]>
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 | ||
58088ad0 | 199 | static void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period) |
d0b27fa7 | 200 | { |
58088ad0 PT |
201 | unsigned long delta; |
202 | ktime_t soft, hard, now; | |
d0b27fa7 | 203 | |
58088ad0 PT |
204 | for (;;) { |
205 | if (hrtimer_active(period_timer)) | |
206 | break; | |
207 | ||
208 | now = hrtimer_cb_get_time(period_timer); | |
209 | hrtimer_forward(period_timer, now, period); | |
210 | ||
211 | soft = hrtimer_get_softexpires(period_timer); | |
212 | hard = hrtimer_get_expires(period_timer); | |
213 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
214 | __hrtimer_start_range_ns(period_timer, soft, delta, | |
215 | HRTIMER_MODE_ABS_PINNED, 0); | |
216 | } | |
217 | } | |
218 | ||
219 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
220 | { | |
cac64d00 | 221 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
222 | return; |
223 | ||
224 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
225 | return; | |
226 | ||
0986b11b | 227 | raw_spin_lock(&rt_b->rt_runtime_lock); |
58088ad0 | 228 | start_bandwidth_timer(&rt_b->rt_period_timer, rt_b->rt_period); |
0986b11b | 229 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
d0b27fa7 PZ |
230 | } |
231 | ||
232 | #ifdef CONFIG_RT_GROUP_SCHED | |
233 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
234 | { | |
235 | hrtimer_cancel(&rt_b->rt_period_timer); | |
236 | } | |
237 | #endif | |
238 | ||
712555ee | 239 | /* |
c4a8849a | 240 | * sched_domains_mutex serializes calls to init_sched_domains, |
712555ee HC |
241 | * detach_destroy_domains and partition_sched_domains. |
242 | */ | |
243 | static DEFINE_MUTEX(sched_domains_mutex); | |
244 | ||
7c941438 | 245 | #ifdef CONFIG_CGROUP_SCHED |
29f59db3 | 246 | |
68318b8e SV |
247 | #include <linux/cgroup.h> |
248 | ||
29f59db3 SV |
249 | struct cfs_rq; |
250 | ||
6f505b16 PZ |
251 | static LIST_HEAD(task_groups); |
252 | ||
ab84d31e PT |
253 | struct cfs_bandwidth { |
254 | #ifdef CONFIG_CFS_BANDWIDTH | |
255 | raw_spinlock_t lock; | |
256 | ktime_t period; | |
ec12cb7f | 257 | u64 quota, runtime; |
a790de99 | 258 | s64 hierarchal_quota; |
a9cf55b2 | 259 | u64 runtime_expires; |
58088ad0 PT |
260 | |
261 | int idle, timer_active; | |
d8b4986d | 262 | struct hrtimer period_timer, slack_timer; |
85dac906 PT |
263 | struct list_head throttled_cfs_rq; |
264 | ||
e8da1b18 NR |
265 | /* statistics */ |
266 | int nr_periods, nr_throttled; | |
267 | u64 throttled_time; | |
ab84d31e PT |
268 | #endif |
269 | }; | |
270 | ||
29f59db3 | 271 | /* task group related information */ |
4cf86d77 | 272 | struct task_group { |
68318b8e | 273 | struct cgroup_subsys_state css; |
6c415b92 | 274 | |
052f1dc7 | 275 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
276 | /* schedulable entities of this group on each cpu */ |
277 | struct sched_entity **se; | |
278 | /* runqueue "owned" by this group on each cpu */ | |
279 | struct cfs_rq **cfs_rq; | |
280 | unsigned long shares; | |
2069dd75 PZ |
281 | |
282 | atomic_t load_weight; | |
052f1dc7 PZ |
283 | #endif |
284 | ||
285 | #ifdef CONFIG_RT_GROUP_SCHED | |
286 | struct sched_rt_entity **rt_se; | |
287 | struct rt_rq **rt_rq; | |
288 | ||
d0b27fa7 | 289 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 290 | #endif |
6b2d7700 | 291 | |
ae8393e5 | 292 | struct rcu_head rcu; |
6f505b16 | 293 | struct list_head list; |
f473aa5e PZ |
294 | |
295 | struct task_group *parent; | |
296 | struct list_head siblings; | |
297 | struct list_head children; | |
5091faa4 MG |
298 | |
299 | #ifdef CONFIG_SCHED_AUTOGROUP | |
300 | struct autogroup *autogroup; | |
301 | #endif | |
ab84d31e PT |
302 | |
303 | struct cfs_bandwidth cfs_bandwidth; | |
29f59db3 SV |
304 | }; |
305 | ||
3d4b47b4 | 306 | /* task_group_lock serializes the addition/removal of task groups */ |
8ed36996 | 307 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 308 | |
e9036b36 CG |
309 | #ifdef CONFIG_FAIR_GROUP_SCHED |
310 | ||
07e06b01 | 311 | # define ROOT_TASK_GROUP_LOAD NICE_0_LOAD |
052f1dc7 | 312 | |
cb4ad1ff | 313 | /* |
2e084786 LJ |
314 | * A weight of 0 or 1 can cause arithmetics problems. |
315 | * A weight of a cfs_rq is the sum of weights of which entities | |
316 | * are queued on this cfs_rq, so a weight of a entity should not be | |
317 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
318 | * (The default weight is 1024 - so there's no practical |
319 | * limitation from this.) | |
320 | */ | |
cd62287e MG |
321 | #define MIN_SHARES (1UL << 1) |
322 | #define MAX_SHARES (1UL << 18) | |
18d95a28 | 323 | |
07e06b01 | 324 | static int root_task_group_load = ROOT_TASK_GROUP_LOAD; |
052f1dc7 PZ |
325 | #endif |
326 | ||
29f59db3 | 327 | /* Default task group. |
3a252015 | 328 | * Every task in system belong to this group at bootup. |
29f59db3 | 329 | */ |
07e06b01 | 330 | struct task_group root_task_group; |
29f59db3 | 331 | |
7c941438 | 332 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 333 | |
6aa645ea IM |
334 | /* CFS-related fields in a runqueue */ |
335 | struct cfs_rq { | |
336 | struct load_weight load; | |
953bfcd1 | 337 | unsigned long nr_running, h_nr_running; |
6aa645ea | 338 | |
6aa645ea | 339 | u64 exec_clock; |
e9acbff6 | 340 | u64 min_vruntime; |
3fe1698b PZ |
341 | #ifndef CONFIG_64BIT |
342 | u64 min_vruntime_copy; | |
343 | #endif | |
6aa645ea IM |
344 | |
345 | struct rb_root tasks_timeline; | |
346 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
347 | |
348 | struct list_head tasks; | |
349 | struct list_head *balance_iterator; | |
350 | ||
351 | /* | |
352 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
353 | * It is set to NULL otherwise (i.e when none are currently running). |
354 | */ | |
ac53db59 | 355 | struct sched_entity *curr, *next, *last, *skip; |
ddc97297 | 356 | |
4934a4d3 | 357 | #ifdef CONFIG_SCHED_DEBUG |
5ac5c4d6 | 358 | unsigned int nr_spread_over; |
4934a4d3 | 359 | #endif |
ddc97297 | 360 | |
62160e3f | 361 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
362 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
363 | ||
41a2d6cf IM |
364 | /* |
365 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
366 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
367 | * (like users, containers etc.) | |
368 | * | |
369 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
370 | * list is used during load balance. | |
371 | */ | |
3d4b47b4 | 372 | int on_list; |
41a2d6cf IM |
373 | struct list_head leaf_cfs_rq_list; |
374 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
375 | |
376 | #ifdef CONFIG_SMP | |
c09595f6 | 377 | /* |
c8cba857 | 378 | * the part of load.weight contributed by tasks |
c09595f6 | 379 | */ |
c8cba857 | 380 | unsigned long task_weight; |
c09595f6 | 381 | |
c8cba857 PZ |
382 | /* |
383 | * h_load = weight * f(tg) | |
384 | * | |
385 | * Where f(tg) is the recursive weight fraction assigned to | |
386 | * this group. | |
387 | */ | |
388 | unsigned long h_load; | |
c09595f6 | 389 | |
c8cba857 | 390 | /* |
3b3d190e PT |
391 | * Maintaining per-cpu shares distribution for group scheduling |
392 | * | |
393 | * load_stamp is the last time we updated the load average | |
394 | * load_last is the last time we updated the load average and saw load | |
395 | * load_unacc_exec_time is currently unaccounted execution time | |
c8cba857 | 396 | */ |
2069dd75 PZ |
397 | u64 load_avg; |
398 | u64 load_period; | |
3b3d190e | 399 | u64 load_stamp, load_last, load_unacc_exec_time; |
f1d239f7 | 400 | |
2069dd75 | 401 | unsigned long load_contribution; |
c09595f6 | 402 | #endif |
ab84d31e PT |
403 | #ifdef CONFIG_CFS_BANDWIDTH |
404 | int runtime_enabled; | |
a9cf55b2 | 405 | u64 runtime_expires; |
ab84d31e | 406 | s64 runtime_remaining; |
85dac906 | 407 | |
e8da1b18 | 408 | u64 throttled_timestamp; |
64660c86 | 409 | int throttled, throttle_count; |
85dac906 | 410 | struct list_head throttled_list; |
ab84d31e | 411 | #endif |
6aa645ea IM |
412 | #endif |
413 | }; | |
1da177e4 | 414 | |
ab84d31e PT |
415 | #ifdef CONFIG_FAIR_GROUP_SCHED |
416 | #ifdef CONFIG_CFS_BANDWIDTH | |
417 | static inline struct cfs_bandwidth *tg_cfs_bandwidth(struct task_group *tg) | |
418 | { | |
419 | return &tg->cfs_bandwidth; | |
420 | } | |
421 | ||
422 | static inline u64 default_cfs_period(void); | |
58088ad0 | 423 | static int do_sched_cfs_period_timer(struct cfs_bandwidth *cfs_b, int overrun); |
d8b4986d PT |
424 | static void do_sched_cfs_slack_timer(struct cfs_bandwidth *cfs_b); |
425 | ||
426 | static enum hrtimer_restart sched_cfs_slack_timer(struct hrtimer *timer) | |
427 | { | |
428 | struct cfs_bandwidth *cfs_b = | |
429 | container_of(timer, struct cfs_bandwidth, slack_timer); | |
430 | do_sched_cfs_slack_timer(cfs_b); | |
431 | ||
432 | return HRTIMER_NORESTART; | |
433 | } | |
58088ad0 PT |
434 | |
435 | static enum hrtimer_restart sched_cfs_period_timer(struct hrtimer *timer) | |
436 | { | |
437 | struct cfs_bandwidth *cfs_b = | |
438 | container_of(timer, struct cfs_bandwidth, period_timer); | |
439 | ktime_t now; | |
440 | int overrun; | |
441 | int idle = 0; | |
442 | ||
443 | for (;;) { | |
444 | now = hrtimer_cb_get_time(timer); | |
445 | overrun = hrtimer_forward(timer, now, cfs_b->period); | |
446 | ||
447 | if (!overrun) | |
448 | break; | |
449 | ||
450 | idle = do_sched_cfs_period_timer(cfs_b, overrun); | |
451 | } | |
452 | ||
453 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
454 | } | |
ab84d31e PT |
455 | |
456 | static void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b) | |
457 | { | |
458 | raw_spin_lock_init(&cfs_b->lock); | |
ec12cb7f | 459 | cfs_b->runtime = 0; |
ab84d31e PT |
460 | cfs_b->quota = RUNTIME_INF; |
461 | cfs_b->period = ns_to_ktime(default_cfs_period()); | |
58088ad0 | 462 | |
85dac906 | 463 | INIT_LIST_HEAD(&cfs_b->throttled_cfs_rq); |
58088ad0 PT |
464 | hrtimer_init(&cfs_b->period_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
465 | cfs_b->period_timer.function = sched_cfs_period_timer; | |
d8b4986d PT |
