]>
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 | 34 | #include <linux/highmem.h> |
1da177e4 LT |
35 | #include <asm/mmu_context.h> |
36 | #include <linux/interrupt.h> | |
c59ede7b | 37 | #include <linux/capability.h> |
1da177e4 LT |
38 | #include <linux/completion.h> |
39 | #include <linux/kernel_stat.h> | |
9a11b49a | 40 | #include <linux/debug_locks.h> |
cdd6c482 | 41 | #include <linux/perf_event.h> |
1da177e4 LT |
42 | #include <linux/security.h> |
43 | #include <linux/notifier.h> | |
44 | #include <linux/profile.h> | |
7dfb7103 | 45 | #include <linux/freezer.h> |
198e2f18 | 46 | #include <linux/vmalloc.h> |
1da177e4 LT |
47 | #include <linux/blkdev.h> |
48 | #include <linux/delay.h> | |
b488893a | 49 | #include <linux/pid_namespace.h> |
1da177e4 LT |
50 | #include <linux/smp.h> |
51 | #include <linux/threads.h> | |
52 | #include <linux/timer.h> | |
53 | #include <linux/rcupdate.h> | |
54 | #include <linux/cpu.h> | |
55 | #include <linux/cpuset.h> | |
56 | #include <linux/percpu.h> | |
b5aadf7f | 57 | #include <linux/proc_fs.h> |
1da177e4 | 58 | #include <linux/seq_file.h> |
969c7921 | 59 | #include <linux/stop_machine.h> |
e692ab53 | 60 | #include <linux/sysctl.h> |
1da177e4 LT |
61 | #include <linux/syscalls.h> |
62 | #include <linux/times.h> | |
8f0ab514 | 63 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 64 | #include <linux/kprobes.h> |
0ff92245 | 65 | #include <linux/delayacct.h> |
dff06c15 | 66 | #include <linux/unistd.h> |
f5ff8422 | 67 | #include <linux/pagemap.h> |
8f4d37ec | 68 | #include <linux/hrtimer.h> |
30914a58 | 69 | #include <linux/tick.h> |
f00b45c1 PZ |
70 | #include <linux/debugfs.h> |
71 | #include <linux/ctype.h> | |
6cd8a4bb | 72 | #include <linux/ftrace.h> |
5a0e3ad6 | 73 | #include <linux/slab.h> |
1da177e4 | 74 | |
5517d86b | 75 | #include <asm/tlb.h> |
838225b4 | 76 | #include <asm/irq_regs.h> |
335d7afb | 77 | #include <asm/mutex.h> |
e6e6685a GC |
78 | #ifdef CONFIG_PARAVIRT |
79 | #include <asm/paravirt.h> | |
80 | #endif | |
1da177e4 | 81 | |
6e0534f2 | 82 | #include "sched_cpupri.h" |
21aa9af0 | 83 | #include "workqueue_sched.h" |
5091faa4 | 84 | #include "sched_autogroup.h" |
6e0534f2 | 85 | |
a8d154b0 | 86 | #define CREATE_TRACE_POINTS |
ad8d75ff | 87 | #include <trace/events/sched.h> |
a8d154b0 | 88 | |
1da177e4 LT |
89 | /* |
90 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
91 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
92 | * and back. | |
93 | */ | |
94 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
95 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
96 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
97 | ||
98 | /* | |
99 | * 'User priority' is the nice value converted to something we | |
100 | * can work with better when scaling various scheduler parameters, | |
101 | * it's a [ 0 ... 39 ] range. | |
102 | */ | |
103 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
104 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
105 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
106 | ||
107 | /* | |
d7876a08 | 108 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 109 | */ |
d6322faf | 110 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 111 | |
6aa645ea IM |
112 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
113 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
114 | ||
1da177e4 LT |
115 | /* |
116 | * These are the 'tuning knobs' of the scheduler: | |
117 | * | |
a4ec24b4 | 118 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
119 | * Timeslices get refilled after they expire. |
120 | */ | |
1da177e4 | 121 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 122 | |
d0b27fa7 PZ |
123 | /* |
124 | * single value that denotes runtime == period, ie unlimited time. | |
125 | */ | |
126 | #define RUNTIME_INF ((u64)~0ULL) | |
127 | ||
e05606d3 IM |
128 | static inline int rt_policy(int policy) |
129 | { | |
63f01241 | 130 | if (policy == SCHED_FIFO || policy == SCHED_RR) |
e05606d3 IM |
131 | return 1; |
132 | return 0; | |
133 | } | |
134 | ||
135 | static inline int task_has_rt_policy(struct task_struct *p) | |
136 | { | |
137 | return rt_policy(p->policy); | |
138 | } | |
139 | ||
1da177e4 | 140 | /* |
6aa645ea | 141 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 142 | */ |
6aa645ea IM |
143 | struct rt_prio_array { |
144 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
145 | struct list_head queue[MAX_RT_PRIO]; | |
146 | }; | |
147 | ||
d0b27fa7 | 148 | struct rt_bandwidth { |
ea736ed5 | 149 | /* nests inside the rq lock: */ |
0986b11b | 150 | raw_spinlock_t rt_runtime_lock; |
ea736ed5 IM |
151 | ktime_t rt_period; |
152 | u64 rt_runtime; | |
153 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
154 | }; |
155 | ||
156 | static struct rt_bandwidth def_rt_bandwidth; | |
157 | ||
158 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
159 | ||
160 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
161 | { | |
162 | struct rt_bandwidth *rt_b = | |
163 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
164 | ktime_t now; | |
165 | int overrun; | |
166 | int idle = 0; | |
167 | ||
168 | for (;;) { | |
169 | now = hrtimer_cb_get_time(timer); | |
170 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
171 | ||
172 | if (!overrun) | |
173 | break; | |
174 | ||
175 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
176 | } | |
177 | ||
178 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
179 | } | |
180 | ||
181 | static | |
182 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
183 | { | |
184 | rt_b->rt_period = ns_to_ktime(period); | |
185 | rt_b->rt_runtime = runtime; | |
186 | ||
0986b11b | 187 | raw_spin_lock_init(&rt_b->rt_runtime_lock); |
ac086bc2 | 188 | |
d0b27fa7 PZ |
189 | hrtimer_init(&rt_b->rt_period_timer, |
190 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
191 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
192 | } |
193 | ||
c8bfff6d KH |
194 | static inline int rt_bandwidth_enabled(void) |
195 | { | |
196 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
197 | } |
198 | ||
199 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
200 | { | |
201 | ktime_t now; | |
202 | ||
cac64d00 | 203 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
204 | return; |
205 | ||
206 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
207 | return; | |
208 | ||
0986b11b | 209 | raw_spin_lock(&rt_b->rt_runtime_lock); |
d0b27fa7 | 210 | for (;;) { |
7f1e2ca9 PZ |
211 | unsigned long delta; |
212 | ktime_t soft, hard; | |
213 | ||
d0b27fa7 PZ |
214 | if (hrtimer_active(&rt_b->rt_period_timer)) |
215 | break; | |
216 | ||
217 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
218 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
219 | |
220 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
221 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
222 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
223 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
5c333864 | 224 | HRTIMER_MODE_ABS_PINNED, 0); |
d0b27fa7 | 225 | } |
0986b11b | 226 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
d0b27fa7 PZ |
227 | } |
228 | ||
229 | #ifdef CONFIG_RT_GROUP_SCHED | |
230 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
231 | { | |
232 | hrtimer_cancel(&rt_b->rt_period_timer); | |
233 | } | |
234 | #endif | |
235 | ||
712555ee | 236 | /* |
c4a8849a | 237 | * sched_domains_mutex serializes calls to init_sched_domains, |
712555ee HC |
238 | * detach_destroy_domains and partition_sched_domains. |
239 | */ | |
240 | static DEFINE_MUTEX(sched_domains_mutex); | |
241 | ||
7c941438 | 242 | #ifdef CONFIG_CGROUP_SCHED |
29f59db3 | 243 | |
68318b8e SV |
244 | #include <linux/cgroup.h> |
245 | ||
29f59db3 SV |
246 | struct cfs_rq; |
247 | ||
6f505b16 PZ |
248 | static LIST_HEAD(task_groups); |
249 | ||
ab84d31e PT |
250 | struct cfs_bandwidth { |
251 | #ifdef CONFIG_CFS_BANDWIDTH | |
252 | raw_spinlock_t lock; | |
253 | ktime_t period; | |
ec12cb7f | 254 | u64 quota, runtime; |
a790de99 | 255 | s64 hierarchal_quota; |
ab84d31e PT |
256 | #endif |
257 | }; | |
258 | ||
29f59db3 | 259 | /* task group related information */ |
4cf86d77 | 260 | struct task_group { |
68318b8e | 261 | struct cgroup_subsys_state css; |
6c415b92 | 262 | |
052f1dc7 | 263 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
264 | /* schedulable entities of this group on each cpu */ |
265 | struct sched_entity **se; | |
266 | /* runqueue "owned" by this group on each cpu */ | |
267 | struct cfs_rq **cfs_rq; | |
268 | unsigned long shares; | |
2069dd75 PZ |
269 | |
270 | atomic_t load_weight; | |
052f1dc7 PZ |
271 | #endif |
272 | ||
273 | #ifdef CONFIG_RT_GROUP_SCHED | |
274 | struct sched_rt_entity **rt_se; | |
275 | struct rt_rq **rt_rq; | |
276 | ||
d0b27fa7 | 277 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 278 | #endif |
6b2d7700 | 279 | |
ae8393e5 | 280 | struct rcu_head rcu; |
6f505b16 | 281 | struct list_head list; |
f473aa5e PZ |
282 | |
283 | struct task_group *parent; | |
284 | struct list_head siblings; | |
285 | struct list_head children; | |
5091faa4 MG |
286 | |
287 | #ifdef CONFIG_SCHED_AUTOGROUP | |
288 | struct autogroup *autogroup; | |
289 | #endif | |
ab84d31e PT |
290 | |
291 | struct cfs_bandwidth cfs_bandwidth; | |
29f59db3 SV |
292 | }; |
293 | ||
3d4b47b4 | 294 | /* task_group_lock serializes the addition/removal of task groups */ |
8ed36996 | 295 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 296 | |
e9036b36 CG |
297 | #ifdef CONFIG_FAIR_GROUP_SCHED |
298 | ||
07e06b01 | 299 | # define ROOT_TASK_GROUP_LOAD NICE_0_LOAD |
052f1dc7 | 300 | |
cb4ad1ff | 301 | /* |
2e084786 LJ |
302 | * A weight of 0 or 1 can cause arithmetics problems. |
303 | * A weight of a cfs_rq is the sum of weights of which entities | |
304 | * are queued on this cfs_rq, so a weight of a entity should not be | |
305 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
306 | * (The default weight is 1024 - so there's no practical |
307 | * limitation from this.) | |
308 | */ | |
cd62287e MG |
309 | #define MIN_SHARES (1UL << 1) |
310 | #define MAX_SHARES (1UL << 18) | |
18d95a28 | 311 | |
07e06b01 | 312 | static int root_task_group_load = ROOT_TASK_GROUP_LOAD; |
052f1dc7 PZ |
313 | #endif |
314 | ||
29f59db3 | 315 | /* Default task group. |
3a252015 | 316 | * Every task in system belong to this group at bootup. |
29f59db3 | 317 | */ |
07e06b01 | 318 | struct task_group root_task_group; |
29f59db3 | 319 | |
7c941438 | 320 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 321 | |
6aa645ea IM |
322 | /* CFS-related fields in a runqueue */ |
323 | struct cfs_rq { | |
324 | struct load_weight load; | |
953bfcd1 | 325 | unsigned long nr_running, h_nr_running; |
6aa645ea | 326 | |
6aa645ea | 327 | u64 exec_clock; |
e9acbff6 | 328 | u64 min_vruntime; |
3fe1698b PZ |
329 | #ifndef CONFIG_64BIT |
330 | u64 min_vruntime_copy; | |
331 | #endif | |
6aa645ea IM |
332 | |
333 | struct rb_root tasks_timeline; | |
334 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
335 | |
336 | struct list_head tasks; | |
337 | struct list_head *balance_iterator; | |
338 | ||
339 | /* | |
340 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
341 | * It is set to NULL otherwise (i.e when none are currently running). |
342 | */ | |
ac53db59 | 343 | struct sched_entity *curr, *next, *last, *skip; |
ddc97297 | 344 | |
4934a4d3 | 345 | #ifdef CONFIG_SCHED_DEBUG |
5ac5c4d6 | 346 | unsigned int nr_spread_over; |
4934a4d3 | 347 | #endif |
ddc97297 | 348 | |
62160e3f | 349 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
350 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
351 | ||
41a2d6cf IM |
352 | /* |
353 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
354 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
355 | * (like users, containers etc.) | |
356 | * | |
357 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
358 | * list is used during load balance. | |
359 | */ | |
3d4b47b4 | 360 | int on_list; |
41a2d6cf IM |
361 | struct list_head leaf_cfs_rq_list; |
362 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
363 | |
364 | #ifdef CONFIG_SMP | |
c09595f6 | 365 | /* |
c8cba857 | 366 | * the part of load.weight contributed by tasks |
c09595f6 | 367 | */ |
c8cba857 | 368 | unsigned long task_weight; |
c09595f6 | 369 | |
c8cba857 PZ |
370 | /* |
371 | * h_load = weight * f(tg) | |
372 | * | |
373 | * Where f(tg) is the recursive weight fraction assigned to | |
374 | * this group. | |
375 | */ | |
376 | unsigned long h_load; | |
c09595f6 | 377 | |
c8cba857 | 378 | /* |
3b3d190e PT |
379 | * Maintaining per-cpu shares distribution for group scheduling |
380 | * | |
381 | * load_stamp is the last time we updated the load average | |
382 | * load_last is the last time we updated the load average and saw load | |
383 | * load_unacc_exec_time is currently unaccounted execution time | |
c8cba857 | 384 | */ |
2069dd75 PZ |
385 | u64 load_avg; |
386 | u64 load_period; | |
3b3d190e | 387 | u64 load_stamp, load_last, load_unacc_exec_time; |
f1d239f7 | 388 | |
2069dd75 | 389 | unsigned long load_contribution; |
c09595f6 | 390 | #endif |
ab84d31e PT |
391 | #ifdef CONFIG_CFS_BANDWIDTH |
392 | int runtime_enabled; | |
393 | s64 runtime_remaining; | |
394 | #endif | |
6aa645ea IM |
395 | #endif |
396 | }; | |
1da177e4 | 397 | |
ab84d31e PT |
398 | #ifdef CONFIG_FAIR_GROUP_SCHED |
399 | #ifdef CONFIG_CFS_BANDWIDTH | |
400 | static inline struct cfs_bandwidth *tg_cfs_bandwidth(struct task_group *tg) | |
401 | { | |
402 | return &tg->cfs_bandwidth; | |
403 | } | |
404 | ||
405 | static inline u64 default_cfs_period(void); | |
406 | ||
407 | static void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b) | |
408 | { | |
409 | raw_spin_lock_init(&cfs_b->lock); | |
ec12cb7f | 410 | cfs_b->runtime = 0; |
ab84d31e PT |
411 | cfs_b->quota = RUNTIME_INF; |
412 | cfs_b->period = ns_to_ktime(default_cfs_period()); | |
413 | } | |
414 | ||
415 | static void init_cfs_rq_runtime(struct cfs_rq *cfs_rq) | |
416 | { | |
417 | cfs_rq->runtime_enabled = 0; | |
418 | } | |
419 | ||
420 | static void destroy_cfs_bandwidth(struct cfs_bandwidth *cfs_b) | |
421 | {} | |
422 | #else | |
423 | static void init_cfs_rq_runtime(struct cfs_rq *cfs_rq) {} | |
424 | static void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b) {} | |
425 | static void destroy_cfs_bandwidth(struct cfs_bandwidth *cfs_b) {} | |
426 | ||
427 | static inline struct cfs_bandwidth *tg_cfs_bandwidth(struct task_group *tg) | |
428 | { | |
429 | return NULL; | |
430 | } | |
431 | #endif /* CONFIG_CFS_BANDWIDTH */ | |
432 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
433 | ||
6aa645ea IM |
434 | /* Real-Time classes' related field in a runqueue: */ |
435 | struct rt_rq { | |
436 | struct rt_prio_array active; | |
63489e45 | 437 | unsigned long rt_nr_running; |
052f1dc7 | 438 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
439 | struct { |
440 | int curr; /* highest queued rt task prio */ | |
398a153b | 441 | #ifdef CONFIG_SMP |
e864c499 | 442 | int next; /* next highest */ |
398a153b | 443 | #endif |
e864c499 | 444 | } highest_prio; |
6f505b16 | 445 | #endif |
fa85ae24 | 446 | #ifdef CONFIG_SMP |
73fe6aae | 447 | unsigned long rt_nr_migratory; |
a1ba4d8b | 448 | unsigned long rt_nr_total; |
a22d7fc1 | 449 | int overloaded; |
917b627d | 450 | struct plist_head pushable_tasks; |
fa85ae24 | 451 | #endif |
6f505b16 | 452 | int rt_throttled; |
fa85ae24 | 453 | u64 rt_time; |
ac086bc2 | 454 | u64 rt_runtime; |
ea736ed5 | 455 | /* Nests inside the rq lock: */ |
0986b11b | 456 | raw_spinlock_t rt_runtime_lock; |
6f505b16 | 457 | |
052f1dc7 | 458 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
459 | unsigned long rt_nr_boosted; |
460 | ||
6f505b16 PZ |
461 | struct rq *rq; |
462 | struct list_head leaf_rt_rq_list; | |
463 | struct task_group *tg; | |
6f505b16 | 464 | #endif |
6aa645ea IM |
465 | }; |
466 | ||
57d885fe GH |
467 | #ifdef CONFIG_SMP |
468 | ||
469 | /* | |
470 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
471 | * variables. Each exclusive cpuset essentially defines an island domain by |
472 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
473 | * exclusive cpuset is created, we also create and attach a new root-domain |
474 | * object. | |
475 | * | |
57d885fe GH |
476 | */ |
477 | struct root_domain { | |
478 | atomic_t refcount; | |
26a148eb | 479 | atomic_t rto_count; |
dce840a0 | 480 | struct rcu_head rcu; |
c6c4927b RR |
481 | cpumask_var_t span; |
482 | cpumask_var_t online; | |
637f5085 | 483 | |
0eab9146 | 484 | /* |
637f5085 GH |
485 | * The "RT overload" flag: it gets set if a CPU has more than |
486 | * one runnable RT task. | |
487 | */ | |
c6c4927b | 488 | cpumask_var_t rto_mask; |
6e0534f2 | 489 | struct cpupri cpupri; |
57d885fe GH |
490 | }; |
491 | ||
dc938520 GH |
492 | /* |
493 | * By default the system creates a single root-domain with all cpus as | |
494 | * members (mimicking the global state we have today). | |
495 | */ | |
57d885fe GH |
496 | static struct root_domain def_root_domain; |
497 | ||
ed2d372c | 498 | #endif /* CONFIG_SMP */ |
57d885fe | 499 | |
1da177e4 LT |
500 | /* |
501 | * This is the main, per-CPU runqueue data structure. | |
502 | * | |
503 | * Locking rule: those places that want to lock multiple runqueues | |
504 | * (such as the load balancing or the thread migration code), lock | |
505 | * acquire operations must be ordered by ascending &runqueue. | |
506 | */ | |
70b97a7f | 507 | struct rq { |
d8016491 | 508 | /* runqueue lock: */ |
05fa785c | 509 | raw_spinlock_t lock; |
1da177e4 LT |
510 | |
511 | /* | |
512 | * nr_running and cpu_load should be in the same cacheline because | |
513 | * remote CPUs use both these fields when doing load calculation. | |
514 | */ | |
515 | unsigned long nr_running; | |
6aa645ea IM |
516 | #define CPU_LOAD_IDX_MAX 5 |
517 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
fdf3e95d | 518 | unsigned long last_load_update_tick; |
46cb4b7c | 519 | #ifdef CONFIG_NO_HZ |
39c0cbe2 | 520 | u64 nohz_stamp; |
83cd4fe2 | 521 | unsigned char nohz_balance_kick; |
46cb4b7c | 522 | #endif |
61eadef6 | 523 | int skip_clock_update; |
a64692a3 | 524 | |
d8016491 IM |
525 | /* capture load from *all* tasks on this cpu: */ |
526 | struct load_weight load; | |
6aa645ea IM |
527 | unsigned long nr_load_updates; |
528 | u64 nr_switches; | |
529 | ||
530 | struct cfs_rq cfs; | |
6f505b16 | 531 | struct rt_rq rt; |
6f505b16 | 532 | |
6aa645ea | 533 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
534 | /* list of leaf cfs_rq on this cpu: */ |
535 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
536 | #endif |
537 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 538 | struct list_head leaf_rt_rq_list; |
1da177e4 | 539 | #endif |
1da177e4 LT |
540 | |
541 | /* | |
542 | * This is part of a global counter where only the total sum | |
543 | * over all CPUs matters. A task can increase this counter on | |
544 | * one CPU and if it got migrated afterwards it may decrease | |
545 | * it on another CPU. Always updated under the runqueue lock: | |
546 | */ | |
547 | unsigned long nr_uninterruptible; | |
548 | ||
34f971f6 | 549 | struct task_struct *curr, *idle, *stop; |
c9819f45 | 550 | unsigned long next_balance; |
1da177e4 | 551 | struct mm_struct *prev_mm; |
6aa645ea | 552 | |
3e51f33f | 553 | u64 clock; |
305e6835 | 554 | u64 clock_task; |
6aa645ea | 555 | |
1da177e4 LT |
556 | atomic_t nr_iowait; |
557 | ||
558 | #ifdef CONFIG_SMP | |
0eab9146 | 559 | struct root_domain *rd; |
1da177e4 LT |
560 | struct sched_domain *sd; |
561 | ||
e51fd5e2 PZ |
562 | unsigned long cpu_power; |
563 | ||
a0a522ce | 564 | unsigned char idle_at_tick; |
1da177e4 | 565 | /* For active balancing */ |
3f029d3c | 566 | int post_schedule; |
1da177e4 LT |
567 | int active_balance; |
568 | int push_cpu; | |
969c7921 | 569 | struct cpu_stop_work active_balance_work; |
d8016491 IM |
570 | /* cpu of this runqueue: */ |
571 | int cpu; | |
1f11eb6a | 572 | int online; |
1da177e4 | 573 | |
e9e9250b PZ |
574 | u64 rt_avg; |
575 | u64 age_stamp; | |
1b9508f6 MG |
576 | u64 idle_stamp; |
577 | u64 avg_idle; | |
1da177e4 LT |
578 | #endif |
579 | ||
aa483808 VP |
580 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
581 | u64 prev_irq_time; | |
582 | #endif | |
e6e6685a GC |
583 | #ifdef CONFIG_PARAVIRT |
584 | u64 prev_steal_time; | |
585 | #endif | |
095c0aa8 GC |
586 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING |
587 | u64 prev_steal_time_rq; | |
588 | #endif | |
aa483808 | 589 | |
dce48a84 TG |
590 | /* calc_load related fields */ |
591 | unsigned long calc_load_update; | |
592 | long calc_load_active; | |
593 | ||
8f4d37ec | 594 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
595 | #ifdef CONFIG_SMP |
596 | int hrtick_csd_pending; | |
597 | struct call_single_data hrtick_csd; | |
598 | #endif | |
8f4d37ec PZ |
599 | struct hrtimer hrtick_timer; |
600 | #endif | |
601 | ||
1da177e4 LT |
602 | #ifdef CONFIG_SCHEDSTATS |
603 | /* latency stats */ | |
604 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
605 | unsigned long long rq_cpu_time; |
606 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
607 | |
608 | /* sys_sched_yield() stats */ | |
480b9434 | 609 | unsigned int yld_count; |
1da177e4 LT |
610 | |
611 | /* schedule() stats */ | |
480b9434 KC |
612 | unsigned int sched_switch; |
613 | unsigned int sched_count; | |
614 | unsigned int sched_goidle; | |
1da177e4 LT |
615 | |
616 | /* try_to_wake_up() stats */ | |
480b9434 KC |
617 | unsigned int ttwu_count; |
618 | unsigned int ttwu_local; | |
1da177e4 | 619 | #endif |
317f3941 PZ |
620 | |
621 | #ifdef CONFIG_SMP | |
622 | struct task_struct *wake_list; | |
623 | #endif | |
1da177e4 LT |
624 | }; |
625 | ||
f34e3b61 | 626 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 627 | |
a64692a3 | 628 | |
1e5a7405 | 629 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); |
dd41f596 | 630 | |
0a2966b4 CL |
631 | static inline int cpu_of(struct rq *rq) |
632 | { | |
633 | #ifdef CONFIG_SMP | |
634 | return rq->cpu; | |
635 | #else | |
636 | return 0; | |
637 | #endif | |
638 | } | |
639 | ||
497f0ab3 | 640 | #define rcu_dereference_check_sched_domain(p) \ |
d11c563d | 641 | rcu_dereference_check((p), \ |
d11c563d PM |
642 | lockdep_is_held(&sched_domains_mutex)) |
643 | ||
674311d5 NP |
644 | /* |
645 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 646 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
647 | * |
648 | * The domain tree of any CPU may only be accessed from within | |
649 | * preempt-disabled sections. | |
650 | */ | |
48f24c4d | 651 | #define for_each_domain(cpu, __sd) \ |
497f0ab3 | 652 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) |
1da177e4 LT |
653 | |
654 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
655 | #define this_rq() (&__get_cpu_var(runqueues)) | |
656 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
657 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 658 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 659 | |
dc61b1d6 PZ |
660 | #ifdef CONFIG_CGROUP_SCHED |
661 | ||
662 | /* | |
663 | * Return the group to which this tasks belongs. | |
664 | * | |
6c6c54e1 PZ |
665 | * We use task_subsys_state_check() and extend the RCU verification with |
666 | * pi->lock and rq->lock because cpu_cgroup_attach() holds those locks for each | |
667 | * task it moves into the cgroup. Therefore by holding either of those locks, | |
668 | * we pin the task to the current cgroup. | |
dc61b1d6 PZ |
669 | */ |
670 | static inline struct task_group *task_group(struct task_struct *p) | |
671 | { | |
5091faa4 | 672 | struct task_group *tg; |
dc61b1d6 PZ |
673 | struct cgroup_subsys_state *css; |
674 | ||
675 | css = task_subsys_state_check(p, cpu_cgroup_subsys_id, | |
6c6c54e1 PZ |
676 | lockdep_is_held(&p->pi_lock) || |
677 | lockdep_is_held(&task_rq(p)->lock)); | |
5091faa4 MG |
678 | tg = container_of(css, struct task_group, css); |
679 | ||
680 | return autogroup_task_group(p, tg); | |
dc61b1d6 PZ |
681 | } |
682 | ||
683 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
684 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
685 | { | |
686 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
687 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; | |
688 | p->se.parent = task_group(p)->se[cpu]; | |
689 | #endif | |
690 | ||
691 | #ifdef CONFIG_RT_GROUP_SCHED | |
692 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; | |
693 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
694 | #endif | |
695 | } | |
696 | ||
697 | #else /* CONFIG_CGROUP_SCHED */ | |
698 | ||
699 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
700 | static inline struct task_group *task_group(struct task_struct *p) | |
701 | { | |
702 | return NULL; | |
703 | } | |
704 | ||
705 | #endif /* CONFIG_CGROUP_SCHED */ | |
706 | ||
fe44d621 | 707 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 708 | |
fe44d621 | 709 | static void update_rq_clock(struct rq *rq) |
3e51f33f | 710 | { |
fe44d621 | 711 | s64 delta; |
305e6835 | 712 | |
61eadef6 | 713 | if (rq->skip_clock_update > 0) |
f26f9aff | 714 | return; |
aa483808 | 715 | |
fe44d621 PZ |
716 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
717 | rq->clock += delta; | |
718 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
719 | } |
720 | ||
bf5c91ba IM |
721 | /* |
722 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
723 | */ | |
724 | #ifdef CONFIG_SCHED_DEBUG | |
725 | # define const_debug __read_mostly | |
726 | #else | |
727 | # define const_debug static const | |
728 | #endif | |
729 | ||
017730c1 | 730 | /** |
1fd06bb1 | 731 | * runqueue_is_locked - Returns true if the current cpu runqueue is locked |
e17b38bf | 732 | * @cpu: the processor in question. |
017730c1 | 733 | * |
017730c1 IM |
734 | * This interface allows printk to be called with the runqueue lock |
735 | * held and know whether or not it is OK to wake up the klogd. | |
736 | */ | |
89f19f04 | 737 | int runqueue_is_locked(int cpu) |
017730c1 | 738 | { |
05fa785c | 739 | return raw_spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
740 | } |
741 | ||
bf5c91ba IM |
742 | /* |
743 | * Debugging: various feature bits | |
744 | */ | |
f00b45c1 PZ |
745 | |
746 | #define SCHED_FEAT(name, enabled) \ | |
747 | __SCHED_FEAT_##name , | |
748 | ||
bf5c91ba | 749 | enum { |
f00b45c1 | 750 | #include "sched_features.h" |
bf5c91ba IM |
751 | }; |
752 | ||
f00b45c1 PZ |
753 | #undef SCHED_FEAT |
754 | ||
755 | #define SCHED_FEAT(name, enabled) \ | |
756 | (1UL << __SCHED_FEAT_##name) * enabled | | |
757 | ||
bf5c91ba | 758 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
759 | #include "sched_features.h" |
760 | 0; | |
761 | ||
762 | #undef SCHED_FEAT | |
763 | ||
764 | #ifdef CONFIG_SCHED_DEBUG | |
765 | #define SCHED_FEAT(name, enabled) \ | |
766 | #name , | |
767 | ||
983ed7a6 | 768 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
769 | #include "sched_features.h" |
770 | NULL | |
771 | }; | |
772 | ||
773 | #undef SCHED_FEAT | |
774 | ||
34f3a814 | 775 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 776 | { |
f00b45c1 PZ |
777 | int i; |
778 | ||
779 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
780 | if (!(sysctl_sched_features & (1UL << i))) |
781 | seq_puts(m, "NO_"); | |
782 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 783 | } |
34f3a814 | 784 | seq_puts(m, "\n"); |
f00b45c1 | 785 | |
34f3a814 | 786 | return 0; |
f00b45c1 PZ |
787 | } |
788 | ||
789 | static ssize_t | |
790 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
791 | size_t cnt, loff_t *ppos) | |
792 | { | |
793 | char buf[64]; | |
7740191c | 794 | char *cmp; |
f00b45c1 PZ |
795 | int neg = 0; |
796 | int i; | |
797 | ||
798 | if (cnt > 63) | |
799 | cnt = 63; | |
800 | ||
801 | if (copy_from_user(&buf, ubuf, cnt)) | |
802 | return -EFAULT; | |
803 | ||
804 | buf[cnt] = 0; | |
7740191c | 805 | cmp = strstrip(buf); |
f00b45c1 | 806 | |
524429c3 | 807 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
808 | neg = 1; |
809 | cmp += 3; | |
810 | } | |
811 | ||
812 | for (i = 0; sched_feat_names[i]; i++) { | |
7740191c | 813 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f00b45c1 PZ |
814 | if (neg) |
815 | sysctl_sched_features &= ~(1UL << i); | |
816 | else | |
817 | sysctl_sched_features |= (1UL << i); | |
818 | break; | |
819 | } | |
820 | } | |
821 | ||
822 | if (!sched_feat_names[i]) | |
823 | return -EINVAL; | |
824 | ||
42994724 | 825 | *ppos += cnt; |
f00b45c1 PZ |
826 | |
827 | return cnt; | |
828 | } | |
829 | ||
34f3a814 LZ |
830 | static int sched_feat_open(struct inode *inode, struct file *filp) |
831 | { | |
832 | return single_open(filp, sched_feat_show, NULL); | |
833 | } | |
834 | ||
828c0950 | 835 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
836 | .open = sched_feat_open, |
837 | .write = sched_feat_write, | |
838 | .read = seq_read, | |
839 | .llseek = seq_lseek, | |
840 | .release = single_release, | |
f00b45c1 PZ |
841 | }; |
842 | ||
843 | static __init int sched_init_debug(void) | |
844 | { | |
f00b45c1 PZ |
845 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
846 | &sched_feat_fops); | |
847 | ||
848 | return 0; | |
849 | } | |
850 | late_initcall(sched_init_debug); | |
851 | ||
852 | #endif | |
853 | ||
854 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 855 | |
b82d9fdd PZ |
856 | /* |
857 | * Number of tasks to iterate in a single balance run. | |
858 | * Limited because this is done with IRQs disabled. | |
859 | */ | |
860 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
861 | ||
e9e9250b PZ |
862 | /* |
863 | * period over which we average the RT time consumption, measured | |
864 | * in ms. | |
865 | * | |
866 | * default: 1s | |
867 | */ | |
868 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
869 | ||
fa85ae24 | 870 | /* |
9f0c1e56 | 871 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
872 | * default: 1s |
873 | */ | |
9f0c1e56 | 874 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 875 | |
6892b75e IM |
876 | static __read_mostly int scheduler_running; |
877 | ||
9f0c1e56 PZ |
878 | /* |
879 | * part of the period that we allow rt tasks to run in us. | |
880 | * default: 0.95s | |
881 | */ | |
882 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 883 | |
d0b27fa7 PZ |
884 | static inline u64 global_rt_period(void) |
885 | { | |
886 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
887 | } | |
888 | ||
889 | static inline u64 global_rt_runtime(void) | |
890 | { | |
e26873bb | 891 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
892 | return RUNTIME_INF; |
893 | ||
894 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
895 | } | |
fa85ae24 | 896 | |
1da177e4 | 897 | #ifndef prepare_arch_switch |
4866cde0 NP |
898 | # define prepare_arch_switch(next) do { } while (0) |
899 | #endif | |
900 | #ifndef finish_arch_switch | |
901 | # define finish_arch_switch(prev) do { } while (0) | |
902 | #endif | |
903 | ||
051a1d1a DA |
904 | static inline int task_current(struct rq *rq, struct task_struct *p) |
905 | { | |
906 | return rq->curr == p; | |
907 | } | |
908 | ||
70b97a7f | 909 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 910 | { |
3ca7a440 PZ |
911 | #ifdef CONFIG_SMP |
912 | return p->on_cpu; | |
913 | #else | |
051a1d1a | 914 | return task_current(rq, p); |
3ca7a440 | 915 | #endif |
4866cde0 NP |
916 | } |
917 | ||
3ca7a440 | 918 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 919 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 | 920 | { |
3ca7a440 PZ |
921 | #ifdef CONFIG_SMP |
922 | /* | |
923 | * We can optimise this out completely for !SMP, because the | |
924 | * SMP rebalancing from interrupt is the only thing that cares | |
925 | * here. | |
926 | */ | |
927 | next->on_cpu = 1; | |
928 | #endif | |
4866cde0 NP |
929 | } |
930 | ||
70b97a7f | 931 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 932 | { |
3ca7a440 PZ |
933 | #ifdef CONFIG_SMP |
934 | /* | |
935 | * After ->on_cpu is cleared, the task can be moved to a different CPU. | |
936 | * We must ensure this doesn't happen until the switch is completely | |
937 | * finished. | |
938 | */ | |
939 | smp_wmb(); | |
940 | prev->on_cpu = 0; | |
941 | #endif | |
da04c035 IM |
942 | #ifdef CONFIG_DEBUG_SPINLOCK |
943 | /* this is a valid case when another task releases the spinlock */ | |
944 | rq->lock.owner = current; | |
945 | #endif | |
8a25d5de IM |
946 | /* |
947 | * If we are tracking spinlock dependencies then we have to | |
948 | * fix up the runqueue lock - which gets 'carried over' from | |
949 | * prev into current: | |
950 | */ | |
951 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
952 | ||
05fa785c | 953 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 NP |
954 | } |
955 | ||
956 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 957 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
958 | { |
959 | #ifdef CONFIG_SMP | |
960 | /* | |
961 | * We can optimise this out completely for !SMP, because the | |
962 | * SMP rebalancing from interrupt is the only thing that cares | |
963 | * here. | |
964 | */ | |
3ca7a440 | 965 | next->on_cpu = 1; |
4866cde0 NP |
966 | #endif |
967 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
05fa785c | 968 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 | 969 | #else |
05fa785c | 970 | raw_spin_unlock(&rq->lock); |
4866cde0 NP |
971 | #endif |
972 | } | |
973 | ||
70b97a7f | 974 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
975 | { |
976 | #ifdef CONFIG_SMP | |
977 | /* | |
3ca7a440 | 978 | * After ->on_cpu is cleared, the task can be moved to a different CPU. |
4866cde0 NP |
979 | * We must ensure this doesn't happen until the switch is completely |
980 | * finished. | |
981 | */ | |
982 | smp_wmb(); | |
3ca7a440 | 983 | prev->on_cpu = 0; |
4866cde0 NP |
984 | #endif |
985 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
986 | local_irq_enable(); | |
1da177e4 | 987 | #endif |
4866cde0 NP |
988 | } |
989 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 990 | |
0970d299 | 991 | /* |
0122ec5b | 992 | * __task_rq_lock - lock the rq @p resides on. |
b29739f9 | 993 | */ |
70b97a7f | 994 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
995 | __acquires(rq->lock) |
996 | { | |
0970d299 PZ |
997 | struct rq *rq; |
998 | ||
0122ec5b PZ |
999 | lockdep_assert_held(&p->pi_lock); |
1000 | ||
3a5c359a | 1001 | for (;;) { |
0970d299 | 1002 | rq = task_rq(p); |
05fa785c | 1003 | raw_spin_lock(&rq->lock); |
65cc8e48 | 1004 | if (likely(rq == task_rq(p))) |
3a5c359a | 1005 | return rq; |
05fa785c | 1006 | raw_spin_unlock(&rq->lock); |
b29739f9 | 1007 | } |
b29739f9 IM |
1008 | } |
1009 | ||
1da177e4 | 1010 | /* |
0122ec5b | 1011 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. |
1da177e4 | 1012 | */ |
70b97a7f | 1013 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
0122ec5b | 1014 | __acquires(p->pi_lock) |
1da177e4 LT |
1015 | __acquires(rq->lock) |
1016 | { | |
70b97a7f | 1017 | struct rq *rq; |
1da177e4 | 1018 | |
3a5c359a | 1019 | for (;;) { |
0122ec5b | 1020 | raw_spin_lock_irqsave(&p->pi_lock, *flags); |
3a5c359a | 1021 | rq = task_rq(p); |
05fa785c | 1022 | raw_spin_lock(&rq->lock); |
65cc8e48 | 1023 | if (likely(rq == task_rq(p))) |
3a5c359a | 1024 | return rq; |
0122ec5b PZ |
1025 | raw_spin_unlock(&rq->lock); |
1026 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 | 1027 | } |
1da177e4 LT |
1028 | } |
1029 | ||
a9957449 | 1030 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
1031 | __releases(rq->lock) |
1032 | { | |
05fa785c | 1033 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
1034 | } |
1035 | ||
0122ec5b PZ |
1036 | static inline void |
1037 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | |
1da177e4 | 1038 | __releases(rq->lock) |
0122ec5b | 1039 | __releases(p->pi_lock) |
1da177e4 | 1040 | { |
0122ec5b PZ |
1041 | raw_spin_unlock(&rq->lock); |
1042 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 LT |
1043 | } |
1044 | ||
1da177e4 | 1045 | /* |
cc2a73b5 | 1046 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 1047 | */ |
a9957449 | 1048 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
1049 | __acquires(rq->lock) |
1050 | { | |
70b97a7f | 1051 | struct rq *rq; |
1da177e4 LT |
1052 | |
1053 | local_irq_disable(); | |
1054 | rq = this_rq(); | |
05fa785c | 1055 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
1056 | |
1057 | return rq; | |
1058 | } | |
1059 | ||
8f4d37ec PZ |
1060 | #ifdef CONFIG_SCHED_HRTICK |
1061 | /* | |
1062 | * Use HR-timers to deliver accurate preemption points. | |
1063 | * | |
1064 | * Its all a bit involved since we cannot program an hrt while holding the | |
1065 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1066 | * reschedule event. | |
1067 | * | |
1068 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1069 | * rq->lock. | |
1070 | */ | |
8f4d37ec PZ |
1071 | |
1072 | /* | |
1073 | * Use hrtick when: | |
1074 | * - enabled by features | |
1075 | * - hrtimer is actually high res | |
1076 | */ | |
1077 | static inline int hrtick_enabled(struct rq *rq) | |
1078 | { | |
1079 | if (!sched_feat(HRTICK)) | |
1080 | return 0; | |
ba42059f | 1081 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1082 | return 0; |
8f4d37ec PZ |
1083 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1084 | } | |
1085 | ||
8f4d37ec PZ |
1086 | static void hrtick_clear(struct rq *rq) |
1087 | { | |
1088 | if (hrtimer_active(&rq->hrtick_timer)) | |
1089 | hrtimer_cancel(&rq->hrtick_timer); | |
1090 | } | |
1091 | ||
8f4d37ec PZ |
1092 | /* |
1093 | * High-resolution timer tick. | |
1094 | * Runs from hardirq context with interrupts disabled. | |
1095 | */ | |
1096 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1097 | { | |
1098 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1099 | ||
1100 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1101 | ||
05fa785c | 1102 | raw_spin_lock(&rq->lock); |
3e51f33f | 1103 | update_rq_clock(rq); |
8f4d37ec | 1104 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 1105 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
1106 | |
1107 | return HRTIMER_NORESTART; | |
1108 | } | |
1109 | ||
95e904c7 | 1110 | #ifdef CONFIG_SMP |
31656519 PZ |
1111 | /* |
1112 | * called from hardirq (IPI) context | |
1113 | */ | |
1114 | static void __hrtick_start(void *arg) | |
b328ca18 | 1115 | { |
31656519 | 1116 | struct rq *rq = arg; |
b328ca18 | 1117 | |
05fa785c | 1118 | raw_spin_lock(&rq->lock); |
31656519 PZ |
1119 | hrtimer_restart(&rq->hrtick_timer); |
1120 | rq->hrtick_csd_pending = 0; | |
05fa785c | 1121 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
1122 | } |
1123 | ||
31656519 PZ |
1124 | /* |
1125 | * Called to set the hrtick timer state. | |
1126 | * | |
1127 | * called with rq->lock held and irqs disabled | |
1128 | */ | |
1129 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1130 | { |
31656519 PZ |
1131 | struct hrtimer *timer = &rq->hrtick_timer; |
1132 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1133 | |
cc584b21 | 1134 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1135 | |
1136 | if (rq == this_rq()) { | |
1137 | hrtimer_restart(timer); | |
1138 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1139 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1140 | rq->hrtick_csd_pending = 1; |
1141 | } | |
b328ca18 PZ |
1142 | } |
1143 | ||
1144 | static int | |
1145 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1146 | { | |
1147 | int cpu = (int)(long)hcpu; | |
1148 | ||
1149 | switch (action) { | |
1150 | case CPU_UP_CANCELED: | |
1151 | case CPU_UP_CANCELED_FROZEN: | |
1152 | case CPU_DOWN_PREPARE: | |
1153 | case CPU_DOWN_PREPARE_FROZEN: | |
1154 | case CPU_DEAD: | |
1155 | case CPU_DEAD_FROZEN: | |
31656519 | 1156 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1157 | return NOTIFY_OK; |
1158 | } | |
1159 | ||
1160 | return NOTIFY_DONE; | |
1161 | } | |
1162 | ||
fa748203 | 1163 | static __init void init_hrtick(void) |
b328ca18 PZ |
1164 | { |
1165 | hotcpu_notifier(hotplug_hrtick, 0); | |
1166 | } | |
31656519 PZ |
1167 | #else |
1168 | /* | |
1169 | * Called to set the hrtick timer state. | |
1170 | * | |
1171 | * called with rq->lock held and irqs disabled | |
1172 | */ | |
1173 | static void hrtick_start(struct rq *rq, u64 delay) | |
1174 | { | |
7f1e2ca9 | 1175 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1176 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1177 | } |
b328ca18 | 1178 | |
006c75f1 | 1179 | static inline void init_hrtick(void) |
8f4d37ec | 1180 | { |
8f4d37ec | 1181 | } |
31656519 | 1182 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1183 | |
31656519 | 1184 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1185 | { |
31656519 PZ |
1186 | #ifdef CONFIG_SMP |
1187 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1188 | |
31656519 PZ |
1189 | rq->hrtick_csd.flags = 0; |
1190 | rq->hrtick_csd.func = __hrtick_start; | |
1191 | rq->hrtick_csd.info = rq; | |
1192 | #endif | |
8f4d37ec | 1193 | |
31656519 PZ |
1194 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1195 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1196 | } |
006c75f1 | 1197 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1198 | static inline void hrtick_clear(struct rq *rq) |
1199 | { | |
1200 | } | |
1201 | ||
8f4d37ec PZ |
1202 | static inline void init_rq_hrtick(struct rq *rq) |
1203 | { | |
1204 | } | |
1205 | ||
b328ca18 PZ |
1206 | static inline void init_hrtick(void) |
1207 | { | |
1208 | } | |
006c75f1 | 1209 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1210 | |
c24d20db IM |
1211 | /* |
1212 | * resched_task - mark a task 'to be rescheduled now'. | |
1213 | * | |
1214 | * On UP this means the setting of the need_resched flag, on SMP it | |
1215 | * might also involve a cross-CPU call to trigger the scheduler on | |
1216 | * the target CPU. | |
1217 | */ | |
1218 | #ifdef CONFIG_SMP | |
1219 | ||
1220 | #ifndef tsk_is_polling | |
1221 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1222 | #endif | |
1223 | ||
31656519 | 1224 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1225 | { |
1226 | int cpu; | |
1227 | ||
05fa785c | 1228 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 1229 | |
5ed0cec0 | 1230 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1231 | return; |
1232 | ||
5ed0cec0 | 1233 | set_tsk_need_resched(p); |
c24d20db IM |
1234 | |
1235 | cpu = task_cpu(p); | |
1236 | if (cpu == smp_processor_id()) | |
1237 | return; | |
1238 | ||
1239 | /* NEED_RESCHED must be visible before we test polling */ | |
1240 | smp_mb(); | |
1241 | if (!tsk_is_polling(p)) | |
1242 | smp_send_reschedule(cpu); | |
1243 | } | |
1244 | ||
1245 | static void resched_cpu(int cpu) | |
1246 | { | |
1247 | struct rq *rq = cpu_rq(cpu); | |
1248 | unsigned long flags; | |
1249 | ||
05fa785c | 1250 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
1251 | return; |
1252 | resched_task(cpu_curr(cpu)); | |
05fa785c | 1253 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 1254 | } |
06d8308c TG |
1255 | |
1256 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
1257 | /* |
1258 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
1259 | * from an idle cpu. This is good for power-savings. | |
1260 | * | |
1261 | * We don't do similar optimization for completely idle system, as | |
1262 | * selecting an idle cpu will add more delays to the timers than intended | |
1263 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
1264 | */ | |
1265 | int get_nohz_timer_target(void) | |
1266 | { | |
1267 | int cpu = smp_processor_id(); | |
1268 | int i; | |
1269 | struct sched_domain *sd; | |
1270 | ||
057f3fad | 1271 | rcu_read_lock(); |
