks are dumped: | |
5139 | */ | |
93335a21 | 5140 | if (!state_filter) |
e59e2ae2 | 5141 | debug_show_all_locks(); |
1da177e4 LT |
5142 | } |
5143 | ||
1df21055 IM |
5144 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5145 | { | |
dd41f596 | 5146 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5147 | } |
5148 | ||
f340c0d1 IM |
5149 | /** |
5150 | * init_idle - set up an idle thread for a given CPU | |
5151 | * @idle: task in question | |
5152 | * @cpu: cpu the idle task belongs to | |
5153 | * | |
5154 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5155 | * flag, to make booting more robust. | |
5156 | */ | |
5c1e1767 | 5157 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5158 | { |
70b97a7f | 5159 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5160 | unsigned long flags; |
5161 | ||
05fa785c | 5162 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 5163 | |
dd41f596 | 5164 | __sched_fork(idle); |
06b83b5f | 5165 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5166 | idle->se.exec_start = sched_clock(); |
5167 | ||
96f874e2 | 5168 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 5169 | __set_task_cpu(idle, cpu); |
1da177e4 | 5170 | |
1da177e4 | 5171 | rq->curr = rq->idle = idle; |
4866cde0 NP |
5172 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5173 | idle->oncpu = 1; | |
5174 | #endif | |
05fa785c | 5175 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5176 | |
5177 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5178 | #if defined(CONFIG_PREEMPT) |
5179 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5180 | #else | |
a1261f54 | 5181 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5182 | #endif |
dd41f596 IM |
5183 | /* |
5184 | * The idle tasks have their own, simple scheduling class: | |
5185 | */ | |
5186 | idle->sched_class = &idle_sched_class; | |
fb52607a | 5187 | ftrace_graph_init_task(idle); |
1da177e4 LT |
5188 | } |
5189 | ||
5190 | /* | |
5191 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5192 | * indicates which cpus entered this state. This is used | |
5193 | * in the rcu update to wait only for active cpus. For system | |
5194 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 5195 | * always be CPU_BITS_NONE. |
1da177e4 | 5196 | */ |
6a7b3dc3 | 5197 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 5198 | |
19978ca6 IM |
5199 | /* |
5200 | * Increase the granularity value when there are more CPUs, | |
5201 | * because with more CPUs the 'effective latency' as visible | |
5202 | * to users decreases. But the relationship is not linear, | |
5203 | * so pick a second-best guess by going with the log2 of the | |
5204 | * number of CPUs. | |
5205 | * | |
5206 | * This idea comes from the SD scheduler of Con Kolivas: | |
5207 | */ | |
acb4a848 | 5208 | static int get_update_sysctl_factor(void) |
19978ca6 | 5209 | { |
4ca3ef71 | 5210 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
5211 | unsigned int factor; |
5212 | ||
5213 | switch (sysctl_sched_tunable_scaling) { | |
5214 | case SCHED_TUNABLESCALING_NONE: | |
5215 | factor = 1; | |
5216 | break; | |
5217 | case SCHED_TUNABLESCALING_LINEAR: | |
5218 | factor = cpus; | |
5219 | break; | |
5220 | case SCHED_TUNABLESCALING_LOG: | |
5221 | default: | |
5222 | factor = 1 + ilog2(cpus); | |
5223 | break; | |
5224 | } | |
19978ca6 | 5225 | |
acb4a848 CE |
5226 | return factor; |
5227 | } | |
19978ca6 | 5228 | |
acb4a848 CE |
5229 | static void update_sysctl(void) |
5230 | { | |
5231 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 5232 | |
0bcdcf28 CE |
5233 | #define SET_SYSCTL(name) \ |
5234 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
5235 | SET_SYSCTL(sched_min_granularity); | |
5236 | SET_SYSCTL(sched_latency); | |
5237 | SET_SYSCTL(sched_wakeup_granularity); | |
5238 | SET_SYSCTL(sched_shares_ratelimit); | |
5239 | #undef SET_SYSCTL | |
5240 | } | |
55cd5340 | 5241 | |
0bcdcf28 CE |
5242 | static inline void sched_init_granularity(void) |
5243 | { | |
5244 | update_sysctl(); | |
19978ca6 IM |
5245 | } |
5246 | ||
1da177e4 LT |
5247 | #ifdef CONFIG_SMP |
5248 | /* | |
5249 | * This is how migration works: | |
5250 | * | |
969c7921 TH |
5251 | * 1) we invoke migration_cpu_stop() on the target CPU using |
5252 | * stop_one_cpu(). | |
5253 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
5254 | * off the CPU) | |
5255 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
5256 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 5257 | * it and puts it into the right queue. |
969c7921 TH |
5258 | * 5) stopper completes and stop_one_cpu() returns and the migration |
5259 | * is done. | |
1da177e4 LT |
5260 | */ |
5261 | ||
5262 | /* | |
5263 | * Change a given task's CPU affinity. Migrate the thread to a | |
5264 | * proper CPU and schedule it away if the CPU it's executing on | |
5265 | * is removed from the allowed bitmask. | |
5266 | * | |
5267 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5268 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5269 | * call is not atomic; no spinlocks may be held. |
5270 | */ | |
96f874e2 | 5271 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
5272 | { |
5273 | unsigned long flags; | |
70b97a7f | 5274 | struct rq *rq; |
969c7921 | 5275 | unsigned int dest_cpu; |
48f24c4d | 5276 | int ret = 0; |
1da177e4 | 5277 | |
65cc8e48 PZ |
5278 | /* |
5279 | * Serialize against TASK_WAKING so that ttwu() and wunt() can | |
5280 | * drop the rq->lock and still rely on ->cpus_allowed. | |
5281 | */ | |
5282 | again: | |
5283 | while (task_is_waking(p)) | |
5284 | cpu_relax(); | |
1da177e4 | 5285 | rq = task_rq_lock(p, &flags); |
65cc8e48 PZ |
5286 | if (task_is_waking(p)) { |
5287 | task_rq_unlock(rq, &flags); | |
5288 | goto again; | |
5289 | } | |
e2912009 | 5290 | |
6ad4c188 | 5291 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
5292 | ret = -EINVAL; |
5293 | goto out; | |
5294 | } | |
5295 | ||
9985b0ba | 5296 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 5297 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
5298 | ret = -EINVAL; |
5299 | goto out; | |
5300 | } | |
5301 | ||
73fe6aae | 5302 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5303 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5304 | else { |
96f874e2 RR |
5305 | cpumask_copy(&p->cpus_allowed, new_mask); |
5306 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
5307 | } |
5308 | ||
1da177e4 | 5309 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 5310 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
5311 | goto out; |
5312 | ||
969c7921 TH |
5313 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
5314 | if (migrate_task(p, dest_cpu)) { | |
5315 | struct migration_arg arg = { p, dest_cpu }; | |
1da177e4 LT |
5316 | /* Need help from migration thread: drop lock and wait. */ |
5317 | task_rq_unlock(rq, &flags); | |
969c7921 | 5318 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
5319 | tlb_migrate_finish(p->mm); |
5320 | return 0; | |
5321 | } | |
5322 | out: | |
5323 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5324 | |
1da177e4 LT |
5325 | return ret; |
5326 | } | |
cd8ba7cd | 5327 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5328 | |
5329 | /* | |
41a2d6cf | 5330 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5331 | * this because either it can't run here any more (set_cpus_allowed() |
5332 | * away from this CPU, or CPU going down), or because we're | |
5333 | * attempting to rebalance this task on exec (sched_exec). | |
5334 | * | |
5335 | * So we race with normal scheduler movements, but that's OK, as long | |
5336 | * as the task is no longer on this CPU. | |
efc30814 KK |
5337 | * |
5338 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5339 | */ |
efc30814 | 5340 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5341 | { |
70b97a7f | 5342 | struct rq *rq_dest, *rq_src; |
e2912009 | 5343 | int ret = 0; |
1da177e4 | 5344 | |
e761b772 | 5345 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5346 | return ret; |
1da177e4 LT |
5347 | |
5348 | rq_src = cpu_rq(src_cpu); | |
5349 | rq_dest = cpu_rq(dest_cpu); | |
5350 | ||
5351 | double_rq_lock(rq_src, rq_dest); | |
5352 | /* Already moved. */ | |
5353 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5354 | goto done; |
1da177e4 | 5355 | /* Affinity changed (again). */ |
96f874e2 | 5356 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 5357 | goto fail; |
1da177e4 | 5358 | |
e2912009 PZ |
5359 | /* |
5360 | * If we're not on a rq, the next wake-up will ensure we're | |
5361 | * placed properly. | |
5362 | */ | |
5363 | if (p->se.on_rq) { | |
2e1cb74a | 5364 | deactivate_task(rq_src, p, 0); |
e2912009 | 5365 | set_task_cpu(p, dest_cpu); |
dd41f596 | 5366 | activate_task(rq_dest, p, 0); |
15afe09b | 5367 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 5368 | } |
b1e38734 | 5369 | done: |
efc30814 | 5370 | ret = 1; |
b1e38734 | 5371 | fail: |
1da177e4 | 5372 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 5373 | return ret; |
1da177e4 LT |
5374 | } |
5375 | ||
5376 | /* | |
969c7921 TH |
5377 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
5378 | * and performs thread migration by bumping thread off CPU then | |
5379 | * 'pushing' onto another runqueue. | |
1da177e4 | 5380 | */ |
969c7921 | 5381 | static int migration_cpu_stop(void *data) |
1da177e4 | 5382 | { |
969c7921 | 5383 | struct migration_arg *arg = data; |
f7b4cddc | 5384 | |
969c7921 TH |
5385 | /* |
5386 | * The original target cpu might have gone down and we might | |
5387 | * be on another cpu but it doesn't matter. | |
5388 | */ | |
f7b4cddc | 5389 | local_irq_disable(); |
969c7921 | 5390 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 5391 | local_irq_enable(); |
1da177e4 | 5392 | return 0; |
f7b4cddc ON |
5393 | } |
5394 | ||
1da177e4 | 5395 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 5396 | /* |
3a4fa0a2 | 5397 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 5398 | */ |
6a1bdc1b | 5399 | void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5400 | { |
1445c08d ON |
5401 | struct rq *rq = cpu_rq(dead_cpu); |
5402 | int needs_cpu, uninitialized_var(dest_cpu); | |
5403 | unsigned long flags; | |
e76bd8d9 | 5404 | |
1445c08d | 5405 | local_irq_save(flags); |
e76bd8d9 | 5406 | |
1445c08d ON |
5407 | raw_spin_lock(&rq->lock); |
5408 | needs_cpu = (task_cpu(p) == dead_cpu) && (p->state != TASK_WAKING); | |
5409 | if (needs_cpu) | |
5410 | dest_cpu = select_fallback_rq(dead_cpu, p); | |
5411 | raw_spin_unlock(&rq->lock); | |
c1804d54 ON |
5412 | /* |
5413 | * It can only fail if we race with set_cpus_allowed(), | |
5414 | * in the racer should migrate the task anyway. | |
5415 | */ | |
1445c08d | 5416 | if (needs_cpu) |
c1804d54 | 5417 | __migrate_task(p, dead_cpu, dest_cpu); |
1445c08d | 5418 | local_irq_restore(flags); |
1da177e4 LT |
5419 | } |
5420 | ||
5421 | /* | |
5422 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5423 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5424 | * for performance reasons the counter is not stricly tracking tasks to | |
5425 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5426 | * to keep the global sum constant after CPU-down: | |
5427 | */ | |
70b97a7f | 5428 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5429 | { |
6ad4c188 | 5430 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 LT |
5431 | unsigned long flags; |
5432 | ||
5433 | local_irq_save(flags); | |
5434 | double_rq_lock(rq_src, rq_dest); | |
5435 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5436 | rq_src->nr_uninterruptible = 0; | |
5437 | double_rq_unlock(rq_src, rq_dest); | |
5438 | local_irq_restore(flags); | |
5439 | } | |
5440 | ||
5441 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5442 | static void migrate_live_tasks(int src_cpu) | |
5443 | { | |
48f24c4d | 5444 | struct task_struct *p, *t; |
1da177e4 | 5445 | |
f7b4cddc | 5446 | read_lock(&tasklist_lock); |
1da177e4 | 5447 | |
48f24c4d IM |
5448 | do_each_thread(t, p) { |
5449 | if (p == current) | |
1da177e4 LT |
5450 | continue; |
5451 | ||
48f24c4d IM |
5452 | if (task_cpu(p) == src_cpu) |
5453 | move_task_off_dead_cpu(src_cpu, p); | |
5454 | } while_each_thread(t, p); | |
1da177e4 | 5455 | |
f7b4cddc | 5456 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5457 | } |
5458 | ||
dd41f596 IM |
5459 | /* |
5460 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5461 | * It does so by boosting its priority to highest possible. |
5462 | * Used by CPU offline code. | |
1da177e4 LT |
5463 | */ |
5464 | void sched_idle_next(void) | |
5465 | { | |
48f24c4d | 5466 | int this_cpu = smp_processor_id(); |
70b97a7f | 5467 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5468 | struct task_struct *p = rq->idle; |
5469 | unsigned long flags; | |
5470 | ||
5471 | /* cpu has to be offline */ | |
48f24c4d | 5472 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5473 | |
48f24c4d IM |
5474 | /* |
5475 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5476 | * and interrupts disabled on the current cpu. | |
1da177e4 | 5477 | */ |
05fa785c | 5478 | raw_spin_lock_irqsave(&rq->lock, flags); |
1da177e4 | 5479 | |
dd41f596 | 5480 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 5481 | |
94bc9a7b | 5482 | activate_task(rq, p, 0); |
1da177e4 | 5483 | |
05fa785c | 5484 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5485 | } |
5486 | ||
48f24c4d IM |
5487 | /* |
5488 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5489 | * offline. |
5490 | */ | |
5491 | void idle_task_exit(void) | |
5492 | { | |
5493 | struct mm_struct *mm = current->active_mm; | |
5494 | ||
5495 | BUG_ON(cpu_online(smp_processor_id())); | |
5496 | ||
5497 | if (mm != &init_mm) | |
5498 | switch_mm(mm, &init_mm, current); | |
5499 | mmdrop(mm); | |
5500 | } | |
5501 | ||
054b9108 | 5502 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5503 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5504 | { |
70b97a7f | 5505 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5506 | |
5507 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5508 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5509 | |
5510 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5511 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5512 | |
48f24c4d | 5513 | get_task_struct(p); |
1da177e4 LT |
5514 | |
5515 | /* | |
5516 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 5517 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
5518 | * fine. |
5519 | */ | |
05fa785c | 5520 | raw_spin_unlock_irq(&rq->lock); |
48f24c4d | 5521 | move_task_off_dead_cpu(dead_cpu, p); |
05fa785c | 5522 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 5523 | |
48f24c4d | 5524 | put_task_struct(p); |
1da177e4 LT |
5525 | } |
5526 | ||
5527 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5528 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5529 | { | |
70b97a7f | 5530 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5531 | struct task_struct *next; |
48f24c4d | 5532 | |
dd41f596 IM |
5533 | for ( ; ; ) { |
5534 | if (!rq->nr_running) | |
5535 | break; | |
b67802ea | 5536 | next = pick_next_task(rq); |
dd41f596 IM |
5537 | if (!next) |
5538 | break; | |
79c53799 | 5539 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 5540 | migrate_dead(dead_cpu, next); |
e692ab53 | 5541 | |
1da177e4 LT |
5542 | } |
5543 | } | |
dce48a84 TG |
5544 | |
5545 | /* | |
5546 | * remove the tasks which were accounted by rq from calc_load_tasks. | |
5547 | */ | |
5548 | static void calc_global_load_remove(struct rq *rq) | |
5549 | { | |
5550 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | |
a468d389 | 5551 | rq->calc_load_active = 0; |
dce48a84 | 5552 | } |
1da177e4 LT |
5553 | #endif /* CONFIG_HOTPLUG_CPU */ |
5554 | ||
e692ab53 NP |
5555 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5556 | ||
5557 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5558 | { |
5559 | .procname = "sched_domain", | |
c57baf1e | 5560 | .mode = 0555, |
e0361851 | 5561 | }, |
56992309 | 5562 | {} |
e692ab53 NP |
5563 | }; |
5564 | ||
5565 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5566 | { |
5567 | .procname = "kernel", | |
c57baf1e | 5568 | .mode = 0555, |
e0361851 AD |
5569 | .child = sd_ctl_dir, |
5570 | }, | |
56992309 | 5571 | {} |
e692ab53 NP |
5572 | }; |
5573 | ||
5574 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5575 | { | |
5576 | struct ctl_table *entry = | |
5cf9f062 | 5577 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5578 | |
e692ab53 NP |
5579 | return entry; |
5580 | } | |
5581 | ||
6382bc90 MM |
5582 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5583 | { | |
cd790076 | 5584 | struct ctl_table *entry; |
6382bc90 | 5585 | |
cd790076 MM |
5586 | /* |
5587 | * In the intermediate directories, both the child directory and | |
5588 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5589 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5590 | * static strings and all have proc handlers. |
5591 | */ | |
5592 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5593 | if (entry->child) |
5594 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5595 | if (entry->proc_handler == NULL) |
5596 | kfree(entry->procname); | |
5597 | } | |
6382bc90 MM |
5598 | |
5599 | kfree(*tablep); | |
5600 | *tablep = NULL; | |
5601 | } | |
5602 | ||
e692ab53 | 5603 | static void |
e0361851 | 5604 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5605 | const char *procname, void *data, int maxlen, |
5606 | mode_t mode, proc_handler *proc_handler) | |
5607 | { | |
e692ab53 NP |
5608 | entry->procname = procname; |
5609 | entry->data = data; | |
5610 | entry->maxlen = maxlen; | |