466 | hrtimer_init(&cfs_b->slack_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
467 | cfs_b->slack_timer.function = sched_cfs_slack_timer; | |
ab84d31e PT |
468 | } |
469 | ||
470 | static void init_cfs_rq_runtime(struct cfs_rq *cfs_rq) | |
471 | { | |
472 | cfs_rq->runtime_enabled = 0; | |
85dac906 | 473 | INIT_LIST_HEAD(&cfs_rq->throttled_list); |
ab84d31e PT |
474 | } |
475 | ||
58088ad0 PT |
476 | /* requires cfs_b->lock, may release to reprogram timer */ |
477 | static void __start_cfs_bandwidth(struct cfs_bandwidth *cfs_b) | |
478 | { | |
479 | /* | |
480 | * The timer may be active because we're trying to set a new bandwidth | |
481 | * period or because we're racing with the tear-down path | |
482 | * (timer_active==0 becomes visible before the hrtimer call-back | |
483 | * terminates). In either case we ensure that it's re-programmed | |
484 | */ | |
485 | while (unlikely(hrtimer_active(&cfs_b->period_timer))) { | |
486 | raw_spin_unlock(&cfs_b->lock); | |
487 | /* ensure cfs_b->lock is available while we wait */ | |
488 | hrtimer_cancel(&cfs_b->period_timer); | |
489 | ||
490 | raw_spin_lock(&cfs_b->lock); | |
491 | /* if someone else restarted the timer then we're done */ | |
492 | if (cfs_b->timer_active) | |
493 | return; | |
494 | } | |
495 | ||
496 | cfs_b->timer_active = 1; | |
497 | start_bandwidth_timer(&cfs_b->period_timer, cfs_b->period); | |
498 | } | |
499 | ||
ab84d31e | 500 | static void destroy_cfs_bandwidth(struct cfs_bandwidth *cfs_b) |
58088ad0 PT |
501 | { |
502 | hrtimer_cancel(&cfs_b->period_timer); | |
d8b4986d | 503 | hrtimer_cancel(&cfs_b->slack_timer); |
58088ad0 | 504 | } |
ab84d31e PT |
505 | #else |
506 | static void init_cfs_rq_runtime(struct cfs_rq *cfs_rq) {} | |
507 | static void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b) {} | |
508 | static void destroy_cfs_bandwidth(struct cfs_bandwidth *cfs_b) {} | |
509 | ||
510 | static inline struct cfs_bandwidth *tg_cfs_bandwidth(struct task_group *tg) | |
511 | { | |
512 | return NULL; | |
513 | } | |
514 | #endif /* CONFIG_CFS_BANDWIDTH */ | |
515 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
516 | ||
6aa645ea IM |
517 | /* Real-Time classes' related field in a runqueue: */ |
518 | struct rt_rq { | |
519 | struct rt_prio_array active; | |
63489e45 | 520 | unsigned long rt_nr_running; |
052f1dc7 | 521 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
522 | struct { |
523 | int curr; /* highest queued rt task prio */ | |
398a153b | 524 | #ifdef CONFIG_SMP |
e864c499 | 525 | int next; /* next highest */ |
398a153b | 526 | #endif |
e864c499 | 527 | } highest_prio; |
6f505b16 | 528 | #endif |
fa85ae24 | 529 | #ifdef CONFIG_SMP |
73fe6aae | 530 | unsigned long rt_nr_migratory; |
a1ba4d8b | 531 | unsigned long rt_nr_total; |
a22d7fc1 | 532 | int overloaded; |
917b627d | 533 | struct plist_head pushable_tasks; |
fa85ae24 | 534 | #endif |
6f505b16 | 535 | int rt_throttled; |
fa85ae24 | 536 | u64 rt_time; |
ac086bc2 | 537 | u64 rt_runtime; |
ea736ed5 | 538 | /* Nests inside the rq lock: */ |
0986b11b | 539 | raw_spinlock_t rt_runtime_lock; |
6f505b16 | 540 | |
052f1dc7 | 541 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
542 | unsigned long rt_nr_boosted; |
543 | ||
6f505b16 PZ |
544 | struct rq *rq; |
545 | struct list_head leaf_rt_rq_list; | |
546 | struct task_group *tg; | |
6f505b16 | 547 | #endif |
6aa645ea IM |
548 | }; |
549 | ||
57d885fe GH |
550 | #ifdef CONFIG_SMP |
551 | ||
552 | /* | |
553 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
554 | * variables. Each exclusive cpuset essentially defines an island domain by |
555 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
556 | * exclusive cpuset is created, we also create and attach a new root-domain |
557 | * object. | |
558 | * | |
57d885fe GH |
559 | */ |
560 | struct root_domain { | |
561 | atomic_t refcount; | |
26a148eb | 562 | atomic_t rto_count; |
dce840a0 | 563 | struct rcu_head rcu; |
c6c4927b RR |
564 | cpumask_var_t span; |
565 | cpumask_var_t online; | |
637f5085 | 566 | |
0eab9146 | 567 | /* |
637f5085 GH |
568 | * The "RT overload" flag: it gets set if a CPU has more than |
569 | * one runnable RT task. | |
570 | */ | |
c6c4927b | 571 | cpumask_var_t rto_mask; |
6e0534f2 | 572 | struct cpupri cpupri; |
57d885fe GH |
573 | }; |
574 | ||
dc938520 GH |
575 | /* |
576 | * By default the system creates a single root-domain with all cpus as | |
577 | * members (mimicking the global state we have today). | |
578 | */ | |
57d885fe GH |
579 | static struct root_domain def_root_domain; |
580 | ||
ed2d372c | 581 | #endif /* CONFIG_SMP */ |
57d885fe | 582 | |
1da177e4 LT |
583 | /* |
584 | * This is the main, per-CPU runqueue data structure. | |
585 | * | |
586 | * Locking rule: those places that want to lock multiple runqueues | |
587 | * (such as the load balancing or the thread migration code), lock | |
588 | * acquire operations must be ordered by ascending &runqueue. | |
589 | */ | |
70b97a7f | 590 | struct rq { |
d8016491 | 591 | /* runqueue lock: */ |
05fa785c | 592 | raw_spinlock_t lock; |
1da177e4 LT |
593 | |
594 | /* | |
595 | * nr_running and cpu_load should be in the same cacheline because | |
596 | * remote CPUs use both these fields when doing load calculation. | |
597 | */ | |
598 | unsigned long nr_running; | |
6aa645ea IM |
599 | #define CPU_LOAD_IDX_MAX 5 |
600 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
fdf3e95d | 601 | unsigned long last_load_update_tick; |
46cb4b7c | 602 | #ifdef CONFIG_NO_HZ |
39c0cbe2 | 603 | u64 nohz_stamp; |
83cd4fe2 | 604 | unsigned char nohz_balance_kick; |
46cb4b7c | 605 | #endif |
61eadef6 | 606 | int skip_clock_update; |
a64692a3 | 607 | |
d8016491 IM |
608 | /* capture load from *all* tasks on this cpu: */ |
609 | struct load_weight load; | |
6aa645ea IM |
610 | unsigned long nr_load_updates; |
611 | u64 nr_switches; | |
612 | ||
613 | struct cfs_rq cfs; | |
6f505b16 | 614 | struct rt_rq rt; |
6f505b16 | 615 | |
6aa645ea | 616 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
617 | /* list of leaf cfs_rq on this cpu: */ |
618 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
619 | #endif |
620 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 621 | struct list_head leaf_rt_rq_list; |
1da177e4 | 622 | #endif |
1da177e4 LT |
623 | |
624 | /* | |
625 | * This is part of a global counter where only the total sum | |
626 | * over all CPUs matters. A task can increase this counter on | |
627 | * one CPU and if it got migrated afterwards it may decrease | |
628 | * it on another CPU. Always updated under the runqueue lock: | |
629 | */ | |
630 | unsigned long nr_uninterruptible; | |
631 | ||
34f971f6 | 632 | struct task_struct *curr, *idle, *stop; |
c9819f45 | 633 | unsigned long next_balance; |
1da177e4 | 634 | struct mm_struct *prev_mm; |
6aa645ea | 635 | |
3e51f33f | 636 | u64 clock; |
305e6835 | 637 | u64 clock_task; |
6aa645ea | 638 | |
1da177e4 LT |
639 | atomic_t nr_iowait; |
640 | ||
641 | #ifdef CONFIG_SMP | |
0eab9146 | 642 | struct root_domain *rd; |
1da177e4 LT |
643 | struct sched_domain *sd; |
644 | ||
e51fd5e2 PZ |
645 | unsigned long cpu_power; |
646 | ||
a0a522ce | 647 | unsigned char idle_at_tick; |
1da177e4 | 648 | /* For active balancing */ |
3f029d3c | 649 | int post_schedule; |
1da177e4 LT |
650 | int active_balance; |
651 | int push_cpu; | |
969c7921 | 652 | struct cpu_stop_work active_balance_work; |
d8016491 IM |
653 | /* cpu of this runqueue: */ |
654 | int cpu; | |
1f11eb6a | 655 | int online; |
1da177e4 | 656 | |
e9e9250b PZ |
657 | u64 rt_avg; |
658 | u64 age_stamp; | |
1b9508f6 MG |
659 | u64 idle_stamp; |
660 | u64 avg_idle; | |
1da177e4 LT |
661 | #endif |
662 | ||
aa483808 VP |
663 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
664 | u64 prev_irq_time; | |
665 | #endif | |
e6e6685a GC |
666 | #ifdef CONFIG_PARAVIRT |
667 | u64 prev_steal_time; | |
668 | #endif | |
095c0aa8 GC |
669 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING |
670 | u64 prev_steal_time_rq; | |
671 | #endif | |
aa483808 | 672 | |
dce48a84 TG |
673 | /* calc_load related fields */ |
674 | unsigned long calc_load_update; | |
675 | long calc_load_active; | |
676 | ||
8f4d37ec | 677 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
678 | #ifdef CONFIG_SMP |
679 | int hrtick_csd_pending; | |
680 | struct call_single_data hrtick_csd; | |
681 | #endif | |
8f4d37ec PZ |
682 | struct hrtimer hrtick_timer; |
683 | #endif | |
684 | ||
1da177e4 LT |
685 | #ifdef CONFIG_SCHEDSTATS |
686 | /* latency stats */ | |
687 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
688 | unsigned long long rq_cpu_time; |
689 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
690 | |
691 | /* sys_sched_yield() stats */ | |
480b9434 | 692 | unsigned int yld_count; |
1da177e4 LT |
693 | |
694 | /* schedule() stats */ | |
480b9434 KC |
695 | unsigned int sched_switch; |
696 | unsigned int sched_count; | |
697 | unsigned int sched_goidle; | |
1da177e4 LT |
698 | |
699 | /* try_to_wake_up() stats */ | |
480b9434 KC |
700 | unsigned int ttwu_count; |
701 | unsigned int ttwu_local; | |
1da177e4 | 702 | #endif |
317f3941 PZ |
703 | |
704 | #ifdef CONFIG_SMP | |
fa14ff4a | 705 | struct llist_head wake_list; |
317f3941 | 706 | #endif |
1da177e4 LT |
707 | }; |
708 | ||
f34e3b61 | 709 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 710 | |
a64692a3 | 711 | |
1e5a7405 | 712 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); |
dd41f596 | 713 | |
0a2966b4 CL |
714 | static inline int cpu_of(struct rq *rq) |
715 | { | |
716 | #ifdef CONFIG_SMP | |
717 | return rq->cpu; | |
718 | #else | |
719 | return 0; | |
720 | #endif | |
721 | } | |
722 | ||
497f0ab3 | 723 | #define rcu_dereference_check_sched_domain(p) \ |
d11c563d | 724 | rcu_dereference_check((p), \ |
d11c563d PM |
725 | lockdep_is_held(&sched_domains_mutex)) |
726 | ||
674311d5 NP |
727 | /* |
728 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 729 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
730 | * |
731 | * The domain tree of any CPU may only be accessed from within | |