83cd4fe2 | 1272 | for_each_domain(cpu, sd) { |
057f3fad PZ |
1273 | for_each_cpu(i, sched_domain_span(sd)) { |
1274 | if (!idle_cpu(i)) { | |
1275 | cpu = i; | |
1276 | goto unlock; | |
1277 | } | |
1278 | } | |
83cd4fe2 | 1279 | } |
057f3fad PZ |
1280 | unlock: |
1281 | rcu_read_unlock(); | |
83cd4fe2 VP |
1282 | return cpu; |
1283 | } | |
06d8308c TG |
1284 | /* |
1285 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1286 | * idle CPU then this timer might expire before the next timer event | |
1287 | * which is scheduled to wake up that CPU. In case of a completely | |
1288 | * idle system the next event might even be infinite time into the | |
1289 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1290 | * leaves the inner idle loop so the newly added timer is taken into | |
1291 | * account when the CPU goes back to idle and evaluates the timer | |
1292 | * wheel for the next timer event. | |
1293 | */ | |
1294 | void wake_up_idle_cpu(int cpu) | |
1295 | { | |
1296 | struct rq *rq = cpu_rq(cpu); | |
1297 | ||
1298 | if (cpu == smp_processor_id()) | |
1299 | return; | |
1300 | ||
1301 | /* | |
1302 | * This is safe, as this function is called with the timer | |
1303 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1304 | * to idle and has not yet set rq->curr to idle then it will | |
1305 | * be serialized on the timer wheel base lock and take the new | |
1306 | * timer into account automatically. | |
1307 | */ | |
1308 | if (rq->curr != rq->idle) | |
1309 | return; | |
1310 | ||
1311 | /* | |
1312 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1313 | * lockless. The worst case is that the other CPU runs the | |
1314 | * idle task through an additional NOOP schedule() | |
1315 | */ | |
5ed0cec0 | 1316 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1317 | |
1318 | /* NEED_RESCHED must be visible before we test polling */ | |
1319 | smp_mb(); | |
1320 | if (!tsk_is_polling(rq->idle)) | |
1321 | smp_send_reschedule(cpu); | |
1322 | } | |
39c0cbe2 | 1323 | |
6d6bc0ad | 1324 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1325 | |
e9e9250b PZ |
1326 | static u64 sched_avg_period(void) |
1327 | { | |
1328 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1329 | } | |
1330 | ||
1331 | static void sched_avg_update(struct rq *rq) | |
1332 | { | |
1333 | s64 period = sched_avg_period(); | |
1334 | ||
1335 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
1336 | /* |
1337 | * Inline assembly required to prevent the compiler | |
1338 | * optimising this loop into a divmod call. | |
1339 | * See __iter_div_u64_rem() for another example of this. | |
1340 | */ | |
1341 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
1342 | rq->age_stamp += period; |
1343 | rq->rt_avg /= 2; | |
1344 | } | |
1345 | } | |
1346 | ||
1347 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1348 | { | |
1349 | rq->rt_avg += rt_delta; | |
1350 | sched_avg_update(rq); | |
1351 | } | |
1352 | ||
6d6bc0ad | 1353 | #else /* !CONFIG_SMP */ |
31656519 | 1354 | static void resched_task(struct task_struct *p) |
c24d20db | 1355 | { |
05fa785c | 1356 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 1357 | set_tsk_need_resched(p); |
c24d20db | 1358 | } |
e9e9250b PZ |
1359 | |
1360 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1361 | { | |
1362 | } | |
da2b71ed SS |
1363 | |
1364 | static void sched_avg_update(struct rq *rq) | |
1365 | { | |
1366 | } | |
6d6bc0ad | 1367 | #endif /* CONFIG_SMP */ |
c24d20db | 1368 | |
45bf76df IM |
1369 | #if BITS_PER_LONG == 32 |
1370 | # define WMULT_CONST (~0UL) | |
1371 | #else | |
1372 | # define WMULT_CONST (1UL << 32) | |
1373 | #endif | |
1374 | ||
1375 | #define WMULT_SHIFT 32 | |
1376 | ||
194081eb IM |
1377 | /* |
1378 | * Shift right and round: | |
1379 | */ | |
cf2ab469 | 1380 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1381 | |
a7be37ac PZ |
1382 | /* |
1383 | * delta *= weight / lw | |
1384 | */ | |
cb1c4fc9 | 1385 | static unsigned long |
45bf76df IM |
1386 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1387 | struct load_weight *lw) | |
1388 | { | |
1389 | u64 tmp; | |
1390 | ||
c8b28116 NR |
1391 | /* |
1392 | * weight can be less than 2^SCHED_LOAD_RESOLUTION for task group sched | |
1393 | * entities since MIN_SHARES = 2. Treat weight as 1 if less than | |
1394 | * 2^SCHED_LOAD_RESOLUTION. | |
1395 | */ | |
1396 | if (likely(weight > (1UL << SCHED_LOAD_RESOLUTION))) | |
1397 | tmp = (u64)delta_exec * scale_load_down(weight); | |
1398 | else | |
1399 | tmp = (u64)delta_exec; | |
db670dac | 1400 | |
7a232e03 | 1401 | if (!lw->inv_weight) { |
c8b28116 NR |
1402 | unsigned long w = scale_load_down(lw->weight); |
1403 | ||
1404 | if (BITS_PER_LONG > 32 && unlikely(w >= WMULT_CONST)) | |
7a232e03 | 1405 | lw->inv_weight = 1; |
c8b28116 NR |
1406 | else if (unlikely(!w)) |
1407 | lw->inv_weight = WMULT_CONST; | |
7a232e03 | 1408 | else |
c8b28116 | 1409 | lw->inv_weight = WMULT_CONST / w; |
7a232e03 | 1410 | } |
45bf76df | 1411 | |
45bf76df IM |
1412 | /* |
1413 | * Check whether we'd overflow the 64-bit multiplication: | |
1414 | */ | |
194081eb | 1415 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1416 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1417 | WMULT_SHIFT/2); |
1418 | else | |
cf2ab469 | 1419 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1420 | |
ecf691da | 1421 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1422 | } |
1423 | ||
1091985b | 1424 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1425 | { |
1426 | lw->weight += inc; | |
e89996ae | 1427 | lw->inv_weight = 0; |
45bf76df IM |
1428 | } |
1429 | ||
1091985b | 1430 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1431 | { |
1432 | lw->weight -= dec; | |
e89996ae | 1433 | lw->inv_weight = 0; |
45bf76df IM |
1434 | } |
1435 | ||
2069dd75 PZ |
1436 | static inline void update_load_set(struct load_weight *lw, unsigned long w) |
1437 | { | |
1438 | lw->weight = w; | |
1439 | lw->inv_weight = 0; | |
1440 | } | |
1441 | ||
2dd73a4f PW |
1442 | /* |
1443 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1444 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1445 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1446 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1447 | * scaled version of the new time slice allocation that they receive on time |
1448 | * slice expiry etc. | |
1449 | */ | |
1450 | ||
cce7ade8 PZ |
1451 | #define WEIGHT_IDLEPRIO 3 |
1452 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1453 | |
1454 | /* | |
1455 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1456 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1457 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1458 | * that remained on nice 0. | |
1459 | * | |
1460 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1461 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1462 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1463 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1464 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1465 | */ |
1466 | static const int prio_to_weight[40] = { | |
254753dc IM |
1467 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1468 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1469 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1470 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1471 | /* 0 */ 1024, 820, 655, 526, 423, | |
1472 | /* 5 */ 335, 272, 215, 172, 137, | |
1473 | /* 10 */ 110, 87, 70, 56, 45, | |
1474 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1475 | }; |
1476 | ||
5714d2de IM |
1477 | /* |
1478 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1479 | * | |
1480 | * In cases where the weight does not change often, we can use the | |
1481 | * precalculated inverse to speed up arithmetics by turning divisions | |
1482 | * into multiplications: | |
1483 | */ | |
dd41f596 | 1484 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1485 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1486 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1487 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1488 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1489 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1490 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1491 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1492 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1493 | }; |
2dd73a4f | 1494 | |
ef12fefa BR |
1495 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1496 | enum cpuacct_stat_index { | |
1497 | CPUACCT_STAT_USER, /* ... user mode */ | |
1498 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1499 | ||
1500 | CPUACCT_STAT_NSTATS, | |
1501 | }; | |
1502 | ||
d842de87 SV |
1503 | #ifdef CONFIG_CGROUP_CPUACCT |
1504 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1505 | static void cpuacct_update_stats(struct task_struct *tsk, |
1506 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1507 | #else |
1508 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1509 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1510 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1511 | #endif |
1512 | ||
18d95a28 PZ |
1513 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1514 | { | |
1515 | update_load_add(&rq->load, load); | |
1516 | } | |
1517 | ||
1518 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1519 | { | |
1520 | update_load_sub(&rq->load, load); | |
1521 | } | |
1522 | ||
a790de99 PT |
1523 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
1524 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
eb755805 | 1525 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1526 | |
1527 | /* | |
1528 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1529 | * leaving it for the final time. | |
1530 | */ | |
eb755805 | 1531 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1532 | { |
1533 | struct task_group *parent, *child; | |
eb755805 | 1534 | int ret; |
c09595f6 PZ |
1535 | |
1536 | rcu_read_lock(); | |
1537 | parent = &root_task_group; | |
1538 | down: | |
eb755805 PZ |
1539 | ret = (*down)(parent, data); |
1540 | if (ret) | |
1541 | goto out_unlock; | |
c09595f6 PZ |
1542 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1543 | parent = child; | |
1544 | goto down; | |
1545 | ||
1546 | up: | |
1547 | continue; | |
1548 | } | |
eb755805 PZ |
1549 | ret = (*up)(parent, data); |
1550 | if (ret) | |
1551 | goto out_unlock; | |
c09595f6 PZ |
1552 | |
1553 | child = parent; | |
1554 | parent = parent->parent; | |
1555 | if (parent) | |
1556 | goto up; | |
eb755805 | 1557 | out_unlock: |
c09595f6 | 1558 | rcu_read_unlock(); |
eb755805 PZ |
1559 | |
1560 | return ret; | |
c09595f6 PZ |
1561 | } |
1562 | ||
eb755805 PZ |
1563 | static int tg_nop(struct task_group *tg, void *data) |
1564 | { | |
1565 | return 0; | |
c09595f6 | 1566 | } |
eb755805 PZ |
1567 | #endif |
1568 | ||
1569 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1570 | /* Used instead of source_load when we know the type == 0 */ |
1571 | static unsigned long weighted_cpuload(const int cpu) | |
1572 | { | |
1573 | return cpu_rq(cpu)->load.weight; | |
1574 | } | |
1575 | ||
1576 | /* | |
1577 | * Return a low guess at the load of a migration-source cpu weighted | |
1578 | * according to the scheduling class and "nice" value. | |
1579 | * | |
1580 | * We want to under-estimate the load of migration sources, to | |
1581 | * balance conservatively. | |
1582 | */ | |
1583 | static unsigned long source_load(int cpu, int type) | |
1584 | { | |
1585 | struct rq *rq = cpu_rq(cpu); | |
1586 | unsigned long total = weighted_cpuload(cpu); | |
1587 | ||
1588 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1589 | return total; | |
1590 | ||
1591 | return min(rq->cpu_load[type-1], total); | |
1592 | } | |
1593 | ||
1594 | /* | |
1595 | * Return a high guess at the load of a migration-target cpu weighted | |
1596 | * according to the scheduling class and "nice" value. | |
1597 | */ | |
1598 | static unsigned long target_load(int cpu, int type) | |
1599 | { | |
1600 | struct rq *rq = cpu_rq(cpu); | |
1601 | unsigned long total = weighted_cpuload(cpu); | |
1602 | ||
1603 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1604 | return total; | |
1605 | ||
1606 | return max(rq->cpu_load[type-1], total); | |
1607 | } | |
1608 | ||
ae154be1 PZ |
1609 | static unsigned long power_of(int cpu) |
1610 | { | |
e51fd5e2 | 1611 | return cpu_rq(cpu)->cpu_power; |
ae154be1 PZ |
1612 | } |
1613 | ||
eb755805 PZ |
1614 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1615 | ||
1616 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1617 | { | |
1618 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1619 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1620 | |
4cd42620 | 1621 | if (nr_running) |
e2b245f8 | 1622 | return rq->load.weight / nr_running; |
eb755805 | 1623 | |
e2b245f8 | 1624 | return 0; |
eb755805 PZ |
1625 | } |
1626 | ||
8f45e2b5 GH |
1627 | #ifdef CONFIG_PREEMPT |
1628 | ||
b78bb868 PZ |
1629 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1630 | ||
70574a99 | 1631 | /* |
8f45e2b5 GH |
1632 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1633 | * way at the expense of forcing extra atomic operations in all | |
1634 | * invocations. This assures that the double_lock is acquired using the | |
1635 | * same underlying policy as the spinlock_t on this architecture, which | |
1636 | * reduces latency compared to the unfair variant below. However, it | |
1637 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1638 | */ |
8f45e2b5 GH |
1639 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1640 | __releases(this_rq->lock) | |
1641 | __acquires(busiest->lock) | |
1642 | __acquires(this_rq->lock) | |
1643 | { | |
05fa785c | 1644 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1645 | double_rq_lock(this_rq, busiest); |
1646 | ||
1647 | return 1; | |
1648 | } | |
1649 | ||
1650 | #else | |
1651 | /* | |
1652 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1653 | * latency by eliminating extra atomic operations when the locks are | |
1654 | * already in proper order on entry. This favors lower cpu-ids and will | |
1655 | * grant the double lock to lower cpus over higher ids under contention, | |
1656 | * regardless of entry order into the function. | |
1657 | */ | |
1658 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1659 | __releases(this_rq->lock) |
1660 | __acquires(busiest->lock) | |
1661 | __acquires(this_rq->lock) | |
1662 | { | |
1663 | int ret = 0; | |
1664 | ||
05fa785c | 1665 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1666 | if (busiest < this_rq) { |
05fa785c TG |
1667 | raw_spin_unlock(&this_rq->lock); |
1668 | raw_spin_lock(&busiest->lock); | |
1669 | raw_spin_lock_nested(&this_rq->lock, | |
1670 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1671 | ret = 1; |
1672 | } else | |
05fa785c TG |
1673 | raw_spin_lock_nested(&busiest->lock, |
1674 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1675 | } |
1676 | return ret; | |
1677 | } | |
1678 | ||
8f45e2b5 GH |
1679 | #endif /* CONFIG_PREEMPT */ |
1680 | ||
1681 | /* | |
1682 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1683 | */ | |
1684 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1685 | { | |
1686 | if (unlikely(!irqs_disabled())) { | |
1687 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1688 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1689 | BUG_ON(1); |
1690 | } | |
1691 | ||
1692 | return _double_lock_balance(this_rq, busiest); | |
1693 | } | |
1694 | ||
70574a99 AD |
1695 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1696 | __releases(busiest->lock) | |
1697 | { | |
05fa785c | 1698 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1699 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1700 | } | |
1e3c88bd PZ |
1701 | |
1702 | /* | |
1703 | * double_rq_lock - safely lock two runqueues | |
1704 | * | |
1705 | * Note this does not disable interrupts like task_rq_lock, | |
1706 | * you need to do so manually before calling. | |
1707 | */ | |
1708 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1709 | __acquires(rq1->lock) | |
1710 | __acquires(rq2->lock) | |
1711 | { | |
1712 | BUG_ON(!irqs_disabled()); | |
1713 | if (rq1 == rq2) { | |
1714 | raw_spin_lock(&rq1->lock); | |
1715 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1716 | } else { | |
1717 | if (rq1 < rq2) { | |
1718 | raw_spin_lock(&rq1->lock); | |
1719 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1720 | } else { | |
1721 | raw_spin_lock(&rq2->lock); | |
1722 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1723 | } | |
1724 | } | |
1e3c88bd PZ |
1725 | } |
1726 | ||
1727 | /* | |
1728 | * double_rq_unlock - safely unlock two runqueues | |
1729 | * | |
1730 | * Note this does not restore interrupts like task_rq_unlock, | |
1731 | * you need to do so manually after calling. | |
1732 | */ | |
1733 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1734 | __releases(rq1->lock) | |
1735 | __releases(rq2->lock) | |
1736 | { | |
1737 | raw_spin_unlock(&rq1->lock); | |
1738 | if (rq1 != rq2) | |
1739 | raw_spin_unlock(&rq2->lock); | |
1740 | else | |
1741 | __release(rq2->lock); | |
1742 | } | |
1743 | ||
d95f4122 MG |
1744 | #else /* CONFIG_SMP */ |
1745 | ||
1746 | /* | |
1747 | * double_rq_lock - safely lock two runqueues | |
1748 | * | |
1749 | * Note this does not disable interrupts like task_rq_lock, | |
1750 | * you need to do so manually before calling. | |
1751 | */ | |
1752 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1753 | __acquires(rq1->lock) | |
1754 | __acquires(rq2->lock) | |
1755 | { | |
1756 | BUG_ON(!irqs_disabled()); | |
1757 | BUG_ON(rq1 != rq2); | |
1758 | raw_spin_lock(&rq1->lock); | |
1759 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1760 | } | |
1761 | ||
1762 | /* | |
1763 | * double_rq_unlock - safely unlock two runqueues | |
1764 | * | |
1765 | * Note this does not restore interrupts like task_rq_unlock, | |
1766 | * you need to do so manually after calling. | |
1767 | */ | |
1768 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1769 | __releases(rq1->lock) | |
1770 | __releases(rq2->lock) | |
1771 | { | |
1772 | BUG_ON(rq1 != rq2); | |
1773 | raw_spin_unlock(&rq1->lock); | |
1774 | __release(rq2->lock); | |
1775 | } | |
1776 | ||
18d95a28 PZ |
1777 | #endif |
1778 | ||
74f5187a | 1779 | static void calc_load_account_idle(struct rq *this_rq); |
0bcdcf28 | 1780 | static void update_sysctl(void); |
acb4a848 | 1781 | static int get_update_sysctl_factor(void); |
fdf3e95d | 1782 | static void update_cpu_load(struct rq *this_rq); |
dce48a84 | 1783 | |
cd29fe6f PZ |
1784 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1785 | { | |
1786 | set_task_rq(p, cpu); | |
1787 | #ifdef CONFIG_SMP | |
1788 | /* | |
1789 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1790 | * successfuly executed on another CPU. We must ensure that updates of | |
1791 | * per-task data have been completed by this moment. | |
1792 | */ | |
1793 | smp_wmb(); | |
1794 | task_thread_info(p)->cpu = cpu; | |
1795 | #endif | |
1796 | } | |
dce48a84 | 1797 | |
1e3c88bd | 1798 | static const struct sched_class rt_sched_class; |
dd41f596 | 1799 | |
34f971f6 | 1800 | #define sched_class_highest (&stop_sched_class) |
1f11eb6a GH |
1801 | #define for_each_class(class) \ |
1802 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1803 | |
1e3c88bd PZ |
1804 | #include "sched_stats.h" |
1805 | ||
c09595f6 | 1806 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1807 | { |
1808 | rq->nr_running++; | |
9c217245 IM |
1809 | } |
1810 | ||
c09595f6 | 1811 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1812 | { |
1813 | rq->nr_running--; | |
9c217245 IM |
1814 | } |
1815 | ||
45bf76df IM |
1816 | static void set_load_weight(struct task_struct *p) |
1817 | { | |
f05998d4 NR |
1818 | int prio = p->static_prio - MAX_RT_PRIO; |
1819 | struct load_weight *load = &p->se.load; | |
1820 | ||
dd41f596 IM |
1821 | /* |
1822 | * SCHED_IDLE tasks get minimal weight: | |
1823 | */ | |
1824 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 1825 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 1826 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
1827 | return; |
1828 | } | |
71f8bd46 | 1829 | |
c8b28116 | 1830 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 1831 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
1832 | } |
1833 | ||
371fd7e7 | 1834 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1835 | { |
a64692a3 | 1836 | update_rq_clock(rq); |
dd41f596 | 1837 | sched_info_queued(p); |
371fd7e7 | 1838 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
1839 | } |
1840 | ||
371fd7e7 | 1841 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1842 | { |
a64692a3 | 1843 | update_rq_clock(rq); |
46ac22ba | 1844 | sched_info_dequeued(p); |
371fd7e7 | 1845 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
1846 | } |
1847 | ||
1e3c88bd PZ |
1848 | /* |
1849 | * activate_task - move a task to the runqueue. | |
1850 | */ | |
371fd7e7 | 1851 | static void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1852 | { |
1853 | if (task_contributes_to_load(p)) | |
1854 | rq->nr_uninterruptible--; | |
1855 | ||
371fd7e7 | 1856 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
1857 | } |
1858 | ||
1859 | /* | |
1860 | * deactivate_task - remove a task from the runqueue. | |
1861 | */ | |
371fd7e7 | 1862 | static void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1863 | { |
1864 | if (task_contributes_to_load(p)) | |
1865 | rq->nr_uninterruptible++; | |
1866 | ||
371fd7e7 | 1867 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1868 | } |
1869 | ||
b52bfee4 VP |
1870 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
1871 | ||
305e6835 VP |
1872 | /* |
1873 | * There are no locks covering percpu hardirq/softirq time. | |
1874 | * They are only modified in account_system_vtime, on corresponding CPU | |
1875 | * with interrupts disabled. So, writes are safe. | |
1876 | * They are read and saved off onto struct rq in update_rq_clock(). | |
1877 | * This may result in other CPU reading this CPU's irq time and can | |
1878 | * race with irq/account_system_vtime on this CPU. We would either get old | |
8e92c201 PZ |
1879 | * or new value with a side effect of accounting a slice of irq time to wrong |
1880 | * task when irq is in progress while we read rq->clock. That is a worthy | |
1881 | * compromise in place of having locks on each irq in account_system_time. | |
305e6835 | 1882 | */ |
b52bfee4 VP |
1883 | static DEFINE_PER_CPU(u64, cpu_hardirq_time); |
1884 | static DEFINE_PER_CPU(u64, cpu_softirq_time); | |
1885 | ||
1886 | static DEFINE_PER_CPU(u64, irq_start_time); | |
1887 | static int sched_clock_irqtime; | |
1888 | ||
1889 | void enable_sched_clock_irqtime(void) | |
1890 | { | |
1891 | sched_clock_irqtime = 1; | |
1892 | } | |
1893 | ||
1894 | void disable_sched_clock_irqtime(void) | |
1895 | { | |
1896 | sched_clock_irqtime = 0; | |
1897 | } | |
1898 | ||
8e92c201 PZ |
1899 | #ifndef CONFIG_64BIT |
1900 | static DEFINE_PER_CPU(seqcount_t, irq_time_seq); | |
1901 | ||
1902 | static inline void irq_time_write_begin(void) | |
1903 | { | |
1904 | __this_cpu_inc(irq_time_seq.sequence); | |
1905 | smp_wmb(); | |
1906 | } | |
1907 | ||
1908 | static inline void irq_time_write_end(void) | |
1909 | { | |
1910 | smp_wmb(); | |
1911 | __this_cpu_inc(irq_time_seq.sequence); | |
1912 | } | |
1913 | ||
1914 | static inline u64 irq_time_read(int cpu) | |
1915 | { | |
1916 | u64 irq_time; | |
1917 | unsigned seq; | |
1918 | ||
1919 | do { | |
1920 | seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); | |
1921 | irq_time = per_cpu(cpu_softirq_time, cpu) + | |
1922 | per_cpu(cpu_hardirq_time, cpu); | |
1923 | } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); | |
1924 | ||
1925 | return irq_time; | |
1926 | } | |
1927 | #else /* CONFIG_64BIT */ | |
1928 | static inline void irq_time_write_begin(void) | |
1929 | { | |
1930 | } | |
1931 | ||
1932 | static inline void irq_time_write_end(void) | |
1933 | { | |
1934 | } | |
1935 | ||
1936 | static inline u64 irq_time_read(int cpu) | |
305e6835 | 1937 | { |
305e6835 VP |
1938 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); |
1939 | } | |
8e92c201 | 1940 | #endif /* CONFIG_64BIT */ |
305e6835 | 1941 | |
fe44d621 PZ |
1942 | /* |
1943 | * Called before incrementing preempt_count on {soft,}irq_enter | |
1944 | * and before decrementing preempt_count on {soft,}irq_exit. | |
1945 | */ | |
b52bfee4 VP |
1946 | void account_system_vtime(struct task_struct *curr) |
1947 | { | |
1948 | unsigned long flags; | |
fe44d621 | 1949 | s64 delta; |
b52bfee4 | 1950 | int cpu; |
b52bfee4 VP |
1951 | |
1952 | if (!sched_clock_irqtime) | |
1953 | return; | |
1954 | ||
1955 | local_irq_save(flags); | |
1956 | ||
b52bfee4 | 1957 | cpu = smp_processor_id(); |
fe44d621 PZ |
1958 | delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time); |
1959 | __this_cpu_add(irq_start_time, delta); | |
1960 | ||
8e92c201 | 1961 | irq_time_write_begin(); |
b52bfee4 VP |
1962 | /* |
1963 | * We do not account for softirq time from ksoftirqd here. | |
1964 | * We want to continue accounting softirq time to ksoftirqd thread | |
1965 | * in that case, so as not to confuse scheduler with a special task | |
1966 | * that do not consume any time, but still wants to run. | |
1967 | */ | |
1968 | if (hardirq_count()) | |
fe44d621 | 1969 | __this_cpu_add(cpu_hardirq_time, delta); |
4dd53d89 | 1970 | else if (in_serving_softirq() && curr != this_cpu_ksoftirqd()) |
fe44d621 | 1971 | __this_cpu_add(cpu_softirq_time, delta); |
b52bfee4 | 1972 | |
8e92c201 | 1973 | irq_time_write_end(); |
b52bfee4 VP |
1974 | local_irq_restore(flags); |
1975 | } | |
b7dadc38 | 1976 | EXPORT_SYMBOL_GPL(account_system_vtime); |
b52bfee4 | 1977 | |
e6e6685a GC |
1978 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
1979 | ||
1980 | #ifdef CONFIG_PARAVIRT | |
1981 | static inline u64 steal_ticks(u64 steal) | |
aa483808 | 1982 | { |
e6e6685a GC |
1983 | if (unlikely(steal > NSEC_PER_SEC)) |
1984 | return div_u64(steal, TICK_NSEC); | |
fe44d621 | 1985 | |
e6e6685a GC |
1986 | return __iter_div_u64_rem(steal, TICK_NSEC, &steal); |
1987 | } | |
1988 | #endif | |
1989 | ||
fe44d621 | 1990 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 1991 | { |
095c0aa8 GC |
1992 | /* |
1993 | * In theory, the compile should just see 0 here, and optimize out the call | |
1994 | * to sched_rt_avg_update. But I don't trust it... | |
1995 | */ | |
1996 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
1997 | s64 steal = 0, irq_delta = 0; | |
1998 | #endif | |
1999 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 2000 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
2001 | |
2002 | /* | |
2003 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
2004 | * this case when a previous update_rq_clock() happened inside a | |
2005 | * {soft,}irq region. | |
2006 | * | |
2007 | * When this happens, we stop ->clock_task and only update the | |
2008 | * prev_irq_time stamp to account for the part that fit, so that a next | |
2009 | * update will consume the rest. This ensures ->clock_task is | |
2010 | * monotonic. | |
2011 | * | |
2012 | * It does however cause some slight miss-attribution of {soft,}irq | |
2013 | * time, a more accurate solution would be to update the irq_time using | |
2014 | * the current rq->clock timestamp, except that would require using | |
2015 | * atomic ops. | |
2016 | */ | |
2017 | if (irq_delta > delta) | |
2018 | irq_delta = delta; | |
2019 | ||
2020 | rq->prev_irq_time += irq_delta; | |
2021 | delta -= irq_delta; | |
095c0aa8 GC |
2022 | #endif |
2023 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
2024 | if (static_branch((¶virt_steal_rq_enabled))) { | |
2025 | u64 st; | |
2026 | ||
2027 | steal = paravirt_steal_clock(cpu_of(rq)); | |
2028 | steal -= rq->prev_steal_time_rq; | |
2029 | ||
2030 | if (unlikely(steal > delta)) | |
2031 | steal = delta; | |
2032 | ||
2033 | st = steal_ticks(steal); | |
2034 | steal = st * TICK_NSEC; | |
2035 | ||
2036 | rq->prev_steal_time_rq += steal; | |
2037 | ||
2038 | delta -= steal; | |
2039 | } | |
2040 | #endif | |
2041 | ||
fe44d621 PZ |
2042 | rq->clock_task += delta; |
2043 | ||
095c0aa8 GC |
2044 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
2045 | if ((irq_delta + steal) && sched_feat(NONTASK_POWER)) | |
2046 | sched_rt_avg_update(rq, irq_delta + steal); | |
2047 | #endif | |
aa483808 VP |
2048 | } |
2049 | ||
095c0aa8 | 2050 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
abb74cef VP |
2051 | static int irqtime_account_hi_update(void) |
2052 | { | |
2053 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
2054 | unsigned long flags; | |
2055 | u64 latest_ns; | |
2056 | int ret = 0; | |
2057 | ||
2058 | local_irq_save(flags); | |
2059 | latest_ns = this_cpu_read(cpu_hardirq_time); | |
2060 | if (cputime64_gt(nsecs_to_cputime64(latest_ns), cpustat->irq)) | |
2061 | ret = 1; | |
2062 | local_irq_restore(flags); | |
2063 | return ret; | |
2064 | } | |
2065 | ||
2066 | static int irqtime_account_si_update(void) | |
2067 | { | |
2068 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
2069 | unsigned long flags; | |
2070 | u64 latest_ns; | |
2071 | int ret = 0; | |
2072 | ||
2073 | local_irq_save(flags); | |
2074 | latest_ns = this_cpu_read(cpu_softirq_time); | |
2075 | if (cputime64_gt(nsecs_to_cputime64(latest_ns), cpustat->softirq)) | |
2076 | ret = 1; | |
2077 | local_irq_restore(flags); | |
2078 | return ret; | |
2079 | } | |
2080 | ||
fe44d621 | 2081 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
305e6835 | 2082 | |
abb74cef VP |
2083 | #define sched_clock_irqtime (0) |
2084 | ||
095c0aa8 | 2085 | #endif |
b52bfee4 | 2086 | |
1e3c88bd PZ |
2087 | #include "sched_idletask.c" |
2088 | #include "sched_fair.c" | |
2089 | #include "sched_rt.c" | |
5091faa4 | 2090 | #include "sched_autogroup.c" |
34f971f6 | 2091 | #include "sched_stoptask.c" |
1e3c88bd PZ |
2092 | #ifdef CONFIG_SCHED_DEBUG |
2093 | # include "sched_debug.c" | |
2094 | #endif | |
2095 | ||
34f971f6 PZ |
2096 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
2097 | { | |
2098 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
2099 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
2100 | ||
2101 | if (stop) { | |
2102 | /* | |
2103 | * Make it appear like a SCHED_FIFO task, its something | |
2104 | * userspace knows about and won't get confused about. | |
2105 | * | |
2106 | * Also, it will make PI more or less work without too | |
2107 | * much confusion -- but then, stop work should not | |
2108 | * rely on PI working anyway. | |
2109 | */ | |
2110 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
2111 | ||
2112 | stop->sched_class = &stop_sched_class; | |
2113 | } | |
2114 | ||
2115 | cpu_rq(cpu)->stop = stop; | |
2116 | ||
2117 | if (old_stop) { | |
2118 | /* | |
2119 | * Reset it back to a normal scheduling class so that | |
2120 | * it can die in pieces. | |
2121 | */ | |
2122 | old_stop->sched_class = &rt_sched_class; | |
2123 | } | |
2124 | } | |
2125 | ||
14531189 | 2126 | /* |
dd41f596 | 2127 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 2128 | */ |
14531189 IM |
2129 | static inline int __normal_prio(struct task_struct *p) |
2130 | { | |
dd41f596 | 2131 | return p->static_prio; |
14531189 IM |
2132 | } |
2133 | ||
b29739f9 IM |
2134 | /* |
2135 | * Calculate the expected normal priority: i.e. priority | |
2136 | * without taking RT-inheritance into account. Might be | |
2137 | * boosted by interactivity modifiers. Changes upon fork, | |
2138 | * setprio syscalls, and whenever the interactivity | |
2139 | * estimator recalculates. | |
2140 | */ | |
36c8b586 | 2141 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
2142 | { |
2143 | int prio; | |
2144 | ||
e05606d3 | 2145 | if (task_has_rt_policy(p)) |
b29739f9 IM |
2146 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
2147 | else | |
2148 | prio = __normal_prio(p); | |
2149 | return prio; | |
2150 | } | |
2151 | ||
2152 | /* | |
2153 | * Calculate the current priority, i.e. the priority | |
2154 | * taken into account by the scheduler. This value might | |
2155 | * be boosted by RT tasks, or might be boosted by | |
2156 | * interactivity modifiers. Will be RT if the task got | |
2157 | * RT-boosted. If not then it returns p->normal_prio. | |
2158 | */ | |
36c8b586 | 2159 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
2160 | { |
2161 | p->normal_prio = normal_prio(p); | |
2162 | /* | |
2163 | * If we are RT tasks or we were boosted to RT priority, | |
2164 | * keep the priority unchanged. Otherwise, update priority | |
2165 | * to the normal priority: | |
2166 | */ | |
2167 | if (!rt_prio(p->prio)) | |
2168 | return p->normal_prio; | |
2169 | return p->prio; | |
2170 | } | |
2171 | ||
1da177e4 LT |
2172 | /** |
2173 | * task_curr - is this task currently executing on a CPU? | |
2174 | * @p: the task in question. | |
2175 | */ | |
36c8b586 | 2176 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
2177 | { |
2178 | return cpu_curr(task_cpu(p)) == p; | |
2179 | } | |
2180 | ||
cb469845 SR |
2181 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
2182 | const struct sched_class *prev_class, | |
da7a735e | 2183 | int oldprio) |
cb469845 SR |
2184 | { |
2185 | if (prev_class != p->sched_class) { | |
2186 | if (prev_class->switched_from) | |
da7a735e PZ |
2187 | prev_class->switched_from(rq, p); |
2188 | p->sched_class->switched_to(rq, p); | |
2189 | } else if (oldprio != p->prio) | |
2190 | p->sched_class->prio_changed(rq, p, oldprio); | |
cb469845 SR |
2191 | } |
2192 | ||
1e5a7405 PZ |
2193 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
2194 | { | |
2195 | const struct sched_class *class; | |
2196 | ||
2197 | if (p->sched_class == rq->curr->sched_class) { | |
2198 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
2199 | } else { | |
2200 | for_each_class(class) { | |
2201 | if (class == rq->curr->sched_class) | |
2202 | break; | |
2203 | if (class == p->sched_class) { | |
2204 | resched_task(rq->curr); | |
2205 | break; | |
2206 | } | |
2207 | } | |
2208 | } | |
2209 | ||
2210 | /* | |
2211 | * A queue event has occurred, and we're going to schedule. In | |
2212 | * this case, we can save a useless back to back clock update. | |
2213 | */ | |
fd2f4419 | 2214 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
2215 | rq->skip_clock_update = 1; |
2216 | } | |
2217 | ||
1da177e4 | 2218 | #ifdef CONFIG_SMP |
cc367732 IM |
2219 | /* |
2220 | * Is this task likely cache-hot: | |
2221 | */ | |
e7693a36 | 2222 | static int |
cc367732 IM |
2223 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
2224 | { | |
2225 | s64 delta; | |
2226 | ||
e6c8fba7 PZ |
2227 | if (p->sched_class != &fair_sched_class) |
2228 | return 0; | |
2229 | ||
ef8002f6 NR |
2230 | if (unlikely(p->policy == SCHED_IDLE)) |
2231 | return 0; | |
2232 | ||
f540a608 IM |
2233 | /* |
2234 | * Buddy candidates are cache hot: | |
2235 | */ | |
f685ceac | 2236 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
2237 | (&p->se == cfs_rq_of(&p->se)->next || |
2238 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2239 | return 1; |
2240 | ||
6bc1665b IM |
2241 | if (sysctl_sched_migration_cost == -1) |
2242 | return 1; | |
2243 | if (sysctl_sched_migration_cost == 0) | |
2244 | return 0; | |
2245 | ||
cc367732 IM |
2246 | delta = now - p->se.exec_start; |
2247 | ||
2248 | return delta < (s64)sysctl_sched_migration_cost; | |
2249 | } | |
2250 | ||
dd41f596 | 2251 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2252 | { |
e2912009 PZ |
2253 | #ifdef CONFIG_SCHED_DEBUG |
2254 | /* | |
2255 | * We should never call set_task_cpu() on a blocked task, | |
2256 | * ttwu() will sort out the placement. | |
2257 | */ | |
077614ee PZ |
2258 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
2259 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
0122ec5b PZ |
2260 | |
2261 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
2262 | /* |
2263 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
2264 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
2265 | * | |
2266 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
2267 | * see set_task_rq(). | |
2268 | * | |
2269 | * Furthermore, all task_rq users should acquire both locks, see | |
2270 | * task_rq_lock(). | |
2271 | */ | |
0122ec5b PZ |
2272 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
2273 | lockdep_is_held(&task_rq(p)->lock))); | |
2274 | #endif | |
e2912009 PZ |
2275 | #endif |
2276 | ||
de1d7286 | 2277 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2278 | |
0c69774e PZ |
2279 | if (task_cpu(p) != new_cpu) { |
2280 | p->se.nr_migrations++; | |
a8b0ca17 | 2281 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); |
0c69774e | 2282 | } |
dd41f596 IM |
2283 | |
2284 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2285 | } |
2286 | ||
969c7921 | 2287 | struct migration_arg { |
36c8b586 | 2288 | struct task_struct *task; |
1da177e4 | 2289 | int dest_cpu; |
70b97a7f | 2290 | }; |
1da177e4 | 2291 | |
969c7921 TH |
2292 | static int migration_cpu_stop(void *data); |
2293 | ||
1da177e4 LT |
2294 | /* |
2295 | * wait_task_inactive - wait for a thread to unschedule. | |
2296 | * | |
85ba2d86 RM |
2297 | * If @match_state is nonzero, it's the @p->state value just checked and |
2298 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2299 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2300 | * we return a positive number (its total switch count). If a second call | |
2301 | * a short while later returns the same number, the caller can be sure that | |
2302 | * @p has remained unscheduled the whole time. | |
2303 | * | |
1da177e4 LT |
2304 | * The caller must ensure that the task *will* unschedule sometime soon, |
2305 | * else this function might spin for a *long* time. This function can't | |
2306 | * be called with interrupts off, or it may introduce deadlock with | |
2307 | * smp_call_function() if an IPI is sent by the same process we are | |
2308 | * waiting to become inactive. | |
2309 | */ | |
85ba2d86 | 2310 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2311 | { |
2312 | unsigned long flags; | |
dd41f596 | 2313 | int running, on_rq; |
85ba2d86 | 2314 | unsigned long ncsw; |
70b97a7f | 2315 | struct rq *rq; |
1da177e4 | 2316 | |
3a5c359a AK |
2317 | for (;;) { |
2318 | /* | |
2319 | * We do the initial early heuristics without holding | |
2320 | * any task-queue locks at all. We'll only try to get | |
2321 | * the runqueue lock when things look like they will | |
2322 | * work out! | |
2323 | */ | |
2324 | rq = task_rq(p); | |
fa490cfd | 2325 | |
3a5c359a AK |
2326 | /* |
2327 | * If the task is actively running on another CPU | |
2328 | * still, just relax and busy-wait without holding | |
2329 | * any locks. | |
2330 | * | |
2331 | * NOTE! Since we don't hold any locks, it's not | |
2332 | * even sure that "rq" stays as the right runqueue! | |
2333 | * But we don't care, since "task_running()" will | |
2334 | * return false if the runqueue has changed and p | |
2335 | * is actually now running somewhere else! | |
2336 | */ | |
85ba2d86 RM |
2337 | while (task_running(rq, p)) { |
2338 | if (match_state && unlikely(p->state != match_state)) | |
2339 | return 0; | |
3a5c359a | 2340 | cpu_relax(); |
85ba2d86 | 2341 | } |
fa490cfd | 2342 | |
3a5c359a AK |
2343 | /* |
2344 | * Ok, time to look more closely! We need the rq | |
2345 | * lock now, to be *sure*. If we're wrong, we'll | |
2346 | * just go back and repeat. | |
2347 | */ | |
2348 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 2349 | trace_sched_wait_task(p); |
3a5c359a | 2350 | running = task_running(rq, p); |
fd2f4419 | 2351 | on_rq = p->on_rq; |
85ba2d86 | 2352 | ncsw = 0; |
f31e11d8 | 2353 | if (!match_state || p->state == match_state) |
93dcf55f | 2354 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 2355 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 2356 | |
85ba2d86 RM |
2357 | /* |
2358 | * If it changed from the expected state, bail out now. | |
2359 | */ | |
2360 | if (unlikely(!ncsw)) | |
2361 | break; | |
2362 | ||
3a5c359a AK |
2363 | /* |
2364 | * Was it really running after all now that we | |
2365 | * checked with the proper locks actually held? | |
2366 | * | |
2367 | * Oops. Go back and try again.. | |
2368 | */ | |
2369 | if (unlikely(running)) { | |
2370 | cpu_relax(); | |
2371 | continue; | |
2372 | } | |
fa490cfd | 2373 | |
3a5c359a AK |
2374 | /* |
2375 | * It's not enough that it's not actively running, | |
2376 | * it must be off the runqueue _entirely_, and not | |
2377 | * preempted! | |
2378 | * | |
80dd99b3 | 2379 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2380 | * running right now), it's preempted, and we should |
2381 | * yield - it could be a while. | |
2382 | */ | |
2383 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
2384 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
2385 | ||
2386 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2387 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
2388 | continue; |
2389 | } | |
fa490cfd | 2390 | |
3a5c359a AK |
2391 | /* |
2392 | * Ahh, all good. It wasn't running, and it wasn't | |
2393 | * runnable, which means that it will never become | |
2394 | * running in the future either. We're all done! | |
2395 | */ | |
2396 | break; | |
2397 | } | |
85ba2d86 RM |
2398 | |
2399 | return ncsw; | |
1da177e4 LT |
2400 | } |
2401 | ||
2402 | /*** | |
2403 | * kick_process - kick a running thread to enter/exit the kernel | |
2404 | * @p: the to-be-kicked thread | |
2405 | * | |
2406 | * Cause a process which is running on another CPU to enter | |
2407 | * kernel-mode, without any delay. (to get signals handled.) | |
2408 | * | |
25985edc | 2409 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
2410 | * because all it wants to ensure is that the remote task enters |
2411 | * the kernel. If the IPI races and the task has been migrated | |
2412 | * to another CPU then no harm is done and the purpose has been | |
2413 | * achieved as well. | |
2414 | */ | |
36c8b586 | 2415 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2416 | { |
2417 | int cpu; | |
2418 | ||
2419 | preempt_disable(); | |
2420 | cpu = task_cpu(p); | |
2421 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2422 | smp_send_reschedule(cpu); | |
2423 | preempt_enable(); | |
2424 | } | |
b43e3521 | 2425 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2426 | #endif /* CONFIG_SMP */ |
1da177e4 | 2427 | |
970b13ba | 2428 | #ifdef CONFIG_SMP |
30da688e | 2429 | /* |
013fdb80 | 2430 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 2431 | */ |
5da9a0fb PZ |
2432 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2433 | { | |
2434 | int dest_cpu; | |
2435 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
2436 | ||
2437 | /* Look for allowed, online CPU in same node. */ | |
2438 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
2439 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
2440 | return dest_cpu; | |
2441 | ||
2442 | /* Any allowed, online CPU? */ | |
2443 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); | |
2444 | if (dest_cpu < nr_cpu_ids) | |
2445 | return dest_cpu; | |
2446 | ||
2447 | /* No more Mr. Nice Guy. */ | |
48c5ccae PZ |
2448 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
2449 | /* | |
2450 | * Don't tell them about moving exiting tasks or | |
2451 | * kernel threads (both mm NULL), since they never | |
2452 | * leave kernel. | |
2453 | */ | |
2454 | if (p->mm && printk_ratelimit()) { | |
2455 | printk(KERN_INFO "process %d (%s) no longer affine to cpu%d\n", | |