5611 | entry->mode = mode; | |
5612 | entry->proc_handler = proc_handler; | |
5613 | } | |
5614 | ||
5615 | static struct ctl_table * | |
5616 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5617 | { | |
a5d8c348 | 5618 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 5619 | |
ad1cdc1d MM |
5620 | if (table == NULL) |
5621 | return NULL; | |
5622 | ||
e0361851 | 5623 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5624 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5625 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5626 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5627 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5628 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5629 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5630 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5631 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5632 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5633 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5634 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5635 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5636 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5637 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5638 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5639 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5640 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5641 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5642 | &sd->cache_nice_tries, |
5643 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5644 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5645 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
5646 | set_table_entry(&table[11], "name", sd->name, |
5647 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
5648 | /* &table[12] is terminator */ | |
e692ab53 NP |
5649 | |
5650 | return table; | |
5651 | } | |
5652 | ||
9a4e7159 | 5653 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5654 | { |
5655 | struct ctl_table *entry, *table; | |
5656 | struct sched_domain *sd; | |
5657 | int domain_num = 0, i; | |
5658 | char buf[32]; | |
5659 | ||
5660 | for_each_domain(cpu, sd) | |
5661 | domain_num++; | |
5662 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5663 | if (table == NULL) |
5664 | return NULL; | |
e692ab53 NP |
5665 | |
5666 | i = 0; | |
5667 | for_each_domain(cpu, sd) { | |
5668 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5669 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5670 | entry->mode = 0555; |
e692ab53 NP |
5671 | entry->child = sd_alloc_ctl_domain_table(sd); |
5672 | entry++; | |
5673 | i++; | |
5674 | } | |
5675 | return table; | |
5676 | } | |
5677 | ||
5678 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5679 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5680 | { |
6ad4c188 | 5681 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5682 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5683 | char buf[32]; | |
5684 | ||
7378547f MM |
5685 | WARN_ON(sd_ctl_dir[0].child); |
5686 | sd_ctl_dir[0].child = entry; | |
5687 | ||
ad1cdc1d MM |
5688 | if (entry == NULL) |
5689 | return; | |
5690 | ||
6ad4c188 | 5691 | for_each_possible_cpu(i) { |
e692ab53 | 5692 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5693 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5694 | entry->mode = 0555; |
e692ab53 | 5695 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5696 | entry++; |
e692ab53 | 5697 | } |
7378547f MM |
5698 | |
5699 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5700 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5701 | } | |
6382bc90 | 5702 | |
7378547f | 5703 | /* may be called multiple times per register */ |
6382bc90 MM |
5704 | static void unregister_sched_domain_sysctl(void) |
5705 | { | |
7378547f MM |
5706 | if (sd_sysctl_header) |
5707 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5708 | sd_sysctl_header = NULL; |
7378547f MM |
5709 | if (sd_ctl_dir[0].child) |
5710 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5711 | } |
e692ab53 | 5712 | #else |
6382bc90 MM |
5713 | static void register_sched_domain_sysctl(void) |
5714 | { | |
5715 | } | |
5716 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5717 | { |
5718 | } | |
5719 | #endif | |
5720 | ||
1f11eb6a GH |
5721 | static void set_rq_online(struct rq *rq) |
5722 | { | |
5723 | if (!rq->online) { | |
5724 | const struct sched_class *class; | |
5725 | ||
c6c4927b | 5726 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5727 | rq->online = 1; |
5728 | ||
5729 | for_each_class(class) { | |
5730 | if (class->rq_online) | |
5731 | class->rq_online(rq); | |
5732 | } | |
5733 | } | |
5734 | } | |
5735 | ||
5736 | static void set_rq_offline(struct rq *rq) | |
5737 | { | |
5738 | if (rq->online) { | |
5739 | const struct sched_class *class; | |
5740 | ||
5741 | for_each_class(class) { | |
5742 | if (class->rq_offline) | |
5743 | class->rq_offline(rq); | |
5744 | } | |
5745 | ||
c6c4927b | 5746 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5747 | rq->online = 0; |
5748 | } | |
5749 | } | |
5750 | ||
1da177e4 LT |
5751 | /* |
5752 | * migration_call - callback that gets triggered when a CPU is added. | |
5753 | * Here we can start up the necessary migration thread for the new CPU. | |
5754 | */ | |
48f24c4d IM |
5755 | static int __cpuinit |
5756 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5757 | { |
48f24c4d | 5758 | int cpu = (long)hcpu; |
1da177e4 | 5759 | unsigned long flags; |
969c7921 | 5760 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5761 | |
5762 | switch (action) { | |
5be9361c | 5763 | |
1da177e4 | 5764 | case CPU_UP_PREPARE: |
8bb78442 | 5765 | case CPU_UP_PREPARE_FROZEN: |
a468d389 | 5766 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5767 | break; |
48f24c4d | 5768 | |
1da177e4 | 5769 | case CPU_ONLINE: |
8bb78442 | 5770 | case CPU_ONLINE_FROZEN: |
1f94ef59 | 5771 | /* Update our root-domain */ |
05fa785c | 5772 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5773 | if (rq->rd) { |
c6c4927b | 5774 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5775 | |
5776 | set_rq_online(rq); | |
1f94ef59 | 5777 | } |
05fa785c | 5778 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5779 | break; |
48f24c4d | 5780 | |
1da177e4 | 5781 | #ifdef CONFIG_HOTPLUG_CPU |
1da177e4 | 5782 | case CPU_DEAD: |
8bb78442 | 5783 | case CPU_DEAD_FROZEN: |
1da177e4 | 5784 | migrate_live_tasks(cpu); |
1da177e4 | 5785 | /* Idle task back to normal (off runqueue, low prio) */ |
05fa785c | 5786 | raw_spin_lock_irq(&rq->lock); |
2e1cb74a | 5787 | deactivate_task(rq, rq->idle, 0); |
dd41f596 IM |
5788 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5789 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 5790 | migrate_dead_tasks(cpu); |
05fa785c | 5791 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 LT |
5792 | migrate_nr_uninterruptible(rq); |
5793 | BUG_ON(rq->nr_running != 0); | |
dce48a84 | 5794 | calc_global_load_remove(rq); |
1da177e4 | 5795 | break; |
57d885fe | 5796 | |
08f503b0 GH |
5797 | case CPU_DYING: |
5798 | case CPU_DYING_FROZEN: | |
57d885fe | 5799 | /* Update our root-domain */ |
05fa785c | 5800 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5801 | if (rq->rd) { |
c6c4927b | 5802 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5803 | set_rq_offline(rq); |
57d885fe | 5804 | } |
05fa785c | 5805 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
57d885fe | 5806 | break; |
1da177e4 LT |
5807 | #endif |
5808 | } | |
5809 | return NOTIFY_OK; | |
5810 | } | |
5811 | ||
f38b0820 PM |
5812 | /* |
5813 | * Register at high priority so that task migration (migrate_all_tasks) | |
5814 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5815 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5816 | */ |
26c2143b | 5817 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5818 | .notifier_call = migration_call, |
5819 | .priority = 10 | |
5820 | }; | |
5821 | ||
7babe8db | 5822 | static int __init migration_init(void) |
1da177e4 LT |
5823 | { |
5824 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5825 | int err; |
48f24c4d IM |
5826 | |
5827 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5828 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5829 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5830 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5831 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5832 | |
a004cd42 | 5833 | return 0; |
1da177e4 | 5834 | } |
7babe8db | 5835 | early_initcall(migration_init); |
1da177e4 LT |
5836 | #endif |
5837 | ||
5838 | #ifdef CONFIG_SMP | |
476f3534 | 5839 | |
3e9830dc | 5840 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5841 | |
f6630114 MT |
5842 | static __read_mostly int sched_domain_debug_enabled; |
5843 | ||
5844 | static int __init sched_domain_debug_setup(char *str) | |
5845 | { | |
5846 | sched_domain_debug_enabled = 1; | |
5847 | ||
5848 | return 0; | |
5849 | } | |
5850 | early_param("sched_debug", sched_domain_debug_setup); | |
5851 | ||
7c16ec58 | 5852 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5853 | struct cpumask *groupmask) |
1da177e4 | 5854 | { |
4dcf6aff | 5855 | struct sched_group *group = sd->groups; |
434d53b0 | 5856 | char str[256]; |
1da177e4 | 5857 | |
968ea6d8 | 5858 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5859 | cpumask_clear(groupmask); |
4dcf6aff IM |
5860 | |
5861 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5862 | ||
5863 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5864 | printk("does not load-balance\n"); |
4dcf6aff | 5865 | if (sd->parent) |
3df0fc5b PZ |
5866 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5867 | " has parent"); | |
4dcf6aff | 5868 | return -1; |
41c7ce9a NP |
5869 | } |
5870 | ||
3df0fc5b | 5871 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5872 | |
758b2cdc | 5873 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5874 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5875 | "CPU%d\n", cpu); | |
4dcf6aff | 5876 | } |
758b2cdc | 5877 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5878 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5879 | " CPU%d\n", cpu); | |
4dcf6aff | 5880 | } |
1da177e4 | 5881 | |
4dcf6aff | 5882 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5883 | do { |
4dcf6aff | 5884 | if (!group) { |
3df0fc5b PZ |
5885 | printk("\n"); |
5886 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5887 | break; |
5888 | } | |
5889 | ||
18a3885f | 5890 | if (!group->cpu_power) { |
3df0fc5b PZ |
5891 | printk(KERN_CONT "\n"); |
5892 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5893 | "set\n"); | |
4dcf6aff IM |
5894 | break; |
5895 | } | |
1da177e4 | 5896 | |
758b2cdc | 5897 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5898 | printk(KERN_CONT "\n"); |
5899 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5900 | break; |
5901 | } | |
1da177e4 | 5902 | |
758b2cdc | 5903 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
5904 | printk(KERN_CONT "\n"); |
5905 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5906 | break; |
5907 | } | |
1da177e4 | 5908 | |
758b2cdc | 5909 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5910 | |
968ea6d8 | 5911 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 5912 | |
3df0fc5b | 5913 | printk(KERN_CONT " %s", str); |
18a3885f | 5914 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
3df0fc5b PZ |
5915 | printk(KERN_CONT " (cpu_power = %d)", |
5916 | group->cpu_power); | |
381512cf | 5917 | } |
1da177e4 | 5918 | |
4dcf6aff IM |
5919 | group = group->next; |
5920 | } while (group != sd->groups); | |
3df0fc5b | 5921 | printk(KERN_CONT "\n"); |
1da177e4 | 5922 | |
758b2cdc | 5923 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5924 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5925 | |
758b2cdc RR |
5926 | if (sd->parent && |
5927 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5928 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5929 | "of domain->span\n"); | |
4dcf6aff IM |
5930 | return 0; |
5931 | } | |
1da177e4 | 5932 | |
4dcf6aff IM |
5933 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5934 | { | |
d5dd3db1 | 5935 | cpumask_var_t groupmask; |
4dcf6aff | 5936 | int level = 0; |
1da177e4 | 5937 | |
f6630114 MT |
5938 | if (!sched_domain_debug_enabled) |
5939 | return; | |
5940 | ||
4dcf6aff IM |
5941 | if (!sd) { |
5942 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5943 | return; | |
5944 | } | |
1da177e4 | 5945 | |
4dcf6aff IM |
5946 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5947 | ||
d5dd3db1 | 5948 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
5949 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
5950 | return; | |
5951 | } | |
5952 | ||
4dcf6aff | 5953 | for (;;) { |
7c16ec58 | 5954 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 5955 | break; |
1da177e4 LT |
5956 | level++; |
5957 | sd = sd->parent; | |
33859f7f | 5958 | if (!sd) |
4dcf6aff IM |
5959 | break; |
5960 | } | |
d5dd3db1 | 5961 | free_cpumask_var(groupmask); |
1da177e4 | 5962 | } |
6d6bc0ad | 5963 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5964 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 5965 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5966 | |
1a20ff27 | 5967 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5968 | { |
758b2cdc | 5969 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5970 | return 1; |
5971 | ||
5972 | /* Following flags need at least 2 groups */ | |
5973 | if (sd->flags & (SD_LOAD_BALANCE | | |
5974 | SD_BALANCE_NEWIDLE | | |
5975 | SD_BALANCE_FORK | | |
89c4710e SS |
5976 | SD_BALANCE_EXEC | |
5977 | SD_SHARE_CPUPOWER | | |
5978 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5979 | if (sd->groups != sd->groups->next) |
5980 | return 0; | |
5981 | } | |
5982 | ||
5983 | /* Following flags don't use groups */ | |
c88d5910 | 5984 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5985 | return 0; |
5986 | ||
5987 | return 1; | |
5988 | } | |
5989 | ||
48f24c4d IM |
5990 | static int |
5991 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5992 | { |
5993 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5994 | ||
5995 | if (sd_degenerate(parent)) | |
5996 | return 1; | |
5997 | ||
758b2cdc | 5998 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5999 | return 0; |
6000 | ||
245af2c7 SS |
6001 | /* Flags needing groups don't count if only 1 group in parent */ |
6002 | if (parent->groups == parent->groups->next) { | |
6003 | pflags &= ~(SD_LOAD_BALANCE | | |
6004 | SD_BALANCE_NEWIDLE | | |
6005 | SD_BALANCE_FORK | | |
89c4710e SS |
6006 | SD_BALANCE_EXEC | |
6007 | SD_SHARE_CPUPOWER | | |
6008 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6009 | if (nr_node_ids == 1) |
6010 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6011 | } |
6012 | if (~cflags & pflags) | |
6013 | return 0; | |
6014 | ||
6015 | return 1; | |
6016 | } | |
6017 | ||
c6c4927b RR |
6018 | static void free_rootdomain(struct root_domain *rd) |
6019 | { | |
047106ad PZ |
6020 | synchronize_sched(); |
6021 | ||
68e74568 RR |
6022 | cpupri_cleanup(&rd->cpupri); |
6023 | ||
c6c4927b RR |
6024 | free_cpumask_var(rd->rto_mask); |
6025 | free_cpumask_var(rd->online); | |
6026 | free_cpumask_var(rd->span); | |
6027 | kfree(rd); | |
6028 | } | |
6029 | ||
57d885fe GH |
6030 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6031 | { | |
a0490fa3 | 6032 | struct root_domain *old_rd = NULL; |
57d885fe | 6033 | unsigned long flags; |
57d885fe | 6034 | |
05fa785c | 6035 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
6036 | |
6037 | if (rq->rd) { | |
a0490fa3 | 6038 | old_rd = rq->rd; |
57d885fe | 6039 | |
c6c4927b | 6040 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6041 | set_rq_offline(rq); |
57d885fe | 6042 | |
c6c4927b | 6043 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6044 | |
a0490fa3 IM |
6045 | /* |
6046 | * If we dont want to free the old_rt yet then | |
6047 | * set old_rd to NULL to skip the freeing later | |
6048 | * in this function: | |
6049 | */ | |
6050 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
6051 | old_rd = NULL; | |
57d885fe GH |
6052 | } |
6053 | ||
6054 | atomic_inc(&rd->refcount); | |
6055 | rq->rd = rd; | |
6056 | ||
c6c4927b | 6057 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 6058 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 6059 | set_rq_online(rq); |
57d885fe | 6060 | |
05fa785c | 6061 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
6062 | |
6063 | if (old_rd) | |
6064 | free_rootdomain(old_rd); | |
57d885fe GH |
6065 | } |
6066 | ||
fd5e1b5d | 6067 | static int init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe | 6068 | { |
36b7b6d4 PE |
6069 | gfp_t gfp = GFP_KERNEL; |
6070 | ||
57d885fe GH |
6071 | memset(rd, 0, sizeof(*rd)); |
6072 | ||
36b7b6d4 PE |
6073 | if (bootmem) |
6074 | gfp = GFP_NOWAIT; | |
c6c4927b | 6075 | |
36b7b6d4 | 6076 | if (!alloc_cpumask_var(&rd->span, gfp)) |
0c910d28 | 6077 | goto out; |
36b7b6d4 | 6078 | if (!alloc_cpumask_var(&rd->online, gfp)) |
c6c4927b | 6079 | goto free_span; |
36b7b6d4 | 6080 | if (!alloc_cpumask_var(&rd->rto_mask, gfp)) |
c6c4927b | 6081 | goto free_online; |
6e0534f2 | 6082 | |
0fb53029 | 6083 | if (cpupri_init(&rd->cpupri, bootmem) != 0) |
68e74568 | 6084 | goto free_rto_mask; |
c6c4927b | 6085 | return 0; |
6e0534f2 | 6086 | |
68e74568 RR |
6087 | free_rto_mask: |
6088 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6089 | free_online: |
6090 | free_cpumask_var(rd->online); | |
6091 | free_span: | |
6092 | free_cpumask_var(rd->span); | |
0c910d28 | 6093 | out: |