732 | * preempt-disabled sections. | |
733 | */ | |
48f24c4d | 734 | #define for_each_domain(cpu, __sd) \ |
497f0ab3 | 735 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) |
1da177e4 LT |
736 | |
737 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
738 | #define this_rq() (&__get_cpu_var(runqueues)) | |
739 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
740 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 741 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 742 | |
dc61b1d6 PZ |
743 | #ifdef CONFIG_CGROUP_SCHED |
744 | ||
745 | /* | |
746 | * Return the group to which this tasks belongs. | |
747 | * | |
6c6c54e1 PZ |
748 | * We use task_subsys_state_check() and extend the RCU verification with |
749 | * pi->lock and rq->lock because cpu_cgroup_attach() holds those locks for each | |
750 | * task it moves into the cgroup. Therefore by holding either of those locks, | |
751 | * we pin the task to the current cgroup. | |
dc61b1d6 PZ |
752 | */ |
753 | static inline struct task_group *task_group(struct task_struct *p) | |
754 | { | |
5091faa4 | 755 | struct task_group *tg; |
dc61b1d6 PZ |
756 | struct cgroup_subsys_state *css; |
757 | ||
758 | css = task_subsys_state_check(p, cpu_cgroup_subsys_id, | |
6c6c54e1 PZ |
759 | lockdep_is_held(&p->pi_lock) || |
760 | lockdep_is_held(&task_rq(p)->lock)); | |
5091faa4 MG |
761 | tg = container_of(css, struct task_group, css); |
762 | ||
763 | return autogroup_task_group(p, tg); | |
dc61b1d6 PZ |
764 | } |
765 | ||
766 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
767 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
768 | { | |
769 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
770 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; | |
771 | p->se.parent = task_group(p)->se[cpu]; | |
772 | #endif | |
773 | ||
774 | #ifdef CONFIG_RT_GROUP_SCHED | |
775 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; | |
776 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
777 | #endif | |
778 | } | |
779 | ||
780 | #else /* CONFIG_CGROUP_SCHED */ | |
781 | ||
782 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
783 | static inline struct task_group *task_group(struct task_struct *p) | |
784 | { | |
785 | return NULL; | |
786 | } | |
787 | ||
788 | #endif /* CONFIG_CGROUP_SCHED */ | |
789 | ||
fe44d621 | 790 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 791 | |
fe44d621 | 792 | static void update_rq_clock(struct rq *rq) |
3e51f33f | 793 | { |
fe44d621 | 794 | s64 delta; |
305e6835 | 795 | |
61eadef6 | 796 | if (rq->skip_clock_update > 0) |
f26f9aff | 797 | return; |
aa483808 | 798 | |
fe44d621 PZ |
799 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
800 | rq->clock += delta; | |
801 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
802 | } |
803 | ||
bf5c91ba IM |
804 | /* |
805 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
806 | */ | |
807 | #ifdef CONFIG_SCHED_DEBUG | |
808 | # define const_debug __read_mostly | |
809 | #else | |
810 | # define const_debug static const | |
811 | #endif | |
812 | ||
017730c1 | 813 | /** |
1fd06bb1 | 814 | * runqueue_is_locked - Returns true if the current cpu runqueue is locked |
e17b38bf | 815 | * @cpu: the processor in question. |
017730c1 | 816 | * |
017730c1 IM |
817 | * This interface allows printk to be called with the runqueue lock |
818 | * held and know whether or not it is OK to wake up the klogd. | |
819 | */ | |
89f19f04 | 820 | int runqueue_is_locked(int cpu) |
017730c1 | 821 | { |
05fa785c | 822 | return raw_spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
823 | } |
824 | ||
bf5c91ba IM |
825 | /* |
826 | * Debugging: various feature bits | |
827 | */ | |
f00b45c1 PZ |
828 | |
829 | #define SCHED_FEAT(name, enabled) \ | |
830 | __SCHED_FEAT_##name , | |
831 | ||
bf5c91ba | 832 | enum { |
f00b45c1 | 833 | #include "sched_features.h" |
bf5c91ba IM |
834 | }; |
835 | ||
f00b45c1 PZ |
836 | #undef SCHED_FEAT |
837 | ||
838 | #define SCHED_FEAT(name, enabled) \ | |
839 | (1UL << __SCHED_FEAT_##name) * enabled | | |
840 | ||
bf5c91ba | 841 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
842 | #include "sched_features.h" |
843 | 0; | |
844 | ||
845 | #undef SCHED_FEAT | |
846 | ||
847 | #ifdef CONFIG_SCHED_DEBUG | |
848 | #define SCHED_FEAT(name, enabled) \ | |
849 | #name , | |
850 | ||
983ed7a6 | 851 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
852 | #include "sched_features.h" |
853 | NULL | |
854 | }; | |
855 | ||
856 | #undef SCHED_FEAT | |
857 | ||
34f3a814 | 858 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 859 | { |
f00b45c1 PZ |
860 | int i; |
861 | ||
862 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
863 | if (!(sysctl_sched_features & (1UL << i))) |
864 | seq_puts(m, "NO_"); | |
865 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 866 | } |
34f3a814 | 867 | seq_puts(m, "\n"); |
f00b45c1 | 868 | |
34f3a814 | 869 | return 0; |
f00b45c1 PZ |
870 | } |
871 | ||
872 | static ssize_t | |
873 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
874 | size_t cnt, loff_t *ppos) | |
875 | { | |
876 | char buf[64]; | |
7740191c | 877 | char *cmp; |
f00b45c1 PZ |
878 | int neg = 0; |
879 | int i; | |
880 | ||
881 | if (cnt > 63) | |
882 | cnt = 63; | |
883 | ||
884 | if (copy_from_user(&buf, ubuf, cnt)) | |
885 | return -EFAULT; | |
886 | ||
887 | buf[cnt] = 0; | |
7740191c | 888 | cmp = strstrip(buf); |
f00b45c1 | 889 | |
524429c3 | 890 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
891 | neg = 1; |
892 | cmp += 3; | |
893 | } | |
894 | ||
895 | for (i = 0; sched_feat_names[i]; i++) { | |
7740191c | 896 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f00b45c1 PZ |
897 | if (neg) |
898 | sysctl_sched_features &= ~(1UL << i); | |
899 | else | |
900 | sysctl_sched_features |= (1UL << i); | |
901 | break; | |
902 | } | |
903 | } | |
904 | ||
905 | if (!sched_feat_names[i]) | |
906 | return -EINVAL; | |
907 | ||
42994724 | 908 | *ppos += cnt; |
f00b45c1 PZ |
909 | |
910 | return cnt; | |
911 | } | |
912 | ||
34f3a814 LZ |
913 | static int sched_feat_open(struct inode *inode, struct file *filp) |
914 | { | |
915 | return single_open(filp, sched_feat_show, NULL); | |
916 | } | |
917 | ||
828c0950 | 918 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
919 | .open = sched_feat_open, |
920 | .write = sched_feat_write, | |
921 | .read = seq_read, | |
922 | .llseek = seq_lseek, | |
923 | .release = single_release, | |
f00b45c1 PZ |
924 | }; |
925 | ||
926 | static __init int sched_init_debug(void) | |
927 | { | |
f00b45c1 PZ |
928 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
929 | &sched_feat_fops); | |
930 | ||
931 | return 0; | |
932 | } | |
933 | late_initcall(sched_init_debug); | |
934 | ||
935 | #endif | |
936 | ||
937 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 938 | |
b82d9fdd PZ |
939 | /* |
940 | * Number of tasks to iterate in a single balance run. | |
941 | * Limited because this is done with IRQs disabled. | |
942 | */ | |
943 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
944 | ||
e9e9250b PZ |
945 | /* |
946 | * period over which we average the RT time consumption, measured | |
947 | * in ms. | |
948 | * | |
949 | * default: 1s | |
950 | */ | |
951 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
952 | ||
fa85ae24 | 953 | /* |
9f0c1e56 | 954 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
955 | * default: 1s |
956 | */ | |
9f0c1e56 | 957 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 958 | |
6892b75e IM |
959 | static __read_mostly int scheduler_running; |
960 | ||
9f0c1e56 PZ |
961 | /* |
962 | * part of the period that we allow rt tasks to run in us. | |
963 | * default: 0.95s | |
964 | */ | |
965 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 966 | |
d0b27fa7 PZ |
967 | static inline u64 global_rt_period(void) |
968 | { | |
969 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
970 | } | |
971 | ||
972 | static inline u64 global_rt_runtime(void) | |
973 | { | |
e26873bb | 974 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
975 | return RUNTIME_INF; |
976 | ||
977 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
978 | } | |
fa85ae24 | 979 | |
1da177e4 | 980 | #ifndef prepare_arch_switch |
4866cde0 NP |
981 | # define prepare_arch_switch(next) do { } while (0) |
982 | #endif | |
983 | #ifndef finish_arch_switch | |
984 | # define finish_arch_switch(prev) do { } while (0) | |
985 | #endif | |
986 | ||
051a1d1a DA |
987 | static inline int task_current(struct rq *rq, struct task_struct *p) |
988 | { | |
989 | return rq->curr == p; | |
990 | } | |
991 | ||
70b97a7f | 992 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 993 | { |
3ca7a440 PZ |
994 | #ifdef CONFIG_SMP |
995 | return p->on_cpu; | |
996 | #else | |
051a1d1a | 997 | return task_current(rq, p); |
3ca7a440 | 998 | #endif |
4866cde0 NP |
999 | } |
1000 | ||
3ca7a440 | 1001 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 1002 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 | 1003 | { |
3ca7a440 PZ |
1004 | #ifdef CONFIG_SMP |
1005 | /* | |
1006 | * We can optimise this out completely for !SMP, because the | |
1007 | * SMP rebalancing from interrupt is the only thing that cares | |
1008 | * here. | |
1009 | */ | |
1010 | next->on_cpu = 1; | |
1011 | #endif | |
4866cde0 NP |
1012 | } |
1013 | ||
70b97a7f | 1014 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 1015 | { |
3ca7a440 PZ |
1016 | #ifdef CONFIG_SMP |
1017 | /* | |
1018 | * After ->on_cpu is cleared, the task can be moved to a different CPU. | |
1019 | * We must ensure this doesn't happen until the switch is completely | |
1020 | * finished. | |
1021 | */ | |
1022 | smp_wmb(); | |
1023 | prev->on_cpu = 0; | |
1024 | #endif | |
da04c035 IM |
1025 | #ifdef CONFIG_DEBUG_SPINLOCK |
1026 | /* this is a valid case when another task releases the spinlock */ | |
1027 | rq->lock.owner = current; | |
1028 | #endif | |
8a25d5de IM |
1029 | /* |
1030 | * If we are tracking spinlock dependencies then we have to | |
1031 | * fix up the runqueue lock - which gets 'carried over' from | |