2456 | task_pid_nr(p), p->comm, cpu); | |
5da9a0fb PZ |
2457 | } |
2458 | ||
2459 | return dest_cpu; | |
2460 | } | |
2461 | ||
e2912009 | 2462 | /* |
013fdb80 | 2463 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 2464 | */ |
970b13ba | 2465 | static inline |
7608dec2 | 2466 | int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 2467 | { |
7608dec2 | 2468 | int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags); |
e2912009 PZ |
2469 | |
2470 | /* | |
2471 | * In order not to call set_task_cpu() on a blocking task we need | |
2472 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2473 | * cpu. | |
2474 | * | |
2475 | * Since this is common to all placement strategies, this lives here. | |
2476 | * | |
2477 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2478 | * not worry about this generic constraint ] | |
2479 | */ | |
2480 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
70f11205 | 2481 | !cpu_online(cpu))) |
5da9a0fb | 2482 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2483 | |
2484 | return cpu; | |
970b13ba | 2485 | } |
09a40af5 MG |
2486 | |
2487 | static void update_avg(u64 *avg, u64 sample) | |
2488 | { | |
2489 | s64 diff = sample - *avg; | |
2490 | *avg += diff >> 3; | |
2491 | } | |
970b13ba PZ |
2492 | #endif |
2493 | ||
d7c01d27 | 2494 | static void |
b84cb5df | 2495 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 2496 | { |
d7c01d27 | 2497 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
2498 | struct rq *rq = this_rq(); |
2499 | ||
d7c01d27 PZ |
2500 | #ifdef CONFIG_SMP |
2501 | int this_cpu = smp_processor_id(); | |
2502 | ||
2503 | if (cpu == this_cpu) { | |
2504 | schedstat_inc(rq, ttwu_local); | |
2505 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
2506 | } else { | |
2507 | struct sched_domain *sd; | |
2508 | ||
2509 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 2510 | rcu_read_lock(); |
d7c01d27 PZ |
2511 | for_each_domain(this_cpu, sd) { |
2512 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
2513 | schedstat_inc(sd, ttwu_wake_remote); | |
2514 | break; | |
2515 | } | |
2516 | } | |
057f3fad | 2517 | rcu_read_unlock(); |
d7c01d27 | 2518 | } |
f339b9dc PZ |
2519 | |
2520 | if (wake_flags & WF_MIGRATED) | |
2521 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
2522 | ||
d7c01d27 PZ |
2523 | #endif /* CONFIG_SMP */ |
2524 | ||
2525 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 2526 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
2527 | |
2528 | if (wake_flags & WF_SYNC) | |
9ed3811a | 2529 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 2530 | |
d7c01d27 PZ |
2531 | #endif /* CONFIG_SCHEDSTATS */ |
2532 | } | |
2533 | ||
2534 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
2535 | { | |
9ed3811a | 2536 | activate_task(rq, p, en_flags); |
fd2f4419 | 2537 | p->on_rq = 1; |
c2f7115e PZ |
2538 | |
2539 | /* if a worker is waking up, notify workqueue */ | |
2540 | if (p->flags & PF_WQ_WORKER) | |
2541 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
2542 | } |
2543 | ||
23f41eeb PZ |
2544 | /* |
2545 | * Mark the task runnable and perform wakeup-preemption. | |
2546 | */ | |
89363381 | 2547 | static void |
23f41eeb | 2548 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 2549 | { |
89363381 | 2550 | trace_sched_wakeup(p, true); |
9ed3811a TH |
2551 | check_preempt_curr(rq, p, wake_flags); |
2552 | ||
2553 | p->state = TASK_RUNNING; | |
2554 | #ifdef CONFIG_SMP | |
2555 | if (p->sched_class->task_woken) | |
2556 | p->sched_class->task_woken(rq, p); | |
2557 | ||
e69c6341 | 2558 | if (rq->idle_stamp) { |
9ed3811a TH |
2559 | u64 delta = rq->clock - rq->idle_stamp; |
2560 | u64 max = 2*sysctl_sched_migration_cost; | |
2561 | ||
2562 | if (delta > max) | |
2563 | rq->avg_idle = max; | |
2564 | else | |
2565 | update_avg(&rq->avg_idle, delta); | |
2566 | rq->idle_stamp = 0; | |
2567 | } | |
2568 | #endif | |
2569 | } | |
2570 | ||
c05fbafb PZ |
2571 | static void |
2572 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
2573 | { | |
2574 | #ifdef CONFIG_SMP | |
2575 | if (p->sched_contributes_to_load) | |
2576 | rq->nr_uninterruptible--; | |
2577 | #endif | |
2578 | ||
2579 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
2580 | ttwu_do_wakeup(rq, p, wake_flags); | |
2581 | } | |
2582 | ||
2583 | /* | |
2584 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
2585 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
2586 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
2587 | * the task is still ->on_rq. | |
2588 | */ | |
2589 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
2590 | { | |
2591 | struct rq *rq; | |
2592 | int ret = 0; | |
2593 | ||
2594 | rq = __task_rq_lock(p); | |
2595 | if (p->on_rq) { | |
2596 | ttwu_do_wakeup(rq, p, wake_flags); | |
2597 | ret = 1; | |
2598 | } | |
2599 | __task_rq_unlock(rq); | |
2600 | ||
2601 | return ret; | |
2602 | } | |
2603 | ||
317f3941 | 2604 | #ifdef CONFIG_SMP |
c5d753a5 | 2605 | static void sched_ttwu_do_pending(struct task_struct *list) |
317f3941 PZ |
2606 | { |
2607 | struct rq *rq = this_rq(); | |
317f3941 PZ |
2608 | |
2609 | raw_spin_lock(&rq->lock); | |
2610 | ||
2611 | while (list) { | |
2612 | struct task_struct *p = list; | |
2613 | list = list->wake_entry; | |
2614 | ttwu_do_activate(rq, p, 0); | |
2615 | } | |
2616 | ||
2617 | raw_spin_unlock(&rq->lock); | |
2618 | } | |
2619 | ||
c5d753a5 PZ |
2620 | #ifdef CONFIG_HOTPLUG_CPU |
2621 | ||
2622 | static void sched_ttwu_pending(void) | |
2623 | { | |
2624 | struct rq *rq = this_rq(); | |
2625 | struct task_struct *list = xchg(&rq->wake_list, NULL); | |
2626 | ||
2627 | if (!list) | |
2628 | return; | |
2629 | ||
2630 | sched_ttwu_do_pending(list); | |
2631 | } | |
2632 | ||
2633 | #endif /* CONFIG_HOTPLUG_CPU */ | |
2634 | ||
317f3941 PZ |
2635 | void scheduler_ipi(void) |
2636 | { | |
c5d753a5 PZ |
2637 | struct rq *rq = this_rq(); |
2638 | struct task_struct *list = xchg(&rq->wake_list, NULL); | |
2639 | ||
2640 | if (!list) | |
2641 | return; | |
2642 | ||
2643 | /* | |
2644 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
2645 | * traditionally all their work was done from the interrupt return | |
2646 | * path. Now that we actually do some work, we need to make sure | |
2647 | * we do call them. | |
2648 | * | |
2649 | * Some archs already do call them, luckily irq_enter/exit nest | |
2650 | * properly. | |
2651 | * | |
2652 | * Arguably we should visit all archs and update all handlers, | |
2653 | * however a fair share of IPIs are still resched only so this would | |
2654 | * somewhat pessimize the simple resched case. | |
2655 | */ | |
2656 | irq_enter(); | |
2657 | sched_ttwu_do_pending(list); | |
2658 | irq_exit(); | |
317f3941 PZ |
2659 | } |
2660 | ||
2661 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
2662 | { | |
2663 | struct rq *rq = cpu_rq(cpu); | |
2664 | struct task_struct *next = rq->wake_list; | |
2665 | ||
2666 | for (;;) { | |
2667 | struct task_struct *old = next; | |
2668 | ||
2669 | p->wake_entry = next; | |
2670 | next = cmpxchg(&rq->wake_list, old, p); | |
2671 | if (next == old) | |
2672 | break; | |
2673 | } | |
2674 | ||
2675 | if (!next) | |
2676 | smp_send_reschedule(cpu); | |
2677 | } | |
d6aa8f85 PZ |
2678 | |
2679 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
2680 | static int ttwu_activate_remote(struct task_struct *p, int wake_flags) | |
2681 | { | |
2682 | struct rq *rq; | |
2683 | int ret = 0; | |
2684 | ||
2685 | rq = __task_rq_lock(p); | |
2686 | if (p->on_cpu) { | |
2687 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); | |
2688 | ttwu_do_wakeup(rq, p, wake_flags); | |
2689 | ret = 1; | |
2690 | } | |
2691 | __task_rq_unlock(rq); | |
2692 | ||
2693 | return ret; | |
2694 | ||
2695 | } | |
2696 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
2697 | #endif /* CONFIG_SMP */ | |
317f3941 | 2698 | |
c05fbafb PZ |
2699 | static void ttwu_queue(struct task_struct *p, int cpu) |
2700 | { | |
2701 | struct rq *rq = cpu_rq(cpu); | |
2702 | ||
17d9f311 | 2703 | #if defined(CONFIG_SMP) |
317f3941 | 2704 | if (sched_feat(TTWU_QUEUE) && cpu != smp_processor_id()) { |
f01114cb | 2705 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
2706 | ttwu_queue_remote(p, cpu); |
2707 | return; | |
2708 | } | |
2709 | #endif | |
2710 | ||
c05fbafb PZ |
2711 | raw_spin_lock(&rq->lock); |
2712 | ttwu_do_activate(rq, p, 0); | |
2713 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
2714 | } |
2715 | ||
2716 | /** | |
1da177e4 | 2717 | * try_to_wake_up - wake up a thread |
9ed3811a | 2718 | * @p: the thread to be awakened |
1da177e4 | 2719 | * @state: the mask of task states that can be woken |
9ed3811a | 2720 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
2721 | * |
2722 | * Put it on the run-queue if it's not already there. The "current" | |
2723 | * thread is always on the run-queue (except when the actual | |
2724 | * re-schedule is in progress), and as such you're allowed to do | |
2725 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2726 | * runnable without the overhead of this. | |
2727 | * | |
9ed3811a TH |
2728 | * Returns %true if @p was woken up, %false if it was already running |
2729 | * or @state didn't match @p's state. | |
1da177e4 | 2730 | */ |
e4a52bcb PZ |
2731 | static int |
2732 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 2733 | { |
1da177e4 | 2734 | unsigned long flags; |
c05fbafb | 2735 | int cpu, success = 0; |
2398f2c6 | 2736 | |
04e2f174 | 2737 | smp_wmb(); |
013fdb80 | 2738 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 2739 | if (!(p->state & state)) |
1da177e4 LT |
2740 | goto out; |
2741 | ||
c05fbafb | 2742 | success = 1; /* we're going to change ->state */ |
1da177e4 | 2743 | cpu = task_cpu(p); |
1da177e4 | 2744 | |
c05fbafb PZ |
2745 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
2746 | goto stat; | |
1da177e4 | 2747 | |
1da177e4 | 2748 | #ifdef CONFIG_SMP |
e9c84311 | 2749 | /* |
c05fbafb PZ |
2750 | * If the owning (remote) cpu is still in the middle of schedule() with |
2751 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 2752 | */ |
e4a52bcb PZ |
2753 | while (p->on_cpu) { |
2754 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
2755 | /* | |
d6aa8f85 PZ |
2756 | * In case the architecture enables interrupts in |
2757 | * context_switch(), we cannot busy wait, since that | |
2758 | * would lead to deadlocks when an interrupt hits and | |
2759 | * tries to wake up @prev. So bail and do a complete | |
2760 | * remote wakeup. | |
e4a52bcb | 2761 | */ |
d6aa8f85 | 2762 | if (ttwu_activate_remote(p, wake_flags)) |
c05fbafb | 2763 | goto stat; |
d6aa8f85 | 2764 | #else |
e4a52bcb | 2765 | cpu_relax(); |
d6aa8f85 | 2766 | #endif |
371fd7e7 | 2767 | } |
0970d299 | 2768 | /* |
e4a52bcb | 2769 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 2770 | */ |
e4a52bcb | 2771 | smp_rmb(); |
1da177e4 | 2772 | |
a8e4f2ea | 2773 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 2774 | p->state = TASK_WAKING; |
e7693a36 | 2775 | |
e4a52bcb | 2776 | if (p->sched_class->task_waking) |
74f8e4b2 | 2777 | p->sched_class->task_waking(p); |
efbbd05a | 2778 | |
7608dec2 | 2779 | cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
2780 | if (task_cpu(p) != cpu) { |
2781 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 2782 | set_task_cpu(p, cpu); |
f339b9dc | 2783 | } |
1da177e4 | 2784 | #endif /* CONFIG_SMP */ |
1da177e4 | 2785 | |
c05fbafb PZ |
2786 | ttwu_queue(p, cpu); |
2787 | stat: | |
b84cb5df | 2788 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 2789 | out: |
013fdb80 | 2790 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
2791 | |
2792 | return success; | |
2793 | } | |
2794 | ||
21aa9af0 TH |
2795 | /** |
2796 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
2797 | * @p: the thread to be awakened | |
2798 | * | |
2acca55e | 2799 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 2800 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 2801 | * the current task. |
21aa9af0 TH |
2802 | */ |
2803 | static void try_to_wake_up_local(struct task_struct *p) | |
2804 | { | |
2805 | struct rq *rq = task_rq(p); | |
21aa9af0 TH |
2806 | |
2807 | BUG_ON(rq != this_rq()); | |
2808 | BUG_ON(p == current); | |
2809 | lockdep_assert_held(&rq->lock); | |
2810 | ||
2acca55e PZ |
2811 | if (!raw_spin_trylock(&p->pi_lock)) { |
2812 | raw_spin_unlock(&rq->lock); | |
2813 | raw_spin_lock(&p->pi_lock); | |
2814 | raw_spin_lock(&rq->lock); | |
2815 | } | |
2816 | ||
21aa9af0 | 2817 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 2818 | goto out; |
21aa9af0 | 2819 | |
fd2f4419 | 2820 | if (!p->on_rq) |
d7c01d27 PZ |
2821 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
2822 | ||
23f41eeb | 2823 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 2824 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
2825 | out: |
2826 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
2827 | } |
2828 | ||
50fa610a DH |
2829 | /** |
2830 | * wake_up_process - Wake up a specific process | |
2831 | * @p: The process to be woken up. | |
2832 | * | |
2833 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2834 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2835 | * running. | |
2836 | * | |
2837 | * It may be assumed that this function implies a write memory barrier before | |
2838 | * changing the task state if and only if any tasks are woken up. | |
2839 | */ | |
7ad5b3a5 | 2840 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2841 | { |
d9514f6c | 2842 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2843 | } |
1da177e4 LT |
2844 | EXPORT_SYMBOL(wake_up_process); |
2845 | ||
7ad5b3a5 | 2846 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2847 | { |
2848 | return try_to_wake_up(p, state, 0); | |
2849 | } | |
2850 | ||
1da177e4 LT |
2851 | /* |
2852 | * Perform scheduler related setup for a newly forked process p. | |
2853 | * p is forked by current. | |
dd41f596 IM |
2854 | * |
2855 | * __sched_fork() is basic setup used by init_idle() too: | |
2856 | */ | |
2857 | static void __sched_fork(struct task_struct *p) | |
2858 | { | |
fd2f4419 PZ |
2859 | p->on_rq = 0; |
2860 | ||
2861 | p->se.on_rq = 0; | |
dd41f596 IM |
2862 | p->se.exec_start = 0; |
2863 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2864 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2865 | p->se.nr_migrations = 0; |
da7a735e | 2866 | p->se.vruntime = 0; |
fd2f4419 | 2867 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
2868 | |
2869 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 2870 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2871 | #endif |
476d139c | 2872 | |
fa717060 | 2873 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 2874 | |
e107be36 AK |
2875 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2876 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2877 | #endif | |
dd41f596 IM |
2878 | } |
2879 | ||
2880 | /* | |
2881 | * fork()/clone()-time setup: | |
2882 | */ | |
3e51e3ed | 2883 | void sched_fork(struct task_struct *p) |
dd41f596 | 2884 | { |
0122ec5b | 2885 | unsigned long flags; |
dd41f596 IM |
2886 | int cpu = get_cpu(); |
2887 | ||
2888 | __sched_fork(p); | |
06b83b5f | 2889 | /* |
0017d735 | 2890 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2891 | * nobody will actually run it, and a signal or other external |
2892 | * event cannot wake it up and insert it on the runqueue either. | |
2893 | */ | |
0017d735 | 2894 | p->state = TASK_RUNNING; |
dd41f596 | 2895 | |
c350a04e MG |
2896 | /* |
2897 | * Make sure we do not leak PI boosting priority to the child. | |
2898 | */ | |
2899 | p->prio = current->normal_prio; | |
2900 | ||
b9dc29e7 MG |
2901 | /* |
2902 | * Revert to default priority/policy on fork if requested. | |
2903 | */ | |
2904 | if (unlikely(p->sched_reset_on_fork)) { | |
c350a04e | 2905 | if (task_has_rt_policy(p)) { |
b9dc29e7 | 2906 | p->policy = SCHED_NORMAL; |
6c697bdf | 2907 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
2908 | p->rt_priority = 0; |
2909 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
2910 | p->static_prio = NICE_TO_PRIO(0); | |
2911 | ||
2912 | p->prio = p->normal_prio = __normal_prio(p); | |
2913 | set_load_weight(p); | |
6c697bdf | 2914 | |
b9dc29e7 MG |
2915 | /* |
2916 | * We don't need the reset flag anymore after the fork. It has | |
2917 | * fulfilled its duty: | |
2918 | */ | |
2919 | p->sched_reset_on_fork = 0; | |
2920 | } | |
ca94c442 | 2921 | |
2ddbf952 HS |
2922 | if (!rt_prio(p->prio)) |
2923 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2924 | |
cd29fe6f PZ |
2925 | if (p->sched_class->task_fork) |
2926 | p->sched_class->task_fork(p); | |
2927 | ||
86951599 PZ |
2928 | /* |
2929 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2930 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2931 | * is ran before sched_fork(). | |
2932 | * | |
2933 | * Silence PROVE_RCU. | |
2934 | */ | |
0122ec5b | 2935 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 2936 | set_task_cpu(p, cpu); |
0122ec5b | 2937 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 2938 | |
52f17b6c | 2939 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2940 | if (likely(sched_info_on())) |
52f17b6c | 2941 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2942 | #endif |
3ca7a440 PZ |
2943 | #if defined(CONFIG_SMP) |
2944 | p->on_cpu = 0; | |
4866cde0 | 2945 | #endif |
bdd4e85d | 2946 | #ifdef CONFIG_PREEMPT_COUNT |
4866cde0 | 2947 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2948 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2949 | #endif |
806c09a7 | 2950 | #ifdef CONFIG_SMP |
917b627d | 2951 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
806c09a7 | 2952 | #endif |
917b627d | 2953 | |
476d139c | 2954 | put_cpu(); |
1da177e4 LT |
2955 | } |
2956 | ||
2957 | /* | |
2958 | * wake_up_new_task - wake up a newly created task for the first time. | |
2959 | * | |
2960 | * This function will do some initial scheduler statistics housekeeping | |
2961 | * that must be done for every newly created context, then puts the task | |
2962 | * on the runqueue and wakes it. | |
2963 | */ | |
3e51e3ed | 2964 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
2965 | { |
2966 | unsigned long flags; | |
dd41f596 | 2967 | struct rq *rq; |
fabf318e | 2968 | |
ab2515c4 | 2969 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
2970 | #ifdef CONFIG_SMP |
2971 | /* | |
2972 | * Fork balancing, do it here and not earlier because: | |
2973 | * - cpus_allowed can change in the fork path | |
2974 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 2975 | */ |
ab2515c4 | 2976 | set_task_cpu(p, select_task_rq(p, SD_BALANCE_FORK, 0)); |
0017d735 PZ |
2977 | #endif |
2978 | ||
ab2515c4 | 2979 | rq = __task_rq_lock(p); |
cd29fe6f | 2980 | activate_task(rq, p, 0); |
fd2f4419 | 2981 | p->on_rq = 1; |
89363381 | 2982 | trace_sched_wakeup_new(p, true); |
a7558e01 | 2983 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2984 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2985 | if (p->sched_class->task_woken) |
2986 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2987 | #endif |
0122ec5b | 2988 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
2989 | } |
2990 | ||
e107be36 AK |
2991 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2992 | ||
2993 | /** | |
80dd99b3 | 2994 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2995 | * @notifier: notifier struct to register |
e107be36 AK |
2996 | */ |
2997 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2998 | { | |
2999 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
3000 | } | |
3001 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
3002 | ||
3003 | /** | |
3004 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 3005 | * @notifier: notifier struct to unregister |
e107be36 AK |
3006 | * |
3007 | * This is safe to call from within a preemption notifier. | |
3008 | */ | |
3009 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
3010 | { | |
3011 | hlist_del(¬ifier->link); | |
3012 | } | |
3013 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
3014 | ||
3015 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
3016 | { | |
3017 | struct preempt_notifier *notifier; | |
3018 | struct hlist_node *node; | |
3019 | ||
3020 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
3021 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
3022 | } | |
3023 | ||
3024 | static void | |
3025 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
3026 | struct task_struct *next) | |
3027 | { | |
3028 | struct preempt_notifier *notifier; | |
3029 | struct hlist_node *node; | |
3030 | ||
3031 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
3032 | notifier->ops->sched_out(notifier, next); | |
3033 | } | |
3034 | ||
6d6bc0ad | 3035 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
3036 | |
3037 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
3038 | { | |
3039 | } | |
3040 | ||
3041 | static void | |
3042 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
3043 | struct task_struct *next) | |
3044 | { | |
3045 | } | |
3046 | ||
6d6bc0ad | 3047 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 3048 | |
4866cde0 NP |
3049 | /** |
3050 | * prepare_task_switch - prepare to switch tasks | |
3051 | * @rq: the runqueue preparing to switch | |
421cee29 | 3052 | * @prev: the current task that is being switched out |
4866cde0 NP |
3053 | * @next: the task we are going to switch to. |
3054 | * | |
3055 | * This is called with the rq lock held and interrupts off. It must | |
3056 | * be paired with a subsequent finish_task_switch after the context | |
3057 | * switch. | |
3058 | * | |
3059 | * prepare_task_switch sets up locking and calls architecture specific | |
3060 | * hooks. | |
3061 | */ | |
e107be36 AK |
3062 | static inline void |
3063 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
3064 | struct task_struct *next) | |
4866cde0 | 3065 | { |
fe4b04fa PZ |
3066 | sched_info_switch(prev, next); |
3067 | perf_event_task_sched_out(prev, next); | |
e107be36 | 3068 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
3069 | prepare_lock_switch(rq, next); |
3070 | prepare_arch_switch(next); | |
fe4b04fa | 3071 | trace_sched_switch(prev, next); |
4866cde0 NP |
3072 | } |
3073 | ||
1da177e4 LT |
3074 | /** |
3075 | * finish_task_switch - clean up after a task-switch | |
344babaa | 3076 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
3077 | * @prev: the thread we just switched away from. |
3078 | * | |
4866cde0 NP |
3079 | * finish_task_switch must be called after the context switch, paired |
3080 | * with a prepare_task_switch call before the context switch. | |
3081 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
3082 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
3083 | * |
3084 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 3085 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
3086 | * with the lock held can cause deadlocks; see schedule() for |
3087 | * details.) | |
3088 | */ | |
a9957449 | 3089 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
3090 | __releases(rq->lock) |
3091 | { | |
1da177e4 | 3092 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 3093 | long prev_state; |
1da177e4 LT |
3094 | |
3095 | rq->prev_mm = NULL; | |
3096 | ||
3097 | /* | |
3098 | * A task struct has one reference for the use as "current". | |
c394cc9f | 3099 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
3100 | * schedule one last time. The schedule call will never return, and |
3101 | * the scheduled task must drop that reference. | |
c394cc9f | 3102 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
3103 | * still held, otherwise prev could be scheduled on another cpu, die |
3104 | * there before we look at prev->state, and then the reference would | |
3105 | * be dropped twice. | |
3106 | * Manfred Spraul <manfred@colorfullife.com> | |
3107 | */ | |
55a101f8 | 3108 | prev_state = prev->state; |
4866cde0 | 3109 | finish_arch_switch(prev); |
8381f65d JI |
3110 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
3111 | local_irq_disable(); | |
3112 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
49f47433 | 3113 | perf_event_task_sched_in(current); |
8381f65d JI |
3114 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
3115 | local_irq_enable(); | |
3116 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 3117 | finish_lock_switch(rq, prev); |
e8fa1362 | 3118 | |
e107be36 | 3119 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
3120 | if (mm) |
3121 | mmdrop(mm); | |
c394cc9f | 3122 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 3123 | /* |
3124 | * Remove function-return probe instances associated with this | |
3125 | * task and put them back on the free list. | |
9761eea8 | 3126 | */ |
c6fd91f0 | 3127 | kprobe_flush_task(prev); |
1da177e4 | 3128 | put_task_struct(prev); |
c6fd91f0 | 3129 | } |
1da177e4 LT |
3130 | } |
3131 | ||
3f029d3c GH |
3132 | #ifdef CONFIG_SMP |
3133 | ||
3134 | /* assumes rq->lock is held */ | |
3135 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
3136 | { | |
3137 | if (prev->sched_class->pre_schedule) | |
3138 | prev->sched_class->pre_schedule(rq, prev); | |
3139 | } | |
3140 | ||
3141 | /* rq->lock is NOT held, but preemption is disabled */ | |
3142 | static inline void post_schedule(struct rq *rq) | |
3143 | { | |
3144 | if (rq->post_schedule) { | |
3145 | unsigned long flags; | |
3146 | ||
05fa785c | 3147 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
3148 | if (rq->curr->sched_class->post_schedule) |
3149 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 3150 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
3151 | |
3152 | rq->post_schedule = 0; | |
3153 | } | |
3154 | } | |
3155 | ||
3156 | #else | |
da19ab51 | 3157 | |
3f029d3c GH |
3158 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
3159 | { | |
3160 | } | |
3161 | ||
3162 | static inline void post_schedule(struct rq *rq) | |
3163 | { | |
1da177e4 LT |
3164 | } |
3165 | ||
3f029d3c GH |
3166 | #endif |
3167 | ||
1da177e4 LT |
3168 | /** |
3169 | * schedule_tail - first thing a freshly forked thread must call. | |
3170 | * @prev: the thread we just switched away from. | |
3171 | */ | |
36c8b586 | 3172 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
3173 | __releases(rq->lock) |
3174 | { | |
70b97a7f IM |
3175 | struct rq *rq = this_rq(); |
3176 | ||
4866cde0 | 3177 | finish_task_switch(rq, prev); |
da19ab51 | 3178 | |
3f029d3c GH |
3179 | /* |
3180 | * FIXME: do we need to worry about rq being invalidated by the | |
3181 | * task_switch? | |
3182 | */ | |
3183 | post_schedule(rq); | |
70b97a7f | 3184 | |
4866cde0 NP |
3185 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
3186 | /* In this case, finish_task_switch does not reenable preemption */ | |
3187 | preempt_enable(); | |
3188 | #endif | |
1da177e4 | 3189 | if (current->set_child_tid) |
b488893a | 3190 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
3191 | } |
3192 | ||
3193 | /* | |
3194 | * context_switch - switch to the new MM and the new | |
3195 | * thread's register state. | |
3196 | */ | |
dd41f596 | 3197 | static inline void |
70b97a7f | 3198 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 3199 | struct task_struct *next) |
1da177e4 | 3200 | { |
dd41f596 | 3201 | struct mm_struct *mm, *oldmm; |
1da177e4 | 3202 | |
e107be36 | 3203 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 3204 | |
dd41f596 IM |
3205 | mm = next->mm; |
3206 | oldmm = prev->active_mm; | |
9226d125 ZA |
3207 | /* |
3208 | * For paravirt, this is coupled with an exit in switch_to to | |
3209 | * combine the page table reload and the switch backend into | |
3210 | * one hypercall. | |
3211 | */ | |
224101ed | 3212 | arch_start_context_switch(prev); |
9226d125 | 3213 | |
31915ab4 | 3214 | if (!mm) { |
1da177e4 LT |
3215 | next->active_mm = oldmm; |
3216 | atomic_inc(&oldmm->mm_count); | |
3217 | enter_lazy_tlb(oldmm, next); | |
3218 | } else | |
3219 | switch_mm(oldmm, mm, next); | |
3220 | ||
31915ab4 | 3221 | if (!prev->mm) { |
1da177e4 | 3222 | prev->active_mm = NULL; |
1da177e4 LT |
3223 | rq->prev_mm = oldmm; |
3224 | } | |
3a5f5e48 IM |
3225 | /* |
3226 | * Since the runqueue lock will be released by the next | |
3227 | * task (which is an invalid locking op but in the case | |
3228 | * of the scheduler it's an obvious special-case), so we | |
3229 | * do an early lockdep release here: | |
3230 | */ | |
3231 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 3232 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 3233 | #endif |
1da177e4 LT |
3234 | |
3235 | /* Here we just switch the register state and the stack. */ | |
3236 | switch_to(prev, next, prev); | |
3237 | ||
dd41f596 IM |
3238 | barrier(); |
3239 | /* | |
3240 | * this_rq must be evaluated again because prev may have moved | |
3241 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
3242 | * frame will be invalid. | |
3243 | */ | |
3244 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
3245 | } |
3246 | ||
3247 | /* | |
3248 | * nr_running, nr_uninterruptible and nr_context_switches: | |
3249 | * | |
3250 | * externally visible scheduler statistics: current number of runnable | |
3251 | * threads, current number of uninterruptible-sleeping threads, total | |
3252 | * number of context switches performed since bootup. | |
3253 | */ | |
3254 | unsigned long nr_running(void) | |
3255 | { | |
3256 | unsigned long i, sum = 0; | |
3257 | ||
3258 | for_each_online_cpu(i) | |
3259 | sum += cpu_rq(i)->nr_running; | |
3260 | ||
3261 | return sum; | |
f711f609 | 3262 | } |
1da177e4 LT |
3263 | |
3264 | unsigned long nr_uninterruptible(void) | |
f711f609 | 3265 | { |
1da177e4 | 3266 | unsigned long i, sum = 0; |
f711f609 | 3267 | |
0a945022 | 3268 | for_each_possible_cpu(i) |
1da177e4 | 3269 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
3270 | |
3271 | /* | |
1da177e4 LT |
3272 | * Since we read the counters lockless, it might be slightly |
3273 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 3274 | */ |
1da177e4 LT |
3275 | if (unlikely((long)sum < 0)) |
3276 | sum = 0; | |
f711f609 | 3277 | |
1da177e4 | 3278 | return sum; |
f711f609 | 3279 | } |
f711f609 | 3280 | |
1da177e4 | 3281 | unsigned long long nr_context_switches(void) |
46cb4b7c | 3282 | { |
cc94abfc SR |
3283 | int i; |
3284 | unsigned long long sum = 0; | |
46cb4b7c | 3285 | |
0a945022 | 3286 | for_each_possible_cpu(i) |
1da177e4 | 3287 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 3288 | |
1da177e4 LT |
3289 | return sum; |
3290 | } | |
483b4ee6 | 3291 | |
1da177e4 LT |
3292 | unsigned long nr_iowait(void) |
3293 | { | |
3294 | unsigned long i, sum = 0; | |
483b4ee6 | 3295 | |
0a945022 | 3296 | for_each_possible_cpu(i) |
1da177e4 | 3297 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 3298 | |
1da177e4 LT |
3299 | return sum; |
3300 | } | |
483b4ee6 | 3301 | |
8c215bd3 | 3302 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 3303 | { |
8c215bd3 | 3304 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
3305 | return atomic_read(&this->nr_iowait); |
3306 | } | |
46cb4b7c | 3307 | |
69d25870 AV |
3308 | unsigned long this_cpu_load(void) |
3309 | { | |
3310 | struct rq *this = this_rq(); | |
3311 | return this->cpu_load[0]; | |
3312 | } | |
e790fb0b | 3313 | |
46cb4b7c | 3314 | |
dce48a84 TG |
3315 | /* Variables and functions for calc_load */ |
3316 | static atomic_long_t calc_load_tasks; | |
3317 | static unsigned long calc_load_update; | |
3318 | unsigned long avenrun[3]; | |
3319 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 3320 | |
74f5187a PZ |
3321 | static long calc_load_fold_active(struct rq *this_rq) |
3322 | { | |
3323 | long nr_active, delta = 0; | |
3324 | ||
3325 | nr_active = this_rq->nr_running; | |
3326 | nr_active += (long) this_rq->nr_uninterruptible; | |
3327 | ||
3328 | if (nr_active != this_rq->calc_load_active) { | |
3329 | delta = nr_active - this_rq->calc_load_active; | |
3330 | this_rq->calc_load_active = nr_active; | |
3331 | } | |
3332 | ||
3333 | return delta; | |
3334 | } | |
3335 | ||
0f004f5a PZ |
3336 | static unsigned long |
3337 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
3338 | { | |
3339 | load *= exp; | |
3340 | load += active * (FIXED_1 - exp); | |
3341 | load += 1UL << (FSHIFT - 1); | |
3342 | return load >> FSHIFT; | |
3343 | } | |
3344 | ||
74f5187a PZ |
3345 | #ifdef CONFIG_NO_HZ |
3346 | /* | |
3347 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | |
3348 | * | |
3349 | * When making the ILB scale, we should try to pull this in as well. | |
3350 | */ | |
3351 | static atomic_long_t calc_load_tasks_idle; | |
3352 | ||
3353 | static void calc_load_account_idle(struct rq *this_rq) | |
3354 | { | |
3355 | long delta; | |
3356 | ||
3357 | delta = calc_load_fold_active(this_rq); | |
3358 | if (delta) | |
3359 | atomic_long_add(delta, &calc_load_tasks_idle); | |
3360 | } | |
3361 | ||
3362 | static long calc_load_fold_idle(void) | |
3363 | { | |
3364 | long delta = 0; | |
3365 | ||
3366 | /* | |
3367 | * Its got a race, we don't care... | |
3368 | */ | |
3369 | if (atomic_long_read(&calc_load_tasks_idle)) | |
3370 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | |
3371 | ||
3372 | return delta; | |
3373 | } | |
0f004f5a PZ |
3374 | |
3375 | /** | |
3376 | * fixed_power_int - compute: x^n, in O(log n) time | |
3377 | * | |
3378 | * @x: base of the power | |
3379 | * @frac_bits: fractional bits of @x | |
3380 | * @n: power to raise @x to. | |
3381 | * | |
3382 | * By exploiting the relation between the definition of the natural power | |
3383 | * function: x^n := x*x*...*x (x multiplied by itself for n times), and | |
3384 | * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i, | |
3385 | * (where: n_i \elem {0, 1}, the binary vector representing n), | |
3386 | * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is | |
3387 | * of course trivially computable in O(log_2 n), the length of our binary | |
3388 | * vector. | |
3389 | */ | |
3390 | static unsigned long | |
3391 | fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n) | |
3392 | { | |
3393 | unsigned long result = 1UL << frac_bits; | |
3394 | ||
3395 | if (n) for (;;) { | |
3396 | if (n & 1) { | |
3397 | result *= x; | |
3398 | result += 1UL << (frac_bits - 1); | |
3399 | result >>= frac_bits; | |
3400 | } | |
3401 | n >>= 1; | |
3402 | if (!n) | |
3403 | break; | |
3404 | x *= x; | |
3405 | x += 1UL << (frac_bits - 1); | |
3406 | x >>= frac_bits; | |
3407 | } | |
3408 | ||
3409 | return result; | |
3410 | } | |
3411 | ||
3412 | /* | |
3413 | * a1 = a0 * e + a * (1 - e) | |
3414 | * | |
3415 | * a2 = a1 * e + a * (1 - e) | |
3416 | * = (a0 * e + a * (1 - e)) * e + a * (1 - e) | |
3417 | * = a0 * e^2 + a * (1 - e) * (1 + e) | |
3418 | * | |
3419 | * a3 = a2 * e + a * (1 - e) | |
3420 | * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e) | |
3421 | * = a0 * e^3 + a * (1 - e) * (1 + e + e^2) | |
3422 | * | |
3423 | * ... | |
3424 | * | |
3425 | * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1] | |
3426 | * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e) | |
3427 | * = a0 * e^n + a * (1 - e^n) | |
3428 | * | |
3429 | * [1] application of the geometric series: | |
3430 | * | |
3431 | * n 1 - x^(n+1) | |
3432 | * S_n := \Sum x^i = ------------- | |
3433 | * i=0 1 - x | |
3434 | */ | |
3435 | static unsigned long | |
3436 | calc_load_n(unsigned long load, unsigned long exp, | |
3437 | unsigned long active, unsigned int n) | |
3438 | { | |
3439 | ||
3440 | return calc_load(load, fixed_power_int(exp, FSHIFT, n), active); | |
3441 | } | |
3442 | ||
3443 | /* | |
3444 | * NO_HZ can leave us missing all per-cpu ticks calling | |
3445 | * calc_load_account_active(), but since an idle CPU folds its delta into | |
3446 | * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold | |
3447 | * in the pending idle delta if our idle period crossed a load cycle boundary. | |
3448 | * | |
3449 | * Once we've updated the global active value, we need to apply the exponential | |
3450 | * weights adjusted to the number of cycles missed. | |
3451 | */ | |
3452 | static void calc_global_nohz(unsigned long ticks) | |
3453 | { | |
3454 | long delta, active, n; | |
3455 | ||
3456 | if (time_before(jiffies, calc_load_update)) | |
3457 | return; | |
3458 | ||
3459 | /* | |
3460 | * If we crossed a calc_load_update boundary, make sure to fold | |
3461 | * any pending idle changes, the respective CPUs might have | |
3462 | * missed the tick driven calc_load_account_active() update | |
3463 | * due to NO_HZ. | |
3464 | */ | |
3465 | delta = calc_load_fold_idle(); | |
3466 | if (delta) | |
3467 | atomic_long_add(delta, &calc_load_tasks); | |
3468 | ||
3469 | /* | |
3470 | * If we were idle for multiple load cycles, apply them. | |
3471 | */ | |
3472 | if (ticks >= LOAD_FREQ) { | |
3473 | n = ticks / LOAD_FREQ; | |
3474 | ||
3475 | active = atomic_long_read(&calc_load_tasks); | |
3476 | active = active > 0 ? active * FIXED_1 : 0; | |
3477 | ||
3478 | avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); | |
3479 | avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); | |
3480 | avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); | |
3481 | ||
3482 | calc_load_update += n * LOAD_FREQ; | |
3483 | } | |
3484 | ||
3485 | /* | |
3486 | * Its possible the remainder of the above division also crosses | |
3487 | * a LOAD_FREQ period, the regular check in calc_global_load() | |
3488 | * which comes after this will take care of that. | |
3489 | * | |
3490 | * Consider us being 11 ticks before a cycle completion, and us | |
3491 | * sleeping for 4*LOAD_FREQ + 22 ticks, then the above code will | |
3492 | * age us 4 cycles, and the test in calc_global_load() will | |
3493 | * pick up the final one. | |
3494 | */ | |
3495 | } | |
74f5187a PZ |
3496 | #else |
3497 | static void calc_load_account_idle(struct rq *this_rq) | |
3498 | { | |
3499 | } | |
3500 | ||
3501 | static inline long calc_load_fold_idle(void) | |
3502 | { | |
3503 | return 0; | |
3504 | } | |
0f004f5a PZ |
3505 | |
3506 | static void calc_global_nohz(unsigned long ticks) | |
3507 | { | |
3508 | } | |
74f5187a PZ |
3509 | #endif |
3510 | ||
2d02494f TG |
3511 | /** |
3512 | * get_avenrun - get the load average array | |
3513 | * @loads: pointer to dest load array | |
3514 | * @offset: offset to add | |
3515 | * @shift: shift count to shift the result left | |
3516 | * | |
3517 | * These values are estimates at best, so no need for locking. | |
3518 | */ | |
3519 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
3520 | { | |
3521 | loads[0] = (avenrun[0] + offset) << shift; | |
3522 | loads[1] = (avenrun[1] + offset) << shift; | |
3523 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 3524 | } |
46cb4b7c | 3525 | |
46cb4b7c | 3526 | /* |
dce48a84 TG |
3527 | * calc_load - update the avenrun load estimates 10 ticks after the |
3528 | * CPUs have updated calc_load_tasks. | |
7835b98b | 3529 | */ |
0f004f5a | 3530 | void calc_global_load(unsigned long ticks) |
7835b98b | 3531 | { |
dce48a84 | 3532 | long active; |
1da177e4 | 3533 | |
0f004f5a PZ |
3534 | calc_global_nohz(ticks); |
3535 | ||
3536 | if (time_before(jiffies, calc_load_update + 10)) | |
dce48a84 | 3537 | return; |
1da177e4 | 3538 | |
dce48a84 TG |
3539 | active = atomic_long_read(&calc_load_tasks); |
3540 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 3541 | |
dce48a84 TG |
3542 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3543 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3544 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 3545 | |
dce48a84 TG |
3546 | calc_load_update += LOAD_FREQ; |
3547 | } | |
1da177e4 | 3548 | |
dce48a84 | 3549 | /* |
74f5187a PZ |
3550 | * Called from update_cpu_load() to periodically update this CPU's |
3551 | * active count. | |
dce48a84 TG |
3552 | */ |
3553 | static void calc_load_account_active(struct rq *this_rq) | |
3554 | { | |
74f5187a | 3555 | long delta; |
08c183f3 | 3556 | |
74f5187a PZ |
3557 | if (time_before(jiffies, this_rq->calc_load_update)) |
3558 | return; | |
783609c6 | 3559 | |
74f5187a PZ |
3560 | delta = calc_load_fold_active(this_rq); |
3561 | delta += calc_load_fold_idle(); | |
3562 | if (delta) | |
dce48a84 | 3563 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
3564 | |
3565 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
3566 | } |
3567 | ||
fdf3e95d VP |
3568 | /* |
3569 | * The exact cpuload at various idx values, calculated at every tick would be | |
3570 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | |
3571 | * | |
3572 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | |
3573 | * on nth tick when cpu may be busy, then we have: | |
3574 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3575 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | |
3576 | * | |
3577 | * decay_load_missed() below does efficient calculation of | |
3578 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3579 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | |
3580 | * | |
3581 | * The calculation is approximated on a 128 point scale. | |
3582 | * degrade_zero_ticks is the number of ticks after which load at any | |
3583 | * particular idx is approximated to be zero. | |
3584 | * degrade_factor is a precomputed table, a row for each load idx. | |
3585 | * Each column corresponds to degradation factor for a power of two ticks, | |
3586 | * based on 128 point scale. | |
3587 | * Example: | |
3588 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | |
3589 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | |
3590 | * | |
3591 | * With this power of 2 load factors, we can degrade the load n times | |
3592 | * by looking at 1 bits in n and doing as many mult/shift instead of | |
3593 | * n mult/shifts needed by the exact degradation. | |
3594 | */ | |
3595 | #define DEGRADE_SHIFT 7 | |
3596 | static const unsigned char | |
3597 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | |
3598 | static const unsigned char | |
3599 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | |
3600 | {0, 0, 0, 0, 0, 0, 0, 0}, | |
3601 | {64, 32, 8, 0, 0, 0, 0, 0}, | |
3602 | {96, 72, 40, 12, 1, 0, 0}, | |
3603 | {112, 98, 75, 43, 15, 1, 0}, | |
3604 | {120, 112, 98, 76, 45, 16, 2} }; | |
3605 | ||
3606 | /* | |
3607 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | |
3608 | * would be when CPU is idle and so we just decay the old load without | |
3609 | * adding any new load. | |
3610 | */ | |
3611 | static unsigned long | |
3612 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | |
3613 | { | |
3614 | int j = 0; | |
3615 | ||
3616 | if (!missed_updates) | |
3617 | return load; | |
3618 | ||
3619 | if (missed_updates >= degrade_zero_ticks[idx]) | |
3620 | return 0; | |
3621 | ||
3622 | if (idx == 1) | |
3623 | return load >> missed_updates; | |
3624 | ||
3625 | while (missed_updates) { | |
3626 | if (missed_updates % 2) | |
3627 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | |
3628 | ||
3629 | missed_updates >>= 1; | |
3630 | j++; | |
3631 | } | |
3632 | return load; | |
3633 | } | |
3634 | ||
46cb4b7c | 3635 | /* |
dd41f596 | 3636 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
3637 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
3638 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 3639 | */ |
dd41f596 | 3640 | static void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 3641 | { |
495eca49 | 3642 | unsigned long this_load = this_rq->load.weight; |
fdf3e95d VP |
3643 | unsigned long curr_jiffies = jiffies; |
3644 | unsigned long pending_updates; | |
dd41f596 | 3645 | int i, scale; |
46cb4b7c | 3646 | |
dd41f596 | 3647 | this_rq->nr_load_updates++; |
46cb4b7c | 3648 | |
fdf3e95d VP |