c6c4927b | 6094 | return -ENOMEM; |
57d885fe GH |
6095 | } |
6096 | ||
6097 | static void init_defrootdomain(void) | |
6098 | { | |
c6c4927b RR |
6099 | init_rootdomain(&def_root_domain, true); |
6100 | ||
57d885fe GH |
6101 | atomic_set(&def_root_domain.refcount, 1); |
6102 | } | |
6103 | ||
dc938520 | 6104 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6105 | { |
6106 | struct root_domain *rd; | |
6107 | ||
6108 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6109 | if (!rd) | |
6110 | return NULL; | |
6111 | ||
c6c4927b RR |
6112 | if (init_rootdomain(rd, false) != 0) { |
6113 | kfree(rd); | |
6114 | return NULL; | |
6115 | } | |
57d885fe GH |
6116 | |
6117 | return rd; | |
6118 | } | |
6119 | ||
1da177e4 | 6120 | /* |
0eab9146 | 6121 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6122 | * hold the hotplug lock. |
6123 | */ | |
0eab9146 IM |
6124 | static void |
6125 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6126 | { |
70b97a7f | 6127 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6128 | struct sched_domain *tmp; |
6129 | ||
669c55e9 PZ |
6130 | for (tmp = sd; tmp; tmp = tmp->parent) |
6131 | tmp->span_weight = cpumask_weight(sched_domain_span(tmp)); | |
6132 | ||
245af2c7 | 6133 | /* Remove the sched domains which do not contribute to scheduling. */ |
f29c9b1c | 6134 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6135 | struct sched_domain *parent = tmp->parent; |
6136 | if (!parent) | |
6137 | break; | |
f29c9b1c | 6138 | |
1a848870 | 6139 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6140 | tmp->parent = parent->parent; |
1a848870 SS |
6141 | if (parent->parent) |
6142 | parent->parent->child = tmp; | |
f29c9b1c LZ |
6143 | } else |
6144 | tmp = tmp->parent; | |
245af2c7 SS |
6145 | } |
6146 | ||
1a848870 | 6147 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6148 | sd = sd->parent; |
1a848870 SS |
6149 | if (sd) |
6150 | sd->child = NULL; | |
6151 | } | |
1da177e4 LT |
6152 | |
6153 | sched_domain_debug(sd, cpu); | |
6154 | ||
57d885fe | 6155 | rq_attach_root(rq, rd); |
674311d5 | 6156 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6157 | } |
6158 | ||
6159 | /* cpus with isolated domains */ | |
dcc30a35 | 6160 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
6161 | |
6162 | /* Setup the mask of cpus configured for isolated domains */ | |
6163 | static int __init isolated_cpu_setup(char *str) | |
6164 | { | |
bdddd296 | 6165 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 6166 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
6167 | return 1; |
6168 | } | |
6169 | ||
8927f494 | 6170 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6171 | |
6172 | /* | |
6711cab4 SS |
6173 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6174 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
6175 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
6176 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
6177 | * |
6178 | * init_sched_build_groups will build a circular linked list of the groups | |
6179 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6180 | * and ->cpu_power to 0. | |
6181 | */ | |
a616058b | 6182 | static void |
96f874e2 RR |
6183 | init_sched_build_groups(const struct cpumask *span, |
6184 | const struct cpumask *cpu_map, | |
6185 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 6186 | struct sched_group **sg, |
96f874e2 RR |
6187 | struct cpumask *tmpmask), |
6188 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
6189 | { |
6190 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6191 | int i; |
6192 | ||
96f874e2 | 6193 | cpumask_clear(covered); |
7c16ec58 | 6194 | |
abcd083a | 6195 | for_each_cpu(i, span) { |
6711cab4 | 6196 | struct sched_group *sg; |
7c16ec58 | 6197 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6198 | int j; |
6199 | ||
758b2cdc | 6200 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
6201 | continue; |
6202 | ||
758b2cdc | 6203 | cpumask_clear(sched_group_cpus(sg)); |
18a3885f | 6204 | sg->cpu_power = 0; |
1da177e4 | 6205 | |
abcd083a | 6206 | for_each_cpu(j, span) { |
7c16ec58 | 6207 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6208 | continue; |
6209 | ||
96f874e2 | 6210 | cpumask_set_cpu(j, covered); |
758b2cdc | 6211 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
6212 | } |
6213 | if (!first) | |
6214 | first = sg; | |
6215 | if (last) | |
6216 | last->next = sg; | |
6217 | last = sg; | |
6218 | } | |
6219 | last->next = first; | |
6220 | } | |
6221 | ||
9c1cfda2 | 6222 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6223 | |
9c1cfda2 | 6224 | #ifdef CONFIG_NUMA |
198e2f18 | 6225 | |
9c1cfda2 JH |
6226 | /** |
6227 | * find_next_best_node - find the next node to include in a sched_domain | |
6228 | * @node: node whose sched_domain we're building | |
6229 | * @used_nodes: nodes already in the sched_domain | |
6230 | * | |
41a2d6cf | 6231 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6232 | * finds the closest node not already in the @used_nodes map. |
6233 | * | |
6234 | * Should use nodemask_t. | |
6235 | */ | |
c5f59f08 | 6236 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6237 | { |
6238 | int i, n, val, min_val, best_node = 0; | |
6239 | ||
6240 | min_val = INT_MAX; | |
6241 | ||
076ac2af | 6242 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6243 | /* Start at @node */ |
076ac2af | 6244 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6245 | |
6246 | if (!nr_cpus_node(n)) | |
6247 | continue; | |
6248 | ||
6249 | /* Skip already used nodes */ | |
c5f59f08 | 6250 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6251 | continue; |
6252 | ||
6253 | /* Simple min distance search */ | |
6254 | val = node_distance(node, n); | |
6255 | ||
6256 | if (val < min_val) { | |
6257 | min_val = val; | |
6258 | best_node = n; | |
6259 | } | |
6260 | } | |
6261 | ||
c5f59f08 | 6262 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6263 | return best_node; |
6264 | } | |
6265 | ||
6266 | /** | |
6267 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6268 | * @node: node whose cpumask we're constructing | |
73486722 | 6269 | * @span: resulting cpumask |
9c1cfda2 | 6270 | * |
41a2d6cf | 6271 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6272 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6273 | * out optimally. | |
6274 | */ | |
96f874e2 | 6275 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 6276 | { |
c5f59f08 | 6277 | nodemask_t used_nodes; |
48f24c4d | 6278 | int i; |
9c1cfda2 | 6279 | |
6ca09dfc | 6280 | cpumask_clear(span); |
c5f59f08 | 6281 | nodes_clear(used_nodes); |
9c1cfda2 | 6282 | |
6ca09dfc | 6283 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 6284 | node_set(node, used_nodes); |
9c1cfda2 JH |
6285 | |
6286 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6287 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6288 | |
6ca09dfc | 6289 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 6290 | } |
9c1cfda2 | 6291 | } |
6d6bc0ad | 6292 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6293 | |
5c45bf27 | 6294 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6295 | |
6c99e9ad RR |
6296 | /* |
6297 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
6298 | * |
6299 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
6300 | * and struct sched_domain. ) | |
6c99e9ad RR |
6301 | */ |
6302 | struct static_sched_group { | |
6303 | struct sched_group sg; | |
6304 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
6305 | }; | |
6306 | ||
6307 | struct static_sched_domain { | |
6308 | struct sched_domain sd; | |
6309 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
6310 | }; | |
6311 | ||
49a02c51 AH |
6312 | struct s_data { |
6313 | #ifdef CONFIG_NUMA | |
6314 | int sd_allnodes; | |
6315 | cpumask_var_t domainspan; | |
6316 | cpumask_var_t covered; | |
6317 | cpumask_var_t notcovered; | |
6318 | #endif | |
6319 | cpumask_var_t nodemask; | |
6320 | cpumask_var_t this_sibling_map; | |
6321 | cpumask_var_t this_core_map; | |
6322 | cpumask_var_t send_covered; | |
6323 | cpumask_var_t tmpmask; | |
6324 | struct sched_group **sched_group_nodes; | |
6325 | struct root_domain *rd; | |
6326 | }; | |
6327 | ||
2109b99e AH |
6328 | enum s_alloc { |
6329 | sa_sched_groups = 0, | |
6330 | sa_rootdomain, | |
6331 | sa_tmpmask, | |
6332 | sa_send_covered, | |
6333 | sa_this_core_map, | |
6334 | sa_this_sibling_map, | |
6335 | sa_nodemask, | |
6336 | sa_sched_group_nodes, | |
6337 | #ifdef CONFIG_NUMA | |
6338 | sa_notcovered, | |
6339 | sa_covered, | |
6340 | sa_domainspan, | |
6341 | #endif | |
6342 | sa_none, | |
6343 | }; | |
6344 | ||
9c1cfda2 | 6345 | /* |
48f24c4d | 6346 | * SMT sched-domains: |
9c1cfda2 | 6347 | */ |
1da177e4 | 6348 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 6349 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
1871e52c | 6350 | static DEFINE_PER_CPU(struct static_sched_group, sched_groups); |
48f24c4d | 6351 | |
41a2d6cf | 6352 | static int |
96f874e2 RR |
6353 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
6354 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 6355 | { |
6711cab4 | 6356 | if (sg) |
1871e52c | 6357 | *sg = &per_cpu(sched_groups, cpu).sg; |
1da177e4 LT |
6358 | return cpu; |
6359 | } | |
6d6bc0ad | 6360 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6361 | |
48f24c4d IM |
6362 | /* |
6363 | * multi-core sched-domains: | |
6364 | */ | |
1e9f28fa | 6365 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
6366 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
6367 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 6368 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
6369 | |
6370 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 6371 | static int |
96f874e2 RR |
6372 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6373 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6374 | { |
6711cab4 | 6375 | int group; |
7c16ec58 | 6376 | |
c69fc56d | 6377 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6378 | group = cpumask_first(mask); |
6711cab4 | 6379 | if (sg) |
6c99e9ad | 6380 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 6381 | return group; |
1e9f28fa SS |
6382 | } |
6383 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 6384 | static int |
96f874e2 RR |
6385 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6386 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 6387 | { |
6711cab4 | 6388 | if (sg) |
6c99e9ad | 6389 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
6390 | return cpu; |
6391 | } | |
6392 | #endif | |
6393 | ||
6c99e9ad RR |
6394 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
6395 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 6396 | |
41a2d6cf | 6397 | static int |
96f874e2 RR |
6398 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
6399 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 6400 | { |
6711cab4 | 6401 | int group; |
48f24c4d | 6402 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 6403 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 6404 | group = cpumask_first(mask); |
1e9f28fa | 6405 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 6406 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6407 | group = cpumask_first(mask); |
1da177e4 | 6408 | #else |
6711cab4 | 6409 | group = cpu; |
1da177e4 | 6410 | #endif |
6711cab4 | 6411 | if (sg) |
6c99e9ad | 6412 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 6413 | return group; |
1da177e4 LT |
6414 | } |
6415 | ||
6416 | #ifdef CONFIG_NUMA | |
1da177e4 | 6417 | /* |
9c1cfda2 JH |
6418 | * The init_sched_build_groups can't handle what we want to do with node |
6419 | * groups, so roll our own. Now each node has its own list of groups which | |
6420 | * gets dynamically allocated. | |
1da177e4 | 6421 | */ |
62ea9ceb | 6422 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 6423 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6424 | |
62ea9ceb | 6425 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 6426 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 6427 | |
96f874e2 RR |
6428 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
6429 | struct sched_group **sg, | |
6430 | struct cpumask *nodemask) | |
9c1cfda2 | 6431 | { |
6711cab4 SS |
6432 | int group; |
6433 | ||
6ca09dfc | 6434 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 6435 | group = cpumask_first(nodemask); |
6711cab4 SS |
6436 | |
6437 | if (sg) | |
6c99e9ad | 6438 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 6439 | return group; |
1da177e4 | 6440 | } |
6711cab4 | 6441 | |
08069033 SS |
6442 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6443 | { | |
6444 | struct sched_group *sg = group_head; | |
6445 | int j; | |
6446 | ||
6447 | if (!sg) | |
6448 | return; | |
3a5c359a | 6449 | do { |
758b2cdc | 6450 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 6451 | struct sched_domain *sd; |
08069033 | 6452 | |
6c99e9ad | 6453 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 6454 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
6455 | /* |
6456 | * Only add "power" once for each | |
6457 | * physical package. | |
6458 | */ | |
6459 | continue; | |
6460 | } | |
08069033 | 6461 | |
18a3885f | 6462 | sg->cpu_power += sd->groups->cpu_power; |
3a5c359a AK |
6463 | } |
6464 | sg = sg->next; | |
6465 | } while (sg != group_head); | |
08069033 | 6466 | } |
0601a88d AH |
6467 | |
6468 | static int build_numa_sched_groups(struct s_data *d, | |
6469 | const struct cpumask *cpu_map, int num) | |
6470 | { | |
6471 | struct sched_domain *sd; | |
6472 | struct sched_group *sg, *prev; | |
6473 | int n, j; | |
6474 | ||
6475 | cpumask_clear(d->covered); | |
6476 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
6477 | if (cpumask_empty(d->nodemask)) { | |
6478 | d->sched_group_nodes[num] = NULL; | |
6479 | goto out; | |
6480 | } | |
6481 | ||
6482 | sched_domain_node_span(num, d->domainspan); | |
6483 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
6484 | ||
6485 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6486 | GFP_KERNEL, num); | |
6487 | if (!sg) { | |
3df0fc5b PZ |
6488 | printk(KERN_WARNING "Can not alloc domain group for node %d\n", |
6489 | num); | |
0601a88d AH |
6490 | return -ENOMEM; |
6491 | } | |
6492 | d->sched_group_nodes[num] = sg; | |
6493 | ||
6494 | for_each_cpu(j, d->nodemask) { | |
6495 | sd = &per_cpu(node_domains, j).sd; | |
6496 | sd->groups = sg; | |
6497 | } | |
6498 | ||
18a3885f | 6499 | sg->cpu_power = 0; |
0601a88d AH |
6500 | cpumask_copy(sched_group_cpus(sg), d->nodemask); |
6501 | sg->next = sg; | |
6502 | cpumask_or(d->covered, d->covered, d->nodemask); | |
6503 | ||
6504 | prev = sg; | |
6505 | for (j = 0; j < nr_node_ids; j++) { | |
6506 | n = (num + j) % nr_node_ids; | |
6507 | cpumask_complement(d->notcovered, d->covered); | |
6508 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
6509 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
6510 | if (cpumask_empty(d->tmpmask)) | |
6511 | break; | |
6512 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
6513 | if (cpumask_empty(d->tmpmask)) | |
6514 | continue; | |
6515 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6516 | GFP_KERNEL, num); | |
6517 | if (!sg) { | |
3df0fc5b PZ |
6518 | printk(KERN_WARNING |
6519 | "Can not alloc domain group for node %d\n", j); | |
0601a88d AH |
6520 | return -ENOMEM; |
6521 | } | |
18a3885f | 6522 | sg->cpu_power = 0; |
0601a88d AH |
6523 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); |
6524 | sg->next = prev->next; | |
6525 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
6526 | prev->next = sg; | |
6527 | prev = sg; | |
6528 | } | |
6529 | out: | |
6530 | return 0; | |
6531 | } | |
6d6bc0ad | 6532 | #endif /* CONFIG_NUMA */ |
1da177e4 | 6533 | |
a616058b | 6534 | #ifdef CONFIG_NUMA |
51888ca2 | 6535 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
6536 | static void free_sched_groups(const struct cpumask *cpu_map, |
6537 | struct cpumask *nodemask) | |
51888ca2 | 6538 | { |
a616058b | 6539 | int cpu, i; |
51888ca2 | 6540 | |
abcd083a | 6541 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
6542 | struct sched_group **sched_group_nodes |
6543 | = sched_group_nodes_bycpu[cpu]; | |
6544 | ||
51888ca2 SV |
6545 | if (!sched_group_nodes) |
6546 | continue; | |
6547 | ||
076ac2af | 6548 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
6549 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
6550 | ||
6ca09dfc | 6551 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 6552 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
6553 | continue; |
6554 | ||
6555 | if (sg == NULL) | |
6556 | continue; | |
6557 | sg = sg->next; | |
6558 | next_sg: | |
6559 | oldsg = sg; | |
6560 | sg = sg->next; | |
6561 | kfree(oldsg); | |
6562 | if (oldsg != sched_group_nodes[i]) | |
6563 | goto next_sg; | |
6564 | } | |
6565 | kfree(sched_group_nodes); | |
6566 | sched_group_nodes_bycpu[cpu] = NULL; | |
6567 | } | |
51888ca2 | 6568 | } |
6d6bc0ad | 6569 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
6570 | static void free_sched_groups(const struct cpumask *cpu_map, |
6571 | struct cpumask *nodemask) | |
a616058b SS |
6572 | { |
6573 | } | |
6d6bc0ad | 6574 | #endif /* CONFIG_NUMA */ |
51888ca2 | 6575 | |
89c4710e SS |
6576 | /* |
6577 | * Initialize sched groups cpu_power. | |
6578 | * | |
6579 | * cpu_power indicates the capacity of sched group, which is used while | |
6580 | * distributing the load between different sched groups in a sched domain. | |
6581 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6582 | * there are asymmetries in the topology. If there are asymmetries, group | |