1032 | * prev into current: | |
1033 | */ | |
1034 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
1035 | ||
05fa785c | 1036 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 NP |
1037 | } |
1038 | ||
1039 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 1040 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
1041 | { |
1042 | #ifdef CONFIG_SMP | |
1043 | /* | |
1044 | * We can optimise this out completely for !SMP, because the | |
1045 | * SMP rebalancing from interrupt is the only thing that cares | |
1046 | * here. | |
1047 | */ | |
3ca7a440 | 1048 | next->on_cpu = 1; |
4866cde0 NP |
1049 | #endif |
1050 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
05fa785c | 1051 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 | 1052 | #else |
05fa785c | 1053 | raw_spin_unlock(&rq->lock); |
4866cde0 NP |
1054 | #endif |
1055 | } | |
1056 | ||
70b97a7f | 1057 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
1058 | { |
1059 | #ifdef CONFIG_SMP | |
1060 | /* | |
3ca7a440 | 1061 | * After ->on_cpu is cleared, the task can be moved to a different CPU. |
4866cde0 NP |
1062 | * We must ensure this doesn't happen until the switch is completely |
1063 | * finished. | |
1064 | */ | |
1065 | smp_wmb(); | |
3ca7a440 | 1066 | prev->on_cpu = 0; |
4866cde0 NP |
1067 | #endif |
1068 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
1069 | local_irq_enable(); | |
1da177e4 | 1070 | #endif |
4866cde0 NP |
1071 | } |
1072 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 1073 | |
0970d299 | 1074 | /* |
0122ec5b | 1075 | * __task_rq_lock - lock the rq @p resides on. |
b29739f9 | 1076 | */ |
70b97a7f | 1077 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
1078 | __acquires(rq->lock) |
1079 | { | |
0970d299 PZ |
1080 | struct rq *rq; |
1081 | ||
0122ec5b PZ |
1082 | lockdep_assert_held(&p->pi_lock); |
1083 | ||
3a5c359a | 1084 | for (;;) { |
0970d299 | 1085 | rq = task_rq(p); |
05fa785c | 1086 | raw_spin_lock(&rq->lock); |
65cc8e48 | 1087 | if (likely(rq == task_rq(p))) |
3a5c359a | 1088 | return rq; |
05fa785c | 1089 | raw_spin_unlock(&rq->lock); |
b29739f9 | 1090 | } |
b29739f9 IM |
1091 | } |
1092 | ||
1da177e4 | 1093 | /* |
0122ec5b | 1094 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. |
1da177e4 | 1095 | */ |
70b97a7f | 1096 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
0122ec5b | 1097 | __acquires(p->pi_lock) |
1da177e4 LT |
1098 | __acquires(rq->lock) |
1099 | { | |
70b97a7f | 1100 | struct rq *rq; |
1da177e4 | 1101 | |
3a5c359a | 1102 | for (;;) { |
0122ec5b | 1103 | raw_spin_lock_irqsave(&p->pi_lock, *flags); |
3a5c359a | 1104 | rq = task_rq(p); |
05fa785c | 1105 | raw_spin_lock(&rq->lock); |
65cc8e48 | 1106 | if (likely(rq == task_rq(p))) |
3a5c359a | 1107 | return rq; |
0122ec5b PZ |
1108 | raw_spin_unlock(&rq->lock); |
1109 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 | 1110 | } |
1da177e4 LT |
1111 | } |
1112 | ||
a9957449 | 1113 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
1114 | __releases(rq->lock) |
1115 | { | |
05fa785c | 1116 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
1117 | } |
1118 | ||
0122ec5b PZ |
1119 | static inline void |
1120 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | |
1da177e4 | 1121 | __releases(rq->lock) |
0122ec5b | 1122 | __releases(p->pi_lock) |
1da177e4 | 1123 | { |
0122ec5b PZ |
1124 | raw_spin_unlock(&rq->lock); |
1125 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 LT |
1126 | } |
1127 | ||
1da177e4 | 1128 | /* |
cc2a73b5 | 1129 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 1130 | */ |
a9957449 | 1131 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
1132 | __acquires(rq->lock) |
1133 | { | |
70b97a7f | 1134 | struct rq *rq; |
1da177e4 LT |
1135 | |
1136 | local_irq_disable(); | |
1137 | rq = this_rq(); | |
05fa785c | 1138 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
1139 | |
1140 | return rq; | |
1141 | } | |
1142 | ||
8f4d37ec PZ |
1143 | #ifdef CONFIG_SCHED_HRTICK |
1144 | /* | |
1145 | * Use HR-timers to deliver accurate preemption points. | |
1146 | * | |
1147 | * Its all a bit involved since we cannot program an hrt while holding the | |
1148 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1149 | * reschedule event. | |
1150 | * | |
1151 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1152 | * rq->lock. | |
1153 | */ | |
8f4d37ec PZ |
1154 | |
1155 | /* | |
1156 | * Use hrtick when: | |
1157 | * - enabled by features | |
1158 | * - hrtimer is actually high res | |
1159 | */ | |
1160 | static inline int hrtick_enabled(struct rq *rq) | |
1161 | { | |
1162 | if (!sched_feat(HRTICK)) | |
1163 | return 0; | |
ba42059f | 1164 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1165 | return 0; |
8f4d37ec PZ |
1166 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1167 | } | |
1168 | ||
8f4d37ec PZ |
1169 | static void hrtick_clear(struct rq *rq) |
1170 | { | |
1171 | if (hrtimer_active(&rq->hrtick_timer)) | |
1172 | hrtimer_cancel(&rq->hrtick_timer); | |
1173 | } | |
1174 | ||
8f4d37ec PZ |
1175 | /* |
1176 | * High-resolution timer tick. | |
1177 | * Runs from hardirq context with interrupts disabled. | |
1178 | */ | |
1179 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1180 | { | |
1181 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1182 | ||
1183 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1184 | ||
05fa785c | 1185 | raw_spin_lock(&rq->lock); |
3e51f33f | 1186 | update_rq_clock(rq); |
8f4d37ec | 1187 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 1188 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
1189 | |
1190 | return HRTIMER_NORESTART; | |
1191 | } | |
1192 | ||
95e904c7 | 1193 | #ifdef CONFIG_SMP |
31656519 PZ |
1194 | /* |
1195 | * called from hardirq (IPI) context | |
1196 | */ | |
1197 | static void __hrtick_start(void *arg) | |
b328ca18 | 1198 | { |
31656519 | 1199 | struct rq *rq = arg; |
b328ca18 | 1200 | |
05fa785c | 1201 | raw_spin_lock(&rq->lock); |
31656519 PZ |
1202 | hrtimer_restart(&rq->hrtick_timer); |
1203 | rq->hrtick_csd_pending = 0; | |
05fa785c | 1204 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
1205 | } |
1206 | ||
31656519 PZ |
1207 | /* |
1208 | * Called to set the hrtick timer state. | |
1209 | * | |
1210 | * called with rq->lock held and irqs disabled | |
1211 | */ | |
1212 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1213 | { |
31656519 PZ |
1214 | struct hrtimer *timer = &rq->hrtick_timer; |
1215 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1216 | |
cc584b21 | 1217 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1218 | |
1219 | if (rq == this_rq()) { | |
1220 | hrtimer_restart(timer); | |
1221 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1222 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1223 | rq->hrtick_csd_pending = 1; |
1224 | } | |
b328ca18 PZ |
1225 | } |
1226 | ||
1227 | static int | |
1228 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1229 | { | |
1230 | int cpu = (int)(long)hcpu; | |
1231 | ||
1232 | switch (action) { | |
1233 | case CPU_UP_CANCELED: | |
1234 | case CPU_UP_CANCELED_FROZEN: | |
1235 | case CPU_DOWN_PREPARE: | |
1236 | case CPU_DOWN_PREPARE_FROZEN: | |
1237 | case CPU_DEAD: | |
1238 | case CPU_DEAD_FROZEN: | |
31656519 | 1239 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1240 | return NOTIFY_OK; |
1241 | } | |
1242 | ||
1243 | return NOTIFY_DONE; | |
1244 | } | |
1245 | ||
fa748203 | 1246 | static __init void init_hrtick(void) |
b328ca18 PZ |
1247 | { |
1248 | hotcpu_notifier(hotplug_hrtick, 0); | |
1249 | } | |
31656519 PZ |
1250 | #else |
1251 | /* | |
1252 | * Called to set the hrtick timer state. | |
1253 | * | |
1254 | * called with rq->lock held and irqs disabled | |
1255 | */ | |
1256 | static void hrtick_start(struct rq *rq, u64 delay) | |
1257 | { | |
7f1e2ca9 | 1258 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1259 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1260 | } |
b328ca18 | 1261 | |
006c75f1 | 1262 | static inline void init_hrtick(void) |
8f4d37ec | 1263 | { |
8f4d37ec | 1264 | } |
31656519 | 1265 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1266 | |
31656519 | 1267 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1268 | { |
31656519 PZ |
1269 | #ifdef CONFIG_SMP |
1270 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1271 | |
31656519 PZ |
1272 | rq->hrtick_csd.flags = 0; |
1273 | rq->hrtick_csd.func = __hrtick_start; | |
1274 | rq->hrtick_csd.info = rq; | |
1275 | #endif | |
8f4d37ec | 1276 | |
31656519 PZ |
1277 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1278 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1279 | } |
006c75f1 | 1280 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1281 | static inline void hrtick_clear(struct rq *rq) |
1282 | { | |
1283 | } | |
1284 | ||
8f4d37ec PZ |
1285 | static inline void init_rq_hrtick(struct rq *rq) |
1286 | { | |
1287 | } | |
1288 | ||
b328ca18 PZ |
1289 | static inline void init_hrtick(void) |
1290 | { | |
1291 | } | |
006c75f1 | 1292 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1293 | |
c24d20db IM |
1294 | /* |
1295 | * resched_task - mark a task 'to be rescheduled now'. | |
1296 | * | |
1297 | * On UP this means the setting of the need_resched flag, on SMP it | |
1298 | * might also involve a cross-CPU call to trigger the scheduler on | |
1299 | * the target CPU. | |
1300 | */ | |
1301 | #ifdef CONFIG_SMP | |
1302 | ||
1303 | #ifndef tsk_is_polling | |
1304 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1305 | #endif | |
1306 | ||
31656519 | 1307 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1308 | { |
1309 | int cpu; | |
1310 | ||
05fa785c | 1311 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 1312 | |
5ed0cec0 | 1313 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1314 | return; |
1315 | ||
5ed0cec0 | 1316 | set_tsk_need_resched(p); |
c24d20db IM |
1317 | |
1318 | cpu = task_cpu(p); | |
1319 | if (cpu == smp_processor_id()) | |
1320 | return; | |
1321 | ||
1322 | /* NEED_RESCHED must be visible before we test polling */ | |