3649 | /* Avoid repeated calls on same jiffy, when moving in and out of idle */ |
3650 | if (curr_jiffies == this_rq->last_load_update_tick) | |
3651 | return; | |
3652 | ||
3653 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
3654 | this_rq->last_load_update_tick = curr_jiffies; | |
3655 | ||
dd41f596 | 3656 | /* Update our load: */ |
fdf3e95d VP |
3657 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
3658 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 3659 | unsigned long old_load, new_load; |
7d1e6a9b | 3660 | |
dd41f596 | 3661 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 3662 | |
dd41f596 | 3663 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 3664 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 3665 | new_load = this_load; |
a25707f3 IM |
3666 | /* |
3667 | * Round up the averaging division if load is increasing. This | |
3668 | * prevents us from getting stuck on 9 if the load is 10, for | |
3669 | * example. | |
3670 | */ | |
3671 | if (new_load > old_load) | |
fdf3e95d VP |
3672 | new_load += scale - 1; |
3673 | ||
3674 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 3675 | } |
da2b71ed SS |
3676 | |
3677 | sched_avg_update(this_rq); | |
fdf3e95d VP |
3678 | } |
3679 | ||
3680 | static void update_cpu_load_active(struct rq *this_rq) | |
3681 | { | |
3682 | update_cpu_load(this_rq); | |
46cb4b7c | 3683 | |
74f5187a | 3684 | calc_load_account_active(this_rq); |
46cb4b7c SS |
3685 | } |
3686 | ||
dd41f596 | 3687 | #ifdef CONFIG_SMP |
8a0be9ef | 3688 | |
46cb4b7c | 3689 | /* |
38022906 PZ |
3690 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3691 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3692 | */ |
38022906 | 3693 | void sched_exec(void) |
46cb4b7c | 3694 | { |
38022906 | 3695 | struct task_struct *p = current; |
1da177e4 | 3696 | unsigned long flags; |
0017d735 | 3697 | int dest_cpu; |
46cb4b7c | 3698 | |
8f42ced9 | 3699 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
7608dec2 | 3700 | dest_cpu = p->sched_class->select_task_rq(p, SD_BALANCE_EXEC, 0); |
0017d735 PZ |
3701 | if (dest_cpu == smp_processor_id()) |
3702 | goto unlock; | |
38022906 | 3703 | |
8f42ced9 | 3704 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 3705 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 3706 | |
8f42ced9 PZ |
3707 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
3708 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
3709 | return; |
3710 | } | |
0017d735 | 3711 | unlock: |
8f42ced9 | 3712 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 3713 | } |
dd41f596 | 3714 | |
1da177e4 LT |
3715 | #endif |
3716 | ||
1da177e4 LT |
3717 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3718 | ||
3719 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3720 | ||
3721 | /* | |
c5f8d995 | 3722 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 3723 | * @p in case that task is currently running. |
c5f8d995 HS |
3724 | * |
3725 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 3726 | */ |
c5f8d995 HS |
3727 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
3728 | { | |
3729 | u64 ns = 0; | |
3730 | ||
3731 | if (task_current(rq, p)) { | |
3732 | update_rq_clock(rq); | |
305e6835 | 3733 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
3734 | if ((s64)ns < 0) |
3735 | ns = 0; | |
3736 | } | |
3737 | ||
3738 | return ns; | |
3739 | } | |
3740 | ||
bb34d92f | 3741 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 3742 | { |
1da177e4 | 3743 | unsigned long flags; |
41b86e9c | 3744 | struct rq *rq; |
bb34d92f | 3745 | u64 ns = 0; |
48f24c4d | 3746 | |
41b86e9c | 3747 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 3748 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 3749 | task_rq_unlock(rq, p, &flags); |
1508487e | 3750 | |
c5f8d995 HS |
3751 | return ns; |
3752 | } | |
f06febc9 | 3753 | |
c5f8d995 HS |
3754 | /* |
3755 | * Return accounted runtime for the task. | |
3756 | * In case the task is currently running, return the runtime plus current's | |
3757 | * pending runtime that have not been accounted yet. | |
3758 | */ | |
3759 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3760 | { | |
3761 | unsigned long flags; | |
3762 | struct rq *rq; | |
3763 | u64 ns = 0; | |
3764 | ||
3765 | rq = task_rq_lock(p, &flags); | |
3766 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 3767 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
3768 | |
3769 | return ns; | |
3770 | } | |
48f24c4d | 3771 | |
c5f8d995 HS |
3772 | /* |
3773 | * Return sum_exec_runtime for the thread group. | |
3774 | * In case the task is currently running, return the sum plus current's | |
3775 | * pending runtime that have not been accounted yet. | |
3776 | * | |
3777 | * Note that the thread group might have other running tasks as well, | |
3778 | * so the return value not includes other pending runtime that other | |
3779 | * running tasks might have. | |
3780 | */ | |
3781 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
3782 | { | |
3783 | struct task_cputime totals; | |
3784 | unsigned long flags; | |
3785 | struct rq *rq; | |
3786 | u64 ns; | |
3787 | ||
3788 | rq = task_rq_lock(p, &flags); | |
3789 | thread_group_cputime(p, &totals); | |
3790 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 3791 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 3792 | |
1da177e4 LT |
3793 | return ns; |
3794 | } | |
3795 | ||
1da177e4 LT |
3796 | /* |
3797 | * Account user cpu time to a process. | |
3798 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 3799 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 3800 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 3801 | */ |
457533a7 MS |
3802 | void account_user_time(struct task_struct *p, cputime_t cputime, |
3803 | cputime_t cputime_scaled) | |
1da177e4 LT |
3804 | { |
3805 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3806 | cputime64_t tmp; | |
3807 | ||
457533a7 | 3808 | /* Add user time to process. */ |
1da177e4 | 3809 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3810 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3811 | account_group_user_time(p, cputime); |
1da177e4 LT |
3812 | |
3813 | /* Add user time to cpustat. */ | |
3814 | tmp = cputime_to_cputime64(cputime); | |
3815 | if (TASK_NICE(p) > 0) | |
3816 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3817 | else | |
3818 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
3819 | |
3820 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
3821 | /* Account for user time used */ |
3822 | acct_update_integrals(p); | |
1da177e4 LT |
3823 | } |
3824 | ||
94886b84 LV |
3825 | /* |
3826 | * Account guest cpu time to a process. | |
3827 | * @p: the process that the cpu time gets accounted to | |
3828 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 3829 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 3830 | */ |
457533a7 MS |
3831 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
3832 | cputime_t cputime_scaled) | |
94886b84 LV |
3833 | { |
3834 | cputime64_t tmp; | |
3835 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3836 | ||
3837 | tmp = cputime_to_cputime64(cputime); | |
3838 | ||
457533a7 | 3839 | /* Add guest time to process. */ |
94886b84 | 3840 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3841 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3842 | account_group_user_time(p, cputime); |
94886b84 LV |
3843 | p->gtime = cputime_add(p->gtime, cputime); |
3844 | ||
457533a7 | 3845 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
3846 | if (TASK_NICE(p) > 0) { |
3847 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3848 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
3849 | } else { | |
3850 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3851 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3852 | } | |
94886b84 LV |
3853 | } |
3854 | ||
70a89a66 VP |
3855 | /* |
3856 | * Account system cpu time to a process and desired cpustat field | |
3857 | * @p: the process that the cpu time gets accounted to | |
3858 | * @cputime: the cpu time spent in kernel space since the last update | |
3859 | * @cputime_scaled: cputime scaled by cpu frequency | |
3860 | * @target_cputime64: pointer to cpustat field that has to be updated | |
3861 | */ | |
3862 | static inline | |
3863 | void __account_system_time(struct task_struct *p, cputime_t cputime, | |
3864 | cputime_t cputime_scaled, cputime64_t *target_cputime64) | |
3865 | { | |
3866 | cputime64_t tmp = cputime_to_cputime64(cputime); | |
3867 | ||
3868 | /* Add system time to process. */ | |
3869 | p->stime = cputime_add(p->stime, cputime); | |
3870 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); | |
3871 | account_group_system_time(p, cputime); | |
3872 | ||
3873 | /* Add system time to cpustat. */ | |
3874 | *target_cputime64 = cputime64_add(*target_cputime64, tmp); | |
3875 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); | |
3876 | ||
3877 | /* Account for system time used */ | |
3878 | acct_update_integrals(p); | |
3879 | } | |
3880 | ||
1da177e4 LT |
3881 | /* |
3882 | * Account system cpu time to a process. | |
3883 | * @p: the process that the cpu time gets accounted to | |
3884 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3885 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 3886 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
3887 | */ |
3888 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 3889 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
3890 | { |
3891 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70a89a66 | 3892 | cputime64_t *target_cputime64; |
1da177e4 | 3893 | |
983ed7a6 | 3894 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 3895 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
3896 | return; |
3897 | } | |
94886b84 | 3898 | |
1da177e4 | 3899 | if (hardirq_count() - hardirq_offset) |
70a89a66 | 3900 | target_cputime64 = &cpustat->irq; |
75e1056f | 3901 | else if (in_serving_softirq()) |
70a89a66 | 3902 | target_cputime64 = &cpustat->softirq; |
1da177e4 | 3903 | else |
70a89a66 | 3904 | target_cputime64 = &cpustat->system; |
ef12fefa | 3905 | |
70a89a66 | 3906 | __account_system_time(p, cputime, cputime_scaled, target_cputime64); |
1da177e4 LT |
3907 | } |
3908 | ||
c66f08be | 3909 | /* |
1da177e4 | 3910 | * Account for involuntary wait time. |
544b4a1f | 3911 | * @cputime: the cpu time spent in involuntary wait |
c66f08be | 3912 | */ |
79741dd3 | 3913 | void account_steal_time(cputime_t cputime) |
c66f08be | 3914 | { |
79741dd3 MS |
3915 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
3916 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
3917 | ||
3918 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
3919 | } |
3920 | ||
1da177e4 | 3921 | /* |
79741dd3 MS |
3922 | * Account for idle time. |
3923 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 3924 | */ |
79741dd3 | 3925 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
3926 | { |
3927 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 3928 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 3929 | struct rq *rq = this_rq(); |
1da177e4 | 3930 | |
79741dd3 MS |
3931 | if (atomic_read(&rq->nr_iowait) > 0) |
3932 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
3933 | else | |
3934 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
3935 | } |
3936 | ||
e6e6685a GC |
3937 | static __always_inline bool steal_account_process_tick(void) |
3938 | { | |
3939 | #ifdef CONFIG_PARAVIRT | |
3940 | if (static_branch(¶virt_steal_enabled)) { | |
3941 | u64 steal, st = 0; | |
3942 | ||
3943 | steal = paravirt_steal_clock(smp_processor_id()); | |
3944 | steal -= this_rq()->prev_steal_time; | |
3945 | ||
3946 | st = steal_ticks(steal); | |
3947 | this_rq()->prev_steal_time += st * TICK_NSEC; | |
3948 | ||
3949 | account_steal_time(st); | |
3950 | return st; | |
3951 | } | |
3952 | #endif | |
3953 | return false; | |
3954 | } | |
3955 | ||
79741dd3 MS |
3956 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
3957 | ||
abb74cef VP |
3958 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
3959 | /* | |
3960 | * Account a tick to a process and cpustat | |
3961 | * @p: the process that the cpu time gets accounted to | |
3962 | * @user_tick: is the tick from userspace | |
3963 | * @rq: the pointer to rq | |
3964 | * | |
3965 | * Tick demultiplexing follows the order | |
3966 | * - pending hardirq update | |
3967 | * - pending softirq update | |
3968 | * - user_time | |
3969 | * - idle_time | |
3970 | * - system time | |
3971 | * - check for guest_time | |
3972 | * - else account as system_time | |
3973 | * | |
3974 | * Check for hardirq is done both for system and user time as there is | |
3975 | * no timer going off while we are on hardirq and hence we may never get an | |
3976 | * opportunity to update it solely in system time. | |
3977 | * p->stime and friends are only updated on system time and not on irq | |
3978 | * softirq as those do not count in task exec_runtime any more. | |
3979 | */ | |
3980 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
3981 | struct rq *rq) | |
3982 | { | |
3983 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); | |
3984 | cputime64_t tmp = cputime_to_cputime64(cputime_one_jiffy); | |
3985 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3986 | ||
e6e6685a GC |
3987 | if (steal_account_process_tick()) |
3988 | return; | |
3989 | ||
abb74cef VP |
3990 | if (irqtime_account_hi_update()) { |
3991 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3992 | } else if (irqtime_account_si_update()) { | |
3993 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
414bee9b VP |
3994 | } else if (this_cpu_ksoftirqd() == p) { |
3995 | /* | |
3996 | * ksoftirqd time do not get accounted in cpu_softirq_time. | |
3997 | * So, we have to handle it separately here. | |
3998 | * Also, p->stime needs to be updated for ksoftirqd. | |
3999 | */ | |
4000 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
4001 | &cpustat->softirq); | |
abb74cef VP |
4002 | } else if (user_tick) { |
4003 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
4004 | } else if (p == rq->idle) { | |
4005 | account_idle_time(cputime_one_jiffy); | |
4006 | } else if (p->flags & PF_VCPU) { /* System time or guest time */ | |
4007 | account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
4008 | } else { | |
4009 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
4010 | &cpustat->system); | |
4011 | } | |
4012 | } | |
4013 | ||
4014 | static void irqtime_account_idle_ticks(int ticks) | |
4015 | { | |
4016 | int i; | |
4017 | struct rq *rq = this_rq(); | |
4018 | ||
4019 | for (i = 0; i < ticks; i++) | |
4020 | irqtime_account_process_tick(current, 0, rq); | |
4021 | } | |
544b4a1f | 4022 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
abb74cef VP |
4023 | static void irqtime_account_idle_ticks(int ticks) {} |
4024 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
4025 | struct rq *rq) {} | |
544b4a1f | 4026 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
79741dd3 MS |
4027 | |
4028 | /* | |
4029 | * Account a single tick of cpu time. | |
4030 | * @p: the process that the cpu time gets accounted to | |
4031 | * @user_tick: indicates if the tick is a user or a system tick | |
4032 | */ | |
4033 | void account_process_tick(struct task_struct *p, int user_tick) | |
4034 | { | |
a42548a1 | 4035 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
4036 | struct rq *rq = this_rq(); |
4037 | ||
abb74cef VP |
4038 | if (sched_clock_irqtime) { |
4039 | irqtime_account_process_tick(p, user_tick, rq); | |
4040 | return; | |
4041 | } | |
4042 | ||
e6e6685a GC |
4043 | if (steal_account_process_tick()) |
4044 | return; | |
4045 | ||
79741dd3 | 4046 | if (user_tick) |
a42548a1 | 4047 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 4048 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 4049 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
4050 | one_jiffy_scaled); |
4051 | else | |
a42548a1 | 4052 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
4053 | } |
4054 | ||
4055 | /* | |
4056 | * Account multiple ticks of steal time. | |
4057 | * @p: the process from which the cpu time has been stolen | |
4058 | * @ticks: number of stolen ticks | |
4059 | */ | |
4060 | void account_steal_ticks(unsigned long ticks) | |
4061 | { | |
4062 | account_steal_time(jiffies_to_cputime(ticks)); | |
4063 | } | |
4064 | ||
4065 | /* | |
4066 | * Account multiple ticks of idle time. | |
4067 | * @ticks: number of stolen ticks | |
4068 | */ | |
4069 | void account_idle_ticks(unsigned long ticks) | |
4070 | { | |
abb74cef VP |
4071 | |
4072 | if (sched_clock_irqtime) { | |
4073 | irqtime_account_idle_ticks(ticks); | |
4074 | return; | |
4075 | } | |
4076 | ||
79741dd3 | 4077 | account_idle_time(jiffies_to_cputime(ticks)); |
1da177e4 LT |
4078 | } |
4079 | ||
79741dd3 MS |
4080 | #endif |
4081 | ||
49048622 BS |
4082 | /* |
4083 | * Use precise platform statistics if available: | |
4084 | */ | |
4085 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 4086 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 4087 | { |
d99ca3b9 HS |
4088 | *ut = p->utime; |
4089 | *st = p->stime; | |
49048622 BS |
4090 | } |
4091 | ||
0cf55e1e | 4092 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 4093 | { |
0cf55e1e HS |
4094 | struct task_cputime cputime; |
4095 | ||
4096 | thread_group_cputime(p, &cputime); | |
4097 | ||
4098 | *ut = cputime.utime; | |
4099 | *st = cputime.stime; | |
49048622 BS |
4100 | } |
4101 | #else | |
761b1d26 HS |
4102 | |
4103 | #ifndef nsecs_to_cputime | |
b7b20df9 | 4104 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
4105 | #endif |
4106 | ||
d180c5bc | 4107 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 4108 | { |
d99ca3b9 | 4109 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
4110 | |
4111 | /* | |
4112 | * Use CFS's precise accounting: | |
4113 | */ | |
d180c5bc | 4114 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
4115 | |
4116 | if (total) { | |
e75e863d | 4117 | u64 temp = rtime; |
d180c5bc | 4118 | |
e75e863d | 4119 | temp *= utime; |
49048622 | 4120 | do_div(temp, total); |
d180c5bc HS |
4121 | utime = (cputime_t)temp; |
4122 | } else | |
4123 | utime = rtime; | |
49048622 | 4124 | |
d180c5bc HS |
4125 | /* |
4126 | * Compare with previous values, to keep monotonicity: | |
4127 | */ | |
761b1d26 | 4128 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 4129 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 4130 | |
d99ca3b9 HS |
4131 | *ut = p->prev_utime; |
4132 | *st = p->prev_stime; | |
49048622 BS |
4133 | } |
4134 | ||
0cf55e1e HS |
4135 | /* |
4136 | * Must be called with siglock held. | |
4137 | */ | |
4138 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 4139 | { |
0cf55e1e HS |
4140 | struct signal_struct *sig = p->signal; |
4141 | struct task_cputime cputime; | |
4142 | cputime_t rtime, utime, total; | |
49048622 | 4143 | |
0cf55e1e | 4144 | thread_group_cputime(p, &cputime); |
49048622 | 4145 | |
0cf55e1e HS |
4146 | total = cputime_add(cputime.utime, cputime.stime); |
4147 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 4148 | |
0cf55e1e | 4149 | if (total) { |
e75e863d | 4150 | u64 temp = rtime; |
49048622 | 4151 | |
e75e863d | 4152 | temp *= cputime.utime; |
0cf55e1e HS |
4153 | do_div(temp, total); |
4154 | utime = (cputime_t)temp; | |
4155 | } else | |
4156 | utime = rtime; | |
4157 | ||
4158 | sig->prev_utime = max(sig->prev_utime, utime); | |
4159 | sig->prev_stime = max(sig->prev_stime, | |
4160 | cputime_sub(rtime, sig->prev_utime)); | |
4161 | ||
4162 | *ut = sig->prev_utime; | |
4163 | *st = sig->prev_stime; | |
49048622 | 4164 | } |
49048622 | 4165 | #endif |
49048622 | 4166 | |
7835b98b CL |
4167 | /* |
4168 | * This function gets called by the timer code, with HZ frequency. | |
4169 | * We call it with interrupts disabled. | |
7835b98b CL |
4170 | */ |
4171 | void scheduler_tick(void) | |
4172 | { | |
7835b98b CL |
4173 | int cpu = smp_processor_id(); |
4174 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4175 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4176 | |
4177 | sched_clock_tick(); | |
dd41f596 | 4178 | |
05fa785c | 4179 | raw_spin_lock(&rq->lock); |
3e51f33f | 4180 | update_rq_clock(rq); |
fdf3e95d | 4181 | update_cpu_load_active(rq); |
fa85ae24 | 4182 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 4183 | raw_spin_unlock(&rq->lock); |
7835b98b | 4184 | |
e9d2b064 | 4185 | perf_event_task_tick(); |
e220d2dc | 4186 | |
e418e1c2 | 4187 | #ifdef CONFIG_SMP |
dd41f596 IM |
4188 | rq->idle_at_tick = idle_cpu(cpu); |
4189 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4190 | #endif |
1da177e4 LT |
4191 | } |
4192 | ||
132380a0 | 4193 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
4194 | { |
4195 | if (in_lock_functions(addr)) { | |
4196 | addr = CALLER_ADDR2; | |
4197 | if (in_lock_functions(addr)) | |
4198 | addr = CALLER_ADDR3; | |
4199 | } | |
4200 | return addr; | |
4201 | } | |
1da177e4 | 4202 | |
7e49fcce SR |
4203 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
4204 | defined(CONFIG_PREEMPT_TRACER)) | |
4205 | ||
43627582 | 4206 | void __kprobes add_preempt_count(int val) |
1da177e4 | 4207 | { |
6cd8a4bb | 4208 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4209 | /* |
4210 | * Underflow? | |
4211 | */ | |
9a11b49a IM |
4212 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4213 | return; | |
6cd8a4bb | 4214 | #endif |
1da177e4 | 4215 | preempt_count() += val; |
6cd8a4bb | 4216 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4217 | /* |
4218 | * Spinlock count overflowing soon? | |
4219 | */ | |
33859f7f MOS |
4220 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4221 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
4222 | #endif |
4223 | if (preempt_count() == val) | |
4224 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4225 | } |
4226 | EXPORT_SYMBOL(add_preempt_count); | |
4227 | ||
43627582 | 4228 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 4229 | { |
6cd8a4bb | 4230 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4231 | /* |
4232 | * Underflow? | |
4233 | */ | |
01e3eb82 | 4234 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 4235 | return; |
1da177e4 LT |
4236 | /* |
4237 | * Is the spinlock portion underflowing? | |
4238 | */ | |
9a11b49a IM |
4239 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4240 | !(preempt_count() & PREEMPT_MASK))) | |
4241 | return; | |
6cd8a4bb | 4242 | #endif |
9a11b49a | 4243 | |
6cd8a4bb SR |
4244 | if (preempt_count() == val) |
4245 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4246 | preempt_count() -= val; |
4247 | } | |
4248 | EXPORT_SYMBOL(sub_preempt_count); | |
4249 | ||
4250 | #endif | |
4251 | ||
4252 | /* | |
dd41f596 | 4253 | * Print scheduling while atomic bug: |
1da177e4 | 4254 | */ |
dd41f596 | 4255 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4256 | { |
838225b4 SS |
4257 | struct pt_regs *regs = get_irq_regs(); |
4258 | ||
3df0fc5b PZ |
4259 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
4260 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 4261 | |
dd41f596 | 4262 | debug_show_held_locks(prev); |
e21f5b15 | 4263 | print_modules(); |
dd41f596 IM |
4264 | if (irqs_disabled()) |
4265 | print_irqtrace_events(prev); | |
838225b4 SS |
4266 | |
4267 | if (regs) | |
4268 | show_regs(regs); | |
4269 | else | |
4270 | dump_stack(); | |
dd41f596 | 4271 | } |
1da177e4 | 4272 | |
dd41f596 IM |
4273 | /* |
4274 | * Various schedule()-time debugging checks and statistics: | |
4275 | */ | |
4276 | static inline void schedule_debug(struct task_struct *prev) | |
4277 | { | |
1da177e4 | 4278 | /* |
41a2d6cf | 4279 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4280 | * schedule() atomically, we ignore that path for now. |
4281 | * Otherwise, whine if we are scheduling when we should not be. | |
4282 | */ | |
3f33a7ce | 4283 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4284 | __schedule_bug(prev); |
4285 | ||
1da177e4 LT |
4286 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4287 | ||
2d72376b | 4288 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
4289 | } |
4290 | ||
6cecd084 | 4291 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 4292 | { |
61eadef6 | 4293 | if (prev->on_rq || rq->skip_clock_update < 0) |
a64692a3 | 4294 | update_rq_clock(rq); |
6cecd084 | 4295 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
4296 | } |
4297 | ||
dd41f596 IM |
4298 | /* |
4299 | * Pick up the highest-prio task: | |
4300 | */ | |
4301 | static inline struct task_struct * | |
b67802ea | 4302 | pick_next_task(struct rq *rq) |
dd41f596 | 4303 | { |
5522d5d5 | 4304 | const struct sched_class *class; |
dd41f596 | 4305 | struct task_struct *p; |
1da177e4 LT |
4306 | |
4307 | /* | |
dd41f596 IM |
4308 | * Optimization: we know that if all tasks are in |
4309 | * the fair class we can call that function directly: | |
1da177e4 | 4310 | */ |
953bfcd1 | 4311 | if (likely(rq->nr_running == rq->cfs.h_nr_running)) { |
fb8d4724 | 4312 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4313 | if (likely(p)) |
4314 | return p; | |
1da177e4 LT |
4315 | } |
4316 | ||
34f971f6 | 4317 | for_each_class(class) { |
fb8d4724 | 4318 | p = class->pick_next_task(rq); |
dd41f596 IM |
4319 | if (p) |
4320 | return p; | |
dd41f596 | 4321 | } |
34f971f6 PZ |
4322 | |
4323 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 4324 | } |
1da177e4 | 4325 | |
dd41f596 IM |
4326 | /* |
4327 | * schedule() is the main scheduler function. | |
4328 | */ | |
ff743345 | 4329 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
4330 | { |
4331 | struct task_struct *prev, *next; | |
67ca7bde | 4332 | unsigned long *switch_count; |
dd41f596 | 4333 | struct rq *rq; |
31656519 | 4334 | int cpu; |
dd41f596 | 4335 | |
ff743345 PZ |
4336 | need_resched: |
4337 | preempt_disable(); | |
dd41f596 IM |
4338 | cpu = smp_processor_id(); |
4339 | rq = cpu_rq(cpu); | |
25502a6c | 4340 | rcu_note_context_switch(cpu); |
dd41f596 | 4341 | prev = rq->curr; |
dd41f596 | 4342 | |
dd41f596 | 4343 | schedule_debug(prev); |
1da177e4 | 4344 | |
31656519 | 4345 | if (sched_feat(HRTICK)) |
f333fdc9 | 4346 | hrtick_clear(rq); |
8f4d37ec | 4347 | |
05fa785c | 4348 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 4349 | |
246d86b5 | 4350 | switch_count = &prev->nivcsw; |
1da177e4 | 4351 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 4352 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 4353 | prev->state = TASK_RUNNING; |
21aa9af0 | 4354 | } else { |
2acca55e PZ |
4355 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
4356 | prev->on_rq = 0; | |
4357 | ||
21aa9af0 | 4358 | /* |
2acca55e PZ |
4359 | * If a worker went to sleep, notify and ask workqueue |
4360 | * whether it wants to wake up a task to maintain | |
4361 | * concurrency. | |
21aa9af0 TH |
4362 | */ |
4363 | if (prev->flags & PF_WQ_WORKER) { | |
4364 | struct task_struct *to_wakeup; | |
4365 | ||
4366 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
4367 | if (to_wakeup) | |
4368 | try_to_wake_up_local(to_wakeup); | |
4369 | } | |
fd2f4419 | 4370 | |
6631e635 | 4371 | /* |
2acca55e PZ |
4372 | * If we are going to sleep and we have plugged IO |
4373 | * queued, make sure to submit it to avoid deadlocks. | |
6631e635 LT |
4374 | */ |
4375 | if (blk_needs_flush_plug(prev)) { | |
4376 | raw_spin_unlock(&rq->lock); | |
a237c1c5 | 4377 | blk_schedule_flush_plug(prev); |
6631e635 LT |
4378 | raw_spin_lock(&rq->lock); |
4379 | } | |
21aa9af0 | 4380 | } |
dd41f596 | 4381 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4382 | } |
4383 | ||
3f029d3c | 4384 | pre_schedule(rq, prev); |
f65eda4f | 4385 | |
dd41f596 | 4386 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4387 | idle_balance(cpu, rq); |
1da177e4 | 4388 | |
df1c99d4 | 4389 | put_prev_task(rq, prev); |
b67802ea | 4390 | next = pick_next_task(rq); |
f26f9aff MG |
4391 | clear_tsk_need_resched(prev); |
4392 | rq->skip_clock_update = 0; | |
1da177e4 | 4393 | |
1da177e4 | 4394 | if (likely(prev != next)) { |
1da177e4 LT |
4395 | rq->nr_switches++; |
4396 | rq->curr = next; | |
4397 | ++*switch_count; | |
4398 | ||
dd41f596 | 4399 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 4400 | /* |
246d86b5 ON |
4401 | * The context switch have flipped the stack from under us |
4402 | * and restored the local variables which were saved when | |
4403 | * this task called schedule() in the past. prev == current | |
4404 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
4405 | */ |
4406 | cpu = smp_processor_id(); | |
4407 | rq = cpu_rq(cpu); | |
1da177e4 | 4408 | } else |
05fa785c | 4409 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 4410 | |
3f029d3c | 4411 | post_schedule(rq); |
1da177e4 | 4412 | |
1da177e4 | 4413 | preempt_enable_no_resched(); |
ff743345 | 4414 | if (need_resched()) |
1da177e4 LT |
4415 | goto need_resched; |
4416 | } | |
1da177e4 LT |
4417 | EXPORT_SYMBOL(schedule); |
4418 | ||
c08f7829 | 4419 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d | 4420 | |
c6eb3dda PZ |
4421 | static inline bool owner_running(struct mutex *lock, struct task_struct *owner) |
4422 | { | |
c6eb3dda | 4423 | if (lock->owner != owner) |
307bf980 | 4424 | return false; |
0d66bf6d PZ |
4425 | |
4426 | /* | |
c6eb3dda PZ |
4427 | * Ensure we emit the owner->on_cpu, dereference _after_ checking |
4428 | * lock->owner still matches owner, if that fails, owner might | |
4429 | * point to free()d memory, if it still matches, the rcu_read_lock() | |
4430 | * ensures the memory stays valid. | |
0d66bf6d | 4431 | */ |
c6eb3dda | 4432 | barrier(); |
0d66bf6d | 4433 | |
307bf980 | 4434 | return owner->on_cpu; |
c6eb3dda | 4435 | } |
0d66bf6d | 4436 | |
c6eb3dda PZ |
4437 | /* |
4438 | * Look out! "owner" is an entirely speculative pointer | |
4439 | * access and not reliable. | |
4440 | */ | |
4441 | int mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner) | |
4442 | { | |
4443 | if (!sched_feat(OWNER_SPIN)) | |
4444 | return 0; | |
0d66bf6d | 4445 | |
307bf980 | 4446 | rcu_read_lock(); |
c6eb3dda PZ |
4447 | while (owner_running(lock, owner)) { |
4448 | if (need_resched()) | |
307bf980 | 4449 | break; |
0d66bf6d | 4450 | |
335d7afb | 4451 | arch_mutex_cpu_relax(); |
0d66bf6d | 4452 | } |
307bf980 | 4453 | rcu_read_unlock(); |
4b402210 | 4454 | |
c6eb3dda | 4455 | /* |
307bf980 TG |
4456 | * We break out the loop above on need_resched() and when the |
4457 | * owner changed, which is a sign for heavy contention. Return | |
4458 | * success only when lock->owner is NULL. | |
c6eb3dda | 4459 | */ |
307bf980 | 4460 | return lock->owner == NULL; |
0d66bf6d PZ |
4461 | } |
4462 | #endif | |
4463 | ||
1da177e4 LT |
4464 | #ifdef CONFIG_PREEMPT |
4465 | /* | |
2ed6e34f | 4466 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4467 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4468 | * occur there and call schedule directly. |
4469 | */ | |
d1f74e20 | 4470 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
4471 | { |
4472 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4473 | |
1da177e4 LT |
4474 | /* |
4475 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4476 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4477 | */ |
beed33a8 | 4478 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4479 | return; |
4480 | ||
3a5c359a | 4481 | do { |
d1f74e20 | 4482 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
3a5c359a | 4483 | schedule(); |
d1f74e20 | 4484 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 4485 | |
3a5c359a AK |
4486 | /* |
4487 | * Check again in case we missed a preemption opportunity | |
4488 | * between schedule and now. | |
4489 | */ | |
4490 | barrier(); | |
5ed0cec0 | 4491 | } while (need_resched()); |
1da177e4 | 4492 | } |
1da177e4 LT |
4493 | EXPORT_SYMBOL(preempt_schedule); |
4494 | ||
4495 | /* | |
2ed6e34f | 4496 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4497 | * off of irq context. |
4498 | * Note, that this is called and return with irqs disabled. This will | |
4499 | * protect us against recursive calling from irq. | |
4500 | */ | |
4501 | asmlinkage void __sched preempt_schedule_irq(void) | |
4502 | { | |
4503 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4504 | |
2ed6e34f | 4505 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4506 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4507 | ||
3a5c359a AK |
4508 | do { |
4509 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4510 | local_irq_enable(); |
4511 | schedule(); | |
4512 | local_irq_disable(); | |
3a5c359a | 4513 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4514 | |
3a5c359a AK |
4515 | /* |
4516 | * Check again in case we missed a preemption opportunity | |
4517 | * between schedule and now. | |
4518 | */ | |
4519 | barrier(); | |
5ed0cec0 | 4520 | } while (need_resched()); |
1da177e4 LT |
4521 | } |
4522 | ||
4523 | #endif /* CONFIG_PREEMPT */ | |
4524 | ||
63859d4f | 4525 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 4526 | void *key) |
1da177e4 | 4527 | { |
63859d4f | 4528 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 4529 | } |
1da177e4 LT |
4530 | EXPORT_SYMBOL(default_wake_function); |
4531 | ||
4532 | /* | |
41a2d6cf IM |
4533 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4534 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4535 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4536 | * | |
4537 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4538 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4539 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4540 | */ | |
78ddb08f | 4541 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 4542 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 4543 | { |
2e45874c | 4544 | wait_queue_t *curr, *next; |
1da177e4 | 4545 | |
2e45874c | 4546 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4547 | unsigned flags = curr->flags; |
4548 | ||
63859d4f | 4549 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 4550 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4551 | break; |
4552 | } | |
4553 | } | |
4554 | ||
4555 | /** | |
4556 | * __wake_up - wake up threads blocked on a waitqueue. | |
4557 | * @q: the waitqueue | |
4558 | * @mode: which threads | |
4559 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4560 | * @key: is directly passed to the wakeup function |
50fa610a DH |
4561 | * |
4562 | * It may be assumed that this function implies a write memory barrier before | |
4563 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4564 | */ |
7ad5b3a5 | 4565 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4566 | int nr_exclusive, void *key) |
1da177e4 LT |
4567 | { |
4568 | unsigned long flags; | |
4569 | ||
4570 | spin_lock_irqsave(&q->lock, flags); | |
4571 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4572 | spin_unlock_irqrestore(&q->lock, flags); | |
4573 | } | |
1da177e4 LT |
4574 | EXPORT_SYMBOL(__wake_up); |
4575 | ||
4576 | /* | |
4577 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4578 | */ | |
7ad5b3a5 | 4579 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4580 | { |
4581 | __wake_up_common(q, mode, 1, 0, NULL); | |
4582 | } | |
22c43c81 | 4583 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 4584 | |
4ede816a DL |
4585 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
4586 | { | |
4587 | __wake_up_common(q, mode, 1, 0, key); | |
4588 | } | |
bf294b41 | 4589 | EXPORT_SYMBOL_GPL(__wake_up_locked_key); |
4ede816a | 4590 | |
1da177e4 | 4591 | /** |
4ede816a | 4592 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4593 | * @q: the waitqueue |
4594 | * @mode: which threads | |
4595 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 4596 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
4597 | * |
4598 | * The sync wakeup differs that the waker knows that it will schedule | |
4599 | * away soon, so while the target thread will be woken up, it will not | |
4600 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4601 | * with each other. This can prevent needless bouncing between CPUs. | |
4602 | * | |
4603 | * On UP it can prevent extra preemption. | |
50fa610a DH |
4604 | * |
4605 | * It may be assumed that this function implies a write memory barrier before | |
4606 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4607 | */ |
4ede816a DL |
4608 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
4609 | int nr_exclusive, void *key) | |
1da177e4 LT |
4610 | { |
4611 | unsigned long flags; | |
7d478721 | 4612 | int wake_flags = WF_SYNC; |
1da177e4 LT |
4613 | |
4614 | if (unlikely(!q)) | |
4615 | return; | |
4616 | ||
4617 | if (unlikely(!nr_exclusive)) | |
7d478721 | 4618 | wake_flags = 0; |
1da177e4 LT |
4619 | |
4620 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 4621 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
4622 | spin_unlock_irqrestore(&q->lock, flags); |
4623 | } | |
4ede816a DL |
4624 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
4625 | ||
4626 | /* | |
4627 | * __wake_up_sync - see __wake_up_sync_key() | |
4628 | */ | |
4629 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
4630 | { | |
4631 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
4632 | } | |
1da177e4 LT |
4633 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
4634 | ||
65eb3dc6 KD |
4635 | /** |
4636 | * complete: - signals a single thread waiting on this completion | |
4637 | * @x: holds the state of this particular completion | |
4638 | * | |
4639 | * This will wake up a single thread waiting on this completion. Threads will be | |
4640 | * awakened in the same order in which they were queued. | |
4641 | * | |
4642 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
4643 | * |
4644 | * It may be assumed that this function implies a write memory barrier before | |
4645 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4646 | */ |
b15136e9 | 4647 | void complete(struct completion *x) |
1da177e4 LT |
4648 | { |
4649 | unsigned long flags; | |
4650 | ||
4651 | spin_lock_irqsave(&x->wait.lock, flags); | |
4652 | x->done++; | |
d9514f6c | 4653 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4654 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4655 | } | |
4656 | EXPORT_SYMBOL(complete); | |
4657 | ||
65eb3dc6 KD |
4658 | /** |
4659 | * complete_all: - signals all threads waiting on this completion | |
4660 | * @x: holds the state of this particular completion | |
4661 | * | |
4662 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
4663 | * |
4664 | * It may be assumed that this function implies a write memory barrier before | |
4665 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4666 | */ |
b15136e9 | 4667 | void complete_all(struct completion *x) |
1da177e4 LT |
4668 | { |
4669 | unsigned long flags; | |
4670 | ||
4671 | spin_lock_irqsave(&x->wait.lock, flags); | |
4672 | x->done += UINT_MAX/2; | |
d9514f6c | 4673 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4674 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4675 | } | |
4676 | EXPORT_SYMBOL(complete_all); | |
4677 | ||
8cbbe86d AK |
4678 | static inline long __sched |
4679 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4680 | { |
1da177e4 LT |
4681 | if (!x->done) { |
4682 | DECLARE_WAITQUEUE(wait, current); | |
4683 | ||
a93d2f17 | 4684 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 4685 | do { |
94d3d824 | 4686 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
4687 | timeout = -ERESTARTSYS; |
4688 | break; | |
8cbbe86d AK |
4689 | } |
4690 | __set_current_state(state); | |
1da177e4 LT |
4691 | spin_unlock_irq(&x->wait.lock); |
4692 | timeout = schedule_timeout(timeout); | |
4693 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4694 | } while (!x->done && timeout); |
1da177e4 | 4695 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4696 | if (!x->done) |
4697 | return timeout; | |
1da177e4 LT |
4698 | } |
4699 | x->done--; | |
ea71a546 | 4700 | return timeout ?: 1; |
1da177e4 | 4701 | } |
1da177e4 | 4702 | |
8cbbe86d AK |
4703 | static long __sched |
4704 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4705 | { |
1da177e4 LT |
4706 | might_sleep(); |
4707 | ||
4708 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4709 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4710 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4711 | return timeout; |
4712 | } | |
1da177e4 | 4713 | |
65eb3dc6 KD |
4714 | /** |
4715 | * wait_for_completion: - waits for completion of a task | |
4716 | * @x: holds the state of this particular completion | |
4717 | * | |
4718 | * This waits to be signaled for completion of a specific task. It is NOT | |
4719 | * interruptible and there is no timeout. | |
4720 | * | |
4721 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4722 | * and interrupt capability. Also see complete(). | |
4723 | */ | |
b15136e9 | 4724 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4725 | { |
4726 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4727 | } |
8cbbe86d | 4728 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4729 | |
65eb3dc6 KD |
4730 | /** |
4731 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4732 | * @x: holds the state of this particular completion | |
4733 | * @timeout: timeout value in jiffies | |
4734 | * | |
4735 | * This waits for either a completion of a specific task to be signaled or for a | |
4736 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4737 | * interruptible. | |
4738 | */ | |
b15136e9 | 4739 | unsigned long __sched |
8cbbe86d | 4740 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4741 | { |
8cbbe86d | 4742 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4743 | } |
8cbbe86d | 4744 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4745 | |
65eb3dc6 KD |
4746 | /** |
4747 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4748 | * @x: holds the state of this particular completion | |
4749 | * | |
4750 | * This waits for completion of a specific task to be signaled. It is | |
4751 | * interruptible. | |
4752 | */ | |
8cbbe86d | 4753 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4754 | { |
51e97990 AK |
4755 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4756 | if (t == -ERESTARTSYS) | |
4757 | return t; | |
4758 | return 0; | |
0fec171c | 4759 | } |
8cbbe86d | 4760 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4761 | |
65eb3dc6 KD |
4762 | /** |
4763 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4764 | * @x: holds the state of this particular completion | |
4765 | * @timeout: timeout value in jiffies | |
4766 | * | |
4767 | * This waits for either a completion of a specific task to be signaled or for a | |
4768 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4769 | */ | |
6bf41237 | 4770 | long __sched |
8cbbe86d AK |
4771 | wait_for_completion_interruptible_timeout(struct completion *x, |
4772 | unsigned long timeout) | |
0fec171c | 4773 | { |
8cbbe86d | 4774 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4775 | } |
8cbbe86d | 4776 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4777 | |
65eb3dc6 KD |
4778 | /** |
4779 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4780 | * @x: holds the state of this particular completion | |
4781 | * | |
4782 | * This waits to be signaled for completion of a specific task. It can be | |
4783 | * interrupted by a kill signal. | |
4784 | */ | |
009e577e MW |
4785 | int __sched wait_for_completion_killable(struct completion *x) |
4786 | { | |
4787 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4788 | if (t == -ERESTARTSYS) | |
4789 | return t; | |
4790 | return 0; | |
4791 | } | |
4792 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4793 | ||
0aa12fb4 SW |
4794 | /** |
4795 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