6583 | * having more cpu_power will pickup more load compared to the group having | |
6584 | * less cpu_power. | |
89c4710e SS |
6585 | */ |
6586 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6587 | { | |
6588 | struct sched_domain *child; | |
6589 | struct sched_group *group; | |
f93e65c1 PZ |
6590 | long power; |
6591 | int weight; | |
89c4710e SS |
6592 | |
6593 | WARN_ON(!sd || !sd->groups); | |
6594 | ||
13318a71 | 6595 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
6596 | return; |
6597 | ||
6598 | child = sd->child; | |
6599 | ||
18a3885f | 6600 | sd->groups->cpu_power = 0; |
5517d86b | 6601 | |
f93e65c1 PZ |
6602 | if (!child) { |
6603 | power = SCHED_LOAD_SCALE; | |
6604 | weight = cpumask_weight(sched_domain_span(sd)); | |
6605 | /* | |
6606 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
6607 | * Usually multiple threads get a better yield out of |
6608 | * that one core than a single thread would have, | |
6609 | * reflect that in sd->smt_gain. | |
f93e65c1 | 6610 | */ |
a52bfd73 PZ |
6611 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
6612 | power *= sd->smt_gain; | |
f93e65c1 | 6613 | power /= weight; |
a52bfd73 PZ |
6614 | power >>= SCHED_LOAD_SHIFT; |
6615 | } | |
18a3885f | 6616 | sd->groups->cpu_power += power; |
89c4710e SS |
6617 | return; |
6618 | } | |
6619 | ||
89c4710e | 6620 | /* |
f93e65c1 | 6621 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
6622 | */ |
6623 | group = child->groups; | |
6624 | do { | |
18a3885f | 6625 | sd->groups->cpu_power += group->cpu_power; |
89c4710e SS |
6626 | group = group->next; |
6627 | } while (group != child->groups); | |
6628 | } | |
6629 | ||
7c16ec58 MT |
6630 | /* |
6631 | * Initializers for schedule domains | |
6632 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6633 | */ | |
6634 | ||
a5d8c348 IM |
6635 | #ifdef CONFIG_SCHED_DEBUG |
6636 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
6637 | #else | |
6638 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
6639 | #endif | |
6640 | ||
7c16ec58 | 6641 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 6642 | |
7c16ec58 MT |
6643 | #define SD_INIT_FUNC(type) \ |
6644 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
6645 | { \ | |
6646 | memset(sd, 0, sizeof(*sd)); \ | |
6647 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 6648 | sd->level = SD_LV_##type; \ |
a5d8c348 | 6649 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
6650 | } |
6651 | ||
6652 | SD_INIT_FUNC(CPU) | |
6653 | #ifdef CONFIG_NUMA | |
6654 | SD_INIT_FUNC(ALLNODES) | |
6655 | SD_INIT_FUNC(NODE) | |
6656 | #endif | |
6657 | #ifdef CONFIG_SCHED_SMT | |
6658 | SD_INIT_FUNC(SIBLING) | |
6659 | #endif | |
6660 | #ifdef CONFIG_SCHED_MC | |
6661 | SD_INIT_FUNC(MC) | |
6662 | #endif | |
6663 | ||
1d3504fc HS |
6664 | static int default_relax_domain_level = -1; |
6665 | ||
6666 | static int __init setup_relax_domain_level(char *str) | |
6667 | { | |
30e0e178 LZ |
6668 | unsigned long val; |
6669 | ||
6670 | val = simple_strtoul(str, NULL, 0); | |
6671 | if (val < SD_LV_MAX) | |
6672 | default_relax_domain_level = val; | |
6673 | ||
1d3504fc HS |
6674 | return 1; |
6675 | } | |
6676 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6677 | ||
6678 | static void set_domain_attribute(struct sched_domain *sd, | |
6679 | struct sched_domain_attr *attr) | |
6680 | { | |
6681 | int request; | |
6682 | ||
6683 | if (!attr || attr->relax_domain_level < 0) { | |
6684 | if (default_relax_domain_level < 0) | |
6685 | return; | |
6686 | else | |
6687 | request = default_relax_domain_level; | |
6688 | } else | |
6689 | request = attr->relax_domain_level; | |
6690 | if (request < sd->level) { | |
6691 | /* turn off idle balance on this domain */ | |
c88d5910 | 6692 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6693 | } else { |
6694 | /* turn on idle balance on this domain */ | |
c88d5910 | 6695 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6696 | } |
6697 | } | |
6698 | ||
2109b99e AH |
6699 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6700 | const struct cpumask *cpu_map) | |
6701 | { | |
6702 | switch (what) { | |
6703 | case sa_sched_groups: | |
6704 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
6705 | d->sched_group_nodes = NULL; | |
6706 | case sa_rootdomain: | |
6707 | free_rootdomain(d->rd); /* fall through */ | |
6708 | case sa_tmpmask: | |
6709 | free_cpumask_var(d->tmpmask); /* fall through */ | |
6710 | case sa_send_covered: | |
6711 | free_cpumask_var(d->send_covered); /* fall through */ | |
6712 | case sa_this_core_map: | |
6713 | free_cpumask_var(d->this_core_map); /* fall through */ | |
6714 | case sa_this_sibling_map: | |
6715 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
6716 | case sa_nodemask: | |
6717 | free_cpumask_var(d->nodemask); /* fall through */ | |
6718 | case sa_sched_group_nodes: | |
d1b55138 | 6719 | #ifdef CONFIG_NUMA |
2109b99e AH |
6720 | kfree(d->sched_group_nodes); /* fall through */ |
6721 | case sa_notcovered: | |
6722 | free_cpumask_var(d->notcovered); /* fall through */ | |
6723 | case sa_covered: | |
6724 | free_cpumask_var(d->covered); /* fall through */ | |
6725 | case sa_domainspan: | |
6726 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 6727 | #endif |
2109b99e AH |
6728 | case sa_none: |
6729 | break; | |
6730 | } | |
6731 | } | |
3404c8d9 | 6732 | |
2109b99e AH |
6733 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6734 | const struct cpumask *cpu_map) | |
6735 | { | |
3404c8d9 | 6736 | #ifdef CONFIG_NUMA |
2109b99e AH |
6737 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
6738 | return sa_none; | |
6739 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
6740 | return sa_domainspan; | |
6741 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
6742 | return sa_covered; | |
6743 | /* Allocate the per-node list of sched groups */ | |
6744 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
6745 | sizeof(struct sched_group *), GFP_KERNEL); | |
6746 | if (!d->sched_group_nodes) { | |
3df0fc5b | 6747 | printk(KERN_WARNING "Can not alloc sched group node list\n"); |
2109b99e | 6748 | return sa_notcovered; |
d1b55138 | 6749 | } |
2109b99e | 6750 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 6751 | #endif |
2109b99e AH |
6752 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
6753 | return sa_sched_group_nodes; | |
6754 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
6755 | return sa_nodemask; | |
6756 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
6757 | return sa_this_sibling_map; | |
6758 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) | |
6759 | return sa_this_core_map; | |
6760 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) | |
6761 | return sa_send_covered; | |
6762 | d->rd = alloc_rootdomain(); | |
6763 | if (!d->rd) { | |
3df0fc5b | 6764 | printk(KERN_WARNING "Cannot alloc root domain\n"); |
2109b99e | 6765 | return sa_tmpmask; |
57d885fe | 6766 | } |
2109b99e AH |
6767 | return sa_rootdomain; |
6768 | } | |
57d885fe | 6769 | |
7f4588f3 AH |
6770 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
6771 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
6772 | { | |
6773 | struct sched_domain *sd = NULL; | |
7c16ec58 | 6774 | #ifdef CONFIG_NUMA |
7f4588f3 | 6775 | struct sched_domain *parent; |
1da177e4 | 6776 | |
7f4588f3 AH |
6777 | d->sd_allnodes = 0; |
6778 | if (cpumask_weight(cpu_map) > | |
6779 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
6780 | sd = &per_cpu(allnodes_domains, i).sd; | |
6781 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 6782 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
6783 | cpumask_copy(sched_domain_span(sd), cpu_map); |
6784 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
6785 | d->sd_allnodes = 1; | |
6786 | } | |
6787 | parent = sd; | |
6788 | ||
6789 | sd = &per_cpu(node_domains, i).sd; | |
6790 | SD_INIT(sd, NODE); | |
6791 | set_domain_attribute(sd, attr); | |
6792 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
6793 | sd->parent = parent; | |
6794 | if (parent) | |
6795 | parent->child = sd; | |
6796 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 6797 | #endif |
7f4588f3 AH |
6798 | return sd; |
6799 | } | |
1da177e4 | 6800 | |
87cce662 AH |
6801 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
6802 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6803 | struct sched_domain *parent, int i) | |
6804 | { | |
6805 | struct sched_domain *sd; | |
6806 | sd = &per_cpu(phys_domains, i).sd; | |
6807 | SD_INIT(sd, CPU); | |
6808 | set_domain_attribute(sd, attr); | |
6809 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
6810 | sd->parent = parent; | |
6811 | if (parent) | |
6812 | parent->child = sd; | |
6813 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
6814 | return sd; | |
6815 | } | |
1da177e4 | 6816 | |
410c4081 AH |
6817 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
6818 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6819 | struct sched_domain *parent, int i) | |
6820 | { | |
6821 | struct sched_domain *sd = parent; | |
1e9f28fa | 6822 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
6823 | sd = &per_cpu(core_domains, i).sd; |
6824 | SD_INIT(sd, MC); | |
6825 | set_domain_attribute(sd, attr); | |
6826 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
6827 | sd->parent = parent; | |
6828 | parent->child = sd; | |
6829 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 6830 | #endif |
410c4081 AH |
6831 | return sd; |
6832 | } | |
1e9f28fa | 6833 | |
d8173535 AH |
6834 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
6835 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6836 | struct sched_domain *parent, int i) | |
6837 | { | |
6838 | struct sched_domain *sd = parent; | |
1da177e4 | 6839 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
6840 | sd = &per_cpu(cpu_domains, i).sd; |
6841 | SD_INIT(sd, SIBLING); | |
6842 | set_domain_attribute(sd, attr); | |
6843 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
6844 | sd->parent = parent; | |
6845 | parent->child = sd; | |
6846 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 6847 | #endif |
d8173535 AH |
6848 | return sd; |
6849 | } | |
1da177e4 | 6850 | |
0e8e85c9 AH |
6851 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
6852 | const struct cpumask *cpu_map, int cpu) | |
6853 | { | |
6854 | switch (l) { | |
1da177e4 | 6855 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
6856 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
6857 | cpumask_and(d->this_sibling_map, cpu_map, | |
6858 | topology_thread_cpumask(cpu)); | |
6859 | if (cpu == cpumask_first(d->this_sibling_map)) | |
6860 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
6861 | &cpu_to_cpu_group, | |
6862 | d->send_covered, d->tmpmask); | |
6863 | break; | |
1da177e4 | 6864 | #endif |
1e9f28fa | 6865 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
6866 | case SD_LV_MC: /* set up multi-core groups */ |
6867 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
6868 | if (cpu == cpumask_first(d->this_core_map)) | |
6869 | init_sched_build_groups(d->this_core_map, cpu_map, | |
6870 | &cpu_to_core_group, | |
6871 | d->send_covered, d->tmpmask); | |
6872 | break; | |
1e9f28fa | 6873 | #endif |
86548096 AH |
6874 | case SD_LV_CPU: /* set up physical groups */ |
6875 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
6876 | if (!cpumask_empty(d->nodemask)) | |
6877 | init_sched_build_groups(d->nodemask, cpu_map, | |
6878 | &cpu_to_phys_group, | |
6879 | d->send_covered, d->tmpmask); | |
6880 | break; | |
1da177e4 | 6881 | #ifdef CONFIG_NUMA |
de616e36 AH |
6882 | case SD_LV_ALLNODES: |
6883 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
6884 | d->send_covered, d->tmpmask); | |
6885 | break; | |
6886 | #endif | |
0e8e85c9 AH |
6887 | default: |
6888 | break; | |
7c16ec58 | 6889 | } |
0e8e85c9 | 6890 | } |
9c1cfda2 | 6891 | |
2109b99e AH |
6892 | /* |
6893 | * Build sched domains for a given set of cpus and attach the sched domains | |
6894 | * to the individual cpus | |
6895 | */ | |
6896 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
6897 | struct sched_domain_attr *attr) | |
6898 | { | |
6899 | enum s_alloc alloc_state = sa_none; | |
6900 | struct s_data d; | |
294b0c96 | 6901 | struct sched_domain *sd; |
2109b99e | 6902 | int i; |
7c16ec58 | 6903 | #ifdef CONFIG_NUMA |
2109b99e | 6904 | d.sd_allnodes = 0; |
7c16ec58 | 6905 | #endif |
9c1cfda2 | 6906 | |
2109b99e AH |
6907 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6908 | if (alloc_state != sa_rootdomain) | |
6909 | goto error; | |
6910 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 6911 | |
1da177e4 | 6912 | /* |
1a20ff27 | 6913 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 6914 | */ |
abcd083a | 6915 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
6916 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
6917 | cpu_map); | |
9761eea8 | 6918 | |
7f4588f3 | 6919 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 6920 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 6921 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 6922 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 6923 | } |
9c1cfda2 | 6924 | |
abcd083a | 6925 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 6926 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
a2af04cd | 6927 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 6928 | } |
9c1cfda2 | 6929 | |
1da177e4 | 6930 | /* Set up physical groups */ |
86548096 AH |
6931 | for (i = 0; i < nr_node_ids; i++) |
6932 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 6933 | |
1da177e4 LT |
6934 | #ifdef CONFIG_NUMA |
6935 | /* Set up node groups */ | |
de616e36 AH |
6936 | if (d.sd_allnodes) |
6937 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 6938 | |
0601a88d AH |
6939 | for (i = 0; i < nr_node_ids; i++) |
6940 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 6941 | goto error; |
1da177e4 LT |
6942 | #endif |
6943 | ||
6944 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 6945 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 6946 | for_each_cpu(i, cpu_map) { |
294b0c96 | 6947 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 6948 | init_sched_groups_power(i, sd); |
5c45bf27 | 6949 | } |
1da177e4 | 6950 | #endif |
1e9f28fa | 6951 | #ifdef CONFIG_SCHED_MC |
abcd083a | 6952 | for_each_cpu(i, cpu_map) { |
294b0c96 | 6953 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 6954 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
6955 | } |
6956 | #endif | |
1e9f28fa | 6957 | |
abcd083a | 6958 | for_each_cpu(i, cpu_map) { |
294b0c96 | 6959 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 6960 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6961 | } |
6962 | ||
9c1cfda2 | 6963 | #ifdef CONFIG_NUMA |
076ac2af | 6964 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 6965 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 6966 | |
49a02c51 | 6967 | if (d.sd_allnodes) { |
6711cab4 | 6968 | struct sched_group *sg; |
f712c0c7 | 6969 | |
96f874e2 | 6970 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 6971 | d.tmpmask); |
f712c0c7 SS |
6972 | init_numa_sched_groups_power(sg); |
6973 | } | |
9c1cfda2 JH |
6974 | #endif |
6975 | ||
1da177e4 | 6976 | /* Attach the domains */ |
abcd083a | 6977 | for_each_cpu(i, cpu_map) { |
1da177e4 | 6978 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 6979 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 6980 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 6981 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 6982 | #else |
6c99e9ad | 6983 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 6984 | #endif |
49a02c51 | 6985 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6986 | } |
51888ca2 | 6987 | |
2109b99e AH |
6988 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
6989 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
6990 | return 0; | |
51888ca2 | 6991 | |
51888ca2 | 6992 | error: |
2109b99e AH |
6993 | __free_domain_allocs(&d, alloc_state, cpu_map); |
6994 | return -ENOMEM; | |
1da177e4 | 6995 | } |
029190c5 | 6996 | |
96f874e2 | 6997 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
6998 | { |
6999 | return __build_sched_domains(cpu_map, NULL); | |
7000 | } | |
7001 | ||
acc3f5d7 | 7002 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7003 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7004 | static struct sched_domain_attr *dattr_cur; |
7005 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7006 | |
7007 | /* | |
7008 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7009 | * cpumask) fails, then fallback to a single sched domain, |
7010 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7011 | */ |
4212823f | 7012 | static cpumask_var_t fallback_doms; |
029190c5 | 7013 | |
ee79d1bd HC |
7014 | /* |
7015 | * arch_update_cpu_topology lets virtualized architectures update the | |
7016 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7017 | * or 0 if it stayed the same. | |
7018 | */ | |
7019 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7020 | { |
ee79d1bd | 7021 | return 0; |
22e52b07 HC |
7022 | } |
7023 | ||
acc3f5d7 RR |
7024 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7025 | { | |
7026 | int i; | |
7027 | cpumask_var_t *doms; | |
7028 | ||
7029 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7030 | if (!doms) | |
7031 | return NULL; | |
7032 | for (i = 0; i < ndoms; i++) { | |
7033 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7034 | free_sched_domains(doms, i); | |
7035 | return NULL; | |
7036 | } | |
7037 | } | |
7038 | return doms; | |
7039 | } | |
7040 | ||