1323 | smp_mb(); | |
1324 | if (!tsk_is_polling(p)) | |
1325 | smp_send_reschedule(cpu); | |
1326 | } | |
1327 | ||
1328 | static void resched_cpu(int cpu) | |
1329 | { | |
1330 | struct rq *rq = cpu_rq(cpu); | |
1331 | unsigned long flags; | |
1332 | ||
05fa785c | 1333 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
1334 | return; |
1335 | resched_task(cpu_curr(cpu)); | |
05fa785c | 1336 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 1337 | } |
06d8308c TG |
1338 | |
1339 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
1340 | /* |
1341 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
1342 | * from an idle cpu. This is good for power-savings. | |
1343 | * | |
1344 | * We don't do similar optimization for completely idle system, as | |
1345 | * selecting an idle cpu will add more delays to the timers than intended | |
1346 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
1347 | */ | |
1348 | int get_nohz_timer_target(void) | |
1349 | { | |
1350 | int cpu = smp_processor_id(); | |
1351 | int i; | |
1352 | struct sched_domain *sd; | |
1353 | ||
057f3fad | 1354 | rcu_read_lock(); |
83cd4fe2 | 1355 | for_each_domain(cpu, sd) { |
057f3fad PZ |
1356 | for_each_cpu(i, sched_domain_span(sd)) { |
1357 | if (!idle_cpu(i)) { | |
1358 | cpu = i; | |
1359 | goto unlock; | |
1360 | } | |
1361 | } | |
83cd4fe2 | 1362 | } |
057f3fad PZ |
1363 | unlock: |
1364 | rcu_read_unlock(); | |
83cd4fe2 VP |
1365 | return cpu; |
1366 | } | |
06d8308c TG |
1367 | /* |
1368 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1369 | * idle CPU then this timer might expire before the next timer event | |
1370 | * which is scheduled to wake up that CPU. In case of a completely | |
1371 | * idle system the next event might even be infinite time into the | |
1372 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1373 | * leaves the inner idle loop so the newly added timer is taken into | |
1374 | * account when the CPU goes back to idle and evaluates the timer | |
1375 | * wheel for the next timer event. | |
1376 | */ | |
1377 | void wake_up_idle_cpu(int cpu) | |
1378 | { | |
1379 | struct rq *rq = cpu_rq(cpu); | |
1380 | ||
1381 | if (cpu == smp_processor_id()) | |
1382 | return; | |
1383 | ||
1384 | /* | |
1385 | * This is safe, as this function is called with the timer | |
1386 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1387 | * to idle and has not yet set rq->curr to idle then it will | |
1388 | * be serialized on the timer wheel base lock and take the new | |
1389 | * timer into account automatically. | |
1390 | */ | |
1391 | if (rq->curr != rq->idle) | |
1392 | return; | |
1393 | ||
1394 | /* | |
1395 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1396 | * lockless. The worst case is that the other CPU runs the | |
1397 | * idle task through an additional NOOP schedule() | |
1398 | */ | |
5ed0cec0 | 1399 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1400 | |
1401 | /* NEED_RESCHED must be visible before we test polling */ | |
1402 | smp_mb(); | |
1403 | if (!tsk_is_polling(rq->idle)) | |
1404 | smp_send_reschedule(cpu); | |
1405 | } | |
39c0cbe2 | 1406 | |
6d6bc0ad | 1407 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1408 | |
e9e9250b PZ |
1409 | static u64 sched_avg_period(void) |
1410 | { | |
1411 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1412 | } | |
1413 | ||
1414 | static void sched_avg_update(struct rq *rq) | |
1415 | { | |
1416 | s64 period = sched_avg_period(); | |
1417 | ||
1418 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
1419 | /* |
1420 | * Inline assembly required to prevent the compiler | |
1421 | * optimising this loop into a divmod call. | |
1422 | * See __iter_div_u64_rem() for another example of this. | |
1423 | */ | |
1424 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
1425 | rq->age_stamp += period; |
1426 | rq->rt_avg /= 2; | |
1427 | } | |
1428 | } | |
1429 | ||
1430 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1431 | { | |
1432 | rq->rt_avg += rt_delta; | |
1433 | sched_avg_update(rq); | |
1434 | } | |
1435 | ||
6d6bc0ad | 1436 | #else /* !CONFIG_SMP */ |
31656519 | 1437 | static void resched_task(struct task_struct *p) |
c24d20db | 1438 | { |
05fa785c | 1439 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 1440 | set_tsk_need_resched(p); |
c24d20db | 1441 | } |
e9e9250b PZ |
1442 | |
1443 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1444 | { | |
1445 | } | |
da2b71ed SS |
1446 | |
1447 | static void sched_avg_update(struct rq *rq) | |
1448 | { | |
1449 | } | |
6d6bc0ad | 1450 | #endif /* CONFIG_SMP */ |
c24d20db | 1451 | |
45bf76df IM |
1452 | #if BITS_PER_LONG == 32 |
1453 | # define WMULT_CONST (~0UL) | |
1454 | #else | |
1455 | # define WMULT_CONST (1UL << 32) | |
1456 | #endif | |
1457 | ||
1458 | #define WMULT_SHIFT 32 | |
1459 | ||
194081eb IM |
1460 | /* |
1461 | * Shift right and round: | |
1462 | */ | |
cf2ab469 | 1463 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1464 | |
a7be37ac PZ |
1465 | /* |
1466 | * delta *= weight / lw | |
1467 | */ | |
cb1c4fc9 | 1468 | static unsigned long |
45bf76df IM |
1469 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1470 | struct load_weight *lw) | |
1471 | { | |
1472 | u64 tmp; | |
1473 | ||
c8b28116 NR |
1474 | /* |
1475 | * weight can be less than 2^SCHED_LOAD_RESOLUTION for task group sched | |
1476 | * entities since MIN_SHARES = 2. Treat weight as 1 if less than | |
1477 | * 2^SCHED_LOAD_RESOLUTION. | |
1478 | */ | |
1479 | if (likely(weight > (1UL << SCHED_LOAD_RESOLUTION))) | |
1480 | tmp = (u64)delta_exec * scale_load_down(weight); | |
1481 | else | |
1482 | tmp = (u64)delta_exec; | |
db670dac | 1483 | |
7a232e03 | 1484 | if (!lw->inv_weight) { |
c8b28116 NR |
1485 | unsigned long w = scale_load_down(lw->weight); |
1486 | ||
1487 | if (BITS_PER_LONG > 32 && unlikely(w >= WMULT_CONST)) | |
7a232e03 | 1488 | lw->inv_weight = 1; |
c8b28116 NR |
1489 | else if (unlikely(!w)) |
1490 | lw->inv_weight = WMULT_CONST; | |
7a232e03 | 1491 | else |
c8b28116 | 1492 | lw->inv_weight = WMULT_CONST / w; |
7a232e03 | 1493 | } |
45bf76df | 1494 | |
45bf76df IM |
1495 | /* |
1496 | * Check whether we'd overflow the 64-bit multiplication: | |
1497 | */ | |
194081eb | 1498 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1499 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1500 | WMULT_SHIFT/2); |
1501 | else | |
cf2ab469 | 1502 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1503 | |
ecf691da | 1504 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1505 | } |
1506 | ||
1091985b | 1507 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1508 | { |
1509 | lw->weight += inc; | |
e89996ae | 1510 | lw->inv_weight = 0; |
45bf76df IM |
1511 | } |
1512 | ||
1091985b | 1513 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1514 | { |
1515 | lw->weight -= dec; | |
e89996ae | 1516 | lw->inv_weight = 0; |
45bf76df IM |
1517 | } |
1518 | ||
2069dd75 PZ |
1519 | static inline void update_load_set(struct load_weight *lw, unsigned long w) |
1520 | { | |
1521 | lw->weight = w; | |
1522 | lw->inv_weight = 0; | |
1523 | } | |
1524 | ||
2dd73a4f PW |
1525 | /* |
1526 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1527 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1528 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1529 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1530 | * scaled version of the new time slice allocation that they receive on time |
1531 | * slice expiry etc. | |
1532 | */ | |
1533 | ||
cce7ade8 PZ |
1534 | #define WEIGHT_IDLEPRIO 3 |
1535 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1536 | |
1537 | /* | |
1538 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1539 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1540 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1541 | * that remained on nice 0. | |
1542 | * | |
1543 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1544 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1545 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1546 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1547 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1548 | */ |
1549 | static const int prio_to_weight[40] = { | |
254753dc IM |
1550 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1551 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1552 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1553 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1554 | /* 0 */ 1024, 820, 655, 526, 423, | |
1555 | /* 5 */ 335, 272, 215, 172, 137, | |
1556 | /* 10 */ 110, 87, 70, 56, 45, | |
1557 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1558 | }; |
1559 | ||
5714d2de IM |
1560 | /* |
1561 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1562 | * | |
1563 | * In cases where the weight does not change often, we can use the | |
1564 | * precalculated inverse to speed up arithmetics by turning divisions | |
1565 | * into multiplications: | |
1566 | */ | |
dd41f596 | 1567 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1568 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1569 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1570 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1571 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1572 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1573 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1574 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1575 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1576 | }; |
2dd73a4f | 1577 | |
ef12fefa BR |
1578 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1579 | enum cpuacct_stat_index { | |
1580 | CPUACCT_STAT_USER, /* ... user mode */ | |
1581 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1582 | ||
1583 | CPUACCT_STAT_NSTATS, | |
1584 | }; | |
1585 | ||
d842de87 SV |
1586 | #ifdef CONFIG_CGROUP_CPUACCT |
1587 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1588 | static void cpuacct_update_stats(struct task_struct *tsk, |
1589 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1590 | #else |