4796 | * @x: holds the state of this particular completion | |
4797 | * @timeout: timeout value in jiffies | |
4798 | * | |
4799 | * This waits for either a completion of a specific task to be | |
4800 | * signaled or for a specified timeout to expire. It can be | |
4801 | * interrupted by a kill signal. The timeout is in jiffies. | |
4802 | */ | |
6bf41237 | 4803 | long __sched |
0aa12fb4 SW |
4804 | wait_for_completion_killable_timeout(struct completion *x, |
4805 | unsigned long timeout) | |
4806 | { | |
4807 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
4808 | } | |
4809 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
4810 | ||
be4de352 DC |
4811 | /** |
4812 | * try_wait_for_completion - try to decrement a completion without blocking | |
4813 | * @x: completion structure | |
4814 | * | |
4815 | * Returns: 0 if a decrement cannot be done without blocking | |
4816 | * 1 if a decrement succeeded. | |
4817 | * | |
4818 | * If a completion is being used as a counting completion, | |
4819 | * attempt to decrement the counter without blocking. This | |
4820 | * enables us to avoid waiting if the resource the completion | |
4821 | * is protecting is not available. | |
4822 | */ | |
4823 | bool try_wait_for_completion(struct completion *x) | |
4824 | { | |
7539a3b3 | 4825 | unsigned long flags; |
be4de352 DC |
4826 | int ret = 1; |
4827 | ||
7539a3b3 | 4828 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4829 | if (!x->done) |
4830 | ret = 0; | |
4831 | else | |
4832 | x->done--; | |
7539a3b3 | 4833 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4834 | return ret; |
4835 | } | |
4836 | EXPORT_SYMBOL(try_wait_for_completion); | |
4837 | ||
4838 | /** | |
4839 | * completion_done - Test to see if a completion has any waiters | |
4840 | * @x: completion structure | |
4841 | * | |
4842 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4843 | * 1 if there are no waiters. | |
4844 | * | |
4845 | */ | |
4846 | bool completion_done(struct completion *x) | |
4847 | { | |
7539a3b3 | 4848 | unsigned long flags; |
be4de352 DC |
4849 | int ret = 1; |
4850 | ||
7539a3b3 | 4851 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4852 | if (!x->done) |
4853 | ret = 0; | |
7539a3b3 | 4854 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4855 | return ret; |
4856 | } | |
4857 | EXPORT_SYMBOL(completion_done); | |
4858 | ||
8cbbe86d AK |
4859 | static long __sched |
4860 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4861 | { |
0fec171c IM |
4862 | unsigned long flags; |
4863 | wait_queue_t wait; | |
4864 | ||
4865 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4866 | |
8cbbe86d | 4867 | __set_current_state(state); |
1da177e4 | 4868 | |
8cbbe86d AK |
4869 | spin_lock_irqsave(&q->lock, flags); |
4870 | __add_wait_queue(q, &wait); | |
4871 | spin_unlock(&q->lock); | |
4872 | timeout = schedule_timeout(timeout); | |
4873 | spin_lock_irq(&q->lock); | |
4874 | __remove_wait_queue(q, &wait); | |
4875 | spin_unlock_irqrestore(&q->lock, flags); | |
4876 | ||
4877 | return timeout; | |
4878 | } | |
4879 | ||
4880 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4881 | { | |
4882 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4883 | } |
1da177e4 LT |
4884 | EXPORT_SYMBOL(interruptible_sleep_on); |
4885 | ||
0fec171c | 4886 | long __sched |
95cdf3b7 | 4887 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4888 | { |
8cbbe86d | 4889 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4890 | } |
1da177e4 LT |
4891 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4892 | ||
0fec171c | 4893 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4894 | { |
8cbbe86d | 4895 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4896 | } |
1da177e4 LT |
4897 | EXPORT_SYMBOL(sleep_on); |
4898 | ||
0fec171c | 4899 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4900 | { |
8cbbe86d | 4901 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4902 | } |
1da177e4 LT |
4903 | EXPORT_SYMBOL(sleep_on_timeout); |
4904 | ||
b29739f9 IM |
4905 | #ifdef CONFIG_RT_MUTEXES |
4906 | ||
4907 | /* | |
4908 | * rt_mutex_setprio - set the current priority of a task | |
4909 | * @p: task | |
4910 | * @prio: prio value (kernel-internal form) | |
4911 | * | |
4912 | * This function changes the 'effective' priority of a task. It does | |
4913 | * not touch ->normal_prio like __setscheduler(). | |
4914 | * | |
4915 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4916 | */ | |
36c8b586 | 4917 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 4918 | { |
83b699ed | 4919 | int oldprio, on_rq, running; |
70b97a7f | 4920 | struct rq *rq; |
83ab0aa0 | 4921 | const struct sched_class *prev_class; |
b29739f9 IM |
4922 | |
4923 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4924 | ||
0122ec5b | 4925 | rq = __task_rq_lock(p); |
b29739f9 | 4926 | |
a8027073 | 4927 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 4928 | oldprio = p->prio; |
83ab0aa0 | 4929 | prev_class = p->sched_class; |
fd2f4419 | 4930 | on_rq = p->on_rq; |
051a1d1a | 4931 | running = task_current(rq, p); |
0e1f3483 | 4932 | if (on_rq) |
69be72c1 | 4933 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4934 | if (running) |
4935 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4936 | |
4937 | if (rt_prio(prio)) | |
4938 | p->sched_class = &rt_sched_class; | |
4939 | else | |
4940 | p->sched_class = &fair_sched_class; | |
4941 | ||
b29739f9 IM |
4942 | p->prio = prio; |
4943 | ||
0e1f3483 HS |
4944 | if (running) |
4945 | p->sched_class->set_curr_task(rq); | |
da7a735e | 4946 | if (on_rq) |
371fd7e7 | 4947 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 | 4948 | |
da7a735e | 4949 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 4950 | __task_rq_unlock(rq); |
b29739f9 IM |
4951 | } |
4952 | ||
4953 | #endif | |
4954 | ||
36c8b586 | 4955 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4956 | { |
dd41f596 | 4957 | int old_prio, delta, on_rq; |
1da177e4 | 4958 | unsigned long flags; |
70b97a7f | 4959 | struct rq *rq; |
1da177e4 LT |
4960 | |
4961 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4962 | return; | |
4963 | /* | |
4964 | * We have to be careful, if called from sys_setpriority(), | |
4965 | * the task might be in the middle of scheduling on another CPU. | |
4966 | */ | |
4967 | rq = task_rq_lock(p, &flags); | |
4968 | /* | |
4969 | * The RT priorities are set via sched_setscheduler(), but we still | |
4970 | * allow the 'normal' nice value to be set - but as expected | |
4971 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4972 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4973 | */ |
e05606d3 | 4974 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4975 | p->static_prio = NICE_TO_PRIO(nice); |
4976 | goto out_unlock; | |
4977 | } | |
fd2f4419 | 4978 | on_rq = p->on_rq; |
c09595f6 | 4979 | if (on_rq) |
69be72c1 | 4980 | dequeue_task(rq, p, 0); |
1da177e4 | 4981 | |
1da177e4 | 4982 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4983 | set_load_weight(p); |
b29739f9 IM |
4984 | old_prio = p->prio; |
4985 | p->prio = effective_prio(p); | |
4986 | delta = p->prio - old_prio; | |
1da177e4 | 4987 | |
dd41f596 | 4988 | if (on_rq) { |
371fd7e7 | 4989 | enqueue_task(rq, p, 0); |
1da177e4 | 4990 | /* |
d5f9f942 AM |
4991 | * If the task increased its priority or is running and |
4992 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4993 | */ |
d5f9f942 | 4994 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4995 | resched_task(rq->curr); |
4996 | } | |
4997 | out_unlock: | |
0122ec5b | 4998 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 4999 | } |
1da177e4 LT |
5000 | EXPORT_SYMBOL(set_user_nice); |
5001 | ||
e43379f1 MM |
5002 | /* |
5003 | * can_nice - check if a task can reduce its nice value | |
5004 | * @p: task | |
5005 | * @nice: nice value | |
5006 | */ | |
36c8b586 | 5007 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 5008 | { |
024f4747 MM |
5009 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
5010 | int nice_rlim = 20 - nice; | |
48f24c4d | 5011 | |
78d7d407 | 5012 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
5013 | capable(CAP_SYS_NICE)); |
5014 | } | |
5015 | ||
1da177e4 LT |
5016 | #ifdef __ARCH_WANT_SYS_NICE |
5017 | ||
5018 | /* | |
5019 | * sys_nice - change the priority of the current process. | |
5020 | * @increment: priority increment | |
5021 | * | |
5022 | * sys_setpriority is a more generic, but much slower function that | |
5023 | * does similar things. | |
5024 | */ | |
5add95d4 | 5025 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 5026 | { |
48f24c4d | 5027 | long nice, retval; |
1da177e4 LT |
5028 | |
5029 | /* | |
5030 | * Setpriority might change our priority at the same moment. | |
5031 | * We don't have to worry. Conceptually one call occurs first | |
5032 | * and we have a single winner. | |
5033 | */ | |
e43379f1 MM |
5034 | if (increment < -40) |
5035 | increment = -40; | |
1da177e4 LT |
5036 | if (increment > 40) |
5037 | increment = 40; | |
5038 | ||
2b8f836f | 5039 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
5040 | if (nice < -20) |
5041 | nice = -20; | |
5042 | if (nice > 19) | |
5043 | nice = 19; | |
5044 | ||
e43379f1 MM |
5045 | if (increment < 0 && !can_nice(current, nice)) |
5046 | return -EPERM; | |
5047 | ||
1da177e4 LT |
5048 | retval = security_task_setnice(current, nice); |
5049 | if (retval) | |
5050 | return retval; | |
5051 | ||
5052 | set_user_nice(current, nice); | |
5053 | return 0; | |
5054 | } | |
5055 | ||
5056 | #endif | |
5057 | ||
5058 | /** | |
5059 | * task_prio - return the priority value of a given task. | |
5060 | * @p: the task in question. | |
5061 | * | |
5062 | * This is the priority value as seen by users in /proc. | |
5063 | * RT tasks are offset by -200. Normal tasks are centered | |
5064 | * around 0, value goes from -16 to +15. | |
5065 | */ | |
36c8b586 | 5066 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
5067 | { |
5068 | return p->prio - MAX_RT_PRIO; | |
5069 | } | |
5070 | ||
5071 | /** | |
5072 | * task_nice - return the nice value of a given task. | |
5073 | * @p: the task in question. | |
5074 | */ | |
36c8b586 | 5075 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
5076 | { |
5077 | return TASK_NICE(p); | |
5078 | } | |
150d8bed | 5079 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
5080 | |
5081 | /** | |
5082 | * idle_cpu - is a given cpu idle currently? | |
5083 | * @cpu: the processor in question. | |
5084 | */ | |
5085 | int idle_cpu(int cpu) | |
5086 | { | |
5087 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
5088 | } | |
5089 | ||
1da177e4 LT |
5090 | /** |
5091 | * idle_task - return the idle task for a given cpu. | |
5092 | * @cpu: the processor in question. | |
5093 | */ | |
36c8b586 | 5094 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
5095 | { |
5096 | return cpu_rq(cpu)->idle; | |
5097 | } | |
5098 | ||
5099 | /** | |
5100 | * find_process_by_pid - find a process with a matching PID value. | |
5101 | * @pid: the pid in question. | |
5102 | */ | |
a9957449 | 5103 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 5104 | { |
228ebcbe | 5105 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
5106 | } |
5107 | ||
5108 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
5109 | static void |
5110 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 5111 | { |
1da177e4 LT |
5112 | p->policy = policy; |
5113 | p->rt_priority = prio; | |
b29739f9 IM |
5114 | p->normal_prio = normal_prio(p); |
5115 | /* we are holding p->pi_lock already */ | |
5116 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
5117 | if (rt_prio(p->prio)) |
5118 | p->sched_class = &rt_sched_class; | |
5119 | else | |
5120 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 5121 | set_load_weight(p); |
1da177e4 LT |
5122 | } |
5123 | ||
c69e8d9c DH |
5124 | /* |
5125 | * check the target process has a UID that matches the current process's | |
5126 | */ | |
5127 | static bool check_same_owner(struct task_struct *p) | |
5128 | { | |
5129 | const struct cred *cred = current_cred(), *pcred; | |
5130 | bool match; | |
5131 | ||
5132 | rcu_read_lock(); | |
5133 | pcred = __task_cred(p); | |
b0e77598 SH |
5134 | if (cred->user->user_ns == pcred->user->user_ns) |
5135 | match = (cred->euid == pcred->euid || | |
5136 | cred->euid == pcred->uid); | |
5137 | else | |
5138 | match = false; | |
c69e8d9c DH |
5139 | rcu_read_unlock(); |
5140 | return match; | |
5141 | } | |
5142 | ||
961ccddd | 5143 | static int __sched_setscheduler(struct task_struct *p, int policy, |
fe7de49f | 5144 | const struct sched_param *param, bool user) |
1da177e4 | 5145 | { |
83b699ed | 5146 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 5147 | unsigned long flags; |
83ab0aa0 | 5148 | const struct sched_class *prev_class; |
70b97a7f | 5149 | struct rq *rq; |
ca94c442 | 5150 | int reset_on_fork; |
1da177e4 | 5151 | |
66e5393a SR |
5152 | /* may grab non-irq protected spin_locks */ |
5153 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
5154 | recheck: |
5155 | /* double check policy once rq lock held */ | |
ca94c442 LP |
5156 | if (policy < 0) { |
5157 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 5158 | policy = oldpolicy = p->policy; |
ca94c442 LP |
5159 | } else { |
5160 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
5161 | policy &= ~SCHED_RESET_ON_FORK; | |
5162 | ||
5163 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
5164 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
5165 | policy != SCHED_IDLE) | |
5166 | return -EINVAL; | |
5167 | } | |
5168 | ||
1da177e4 LT |
5169 | /* |
5170 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
5171 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
5172 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
5173 | */ |
5174 | if (param->sched_priority < 0 || | |
95cdf3b7 | 5175 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 5176 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 5177 | return -EINVAL; |
e05606d3 | 5178 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
5179 | return -EINVAL; |
5180 | ||
37e4ab3f OC |
5181 | /* |
5182 | * Allow unprivileged RT tasks to decrease priority: | |
5183 | */ | |
961ccddd | 5184 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 5185 | if (rt_policy(policy)) { |
a44702e8 ON |
5186 | unsigned long rlim_rtprio = |
5187 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
5188 | |
5189 | /* can't set/change the rt policy */ | |
5190 | if (policy != p->policy && !rlim_rtprio) | |
5191 | return -EPERM; | |
5192 | ||
5193 | /* can't increase priority */ | |
5194 | if (param->sched_priority > p->rt_priority && | |
5195 | param->sched_priority > rlim_rtprio) | |
5196 | return -EPERM; | |
5197 | } | |
c02aa73b | 5198 | |
dd41f596 | 5199 | /* |
c02aa73b DH |
5200 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
5201 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 5202 | */ |
c02aa73b DH |
5203 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
5204 | if (!can_nice(p, TASK_NICE(p))) | |
5205 | return -EPERM; | |
5206 | } | |
5fe1d75f | 5207 | |
37e4ab3f | 5208 | /* can't change other user's priorities */ |
c69e8d9c | 5209 | if (!check_same_owner(p)) |
37e4ab3f | 5210 | return -EPERM; |
ca94c442 LP |
5211 | |
5212 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
5213 | if (p->sched_reset_on_fork && !reset_on_fork) | |
5214 | return -EPERM; | |
37e4ab3f | 5215 | } |
1da177e4 | 5216 | |
725aad24 | 5217 | if (user) { |
b0ae1981 | 5218 | retval = security_task_setscheduler(p); |
725aad24 JF |
5219 | if (retval) |
5220 | return retval; | |
5221 | } | |
5222 | ||
b29739f9 IM |
5223 | /* |
5224 | * make sure no PI-waiters arrive (or leave) while we are | |
5225 | * changing the priority of the task: | |
0122ec5b | 5226 | * |
25985edc | 5227 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
5228 | * runqueue lock must be held. |
5229 | */ | |
0122ec5b | 5230 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 5231 | |
34f971f6 PZ |
5232 | /* |
5233 | * Changing the policy of the stop threads its a very bad idea | |
5234 | */ | |
5235 | if (p == rq->stop) { | |
0122ec5b | 5236 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
5237 | return -EINVAL; |
5238 | } | |
5239 | ||
a51e9198 DF |
5240 | /* |
5241 | * If not changing anything there's no need to proceed further: | |
5242 | */ | |
5243 | if (unlikely(policy == p->policy && (!rt_policy(policy) || | |
5244 | param->sched_priority == p->rt_priority))) { | |
5245 | ||
5246 | __task_rq_unlock(rq); | |
5247 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
5248 | return 0; | |
5249 | } | |
5250 | ||
dc61b1d6 PZ |
5251 | #ifdef CONFIG_RT_GROUP_SCHED |
5252 | if (user) { | |
5253 | /* | |
5254 | * Do not allow realtime tasks into groups that have no runtime | |
5255 | * assigned. | |
5256 | */ | |
5257 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
5258 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
5259 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 5260 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
5261 | return -EPERM; |
5262 | } | |
5263 | } | |
5264 | #endif | |
5265 | ||
1da177e4 LT |
5266 | /* recheck policy now with rq lock held */ |
5267 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5268 | policy = oldpolicy = -1; | |
0122ec5b | 5269 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
5270 | goto recheck; |
5271 | } | |
fd2f4419 | 5272 | on_rq = p->on_rq; |
051a1d1a | 5273 | running = task_current(rq, p); |
0e1f3483 | 5274 | if (on_rq) |
2e1cb74a | 5275 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5276 | if (running) |
5277 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5278 | |
ca94c442 LP |
5279 | p->sched_reset_on_fork = reset_on_fork; |
5280 | ||
1da177e4 | 5281 | oldprio = p->prio; |
83ab0aa0 | 5282 | prev_class = p->sched_class; |
dd41f596 | 5283 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5284 | |
0e1f3483 HS |
5285 | if (running) |
5286 | p->sched_class->set_curr_task(rq); | |
da7a735e | 5287 | if (on_rq) |
dd41f596 | 5288 | activate_task(rq, p, 0); |
cb469845 | 5289 | |
da7a735e | 5290 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 5291 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 5292 | |
95e02ca9 TG |
5293 | rt_mutex_adjust_pi(p); |
5294 | ||
1da177e4 LT |
5295 | return 0; |
5296 | } | |
961ccddd RR |
5297 | |
5298 | /** | |
5299 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5300 | * @p: the task in question. | |
5301 | * @policy: new policy. | |
5302 | * @param: structure containing the new RT priority. | |
5303 | * | |
5304 | * NOTE that the task may be already dead. | |
5305 | */ | |
5306 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 5307 | const struct sched_param *param) |
961ccddd RR |
5308 | { |
5309 | return __sched_setscheduler(p, policy, param, true); | |
5310 | } | |
1da177e4 LT |
5311 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5312 | ||
961ccddd RR |
5313 | /** |
5314 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5315 | * @p: the task in question. | |
5316 | * @policy: new policy. | |
5317 | * @param: structure containing the new RT priority. | |
5318 | * | |
5319 | * Just like sched_setscheduler, only don't bother checking if the | |
5320 | * current context has permission. For example, this is needed in | |
5321 | * stop_machine(): we create temporary high priority worker threads, | |
5322 | * but our caller might not have that capability. | |
5323 | */ | |
5324 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 5325 | const struct sched_param *param) |
961ccddd RR |
5326 | { |
5327 | return __sched_setscheduler(p, policy, param, false); | |
5328 | } | |
5329 | ||
95cdf3b7 IM |
5330 | static int |
5331 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5332 | { |
1da177e4 LT |
5333 | struct sched_param lparam; |
5334 | struct task_struct *p; | |
36c8b586 | 5335 | int retval; |
1da177e4 LT |
5336 | |
5337 | if (!param || pid < 0) | |
5338 | return -EINVAL; | |
5339 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5340 | return -EFAULT; | |
5fe1d75f ON |
5341 | |
5342 | rcu_read_lock(); | |
5343 | retval = -ESRCH; | |
1da177e4 | 5344 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5345 | if (p != NULL) |
5346 | retval = sched_setscheduler(p, policy, &lparam); | |
5347 | rcu_read_unlock(); | |
36c8b586 | 5348 | |
1da177e4 LT |
5349 | return retval; |
5350 | } | |
5351 | ||
5352 | /** | |
5353 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5354 | * @pid: the pid in question. | |
5355 | * @policy: new policy. | |
5356 | * @param: structure containing the new RT priority. | |
5357 | */ | |
5add95d4 HC |
5358 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
5359 | struct sched_param __user *, param) | |
1da177e4 | 5360 | { |
c21761f1 JB |
5361 | /* negative values for policy are not valid */ |
5362 | if (policy < 0) | |
5363 | return -EINVAL; | |
5364 | ||
1da177e4 LT |
5365 | return do_sched_setscheduler(pid, policy, param); |
5366 | } | |
5367 | ||
5368 | /** | |
5369 | * sys_sched_setparam - set/change the RT priority of a thread | |
5370 | * @pid: the pid in question. | |
5371 | * @param: structure containing the new RT priority. | |
5372 | */ | |
5add95d4 | 5373 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5374 | { |
5375 | return do_sched_setscheduler(pid, -1, param); | |
5376 | } | |
5377 | ||
5378 | /** | |
5379 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5380 | * @pid: the pid in question. | |
5381 | */ | |
5add95d4 | 5382 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 5383 | { |
36c8b586 | 5384 | struct task_struct *p; |
3a5c359a | 5385 | int retval; |
1da177e4 LT |
5386 | |
5387 | if (pid < 0) | |
3a5c359a | 5388 | return -EINVAL; |
1da177e4 LT |
5389 | |
5390 | retval = -ESRCH; | |
5fe85be0 | 5391 | rcu_read_lock(); |
1da177e4 LT |
5392 | p = find_process_by_pid(pid); |
5393 | if (p) { | |
5394 | retval = security_task_getscheduler(p); | |
5395 | if (!retval) | |
ca94c442 LP |
5396 | retval = p->policy |
5397 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 5398 | } |
5fe85be0 | 5399 | rcu_read_unlock(); |
1da177e4 LT |
5400 | return retval; |
5401 | } | |
5402 | ||
5403 | /** | |
ca94c442 | 5404 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
5405 | * @pid: the pid in question. |
5406 | * @param: structure containing the RT priority. | |
5407 | */ | |
5add95d4 | 5408 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5409 | { |
5410 | struct sched_param lp; | |
36c8b586 | 5411 | struct task_struct *p; |
3a5c359a | 5412 | int retval; |
1da177e4 LT |
5413 | |
5414 | if (!param || pid < 0) | |
3a5c359a | 5415 | return -EINVAL; |
1da177e4 | 5416 | |
5fe85be0 | 5417 | rcu_read_lock(); |
1da177e4 LT |
5418 | p = find_process_by_pid(pid); |
5419 | retval = -ESRCH; | |
5420 | if (!p) | |
5421 | goto out_unlock; | |
5422 | ||
5423 | retval = security_task_getscheduler(p); | |
5424 | if (retval) | |
5425 | goto out_unlock; | |
5426 | ||
5427 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 5428 | rcu_read_unlock(); |
1da177e4 LT |
5429 | |
5430 | /* | |
5431 | * This one might sleep, we cannot do it with a spinlock held ... | |
5432 | */ | |
5433 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5434 | ||
1da177e4 LT |
5435 | return retval; |
5436 | ||
5437 | out_unlock: | |
5fe85be0 | 5438 | rcu_read_unlock(); |
1da177e4 LT |
5439 | return retval; |
5440 | } | |
5441 | ||
96f874e2 | 5442 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 5443 | { |
5a16f3d3 | 5444 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
5445 | struct task_struct *p; |
5446 | int retval; | |
1da177e4 | 5447 | |
95402b38 | 5448 | get_online_cpus(); |
23f5d142 | 5449 | rcu_read_lock(); |
1da177e4 LT |
5450 | |
5451 | p = find_process_by_pid(pid); | |
5452 | if (!p) { | |
23f5d142 | 5453 | rcu_read_unlock(); |
95402b38 | 5454 | put_online_cpus(); |
1da177e4 LT |
5455 | return -ESRCH; |
5456 | } | |
5457 | ||
23f5d142 | 5458 | /* Prevent p going away */ |
1da177e4 | 5459 | get_task_struct(p); |
23f5d142 | 5460 | rcu_read_unlock(); |
1da177e4 | 5461 | |
5a16f3d3 RR |
5462 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
5463 | retval = -ENOMEM; | |
5464 | goto out_put_task; | |
5465 | } | |
5466 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
5467 | retval = -ENOMEM; | |
5468 | goto out_free_cpus_allowed; | |
5469 | } | |
1da177e4 | 5470 | retval = -EPERM; |
b0e77598 | 5471 | if (!check_same_owner(p) && !task_ns_capable(p, CAP_SYS_NICE)) |
1da177e4 LT |
5472 | goto out_unlock; |
5473 | ||
b0ae1981 | 5474 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
5475 | if (retval) |
5476 | goto out_unlock; | |
5477 | ||
5a16f3d3 RR |
5478 | cpuset_cpus_allowed(p, cpus_allowed); |
5479 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 5480 | again: |
5a16f3d3 | 5481 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 5482 | |
8707d8b8 | 5483 | if (!retval) { |
5a16f3d3 RR |
5484 | cpuset_cpus_allowed(p, cpus_allowed); |
5485 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
5486 | /* |
5487 | * We must have raced with a concurrent cpuset | |
5488 | * update. Just reset the cpus_allowed to the | |
5489 | * cpuset's cpus_allowed | |
5490 | */ | |
5a16f3d3 | 5491 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
5492 | goto again; |
5493 | } | |
5494 | } | |
1da177e4 | 5495 | out_unlock: |
5a16f3d3 RR |
5496 | free_cpumask_var(new_mask); |
5497 | out_free_cpus_allowed: | |
5498 | free_cpumask_var(cpus_allowed); | |
5499 | out_put_task: | |
1da177e4 | 5500 | put_task_struct(p); |
95402b38 | 5501 | put_online_cpus(); |
1da177e4 LT |
5502 | return retval; |
5503 | } | |
5504 | ||
5505 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 5506 | struct cpumask *new_mask) |
1da177e4 | 5507 | { |
96f874e2 RR |
5508 | if (len < cpumask_size()) |
5509 | cpumask_clear(new_mask); | |
5510 | else if (len > cpumask_size()) | |
5511 | len = cpumask_size(); | |
5512 | ||
1da177e4 LT |
5513 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
5514 | } | |
5515 | ||
5516 | /** | |
5517 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5518 | * @pid: pid of the process | |
5519 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5520 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5521 | */ | |
5add95d4 HC |
5522 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
5523 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 5524 | { |
5a16f3d3 | 5525 | cpumask_var_t new_mask; |
1da177e4 LT |
5526 | int retval; |
5527 | ||
5a16f3d3 RR |
5528 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
5529 | return -ENOMEM; | |
1da177e4 | 5530 | |
5a16f3d3 RR |
5531 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
5532 | if (retval == 0) | |
5533 | retval = sched_setaffinity(pid, new_mask); | |
5534 | free_cpumask_var(new_mask); | |
5535 | return retval; | |
1da177e4 LT |
5536 | } |
5537 | ||
96f874e2 | 5538 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 5539 | { |
36c8b586 | 5540 | struct task_struct *p; |
31605683 | 5541 | unsigned long flags; |
1da177e4 | 5542 | int retval; |
1da177e4 | 5543 | |
95402b38 | 5544 | get_online_cpus(); |
23f5d142 | 5545 | rcu_read_lock(); |
1da177e4 LT |
5546 | |
5547 | retval = -ESRCH; | |
5548 | p = find_process_by_pid(pid); | |
5549 | if (!p) | |
5550 | goto out_unlock; | |
5551 | ||
e7834f8f DQ |
5552 | retval = security_task_getscheduler(p); |
5553 | if (retval) | |
5554 | goto out_unlock; | |
5555 | ||
013fdb80 | 5556 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
96f874e2 | 5557 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
013fdb80 | 5558 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
5559 | |
5560 | out_unlock: | |
23f5d142 | 5561 | rcu_read_unlock(); |
95402b38 | 5562 | put_online_cpus(); |
1da177e4 | 5563 | |
9531b62f | 5564 | return retval; |
1da177e4 LT |
5565 | } |
5566 | ||
5567 | /** | |
5568 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5569 | * @pid: pid of the process | |
5570 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5571 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5572 | */ | |
5add95d4 HC |
5573 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
5574 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
5575 | { |
5576 | int ret; | |
f17c8607 | 5577 | cpumask_var_t mask; |
1da177e4 | 5578 | |
84fba5ec | 5579 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
5580 | return -EINVAL; |
5581 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
5582 | return -EINVAL; |
5583 | ||
f17c8607 RR |
5584 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
5585 | return -ENOMEM; | |
1da177e4 | 5586 | |
f17c8607 RR |
5587 | ret = sched_getaffinity(pid, mask); |
5588 | if (ret == 0) { | |
8bc037fb | 5589 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
5590 | |
5591 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
5592 | ret = -EFAULT; |
5593 | else | |
cd3d8031 | 5594 | ret = retlen; |
f17c8607 RR |
5595 | } |
5596 | free_cpumask_var(mask); | |
1da177e4 | 5597 | |
f17c8607 | 5598 | return ret; |
1da177e4 LT |
5599 | } |
5600 | ||
5601 | /** | |
5602 | * sys_sched_yield - yield the current processor to other threads. | |
5603 | * | |
dd41f596 IM |
5604 | * This function yields the current CPU to other tasks. If there are no |
5605 | * other threads running on this CPU then this function will return. | |
1da177e4 | 5606 | */ |
5add95d4 | 5607 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 5608 | { |
70b97a7f | 5609 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5610 | |
2d72376b | 5611 | schedstat_inc(rq, yld_count); |
4530d7ab | 5612 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5613 | |
5614 | /* | |
5615 | * Since we are going to call schedule() anyway, there's | |
5616 | * no need to preempt or enable interrupts: | |
5617 | */ | |
5618 | __release(rq->lock); | |
8a25d5de | 5619 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 5620 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
5621 | preempt_enable_no_resched(); |
5622 | ||
5623 | schedule(); | |
5624 | ||
5625 | return 0; | |
5626 | } | |
5627 | ||
d86ee480 PZ |
5628 | static inline int should_resched(void) |
5629 | { | |
5630 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
5631 | } | |
5632 | ||
e7b38404 | 5633 | static void __cond_resched(void) |
1da177e4 | 5634 | { |
e7aaaa69 FW |
5635 | add_preempt_count(PREEMPT_ACTIVE); |
5636 | schedule(); | |
5637 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
5638 | } |
5639 | ||
02b67cc3 | 5640 | int __sched _cond_resched(void) |
1da177e4 | 5641 | { |
d86ee480 | 5642 | if (should_resched()) { |
1da177e4 LT |
5643 | __cond_resched(); |
5644 | return 1; | |
5645 | } | |
5646 | return 0; | |
5647 | } | |
02b67cc3 | 5648 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5649 | |
5650 | /* | |
613afbf8 | 5651 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
5652 | * call schedule, and on return reacquire the lock. |
5653 | * | |
41a2d6cf | 5654 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5655 | * operations here to prevent schedule() from being called twice (once via |
5656 | * spin_unlock(), once by hand). | |
5657 | */ | |
613afbf8 | 5658 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5659 | { |
d86ee480 | 5660 | int resched = should_resched(); |
6df3cecb JK |
5661 | int ret = 0; |
5662 | ||
f607c668 PZ |
5663 | lockdep_assert_held(lock); |
5664 | ||
95c354fe | 5665 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5666 | spin_unlock(lock); |
d86ee480 | 5667 | if (resched) |
95c354fe NP |
5668 | __cond_resched(); |
5669 | else | |
5670 | cpu_relax(); | |
6df3cecb | 5671 | ret = 1; |
1da177e4 | 5672 | spin_lock(lock); |
1da177e4 | 5673 | } |
6df3cecb | 5674 | return ret; |
1da177e4 | 5675 | } |
613afbf8 | 5676 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 5677 | |
613afbf8 | 5678 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
5679 | { |
5680 | BUG_ON(!in_softirq()); | |
5681 | ||
d86ee480 | 5682 | if (should_resched()) { |
98d82567 | 5683 | local_bh_enable(); |
1da177e4 LT |
5684 | __cond_resched(); |
5685 | local_bh_disable(); | |
5686 | return 1; | |
5687 | } | |
5688 | return 0; | |
5689 | } | |
613afbf8 | 5690 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 5691 | |
1da177e4 LT |
5692 | /** |
5693 | * yield - yield the current processor to other threads. | |
5694 | * | |
72fd4a35 | 5695 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5696 | * thread runnable and calls sys_sched_yield(). |
5697 | */ | |
5698 | void __sched yield(void) | |
5699 | { | |
5700 | set_current_state(TASK_RUNNING); | |
5701 | sys_sched_yield(); | |
5702 | } | |
1da177e4 LT |
5703 | EXPORT_SYMBOL(yield); |
5704 | ||
d95f4122 MG |
5705 | /** |
5706 | * yield_to - yield the current processor to another thread in | |
5707 | * your thread group, or accelerate that thread toward the | |
5708 | * processor it's on. | |
16addf95 RD |
5709 | * @p: target task |
5710 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
5711 | * |
5712 | * It's the caller's job to ensure that the target task struct | |
5713 | * can't go away on us before we can do any checks. | |
5714 | * | |
5715 | * Returns true if we indeed boosted the target task. | |
5716 | */ | |
5717 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
5718 | { | |
5719 | struct task_struct *curr = current; | |
5720 | struct rq *rq, *p_rq; | |
5721 | unsigned long flags; | |
5722 | bool yielded = 0; | |
5723 | ||
5724 | local_irq_save(flags); | |
5725 | rq = this_rq(); | |
5726 | ||
5727 | again: | |
5728 | p_rq = task_rq(p); | |
5729 | double_rq_lock(rq, p_rq); | |
5730 | while (task_rq(p) != p_rq) { | |
5731 | double_rq_unlock(rq, p_rq); | |
5732 | goto again; | |
5733 | } | |
5734 | ||
5735 | if (!curr->sched_class->yield_to_task) | |
5736 | goto out; | |
5737 | ||
5738 | if (curr->sched_class != p->sched_class) | |
5739 | goto out; | |
5740 | ||
5741 | if (task_running(p_rq, p) || p->state) | |
5742 | goto out; | |
5743 | ||
5744 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 5745 | if (yielded) { |
d95f4122 | 5746 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
5747 | /* |
5748 | * Make p's CPU reschedule; pick_next_entity takes care of | |
5749 | * fairness. | |
5750 | */ | |
5751 | if (preempt && rq != p_rq) | |
5752 | resched_task(p_rq->curr); | |
5753 | } | |
d95f4122 MG |
5754 | |
5755 | out: | |
5756 | double_rq_unlock(rq, p_rq); | |
5757 | local_irq_restore(flags); | |
5758 | ||
5759 | if (yielded) | |
5760 | schedule(); | |
5761 | ||
5762 | return yielded; | |
5763 | } | |
5764 | EXPORT_SYMBOL_GPL(yield_to); | |
5765 | ||
1da177e4 | 5766 | /* |
41a2d6cf | 5767 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5768 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
5769 | */ |
5770 | void __sched io_schedule(void) | |
5771 | { | |
54d35f29 | 5772 | struct rq *rq = raw_rq(); |
1da177e4 | 5773 | |
0ff92245 | 5774 | delayacct_blkio_start(); |
1da177e4 | 5775 | atomic_inc(&rq->nr_iowait); |
73c10101 | 5776 | blk_flush_plug(current); |
8f0dfc34 | 5777 | current->in_iowait = 1; |
1da177e4 | 5778 | schedule(); |
8f0dfc34 | 5779 | current->in_iowait = 0; |
1da177e4 | 5780 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5781 | delayacct_blkio_end(); |
1da177e4 | 5782 | } |
1da177e4 LT |
5783 | EXPORT_SYMBOL(io_schedule); |
5784 | ||
5785 | long __sched io_schedule_timeout(long timeout) | |
5786 | { | |
54d35f29 | 5787 | struct rq *rq = raw_rq(); |
1da177e4 LT |
5788 | long ret; |
5789 | ||
0ff92245 | 5790 | delayacct_blkio_start(); |
1da177e4 | 5791 | atomic_inc(&rq->nr_iowait); |
73c10101 | 5792 | blk_flush_plug(current); |
8f0dfc34 | 5793 | current->in_iowait = 1; |
1da177e4 | 5794 | ret = schedule_timeout(timeout); |
8f0dfc34 | 5795 | current->in_iowait = 0; |
1da177e4 | 5796 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5797 | delayacct_blkio_end(); |
1da177e4 LT |
5798 | return ret; |
5799 | } | |
5800 | ||
5801 | /** | |
5802 | * sys_sched_get_priority_max - return maximum RT priority. | |
5803 | * @policy: scheduling class. | |
5804 | * | |
5805 | * this syscall returns the maximum rt_priority that can be used | |
5806 | * by a given scheduling class. | |
5807 | */ | |
5add95d4 | 5808 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5809 | { |
5810 | int ret = -EINVAL; | |
5811 | ||
5812 | switch (policy) { | |
5813 | case SCHED_FIFO: | |
5814 | case SCHED_RR: | |
5815 | ret = MAX_USER_RT_PRIO-1; | |
5816 | break; | |
5817 | case SCHED_NORMAL: | |
b0a9499c | 5818 | case SCHED_BATCH: |
dd41f596 | 5819 | case SCHED_IDLE: |
1da177e4 LT |
5820 | ret = 0; |
5821 | break; | |
5822 | } | |
5823 | return ret; | |
5824 | } | |
5825 | ||
5826 | /** | |
5827 | * sys_sched_get_priority_min - return minimum RT priority. | |
5828 | * @policy: scheduling class. | |
5829 | * | |
5830 | * this syscall returns the minimum rt_priority that can be used | |
5831 | * by a given scheduling class. | |
5832 | */ | |
5add95d4 | 5833 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5834 | { |
5835 | int ret = -EINVAL; | |
5836 | ||
5837 | switch (policy) { | |
5838 | case SCHED_FIFO: | |
5839 | case SCHED_RR: | |
5840 | ret = 1; | |
5841 | break; | |
5842 | case SCHED_NORMAL: | |
b0a9499c | 5843 | case SCHED_BATCH: |
dd41f596 | 5844 | case SCHED_IDLE: |
1da177e4 LT |
5845 | ret = 0; |
5846 | } | |
5847 | return ret; | |
5848 | } | |
5849 | ||
5850 | /** | |
5851 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5852 | * @pid: pid of the process. | |
5853 | * @interval: userspace pointer to the timeslice value. | |
5854 | * | |
5855 | * this syscall writes the default timeslice value of a given process | |
5856 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5857 | */ | |
17da2bd9 | 5858 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5859 | struct timespec __user *, interval) |
1da177e4 | 5860 | { |
36c8b586 | 5861 | struct task_struct *p; |
a4ec24b4 | 5862 | unsigned int time_slice; |
dba091b9 TG |
5863 | unsigned long flags; |
5864 | struct rq *rq; | |
3a5c359a | 5865 | int retval; |
1da177e4 | 5866 | struct timespec t; |
1da177e4 LT |
5867 | |
5868 | if (pid < 0) | |
3a5c359a | 5869 | return -EINVAL; |
1da177e4 LT |
5870 | |
5871 | retval = -ESRCH; | |
1a551ae7 | 5872 | rcu_read_lock(); |
1da177e4 LT |
5873 | p = find_process_by_pid(pid); |
5874 | if (!p) | |
5875 | goto out_unlock; | |
5876 | ||
5877 | retval = security_task_getscheduler(p); | |
5878 | if (retval) | |
5879 | goto out_unlock; | |
5880 | ||
dba091b9 TG |
5881 | rq = task_rq_lock(p, &flags); |
5882 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 5883 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 5884 | |
1a551ae7 | 5885 | rcu_read_unlock(); |
a4ec24b4 | 5886 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5887 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5888 | return retval; |
3a5c359a | 5889 | |
1da177e4 | 5890 | out_unlock: |
1a551ae7 | 5891 | rcu_read_unlock(); |
1da177e4 LT |
5892 | return retval; |
5893 | } | |
5894 | ||
7c731e0a | 5895 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5896 | |
82a1fcb9 | 5897 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5898 | { |
1da177e4 | 5899 | unsigned long free = 0; |
36c8b586 | 5900 | unsigned state; |
1da177e4 | 5901 | |
1da177e4 | 5902 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 5903 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 5904 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5905 | #if BITS_PER_LONG == 32 |
1da177e4 | 5906 | if (state == TASK_RUNNING) |
3df0fc5b | 5907 | printk(KERN_CONT " running "); |
1da177e4 | 5908 | else |
3df0fc5b | 5909 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5910 | #else |
5911 | if (state == TASK_RUNNING) | |
3df0fc5b | 5912 | printk(KERN_CONT " running task "); |
1da177e4 | 5913 | else |
3df0fc5b | 5914 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5915 | #endif |
5916 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5917 | free = stack_not_used(p); |
1da177e4 | 5918 | #endif |
3df0fc5b | 5919 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
5920 | task_pid_nr(p), task_pid_nr(p->real_parent), |
5921 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 5922 | |
5fb5e6de | 5923 | show_stack(p, NULL); |
1da177e4 LT |
5924 | } |
5925 | ||
e59e2ae2 | 5926 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5927 | { |
36c8b586 | 5928 | struct task_struct *g, *p; |
1da177e4 | 5929 | |
4bd77321 | 5930 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5931 | printk(KERN_INFO |
5932 | " task PC stack pid father\n"); | |
1da177e4 | 5933 | #else |
3df0fc5b PZ |
5934 | printk(KERN_INFO |
5935 | " task PC stack pid father\n"); | |
1da177e4 LT |
5936 | #endif |
5937 | read_lock(&tasklist_lock); | |
5938 | do_each_thread(g, p) { | |
5939 | /* | |
5940 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 5941 | * console might take a lot of time: |
1da177e4 LT |
5942 | */ |
5943 | touch_nmi_watchdog(); | |
39bc89fd | 5944 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5945 | sched_show_task(p); |
1da177e4 LT |
5946 | } while_each_thread(g, p); |
5947 | ||
04c9167f JF |
5948 | touch_all_softlockup_watchdogs(); |
5949 | ||
dd41f596 IM |
5950 | #ifdef CONFIG_SCHED_DEBUG |
5951 | sysrq_sched_debug_show(); | |
5952 | #endif | |
1da177e4 | 5953 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5954 | /* |
5955 | * Only show locks if all tasks are dumped: | |
5956 | */ | |
93335a21 | 5957 | if (!state_filter) |
e59e2ae2 | 5958 | debug_show_all_locks(); |
1da177e4 LT |
5959 | } |
5960 | ||
1df21055 IM |
5961 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5962 | { | |
dd41f596 | 5963 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5964 | } |
5965 | ||
f340c0d1 IM |
5966 | /** |
5967 | * init_idle - set up an idle thread for a given CPU | |
5968 | * @idle: task in question | |
5969 | * @cpu: cpu the idle task belongs to | |
5970 | * | |
5971 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5972 | * flag, to make booting more robust. | |
5973 | */ | |
5c1e1767 | 5974 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5975 | { |
70b97a7f | 5976 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5977 | unsigned long flags; |
5978 | ||
05fa785c | 5979 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 5980 | |
dd41f596 | 5981 | __sched_fork(idle); |
06b83b5f | 5982 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5983 | idle->se.exec_start = sched_clock(); |
5984 | ||
1e1b6c51 | 5985 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
5986 | /* |
5987 | * We're having a chicken and egg problem, even though we are | |
5988 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
5989 | * lockdep check in task_group() will fail. | |
5990 | * | |
5991 | * Similar case to sched_fork(). / Alternatively we could | |
5992 | * use task_rq_lock() here and obtain the other rq->lock. | |
5993 | * | |
5994 | * Silence PROVE_RCU | |
5995 | */ | |
5996 | rcu_read_lock(); | |
dd41f596 | 5997 | __set_task_cpu(idle, cpu); |
6506cf6c | 5998 | rcu_read_unlock(); |
1da177e4 | 5999 | |
1da177e4 | 6000 | rq->curr = rq->idle = idle; |
3ca7a440 PZ |
6001 | #if defined(CONFIG_SMP) |
6002 | idle->on_cpu = 1; | |
4866cde0 | 6003 | #endif |
05fa785c | 6004 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
6005 | |
6006 | /* Set the preempt count _outside_ the spinlocks! */ | |
a1261f54 | 6007 | task_thread_info(idle)->preempt_count = 0; |
625f2a37 | 6008 | |
dd41f596 IM |
6009 | /* |
6010 | * The idle tasks have their own, simple scheduling class: | |
6011 | */ | |
6012 | idle->sched_class = &idle_sched_class; | |
868baf07 | 6013 | ftrace_graph_init_idle_task(idle, cpu); |