7041 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7042 | { | |
7043 | unsigned int i; | |
7044 | for (i = 0; i < ndoms; i++) | |
7045 | free_cpumask_var(doms[i]); | |
7046 | kfree(doms); | |
7047 | } | |
7048 | ||
1a20ff27 | 7049 | /* |
41a2d6cf | 7050 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7051 | * For now this just excludes isolated cpus, but could be used to |
7052 | * exclude other special cases in the future. | |
1a20ff27 | 7053 | */ |
96f874e2 | 7054 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7055 | { |
7378547f MM |
7056 | int err; |
7057 | ||
22e52b07 | 7058 | arch_update_cpu_topology(); |
029190c5 | 7059 | ndoms_cur = 1; |
acc3f5d7 | 7060 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7061 | if (!doms_cur) |
acc3f5d7 RR |
7062 | doms_cur = &fallback_doms; |
7063 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 7064 | dattr_cur = NULL; |
acc3f5d7 | 7065 | err = build_sched_domains(doms_cur[0]); |
6382bc90 | 7066 | register_sched_domain_sysctl(); |
7378547f MM |
7067 | |
7068 | return err; | |
1a20ff27 DG |
7069 | } |
7070 | ||
96f874e2 RR |
7071 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
7072 | struct cpumask *tmpmask) | |
1da177e4 | 7073 | { |
7c16ec58 | 7074 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7075 | } |
1da177e4 | 7076 | |
1a20ff27 DG |
7077 | /* |
7078 | * Detach sched domains from a group of cpus specified in cpu_map | |
7079 | * These cpus will now be attached to the NULL domain | |
7080 | */ | |
96f874e2 | 7081 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7082 | { |
96f874e2 RR |
7083 | /* Save because hotplug lock held. */ |
7084 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
7085 | int i; |
7086 | ||
abcd083a | 7087 | for_each_cpu(i, cpu_map) |
57d885fe | 7088 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7089 | synchronize_sched(); |
96f874e2 | 7090 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
7091 | } |
7092 | ||
1d3504fc HS |
7093 | /* handle null as "default" */ |
7094 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7095 | struct sched_domain_attr *new, int idx_new) | |
7096 | { | |
7097 | struct sched_domain_attr tmp; | |
7098 | ||
7099 | /* fast path */ | |
7100 | if (!new && !cur) | |
7101 | return 1; | |
7102 | ||
7103 | tmp = SD_ATTR_INIT; | |
7104 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7105 | new ? (new + idx_new) : &tmp, | |
7106 | sizeof(struct sched_domain_attr)); | |
7107 | } | |
7108 | ||
029190c5 PJ |
7109 | /* |
7110 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7111 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7112 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7113 | * It destroys each deleted domain and builds each new domain. | |
7114 | * | |
acc3f5d7 | 7115 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7116 | * The masks don't intersect (don't overlap.) We should setup one |
7117 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7118 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7119 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7120 | * it as it is. | |
7121 | * | |
acc3f5d7 RR |
7122 | * The passed in 'doms_new' should be allocated using |
7123 | * alloc_sched_domains. This routine takes ownership of it and will | |
7124 | * free_sched_domains it when done with it. If the caller failed the | |
7125 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7126 | * and partition_sched_domains() will fallback to the single partition | |
7127 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7128 | * |
96f874e2 | 7129 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7130 | * ndoms_new == 0 is a special case for destroying existing domains, |
7131 | * and it will not create the default domain. | |
dfb512ec | 7132 | * |
029190c5 PJ |
7133 | * Call with hotplug lock held |
7134 | */ | |
acc3f5d7 | 7135 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7136 | struct sched_domain_attr *dattr_new) |
029190c5 | 7137 | { |
dfb512ec | 7138 | int i, j, n; |
d65bd5ec | 7139 | int new_topology; |
029190c5 | 7140 | |
712555ee | 7141 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7142 | |
7378547f MM |
7143 | /* always unregister in case we don't destroy any domains */ |
7144 | unregister_sched_domain_sysctl(); | |
7145 | ||
d65bd5ec HC |
7146 | /* Let architecture update cpu core mappings. */ |
7147 | new_topology = arch_update_cpu_topology(); | |
7148 | ||
dfb512ec | 7149 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7150 | |
7151 | /* Destroy deleted domains */ | |
7152 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7153 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7154 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7155 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7156 | goto match1; |
7157 | } | |
7158 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7159 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7160 | match1: |
7161 | ; | |
7162 | } | |
7163 | ||
e761b772 MK |
7164 | if (doms_new == NULL) { |
7165 | ndoms_cur = 0; | |
acc3f5d7 | 7166 | doms_new = &fallback_doms; |
6ad4c188 | 7167 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7168 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7169 | } |
7170 | ||
029190c5 PJ |
7171 | /* Build new domains */ |
7172 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7173 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7174 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7175 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7176 | goto match2; |
7177 | } | |
7178 | /* no match - add a new doms_new */ | |
acc3f5d7 | 7179 | __build_sched_domains(doms_new[i], |
1d3504fc | 7180 | dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7181 | match2: |
7182 | ; | |
7183 | } | |
7184 | ||
7185 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7186 | if (doms_cur != &fallback_doms) |
7187 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7188 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7189 | doms_cur = doms_new; |
1d3504fc | 7190 | dattr_cur = dattr_new; |
029190c5 | 7191 | ndoms_cur = ndoms_new; |
7378547f MM |
7192 | |
7193 | register_sched_domain_sysctl(); | |
a1835615 | 7194 | |
712555ee | 7195 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7196 | } |
7197 | ||
5c45bf27 | 7198 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 7199 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 7200 | { |
95402b38 | 7201 | get_online_cpus(); |
dfb512ec MK |
7202 | |
7203 | /* Destroy domains first to force the rebuild */ | |
7204 | partition_sched_domains(0, NULL, NULL); | |
7205 | ||
e761b772 | 7206 | rebuild_sched_domains(); |
95402b38 | 7207 | put_online_cpus(); |
5c45bf27 SS |
7208 | } |
7209 | ||
7210 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7211 | { | |
afb8a9b7 | 7212 | unsigned int level = 0; |
5c45bf27 | 7213 | |
afb8a9b7 GS |
7214 | if (sscanf(buf, "%u", &level) != 1) |
7215 | return -EINVAL; | |
7216 | ||
7217 | /* | |
7218 | * level is always be positive so don't check for | |
7219 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7220 | * What happens on 0 or 1 byte write, | |
7221 | * need to check for count as well? | |
7222 | */ | |
7223 | ||
7224 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7225 | return -EINVAL; |
7226 | ||
7227 | if (smt) | |
afb8a9b7 | 7228 | sched_smt_power_savings = level; |
5c45bf27 | 7229 | else |
afb8a9b7 | 7230 | sched_mc_power_savings = level; |
5c45bf27 | 7231 | |
c70f22d2 | 7232 | arch_reinit_sched_domains(); |
5c45bf27 | 7233 | |
c70f22d2 | 7234 | return count; |
5c45bf27 SS |
7235 | } |
7236 | ||
5c45bf27 | 7237 | #ifdef CONFIG_SCHED_MC |
f718cd4a | 7238 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
c9be0a36 | 7239 | struct sysdev_class_attribute *attr, |
f718cd4a | 7240 | char *page) |
5c45bf27 SS |
7241 | { |
7242 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7243 | } | |
f718cd4a | 7244 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
c9be0a36 | 7245 | struct sysdev_class_attribute *attr, |
48f24c4d | 7246 | const char *buf, size_t count) |
5c45bf27 SS |
7247 | { |
7248 | return sched_power_savings_store(buf, count, 0); | |
7249 | } | |
f718cd4a AK |
7250 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7251 | sched_mc_power_savings_show, | |
7252 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7253 | #endif |
7254 | ||
7255 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a | 7256 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
c9be0a36 | 7257 | struct sysdev_class_attribute *attr, |
f718cd4a | 7258 | char *page) |
5c45bf27 SS |
7259 | { |
7260 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7261 | } | |
f718cd4a | 7262 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
c9be0a36 | 7263 | struct sysdev_class_attribute *attr, |
48f24c4d | 7264 | const char *buf, size_t count) |
5c45bf27 SS |
7265 | { |
7266 | return sched_power_savings_store(buf, count, 1); | |
7267 | } | |
f718cd4a AK |
7268 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7269 | sched_smt_power_savings_show, | |
6707de00 AB |
7270 | sched_smt_power_savings_store); |
7271 | #endif | |
7272 | ||
39aac648 | 7273 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
7274 | { |
7275 | int err = 0; | |
7276 | ||
7277 | #ifdef CONFIG_SCHED_SMT | |
7278 | if (smt_capable()) | |
7279 | err = sysfs_create_file(&cls->kset.kobj, | |
7280 | &attr_sched_smt_power_savings.attr); | |
7281 | #endif | |
7282 | #ifdef CONFIG_SCHED_MC | |
7283 | if (!err && mc_capable()) | |
7284 | err = sysfs_create_file(&cls->kset.kobj, | |
7285 | &attr_sched_mc_power_savings.attr); | |
7286 | #endif | |
7287 | return err; | |
7288 | } | |
6d6bc0ad | 7289 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7290 | |
e761b772 | 7291 | #ifndef CONFIG_CPUSETS |
1da177e4 | 7292 | /* |
e761b772 MK |
7293 | * Add online and remove offline CPUs from the scheduler domains. |
7294 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
7295 | */ |
7296 | static int update_sched_domains(struct notifier_block *nfb, | |
7297 | unsigned long action, void *hcpu) | |
e761b772 MK |
7298 | { |
7299 | switch (action) { | |
7300 | case CPU_ONLINE: | |
7301 | case CPU_ONLINE_FROZEN: | |
6ad4c188 PZ |
7302 | case CPU_DOWN_PREPARE: |
7303 | case CPU_DOWN_PREPARE_FROZEN: | |
7304 | case CPU_DOWN_FAILED: | |
7305 | case CPU_DOWN_FAILED_FROZEN: | |
dfb512ec | 7306 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
7307 | return NOTIFY_OK; |
7308 | ||
7309 | default: | |
7310 | return NOTIFY_DONE; | |
7311 | } | |
7312 | } | |
7313 | #endif | |
7314 | ||
7315 | static int update_runtime(struct notifier_block *nfb, | |
7316 | unsigned long action, void *hcpu) | |
1da177e4 | 7317 | { |
7def2be1 PZ |
7318 | int cpu = (int)(long)hcpu; |
7319 | ||
1da177e4 | 7320 | switch (action) { |
1da177e4 | 7321 | case CPU_DOWN_PREPARE: |
8bb78442 | 7322 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7323 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7324 | return NOTIFY_OK; |
7325 | ||
1da177e4 | 7326 | case CPU_DOWN_FAILED: |
8bb78442 | 7327 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7328 | case CPU_ONLINE: |
8bb78442 | 7329 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7330 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7331 | return NOTIFY_OK; |
7332 | ||
1da177e4 LT |
7333 | default: |
7334 | return NOTIFY_DONE; | |
7335 | } | |
1da177e4 | 7336 | } |
1da177e4 LT |
7337 | |
7338 | void __init sched_init_smp(void) | |
7339 | { | |
dcc30a35 RR |
7340 | cpumask_var_t non_isolated_cpus; |
7341 | ||
7342 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7343 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7344 | |
434d53b0 MT |
7345 | #if defined(CONFIG_NUMA) |
7346 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7347 | GFP_KERNEL); | |
7348 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7349 | #endif | |
95402b38 | 7350 | get_online_cpus(); |
712555ee | 7351 | mutex_lock(&sched_domains_mutex); |
6ad4c188 | 7352 | arch_init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7353 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7354 | if (cpumask_empty(non_isolated_cpus)) | |
7355 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7356 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7357 | put_online_cpus(); |
e761b772 MK |
7358 | |
7359 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
7360 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7361 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
7362 | #endif |
7363 | ||
7364 | /* RT runtime code needs to handle some hotplug events */ | |
7365 | hotcpu_notifier(update_runtime, 0); | |
7366 | ||
b328ca18 | 7367 | init_hrtick(); |
5c1e1767 NP |
7368 | |
7369 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7370 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7371 | BUG(); |
19978ca6 | 7372 | sched_init_granularity(); |
dcc30a35 | 7373 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7374 | |
0e3900e6 | 7375 | init_sched_rt_class(); |
1da177e4 LT |
7376 | } |
7377 | #else | |
7378 | void __init sched_init_smp(void) | |
7379 | { | |
19978ca6 | 7380 | sched_init_granularity(); |
1da177e4 LT |
7381 | } |
7382 | #endif /* CONFIG_SMP */ | |
7383 | ||
cd1bb94b AB |
7384 | const_debug unsigned int sysctl_timer_migration = 1; |
7385 | ||
1da177e4 LT |
7386 | int in_sched_functions(unsigned long addr) |
7387 | { | |
1da177e4 LT |
7388 | return in_lock_functions(addr) || |
7389 | (addr >= (unsigned long)__sched_text_start | |
7390 | && addr < (unsigned long)__sched_text_end); | |
7391 | } | |
7392 | ||
a9957449 | 7393 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7394 | { |
7395 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7396 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7397 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7398 | cfs_rq->rq = rq; | |
7399 | #endif | |
67e9fb2a | 7400 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7401 | } |
7402 | ||
fa85ae24 PZ |
7403 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7404 | { | |
7405 | struct rt_prio_array *array; | |
7406 | int i; | |
7407 | ||
7408 | array = &rt_rq->active; | |
7409 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7410 | INIT_LIST_HEAD(array->queue + i); | |
7411 | __clear_bit(i, array->bitmap); | |
7412 | } | |
7413 | /* delimiter for bitsearch: */ | |
7414 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7415 | ||
052f1dc7 | 7416 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 7417 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 7418 | #ifdef CONFIG_SMP |
e864c499 | 7419 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 7420 | #endif |
48d5e258 | 7421 | #endif |
fa85ae24 PZ |
7422 | #ifdef CONFIG_SMP |
7423 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 7424 | rt_rq->overloaded = 0; |
05fa785c | 7425 | plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
7426 | #endif |
7427 | ||
7428 | rt_rq->rt_time = 0; | |
7429 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 7430 | rt_rq->rt_runtime = 0; |
0986b11b | 7431 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
6f505b16 | 7432 | |
052f1dc7 | 7433 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7434 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7435 | rt_rq->rq = rq; |
7436 | #endif | |
fa85ae24 PZ |
7437 | } |
7438 | ||
6f505b16 | 7439 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
7440 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
7441 | struct sched_entity *se, int cpu, int add, | |
7442 | struct sched_entity *parent) | |
6f505b16 | 7443 | { |
ec7dc8ac | 7444 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7445 | tg->cfs_rq[cpu] = cfs_rq; |
7446 | init_cfs_rq(cfs_rq, rq); | |
7447 | cfs_rq->tg = tg; | |
7448 | if (add) | |
7449 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7450 | ||
7451 | tg->se[cpu] = se; | |
354d60c2 DG |
7452 | /* se could be NULL for init_task_group */ |
7453 | if (!se) | |
7454 | return; | |
7455 | ||
ec7dc8ac DG |
7456 | if (!parent) |
7457 | se->cfs_rq = &rq->cfs; | |
7458 | else | |
7459 | se->cfs_rq = parent->my_q; | |
7460 | ||
6f505b16 PZ |
7461 | se->my_q = cfs_rq; |
7462 | se->load.weight = tg->shares; | |
e05510d0 | 7463 | se->load.inv_weight = 0; |
ec7dc8ac | 7464 | se->parent = parent; |
6f505b16 | 7465 | } |
052f1dc7 | 7466 | #endif |
6f505b16 | 7467 | |
052f1dc7 | 7468 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
7469 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
7470 | struct sched_rt_entity *rt_se, int cpu, int add, | |
7471 | struct sched_rt_entity *parent) | |
6f505b16 | 7472 | { |
ec7dc8ac DG |
7473 | struct rq *rq = cpu_rq(cpu); |
7474 | ||
6f505b16 PZ |
7475 | tg->rt_rq[cpu] = rt_rq; |
7476 | init_rt_rq(rt_rq, rq); | |
7477 | rt_rq->tg = tg; | |
ac086bc2 | 7478 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7479 | if (add) |
7480 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
7481 | ||
7482 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7483 | if (!rt_se) |
7484 | return; | |
7485 | ||
ec7dc8ac DG |
7486 | if (!parent) |
7487 | rt_se->rt_rq = &rq->rt; | |
7488 | else | |
7489 | rt_se->rt_rq = parent->my_q; | |
7490 | ||
6f505b16 | 7491 | rt_se->my_q = rt_rq; |
ec7dc8ac | 7492 | rt_se->parent = parent; |
6f505b16 PZ |
7493 | INIT_LIST_HEAD(&rt_se->run_list); |
7494 | } | |
7495 | #endif | |
7496 | ||
1da177e4 LT |
7497 | void __init sched_init(void) |
7498 | { | |
dd41f596 | 7499 | int i, j; |
434d53b0 MT |
7500 | unsigned long alloc_size = 0, ptr; |
7501 | ||
7502 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7503 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7504 | #endif | |