1591 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1592 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1593 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1594 | #endif |
1595 | ||
18d95a28 PZ |
1596 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1597 | { | |
1598 | update_load_add(&rq->load, load); | |
1599 | } | |
1600 | ||
1601 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1602 | { | |
1603 | update_load_sub(&rq->load, load); | |
1604 | } | |
1605 | ||
a790de99 PT |
1606 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
1607 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
eb755805 | 1608 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1609 | |
1610 | /* | |
8277434e PT |
1611 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
1612 | * node and @up when leaving it for the final time. | |
1613 | * | |
1614 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 1615 | */ |
8277434e PT |
1616 | static int walk_tg_tree_from(struct task_group *from, |
1617 | tg_visitor down, tg_visitor up, void *data) | |
c09595f6 PZ |
1618 | { |
1619 | struct task_group *parent, *child; | |
eb755805 | 1620 | int ret; |
c09595f6 | 1621 | |
8277434e PT |
1622 | parent = from; |
1623 | ||
c09595f6 | 1624 | down: |
eb755805 PZ |
1625 | ret = (*down)(parent, data); |
1626 | if (ret) | |
8277434e | 1627 | goto out; |
c09595f6 PZ |
1628 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1629 | parent = child; | |
1630 | goto down; | |
1631 | ||
1632 | up: | |
1633 | continue; | |
1634 | } | |
eb755805 | 1635 | ret = (*up)(parent, data); |
8277434e PT |
1636 | if (ret || parent == from) |
1637 | goto out; | |
c09595f6 PZ |
1638 | |
1639 | child = parent; | |
1640 | parent = parent->parent; | |
1641 | if (parent) | |
1642 | goto up; | |
8277434e | 1643 | out: |
eb755805 | 1644 | return ret; |
c09595f6 PZ |
1645 | } |
1646 | ||
8277434e PT |
1647 | /* |
1648 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1649 | * leaving it for the final time. | |
1650 | * | |
1651 | * Caller must hold rcu_lock or sufficient equivalent. | |
1652 | */ | |
1653 | ||
1654 | static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) | |
1655 | { | |
1656 | return walk_tg_tree_from(&root_task_group, down, up, data); | |
1657 | } | |
1658 | ||
eb755805 PZ |
1659 | static int tg_nop(struct task_group *tg, void *data) |
1660 | { | |
1661 | return 0; | |
c09595f6 | 1662 | } |
eb755805 PZ |
1663 | #endif |
1664 | ||
1665 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1666 | /* Used instead of source_load when we know the type == 0 */ |
1667 | static unsigned long weighted_cpuload(const int cpu) | |
1668 | { | |
1669 | return cpu_rq(cpu)->load.weight; | |
1670 | } | |
1671 | ||
1672 | /* | |
1673 | * Return a low guess at the load of a migration-source cpu weighted | |
1674 | * according to the scheduling class and "nice" value. | |
1675 | * | |
1676 | * We want to under-estimate the load of migration sources, to | |
1677 | * balance conservatively. | |
1678 | */ | |
1679 | static unsigned long source_load(int cpu, int type) | |
1680 | { | |
1681 | struct rq *rq = cpu_rq(cpu); | |
1682 | unsigned long total = weighted_cpuload(cpu); | |
1683 | ||
1684 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1685 | return total; | |
1686 | ||
1687 | return min(rq->cpu_load[type-1], total); | |
1688 | } | |
1689 | ||
1690 | /* | |
1691 | * Return a high guess at the load of a migration-target cpu weighted | |
1692 | * according to the scheduling class and "nice" value. | |
1693 | */ | |
1694 | static unsigned long target_load(int cpu, int type) | |
1695 | { | |
1696 | struct rq *rq = cpu_rq(cpu); | |
1697 | unsigned long total = weighted_cpuload(cpu); | |
1698 | ||
1699 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1700 | return total; | |
1701 | ||
1702 | return max(rq->cpu_load[type-1], total); | |
1703 | } | |
1704 | ||
ae154be1 PZ |
1705 | static unsigned long power_of(int cpu) |
1706 | { | |
e51fd5e2 | 1707 | return cpu_rq(cpu)->cpu_power; |
ae154be1 PZ |
1708 | } |
1709 | ||
eb755805 PZ |
1710 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1711 | ||
1712 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1713 | { | |
1714 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1715 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1716 | |
4cd42620 | 1717 | if (nr_running) |
e2b245f8 | 1718 | return rq->load.weight / nr_running; |
eb755805 | 1719 | |
e2b245f8 | 1720 | return 0; |
eb755805 PZ |
1721 | } |
1722 | ||
8f45e2b5 GH |
1723 | #ifdef CONFIG_PREEMPT |
1724 | ||
b78bb868 PZ |
1725 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1726 | ||
70574a99 | 1727 | /* |
8f45e2b5 GH |
1728 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1729 | * way at the expense of forcing extra atomic operations in all | |
1730 | * invocations. This assures that the double_lock is acquired using the | |
1731 | * same underlying policy as the spinlock_t on this architecture, which | |
1732 | * reduces latency compared to the unfair variant below. However, it | |
1733 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1734 | */ |
8f45e2b5 GH |
1735 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1736 | __releases(this_rq->lock) | |
1737 | __acquires(busiest->lock) | |
1738 | __acquires(this_rq->lock) | |
1739 | { | |
05fa785c | 1740 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1741 | double_rq_lock(this_rq, busiest); |
1742 | ||
1743 | return 1; | |
1744 | } | |
1745 | ||
1746 | #else | |
1747 | /* | |
1748 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1749 | * latency by eliminating extra atomic operations when the locks are | |
1750 | * already in proper order on entry. This favors lower cpu-ids and will | |
1751 | * grant the double lock to lower cpus over higher ids under contention, | |
1752 | * regardless of entry order into the function. | |
1753 | */ | |
1754 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1755 | __releases(this_rq->lock) |
1756 | __acquires(busiest->lock) | |
1757 | __acquires(this_rq->lock) | |
1758 | { | |
1759 | int ret = 0; | |
1760 | ||
05fa785c | 1761 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1762 | if (busiest < this_rq) { |
05fa785c TG |
1763 | raw_spin_unlock(&this_rq->lock); |
1764 | raw_spin_lock(&busiest->lock); | |
1765 | raw_spin_lock_nested(&this_rq->lock, | |
1766 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1767 | ret = 1; |
1768 | } else | |
05fa785c TG |
1769 | raw_spin_lock_nested(&busiest->lock, |
1770 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1771 | } |
1772 | return ret; | |
1773 | } | |
1774 | ||
8f45e2b5 GH |
1775 | #endif /* CONFIG_PREEMPT */ |
1776 | ||
1777 | /* | |
1778 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1779 | */ | |
1780 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1781 | { | |
1782 | if (unlikely(!irqs_disabled())) { | |
1783 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1784 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1785 | BUG_ON(1); |
1786 | } | |
1787 | ||
1788 | return _double_lock_balance(this_rq, busiest); | |
1789 | } | |
1790 | ||
70574a99 AD |
1791 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1792 | __releases(busiest->lock) | |
1793 | { | |
05fa785c | 1794 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1795 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1796 | } | |
1e3c88bd PZ |
1797 | |
1798 | /* | |
1799 | * double_rq_lock - safely lock two runqueues | |
1800 | * | |
1801 | * Note this does not disable interrupts like task_rq_lock, | |
1802 | * you need to do so manually before calling. | |
1803 | */ | |
1804 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1805 | __acquires(rq1->lock) | |
1806 | __acquires(rq2->lock) | |
1807 | { | |
1808 | BUG_ON(!irqs_disabled()); | |
1809 | if (rq1 == rq2) { | |
1810 | raw_spin_lock(&rq1->lock); | |
1811 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1812 | } else { | |
1813 | if (rq1 < rq2) { | |
1814 | raw_spin_lock(&rq1->lock); | |
1815 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1816 | } else { | |
1817 | raw_spin_lock(&rq2->lock); | |
1818 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1819 | } | |
1820 | } | |
1e3c88bd PZ |
1821 | } |
1822 | ||
1823 | /* | |
1824 | * double_rq_unlock - safely unlock two runqueues | |
1825 | * | |
1826 | * Note this does not restore interrupts like task_rq_unlock, | |
1827 | * you need to do so manually after calling. | |
1828 | */ | |
1829 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1830 | __releases(rq1->lock) | |
1831 | __releases(rq2->lock) | |
1832 | { | |
1833 | raw_spin_unlock(&rq1->lock); | |
1834 | if (rq1 != rq2) | |
1835 | raw_spin_unlock(&rq2->lock); | |
1836 | else | |
1837 | __release(rq2->lock); | |
1838 | } | |
1839 | ||
d95f4122 MG |
1840 | #else /* CONFIG_SMP */ |
1841 | ||
1842 | /* | |
1843 | * double_rq_lock - safely lock two runqueues | |
1844 | * | |
1845 | * Note this does not disable interrupts like task_rq_lock, | |
1846 | * you need to do so manually before calling. | |
1847 | */ | |
1848 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1849 | __acquires(rq1->lock) | |
1850 | __acquires(rq2->lock) | |
1851 | { | |
1852 | BUG_ON(!irqs_disabled()); | |
1853 | BUG_ON(rq1 != rq2); | |
1854 | raw_spin_lock(&rq1->lock); | |
1855 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1856 | } | |
1857 | ||
1858 | /* | |
1859 | * double_rq_unlock - safely unlock two runqueues | |
1860 | * | |
1861 | * Note this does not restore interrupts like task_rq_unlock, | |
1862 | * you need to do so manually after calling. | |
1863 | */ | |
1864 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1865 | __releases(rq1->lock) | |
1866 | __releases(rq2->lock) | |
1867 | { | |
1868 | BUG_ON(rq1 != rq2); | |
1869 | raw_spin_unlock(&rq1->lock); | |
1870 | __release(rq2->lock); | |
1871 | } | |
1872 | ||
18d95a28 PZ |
1873 | #endif |
1874 | ||
74f5187a | 1875 | static void calc_load_account_idle(struct rq *this_rq); |