1da177e4 LT |
6014 | } |
6015 | ||
6016 | /* | |
6017 | * In a system that switches off the HZ timer nohz_cpu_mask | |
6018 | * indicates which cpus entered this state. This is used | |
6019 | * in the rcu update to wait only for active cpus. For system | |
6020 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 6021 | * always be CPU_BITS_NONE. |
1da177e4 | 6022 | */ |
6a7b3dc3 | 6023 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 6024 | |
19978ca6 IM |
6025 | /* |
6026 | * Increase the granularity value when there are more CPUs, | |
6027 | * because with more CPUs the 'effective latency' as visible | |
6028 | * to users decreases. But the relationship is not linear, | |
6029 | * so pick a second-best guess by going with the log2 of the | |
6030 | * number of CPUs. | |
6031 | * | |
6032 | * This idea comes from the SD scheduler of Con Kolivas: | |
6033 | */ | |
acb4a848 | 6034 | static int get_update_sysctl_factor(void) |
19978ca6 | 6035 | { |
4ca3ef71 | 6036 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
6037 | unsigned int factor; |
6038 | ||
6039 | switch (sysctl_sched_tunable_scaling) { | |
6040 | case SCHED_TUNABLESCALING_NONE: | |
6041 | factor = 1; | |
6042 | break; | |
6043 | case SCHED_TUNABLESCALING_LINEAR: | |
6044 | factor = cpus; | |
6045 | break; | |
6046 | case SCHED_TUNABLESCALING_LOG: | |
6047 | default: | |
6048 | factor = 1 + ilog2(cpus); | |
6049 | break; | |
6050 | } | |
19978ca6 | 6051 | |
acb4a848 CE |
6052 | return factor; |
6053 | } | |
19978ca6 | 6054 | |
acb4a848 CE |
6055 | static void update_sysctl(void) |
6056 | { | |
6057 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 6058 | |
0bcdcf28 CE |
6059 | #define SET_SYSCTL(name) \ |
6060 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
6061 | SET_SYSCTL(sched_min_granularity); | |
6062 | SET_SYSCTL(sched_latency); | |
6063 | SET_SYSCTL(sched_wakeup_granularity); | |
0bcdcf28 CE |
6064 | #undef SET_SYSCTL |
6065 | } | |
55cd5340 | 6066 | |
0bcdcf28 CE |
6067 | static inline void sched_init_granularity(void) |
6068 | { | |
6069 | update_sysctl(); | |
19978ca6 IM |
6070 | } |
6071 | ||
1da177e4 | 6072 | #ifdef CONFIG_SMP |
1e1b6c51 KM |
6073 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
6074 | { | |
6075 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
6076 | p->sched_class->set_cpus_allowed(p, new_mask); | |
6077 | else { | |
6078 | cpumask_copy(&p->cpus_allowed, new_mask); | |
6079 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
6080 | } | |
6081 | } | |
6082 | ||
1da177e4 LT |
6083 | /* |
6084 | * This is how migration works: | |
6085 | * | |
969c7921 TH |
6086 | * 1) we invoke migration_cpu_stop() on the target CPU using |
6087 | * stop_one_cpu(). | |
6088 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
6089 | * off the CPU) | |
6090 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
6091 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 6092 | * it and puts it into the right queue. |
969c7921 TH |
6093 | * 5) stopper completes and stop_one_cpu() returns and the migration |
6094 | * is done. | |
1da177e4 LT |
6095 | */ |
6096 | ||
6097 | /* | |
6098 | * Change a given task's CPU affinity. Migrate the thread to a | |
6099 | * proper CPU and schedule it away if the CPU it's executing on | |
6100 | * is removed from the allowed bitmask. | |
6101 | * | |
6102 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 6103 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
6104 | * call is not atomic; no spinlocks may be held. |
6105 | */ | |
96f874e2 | 6106 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
6107 | { |
6108 | unsigned long flags; | |
70b97a7f | 6109 | struct rq *rq; |
969c7921 | 6110 | unsigned int dest_cpu; |
48f24c4d | 6111 | int ret = 0; |
1da177e4 LT |
6112 | |
6113 | rq = task_rq_lock(p, &flags); | |
e2912009 | 6114 | |
db44fc01 YZ |
6115 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
6116 | goto out; | |
6117 | ||
6ad4c188 | 6118 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
6119 | ret = -EINVAL; |
6120 | goto out; | |
6121 | } | |
6122 | ||
db44fc01 | 6123 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current)) { |
9985b0ba DR |
6124 | ret = -EINVAL; |
6125 | goto out; | |
6126 | } | |
6127 | ||
1e1b6c51 | 6128 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 6129 | |
1da177e4 | 6130 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 6131 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
6132 | goto out; |
6133 | ||
969c7921 | 6134 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
bd8e7dde | 6135 | if (p->on_rq) { |
969c7921 | 6136 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 6137 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 6138 | task_rq_unlock(rq, p, &flags); |
969c7921 | 6139 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
6140 | tlb_migrate_finish(p->mm); |
6141 | return 0; | |
6142 | } | |
6143 | out: | |
0122ec5b | 6144 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 6145 | |
1da177e4 LT |
6146 | return ret; |
6147 | } | |
cd8ba7cd | 6148 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
6149 | |
6150 | /* | |
41a2d6cf | 6151 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
6152 | * this because either it can't run here any more (set_cpus_allowed() |
6153 | * away from this CPU, or CPU going down), or because we're | |
6154 | * attempting to rebalance this task on exec (sched_exec). | |
6155 | * | |
6156 | * So we race with normal scheduler movements, but that's OK, as long | |
6157 | * as the task is no longer on this CPU. | |
efc30814 KK |
6158 | * |
6159 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 6160 | */ |
efc30814 | 6161 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 6162 | { |
70b97a7f | 6163 | struct rq *rq_dest, *rq_src; |
e2912009 | 6164 | int ret = 0; |
1da177e4 | 6165 | |
e761b772 | 6166 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 6167 | return ret; |
1da177e4 LT |
6168 | |
6169 | rq_src = cpu_rq(src_cpu); | |
6170 | rq_dest = cpu_rq(dest_cpu); | |
6171 | ||
0122ec5b | 6172 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
6173 | double_rq_lock(rq_src, rq_dest); |
6174 | /* Already moved. */ | |
6175 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 6176 | goto done; |
1da177e4 | 6177 | /* Affinity changed (again). */ |
96f874e2 | 6178 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 6179 | goto fail; |
1da177e4 | 6180 | |
e2912009 PZ |
6181 | /* |
6182 | * If we're not on a rq, the next wake-up will ensure we're | |
6183 | * placed properly. | |
6184 | */ | |
fd2f4419 | 6185 | if (p->on_rq) { |
2e1cb74a | 6186 | deactivate_task(rq_src, p, 0); |
e2912009 | 6187 | set_task_cpu(p, dest_cpu); |
dd41f596 | 6188 | activate_task(rq_dest, p, 0); |
15afe09b | 6189 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 6190 | } |
b1e38734 | 6191 | done: |
efc30814 | 6192 | ret = 1; |
b1e38734 | 6193 | fail: |
1da177e4 | 6194 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 6195 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 6196 | return ret; |
1da177e4 LT |
6197 | } |
6198 | ||
6199 | /* | |
969c7921 TH |
6200 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
6201 | * and performs thread migration by bumping thread off CPU then | |
6202 | * 'pushing' onto another runqueue. | |
1da177e4 | 6203 | */ |
969c7921 | 6204 | static int migration_cpu_stop(void *data) |
1da177e4 | 6205 | { |
969c7921 | 6206 | struct migration_arg *arg = data; |
f7b4cddc | 6207 | |
969c7921 TH |
6208 | /* |
6209 | * The original target cpu might have gone down and we might | |
6210 | * be on another cpu but it doesn't matter. | |
6211 | */ | |
f7b4cddc | 6212 | local_irq_disable(); |
969c7921 | 6213 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 6214 | local_irq_enable(); |
1da177e4 | 6215 | return 0; |
f7b4cddc ON |
6216 | } |
6217 | ||
1da177e4 | 6218 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 6219 | |
054b9108 | 6220 | /* |
48c5ccae PZ |
6221 | * Ensures that the idle task is using init_mm right before its cpu goes |
6222 | * offline. | |
054b9108 | 6223 | */ |
48c5ccae | 6224 | void idle_task_exit(void) |
1da177e4 | 6225 | { |
48c5ccae | 6226 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 6227 | |
48c5ccae | 6228 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 6229 | |
48c5ccae PZ |
6230 | if (mm != &init_mm) |
6231 | switch_mm(mm, &init_mm, current); | |
6232 | mmdrop(mm); | |
1da177e4 LT |
6233 | } |
6234 | ||
6235 | /* | |
6236 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
6237 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
6238 | * for performance reasons the counter is not stricly tracking tasks to | |
6239 | * their home CPUs. So we just add the counter to another CPU's counter, | |
6240 | * to keep the global sum constant after CPU-down: | |
6241 | */ | |
70b97a7f | 6242 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 6243 | { |
6ad4c188 | 6244 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 | 6245 | |
1da177e4 LT |
6246 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; |
6247 | rq_src->nr_uninterruptible = 0; | |
1da177e4 LT |
6248 | } |
6249 | ||
dd41f596 | 6250 | /* |
48c5ccae | 6251 | * remove the tasks which were accounted by rq from calc_load_tasks. |
1da177e4 | 6252 | */ |
48c5ccae | 6253 | static void calc_global_load_remove(struct rq *rq) |
1da177e4 | 6254 | { |
48c5ccae PZ |
6255 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); |
6256 | rq->calc_load_active = 0; | |
1da177e4 LT |
6257 | } |
6258 | ||
48f24c4d | 6259 | /* |
48c5ccae PZ |
6260 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
6261 | * try_to_wake_up()->select_task_rq(). | |
6262 | * | |
6263 | * Called with rq->lock held even though we'er in stop_machine() and | |
6264 | * there's no concurrency possible, we hold the required locks anyway | |
6265 | * because of lock validation efforts. | |
1da177e4 | 6266 | */ |
48c5ccae | 6267 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 6268 | { |
70b97a7f | 6269 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
6270 | struct task_struct *next, *stop = rq->stop; |
6271 | int dest_cpu; | |
1da177e4 LT |
6272 | |
6273 | /* | |
48c5ccae PZ |
6274 | * Fudge the rq selection such that the below task selection loop |
6275 | * doesn't get stuck on the currently eligible stop task. | |
6276 | * | |
6277 | * We're currently inside stop_machine() and the rq is either stuck | |
6278 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
6279 | * either way we should never end up calling schedule() until we're | |
6280 | * done here. | |
1da177e4 | 6281 | */ |
48c5ccae | 6282 | rq->stop = NULL; |
48f24c4d | 6283 | |
dd41f596 | 6284 | for ( ; ; ) { |
48c5ccae PZ |
6285 | /* |
6286 | * There's this thread running, bail when that's the only | |
6287 | * remaining thread. | |
6288 | */ | |
6289 | if (rq->nr_running == 1) | |
dd41f596 | 6290 | break; |
48c5ccae | 6291 | |
b67802ea | 6292 | next = pick_next_task(rq); |
48c5ccae | 6293 | BUG_ON(!next); |
79c53799 | 6294 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 6295 | |
48c5ccae PZ |
6296 | /* Find suitable destination for @next, with force if needed. */ |
6297 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
6298 | raw_spin_unlock(&rq->lock); | |
6299 | ||
6300 | __migrate_task(next, dead_cpu, dest_cpu); | |
6301 | ||
6302 | raw_spin_lock(&rq->lock); | |
1da177e4 | 6303 | } |
dce48a84 | 6304 | |
48c5ccae | 6305 | rq->stop = stop; |
dce48a84 | 6306 | } |
48c5ccae | 6307 | |
1da177e4 LT |
6308 | #endif /* CONFIG_HOTPLUG_CPU */ |
6309 | ||
e692ab53 NP |
6310 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6311 | ||
6312 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6313 | { |
6314 | .procname = "sched_domain", | |
c57baf1e | 6315 | .mode = 0555, |
e0361851 | 6316 | }, |
56992309 | 6317 | {} |
e692ab53 NP |
6318 | }; |
6319 | ||
6320 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
6321 | { |
6322 | .procname = "kernel", | |
c57baf1e | 6323 | .mode = 0555, |
e0361851 AD |
6324 | .child = sd_ctl_dir, |
6325 | }, | |
56992309 | 6326 | {} |
e692ab53 NP |
6327 | }; |
6328 | ||
6329 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6330 | { | |
6331 | struct ctl_table *entry = | |
5cf9f062 | 6332 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6333 | |
e692ab53 NP |
6334 | return entry; |
6335 | } | |
6336 | ||
6382bc90 MM |
6337 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6338 | { | |
cd790076 | 6339 | struct ctl_table *entry; |
6382bc90 | 6340 | |
cd790076 MM |
6341 | /* |
6342 | * In the intermediate directories, both the child directory and | |
6343 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6344 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6345 | * static strings and all have proc handlers. |
6346 | */ | |
6347 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6348 | if (entry->child) |
6349 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6350 | if (entry->proc_handler == NULL) |
6351 | kfree(entry->procname); | |
6352 | } | |
6382bc90 MM |
6353 | |
6354 | kfree(*tablep); | |
6355 | *tablep = NULL; | |
6356 | } | |
6357 | ||
e692ab53 | 6358 | static void |
e0361851 | 6359 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6360 | const char *procname, void *data, int maxlen, |
6361 | mode_t mode, proc_handler *proc_handler) | |
6362 | { | |
e692ab53 NP |
6363 | entry->procname = procname; |
6364 | entry->data = data; | |
6365 | entry->maxlen = maxlen; | |
6366 | entry->mode = mode; | |
6367 | entry->proc_handler = proc_handler; | |
6368 | } | |
6369 | ||
6370 | static struct ctl_table * | |
6371 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6372 | { | |
a5d8c348 | 6373 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 6374 | |
ad1cdc1d MM |
6375 | if (table == NULL) |
6376 | return NULL; | |
6377 | ||
e0361851 | 6378 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6379 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6380 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6381 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6382 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6383 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6384 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6385 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6386 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6387 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6388 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6389 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6390 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6391 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6392 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6393 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6394 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6395 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6396 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6397 | &sd->cache_nice_tries, |
6398 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6399 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6400 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
6401 | set_table_entry(&table[11], "name", sd->name, |
6402 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
6403 | /* &table[12] is terminator */ | |
e692ab53 NP |
6404 | |
6405 | return table; | |
6406 | } | |
6407 | ||
9a4e7159 | 6408 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6409 | { |
6410 | struct ctl_table *entry, *table; | |
6411 | struct sched_domain *sd; | |
6412 | int domain_num = 0, i; | |
6413 | char buf[32]; | |
6414 | ||
6415 | for_each_domain(cpu, sd) | |
6416 | domain_num++; | |
6417 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6418 | if (table == NULL) |
6419 | return NULL; | |
e692ab53 NP |
6420 | |
6421 | i = 0; | |
6422 | for_each_domain(cpu, sd) { | |
6423 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6424 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6425 | entry->mode = 0555; |
e692ab53 NP |
6426 | entry->child = sd_alloc_ctl_domain_table(sd); |
6427 | entry++; | |
6428 | i++; | |
6429 | } | |
6430 | return table; | |
6431 | } | |
6432 | ||
6433 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6434 | static void register_sched_domain_sysctl(void) |
e692ab53 | 6435 | { |
6ad4c188 | 6436 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
6437 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
6438 | char buf[32]; | |
6439 | ||
7378547f MM |
6440 | WARN_ON(sd_ctl_dir[0].child); |
6441 | sd_ctl_dir[0].child = entry; | |
6442 | ||
ad1cdc1d MM |
6443 | if (entry == NULL) |
6444 | return; | |
6445 | ||
6ad4c188 | 6446 | for_each_possible_cpu(i) { |
e692ab53 | 6447 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6448 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6449 | entry->mode = 0555; |
e692ab53 | 6450 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6451 | entry++; |
e692ab53 | 6452 | } |
7378547f MM |
6453 | |
6454 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6455 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6456 | } | |
6382bc90 | 6457 | |
7378547f | 6458 | /* may be called multiple times per register */ |
6382bc90 MM |
6459 | static void unregister_sched_domain_sysctl(void) |
6460 | { | |
7378547f MM |
6461 | if (sd_sysctl_header) |
6462 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6463 | sd_sysctl_header = NULL; |
7378547f MM |
6464 | if (sd_ctl_dir[0].child) |
6465 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6466 | } |
e692ab53 | 6467 | #else |
6382bc90 MM |
6468 | static void register_sched_domain_sysctl(void) |
6469 | { | |
6470 | } | |
6471 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6472 | { |
6473 | } | |
6474 | #endif | |
6475 | ||
1f11eb6a GH |
6476 | static void set_rq_online(struct rq *rq) |
6477 | { | |
6478 | if (!rq->online) { | |
6479 | const struct sched_class *class; | |
6480 | ||
c6c4927b | 6481 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6482 | rq->online = 1; |
6483 | ||
6484 | for_each_class(class) { | |
6485 | if (class->rq_online) | |
6486 | class->rq_online(rq); | |
6487 | } | |
6488 | } | |
6489 | } | |
6490 | ||
6491 | static void set_rq_offline(struct rq *rq) | |
6492 | { | |
6493 | if (rq->online) { | |
6494 | const struct sched_class *class; | |
6495 | ||
6496 | for_each_class(class) { | |
6497 | if (class->rq_offline) | |
6498 | class->rq_offline(rq); | |
6499 | } | |
6500 | ||
c6c4927b | 6501 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6502 | rq->online = 0; |
6503 | } | |
6504 | } | |
6505 | ||
1da177e4 LT |
6506 | /* |
6507 | * migration_call - callback that gets triggered when a CPU is added. | |
6508 | * Here we can start up the necessary migration thread for the new CPU. | |
6509 | */ | |
48f24c4d IM |
6510 | static int __cpuinit |
6511 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6512 | { |
48f24c4d | 6513 | int cpu = (long)hcpu; |
1da177e4 | 6514 | unsigned long flags; |
969c7921 | 6515 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 6516 | |
48c5ccae | 6517 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 6518 | |
1da177e4 | 6519 | case CPU_UP_PREPARE: |
a468d389 | 6520 | rq->calc_load_update = calc_load_update; |
1da177e4 | 6521 | break; |
48f24c4d | 6522 | |
1da177e4 | 6523 | case CPU_ONLINE: |
1f94ef59 | 6524 | /* Update our root-domain */ |
05fa785c | 6525 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 6526 | if (rq->rd) { |
c6c4927b | 6527 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
6528 | |
6529 | set_rq_online(rq); | |
1f94ef59 | 6530 | } |
05fa785c | 6531 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 6532 | break; |
48f24c4d | 6533 | |
1da177e4 | 6534 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 6535 | case CPU_DYING: |
317f3941 | 6536 | sched_ttwu_pending(); |
57d885fe | 6537 | /* Update our root-domain */ |
05fa785c | 6538 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 6539 | if (rq->rd) { |
c6c4927b | 6540 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 6541 | set_rq_offline(rq); |
57d885fe | 6542 | } |
48c5ccae PZ |
6543 | migrate_tasks(cpu); |
6544 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 6545 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
48c5ccae PZ |
6546 | |
6547 | migrate_nr_uninterruptible(rq); | |
6548 | calc_global_load_remove(rq); | |
57d885fe | 6549 | break; |
1da177e4 LT |
6550 | #endif |
6551 | } | |
49c022e6 PZ |
6552 | |
6553 | update_max_interval(); | |
6554 | ||
1da177e4 LT |
6555 | return NOTIFY_OK; |
6556 | } | |
6557 | ||
f38b0820 PM |
6558 | /* |
6559 | * Register at high priority so that task migration (migrate_all_tasks) | |
6560 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 6561 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 6562 | */ |
26c2143b | 6563 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 6564 | .notifier_call = migration_call, |
50a323b7 | 6565 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
6566 | }; |
6567 | ||
3a101d05 TH |
6568 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
6569 | unsigned long action, void *hcpu) | |
6570 | { | |
6571 | switch (action & ~CPU_TASKS_FROZEN) { | |
6572 | case CPU_ONLINE: | |
6573 | case CPU_DOWN_FAILED: | |
6574 | set_cpu_active((long)hcpu, true); | |
6575 | return NOTIFY_OK; | |
6576 | default: | |
6577 | return NOTIFY_DONE; | |
6578 | } | |
6579 | } | |
6580 | ||
6581 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
6582 | unsigned long action, void *hcpu) | |
6583 | { | |
6584 | switch (action & ~CPU_TASKS_FROZEN) { | |
6585 | case CPU_DOWN_PREPARE: | |
6586 | set_cpu_active((long)hcpu, false); | |
6587 | return NOTIFY_OK; | |
6588 | default: | |
6589 | return NOTIFY_DONE; | |
6590 | } | |
6591 | } | |
6592 | ||
7babe8db | 6593 | static int __init migration_init(void) |
1da177e4 LT |
6594 | { |
6595 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6596 | int err; |
48f24c4d | 6597 | |
3a101d05 | 6598 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
6599 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6600 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6601 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6602 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 6603 | |
3a101d05 TH |
6604 | /* Register cpu active notifiers */ |
6605 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
6606 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
6607 | ||
a004cd42 | 6608 | return 0; |
1da177e4 | 6609 | } |
7babe8db | 6610 | early_initcall(migration_init); |
1da177e4 LT |
6611 | #endif |
6612 | ||
6613 | #ifdef CONFIG_SMP | |
476f3534 | 6614 | |
4cb98839 PZ |
6615 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
6616 | ||
3e9830dc | 6617 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6618 | |
f6630114 MT |
6619 | static __read_mostly int sched_domain_debug_enabled; |
6620 | ||
6621 | static int __init sched_domain_debug_setup(char *str) | |
6622 | { | |
6623 | sched_domain_debug_enabled = 1; | |
6624 | ||
6625 | return 0; | |
6626 | } | |
6627 | early_param("sched_debug", sched_domain_debug_setup); | |
6628 | ||
7c16ec58 | 6629 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 6630 | struct cpumask *groupmask) |
1da177e4 | 6631 | { |
4dcf6aff | 6632 | struct sched_group *group = sd->groups; |
434d53b0 | 6633 | char str[256]; |
1da177e4 | 6634 | |
968ea6d8 | 6635 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 6636 | cpumask_clear(groupmask); |
4dcf6aff IM |
6637 | |
6638 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6639 | ||
6640 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 6641 | printk("does not load-balance\n"); |
4dcf6aff | 6642 | if (sd->parent) |
3df0fc5b PZ |
6643 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
6644 | " has parent"); | |
4dcf6aff | 6645 | return -1; |
41c7ce9a NP |
6646 | } |
6647 | ||
3df0fc5b | 6648 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 6649 | |
758b2cdc | 6650 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
6651 | printk(KERN_ERR "ERROR: domain->span does not contain " |
6652 | "CPU%d\n", cpu); | |
4dcf6aff | 6653 | } |
758b2cdc | 6654 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
6655 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
6656 | " CPU%d\n", cpu); | |
4dcf6aff | 6657 | } |
1da177e4 | 6658 | |
4dcf6aff | 6659 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6660 | do { |
4dcf6aff | 6661 | if (!group) { |
3df0fc5b PZ |
6662 | printk("\n"); |
6663 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6664 | break; |
6665 | } | |
6666 | ||
9c3f75cb | 6667 | if (!group->sgp->power) { |
3df0fc5b PZ |
6668 | printk(KERN_CONT "\n"); |
6669 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6670 | "set\n"); | |
4dcf6aff IM |
6671 | break; |
6672 | } | |
1da177e4 | 6673 | |
758b2cdc | 6674 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
6675 | printk(KERN_CONT "\n"); |
6676 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
6677 | break; |
6678 | } | |
1da177e4 | 6679 | |
758b2cdc | 6680 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
6681 | printk(KERN_CONT "\n"); |
6682 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
6683 | break; |
6684 | } | |
1da177e4 | 6685 | |
758b2cdc | 6686 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 6687 | |
968ea6d8 | 6688 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 6689 | |
3df0fc5b | 6690 | printk(KERN_CONT " %s", str); |
9c3f75cb | 6691 | if (group->sgp->power != SCHED_POWER_SCALE) { |
3df0fc5b | 6692 | printk(KERN_CONT " (cpu_power = %d)", |
9c3f75cb | 6693 | group->sgp->power); |
381512cf | 6694 | } |
1da177e4 | 6695 | |
4dcf6aff IM |
6696 | group = group->next; |
6697 | } while (group != sd->groups); | |
3df0fc5b | 6698 | printk(KERN_CONT "\n"); |
1da177e4 | 6699 | |
758b2cdc | 6700 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 6701 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6702 | |
758b2cdc RR |
6703 | if (sd->parent && |
6704 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
6705 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6706 | "of domain->span\n"); | |
4dcf6aff IM |
6707 | return 0; |
6708 | } | |
1da177e4 | 6709 | |
4dcf6aff IM |
6710 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6711 | { | |
6712 | int level = 0; | |
1da177e4 | 6713 | |
f6630114 MT |
6714 | if (!sched_domain_debug_enabled) |
6715 | return; | |
6716 | ||
4dcf6aff IM |
6717 | if (!sd) { |
6718 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6719 | return; | |
6720 | } | |
1da177e4 | 6721 | |
4dcf6aff IM |
6722 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6723 | ||
6724 | for (;;) { | |
4cb98839 | 6725 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 6726 | break; |
1da177e4 LT |
6727 | level++; |
6728 | sd = sd->parent; | |
33859f7f | 6729 | if (!sd) |
4dcf6aff IM |
6730 | break; |
6731 | } | |
1da177e4 | 6732 | } |
6d6bc0ad | 6733 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6734 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6735 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6736 | |
1a20ff27 | 6737 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 6738 | { |
758b2cdc | 6739 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
6740 | return 1; |
6741 | ||
6742 | /* Following flags need at least 2 groups */ | |
6743 | if (sd->flags & (SD_LOAD_BALANCE | | |
6744 | SD_BALANCE_NEWIDLE | | |
6745 | SD_BALANCE_FORK | | |
89c4710e SS |
6746 | SD_BALANCE_EXEC | |
6747 | SD_SHARE_CPUPOWER | | |
6748 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6749 | if (sd->groups != sd->groups->next) |
6750 | return 0; | |
6751 | } | |
6752 | ||
6753 | /* Following flags don't use groups */ | |
c88d5910 | 6754 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
6755 | return 0; |
6756 | ||
6757 | return 1; | |
6758 | } | |
6759 | ||
48f24c4d IM |
6760 | static int |
6761 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6762 | { |
6763 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6764 | ||
6765 | if (sd_degenerate(parent)) | |
6766 | return 1; | |
6767 | ||
758b2cdc | 6768 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
6769 | return 0; |
6770 | ||
245af2c7 SS |
6771 | /* Flags needing groups don't count if only 1 group in parent */ |
6772 | if (parent->groups == parent->groups->next) { | |
6773 | pflags &= ~(SD_LOAD_BALANCE | | |
6774 | SD_BALANCE_NEWIDLE | | |
6775 | SD_BALANCE_FORK | | |
89c4710e SS |
6776 | SD_BALANCE_EXEC | |
6777 | SD_SHARE_CPUPOWER | | |
6778 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6779 | if (nr_node_ids == 1) |
6780 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6781 | } |
6782 | if (~cflags & pflags) | |
6783 | return 0; | |
6784 | ||
6785 | return 1; | |
6786 | } | |
6787 | ||
dce840a0 | 6788 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 6789 | { |
dce840a0 | 6790 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 6791 | |
68e74568 | 6792 | cpupri_cleanup(&rd->cpupri); |
c6c4927b RR |
6793 | free_cpumask_var(rd->rto_mask); |
6794 | free_cpumask_var(rd->online); | |
6795 | free_cpumask_var(rd->span); | |
6796 | kfree(rd); | |
6797 | } | |
6798 | ||
57d885fe GH |
6799 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6800 | { | |
a0490fa3 | 6801 | struct root_domain *old_rd = NULL; |
57d885fe | 6802 | unsigned long flags; |
57d885fe | 6803 | |
05fa785c | 6804 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
6805 | |
6806 | if (rq->rd) { | |
a0490fa3 | 6807 | old_rd = rq->rd; |
57d885fe | 6808 | |
c6c4927b | 6809 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6810 | set_rq_offline(rq); |
57d885fe | 6811 | |
c6c4927b | 6812 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6813 | |
a0490fa3 IM |
6814 | /* |
6815 | * If we dont want to free the old_rt yet then | |
6816 | * set old_rd to NULL to skip the freeing later | |
6817 | * in this function: | |
6818 | */ | |
6819 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
6820 | old_rd = NULL; | |
57d885fe GH |
6821 | } |
6822 | ||
6823 | atomic_inc(&rd->refcount); | |
6824 | rq->rd = rd; | |
6825 | ||
c6c4927b | 6826 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 6827 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 6828 | set_rq_online(rq); |
57d885fe | 6829 | |
05fa785c | 6830 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
6831 | |
6832 | if (old_rd) | |
dce840a0 | 6833 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
6834 | } |
6835 | ||
68c38fc3 | 6836 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6837 | { |
6838 | memset(rd, 0, sizeof(*rd)); | |
6839 | ||
68c38fc3 | 6840 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 6841 | goto out; |
68c38fc3 | 6842 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 6843 | goto free_span; |
68c38fc3 | 6844 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 6845 | goto free_online; |
6e0534f2 | 6846 | |
68c38fc3 | 6847 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 6848 | goto free_rto_mask; |
c6c4927b | 6849 | return 0; |
6e0534f2 | 6850 | |
68e74568 RR |
6851 | free_rto_mask: |
6852 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6853 | free_online: |
6854 | free_cpumask_var(rd->online); | |
6855 | free_span: | |
6856 | free_cpumask_var(rd->span); | |
0c910d28 | 6857 | out: |
c6c4927b | 6858 | return -ENOMEM; |
57d885fe GH |
6859 | } |
6860 | ||
6861 | static void init_defrootdomain(void) | |
6862 | { | |
68c38fc3 | 6863 | init_rootdomain(&def_root_domain); |
c6c4927b | 6864 | |
57d885fe GH |
6865 | atomic_set(&def_root_domain.refcount, 1); |
6866 | } | |
6867 | ||
dc938520 | 6868 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6869 | { |
6870 | struct root_domain *rd; | |
6871 | ||
6872 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6873 | if (!rd) | |
6874 | return NULL; | |
6875 | ||
68c38fc3 | 6876 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
6877 | kfree(rd); |
6878 | return NULL; | |
6879 | } | |
57d885fe GH |
6880 | |
6881 | return rd; | |
6882 | } | |
6883 | ||
e3589f6c PZ |
6884 | static void free_sched_groups(struct sched_group *sg, int free_sgp) |
6885 | { | |
6886 | struct sched_group *tmp, *first; | |
6887 | ||
6888 | if (!sg) | |
6889 | return; | |
6890 | ||
6891 | first = sg; | |
6892 | do { | |
6893 | tmp = sg->next; | |
6894 | ||
6895 | if (free_sgp && atomic_dec_and_test(&sg->sgp->ref)) | |
6896 | kfree(sg->sgp); | |
6897 | ||
6898 | kfree(sg); | |
6899 | sg = tmp; | |
6900 | } while (sg != first); | |
6901 | } | |
6902 | ||
dce840a0 PZ |
6903 | static void free_sched_domain(struct rcu_head *rcu) |
6904 | { | |
6905 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
6906 | |
6907 | /* | |
6908 | * If its an overlapping domain it has private groups, iterate and | |
6909 | * nuke them all. | |
6910 | */ | |
6911 | if (sd->flags & SD_OVERLAP) { | |
6912 | free_sched_groups(sd->groups, 1); | |
6913 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
9c3f75cb | 6914 | kfree(sd->groups->sgp); |
dce840a0 | 6915 | kfree(sd->groups); |
9c3f75cb | 6916 | } |
dce840a0 PZ |
6917 | kfree(sd); |
6918 | } | |
6919 | ||
6920 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
6921 | { | |
6922 | call_rcu(&sd->rcu, free_sched_domain); | |
6923 | } | |
6924 | ||
6925 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
6926 | { | |
6927 | for (; sd; sd = sd->parent) | |
6928 | destroy_sched_domain(sd, cpu); | |
6929 | } | |
6930 | ||
1da177e4 | 6931 | /* |
0eab9146 | 6932 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6933 | * hold the hotplug lock. |
6934 | */ | |
0eab9146 IM |
6935 | static void |
6936 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6937 | { |
70b97a7f | 6938 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6939 | struct sched_domain *tmp; |
6940 | ||
6941 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 6942 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6943 | struct sched_domain *parent = tmp->parent; |
6944 | if (!parent) | |
6945 | break; | |
f29c9b1c | 6946 | |
1a848870 | 6947 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6948 | tmp->parent = parent->parent; |
1a848870 SS |
6949 | if (parent->parent) |
6950 | parent->parent->child = tmp; | |
dce840a0 | 6951 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
6952 | } else |
6953 | tmp = tmp->parent; | |
245af2c7 SS |
6954 | } |
6955 | ||
1a848870 | 6956 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 6957 | tmp = sd; |
245af2c7 | 6958 | sd = sd->parent; |
dce840a0 | 6959 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
6960 | if (sd) |
6961 | sd->child = NULL; | |
6962 | } | |
1da177e4 | 6963 | |
4cb98839 | 6964 | sched_domain_debug(sd, cpu); |
1da177e4 | 6965 | |
57d885fe | 6966 | rq_attach_root(rq, rd); |
dce840a0 | 6967 | tmp = rq->sd; |
674311d5 | 6968 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 6969 | destroy_sched_domains(tmp, cpu); |
1da177e4 LT |
6970 | } |
6971 | ||
6972 | /* cpus with isolated domains */ | |
dcc30a35 | 6973 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
6974 | |
6975 | /* Setup the mask of cpus configured for isolated domains */ | |
6976 | static int __init isolated_cpu_setup(char *str) | |
6977 | { | |
bdddd296 | 6978 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 6979 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
6980 | return 1; |
6981 | } | |
6982 | ||
8927f494 | 6983 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 6984 | |
9c1cfda2 | 6985 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6986 | |
9c1cfda2 | 6987 | #ifdef CONFIG_NUMA |
198e2f18 | 6988 | |
9c1cfda2 JH |
6989 | /** |
6990 | * find_next_best_node - find the next node to include in a sched_domain | |
6991 | * @node: node whose sched_domain we're building | |
6992 | * @used_nodes: nodes already in the sched_domain | |
6993 | * | |
41a2d6cf | 6994 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6995 | * finds the closest node not already in the @used_nodes map. |
6996 | * | |
6997 | * Should use nodemask_t. | |
6998 | */ | |
c5f59f08 | 6999 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 | 7000 | { |
7142d17e | 7001 | int i, n, val, min_val, best_node = -1; |
9c1cfda2 JH |
7002 | |
7003 | min_val = INT_MAX; | |
7004 | ||
076ac2af | 7005 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 7006 | /* Start at @node */ |
076ac2af | 7007 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
7008 | |
7009 | if (!nr_cpus_node(n)) | |
7010 | continue; | |
7011 | ||
7012 | /* Skip already used nodes */ | |
c5f59f08 | 7013 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
7014 | continue; |
7015 | ||
7016 | /* Simple min distance search */ | |
7017 | val = node_distance(node, n); | |
7018 | ||
7019 | if (val < min_val) { | |
7020 | min_val = val; | |
7021 | best_node = n; | |
7022 | } | |
7023 | } | |
7024 | ||
7142d17e HD |
7025 | if (best_node != -1) |
7026 | node_set(best_node, *used_nodes); | |
9c1cfda2 JH |
7027 | return best_node; |
7028 | } | |
7029 | ||
7030 | /** | |
7031 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
7032 | * @node: node whose cpumask we're constructing | |
73486722 | 7033 | * @span: resulting cpumask |
9c1cfda2 | 7034 | * |
41a2d6cf | 7035 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
7036 | * should be one that prevents unnecessary balancing, but also spreads tasks |
7037 | * out optimally. | |
7038 | */ | |
96f874e2 | 7039 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 7040 | { |
c5f59f08 | 7041 | nodemask_t used_nodes; |
48f24c4d | 7042 | int i; |
9c1cfda2 | 7043 | |
6ca09dfc | 7044 | cpumask_clear(span); |
c5f59f08 | 7045 | nodes_clear(used_nodes); |
9c1cfda2 | 7046 | |
6ca09dfc | 7047 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 7048 | node_set(node, used_nodes); |
9c1cfda2 JH |
7049 | |
7050 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 7051 | int next_node = find_next_best_node(node, &used_nodes); |
7142d17e HD |
7052 | if (next_node < 0) |
7053 | break; | |
6ca09dfc | 7054 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 7055 | } |
9c1cfda2 | 7056 | } |
d3081f52 PZ |
7057 | |
7058 | static const struct cpumask *cpu_node_mask(int cpu) | |
7059 | { | |
7060 | lockdep_assert_held(&sched_domains_mutex); | |
7061 | ||
7062 | sched_domain_node_span(cpu_to_node(cpu), sched_domains_tmpmask); | |
7063 | ||
7064 | return sched_domains_tmpmask; | |
7065 | } | |
2c402dc3 PZ |
7066 | |
7067 | static const struct cpumask *cpu_allnodes_mask(int cpu) | |
7068 | { | |
7069 | return cpu_possible_mask; | |
7070 | } | |
6d6bc0ad | 7071 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 7072 | |
d3081f52 PZ |
7073 | static const struct cpumask *cpu_cpu_mask(int cpu) |
7074 | { | |
7075 | return cpumask_of_node(cpu_to_node(cpu)); | |
7076 | } | |
7077 | ||
5c45bf27 | 7078 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 7079 | |
dce840a0 PZ |
7080 | struct sd_data { |
7081 | struct sched_domain **__percpu sd; | |
7082 | struct sched_group **__percpu sg; | |
9c3f75cb | 7083 | struct sched_group_power **__percpu sgp; |
dce840a0 PZ |
7084 | }; |
7085 | ||
49a02c51 | 7086 | struct s_data { |
21d42ccf | 7087 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
7088 | struct root_domain *rd; |
7089 | }; | |
7090 | ||
2109b99e | 7091 | enum s_alloc { |
2109b99e | 7092 | sa_rootdomain, |
21d42ccf | 7093 | sa_sd, |
dce840a0 | 7094 | sa_sd_storage, |
2109b99e AH |
7095 | sa_none, |
7096 | }; | |
7097 | ||
54ab4ff4 PZ |
7098 | struct sched_domain_topology_level; |
7099 | ||
7100 | typedef struct sched_domain *(*sched_domain_init_f)(struct sched_domain_topology_level *tl, int cpu); | |
eb7a74e6 PZ |
7101 | typedef const struct cpumask *(*sched_domain_mask_f)(int cpu); |
7102 | ||
e3589f6c PZ |
7103 | #define SDTL_OVERLAP 0x01 |
7104 | ||
eb7a74e6 | 7105 | struct sched_domain_topology_level { |
2c402dc3 PZ |
7106 | sched_domain_init_f init; |
7107 | sched_domain_mask_f mask; | |
e3589f6c | 7108 | int flags; |
54ab4ff4 | 7109 | struct sd_data data; |
eb7a74e6 PZ |
7110 | }; |
7111 | ||
e3589f6c PZ |
7112 | static int |
7113 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
7114 | { | |
7115 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
7116 | const struct cpumask *span = sched_domain_span(sd); | |
7117 | struct cpumask *covered = sched_domains_tmpmask; | |
7118 | struct sd_data *sdd = sd->private; | |
7119 | struct sched_domain *child; | |
7120 | int i; | |
7121 | ||
7122 | cpumask_clear(covered); | |
7123 | ||
7124 | for_each_cpu(i, span) { | |
7125 | struct cpumask *sg_span; | |
7126 | ||
7127 | if (cpumask_test_cpu(i, covered)) | |
7128 | continue; | |
7129 | ||
7130 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
7131 | GFP_KERNEL, cpu_to_node(i)); | |
7132 | ||
7133 | if (!sg) | |
7134 | goto fail; | |
7135 | ||
7136 | sg_span = sched_group_cpus(sg); | |
7137 | ||
7138 | child = *per_cpu_ptr(sdd->sd, i); | |
7139 | if (child->child) { | |
7140 | child = child->child; | |
7141 | cpumask_copy(sg_span, sched_domain_span(child)); | |
7142 | } else | |
7143 | cpumask_set_cpu(i, sg_span); | |
7144 | ||
7145 | cpumask_or(covered, covered, sg_span); | |
7146 | ||
7147 | sg->sgp = *per_cpu_ptr(sdd->sgp, cpumask_first(sg_span)); | |
7148 | atomic_inc(&sg->sgp->ref); | |
7149 | ||
7150 | if (cpumask_test_cpu(cpu, sg_span)) | |
7151 | groups = sg; | |
7152 | ||
7153 | if (!first) | |
7154 | first = sg; | |
7155 | if (last) | |
7156 | last->next = sg; | |
7157 | last = sg; | |
7158 | last->next = first; | |
7159 | } | |
7160 | sd->groups = groups; | |
7161 | ||
7162 | return 0; | |
7163 | ||
7164 | fail: | |
7165 | free_sched_groups(first, 0); | |
7166 | ||
7167 | return -ENOMEM; | |
7168 | } | |
7169 | ||
dce840a0 | 7170 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 7171 | { |
dce840a0 PZ |
7172 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
7173 | struct sched_domain *child = sd->child; | |
1da177e4 | 7174 | |
dce840a0 PZ |
7175 | if (child) |
7176 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 7177 | |
9c3f75cb | 7178 | if (sg) { |
dce840a0 | 7179 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
9c3f75cb | 7180 | (*sg)->sgp = *per_cpu_ptr(sdd->sgp, cpu); |
e3589f6c | 7181 | atomic_set(&(*sg)->sgp->ref, 1); /* for claim_allocations */ |
9c3f75cb | 7182 | } |
dce840a0 PZ |
7183 | |
7184 | return cpu; | |
1e9f28fa | 7185 | } |
1e9f28fa | 7186 | |
01a08546 | 7187 | /* |
dce840a0 PZ |
7188 | * build_sched_groups will build a circular linked list of the groups |
7189 | * covered by the given span, and will set each group's ->cpumask correctly, | |
7190 | * and ->cpu_power to 0. | |
e3589f6c PZ |
7191 | * |
7192 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 7193 | */ |
e3589f6c PZ |
7194 | static int |
7195 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 7196 | { |
dce840a0 PZ |
7197 | struct sched_group *first = NULL, *last = NULL; |
7198 | struct sd_data *sdd = sd->private; | |
7199 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 7200 | struct cpumask *covered; |
dce840a0 | 7201 | int i; |
9c1cfda2 | 7202 | |
e3589f6c PZ |
7203 | get_group(cpu, sdd, &sd->groups); |
7204 | atomic_inc(&sd->groups->ref); | |
7205 | ||
7206 | if (cpu != cpumask_first(sched_domain_span(sd))) | |
7207 | return 0; | |
7208 | ||
f96225fd PZ |
7209 | lockdep_assert_held(&sched_domains_mutex); |
7210 | covered = sched_domains_tmpmask; | |
7211 | ||
dce840a0 | 7212 | cpumask_clear(covered); |
6711cab4 | 7213 | |
dce840a0 PZ |
7214 | for_each_cpu(i, span) { |
7215 | struct sched_group *sg; | |
7216 | int group = get_group(i, sdd, &sg); | |
7217 | int j; | |
6711cab4 | 7218 | |
dce840a0 PZ |
7219 | if (cpumask_test_cpu(i, covered)) |
7220 | continue; | |
6711cab4 | 7221 | |