7505 | #ifdef CONFIG_RT_GROUP_SCHED | |
7506 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7507 | #endif |
df7c8e84 | 7508 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 7509 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 7510 | #endif |
434d53b0 | 7511 | if (alloc_size) { |
36b7b6d4 | 7512 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7513 | |
7514 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7515 | init_task_group.se = (struct sched_entity **)ptr; | |
7516 | ptr += nr_cpu_ids * sizeof(void **); | |
7517 | ||
7518 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7519 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7520 | |
6d6bc0ad | 7521 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 MT |
7522 | #ifdef CONFIG_RT_GROUP_SCHED |
7523 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7524 | ptr += nr_cpu_ids * sizeof(void **); | |
7525 | ||
7526 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
7527 | ptr += nr_cpu_ids * sizeof(void **); |
7528 | ||
6d6bc0ad | 7529 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
7530 | #ifdef CONFIG_CPUMASK_OFFSTACK |
7531 | for_each_possible_cpu(i) { | |
7532 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
7533 | ptr += cpumask_size(); | |
7534 | } | |
7535 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 7536 | } |
dd41f596 | 7537 | |
57d885fe GH |
7538 | #ifdef CONFIG_SMP |
7539 | init_defrootdomain(); | |
7540 | #endif | |
7541 | ||
d0b27fa7 PZ |
7542 | init_rt_bandwidth(&def_rt_bandwidth, |
7543 | global_rt_period(), global_rt_runtime()); | |
7544 | ||
7545 | #ifdef CONFIG_RT_GROUP_SCHED | |
7546 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
7547 | global_rt_period(), global_rt_runtime()); | |
6d6bc0ad | 7548 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7549 | |
7c941438 | 7550 | #ifdef CONFIG_CGROUP_SCHED |
6f505b16 | 7551 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
7552 | INIT_LIST_HEAD(&init_task_group.children); |
7553 | ||
7c941438 | 7554 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7555 | |
4a6cc4bd JK |
7556 | #if defined CONFIG_FAIR_GROUP_SCHED && defined CONFIG_SMP |
7557 | update_shares_data = __alloc_percpu(nr_cpu_ids * sizeof(unsigned long), | |
7558 | __alignof__(unsigned long)); | |
7559 | #endif | |
0a945022 | 7560 | for_each_possible_cpu(i) { |
70b97a7f | 7561 | struct rq *rq; |
1da177e4 LT |
7562 | |
7563 | rq = cpu_rq(i); | |
05fa785c | 7564 | raw_spin_lock_init(&rq->lock); |
7897986b | 7565 | rq->nr_running = 0; |
dce48a84 TG |
7566 | rq->calc_load_active = 0; |
7567 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 7568 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7569 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7570 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 7571 | init_task_group.shares = init_task_group_load; |
6f505b16 | 7572 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
7573 | #ifdef CONFIG_CGROUP_SCHED |
7574 | /* | |
7575 | * How much cpu bandwidth does init_task_group get? | |
7576 | * | |
7577 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7578 | * gets 100% of the cpu resources in the system. This overall | |
7579 | * system cpu resource is divided among the tasks of | |
7580 | * init_task_group and its child task-groups in a fair manner, | |
7581 | * based on each entity's (task or task-group's) weight | |
7582 | * (se->load.weight). | |
7583 | * | |
7584 | * In other words, if init_task_group has 10 tasks of weight | |
7585 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
7586 | * then A0's share of the cpu resource is: | |
7587 | * | |
0d905bca | 7588 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
7589 | * |
7590 | * We achieve this by letting init_task_group's tasks sit | |
7591 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
7592 | */ | |
ec7dc8ac | 7593 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
052f1dc7 | 7594 | #endif |
354d60c2 DG |
7595 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7596 | ||
7597 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7598 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7599 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 7600 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 7601 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 7602 | #endif |
dd41f596 | 7603 | #endif |
1da177e4 | 7604 | |
dd41f596 IM |
7605 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7606 | rq->cpu_load[j] = 0; | |
1da177e4 | 7607 | #ifdef CONFIG_SMP |
41c7ce9a | 7608 | rq->sd = NULL; |
57d885fe | 7609 | rq->rd = NULL; |
e51fd5e2 | 7610 | rq->cpu_power = SCHED_LOAD_SCALE; |
3f029d3c | 7611 | rq->post_schedule = 0; |
1da177e4 | 7612 | rq->active_balance = 0; |
dd41f596 | 7613 | rq->next_balance = jiffies; |
1da177e4 | 7614 | rq->push_cpu = 0; |
0a2966b4 | 7615 | rq->cpu = i; |
1f11eb6a | 7616 | rq->online = 0; |
eae0c9df MG |
7617 | rq->idle_stamp = 0; |
7618 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
dc938520 | 7619 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 7620 | #endif |
8f4d37ec | 7621 | init_rq_hrtick(rq); |
1da177e4 | 7622 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7623 | } |
7624 | ||
2dd73a4f | 7625 | set_load_weight(&init_task); |
b50f60ce | 7626 | |
e107be36 AK |
7627 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7628 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7629 | #endif | |
7630 | ||
c9819f45 | 7631 | #ifdef CONFIG_SMP |
962cf36c | 7632 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
7633 | #endif |
7634 | ||
b50f60ce | 7635 | #ifdef CONFIG_RT_MUTEXES |
1d615482 | 7636 | plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock); |
b50f60ce HC |
7637 | #endif |
7638 | ||
1da177e4 LT |
7639 | /* |
7640 | * The boot idle thread does lazy MMU switching as well: | |
7641 | */ | |
7642 | atomic_inc(&init_mm.mm_count); | |
7643 | enter_lazy_tlb(&init_mm, current); | |
7644 | ||
7645 | /* | |
7646 | * Make us the idle thread. Technically, schedule() should not be | |
7647 | * called from this thread, however somewhere below it might be, | |
7648 | * but because we are the idle thread, we just pick up running again | |
7649 | * when this runqueue becomes "idle". | |
7650 | */ | |
7651 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7652 | |
7653 | calc_load_update = jiffies + LOAD_FREQ; | |
7654 | ||
dd41f596 IM |
7655 | /* |
7656 | * During early bootup we pretend to be a normal task: | |
7657 | */ | |
7658 | current->sched_class = &fair_sched_class; | |
6892b75e | 7659 | |
6a7b3dc3 | 7660 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 7661 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 7662 | #ifdef CONFIG_SMP |
7d1e6a9b | 7663 | #ifdef CONFIG_NO_HZ |
49557e62 | 7664 | zalloc_cpumask_var(&nohz.cpu_mask, GFP_NOWAIT); |
4bdddf8f | 7665 | alloc_cpumask_var(&nohz.ilb_grp_nohz_mask, GFP_NOWAIT); |
7d1e6a9b | 7666 | #endif |
bdddd296 RR |
7667 | /* May be allocated at isolcpus cmdline parse time */ |
7668 | if (cpu_isolated_map == NULL) | |
7669 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 7670 | #endif /* SMP */ |
6a7b3dc3 | 7671 | |
cdd6c482 | 7672 | perf_event_init(); |
0d905bca | 7673 | |
6892b75e | 7674 | scheduler_running = 1; |
1da177e4 LT |
7675 | } |
7676 | ||
7677 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
7678 | static inline int preempt_count_equals(int preempt_offset) |
7679 | { | |
234da7bc | 7680 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 FW |
7681 | |
7682 | return (nested == PREEMPT_INATOMIC_BASE + preempt_offset); | |
7683 | } | |
7684 | ||
d894837f | 7685 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7686 | { |
48f24c4d | 7687 | #ifdef in_atomic |
1da177e4 LT |
7688 | static unsigned long prev_jiffy; /* ratelimiting */ |
7689 | ||
e4aafea2 FW |
7690 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
7691 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
7692 | return; |
7693 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7694 | return; | |
7695 | prev_jiffy = jiffies; | |
7696 | ||
3df0fc5b PZ |
7697 | printk(KERN_ERR |
7698 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7699 | file, line); | |
7700 | printk(KERN_ERR | |
7701 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7702 | in_atomic(), irqs_disabled(), | |
7703 | current->pid, current->comm); | |
aef745fc IM |
7704 | |
7705 | debug_show_held_locks(current); | |
7706 | if (irqs_disabled()) | |
7707 | print_irqtrace_events(current); | |
7708 | dump_stack(); | |
1da177e4 LT |
7709 | #endif |
7710 | } | |
7711 | EXPORT_SYMBOL(__might_sleep); | |
7712 | #endif | |
7713 | ||
7714 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7715 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7716 | { | |
7717 | int on_rq; | |
3e51f33f | 7718 | |
3a5e4dc1 AK |
7719 | on_rq = p->se.on_rq; |
7720 | if (on_rq) | |
7721 | deactivate_task(rq, p, 0); | |
7722 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
7723 | if (on_rq) { | |
7724 | activate_task(rq, p, 0); | |
7725 | resched_task(rq->curr); | |
7726 | } | |
7727 | } | |
7728 | ||
1da177e4 LT |
7729 | void normalize_rt_tasks(void) |
7730 | { | |
a0f98a1c | 7731 | struct task_struct *g, *p; |
1da177e4 | 7732 | unsigned long flags; |
70b97a7f | 7733 | struct rq *rq; |
1da177e4 | 7734 | |
4cf5d77a | 7735 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7736 | do_each_thread(g, p) { |
178be793 IM |
7737 | /* |
7738 | * Only normalize user tasks: | |
7739 | */ | |
7740 | if (!p->mm) | |
7741 | continue; | |
7742 | ||
6cfb0d5d | 7743 | p->se.exec_start = 0; |
6cfb0d5d | 7744 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7745 | p->se.statistics.wait_start = 0; |
7746 | p->se.statistics.sleep_start = 0; | |
7747 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7748 | #endif |
dd41f596 IM |
7749 | |
7750 | if (!rt_task(p)) { | |
7751 | /* | |
7752 | * Renice negative nice level userspace | |
7753 | * tasks back to 0: | |
7754 | */ | |
7755 | if (TASK_NICE(p) < 0 && p->mm) | |
7756 | set_user_nice(p, 0); | |
1da177e4 | 7757 | continue; |
dd41f596 | 7758 | } |
1da177e4 | 7759 | |
1d615482 | 7760 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7761 | rq = __task_rq_lock(p); |
1da177e4 | 7762 | |
178be793 | 7763 | normalize_task(rq, p); |
3a5e4dc1 | 7764 | |
b29739f9 | 7765 | __task_rq_unlock(rq); |
1d615482 | 7766 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7767 | } while_each_thread(g, p); |
7768 | ||
4cf5d77a | 7769 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7770 | } |
7771 | ||
7772 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7773 | |
67fc4e0c | 7774 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7775 | /* |
67fc4e0c | 7776 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7777 | * |
7778 | * They can only be called when the whole system has been | |
7779 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7780 | * activity can take place. Using them for anything else would | |
7781 | * be a serious bug, and as a result, they aren't even visible | |
7782 | * under any other configuration. | |
7783 | */ | |
7784 | ||
7785 | /** | |
7786 | * curr_task - return the current task for a given cpu. | |
7787 | * @cpu: the processor in question. | |
7788 | * | |
7789 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7790 | */ | |
36c8b586 | 7791 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7792 | { |
7793 | return cpu_curr(cpu); | |
7794 | } | |
7795 | ||
67fc4e0c JW |
7796 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7797 | ||
7798 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7799 | /** |
7800 | * set_curr_task - set the current task for a given cpu. | |
7801 | * @cpu: the processor in question. | |
7802 | * @p: the task pointer to set. | |
7803 | * | |
7804 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7805 | * are serviced on a separate stack. It allows the architecture to switch the |
7806 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7807 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7808 | * and caller must save the original value of the current task (see | |
7809 | * curr_task() above) and restore that value before reenabling interrupts and | |
7810 | * re-starting the system. | |
7811 | * | |
7812 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7813 | */ | |
36c8b586 | 7814 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7815 | { |
7816 | cpu_curr(cpu) = p; | |
7817 | } | |
7818 | ||
7819 | #endif | |
29f59db3 | 7820 | |
bccbe08a PZ |
7821 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7822 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
7823 | { |
7824 | int i; | |
7825 | ||
7826 | for_each_possible_cpu(i) { | |
7827 | if (tg->cfs_rq) | |
7828 | kfree(tg->cfs_rq[i]); | |
7829 | if (tg->se) | |
7830 | kfree(tg->se[i]); | |
6f505b16 PZ |
7831 | } |
7832 | ||
7833 | kfree(tg->cfs_rq); | |
7834 | kfree(tg->se); | |
6f505b16 PZ |
7835 | } |
7836 | ||
ec7dc8ac DG |
7837 | static |
7838 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 7839 | { |
29f59db3 | 7840 | struct cfs_rq *cfs_rq; |
eab17229 | 7841 | struct sched_entity *se; |
9b5b7751 | 7842 | struct rq *rq; |
29f59db3 SV |
7843 | int i; |
7844 | ||
434d53b0 | 7845 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
7846 | if (!tg->cfs_rq) |
7847 | goto err; | |
434d53b0 | 7848 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
7849 | if (!tg->se) |
7850 | goto err; | |
052f1dc7 PZ |
7851 | |
7852 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
7853 | |
7854 | for_each_possible_cpu(i) { | |
9b5b7751 | 7855 | rq = cpu_rq(i); |
29f59db3 | 7856 | |
eab17229 LZ |
7857 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
7858 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
7859 | if (!cfs_rq) |
7860 | goto err; | |
7861 | ||
eab17229 LZ |
7862 | se = kzalloc_node(sizeof(struct sched_entity), |
7863 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 7864 | if (!se) |
dfc12eb2 | 7865 | goto err_free_rq; |
29f59db3 | 7866 | |
eab17229 | 7867 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
7868 | } |
7869 | ||
7870 | return 1; | |
7871 | ||
dfc12eb2 PC |
7872 | err_free_rq: |
7873 | kfree(cfs_rq); | |
bccbe08a PZ |
7874 | err: |
7875 | return 0; | |
7876 | } | |
7877 | ||
7878 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
7879 | { | |
7880 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
7881 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
7882 | } | |
7883 | ||
7884 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
7885 | { | |
7886 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
7887 | } | |
6d6bc0ad | 7888 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
7889 | static inline void free_fair_sched_group(struct task_group *tg) |
7890 | { | |
7891 | } | |
7892 | ||
ec7dc8ac DG |
7893 | static inline |
7894 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
7895 | { |
7896 | return 1; | |
7897 | } | |
7898 | ||
7899 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
7900 | { | |
7901 | } | |
7902 | ||
7903 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
7904 | { | |
7905 | } | |
6d6bc0ad | 7906 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
7907 | |
7908 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
7909 | static void free_rt_sched_group(struct task_group *tg) |
7910 | { | |
7911 | int i; | |
7912 | ||
d0b27fa7 PZ |
7913 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
7914 | ||
bccbe08a PZ |
7915 | for_each_possible_cpu(i) { |
7916 | if (tg->rt_rq) | |
7917 | kfree(tg->rt_rq[i]); | |
7918 | if (tg->rt_se) | |
7919 | kfree(tg->rt_se[i]); | |
7920 | } | |
7921 | ||
7922 | kfree(tg->rt_rq); | |
7923 | kfree(tg->rt_se); | |
7924 | } | |
7925 | ||
ec7dc8ac DG |
7926 | static |
7927 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
7928 | { |
7929 | struct rt_rq *rt_rq; | |
eab17229 | 7930 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
7931 | struct rq *rq; |
7932 | int i; | |
7933 | ||
434d53b0 | 7934 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
7935 | if (!tg->rt_rq) |
7936 | goto err; | |
434d53b0 | 7937 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
7938 | if (!tg->rt_se) |
7939 | goto err; | |
7940 | ||
d0b27fa7 PZ |
7941 | init_rt_bandwidth(&tg->rt_bandwidth, |
7942 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
7943 | |
7944 | for_each_possible_cpu(i) { | |
7945 | rq = cpu_rq(i); | |
7946 | ||
eab17229 LZ |
7947 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
7948 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
7949 | if (!rt_rq) |
7950 | goto err; | |
29f59db3 | 7951 | |
eab17229 LZ |
7952 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
7953 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 7954 | if (!rt_se) |
dfc12eb2 | 7955 | goto err_free_rq; |
29f59db3 | 7956 | |
eab17229 | 7957 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
7958 | } |
7959 | ||
bccbe08a PZ |
7960 | return 1; |
7961 | ||
dfc12eb2 PC |
7962 | err_free_rq: |
7963 | kfree(rt_rq); | |
bccbe08a PZ |
7964 | err: |
7965 | return 0; | |
7966 | } | |
7967 | ||
7968 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
7969 | { | |
7970 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
7971 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
7972 | } | |
7973 | ||
7974 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
7975 | { | |
7976 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
7977 | } | |
6d6bc0ad | 7978 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
7979 | static inline void free_rt_sched_group(struct task_group *tg) |
7980 | { | |
7981 | } | |