0bcdcf28 | 1876 | static void update_sysctl(void); |
acb4a848 | 1877 | static int get_update_sysctl_factor(void); |
fdf3e95d | 1878 | static void update_cpu_load(struct rq *this_rq); |
dce48a84 | 1879 | |
cd29fe6f PZ |
1880 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1881 | { | |
1882 | set_task_rq(p, cpu); | |
1883 | #ifdef CONFIG_SMP | |
1884 | /* | |
1885 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1886 | * successfuly executed on another CPU. We must ensure that updates of | |
1887 | * per-task data have been completed by this moment. | |
1888 | */ | |
1889 | smp_wmb(); | |
1890 | task_thread_info(p)->cpu = cpu; | |
1891 | #endif | |
1892 | } | |
dce48a84 | 1893 | |
1e3c88bd | 1894 | static const struct sched_class rt_sched_class; |
dd41f596 | 1895 | |
34f971f6 | 1896 | #define sched_class_highest (&stop_sched_class) |
1f11eb6a GH |
1897 | #define for_each_class(class) \ |
1898 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1899 | |
1e3c88bd PZ |
1900 | #include "sched_stats.h" |
1901 | ||
c09595f6 | 1902 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1903 | { |
1904 | rq->nr_running++; | |
9c217245 IM |
1905 | } |
1906 | ||
c09595f6 | 1907 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1908 | { |
1909 | rq->nr_running--; | |
9c217245 IM |
1910 | } |
1911 | ||
45bf76df IM |
1912 | static void set_load_weight(struct task_struct *p) |
1913 | { | |
f05998d4 NR |
1914 | int prio = p->static_prio - MAX_RT_PRIO; |
1915 | struct load_weight *load = &p->se.load; | |
1916 | ||
dd41f596 IM |
1917 | /* |
1918 | * SCHED_IDLE tasks get minimal weight: | |
1919 | */ | |
1920 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 1921 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 1922 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
1923 | return; |
1924 | } | |
71f8bd46 | 1925 | |
c8b28116 | 1926 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 1927 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
1928 | } |
1929 | ||
371fd7e7 | 1930 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1931 | { |
a64692a3 | 1932 | update_rq_clock(rq); |
dd41f596 | 1933 | sched_info_queued(p); |
371fd7e7 | 1934 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
1935 | } |
1936 | ||
371fd7e7 | 1937 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1938 | { |
a64692a3 | 1939 | update_rq_clock(rq); |
46ac22ba | 1940 | sched_info_dequeued(p); |
371fd7e7 | 1941 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
1942 | } |
1943 | ||
1e3c88bd PZ |
1944 | /* |
1945 | * activate_task - move a task to the runqueue. | |
1946 | */ | |
371fd7e7 | 1947 | static void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1948 | { |
1949 | if (task_contributes_to_load(p)) | |
1950 | rq->nr_uninterruptible--; | |
1951 | ||
371fd7e7 | 1952 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
1953 | } |
1954 | ||
1955 | /* | |
1956 | * deactivate_task - remove a task from the runqueue. | |
1957 | */ | |
371fd7e7 | 1958 | static void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1959 | { |
1960 | if (task_contributes_to_load(p)) | |
1961 | rq->nr_uninterruptible++; | |
1962 | ||
371fd7e7 | 1963 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1964 | } |
1965 | ||
b52bfee4 VP |
1966 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
1967 | ||
305e6835 VP |
1968 | /* |
1969 | * There are no locks covering percpu hardirq/softirq time. | |
1970 | * They are only modified in account_system_vtime, on corresponding CPU | |
1971 | * with interrupts disabled. So, writes are safe. | |
1972 | * They are read and saved off onto struct rq in update_rq_clock(). | |
1973 | * This may result in other CPU reading this CPU's irq time and can | |
1974 | * race with irq/account_system_vtime on this CPU. We would either get old | |
8e92c201 PZ |
1975 | * or new value with a side effect of accounting a slice of irq time to wrong |
1976 | * task when irq is in progress while we read rq->clock. That is a worthy | |
1977 | * compromise in place of having locks on each irq in account_system_time. | |
305e6835 | 1978 | */ |
b52bfee4 VP |
1979 | static DEFINE_PER_CPU(u64, cpu_hardirq_time); |
1980 | static DEFINE_PER_CPU(u64, cpu_softirq_time); | |
1981 | ||
1982 | static DEFINE_PER_CPU(u64, irq_start_time); | |
1983 | static int sched_clock_irqtime; | |
1984 | ||
1985 | void enable_sched_clock_irqtime(void) | |
1986 | { | |
1987 | sched_clock_irqtime = 1; | |
1988 | } | |
1989 | ||
1990 | void disable_sched_clock_irqtime(void) | |
1991 | { | |
1992 | sched_clock_irqtime = 0; | |
1993 | } | |
1994 | ||
8e92c201 PZ |
1995 | #ifndef CONFIG_64BIT |
1996 | static DEFINE_PER_CPU(seqcount_t, irq_time_seq); | |
1997 | ||
1998 | static inline void irq_time_write_begin(void) | |
1999 | { | |
2000 | __this_cpu_inc(irq_time_seq.sequence); | |
2001 | smp_wmb(); | |
2002 | } | |
2003 | ||
2004 | static inline void irq_time_write_end(void) | |
2005 | { | |
2006 | smp_wmb(); | |
2007 | __this_cpu_inc(irq_time_seq.sequence); | |
2008 | } | |
2009 | ||
2010 | static inline u64 irq_time_read(int cpu) | |
2011 | { | |
2012 | u64 irq_time; | |
2013 | unsigned seq; | |
2014 | ||
2015 | do { | |
2016 | seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); | |
2017 | irq_time = per_cpu(cpu_softirq_time, cpu) + | |
2018 | per_cpu(cpu_hardirq_time, cpu); | |
2019 | } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); | |
2020 | ||
2021 | return irq_time; | |
2022 | } | |
2023 | #else /* CONFIG_64BIT */ | |
2024 | static inline void irq_time_write_begin(void) | |
2025 | { | |
2026 | } | |
2027 | ||
2028 | static inline void irq_time_write_end(void) | |
2029 | { | |
2030 | } | |
2031 | ||
2032 | static inline u64 irq_time_read(int cpu) | |
305e6835 | 2033 | { |
305e6835 VP |
2034 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); |
2035 | } | |
8e92c201 | 2036 | #endif /* CONFIG_64BIT */ |
305e6835 | 2037 | |
fe44d621 PZ |
2038 | /* |
2039 | * Called before incrementing preempt_count on {soft,}irq_enter | |
2040 | * and before decrementing preempt_count on {soft,}irq_exit. | |
2041 | */ | |
b52bfee4 VP |
2042 | void account_system_vtime(struct task_struct *curr) |
2043 | { | |
2044 | unsigned long flags; | |
fe44d621 | 2045 | s64 delta; |
b52bfee4 | 2046 | int cpu; |
b52bfee4 VP |
2047 | |
2048 | if (!sched_clock_irqtime) | |
2049 | return; | |
2050 | ||
2051 | local_irq_save(flags); | |
2052 | ||
b52bfee4 | 2053 | cpu = smp_processor_id(); |
fe44d621 PZ |
2054 | delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time); |
2055 | __this_cpu_add(irq_start_time, delta); | |
2056 | ||
8e92c201 | 2057 | irq_time_write_begin(); |
b52bfee4 VP |
2058 | /* |
2059 | * We do not account for softirq time from ksoftirqd here. | |
2060 | * We want to continue accounting softirq time to ksoftirqd thread | |
2061 | * in that case, so as not to confuse scheduler with a special task | |
2062 | * that do not consume any time, but still wants to run. | |
2063 | */ | |
2064 | if (hardirq_count()) | |
fe44d621 | 2065 | __this_cpu_add(cpu_hardirq_time, delta); |
4dd53d89 | 2066 | else if (in_serving_softirq() && curr != this_cpu_ksoftirqd()) |
fe44d621 | 2067 | __this_cpu_add(cpu_softirq_time, delta); |
b52bfee4 | 2068 | |
8e92c201 | 2069 | irq_time_write_end(); |
b52bfee4 VP |
2070 | local_irq_restore(flags); |
2071 | } | |
b7dadc38 | 2072 | EXPORT_SYMBOL_GPL(account_system_vtime); |
b52bfee4 | 2073 | |
e6e6685a GC |
2074 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
2075 | ||
2076 | #ifdef CONFIG_PARAVIRT | |
2077 | static inline u64 steal_ticks(u64 steal) | |
aa483808 | 2078 | { |
e6e6685a GC |
2079 | if (unlikely(steal > NSEC_PER_SEC)) |
2080 | return div_u64(steal, TICK_NSEC); | |
fe44d621 | 2081 | |
e6e6685a GC |
2082 | return __iter_div_u64_rem(steal, TICK_NSEC, &steal); |
2083 | } | |
2084 | #endif | |
2085 | ||
fe44d621 | 2086 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 2087 | { |
095c0aa8 GC |
2088 | /* |
2089 | * In theory, the compile should just see 0 here, and optimize out the call | |
2090 | * to sched_rt_avg_update. But I don't trust it... | |
2091 | */ | |
2092 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
2093 | s64 steal = 0, irq_delta = 0; | |
2094 | #endif | |
2095 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 2096 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
2097 | |
2098 | /* | |
2099 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
2100 | * this case when a previous update_rq_clock() happened inside a | |
2101 | * {soft,}irq region. | |
2102 | * | |
2103 | * When this happens, we stop ->clock_task and only update the | |
2104 | * prev_irq_time stamp to account for the part that fit, so that a next | |
2105 | * update will consume the rest. This ensures ->clock_task is | |
2106 | * monotonic. | |
2107 | * | |
2108 | * It does however cause some slight miss-attribution of {soft,}irq | |
2109 | * time, a more accurate solution would be to update the irq_time using | |
2110 | * the current rq->clock timestamp, except that would require using | |
2111 | * atomic ops. | |
2112 | */ | |
2113 | if (irq_delta > delta) | |
2114 | irq_delta = delta; | |
2115 | ||
2116 | rq->prev_irq_time += irq_delta; | |
2117 | delta -= irq_delta; | |
095c0aa8 GC |
2118 | #endif |
2119 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
2120 | if (static_branch((¶virt_steal_rq_enabled))) { | |
2121 | u64 st; | |
2122 | ||
2123 | steal = paravirt_steal_clock(cpu_of(rq)); | |
2124 | steal -= rq->prev_steal_time_rq; | |
2125 | ||
2126 | if (unlikely(steal > delta)) | |
2127 | steal = delta; | |
2128 | ||
2129 | st = steal_ticks(steal); | |
2130 | steal = st * TICK_NSEC; | |
2131 | ||
2132 | rq->prev_steal_time_rq += steal; | |
2133 | ||
2134 | delta -= steal; | |
2135 | } | |
2136 | #endif | |
2137 | ||
fe44d621 PZ |
2138 | rq->clock_task += delta; |
2139 | ||
095c0aa8 GC |
2140 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
2141 | if ((irq_delta + steal) && sched_feat(NONTASK_POWER)) | |
2142 | sched_rt_avg_update(rq, irq_delta + steal); | |