dce840a0 | 7222 | cpumask_clear(sched_group_cpus(sg)); |
9c3f75cb | 7223 | sg->sgp->power = 0; |
0601a88d | 7224 | |
dce840a0 PZ |
7225 | for_each_cpu(j, span) { |
7226 | if (get_group(j, sdd, NULL) != group) | |
7227 | continue; | |
0601a88d | 7228 | |
dce840a0 PZ |
7229 | cpumask_set_cpu(j, covered); |
7230 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
7231 | } | |
0601a88d | 7232 | |
dce840a0 PZ |
7233 | if (!first) |
7234 | first = sg; | |
7235 | if (last) | |
7236 | last->next = sg; | |
7237 | last = sg; | |
7238 | } | |
7239 | last->next = first; | |
e3589f6c PZ |
7240 | |
7241 | return 0; | |
0601a88d | 7242 | } |
51888ca2 | 7243 | |
89c4710e SS |
7244 | /* |
7245 | * Initialize sched groups cpu_power. | |
7246 | * | |
7247 | * cpu_power indicates the capacity of sched group, which is used while | |
7248 | * distributing the load between different sched groups in a sched domain. | |
7249 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
7250 | * there are asymmetries in the topology. If there are asymmetries, group | |
7251 | * having more cpu_power will pickup more load compared to the group having | |
7252 | * less cpu_power. | |
89c4710e SS |
7253 | */ |
7254 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
7255 | { | |
e3589f6c | 7256 | struct sched_group *sg = sd->groups; |
89c4710e | 7257 | |
e3589f6c PZ |
7258 | WARN_ON(!sd || !sg); |
7259 | ||
7260 | do { | |
7261 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
7262 | sg = sg->next; | |
7263 | } while (sg != sd->groups); | |
89c4710e | 7264 | |
e3589f6c PZ |
7265 | if (cpu != group_first_cpu(sg)) |
7266 | return; | |
aae6d3dd | 7267 | |
d274cb30 | 7268 | update_group_power(sd, cpu); |
89c4710e SS |
7269 | } |
7270 | ||
7c16ec58 MT |
7271 | /* |
7272 | * Initializers for schedule domains | |
7273 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7274 | */ | |
7275 | ||
a5d8c348 IM |
7276 | #ifdef CONFIG_SCHED_DEBUG |
7277 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
7278 | #else | |
7279 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
7280 | #endif | |
7281 | ||
54ab4ff4 PZ |
7282 | #define SD_INIT_FUNC(type) \ |
7283 | static noinline struct sched_domain * \ | |
7284 | sd_init_##type(struct sched_domain_topology_level *tl, int cpu) \ | |
7285 | { \ | |
7286 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); \ | |
7287 | *sd = SD_##type##_INIT; \ | |
54ab4ff4 PZ |
7288 | SD_INIT_NAME(sd, type); \ |
7289 | sd->private = &tl->data; \ | |
7290 | return sd; \ | |
7c16ec58 MT |
7291 | } |
7292 | ||
7293 | SD_INIT_FUNC(CPU) | |
7294 | #ifdef CONFIG_NUMA | |
7295 | SD_INIT_FUNC(ALLNODES) | |
7296 | SD_INIT_FUNC(NODE) | |
7297 | #endif | |
7298 | #ifdef CONFIG_SCHED_SMT | |
7299 | SD_INIT_FUNC(SIBLING) | |
7300 | #endif | |
7301 | #ifdef CONFIG_SCHED_MC | |
7302 | SD_INIT_FUNC(MC) | |
7303 | #endif | |
01a08546 HC |
7304 | #ifdef CONFIG_SCHED_BOOK |
7305 | SD_INIT_FUNC(BOOK) | |
7306 | #endif | |
7c16ec58 | 7307 | |
1d3504fc | 7308 | static int default_relax_domain_level = -1; |
60495e77 | 7309 | int sched_domain_level_max; |
1d3504fc HS |
7310 | |
7311 | static int __init setup_relax_domain_level(char *str) | |
7312 | { | |
30e0e178 LZ |
7313 | unsigned long val; |
7314 | ||
7315 | val = simple_strtoul(str, NULL, 0); | |
60495e77 | 7316 | if (val < sched_domain_level_max) |
30e0e178 LZ |
7317 | default_relax_domain_level = val; |
7318 | ||
1d3504fc HS |
7319 | return 1; |
7320 | } | |
7321 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7322 | ||
7323 | static void set_domain_attribute(struct sched_domain *sd, | |
7324 | struct sched_domain_attr *attr) | |
7325 | { | |
7326 | int request; | |
7327 | ||
7328 | if (!attr || attr->relax_domain_level < 0) { | |
7329 | if (default_relax_domain_level < 0) | |
7330 | return; | |
7331 | else | |
7332 | request = default_relax_domain_level; | |
7333 | } else | |
7334 | request = attr->relax_domain_level; | |
7335 | if (request < sd->level) { | |
7336 | /* turn off idle balance on this domain */ | |
c88d5910 | 7337 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7338 | } else { |
7339 | /* turn on idle balance on this domain */ | |
c88d5910 | 7340 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7341 | } |
7342 | } | |
7343 | ||
54ab4ff4 PZ |
7344 | static void __sdt_free(const struct cpumask *cpu_map); |
7345 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
7346 | ||
2109b99e AH |
7347 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
7348 | const struct cpumask *cpu_map) | |
7349 | { | |
7350 | switch (what) { | |
2109b99e | 7351 | case sa_rootdomain: |
822ff793 PZ |
7352 | if (!atomic_read(&d->rd->refcount)) |
7353 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
7354 | case sa_sd: |
7355 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 7356 | case sa_sd_storage: |
54ab4ff4 | 7357 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
7358 | case sa_none: |
7359 | break; | |
7360 | } | |
7361 | } | |
3404c8d9 | 7362 | |
2109b99e AH |
7363 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
7364 | const struct cpumask *cpu_map) | |
7365 | { | |
dce840a0 PZ |
7366 | memset(d, 0, sizeof(*d)); |
7367 | ||
54ab4ff4 PZ |
7368 | if (__sdt_alloc(cpu_map)) |
7369 | return sa_sd_storage; | |
dce840a0 PZ |
7370 | d->sd = alloc_percpu(struct sched_domain *); |
7371 | if (!d->sd) | |
7372 | return sa_sd_storage; | |
2109b99e | 7373 | d->rd = alloc_rootdomain(); |
dce840a0 | 7374 | if (!d->rd) |
21d42ccf | 7375 | return sa_sd; |
2109b99e AH |
7376 | return sa_rootdomain; |
7377 | } | |
57d885fe | 7378 | |
dce840a0 PZ |
7379 | /* |
7380 | * NULL the sd_data elements we've used to build the sched_domain and | |
7381 | * sched_group structure so that the subsequent __free_domain_allocs() | |
7382 | * will not free the data we're using. | |
7383 | */ | |
7384 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
7385 | { | |
7386 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
7387 | |
7388 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
7389 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
7390 | ||
e3589f6c | 7391 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 7392 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c PZ |
7393 | |
7394 | if (atomic_read(&(*per_cpu_ptr(sdd->sgp, cpu))->ref)) | |
9c3f75cb | 7395 | *per_cpu_ptr(sdd->sgp, cpu) = NULL; |
dce840a0 PZ |
7396 | } |
7397 | ||
2c402dc3 PZ |
7398 | #ifdef CONFIG_SCHED_SMT |
7399 | static const struct cpumask *cpu_smt_mask(int cpu) | |
7f4588f3 | 7400 | { |
2c402dc3 | 7401 | return topology_thread_cpumask(cpu); |
3bd65a80 | 7402 | } |
2c402dc3 | 7403 | #endif |
7f4588f3 | 7404 | |
d069b916 PZ |
7405 | /* |
7406 | * Topology list, bottom-up. | |
7407 | */ | |
2c402dc3 | 7408 | static struct sched_domain_topology_level default_topology[] = { |
d069b916 PZ |
7409 | #ifdef CONFIG_SCHED_SMT |
7410 | { sd_init_SIBLING, cpu_smt_mask, }, | |
01a08546 | 7411 | #endif |
1e9f28fa | 7412 | #ifdef CONFIG_SCHED_MC |
2c402dc3 | 7413 | { sd_init_MC, cpu_coregroup_mask, }, |
1e9f28fa | 7414 | #endif |
d069b916 PZ |
7415 | #ifdef CONFIG_SCHED_BOOK |
7416 | { sd_init_BOOK, cpu_book_mask, }, | |
7417 | #endif | |
7418 | { sd_init_CPU, cpu_cpu_mask, }, | |
7419 | #ifdef CONFIG_NUMA | |
e3589f6c | 7420 | { sd_init_NODE, cpu_node_mask, SDTL_OVERLAP, }, |
d069b916 | 7421 | { sd_init_ALLNODES, cpu_allnodes_mask, }, |
1da177e4 | 7422 | #endif |
eb7a74e6 PZ |
7423 | { NULL, }, |
7424 | }; | |
7425 | ||
7426 | static struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
7427 | ||
54ab4ff4 PZ |
7428 | static int __sdt_alloc(const struct cpumask *cpu_map) |
7429 | { | |
7430 | struct sched_domain_topology_level *tl; | |
7431 | int j; | |
7432 | ||
7433 | for (tl = sched_domain_topology; tl->init; tl++) { | |
7434 | struct sd_data *sdd = &tl->data; | |
7435 | ||
7436 | sdd->sd = alloc_percpu(struct sched_domain *); | |
7437 | if (!sdd->sd) | |
7438 | return -ENOMEM; | |
7439 | ||
7440 | sdd->sg = alloc_percpu(struct sched_group *); | |
7441 | if (!sdd->sg) | |
7442 | return -ENOMEM; | |
7443 | ||
9c3f75cb PZ |
7444 | sdd->sgp = alloc_percpu(struct sched_group_power *); |
7445 | if (!sdd->sgp) | |
7446 | return -ENOMEM; | |
7447 | ||
54ab4ff4 PZ |
7448 | for_each_cpu(j, cpu_map) { |
7449 | struct sched_domain *sd; | |
7450 | struct sched_group *sg; | |
9c3f75cb | 7451 | struct sched_group_power *sgp; |
54ab4ff4 PZ |
7452 | |
7453 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
7454 | GFP_KERNEL, cpu_to_node(j)); | |
7455 | if (!sd) | |
7456 | return -ENOMEM; | |
7457 | ||
7458 | *per_cpu_ptr(sdd->sd, j) = sd; | |
7459 | ||
7460 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
7461 | GFP_KERNEL, cpu_to_node(j)); | |
7462 | if (!sg) | |
7463 | return -ENOMEM; | |
7464 | ||
7465 | *per_cpu_ptr(sdd->sg, j) = sg; | |
9c3f75cb PZ |
7466 | |
7467 | sgp = kzalloc_node(sizeof(struct sched_group_power), | |
7468 | GFP_KERNEL, cpu_to_node(j)); | |
7469 | if (!sgp) | |
7470 | return -ENOMEM; | |
7471 | ||
7472 | *per_cpu_ptr(sdd->sgp, j) = sgp; | |
54ab4ff4 PZ |
7473 | } |
7474 | } | |
7475 | ||
7476 | return 0; | |
7477 | } | |
7478 | ||
7479 | static void __sdt_free(const struct cpumask *cpu_map) | |
7480 | { | |
7481 | struct sched_domain_topology_level *tl; | |
7482 | int j; | |
7483 | ||
7484 | for (tl = sched_domain_topology; tl->init; tl++) { | |
7485 | struct sd_data *sdd = &tl->data; | |
7486 | ||
7487 | for_each_cpu(j, cpu_map) { | |
e3589f6c PZ |
7488 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, j); |
7489 | if (sd && (sd->flags & SD_OVERLAP)) | |
7490 | free_sched_groups(sd->groups, 0); | |
54ab4ff4 | 7491 | kfree(*per_cpu_ptr(sdd->sg, j)); |
9c3f75cb | 7492 | kfree(*per_cpu_ptr(sdd->sgp, j)); |
54ab4ff4 PZ |
7493 | } |
7494 | free_percpu(sdd->sd); | |
7495 | free_percpu(sdd->sg); | |
9c3f75cb | 7496 | free_percpu(sdd->sgp); |
54ab4ff4 PZ |
7497 | } |
7498 | } | |
7499 | ||
2c402dc3 PZ |
7500 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
7501 | struct s_data *d, const struct cpumask *cpu_map, | |
d069b916 | 7502 | struct sched_domain_attr *attr, struct sched_domain *child, |
2c402dc3 PZ |
7503 | int cpu) |
7504 | { | |
54ab4ff4 | 7505 | struct sched_domain *sd = tl->init(tl, cpu); |
2c402dc3 | 7506 | if (!sd) |
d069b916 | 7507 | return child; |
2c402dc3 PZ |
7508 | |
7509 | set_domain_attribute(sd, attr); | |
7510 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); | |
60495e77 PZ |
7511 | if (child) { |
7512 | sd->level = child->level + 1; | |
7513 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 7514 | child->parent = sd; |
60495e77 | 7515 | } |
d069b916 | 7516 | sd->child = child; |
2c402dc3 PZ |
7517 | |
7518 | return sd; | |
7519 | } | |
7520 | ||
2109b99e AH |
7521 | /* |
7522 | * Build sched domains for a given set of cpus and attach the sched domains | |
7523 | * to the individual cpus | |
7524 | */ | |
dce840a0 PZ |
7525 | static int build_sched_domains(const struct cpumask *cpu_map, |
7526 | struct sched_domain_attr *attr) | |
2109b99e AH |
7527 | { |
7528 | enum s_alloc alloc_state = sa_none; | |
dce840a0 | 7529 | struct sched_domain *sd; |
2109b99e | 7530 | struct s_data d; |
822ff793 | 7531 | int i, ret = -ENOMEM; |
9c1cfda2 | 7532 | |
2109b99e AH |
7533 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
7534 | if (alloc_state != sa_rootdomain) | |
7535 | goto error; | |
9c1cfda2 | 7536 | |
dce840a0 | 7537 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 7538 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
7539 | struct sched_domain_topology_level *tl; |
7540 | ||
3bd65a80 | 7541 | sd = NULL; |
e3589f6c | 7542 | for (tl = sched_domain_topology; tl->init; tl++) { |
2c402dc3 | 7543 | sd = build_sched_domain(tl, &d, cpu_map, attr, sd, i); |
e3589f6c PZ |
7544 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
7545 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
7546 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
7547 | break; | |
e3589f6c | 7548 | } |
d274cb30 | 7549 | |
d069b916 PZ |
7550 | while (sd->child) |
7551 | sd = sd->child; | |
7552 | ||
21d42ccf | 7553 | *per_cpu_ptr(d.sd, i) = sd; |
dce840a0 PZ |
7554 | } |
7555 | ||
7556 | /* Build the groups for the domains */ | |
7557 | for_each_cpu(i, cpu_map) { | |
7558 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
7559 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
7560 | if (sd->flags & SD_OVERLAP) { |
7561 | if (build_overlap_sched_groups(sd, i)) | |
7562 | goto error; | |
7563 | } else { | |
7564 | if (build_sched_groups(sd, i)) | |
7565 | goto error; | |
7566 | } | |
1cf51902 | 7567 | } |
a06dadbe | 7568 | } |
9c1cfda2 | 7569 | |
1da177e4 | 7570 | /* Calculate CPU power for physical packages and nodes */ |
a9c9a9b6 PZ |
7571 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
7572 | if (!cpumask_test_cpu(i, cpu_map)) | |
7573 | continue; | |
9c1cfda2 | 7574 | |
dce840a0 PZ |
7575 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
7576 | claim_allocations(i, sd); | |
cd4ea6ae | 7577 | init_sched_groups_power(i, sd); |
dce840a0 | 7578 | } |
f712c0c7 | 7579 | } |
9c1cfda2 | 7580 | |
1da177e4 | 7581 | /* Attach the domains */ |
dce840a0 | 7582 | rcu_read_lock(); |
abcd083a | 7583 | for_each_cpu(i, cpu_map) { |
21d42ccf | 7584 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 7585 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 7586 | } |
dce840a0 | 7587 | rcu_read_unlock(); |
51888ca2 | 7588 | |
822ff793 | 7589 | ret = 0; |
51888ca2 | 7590 | error: |
2109b99e | 7591 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 7592 | return ret; |
1da177e4 | 7593 | } |
029190c5 | 7594 | |
acc3f5d7 | 7595 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7596 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7597 | static struct sched_domain_attr *dattr_cur; |
7598 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7599 | |
7600 | /* | |
7601 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7602 | * cpumask) fails, then fallback to a single sched domain, |
7603 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7604 | */ |
4212823f | 7605 | static cpumask_var_t fallback_doms; |
029190c5 | 7606 | |
ee79d1bd HC |
7607 | /* |
7608 | * arch_update_cpu_topology lets virtualized architectures update the | |
7609 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7610 | * or 0 if it stayed the same. | |
7611 | */ | |
7612 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7613 | { |
ee79d1bd | 7614 | return 0; |
22e52b07 HC |
7615 | } |
7616 | ||
acc3f5d7 RR |
7617 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7618 | { | |
7619 | int i; | |
7620 | cpumask_var_t *doms; | |
7621 | ||
7622 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7623 | if (!doms) | |
7624 | return NULL; | |
7625 | for (i = 0; i < ndoms; i++) { | |
7626 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7627 | free_sched_domains(doms, i); | |
7628 | return NULL; | |
7629 | } | |
7630 | } | |
7631 | return doms; | |
7632 | } | |
7633 | ||
7634 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7635 | { | |
7636 | unsigned int i; | |
7637 | for (i = 0; i < ndoms; i++) | |
7638 | free_cpumask_var(doms[i]); | |
7639 | kfree(doms); | |
7640 | } | |
7641 | ||
1a20ff27 | 7642 | /* |
41a2d6cf | 7643 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7644 | * For now this just excludes isolated cpus, but could be used to |
7645 | * exclude other special cases in the future. | |
1a20ff27 | 7646 | */ |
c4a8849a | 7647 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7648 | { |
7378547f MM |
7649 | int err; |
7650 | ||
22e52b07 | 7651 | arch_update_cpu_topology(); |
029190c5 | 7652 | ndoms_cur = 1; |
acc3f5d7 | 7653 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7654 | if (!doms_cur) |
acc3f5d7 RR |
7655 | doms_cur = &fallback_doms; |
7656 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 7657 | dattr_cur = NULL; |
dce840a0 | 7658 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 7659 | register_sched_domain_sysctl(); |
7378547f MM |
7660 | |
7661 | return err; | |
1a20ff27 DG |
7662 | } |
7663 | ||
1a20ff27 DG |
7664 | /* |
7665 | * Detach sched domains from a group of cpus specified in cpu_map | |
7666 | * These cpus will now be attached to the NULL domain | |
7667 | */ | |
96f874e2 | 7668 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
7669 | { |
7670 | int i; | |
7671 | ||
dce840a0 | 7672 | rcu_read_lock(); |
abcd083a | 7673 | for_each_cpu(i, cpu_map) |
57d885fe | 7674 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 7675 | rcu_read_unlock(); |
1a20ff27 DG |
7676 | } |
7677 | ||
1d3504fc HS |
7678 | /* handle null as "default" */ |
7679 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7680 | struct sched_domain_attr *new, int idx_new) | |
7681 | { | |
7682 | struct sched_domain_attr tmp; | |
7683 | ||
7684 | /* fast path */ | |
7685 | if (!new && !cur) | |
7686 | return 1; | |
7687 | ||
7688 | tmp = SD_ATTR_INIT; | |
7689 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7690 | new ? (new + idx_new) : &tmp, | |
7691 | sizeof(struct sched_domain_attr)); | |
7692 | } | |
7693 | ||
029190c5 PJ |
7694 | /* |
7695 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7696 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7697 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7698 | * It destroys each deleted domain and builds each new domain. | |
7699 | * | |
acc3f5d7 | 7700 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7701 | * The masks don't intersect (don't overlap.) We should setup one |
7702 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7703 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7704 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7705 | * it as it is. | |
7706 | * | |
acc3f5d7 RR |
7707 | * The passed in 'doms_new' should be allocated using |
7708 | * alloc_sched_domains. This routine takes ownership of it and will | |
7709 | * free_sched_domains it when done with it. If the caller failed the | |
7710 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7711 | * and partition_sched_domains() will fallback to the single partition | |
7712 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7713 | * |
96f874e2 | 7714 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7715 | * ndoms_new == 0 is a special case for destroying existing domains, |
7716 | * and it will not create the default domain. | |
dfb512ec | 7717 | * |
029190c5 PJ |
7718 | * Call with hotplug lock held |
7719 | */ | |
acc3f5d7 | 7720 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7721 | struct sched_domain_attr *dattr_new) |
029190c5 | 7722 | { |
dfb512ec | 7723 | int i, j, n; |
d65bd5ec | 7724 | int new_topology; |
029190c5 | 7725 | |
712555ee | 7726 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7727 | |
7378547f MM |
7728 | /* always unregister in case we don't destroy any domains */ |
7729 | unregister_sched_domain_sysctl(); | |
7730 | ||
d65bd5ec HC |
7731 | /* Let architecture update cpu core mappings. */ |
7732 | new_topology = arch_update_cpu_topology(); | |
7733 | ||
dfb512ec | 7734 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7735 | |
7736 | /* Destroy deleted domains */ | |
7737 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7738 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7739 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7740 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7741 | goto match1; |
7742 | } | |
7743 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7744 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7745 | match1: |
7746 | ; | |
7747 | } | |
7748 | ||
e761b772 MK |
7749 | if (doms_new == NULL) { |
7750 | ndoms_cur = 0; | |
acc3f5d7 | 7751 | doms_new = &fallback_doms; |
6ad4c188 | 7752 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7753 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7754 | } |
7755 | ||
029190c5 PJ |
7756 | /* Build new domains */ |
7757 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7758 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7759 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7760 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7761 | goto match2; |
7762 | } | |
7763 | /* no match - add a new doms_new */ | |
dce840a0 | 7764 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7765 | match2: |
7766 | ; | |
7767 | } | |
7768 | ||
7769 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7770 | if (doms_cur != &fallback_doms) |
7771 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7772 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7773 | doms_cur = doms_new; |
1d3504fc | 7774 | dattr_cur = dattr_new; |
029190c5 | 7775 | ndoms_cur = ndoms_new; |
7378547f MM |
7776 | |
7777 | register_sched_domain_sysctl(); | |
a1835615 | 7778 | |
712555ee | 7779 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7780 | } |
7781 | ||
5c45bf27 | 7782 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c4a8849a | 7783 | static void reinit_sched_domains(void) |
5c45bf27 | 7784 | { |
95402b38 | 7785 | get_online_cpus(); |
dfb512ec MK |
7786 | |
7787 | /* Destroy domains first to force the rebuild */ | |
7788 | partition_sched_domains(0, NULL, NULL); | |
7789 | ||
e761b772 | 7790 | rebuild_sched_domains(); |
95402b38 | 7791 | put_online_cpus(); |
5c45bf27 SS |
7792 | } |
7793 | ||
7794 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7795 | { | |
afb8a9b7 | 7796 | unsigned int level = 0; |
5c45bf27 | 7797 | |
afb8a9b7 GS |
7798 | if (sscanf(buf, "%u", &level) != 1) |
7799 | return -EINVAL; | |
7800 | ||
7801 | /* | |
7802 | * level is always be positive so don't check for | |
7803 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7804 | * What happens on 0 or 1 byte write, | |
7805 | * need to check for count as well? | |
7806 | */ | |
7807 | ||
7808 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7809 | return -EINVAL; |
7810 | ||
7811 | if (smt) | |
afb8a9b7 | 7812 | sched_smt_power_savings = level; |
5c45bf27 | 7813 | else |
afb8a9b7 | 7814 | sched_mc_power_savings = level; |
5c45bf27 | 7815 | |
c4a8849a | 7816 | reinit_sched_domains(); |
5c45bf27 | 7817 | |
c70f22d2 | 7818 | return count; |
5c45bf27 SS |
7819 | } |
7820 | ||
5c45bf27 | 7821 | #ifdef CONFIG_SCHED_MC |
f718cd4a | 7822 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
c9be0a36 | 7823 | struct sysdev_class_attribute *attr, |
f718cd4a | 7824 | char *page) |
5c45bf27 SS |
7825 | { |
7826 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7827 | } | |
f718cd4a | 7828 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
c9be0a36 | 7829 | struct sysdev_class_attribute *attr, |
48f24c4d | 7830 | const char *buf, size_t count) |
5c45bf27 SS |
7831 | { |
7832 | return sched_power_savings_store(buf, count, 0); | |
7833 | } | |
f718cd4a AK |
7834 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7835 | sched_mc_power_savings_show, | |
7836 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7837 | #endif |
7838 | ||
7839 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a | 7840 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
c9be0a36 | 7841 | struct sysdev_class_attribute *attr, |
f718cd4a | 7842 | char *page) |
5c45bf27 SS |
7843 | { |
7844 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7845 | } | |
f718cd4a | 7846 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
c9be0a36 | 7847 | struct sysdev_class_attribute *attr, |
48f24c4d | 7848 | const char *buf, size_t count) |
5c45bf27 SS |
7849 | { |
7850 | return sched_power_savings_store(buf, count, 1); | |
7851 | } | |
f718cd4a AK |
7852 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7853 | sched_smt_power_savings_show, | |
6707de00 AB |
7854 | sched_smt_power_savings_store); |
7855 | #endif | |
7856 | ||
39aac648 | 7857 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
7858 | { |
7859 | int err = 0; | |
7860 | ||
7861 | #ifdef CONFIG_SCHED_SMT | |
7862 | if (smt_capable()) | |
7863 | err = sysfs_create_file(&cls->kset.kobj, | |
7864 | &attr_sched_smt_power_savings.attr); | |
7865 | #endif | |
7866 | #ifdef CONFIG_SCHED_MC | |
7867 | if (!err && mc_capable()) | |
7868 | err = sysfs_create_file(&cls->kset.kobj, | |
7869 | &attr_sched_mc_power_savings.attr); | |
7870 | #endif | |
7871 | return err; | |
7872 | } | |
6d6bc0ad | 7873 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7874 | |
1da177e4 | 7875 | /* |
3a101d05 TH |
7876 | * Update cpusets according to cpu_active mask. If cpusets are |
7877 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7878 | * around partition_sched_domains(). | |
1da177e4 | 7879 | */ |
0b2e918a TH |
7880 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
7881 | void *hcpu) | |
e761b772 | 7882 | { |
3a101d05 | 7883 | switch (action & ~CPU_TASKS_FROZEN) { |
e761b772 | 7884 | case CPU_ONLINE: |
6ad4c188 | 7885 | case CPU_DOWN_FAILED: |
3a101d05 | 7886 | cpuset_update_active_cpus(); |
e761b772 | 7887 | return NOTIFY_OK; |
3a101d05 TH |
7888 | default: |
7889 | return NOTIFY_DONE; | |
7890 | } | |
7891 | } | |
e761b772 | 7892 | |
0b2e918a TH |
7893 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
7894 | void *hcpu) | |
3a101d05 TH |
7895 | { |
7896 | switch (action & ~CPU_TASKS_FROZEN) { | |
7897 | case CPU_DOWN_PREPARE: | |
7898 | cpuset_update_active_cpus(); | |
7899 | return NOTIFY_OK; | |
e761b772 MK |
7900 | default: |
7901 | return NOTIFY_DONE; | |
7902 | } | |
7903 | } | |
e761b772 MK |
7904 | |
7905 | static int update_runtime(struct notifier_block *nfb, | |
7906 | unsigned long action, void *hcpu) | |
1da177e4 | 7907 | { |
7def2be1 PZ |
7908 | int cpu = (int)(long)hcpu; |
7909 | ||
1da177e4 | 7910 | switch (action) { |
1da177e4 | 7911 | case CPU_DOWN_PREPARE: |
8bb78442 | 7912 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7913 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7914 | return NOTIFY_OK; |
7915 | ||
1da177e4 | 7916 | case CPU_DOWN_FAILED: |
8bb78442 | 7917 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7918 | case CPU_ONLINE: |
8bb78442 | 7919 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7920 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7921 | return NOTIFY_OK; |
7922 | ||
1da177e4 LT |
7923 | default: |
7924 | return NOTIFY_DONE; | |
7925 | } | |
1da177e4 | 7926 | } |
1da177e4 LT |
7927 | |
7928 | void __init sched_init_smp(void) | |
7929 | { | |
dcc30a35 RR |
7930 | cpumask_var_t non_isolated_cpus; |
7931 | ||
7932 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7933 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7934 | |
95402b38 | 7935 | get_online_cpus(); |
712555ee | 7936 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 7937 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7938 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7939 | if (cpumask_empty(non_isolated_cpus)) | |
7940 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7941 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7942 | put_online_cpus(); |
e761b772 | 7943 | |
3a101d05 TH |
7944 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
7945 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
7946 | |
7947 | /* RT runtime code needs to handle some hotplug events */ | |
7948 | hotcpu_notifier(update_runtime, 0); | |
7949 | ||
b328ca18 | 7950 | init_hrtick(); |
5c1e1767 NP |
7951 | |
7952 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7953 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7954 | BUG(); |
19978ca6 | 7955 | sched_init_granularity(); |
dcc30a35 | 7956 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7957 | |
0e3900e6 | 7958 | init_sched_rt_class(); |
1da177e4 LT |
7959 | } |
7960 | #else | |
7961 | void __init sched_init_smp(void) | |
7962 | { | |
19978ca6 | 7963 | sched_init_granularity(); |
1da177e4 LT |
7964 | } |
7965 | #endif /* CONFIG_SMP */ | |
7966 | ||
cd1bb94b AB |
7967 | const_debug unsigned int sysctl_timer_migration = 1; |
7968 | ||
1da177e4 LT |
7969 | int in_sched_functions(unsigned long addr) |
7970 | { | |
1da177e4 LT |
7971 | return in_lock_functions(addr) || |
7972 | (addr >= (unsigned long)__sched_text_start | |
7973 | && addr < (unsigned long)__sched_text_end); | |
7974 | } | |
7975 | ||
acb5a9ba | 7976 | static void init_cfs_rq(struct cfs_rq *cfs_rq) |
dd41f596 IM |
7977 | { |
7978 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7979 | INIT_LIST_HEAD(&cfs_rq->tasks); |
67e9fb2a | 7980 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
c64be78f PZ |
7981 | #ifndef CONFIG_64BIT |
7982 | cfs_rq->min_vruntime_copy = cfs_rq->min_vruntime; | |
7983 | #endif | |
dd41f596 IM |
7984 | } |
7985 | ||
fa85ae24 PZ |
7986 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7987 | { | |
7988 | struct rt_prio_array *array; | |
7989 | int i; | |
7990 | ||
7991 | array = &rt_rq->active; | |
7992 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7993 | INIT_LIST_HEAD(array->queue + i); | |
7994 | __clear_bit(i, array->bitmap); | |
7995 | } | |
7996 | /* delimiter for bitsearch: */ | |
7997 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7998 | ||
acb5a9ba | 7999 | #if defined CONFIG_SMP |
e864c499 GH |
8000 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
8001 | rt_rq->highest_prio.next = MAX_RT_PRIO; | |
fa85ae24 | 8002 | rt_rq->rt_nr_migratory = 0; |
fa85ae24 | 8003 | rt_rq->overloaded = 0; |
732375c6 | 8004 | plist_head_init(&rt_rq->pushable_tasks); |
fa85ae24 PZ |
8005 | #endif |
8006 | ||
8007 | rt_rq->rt_time = 0; | |
8008 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 8009 | rt_rq->rt_runtime = 0; |
0986b11b | 8010 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
fa85ae24 PZ |
8011 | } |
8012 | ||
6f505b16 | 8013 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac | 8014 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
3d4b47b4 | 8015 | struct sched_entity *se, int cpu, |
ec7dc8ac | 8016 | struct sched_entity *parent) |
6f505b16 | 8017 | { |
ec7dc8ac | 8018 | struct rq *rq = cpu_rq(cpu); |
acb5a9ba | 8019 | |
6f505b16 | 8020 | cfs_rq->tg = tg; |
acb5a9ba JS |
8021 | cfs_rq->rq = rq; |
8022 | #ifdef CONFIG_SMP | |
8023 | /* allow initial update_cfs_load() to truncate */ | |
8024 | cfs_rq->load_stamp = 1; | |
8025 | #endif | |
ab84d31e | 8026 | init_cfs_rq_runtime(cfs_rq); |
6f505b16 | 8027 | |
acb5a9ba | 8028 | tg->cfs_rq[cpu] = cfs_rq; |
6f505b16 | 8029 | tg->se[cpu] = se; |
acb5a9ba | 8030 | |
07e06b01 | 8031 | /* se could be NULL for root_task_group */ |
354d60c2 DG |
8032 | if (!se) |
8033 | return; | |
8034 | ||
ec7dc8ac DG |
8035 | if (!parent) |
8036 | se->cfs_rq = &rq->cfs; | |
8037 | else | |
8038 | se->cfs_rq = parent->my_q; | |
8039 | ||
6f505b16 | 8040 | se->my_q = cfs_rq; |
9437178f | 8041 | update_load_set(&se->load, 0); |
ec7dc8ac | 8042 | se->parent = parent; |
6f505b16 | 8043 | } |
052f1dc7 | 8044 | #endif |
6f505b16 | 8045 | |
052f1dc7 | 8046 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac | 8047 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
3d4b47b4 | 8048 | struct sched_rt_entity *rt_se, int cpu, |
ec7dc8ac | 8049 | struct sched_rt_entity *parent) |
6f505b16 | 8050 | { |
ec7dc8ac DG |
8051 | struct rq *rq = cpu_rq(cpu); |
8052 | ||
acb5a9ba JS |
8053 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
8054 | rt_rq->rt_nr_boosted = 0; | |
8055 | rt_rq->rq = rq; | |
6f505b16 | 8056 | rt_rq->tg = tg; |
6f505b16 | 8057 | |
acb5a9ba | 8058 | tg->rt_rq[cpu] = rt_rq; |
6f505b16 | 8059 | tg->rt_se[cpu] = rt_se; |
acb5a9ba | 8060 | |
354d60c2 DG |
8061 | if (!rt_se) |
8062 | return; | |
8063 | ||
ec7dc8ac DG |
8064 | if (!parent) |
8065 | rt_se->rt_rq = &rq->rt; | |
8066 | else | |
8067 | rt_se->rt_rq = parent->my_q; | |
8068 | ||
6f505b16 | 8069 | rt_se->my_q = rt_rq; |
ec7dc8ac | 8070 | rt_se->parent = parent; |
6f505b16 PZ |
8071 | INIT_LIST_HEAD(&rt_se->run_list); |
8072 | } | |
8073 | #endif | |
8074 | ||
1da177e4 LT |
8075 | void __init sched_init(void) |
8076 | { | |
dd41f596 | 8077 | int i, j; |
434d53b0 MT |
8078 | unsigned long alloc_size = 0, ptr; |
8079 | ||
8080 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8081 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
8082 | #endif | |
8083 | #ifdef CONFIG_RT_GROUP_SCHED | |
8084 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 8085 | #endif |
df7c8e84 | 8086 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 8087 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 8088 | #endif |
434d53b0 | 8089 | if (alloc_size) { |
36b7b6d4 | 8090 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
8091 | |
8092 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 8093 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
8094 | ptr += nr_cpu_ids * sizeof(void **); |
8095 | ||
07e06b01 | 8096 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 8097 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 8098 | |
6d6bc0ad | 8099 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 8100 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 8101 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
8102 | ptr += nr_cpu_ids * sizeof(void **); |
8103 | ||
07e06b01 | 8104 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
8105 | ptr += nr_cpu_ids * sizeof(void **); |
8106 | ||
6d6bc0ad | 8107 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
8108 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8109 | for_each_possible_cpu(i) { | |
8110 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
8111 | ptr += cpumask_size(); | |
8112 | } | |
8113 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 8114 | } |
dd41f596 | 8115 | |
57d885fe GH |
8116 | #ifdef CONFIG_SMP |
8117 | init_defrootdomain(); | |
8118 | #endif | |
8119 | ||
d0b27fa7 PZ |
8120 | init_rt_bandwidth(&def_rt_bandwidth, |
8121 | global_rt_period(), global_rt_runtime()); | |
8122 | ||
8123 | #ifdef CONFIG_RT_GROUP_SCHED | |
07e06b01 | 8124 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 8125 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 8126 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 8127 | |
7c941438 | 8128 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
8129 | list_add(&root_task_group.list, &task_groups); |
8130 | INIT_LIST_HEAD(&root_task_group.children); | |
5091faa4 | 8131 | autogroup_init(&init_task); |
7c941438 | 8132 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 8133 | |
0a945022 | 8134 | for_each_possible_cpu(i) { |
70b97a7f | 8135 | struct rq *rq; |
1da177e4 LT |
8136 | |
8137 | rq = cpu_rq(i); | |
05fa785c | 8138 | raw_spin_lock_init(&rq->lock); |
7897986b | 8139 | rq->nr_running = 0; |
dce48a84 TG |
8140 | rq->calc_load_active = 0; |
8141 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 8142 | init_cfs_rq(&rq->cfs); |
6f505b16 | 8143 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 8144 | #ifdef CONFIG_FAIR_GROUP_SCHED |
07e06b01 | 8145 | root_task_group.shares = root_task_group_load; |
6f505b16 | 8146 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 8147 | /* |
07e06b01 | 8148 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
8149 | * |
8150 | * In case of task-groups formed thr' the cgroup filesystem, it | |
8151 | * gets 100% of the cpu resources in the system. This overall | |
8152 | * system cpu resource is divided among the tasks of | |
07e06b01 | 8153 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
8154 | * based on each entity's (task or task-group's) weight |
8155 | * (se->load.weight). | |
8156 | * | |
07e06b01 | 8157 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
8158 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
8159 | * then A0's share of the cpu resource is: | |
8160 | * | |
0d905bca | 8161 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 8162 | * |
07e06b01 YZ |
8163 | * We achieve this by letting root_task_group's tasks sit |
8164 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 8165 | */ |
ab84d31e | 8166 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 8167 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
8168 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8169 | ||
8170 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8171 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8172 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
07e06b01 | 8173 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 8174 | #endif |
1da177e4 | 8175 | |
dd41f596 IM |
8176 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
8177 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
8178 | |
8179 | rq->last_load_update_tick = jiffies; | |
8180 | ||
1da177e4 | 8181 | #ifdef CONFIG_SMP |
41c7ce9a | 8182 | rq->sd = NULL; |
57d885fe | 8183 | rq->rd = NULL; |
1399fa78 | 8184 | rq->cpu_power = SCHED_POWER_SCALE; |
3f029d3c | 8185 | rq->post_schedule = 0; |
1da177e4 | 8186 | rq->active_balance = 0; |
dd41f596 | 8187 | rq->next_balance = jiffies; |
1da177e4 | 8188 | rq->push_cpu = 0; |
0a2966b4 | 8189 | rq->cpu = i; |
1f11eb6a | 8190 | rq->online = 0; |
eae0c9df MG |
8191 | rq->idle_stamp = 0; |
8192 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
dc938520 | 8193 | rq_attach_root(rq, &def_root_domain); |
83cd4fe2 VP |
8194 | #ifdef CONFIG_NO_HZ |
8195 | rq->nohz_balance_kick = 0; | |
8196 | init_sched_softirq_csd(&per_cpu(remote_sched_softirq_cb, i)); | |
8197 | #endif | |
1da177e4 | 8198 | #endif |
8f4d37ec | 8199 | init_rq_hrtick(rq); |
1da177e4 | 8200 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8201 | } |
8202 | ||
2dd73a4f | 8203 | set_load_weight(&init_task); |
b50f60ce | 8204 | |
e107be36 AK |
8205 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8206 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8207 | #endif | |
8208 | ||
c9819f45 | 8209 | #ifdef CONFIG_SMP |
962cf36c | 8210 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
8211 | #endif |
8212 | ||
b50f60ce | 8213 | #ifdef CONFIG_RT_MUTEXES |
732375c6 | 8214 | plist_head_init(&init_task.pi_waiters); |
b50f60ce HC |
8215 | #endif |
8216 | ||
1da177e4 LT |
8217 | /* |
8218 | * The boot idle thread does lazy MMU switching as well: | |
8219 | */ | |
8220 | atomic_inc(&init_mm.mm_count); | |
8221 | enter_lazy_tlb(&init_mm, current); | |
8222 | ||
8223 | /* | |
8224 | * Make us the idle thread. Technically, schedule() should not be | |
8225 | * called from this thread, however somewhere below it might be, | |
8226 | * but because we are the idle thread, we just pick up running again | |
8227 | * when this runqueue becomes "idle". | |
8228 | */ | |
8229 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
8230 | |
8231 | calc_load_update = jiffies + LOAD_FREQ; | |
8232 | ||
dd41f596 IM |
8233 | /* |
8234 | * During early bootup we pretend to be a normal task: | |
8235 | */ | |
8236 | current->sched_class = &fair_sched_class; | |
6892b75e | 8237 | |
6a7b3dc3 | 8238 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 8239 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 8240 | #ifdef CONFIG_SMP |
4cb98839 | 8241 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
7d1e6a9b | 8242 | #ifdef CONFIG_NO_HZ |
83cd4fe2 VP |
8243 | zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT); |
8244 | alloc_cpumask_var(&nohz.grp_idle_mask, GFP_NOWAIT); | |
8245 | atomic_set(&nohz.load_balancer, nr_cpu_ids); | |
8246 | atomic_set(&nohz.first_pick_cpu, nr_cpu_ids); | |
8247 | atomic_set(&nohz.second_pick_cpu, nr_cpu_ids); | |
7d1e6a9b | 8248 | #endif |
bdddd296 RR |
8249 | /* May be allocated at isolcpus cmdline parse time */ |
8250 | if (cpu_isolated_map == NULL) | |
8251 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 8252 | #endif /* SMP */ |
6a7b3dc3 | 8253 | |
6892b75e | 8254 | scheduler_running = 1; |
1da177e4 LT |
8255 | } |
8256 | ||
d902db1e | 8257 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
8258 | static inline int preempt_count_equals(int preempt_offset) |
8259 | { | |
234da7bc | 8260 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 8261 | |
4ba8216c | 8262 | return (nested == preempt_offset); |
e4aafea2 FW |
8263 | } |
8264 | ||
d894837f | 8265 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 8266 | { |
1da177e4 LT |
8267 | static unsigned long prev_jiffy; /* ratelimiting */ |
8268 | ||
e4aafea2 FW |
8269 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
8270 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
8271 | return; |
8272 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8273 | return; | |
8274 | prev_jiffy = jiffies; | |
8275 | ||
3df0fc5b PZ |
8276 | printk(KERN_ERR |
8277 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
8278 | file, line); | |
8279 | printk(KERN_ERR | |
8280 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
8281 | in_atomic(), irqs_disabled(), | |
8282 | current->pid, current->comm); | |
aef745fc IM |
8283 | |
8284 | debug_show_held_locks(current); | |
8285 | if (irqs_disabled()) | |
8286 | print_irqtrace_events(current); | |
8287 | dump_stack(); | |
1da177e4 LT |
8288 | } |
8289 | EXPORT_SYMBOL(__might_sleep); | |
8290 | #endif | |
8291 | ||
8292 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8293 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8294 | { | |
da7a735e PZ |
8295 | const struct sched_class *prev_class = p->sched_class; |
8296 | int old_prio = p->prio; | |
3a5e4dc1 | 8297 | int on_rq; |
3e51f33f | 8298 | |
fd2f4419 | 8299 | on_rq = p->on_rq; |
3a5e4dc1 AK |
8300 | if (on_rq) |
8301 | deactivate_task(rq, p, 0); | |
8302 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8303 | if (on_rq) { | |
8304 | activate_task(rq, p, 0); | |
8305 | resched_task(rq->curr); | |
8306 | } | |
da7a735e PZ |
8307 | |
8308 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
8309 | } |
8310 | ||
1da177e4 LT |
8311 | void normalize_rt_tasks(void) |
8312 | { | |
a0f98a1c | 8313 | struct task_struct *g, *p; |
1da177e4 | 8314 | unsigned long flags; |
70b97a7f | 8315 | struct rq *rq; |
1da177e4 | 8316 | |
4cf5d77a | 8317 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8318 | do_each_thread(g, p) { |
178be793 IM |
8319 | /* |
8320 | * Only normalize user tasks: | |
8321 | */ | |
8322 | if (!p->mm) | |
8323 | continue; | |
8324 | ||
6cfb0d5d | 8325 | p->se.exec_start = 0; |
6cfb0d5d | 8326 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
8327 | p->se.statistics.wait_start = 0; |
8328 | p->se.statistics.sleep_start = 0; | |
8329 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 8330 | #endif |
dd41f596 IM |
8331 | |
8332 | if (!rt_task(p)) { | |
8333 | /* | |
8334 | * Renice negative nice level userspace | |
8335 | * tasks back to 0: | |
8336 | */ | |
8337 | if (TASK_NICE(p) < 0 && p->mm) | |
8338 | set_user_nice(p, 0); | |
1da177e4 | 8339 | continue; |
dd41f596 | 8340 | } |
1da177e4 | 8341 | |
1d615482 | 8342 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 8343 | rq = __task_rq_lock(p); |
1da177e4 | 8344 | |
178be793 | 8345 | normalize_task(rq, p); |
3a5e4dc1 | 8346 | |
b29739f9 | 8347 | __task_rq_unlock(rq); |
1d615482 | 8348 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8349 | } while_each_thread(g, p); |
8350 | ||
4cf5d77a | 8351 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8352 | } |
8353 | ||
8354 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 8355 | |
67fc4e0c | 8356 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 8357 | /* |
67fc4e0c | 8358 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
8359 | * |
8360 | * They can only be called when the whole system has been | |
8361 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8362 | * activity can take place. Using them for anything else would | |
8363 | * be a serious bug, and as a result, they aren't even visible | |
8364 | * under any other configuration. | |
8365 | */ | |
8366 | ||
8367 | /** | |
8368 | * curr_task - return the current task for a given cpu. | |
8369 | * @cpu: the processor in question. | |