7982 | ||
ec7dc8ac DG |
7983 | static inline |
7984 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
7985 | { |
7986 | return 1; | |
7987 | } | |
7988 | ||
7989 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
7990 | { | |
7991 | } | |
7992 | ||
7993 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
7994 | { | |
7995 | } | |
6d6bc0ad | 7996 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 7997 | |
7c941438 | 7998 | #ifdef CONFIG_CGROUP_SCHED |
bccbe08a PZ |
7999 | static void free_sched_group(struct task_group *tg) |
8000 | { | |
8001 | free_fair_sched_group(tg); | |
8002 | free_rt_sched_group(tg); | |
8003 | kfree(tg); | |
8004 | } | |
8005 | ||
8006 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8007 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8008 | { |
8009 | struct task_group *tg; | |
8010 | unsigned long flags; | |
8011 | int i; | |
8012 | ||
8013 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8014 | if (!tg) | |
8015 | return ERR_PTR(-ENOMEM); | |
8016 | ||
ec7dc8ac | 8017 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8018 | goto err; |
8019 | ||
ec7dc8ac | 8020 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8021 | goto err; |
8022 | ||
8ed36996 | 8023 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8024 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8025 | register_fair_sched_group(tg, i); |
8026 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8027 | } |
6f505b16 | 8028 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8029 | |
8030 | WARN_ON(!parent); /* root should already exist */ | |
8031 | ||
8032 | tg->parent = parent; | |
f473aa5e | 8033 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8034 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8035 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8036 | |
9b5b7751 | 8037 | return tg; |
29f59db3 SV |
8038 | |
8039 | err: | |
6f505b16 | 8040 | free_sched_group(tg); |
29f59db3 SV |
8041 | return ERR_PTR(-ENOMEM); |
8042 | } | |
8043 | ||
9b5b7751 | 8044 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8045 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8046 | { |
29f59db3 | 8047 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8048 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8049 | } |
8050 | ||
9b5b7751 | 8051 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8052 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8053 | { |
8ed36996 | 8054 | unsigned long flags; |
9b5b7751 | 8055 | int i; |
29f59db3 | 8056 | |
8ed36996 | 8057 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8058 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8059 | unregister_fair_sched_group(tg, i); |
8060 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8061 | } |
6f505b16 | 8062 | list_del_rcu(&tg->list); |
f473aa5e | 8063 | list_del_rcu(&tg->siblings); |
8ed36996 | 8064 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8065 | |
9b5b7751 | 8066 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8067 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8068 | } |
8069 | ||
9b5b7751 | 8070 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8071 | * The caller of this function should have put the task in its new group |
8072 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8073 | * reflect its new group. | |
9b5b7751 SV |
8074 | */ |
8075 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8076 | { |
8077 | int on_rq, running; | |
8078 | unsigned long flags; | |
8079 | struct rq *rq; | |
8080 | ||
8081 | rq = task_rq_lock(tsk, &flags); | |
8082 | ||
051a1d1a | 8083 | running = task_current(rq, tsk); |
29f59db3 SV |
8084 | on_rq = tsk->se.on_rq; |
8085 | ||
0e1f3483 | 8086 | if (on_rq) |
29f59db3 | 8087 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8088 | if (unlikely(running)) |
8089 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8090 | |
6f505b16 | 8091 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8092 | |
810b3817 PZ |
8093 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8094 | if (tsk->sched_class->moved_group) | |
88ec22d3 | 8095 | tsk->sched_class->moved_group(tsk, on_rq); |
810b3817 PZ |
8096 | #endif |
8097 | ||
0e1f3483 HS |
8098 | if (unlikely(running)) |
8099 | tsk->sched_class->set_curr_task(rq); | |
8100 | if (on_rq) | |
371fd7e7 | 8101 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8102 | |
29f59db3 SV |
8103 | task_rq_unlock(rq, &flags); |
8104 | } | |
7c941438 | 8105 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8106 | |
052f1dc7 | 8107 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8108 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8109 | { |
8110 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8111 | int on_rq; |
8112 | ||
29f59db3 | 8113 | on_rq = se->on_rq; |
62fb1851 | 8114 | if (on_rq) |
29f59db3 SV |
8115 | dequeue_entity(cfs_rq, se, 0); |
8116 | ||
8117 | se->load.weight = shares; | |
e05510d0 | 8118 | se->load.inv_weight = 0; |
29f59db3 | 8119 | |
62fb1851 | 8120 | if (on_rq) |
29f59db3 | 8121 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8122 | } |
62fb1851 | 8123 | |
c09595f6 PZ |
8124 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8125 | { | |
8126 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8127 | struct rq *rq = cfs_rq->rq; | |
8128 | unsigned long flags; | |
8129 | ||
05fa785c | 8130 | raw_spin_lock_irqsave(&rq->lock, flags); |
c09595f6 | 8131 | __set_se_shares(se, shares); |
05fa785c | 8132 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
29f59db3 SV |
8133 | } |
8134 | ||
8ed36996 PZ |
8135 | static DEFINE_MUTEX(shares_mutex); |
8136 | ||
4cf86d77 | 8137 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8138 | { |
8139 | int i; | |
8ed36996 | 8140 | unsigned long flags; |
c61935fd | 8141 | |
ec7dc8ac DG |
8142 | /* |
8143 | * We can't change the weight of the root cgroup. | |
8144 | */ | |
8145 | if (!tg->se[0]) | |
8146 | return -EINVAL; | |
8147 | ||
18d95a28 PZ |
8148 | if (shares < MIN_SHARES) |
8149 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8150 | else if (shares > MAX_SHARES) |
8151 | shares = MAX_SHARES; | |
62fb1851 | 8152 | |
8ed36996 | 8153 | mutex_lock(&shares_mutex); |
9b5b7751 | 8154 | if (tg->shares == shares) |
5cb350ba | 8155 | goto done; |
29f59db3 | 8156 | |
8ed36996 | 8157 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8158 | for_each_possible_cpu(i) |
8159 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8160 | list_del_rcu(&tg->siblings); |
8ed36996 | 8161 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8162 | |
8163 | /* wait for any ongoing reference to this group to finish */ | |
8164 | synchronize_sched(); | |
8165 | ||
8166 | /* | |
8167 | * Now we are free to modify the group's share on each cpu | |
8168 | * w/o tripping rebalance_share or load_balance_fair. | |
8169 | */ | |
9b5b7751 | 8170 | tg->shares = shares; |
c09595f6 PZ |
8171 | for_each_possible_cpu(i) { |
8172 | /* | |
8173 | * force a rebalance | |
8174 | */ | |
8175 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8176 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8177 | } |
29f59db3 | 8178 | |
6b2d7700 SV |
8179 | /* |
8180 | * Enable load balance activity on this group, by inserting it back on | |
8181 | * each cpu's rq->leaf_cfs_rq_list. | |
8182 | */ | |
8ed36996 | 8183 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8184 | for_each_possible_cpu(i) |
8185 | register_fair_sched_group(tg, i); | |
f473aa5e | 8186 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8187 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8188 | done: |
8ed36996 | 8189 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8190 | return 0; |
29f59db3 SV |
8191 | } |
8192 | ||
5cb350ba DG |
8193 | unsigned long sched_group_shares(struct task_group *tg) |
8194 | { | |
8195 | return tg->shares; | |
8196 | } | |
052f1dc7 | 8197 | #endif |
5cb350ba | 8198 | |
052f1dc7 | 8199 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8200 | /* |
9f0c1e56 | 8201 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8202 | */ |
9f0c1e56 PZ |
8203 | static DEFINE_MUTEX(rt_constraints_mutex); |
8204 | ||
8205 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8206 | { | |
8207 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8208 | return 1ULL << 20; |
9f0c1e56 | 8209 | |
9a7e0b18 | 8210 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8211 | } |
8212 | ||
9a7e0b18 PZ |
8213 | /* Must be called with tasklist_lock held */ |
8214 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8215 | { |
9a7e0b18 | 8216 | struct task_struct *g, *p; |
b40b2e8e | 8217 | |
9a7e0b18 PZ |
8218 | do_each_thread(g, p) { |
8219 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8220 | return 1; | |
8221 | } while_each_thread(g, p); | |
b40b2e8e | 8222 | |
9a7e0b18 PZ |
8223 | return 0; |
8224 | } | |
b40b2e8e | 8225 | |
9a7e0b18 PZ |
8226 | struct rt_schedulable_data { |
8227 | struct task_group *tg; | |
8228 | u64 rt_period; | |
8229 | u64 rt_runtime; | |
8230 | }; | |
b40b2e8e | 8231 | |
9a7e0b18 PZ |
8232 | static int tg_schedulable(struct task_group *tg, void *data) |
8233 | { | |
8234 | struct rt_schedulable_data *d = data; | |
8235 | struct task_group *child; | |
8236 | unsigned long total, sum = 0; | |
8237 | u64 period, runtime; | |
b40b2e8e | 8238 | |
9a7e0b18 PZ |
8239 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8240 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8241 | |
9a7e0b18 PZ |
8242 | if (tg == d->tg) { |
8243 | period = d->rt_period; | |
8244 | runtime = d->rt_runtime; | |
b40b2e8e | 8245 | } |
b40b2e8e | 8246 | |
4653f803 PZ |
8247 | /* |
8248 | * Cannot have more runtime than the period. | |
8249 | */ | |
8250 | if (runtime > period && runtime != RUNTIME_INF) | |
8251 | return -EINVAL; | |
6f505b16 | 8252 | |
4653f803 PZ |
8253 | /* |
8254 | * Ensure we don't starve existing RT tasks. | |
8255 | */ | |
9a7e0b18 PZ |
8256 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8257 | return -EBUSY; | |
6f505b16 | 8258 | |
9a7e0b18 | 8259 | total = to_ratio(period, runtime); |
6f505b16 | 8260 | |
4653f803 PZ |
8261 | /* |
8262 | * Nobody can have more than the global setting allows. | |
8263 | */ | |
8264 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8265 | return -EINVAL; | |
6f505b16 | 8266 | |
4653f803 PZ |
8267 | /* |
8268 | * The sum of our children's runtime should not exceed our own. | |
8269 | */ | |
9a7e0b18 PZ |
8270 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8271 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8272 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8273 | |
9a7e0b18 PZ |
8274 | if (child == d->tg) { |
8275 | period = d->rt_period; | |
8276 | runtime = d->rt_runtime; | |
8277 | } | |
6f505b16 | 8278 | |
9a7e0b18 | 8279 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8280 | } |
6f505b16 | 8281 | |
9a7e0b18 PZ |
8282 | if (sum > total) |
8283 | return -EINVAL; | |
8284 | ||
8285 | return 0; | |
6f505b16 PZ |
8286 | } |
8287 | ||
9a7e0b18 | 8288 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8289 | { |
9a7e0b18 PZ |
8290 | struct rt_schedulable_data data = { |
8291 | .tg = tg, | |
8292 | .rt_period = period, | |
8293 | .rt_runtime = runtime, | |
8294 | }; | |
8295 | ||
8296 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8297 | } |
8298 | ||
d0b27fa7 PZ |
8299 | static int tg_set_bandwidth(struct task_group *tg, |
8300 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8301 | { |
ac086bc2 | 8302 | int i, err = 0; |
9f0c1e56 | 8303 | |
9f0c1e56 | 8304 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8305 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8306 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8307 | if (err) | |
9f0c1e56 | 8308 | goto unlock; |
ac086bc2 | 8309 | |
0986b11b | 8310 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8311 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8312 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8313 | |
8314 | for_each_possible_cpu(i) { | |
8315 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8316 | ||
0986b11b | 8317 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8318 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8319 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8320 | } |
0986b11b | 8321 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
9f0c1e56 | 8322 | unlock: |
521f1a24 | 8323 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8324 | mutex_unlock(&rt_constraints_mutex); |
8325 | ||
8326 | return err; | |
6f505b16 PZ |
8327 | } |
8328 | ||
d0b27fa7 PZ |
8329 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8330 | { | |
8331 | u64 rt_runtime, rt_period; | |
8332 | ||
8333 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8334 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8335 | if (rt_runtime_us < 0) | |
8336 | rt_runtime = RUNTIME_INF; | |
8337 | ||
8338 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8339 | } | |
8340 | ||
9f0c1e56 PZ |
8341 | long sched_group_rt_runtime(struct task_group *tg) |
8342 | { | |
8343 | u64 rt_runtime_us; | |
8344 | ||
d0b27fa7 | 8345 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8346 | return -1; |
8347 | ||
d0b27fa7 | 8348 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8349 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8350 | return rt_runtime_us; | |
8351 | } | |
d0b27fa7 PZ |
8352 | |
8353 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8354 | { | |
8355 | u64 rt_runtime, rt_period; | |
8356 | ||
8357 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8358 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8359 | ||
619b0488 R |
8360 | if (rt_period == 0) |
8361 | return -EINVAL; | |
8362 | ||
d0b27fa7 PZ |
8363 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8364 | } | |
8365 | ||
8366 | long sched_group_rt_period(struct task_group *tg) | |
8367 | { | |
8368 | u64 rt_period_us; | |
8369 | ||
8370 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8371 | do_div(rt_period_us, NSEC_PER_USEC); | |
8372 | return rt_period_us; | |
8373 | } | |
8374 | ||
8375 | static int sched_rt_global_constraints(void) | |
8376 | { | |
4653f803 | 8377 | u64 runtime, period; |
d0b27fa7 PZ |
8378 | int ret = 0; |
8379 | ||
ec5d4989 HS |
8380 | if (sysctl_sched_rt_period <= 0) |
8381 | return -EINVAL; | |
8382 | ||
4653f803 PZ |
8383 | runtime = global_rt_runtime(); |
8384 | period = global_rt_period(); | |
8385 | ||
8386 | /* | |
8387 | * Sanity check on the sysctl variables. | |
8388 | */ | |
8389 | if (runtime > period && runtime != RUNTIME_INF) | |
8390 | return -EINVAL; | |
10b612f4 | 8391 | |
d0b27fa7 | 8392 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 8393 | read_lock(&tasklist_lock); |
4653f803 | 8394 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 8395 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
8396 | mutex_unlock(&rt_constraints_mutex); |
8397 | ||
8398 | return ret; | |
8399 | } | |
54e99124 DG |
8400 | |
8401 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
8402 | { | |
8403 | /* Don't accept realtime tasks when there is no way for them to run */ | |
8404 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
8405 | return 0; | |
8406 | ||
8407 | return 1; | |
8408 | } | |
8409 | ||
6d6bc0ad | 8410 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8411 | static int sched_rt_global_constraints(void) |
8412 | { | |
ac086bc2 PZ |
8413 | unsigned long flags; |
8414 | int i; | |
8415 | ||
ec5d4989 HS |
8416 | if (sysctl_sched_rt_period <= 0) |
8417 | return -EINVAL; | |
8418 | ||
60aa605d PZ |
8419 | /* |
8420 | * There's always some RT tasks in the root group | |
8421 | * -- migration, kstopmachine etc.. | |
8422 | */ | |
8423 | if (sysctl_sched_rt_runtime == 0) | |
8424 | return -EBUSY; | |
8425 | ||
0986b11b | 8426 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
8427 | for_each_possible_cpu(i) { |
8428 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8429 | ||
0986b11b | 8430 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8431 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 8432 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8433 | } |
0986b11b | 8434 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 8435 | |
d0b27fa7 PZ |
8436 | return 0; |
8437 | } | |
6d6bc0ad | 8438 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8439 | |
8440 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 8441 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8442 | loff_t *ppos) |
8443 | { | |
8444 | int ret; | |
8445 | int old_period, old_runtime; | |
8446 | static DEFINE_MUTEX(mutex); | |
8447 | ||
8448 | mutex_lock(&mutex); | |
8449 | old_period = sysctl_sched_rt_period; | |
8450 | old_runtime = sysctl_sched_rt_runtime; | |
8451 | ||
8d65af78 | 8452 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
8453 | |
8454 | if (!ret && write) { | |
8455 | ret = sched_rt_global_constraints(); | |
8456 | if (ret) { | |
8457 | sysctl_sched_rt_period = old_period; | |
8458 | sysctl_sched_rt_runtime = old_runtime; | |
8459 | } else { | |
8460 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8461 | def_rt_bandwidth.rt_period = | |
8462 | ns_to_ktime(global_rt_period()); | |
8463 | } | |
8464 | } | |
8465 | mutex_unlock(&mutex); | |
8466 | ||
8467 | return ret; | |
8468 | } | |
68318b8e | 8469 | |
052f1dc7 | 8470 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8471 | |
8472 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8473 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8474 | { |
2b01dfe3 PM |
8475 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8476 | struct task_group, css); | |
68318b8e SV |
8477 | } |
8478 | ||
8479 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8480 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8481 | { |
ec7dc8ac | 8482 | struct task_group *tg, *parent; |
68318b8e | 8483 | |
2b01dfe3 | 8484 | if (!cgrp->parent) { |
68318b8e | 8485 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
8486 | return &init_task_group.css; |
8487 | } | |
8488 | ||
ec7dc8ac DG |
8489 | parent = cgroup_tg(cgrp->parent); |
8490 | tg = sched_create_group(parent); | |
68318b8e SV |
8491 | if (IS_ERR(tg)) |
8492 | return ERR_PTR(-ENOMEM); | |
8493 | ||
68318b8e SV |
8494 | return &tg->css; |
8495 | } | |
8496 | ||
41a2d6cf IM |
8497 | static void |
8498 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8499 | { |
2b01dfe3 | 8500 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8501 | |
8502 | sched_destroy_group(tg); | |
8503 | } | |
8504 | ||
41a2d6cf | 8505 | static int |
be367d09 | 8506 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 8507 | { |
b68aa230 | 8508 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 8509 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
8510 | return -EINVAL; |
8511 | #else | |
68318b8e SV |
8512 | /* We don't support RT-tasks being in separate groups */ |
8513 | if (tsk->sched_class != &fair_sched_class) | |
8514 | return -EINVAL; | |
b68aa230 | 8515 | #endif |
be367d09 BB |
8516 | return 0; |
8517 | } | |
68318b8e | 8518 | |
be367d09 BB |
8519 | static int |
8520 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8521 | struct task_struct *tsk, bool threadgroup) | |
8522 | { | |
8523 | int retval = cpu_cgroup_can_attach_task(cgrp, tsk); | |
8524 | if (retval) | |
8525 | return retval; | |
8526 | if (threadgroup) { | |
8527 | struct task_struct *c; | |
8528 | rcu_read_lock(); | |
8529 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8530 | retval = cpu_cgroup_can_attach_task(cgrp, c); | |
8531 | if (retval) { | |
8532 | rcu_read_unlock(); | |
8533 | return retval; | |
8534 | } | |
8535 | } | |
8536 | rcu_read_unlock(); | |
8537 | } | |
68318b8e SV |
8538 | return 0; |
8539 | } | |
8540 | ||
8541 | static void | |
2b01dfe3 | 8542 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
be367d09 BB |
8543 | struct cgroup *old_cont, struct task_struct *tsk, |
8544 | bool threadgroup) | |
68318b8e SV |
8545 | { |
8546 | sched_move_task(tsk); | |
be367d09 BB |
8547 | if (threadgroup) { |
8548 | struct task_struct *c; | |
8549 | rcu_read_lock(); | |
8550 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8551 | sched_move_task(c); | |
8552 | } | |
8553 | rcu_read_unlock(); | |
8554 | } | |
68318b8e SV |
8555 | } |
8556 | ||
052f1dc7 | 8557 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8558 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8559 | u64 shareval) |
68318b8e | 8560 | { |
2b01dfe3 | 8561 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8562 | } |
8563 | ||
f4c753b7 | 8564 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8565 | { |
2b01dfe3 | 8566 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8567 | |
8568 | return (u64) tg->shares; | |
8569 | } | |
6d6bc0ad | 8570 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8571 | |
052f1dc7 | 8572 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8573 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8574 | s64 val) |
6f505b16 | 8575 | { |
06ecb27c | 8576 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8577 | } |
8578 | ||
06ecb27c | 8579 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8580 | { |
06ecb27c | 8581 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8582 | } |
d0b27fa7 PZ |
8583 | |
8584 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8585 | u64 rt_period_us) | |
8586 | { | |
8587 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8588 | } | |
8589 | ||
8590 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8591 | { | |
8592 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8593 | } | |
6d6bc0ad | 8594 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8595 | |
fe5c7cc2 | 8596 | static struct cftype cpu_files[] = { |
052f1dc7 | 8597 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8598 | { |
8599 | .name = "shares", | |
f4c753b7 PM |
8600 | .read_u64 = cpu_shares_read_u64, |
8601 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8602 | }, |
052f1dc7 PZ |
8603 | #endif |
8604 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8605 | { |
9f0c1e56 | 8606 | .name = "rt_runtime_us", |
06ecb27c PM |
8607 | .read_s64 = cpu_rt_runtime_read, |
8608 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8609 | }, |
d0b27fa7 PZ |
8610 | { |
8611 | .name = "rt_period_us", | |
f4c753b7 PM |
8612 | .read_u64 = cpu_rt_period_read_uint, |
8613 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8614 | }, |
052f1dc7 | 8615 | #endif |
68318b8e SV |
8616 | }; |
8617 | ||
8618 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8619 | { | |
fe5c7cc2 | 8620 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8621 | } |
8622 | ||
8623 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8624 | .name = "cpu", |
8625 | .create = cpu_cgroup_create, | |
8626 | .destroy = cpu_cgroup_destroy, | |
8627 | .can_attach = cpu_cgroup_can_attach, | |
8628 | .attach = cpu_cgroup_attach, | |
8629 | .populate = cpu_cgroup_populate, | |
8630 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8631 | .early_init = 1, |
8632 | }; | |
8633 | ||
052f1dc7 | 8634 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8635 | |
8636 | #ifdef CONFIG_CGROUP_CPUACCT | |
8637 | ||
8638 | /* | |
8639 | * CPU accounting code for task groups. | |
8640 | * | |
8641 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8642 | * (balbir@in.ibm.com). | |
8643 | */ | |
8644 | ||
934352f2 | 8645 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
8646 | struct cpuacct { |
8647 | struct cgroup_subsys_state css; | |
8648 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43cf38eb | 8649 | u64 __percpu *cpuusage; |
ef12fefa | 8650 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 8651 | struct cpuacct *parent; |
d842de87 SV |
8652 | }; |
8653 | ||
8654 | struct cgroup_subsys cpuacct_subsys; | |
8655 | ||
8656 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8657 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8658 | { |
32cd756a | 8659 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8660 | struct cpuacct, css); |
8661 | } | |
8662 | ||
8663 | /* return cpu accounting group to which this task belongs */ | |
8664 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8665 | { | |
8666 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8667 | struct cpuacct, css); | |
8668 | } | |
8669 | ||
8670 | /* create a new cpu accounting group */ | |
8671 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8672 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
8673 | { |
8674 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 8675 | int i; |
d842de87 SV |
8676 | |
8677 | if (!ca) | |
ef12fefa | 8678 | goto out; |
d842de87 SV |
8679 | |
8680 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
8681 | if (!ca->cpuusage) |
8682 | goto out_free_ca; | |
8683 | ||
8684 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
8685 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
8686 | goto out_free_counters; | |
d842de87 | 8687 | |
934352f2 BR |
8688 | if (cgrp->parent) |
8689 | ca->parent = cgroup_ca(cgrp->parent); | |
8690 | ||
d842de87 | 8691 | return &ca->css; |
ef12fefa BR |
8692 | |
8693 | out_free_counters: | |
8694 | while (--i >= 0) | |
8695 | percpu_counter_destroy(&ca->cpustat[i]); | |
8696 | free_percpu(ca->cpuusage); | |
8697 | out_free_ca: | |
8698 | kfree(ca); | |
8699 | out: | |
8700 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
8701 | } |
8702 | ||
8703 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 8704 | static void |
32cd756a | 8705 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8706 | { |
32cd756a | 8707 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 8708 | int i; |
d842de87 | 8709 | |
ef12fefa BR |
8710 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
8711 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
8712 | free_percpu(ca->cpuusage); |
8713 | kfree(ca); | |
8714 | } | |
8715 | ||
720f5498 KC |
8716 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
8717 | { | |
b36128c8 | 8718 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8719 | u64 data; |
8720 | ||
8721 | #ifndef CONFIG_64BIT | |
8722 | /* | |
8723 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
8724 | */ | |
05fa785c | 8725 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8726 | data = *cpuusage; |
05fa785c | 8727 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8728 | #else |
8729 | data = *cpuusage; | |
8730 | #endif | |
8731 | ||
8732 | return data; | |
8733 | } | |
8734 | ||
8735 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
8736 | { | |
b36128c8 | 8737 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8738 | |
8739 | #ifndef CONFIG_64BIT | |
8740 | /* | |
8741 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
8742 | */ | |
05fa785c | 8743 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8744 | *cpuusage = val; |
05fa785c | 8745 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8746 | #else |
8747 | *cpuusage = val; | |
8748 | #endif | |
8749 | } | |
8750 | ||
d842de87 | 8751 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 8752 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 8753 | { |
32cd756a | 8754 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8755 | u64 totalcpuusage = 0; |
8756 | int i; | |
8757 | ||
720f5498 KC |
8758 | for_each_present_cpu(i) |
8759 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
8760 | |
8761 | return totalcpuusage; | |
8762 | } | |
8763 | ||
0297b803 DG |
8764 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
8765 | u64 reset) | |
8766 | { | |
8767 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8768 | int err = 0; | |
8769 | int i; | |
8770 | ||
8771 | if (reset) { | |
8772 | err = -EINVAL; | |
8773 | goto out; | |
8774 | } | |
8775 | ||
720f5498 KC |
8776 | for_each_present_cpu(i) |
8777 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 8778 | |
0297b803 DG |
8779 | out: |
8780 | return err; | |
8781 | } | |
8782 | ||
e9515c3c KC |
8783 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
8784 | struct seq_file *m) | |
8785 | { | |
8786 | struct cpuacct *ca = cgroup_ca(cgroup); | |
8787 | u64 percpu; | |
8788 | int i; | |
8789 | ||
8790 | for_each_present_cpu(i) { | |
8791 | percpu = cpuacct_cpuusage_read(ca, i); | |
8792 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
8793 | } | |
8794 | seq_printf(m, "\n"); | |
8795 | return 0; | |
8796 | } | |
8797 | ||
ef12fefa BR |
8798 | static const char *cpuacct_stat_desc[] = { |
8799 | [CPUACCT_STAT_USER] = "user", | |
8800 | [CPUACCT_STAT_SYSTEM] = "system", | |
8801 | }; | |
8802 | ||
8803 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
8804 | struct cgroup_map_cb *cb) | |
8805 | { | |
8806 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8807 | int i; | |
8808 | ||
8809 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
8810 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
8811 | val = cputime64_to_clock_t(val); | |
8812 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
8813 | } | |
8814 | return 0; | |
8815 | } | |
8816 | ||
d842de87 SV |
8817 | static struct cftype files[] = { |
8818 | { | |
8819 | .name = "usage", | |
f4c753b7 PM |
8820 | .read_u64 = cpuusage_read, |
8821 | .write_u64 = cpuusage_write, | |
d842de87 | 8822 | }, |
e9515c3c KC |
8823 | { |
8824 | .name = "usage_percpu", | |
8825 | .read_seq_string = cpuacct_percpu_seq_read, | |
8826 | }, | |
ef12fefa BR |
8827 | { |
8828 | .name = "stat", | |
8829 | .read_map = cpuacct_stats_show, | |
8830 | }, | |
d842de87 SV |
8831 | }; |
8832 | ||
32cd756a | 8833 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8834 | { |
32cd756a | 8835 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
8836 | } |
8837 | ||
8838 | /* | |
8839 | * charge this task's execution time to its accounting group. | |
8840 | * | |
8841 | * called with rq->lock held. | |
8842 | */ | |
8843 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
8844 | { | |
8845 | struct cpuacct *ca; | |
934352f2 | 8846 | int cpu; |
d842de87 | 8847 | |
c40c6f85 | 8848 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
8849 | return; |
8850 | ||
934352f2 | 8851 | cpu = task_cpu(tsk); |
a18b83b7 BR |
8852 | |
8853 | rcu_read_lock(); | |
8854 | ||
d842de87 | 8855 | ca = task_ca(tsk); |
d842de87 | 8856 | |
934352f2 | 8857 | for (; ca; ca = ca->parent) { |
b36128c8 | 8858 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
8859 | *cpuusage += cputime; |
8860 | } | |
a18b83b7 BR |
8861 | |
8862 | rcu_read_unlock(); | |
d842de87 SV |
8863 | } |
8864 | ||
fa535a77 AB |
8865 | /* |
8866 | * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large | |
8867 | * in cputime_t units. As a result, cpuacct_update_stats calls | |
8868 | * percpu_counter_add with values large enough to always overflow the | |
8869 | * per cpu batch limit causing bad SMP scalability. | |
8870 | * | |
8871 | * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we | |
8872 | * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled | |
8873 | * and enabled. We cap it at INT_MAX which is the largest allowed batch value. | |
8874 | */ | |
8875 | #ifdef CONFIG_SMP | |
8876 | #define CPUACCT_BATCH \ | |
8877 | min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX) | |
8878 | #else | |
8879 | #define CPUACCT_BATCH 0 | |
8880 | #endif | |
8881 | ||
ef12fefa BR |
8882 | /* |
8883 | * Charge the system/user time to the task's accounting group. | |
8884 | */ | |
8885 | static void cpuacct_update_stats(struct task_struct *tsk, | |
8886 | enum cpuacct_stat_index idx, cputime_t val) | |
8887 | { | |
8888 | struct cpuacct *ca; | |
fa535a77 | 8889 | int batch = CPUACCT_BATCH; |
ef12fefa BR |
8890 | |
8891 | if (unlikely(!cpuacct_subsys.active)) | |
8892 | return; | |
8893 | ||
8894 | rcu_read_lock(); | |
8895 | ca = task_ca(tsk); | |
8896 | ||
8897 | do { | |
fa535a77 | 8898 | __percpu_counter_add(&ca->cpustat[idx], val, batch); |
ef12fefa BR |
8899 | ca = ca->parent; |
8900 | } while (ca); | |
8901 | rcu_read_unlock(); | |
8902 | } | |
8903 | ||
d842de87 SV |
8904 | struct cgroup_subsys cpuacct_subsys = { |
8905 | .name = "cpuacct", | |
8906 | .create = cpuacct_create, | |
8907 | .destroy = cpuacct_destroy, | |
8908 | .populate = cpuacct_populate, | |
8909 | .subsys_id = cpuacct_subsys_id, | |
8910 | }; | |
8911 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf PM |
8912 | |
8913 | #ifndef CONFIG_SMP | |
8914 | ||
03b042bf PM |
8915 | void synchronize_sched_expedited(void) |
8916 | { | |
fc390cde | 8917 | barrier(); |
03b042bf PM |
8918 | } |
8919 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
8920 | ||
8921 | #else /* #ifndef CONFIG_SMP */ | |
8922 | ||
cc631fb7 | 8923 | static atomic_t synchronize_sched_expedited_count = ATOMIC_INIT(0); |
03b042bf | 8924 | |
cc631fb7 | 8925 | static int synchronize_sched_expedited_cpu_stop(void *data) |
03b042bf | 8926 | { |
969c7921 TH |
8927 | /* |
8928 | * There must be a full memory barrier on each affected CPU | |
8929 | * between the time that try_stop_cpus() is called and the | |
8930 | * time that it returns. | |
8931 | * | |
8932 | * In the current initial implementation of cpu_stop, the | |
8933 | * above condition is already met when the control reaches | |
8934 | * this point and the following smp_mb() is not strictly | |
8935 | * necessary. Do smp_mb() anyway for documentation and | |
8936 | * robustness against future implementation changes. | |
8937 | */ | |
cc631fb7 | 8938 | smp_mb(); /* See above comment block. */ |
969c7921 | 8939 | return 0; |
03b042bf | 8940 | } |
03b042bf PM |
8941 | |
8942 | /* | |
8943 | * Wait for an rcu-sched grace period to elapse, but use "big hammer" | |
8944 | * approach to force grace period to end quickly. This consumes | |
8945 | * significant time on all CPUs, and is thus not recommended for | |
8946 | * any sort of common-case code. | |
8947 | * | |
8948 | * Note that it is illegal to call this function while holding any | |
8949 | * lock that is acquired by a CPU-hotplug notifier. Failing to | |
8950 | * observe this restriction will result in deadlock. | |
8951 | */ | |
8952 | void synchronize_sched_expedited(void) | |
8953 | { | |
969c7921 | 8954 | int snap, trycount = 0; |
03b042bf PM |
8955 | |
8956 | smp_mb(); /* ensure prior mod happens before capturing snap. */ | |
969c7921 | 8957 | snap = atomic_read(&synchronize_sched_expedited_count) + 1; |
03b042bf | 8958 | get_online_cpus(); |
969c7921 TH |
8959 | while (try_stop_cpus(cpu_online_mask, |
8960 | synchronize_sched_expedited_cpu_stop, | |
94458d5e | 8961 | NULL) == -EAGAIN) { |
03b042bf PM |
8962 | put_online_cpus(); |
8963 | if (trycount++ < 10) | |
8964 | udelay(trycount * num_online_cpus()); | |
8965 | else { | |
8966 | synchronize_sched(); | |
8967 | return; | |
8968 | } | |
969c7921 | 8969 | if (atomic_read(&synchronize_sched_expedited_count) - snap > 0) { |
03b042bf PM |
8970 | smp_mb(); /* ensure test happens before caller kfree */ |
8971 | return; | |
8972 | } | |
8973 | get_online_cpus(); | |
8974 | } | |
969c7921 | 8975 | atomic_inc(&synchronize_sched_expedited_count); |
cc631fb7 | 8976 | smp_mb__after_atomic_inc(); /* ensure post-GP actions seen after GP. */ |
03b042bf | 8977 | put_online_cpus(); |
03b042bf PM |
8978 | } |
8979 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
8980 | ||
8981 | #endif /* #else #ifndef CONFIG_SMP */ |