8370 | * | |
8371 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8372 | */ | |
36c8b586 | 8373 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8374 | { |
8375 | return cpu_curr(cpu); | |
8376 | } | |
8377 | ||
67fc4e0c JW |
8378 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
8379 | ||
8380 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
8381 | /** |
8382 | * set_curr_task - set the current task for a given cpu. | |
8383 | * @cpu: the processor in question. | |
8384 | * @p: the task pointer to set. | |
8385 | * | |
8386 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8387 | * are serviced on a separate stack. It allows the architecture to switch the |
8388 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8389 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8390 | * and caller must save the original value of the current task (see | |
8391 | * curr_task() above) and restore that value before reenabling interrupts and | |
8392 | * re-starting the system. | |
8393 | * | |
8394 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8395 | */ | |
36c8b586 | 8396 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8397 | { |
8398 | cpu_curr(cpu) = p; | |
8399 | } | |
8400 | ||
8401 | #endif | |
29f59db3 | 8402 | |
bccbe08a PZ |
8403 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8404 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8405 | { |
8406 | int i; | |
8407 | ||
ab84d31e PT |
8408 | destroy_cfs_bandwidth(tg_cfs_bandwidth(tg)); |
8409 | ||
6f505b16 PZ |
8410 | for_each_possible_cpu(i) { |
8411 | if (tg->cfs_rq) | |
8412 | kfree(tg->cfs_rq[i]); | |
8413 | if (tg->se) | |
8414 | kfree(tg->se[i]); | |
6f505b16 PZ |
8415 | } |
8416 | ||
8417 | kfree(tg->cfs_rq); | |
8418 | kfree(tg->se); | |
6f505b16 PZ |
8419 | } |
8420 | ||
ec7dc8ac DG |
8421 | static |
8422 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8423 | { |
29f59db3 | 8424 | struct cfs_rq *cfs_rq; |
eab17229 | 8425 | struct sched_entity *se; |
29f59db3 SV |
8426 | int i; |
8427 | ||
434d53b0 | 8428 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8429 | if (!tg->cfs_rq) |
8430 | goto err; | |
434d53b0 | 8431 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8432 | if (!tg->se) |
8433 | goto err; | |
052f1dc7 PZ |
8434 | |
8435 | tg->shares = NICE_0_LOAD; | |
29f59db3 | 8436 | |
ab84d31e PT |
8437 | init_cfs_bandwidth(tg_cfs_bandwidth(tg)); |
8438 | ||
29f59db3 | 8439 | for_each_possible_cpu(i) { |
eab17229 LZ |
8440 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
8441 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
8442 | if (!cfs_rq) |
8443 | goto err; | |
8444 | ||
eab17229 LZ |
8445 | se = kzalloc_node(sizeof(struct sched_entity), |
8446 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 8447 | if (!se) |
dfc12eb2 | 8448 | goto err_free_rq; |
29f59db3 | 8449 | |
acb5a9ba | 8450 | init_cfs_rq(cfs_rq); |
3d4b47b4 | 8451 | init_tg_cfs_entry(tg, cfs_rq, se, i, parent->se[i]); |
bccbe08a PZ |
8452 | } |
8453 | ||
8454 | return 1; | |
8455 | ||
49246274 | 8456 | err_free_rq: |
dfc12eb2 | 8457 | kfree(cfs_rq); |
49246274 | 8458 | err: |
bccbe08a PZ |
8459 | return 0; |
8460 | } | |
8461 | ||
bccbe08a PZ |
8462 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8463 | { | |
3d4b47b4 PZ |
8464 | struct rq *rq = cpu_rq(cpu); |
8465 | unsigned long flags; | |
3d4b47b4 PZ |
8466 | |
8467 | /* | |
8468 | * Only empty task groups can be destroyed; so we can speculatively | |
8469 | * check on_list without danger of it being re-added. | |
8470 | */ | |
8471 | if (!tg->cfs_rq[cpu]->on_list) | |
8472 | return; | |
8473 | ||
8474 | raw_spin_lock_irqsave(&rq->lock, flags); | |
822bc180 | 8475 | list_del_leaf_cfs_rq(tg->cfs_rq[cpu]); |
3d4b47b4 | 8476 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
bccbe08a | 8477 | } |
5f817d67 | 8478 | #else /* !CONFIG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8479 | static inline void free_fair_sched_group(struct task_group *tg) |
8480 | { | |
8481 | } | |
8482 | ||
ec7dc8ac DG |
8483 | static inline |
8484 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8485 | { |
8486 | return 1; | |
8487 | } | |
8488 | ||
bccbe08a PZ |
8489 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8490 | { | |
8491 | } | |
6d6bc0ad | 8492 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8493 | |
8494 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8495 | static void free_rt_sched_group(struct task_group *tg) |
8496 | { | |
8497 | int i; | |
8498 | ||
99bc5242 BL |
8499 | if (tg->rt_se) |
8500 | destroy_rt_bandwidth(&tg->rt_bandwidth); | |
d0b27fa7 | 8501 | |
bccbe08a PZ |
8502 | for_each_possible_cpu(i) { |
8503 | if (tg->rt_rq) | |
8504 | kfree(tg->rt_rq[i]); | |
8505 | if (tg->rt_se) | |
8506 | kfree(tg->rt_se[i]); | |
8507 | } | |
8508 | ||
8509 | kfree(tg->rt_rq); | |
8510 | kfree(tg->rt_se); | |
8511 | } | |
8512 | ||
ec7dc8ac DG |
8513 | static |
8514 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8515 | { |
8516 | struct rt_rq *rt_rq; | |
eab17229 | 8517 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8518 | int i; |
8519 | ||
434d53b0 | 8520 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8521 | if (!tg->rt_rq) |
8522 | goto err; | |
434d53b0 | 8523 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8524 | if (!tg->rt_se) |
8525 | goto err; | |
8526 | ||
d0b27fa7 PZ |
8527 | init_rt_bandwidth(&tg->rt_bandwidth, |
8528 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8529 | |
8530 | for_each_possible_cpu(i) { | |
eab17229 LZ |
8531 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8532 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8533 | if (!rt_rq) |
8534 | goto err; | |
29f59db3 | 8535 | |
eab17229 LZ |
8536 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8537 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 8538 | if (!rt_se) |
dfc12eb2 | 8539 | goto err_free_rq; |
29f59db3 | 8540 | |
acb5a9ba JS |
8541 | init_rt_rq(rt_rq, cpu_rq(i)); |
8542 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; | |
3d4b47b4 | 8543 | init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]); |
29f59db3 SV |
8544 | } |
8545 | ||
bccbe08a PZ |
8546 | return 1; |
8547 | ||
49246274 | 8548 | err_free_rq: |
dfc12eb2 | 8549 | kfree(rt_rq); |
49246274 | 8550 | err: |
bccbe08a PZ |
8551 | return 0; |
8552 | } | |
6d6bc0ad | 8553 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8554 | static inline void free_rt_sched_group(struct task_group *tg) |
8555 | { | |
8556 | } | |
8557 | ||
ec7dc8ac DG |
8558 | static inline |
8559 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8560 | { |
8561 | return 1; | |
8562 | } | |
6d6bc0ad | 8563 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8564 | |
7c941438 | 8565 | #ifdef CONFIG_CGROUP_SCHED |
bccbe08a PZ |
8566 | static void free_sched_group(struct task_group *tg) |
8567 | { | |
8568 | free_fair_sched_group(tg); | |
8569 | free_rt_sched_group(tg); | |
e9aa1dd1 | 8570 | autogroup_free(tg); |
bccbe08a PZ |
8571 | kfree(tg); |
8572 | } | |
8573 | ||
8574 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8575 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8576 | { |
8577 | struct task_group *tg; | |
8578 | unsigned long flags; | |
bccbe08a PZ |
8579 | |
8580 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8581 | if (!tg) | |
8582 | return ERR_PTR(-ENOMEM); | |
8583 | ||
ec7dc8ac | 8584 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8585 | goto err; |
8586 | ||
ec7dc8ac | 8587 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8588 | goto err; |
8589 | ||
8ed36996 | 8590 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 8591 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8592 | |
8593 | WARN_ON(!parent); /* root should already exist */ | |
8594 | ||
8595 | tg->parent = parent; | |
f473aa5e | 8596 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8597 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8598 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8599 | |
9b5b7751 | 8600 | return tg; |
29f59db3 SV |
8601 | |
8602 | err: | |
6f505b16 | 8603 | free_sched_group(tg); |
29f59db3 SV |
8604 | return ERR_PTR(-ENOMEM); |
8605 | } | |
8606 | ||
9b5b7751 | 8607 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8608 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8609 | { |
29f59db3 | 8610 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8611 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8612 | } |
8613 | ||
9b5b7751 | 8614 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8615 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8616 | { |
8ed36996 | 8617 | unsigned long flags; |
9b5b7751 | 8618 | int i; |
29f59db3 | 8619 | |
3d4b47b4 PZ |
8620 | /* end participation in shares distribution */ |
8621 | for_each_possible_cpu(i) | |
bccbe08a | 8622 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
8623 | |
8624 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 8625 | list_del_rcu(&tg->list); |
f473aa5e | 8626 | list_del_rcu(&tg->siblings); |
8ed36996 | 8627 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8628 | |
9b5b7751 | 8629 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8630 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8631 | } |
8632 | ||
9b5b7751 | 8633 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8634 | * The caller of this function should have put the task in its new group |
8635 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8636 | * reflect its new group. | |
9b5b7751 SV |
8637 | */ |
8638 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8639 | { |
8640 | int on_rq, running; | |
8641 | unsigned long flags; | |
8642 | struct rq *rq; | |
8643 | ||
8644 | rq = task_rq_lock(tsk, &flags); | |
8645 | ||
051a1d1a | 8646 | running = task_current(rq, tsk); |
fd2f4419 | 8647 | on_rq = tsk->on_rq; |
29f59db3 | 8648 | |
0e1f3483 | 8649 | if (on_rq) |
29f59db3 | 8650 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8651 | if (unlikely(running)) |
8652 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8653 | |
810b3817 | 8654 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
8655 | if (tsk->sched_class->task_move_group) |
8656 | tsk->sched_class->task_move_group(tsk, on_rq); | |
8657 | else | |
810b3817 | 8658 | #endif |
b2b5ce02 | 8659 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 8660 | |
0e1f3483 HS |
8661 | if (unlikely(running)) |
8662 | tsk->sched_class->set_curr_task(rq); | |
8663 | if (on_rq) | |
371fd7e7 | 8664 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8665 | |
0122ec5b | 8666 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 8667 | } |
7c941438 | 8668 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8669 | |
052f1dc7 | 8670 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8ed36996 PZ |
8671 | static DEFINE_MUTEX(shares_mutex); |
8672 | ||
4cf86d77 | 8673 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8674 | { |
8675 | int i; | |
8ed36996 | 8676 | unsigned long flags; |
c61935fd | 8677 | |
ec7dc8ac DG |
8678 | /* |
8679 | * We can't change the weight of the root cgroup. | |
8680 | */ | |
8681 | if (!tg->se[0]) | |
8682 | return -EINVAL; | |
8683 | ||
cd62287e | 8684 | shares = clamp(shares, scale_load(MIN_SHARES), scale_load(MAX_SHARES)); |
62fb1851 | 8685 | |
8ed36996 | 8686 | mutex_lock(&shares_mutex); |
9b5b7751 | 8687 | if (tg->shares == shares) |
5cb350ba | 8688 | goto done; |
29f59db3 | 8689 | |
9b5b7751 | 8690 | tg->shares = shares; |
c09595f6 | 8691 | for_each_possible_cpu(i) { |
9437178f PT |
8692 | struct rq *rq = cpu_rq(i); |
8693 | struct sched_entity *se; | |
8694 | ||
8695 | se = tg->se[i]; | |
8696 | /* Propagate contribution to hierarchy */ | |
8697 | raw_spin_lock_irqsave(&rq->lock, flags); | |
8698 | for_each_sched_entity(se) | |
6d5ab293 | 8699 | update_cfs_shares(group_cfs_rq(se)); |
9437178f | 8700 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c09595f6 | 8701 | } |
29f59db3 | 8702 | |
5cb350ba | 8703 | done: |
8ed36996 | 8704 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8705 | return 0; |
29f59db3 SV |
8706 | } |
8707 | ||
5cb350ba DG |
8708 | unsigned long sched_group_shares(struct task_group *tg) |
8709 | { | |
8710 | return tg->shares; | |
8711 | } | |
052f1dc7 | 8712 | #endif |
5cb350ba | 8713 | |
a790de99 | 8714 | #if defined(CONFIG_RT_GROUP_SCHED) || defined(CONFIG_CFS_BANDWIDTH) |
9f0c1e56 PZ |
8715 | static unsigned long to_ratio(u64 period, u64 runtime) |
8716 | { | |
8717 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8718 | return 1ULL << 20; |
9f0c1e56 | 8719 | |
9a7e0b18 | 8720 | return div64_u64(runtime << 20, period); |
9f0c1e56 | 8721 | } |
a790de99 PT |
8722 | #endif |
8723 | ||
8724 | #ifdef CONFIG_RT_GROUP_SCHED | |
8725 | /* | |
8726 | * Ensure that the real time constraints are schedulable. | |
8727 | */ | |
8728 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 8729 | |
9a7e0b18 PZ |
8730 | /* Must be called with tasklist_lock held */ |
8731 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8732 | { |
9a7e0b18 | 8733 | struct task_struct *g, *p; |
b40b2e8e | 8734 | |
9a7e0b18 PZ |
8735 | do_each_thread(g, p) { |
8736 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8737 | return 1; | |
8738 | } while_each_thread(g, p); | |
b40b2e8e | 8739 | |
9a7e0b18 PZ |
8740 | return 0; |
8741 | } | |
b40b2e8e | 8742 | |
9a7e0b18 PZ |
8743 | struct rt_schedulable_data { |
8744 | struct task_group *tg; | |
8745 | u64 rt_period; | |
8746 | u64 rt_runtime; | |
8747 | }; | |
b40b2e8e | 8748 | |
a790de99 | 8749 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
8750 | { |
8751 | struct rt_schedulable_data *d = data; | |
8752 | struct task_group *child; | |
8753 | unsigned long total, sum = 0; | |
8754 | u64 period, runtime; | |
b40b2e8e | 8755 | |
9a7e0b18 PZ |
8756 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8757 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8758 | |
9a7e0b18 PZ |
8759 | if (tg == d->tg) { |
8760 | period = d->rt_period; | |
8761 | runtime = d->rt_runtime; | |
b40b2e8e | 8762 | } |
b40b2e8e | 8763 | |
4653f803 PZ |
8764 | /* |
8765 | * Cannot have more runtime than the period. | |
8766 | */ | |
8767 | if (runtime > period && runtime != RUNTIME_INF) | |
8768 | return -EINVAL; | |
6f505b16 | 8769 | |
4653f803 PZ |
8770 | /* |
8771 | * Ensure we don't starve existing RT tasks. | |
8772 | */ | |
9a7e0b18 PZ |
8773 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8774 | return -EBUSY; | |
6f505b16 | 8775 | |
9a7e0b18 | 8776 | total = to_ratio(period, runtime); |
6f505b16 | 8777 | |
4653f803 PZ |
8778 | /* |
8779 | * Nobody can have more than the global setting allows. | |
8780 | */ | |
8781 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8782 | return -EINVAL; | |
6f505b16 | 8783 | |
4653f803 PZ |
8784 | /* |
8785 | * The sum of our children's runtime should not exceed our own. | |
8786 | */ | |
9a7e0b18 PZ |
8787 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8788 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8789 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8790 | |
9a7e0b18 PZ |
8791 | if (child == d->tg) { |
8792 | period = d->rt_period; | |
8793 | runtime = d->rt_runtime; | |
8794 | } | |
6f505b16 | 8795 | |
9a7e0b18 | 8796 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8797 | } |
6f505b16 | 8798 | |
9a7e0b18 PZ |
8799 | if (sum > total) |
8800 | return -EINVAL; | |
8801 | ||
8802 | return 0; | |
6f505b16 PZ |
8803 | } |
8804 | ||
9a7e0b18 | 8805 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8806 | { |
9a7e0b18 PZ |
8807 | struct rt_schedulable_data data = { |
8808 | .tg = tg, | |
8809 | .rt_period = period, | |
8810 | .rt_runtime = runtime, | |
8811 | }; | |
8812 | ||
a790de99 | 8813 | return walk_tg_tree(tg_rt_schedulable, tg_nop, &data); |
521f1a24 DG |
8814 | } |
8815 | ||
ab84d31e | 8816 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 8817 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 8818 | { |
ac086bc2 | 8819 | int i, err = 0; |
9f0c1e56 | 8820 | |
9f0c1e56 | 8821 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8822 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8823 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8824 | if (err) | |
9f0c1e56 | 8825 | goto unlock; |
ac086bc2 | 8826 | |
0986b11b | 8827 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8828 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8829 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8830 | |
8831 | for_each_possible_cpu(i) { | |
8832 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8833 | ||
0986b11b | 8834 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8835 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8836 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8837 | } |
0986b11b | 8838 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 8839 | unlock: |
521f1a24 | 8840 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8841 | mutex_unlock(&rt_constraints_mutex); |
8842 | ||
8843 | return err; | |
6f505b16 PZ |
8844 | } |
8845 | ||
d0b27fa7 PZ |
8846 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8847 | { | |
8848 | u64 rt_runtime, rt_period; | |
8849 | ||
8850 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8851 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8852 | if (rt_runtime_us < 0) | |
8853 | rt_runtime = RUNTIME_INF; | |
8854 | ||
ab84d31e | 8855 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
8856 | } |
8857 | ||
9f0c1e56 PZ |
8858 | long sched_group_rt_runtime(struct task_group *tg) |
8859 | { | |
8860 | u64 rt_runtime_us; | |
8861 | ||
d0b27fa7 | 8862 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8863 | return -1; |
8864 | ||
d0b27fa7 | 8865 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8866 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8867 | return rt_runtime_us; | |
8868 | } | |
d0b27fa7 PZ |
8869 | |
8870 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8871 | { | |
8872 | u64 rt_runtime, rt_period; | |
8873 | ||
8874 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8875 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8876 | ||
619b0488 R |
8877 | if (rt_period == 0) |
8878 | return -EINVAL; | |
8879 | ||
ab84d31e | 8880 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
8881 | } |
8882 | ||
8883 | long sched_group_rt_period(struct task_group *tg) | |
8884 | { | |
8885 | u64 rt_period_us; | |
8886 | ||
8887 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8888 | do_div(rt_period_us, NSEC_PER_USEC); | |
8889 | return rt_period_us; | |
8890 | } | |
8891 | ||
8892 | static int sched_rt_global_constraints(void) | |
8893 | { | |
4653f803 | 8894 | u64 runtime, period; |
d0b27fa7 PZ |
8895 | int ret = 0; |
8896 | ||
ec5d4989 HS |
8897 | if (sysctl_sched_rt_period <= 0) |
8898 | return -EINVAL; | |
8899 | ||
4653f803 PZ |
8900 | runtime = global_rt_runtime(); |
8901 | period = global_rt_period(); | |
8902 | ||
8903 | /* | |
8904 | * Sanity check on the sysctl variables. | |
8905 | */ | |
8906 | if (runtime > period && runtime != RUNTIME_INF) | |
8907 | return -EINVAL; | |
10b612f4 | 8908 | |
d0b27fa7 | 8909 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 8910 | read_lock(&tasklist_lock); |
4653f803 | 8911 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 8912 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
8913 | mutex_unlock(&rt_constraints_mutex); |
8914 | ||
8915 | return ret; | |
8916 | } | |
54e99124 DG |
8917 | |
8918 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
8919 | { | |
8920 | /* Don't accept realtime tasks when there is no way for them to run */ | |
8921 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
8922 | return 0; | |
8923 | ||
8924 | return 1; | |
8925 | } | |
8926 | ||
6d6bc0ad | 8927 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8928 | static int sched_rt_global_constraints(void) |
8929 | { | |
ac086bc2 PZ |
8930 | unsigned long flags; |
8931 | int i; | |
8932 | ||
ec5d4989 HS |
8933 | if (sysctl_sched_rt_period <= 0) |
8934 | return -EINVAL; | |
8935 | ||
60aa605d PZ |
8936 | /* |
8937 | * There's always some RT tasks in the root group | |
8938 | * -- migration, kstopmachine etc.. | |
8939 | */ | |
8940 | if (sysctl_sched_rt_runtime == 0) | |
8941 | return -EBUSY; | |
8942 | ||
0986b11b | 8943 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
8944 | for_each_possible_cpu(i) { |
8945 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8946 | ||
0986b11b | 8947 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8948 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 8949 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8950 | } |
0986b11b | 8951 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 8952 | |
d0b27fa7 PZ |
8953 | return 0; |
8954 | } | |
6d6bc0ad | 8955 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8956 | |
8957 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 8958 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8959 | loff_t *ppos) |
8960 | { | |
8961 | int ret; | |
8962 | int old_period, old_runtime; | |
8963 | static DEFINE_MUTEX(mutex); | |
8964 | ||
8965 | mutex_lock(&mutex); | |
8966 | old_period = sysctl_sched_rt_period; | |
8967 | old_runtime = sysctl_sched_rt_runtime; | |
8968 | ||
8d65af78 | 8969 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
8970 | |
8971 | if (!ret && write) { | |
8972 | ret = sched_rt_global_constraints(); | |
8973 | if (ret) { | |
8974 | sysctl_sched_rt_period = old_period; | |
8975 | sysctl_sched_rt_runtime = old_runtime; | |
8976 | } else { | |
8977 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8978 | def_rt_bandwidth.rt_period = | |
8979 | ns_to_ktime(global_rt_period()); | |
8980 | } | |
8981 | } | |
8982 | mutex_unlock(&mutex); | |
8983 | ||
8984 | return ret; | |
8985 | } | |
68318b8e | 8986 | |
052f1dc7 | 8987 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8988 | |
8989 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8990 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8991 | { |
2b01dfe3 PM |
8992 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8993 | struct task_group, css); | |
68318b8e SV |
8994 | } |
8995 | ||
8996 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8997 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8998 | { |
ec7dc8ac | 8999 | struct task_group *tg, *parent; |
68318b8e | 9000 | |
2b01dfe3 | 9001 | if (!cgrp->parent) { |
68318b8e | 9002 | /* This is early initialization for the top cgroup */ |
07e06b01 | 9003 | return &root_task_group.css; |
68318b8e SV |
9004 | } |
9005 | ||
ec7dc8ac DG |
9006 | parent = cgroup_tg(cgrp->parent); |
9007 | tg = sched_create_group(parent); | |
68318b8e SV |
9008 | if (IS_ERR(tg)) |
9009 | return ERR_PTR(-ENOMEM); | |
9010 | ||
68318b8e SV |
9011 | return &tg->css; |
9012 | } | |
9013 | ||
41a2d6cf IM |
9014 | static void |
9015 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 9016 | { |
2b01dfe3 | 9017 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9018 | |
9019 | sched_destroy_group(tg); | |
9020 | } | |
9021 | ||
41a2d6cf | 9022 | static int |
be367d09 | 9023 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 9024 | { |
b68aa230 | 9025 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 9026 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
9027 | return -EINVAL; |
9028 | #else | |
68318b8e SV |
9029 | /* We don't support RT-tasks being in separate groups */ |
9030 | if (tsk->sched_class != &fair_sched_class) | |
9031 | return -EINVAL; | |
b68aa230 | 9032 | #endif |
be367d09 BB |
9033 | return 0; |
9034 | } | |
68318b8e | 9035 | |
68318b8e | 9036 | static void |
f780bdb7 | 9037 | cpu_cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e SV |
9038 | { |
9039 | sched_move_task(tsk); | |
9040 | } | |
9041 | ||
068c5cc5 | 9042 | static void |
d41d5a01 PZ |
9043 | cpu_cgroup_exit(struct cgroup_subsys *ss, struct cgroup *cgrp, |
9044 | struct cgroup *old_cgrp, struct task_struct *task) | |
068c5cc5 PZ |
9045 | { |
9046 | /* | |
9047 | * cgroup_exit() is called in the copy_process() failure path. | |
9048 | * Ignore this case since the task hasn't ran yet, this avoids | |
9049 | * trying to poke a half freed task state from generic code. | |
9050 | */ | |
9051 | if (!(task->flags & PF_EXITING)) | |
9052 | return; | |
9053 | ||
9054 | sched_move_task(task); | |
9055 | } | |
9056 | ||
052f1dc7 | 9057 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 9058 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 9059 | u64 shareval) |
68318b8e | 9060 | { |
c8b28116 | 9061 | return sched_group_set_shares(cgroup_tg(cgrp), scale_load(shareval)); |
68318b8e SV |
9062 | } |
9063 | ||
f4c753b7 | 9064 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 9065 | { |
2b01dfe3 | 9066 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e | 9067 | |
c8b28116 | 9068 | return (u64) scale_load_down(tg->shares); |
68318b8e | 9069 | } |
ab84d31e PT |
9070 | |
9071 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
9072 | static DEFINE_MUTEX(cfs_constraints_mutex); |
9073 | ||
ab84d31e PT |
9074 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
9075 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
9076 | ||
a790de99 PT |
9077 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
9078 | ||
ab84d31e PT |
9079 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
9080 | { | |
a790de99 | 9081 | int i, ret = 0; |
ab84d31e | 9082 | struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(tg); |
ab84d31e PT |
9083 | |
9084 | if (tg == &root_task_group) | |
9085 | return -EINVAL; | |
9086 | ||
9087 | /* | |
9088 | * Ensure we have at some amount of bandwidth every period. This is | |
9089 | * to prevent reaching a state of large arrears when throttled via | |
9090 | * entity_tick() resulting in prolonged exit starvation. | |
9091 | */ | |
9092 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
9093 | return -EINVAL; | |
9094 | ||
9095 | /* | |
9096 | * Likewise, bound things on the otherside by preventing insane quota | |
9097 | * periods. This also allows us to normalize in computing quota | |
9098 | * feasibility. | |
9099 | */ | |
9100 | if (period > max_cfs_quota_period) | |
9101 | return -EINVAL; | |
9102 | ||
a790de99 PT |
9103 | mutex_lock(&cfs_constraints_mutex); |
9104 | ret = __cfs_schedulable(tg, period, quota); | |
9105 | if (ret) | |
9106 | goto out_unlock; | |
9107 | ||
ab84d31e PT |
9108 | raw_spin_lock_irq(&cfs_b->lock); |
9109 | cfs_b->period = ns_to_ktime(period); | |
9110 | cfs_b->quota = quota; | |
ec12cb7f | 9111 | cfs_b->runtime = quota; |
ab84d31e PT |
9112 | raw_spin_unlock_irq(&cfs_b->lock); |
9113 | ||
9114 | for_each_possible_cpu(i) { | |
9115 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; | |
9116 | struct rq *rq = rq_of(cfs_rq); | |
9117 | ||
9118 | raw_spin_lock_irq(&rq->lock); | |
9119 | cfs_rq->runtime_enabled = quota != RUNTIME_INF; | |
9120 | cfs_rq->runtime_remaining = 0; | |
9121 | raw_spin_unlock_irq(&rq->lock); | |
9122 | } | |
a790de99 PT |
9123 | out_unlock: |
9124 | mutex_unlock(&cfs_constraints_mutex); | |
ab84d31e | 9125 | |
a790de99 | 9126 | return ret; |
ab84d31e PT |
9127 | } |
9128 | ||
9129 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
9130 | { | |
9131 | u64 quota, period; | |
9132 | ||
9133 | period = ktime_to_ns(tg_cfs_bandwidth(tg)->period); | |
9134 | if (cfs_quota_us < 0) | |
9135 | quota = RUNTIME_INF; | |
9136 | else | |
9137 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
9138 | ||
9139 | return tg_set_cfs_bandwidth(tg, period, quota); | |
9140 | } | |
9141 | ||
9142 | long tg_get_cfs_quota(struct task_group *tg) | |
9143 | { | |
9144 | u64 quota_us; | |
9145 | ||
9146 | if (tg_cfs_bandwidth(tg)->quota == RUNTIME_INF) | |
9147 | return -1; | |
9148 | ||
9149 | quota_us = tg_cfs_bandwidth(tg)->quota; | |
9150 | do_div(quota_us, NSEC_PER_USEC); | |
9151 | ||
9152 | return quota_us; | |
9153 | } | |
9154 | ||
9155 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
9156 | { | |
9157 | u64 quota, period; | |
9158 | ||
9159 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
9160 | quota = tg_cfs_bandwidth(tg)->quota; | |
9161 | ||
9162 | if (period <= 0) | |
9163 | return -EINVAL; | |
9164 | ||
9165 | return tg_set_cfs_bandwidth(tg, period, quota); | |
9166 | } | |
9167 | ||
9168 | long tg_get_cfs_period(struct task_group *tg) | |
9169 | { | |
9170 | u64 cfs_period_us; | |
9171 | ||
9172 | cfs_period_us = ktime_to_ns(tg_cfs_bandwidth(tg)->period); | |
9173 | do_div(cfs_period_us, NSEC_PER_USEC); | |
9174 | ||
9175 | return cfs_period_us; | |
9176 | } | |
9177 | ||
9178 | static s64 cpu_cfs_quota_read_s64(struct cgroup *cgrp, struct cftype *cft) | |
9179 | { | |
9180 | return tg_get_cfs_quota(cgroup_tg(cgrp)); | |
9181 | } | |
9182 | ||
9183 | static int cpu_cfs_quota_write_s64(struct cgroup *cgrp, struct cftype *cftype, | |
9184 | s64 cfs_quota_us) | |
9185 | { | |
9186 | return tg_set_cfs_quota(cgroup_tg(cgrp), cfs_quota_us); | |
9187 | } | |
9188 | ||
9189 | static u64 cpu_cfs_period_read_u64(struct cgroup *cgrp, struct cftype *cft) | |
9190 | { | |
9191 | return tg_get_cfs_period(cgroup_tg(cgrp)); | |
9192 | } | |
9193 | ||
9194 | static int cpu_cfs_period_write_u64(struct cgroup *cgrp, struct cftype *cftype, | |
9195 | u64 cfs_period_us) | |
9196 | { | |
9197 | return tg_set_cfs_period(cgroup_tg(cgrp), cfs_period_us); | |
9198 | } | |
9199 | ||
a790de99 PT |
9200 | struct cfs_schedulable_data { |
9201 | struct task_group *tg; | |
9202 | u64 period, quota; | |
9203 | }; | |
9204 | ||
9205 | /* | |
9206 | * normalize group quota/period to be quota/max_period | |
9207 | * note: units are usecs | |
9208 | */ | |
9209 | static u64 normalize_cfs_quota(struct task_group *tg, | |
9210 | struct cfs_schedulable_data *d) | |
9211 | { | |
9212 | u64 quota, period; | |
9213 | ||
9214 | if (tg == d->tg) { | |
9215 | period = d->period; | |
9216 | quota = d->quota; | |
9217 | } else { | |
9218 | period = tg_get_cfs_period(tg); | |
9219 | quota = tg_get_cfs_quota(tg); | |
9220 | } | |
9221 | ||
9222 | /* note: these should typically be equivalent */ | |
9223 | if (quota == RUNTIME_INF || quota == -1) | |
9224 | return RUNTIME_INF; | |
9225 | ||
9226 | return to_ratio(period, quota); | |
9227 | } | |
9228 | ||
9229 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
9230 | { | |
9231 | struct cfs_schedulable_data *d = data; | |
9232 | struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(tg); | |
9233 | s64 quota = 0, parent_quota = -1; | |
9234 | ||
9235 | if (!tg->parent) { | |
9236 | quota = RUNTIME_INF; | |
9237 | } else { | |
9238 | struct cfs_bandwidth *parent_b = tg_cfs_bandwidth(tg->parent); | |
9239 | ||
9240 | quota = normalize_cfs_quota(tg, d); | |
9241 | parent_quota = parent_b->hierarchal_quota; | |
9242 | ||
9243 | /* | |
9244 | * ensure max(child_quota) <= parent_quota, inherit when no | |
9245 | * limit is set | |
9246 | */ | |
9247 | if (quota == RUNTIME_INF) | |
9248 | quota = parent_quota; | |
9249 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
9250 | return -EINVAL; | |
9251 | } | |
9252 | cfs_b->hierarchal_quota = quota; | |
9253 | ||
9254 | return 0; | |
9255 | } | |
9256 | ||
9257 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
9258 | { | |
9259 | struct cfs_schedulable_data data = { | |
9260 | .tg = tg, | |
9261 | .period = period, | |
9262 | .quota = quota, | |
9263 | }; | |
9264 | ||
9265 | if (quota != RUNTIME_INF) { | |
9266 | do_div(data.period, NSEC_PER_USEC); | |
9267 | do_div(data.quota, NSEC_PER_USEC); | |
9268 | } | |
9269 | ||
9270 | return walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
9271 | } | |
ab84d31e | 9272 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 9273 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 9274 | |
052f1dc7 | 9275 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 9276 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 9277 | s64 val) |
6f505b16 | 9278 | { |
06ecb27c | 9279 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
9280 | } |
9281 | ||
06ecb27c | 9282 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 9283 | { |
06ecb27c | 9284 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 9285 | } |
d0b27fa7 PZ |
9286 | |
9287 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
9288 | u64 rt_period_us) | |
9289 | { | |
9290 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
9291 | } | |
9292 | ||
9293 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
9294 | { | |
9295 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
9296 | } | |
6d6bc0ad | 9297 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 9298 | |
fe5c7cc2 | 9299 | static struct cftype cpu_files[] = { |
052f1dc7 | 9300 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
9301 | { |
9302 | .name = "shares", | |
f4c753b7 PM |
9303 | .read_u64 = cpu_shares_read_u64, |
9304 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 9305 | }, |
052f1dc7 | 9306 | #endif |
ab84d31e PT |
9307 | #ifdef CONFIG_CFS_BANDWIDTH |
9308 | { | |
9309 | .name = "cfs_quota_us", | |
9310 | .read_s64 = cpu_cfs_quota_read_s64, | |
9311 | .write_s64 = cpu_cfs_quota_write_s64, | |
9312 | }, | |
9313 | { | |
9314 | .name = "cfs_period_us", | |
9315 | .read_u64 = cpu_cfs_period_read_u64, | |
9316 | .write_u64 = cpu_cfs_period_write_u64, | |
9317 | }, | |
9318 | #endif | |
052f1dc7 | 9319 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9320 | { |
9f0c1e56 | 9321 | .name = "rt_runtime_us", |
06ecb27c PM |
9322 | .read_s64 = cpu_rt_runtime_read, |
9323 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 9324 | }, |
d0b27fa7 PZ |
9325 | { |
9326 | .name = "rt_period_us", | |
f4c753b7 PM |
9327 | .read_u64 = cpu_rt_period_read_uint, |
9328 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 9329 | }, |
052f1dc7 | 9330 | #endif |
68318b8e SV |
9331 | }; |
9332 | ||
9333 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
9334 | { | |
fe5c7cc2 | 9335 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
9336 | } |
9337 | ||
9338 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
9339 | .name = "cpu", |
9340 | .create = cpu_cgroup_create, | |
9341 | .destroy = cpu_cgroup_destroy, | |
f780bdb7 BB |
9342 | .can_attach_task = cpu_cgroup_can_attach_task, |
9343 | .attach_task = cpu_cgroup_attach_task, | |
068c5cc5 | 9344 | .exit = cpu_cgroup_exit, |
38605cae IM |
9345 | .populate = cpu_cgroup_populate, |
9346 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
9347 | .early_init = 1, |
9348 | }; | |
9349 | ||
052f1dc7 | 9350 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
9351 | |
9352 | #ifdef CONFIG_CGROUP_CPUACCT | |
9353 | ||
9354 | /* | |
9355 | * CPU accounting code for task groups. | |
9356 | * | |
9357 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
9358 | * (balbir@in.ibm.com). | |
9359 | */ | |
9360 | ||
934352f2 | 9361 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
9362 | struct cpuacct { |
9363 | struct cgroup_subsys_state css; | |
9364 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43cf38eb | 9365 | u64 __percpu *cpuusage; |
ef12fefa | 9366 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 9367 | struct cpuacct *parent; |
d842de87 SV |
9368 | }; |
9369 | ||
9370 | struct cgroup_subsys cpuacct_subsys; | |
9371 | ||
9372 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 9373 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 9374 | { |
32cd756a | 9375 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
9376 | struct cpuacct, css); |
9377 | } | |
9378 | ||
9379 | /* return cpu accounting group to which this task belongs */ | |
9380 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
9381 | { | |
9382 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
9383 | struct cpuacct, css); | |
9384 | } | |
9385 | ||
9386 | /* create a new cpu accounting group */ | |
9387 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 9388 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
9389 | { |
9390 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 9391 | int i; |
d842de87 SV |
9392 | |
9393 | if (!ca) | |
ef12fefa | 9394 | goto out; |
d842de87 SV |
9395 | |
9396 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
9397 | if (!ca->cpuusage) |
9398 | goto out_free_ca; | |
9399 | ||
9400 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
9401 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
9402 | goto out_free_counters; | |
d842de87 | 9403 | |
934352f2 BR |
9404 | if (cgrp->parent) |
9405 | ca->parent = cgroup_ca(cgrp->parent); | |
9406 | ||
d842de87 | 9407 | return &ca->css; |
ef12fefa BR |
9408 | |
9409 | out_free_counters: | |
9410 | while (--i >= 0) | |
9411 | percpu_counter_destroy(&ca->cpustat[i]); | |
9412 | free_percpu(ca->cpuusage); | |
9413 | out_free_ca: | |
9414 | kfree(ca); | |
9415 | out: | |
9416 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
9417 | } |
9418 | ||
9419 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9420 | static void |
32cd756a | 9421 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9422 | { |
32cd756a | 9423 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 9424 | int i; |
d842de87 | 9425 | |
ef12fefa BR |
9426 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
9427 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
9428 | free_percpu(ca->cpuusage); |
9429 | kfree(ca); | |
9430 | } | |
9431 | ||
720f5498 KC |
9432 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
9433 | { | |
b36128c8 | 9434 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
9435 | u64 data; |
9436 | ||
9437 | #ifndef CONFIG_64BIT | |
9438 | /* | |
9439 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
9440 | */ | |
05fa785c | 9441 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 9442 | data = *cpuusage; |
05fa785c | 9443 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
9444 | #else |
9445 | data = *cpuusage; | |
9446 | #endif | |
9447 | ||
9448 | return data; | |
9449 | } | |
9450 | ||
9451 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
9452 | { | |
b36128c8 | 9453 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
9454 | |
9455 | #ifndef CONFIG_64BIT | |
9456 | /* | |
9457 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
9458 | */ | |
05fa785c | 9459 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 9460 | *cpuusage = val; |
05fa785c | 9461 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
9462 | #else |
9463 | *cpuusage = val; | |
9464 | #endif | |
9465 | } | |
9466 | ||
d842de87 | 9467 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 9468 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9469 | { |
32cd756a | 9470 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9471 | u64 totalcpuusage = 0; |
9472 | int i; | |
9473 | ||
720f5498 KC |
9474 | for_each_present_cpu(i) |
9475 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
9476 | |
9477 | return totalcpuusage; | |
9478 | } | |
9479 | ||
0297b803 DG |
9480 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9481 | u64 reset) | |
9482 | { | |
9483 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9484 | int err = 0; | |
9485 | int i; | |
9486 | ||
9487 | if (reset) { | |
9488 | err = -EINVAL; | |
9489 | goto out; | |
9490 | } | |
9491 | ||
720f5498 KC |
9492 | for_each_present_cpu(i) |
9493 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 9494 | |
0297b803 DG |
9495 | out: |
9496 | return err; | |
9497 | } | |
9498 | ||
e9515c3c KC |
9499 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
9500 | struct seq_file *m) | |
9501 | { | |
9502 | struct cpuacct *ca = cgroup_ca(cgroup); | |
9503 | u64 percpu; | |
9504 | int i; | |
9505 | ||
9506 | for_each_present_cpu(i) { | |
9507 | percpu = cpuacct_cpuusage_read(ca, i); | |
9508 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
9509 | } | |
9510 | seq_printf(m, "\n"); | |
9511 | return 0; | |
9512 | } | |
9513 | ||
ef12fefa BR |
9514 | static const char *cpuacct_stat_desc[] = { |
9515 | [CPUACCT_STAT_USER] = "user", | |
9516 | [CPUACCT_STAT_SYSTEM] = "system", | |
9517 | }; | |
9518 | ||
9519 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
9520 | struct cgroup_map_cb *cb) | |
9521 | { | |
9522 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9523 | int i; | |
9524 | ||
9525 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
9526 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
9527 | val = cputime64_to_clock_t(val); | |
9528 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
9529 | } | |
9530 | return 0; | |
9531 | } | |
9532 | ||
d842de87 SV |
9533 | static struct cftype files[] = { |
9534 | { | |
9535 | .name = "usage", | |
f4c753b7 PM |
9536 | .read_u64 = cpuusage_read, |
9537 | .write_u64 = cpuusage_write, | |
d842de87 | 9538 | }, |
e9515c3c KC |
9539 | { |
9540 | .name = "usage_percpu", | |
9541 | .read_seq_string = cpuacct_percpu_seq_read, | |
9542 | }, | |
ef12fefa BR |
9543 | { |
9544 | .name = "stat", | |
9545 | .read_map = cpuacct_stats_show, | |
9546 | }, | |
d842de87 SV |
9547 | }; |
9548 | ||
32cd756a | 9549 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9550 | { |
32cd756a | 9551 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9552 | } |
9553 | ||
9554 | /* | |
9555 | * charge this task's execution time to its accounting group. | |
9556 | * | |
9557 | * called with rq->lock held. | |
9558 | */ | |
9559 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9560 | { | |
9561 | struct cpuacct *ca; | |
934352f2 | 9562 | int cpu; |
d842de87 | 9563 | |
c40c6f85 | 9564 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
9565 | return; |
9566 | ||
934352f2 | 9567 | cpu = task_cpu(tsk); |
a18b83b7 BR |
9568 | |
9569 | rcu_read_lock(); | |
9570 | ||
d842de87 | 9571 | ca = task_ca(tsk); |
d842de87 | 9572 | |
934352f2 | 9573 | for (; ca; ca = ca->parent) { |
b36128c8 | 9574 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
9575 | *cpuusage += cputime; |
9576 | } | |
a18b83b7 BR |
9577 | |
9578 | rcu_read_unlock(); | |
d842de87 SV |
9579 | } |
9580 | ||
fa535a77 AB |
9581 | /* |
9582 | * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large | |
9583 | * in cputime_t units. As a result, cpuacct_update_stats calls | |
9584 | * percpu_counter_add with values large enough to always overflow the | |
9585 | * per cpu batch limit causing bad SMP scalability. | |
9586 | * | |
9587 | * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we | |
9588 | * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled | |
9589 | * and enabled. We cap it at INT_MAX which is the largest allowed batch value. | |
9590 | */ | |
9591 | #ifdef CONFIG_SMP | |
9592 | #define CPUACCT_BATCH \ | |
9593 | min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX) | |
9594 | #else | |
9595 | #define CPUACCT_BATCH 0 | |
9596 | #endif | |
9597 | ||
ef12fefa BR |
9598 | /* |
9599 | * Charge the system/user time to the task's accounting group. | |
9600 | */ | |
9601 | static void cpuacct_update_stats(struct task_struct *tsk, | |
9602 | enum cpuacct_stat_index idx, cputime_t val) | |
9603 | { | |
9604 | struct cpuacct *ca; | |
fa535a77 | 9605 | int batch = CPUACCT_BATCH; |
ef12fefa BR |
9606 | |
9607 | if (unlikely(!cpuacct_subsys.active)) | |
9608 | return; | |
9609 | ||
9610 | rcu_read_lock(); | |
9611 | ca = task_ca(tsk); | |
9612 | ||
9613 | do { | |
fa535a77 | 9614 | __percpu_counter_add(&ca->cpustat[idx], val, batch); |
ef12fefa BR |
9615 | ca = ca->parent; |
9616 | } while (ca); | |
9617 | rcu_read_unlock(); | |
9618 | } | |
9619 | ||
d842de87 SV |
9620 | struct cgroup_subsys cpuacct_subsys = { |
9621 | .name = "cpuacct", | |
9622 | .create = cpuacct_create, | |
9623 | .destroy = cpuacct_destroy, | |
9624 | .populate = cpuacct_populate, | |
9625 | .subsys_id = cpuacct_subsys_id, | |
9626 | }; | |
9627 | #endif /* CONFIG_CGROUP_CPUACCT */ |