2143 | #endif | |
aa483808 VP |
2144 | } |
2145 | ||
095c0aa8 | 2146 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
abb74cef VP |
2147 | static int irqtime_account_hi_update(void) |
2148 | { | |
2149 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
2150 | unsigned long flags; | |
2151 | u64 latest_ns; | |
2152 | int ret = 0; | |
2153 | ||
2154 | local_irq_save(flags); | |
2155 | latest_ns = this_cpu_read(cpu_hardirq_time); | |
2156 | if (cputime64_gt(nsecs_to_cputime64(latest_ns), cpustat->irq)) | |
2157 | ret = 1; | |
2158 | local_irq_restore(flags); | |
2159 | return ret; | |
2160 | } | |
2161 | ||
2162 | static int irqtime_account_si_update(void) | |
2163 | { | |
2164 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
2165 | unsigned long flags; | |
2166 | u64 latest_ns; | |
2167 | int ret = 0; | |
2168 | ||
2169 | local_irq_save(flags); | |
2170 | latest_ns = this_cpu_read(cpu_softirq_time); | |
2171 | if (cputime64_gt(nsecs_to_cputime64(latest_ns), cpustat->softirq)) | |
2172 | ret = 1; | |
2173 | local_irq_restore(flags); | |
2174 | return ret; | |
2175 | } | |
2176 | ||
fe44d621 | 2177 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
305e6835 | 2178 | |
abb74cef VP |
2179 | #define sched_clock_irqtime (0) |
2180 | ||
095c0aa8 | 2181 | #endif |
b52bfee4 | 2182 | |
1e3c88bd PZ |
2183 | #include "sched_idletask.c" |
2184 | #include "sched_fair.c" | |
2185 | #include "sched_rt.c" | |
5091faa4 | 2186 | #include "sched_autogroup.c" |
34f971f6 | 2187 | #include "sched_stoptask.c" |
1e3c88bd PZ |
2188 | #ifdef CONFIG_SCHED_DEBUG |
2189 | # include "sched_debug.c" | |
2190 | #endif | |
2191 | ||
34f971f6 PZ |
2192 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
2193 | { | |
2194 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
2195 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
2196 | ||
2197 | if (stop) { | |
2198 | /* | |
2199 | * Make it appear like a SCHED_FIFO task, its something | |
2200 | * userspace knows about and won't get confused about. | |
2201 | * | |
2202 | * Also, it will make PI more or less work without too | |
2203 | * much confusion -- but then, stop work should not | |
2204 | * rely on PI working anyway. | |
2205 | */ | |
2206 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
2207 | ||
2208 | stop->sched_class = &stop_sched_class; | |
2209 | } | |
2210 | ||
2211 | cpu_rq(cpu)->stop = stop; | |
2212 | ||
2213 | if (old_stop) { | |
2214 | /* | |
2215 | * Reset it back to a normal scheduling class so that | |
2216 | * it can die in pieces. | |
2217 | */ | |
2218 | old_stop->sched_class = &rt_sched_class; | |
2219 | } | |
2220 | } | |
2221 | ||
14531189 | 2222 | /* |
dd41f596 | 2223 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 2224 | */ |
14531189 IM |
2225 | static inline int __normal_prio(struct task_struct *p) |
2226 | { | |
dd41f596 | 2227 | return p->static_prio; |
14531189 IM |
2228 | } |
2229 | ||
b29739f9 IM |
2230 | /* |
2231 | * Calculate the expected normal priority: i.e. priority | |
2232 | * without taking RT-inheritance into account. Might be | |
2233 | * boosted by interactivity modifiers. Changes upon fork, | |
2234 | * setprio syscalls, and whenever the interactivity | |
2235 | * estimator recalculates. | |
2236 | */ | |
36c8b586 | 2237 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
2238 | { |
2239 | int prio; | |
2240 | ||
e05606d3 | 2241 | if (task_has_rt_policy(p)) |
b29739f9 IM |
2242 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
2243 | else | |
2244 | prio = __normal_prio(p); | |
2245 | return prio; | |
2246 | } | |
2247 | ||
2248 | /* | |
2249 | * Calculate the current priority, i.e. the priority | |
2250 | * taken into account by the scheduler. This value might | |
2251 | * be boosted by RT tasks, or might be boosted by | |
2252 | * interactivity modifiers. Will be RT if the task got | |
2253 | * RT-boosted. If not then it returns p->normal_prio. | |
2254 | */ | |
36c8b586 | 2255 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
2256 | { |
2257 | p->normal_prio = normal_prio(p); | |
2258 | /* | |
2259 | * If we are RT tasks or we were boosted to RT priority, | |
2260 | * keep the priority unchanged. Otherwise, update priority | |
2261 | * to the normal priority: | |
2262 | */ | |
2263 | if (!rt_prio(p->prio)) | |
2264 | return p->normal_prio; | |
2265 | return p->prio; | |
2266 | } | |
2267 | ||
1da177e4 LT |
2268 | /** |
2269 | * task_curr - is this task currently executing on a CPU? | |
2270 | * @p: the task in question. | |
2271 | */ | |
36c8b586 | 2272 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
2273 | { |
2274 | return cpu_curr(task_cpu(p)) == p; | |
2275 | } | |
2276 | ||
cb469845 SR |
2277 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
2278 | const struct sched_class *prev_class, | |
da7a735e | 2279 | int oldprio) |
cb469845 SR |
2280 | { |
2281 | if (prev_class != p->sched_class) { | |
2282 | if (prev_class->switched_from) | |
da7a735e PZ |
2283 | prev_class->switched_from(rq, p); |
2284 | p->sched_class->switched_to(rq, p); | |
2285 | } else if (oldprio != p->prio) | |
2286 | p->sched_class->prio_changed(rq, p, oldprio); | |
cb469845 SR |
2287 | } |
2288 | ||
1e5a7405 PZ |
2289 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
2290 | { | |
2291 | const struct sched_class *class; | |
2292 | ||
2293 | if (p->sched_class == rq->curr->sched_class) { | |
2294 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
2295 | } else { | |
2296 | for_each_class(class) { | |
2297 | if (class == rq->curr->sched_class) | |
2298 | break; | |
2299 | if (class == p->sched_class) { | |
2300 | resched_task(rq->curr); | |
2301 | break; | |
2302 | } | |
2303 | } | |
2304 | } | |
2305 | ||
2306 | /* | |
2307 | * A queue event has occurred, and we're going to schedule. In | |
2308 | * this case, we can save a useless back to back clock update. | |
2309 | */ | |
fd2f4419 | 2310 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
2311 | rq->skip_clock_update = 1; |
2312 | } | |
2313 | ||
1da177e4 | 2314 | #ifdef CONFIG_SMP |
cc367732 IM |
2315 | /* |
2316 | * Is this task likely cache-hot: | |
2317 | */ | |
e7693a36 | 2318 | static int |
cc367732 IM |
2319 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
2320 | { | |
2321 | s64 delta; | |
2322 | ||
e6c8fba7 PZ |
2323 | if (p->sched_class != &fair_sched_class) |
2324 | return 0; | |
2325 | ||
ef8002f6 NR |
2326 | if (unlikely(p->policy == SCHED_IDLE)) |
2327 | return 0; | |
2328 | ||
f540a608 IM |
2329 | /* |
2330 | * Buddy candidates are cache hot: | |
2331 | */ | |
f685ceac | 2332 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
2333 | (&p->se == cfs_rq_of(&p->se)->next || |
2334 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2335 | return 1; |
2336 | ||
6bc1665b IM |
2337 | if (sysctl_sched_migration_cost == -1) |
2338 | return 1; | |
2339 | if (sysctl_sched_migration_cost == 0) | |
2340 | return 0; | |
2341 | ||
cc367732 IM |
2342 | delta = now - p->se.exec_start; |
2343 | ||
2344 | return delta < (s64)sysctl_sched_migration_cost; | |
2345 | } | |
2346 | ||
dd41f596 | 2347 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2348 | { |
e2912009 PZ |
2349 | #ifdef CONFIG_SCHED_DEBUG |
2350 | /* | |
2351 | * We should never call set_task_cpu() on a blocked task, | |
2352 | * ttwu() will sort out the placement. | |
2353 | */ | |
077614ee PZ |
2354 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
2355 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
0122ec5b PZ |
2356 | |
2357 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
2358 | /* |
2359 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
2360 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
2361 | * | |
2362 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
2363 | * see set_task_rq(). | |
2364 | * | |
2365 | * Furthermore, all task_rq users should acquire both locks, see | |
2366 | * task_rq_lock(). | |
2367 | */ | |
0122ec5b PZ |
2368 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
2369 | lockdep_is_held(&task_rq(p)->lock))); | |
2370 | #endif | |
e2912009 PZ |
2371 | #endif |
2372 | ||
de1d7286 | 2373 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2374 | |
0c69774e PZ |
2375 | if (task_cpu(p) != new_cpu) { |
2376 | p->se.nr_migrations++; | |
a8b0ca17 | 2377 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); |
0c69774e | 2378 | } |
dd41f596 IM |
2379 | |
2380 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2381 | } |
2382 | ||
969c7921 | 2383 | struct migration_arg { |
36c8b586 | 2384 | struct task_struct *task; |
1da177e4 | 2385 | int dest_cpu; |
70b97a7f | 2386 | }; |
1da177e4 | 2387 | |
969c7921 TH |
2388 | static int migration_cpu_stop(void *data); |
2389 | ||
1da177e4 LT |
2390 | /* |
2391 | * wait_task_inactive - wait for a thread to unschedule. | |
2392 | * | |
85ba2d86 RM |
2393 | * If @match_state is nonzero, it's the @p->state value just checked and |
2394 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2395 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2396 | * we return a positive number (its total switch count). If a second call | |
2397 | * a short while later returns the same number, the caller can be sure that | |
2398 | * @p has remained unscheduled the whole time. | |
2399 | * | |
1da177e4 LT |
2400 | * The caller must ensure that the task *will* unschedule sometime soon, |
2401 | * else this function might spin for a *long* time. This function can't | |
2402 | * be called with interrupts off, or it may introduce deadlock with | |
2403 | * smp_call_function() if an IPI is sent by the same process we are | |
2404 | * waiting to become inactive. | |
2405 | */ | |
85ba2d86 | 2406 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2407 | { |
2408 | unsigned long flags; | |
dd41f596 | 2409 | int running, on_rq; |
85ba2d86 | 2410 | unsigned long ncsw; |
70b97a7f | 2411 | struct rq *rq; |
1da177e4 | 2412 | |
3a5c359a AK |
2413 | for (;;) { |
2414 | /* | |
2415 | * We do the initial early heuristics without holding | |
2416 | * any task-queue locks at all. We'll only try to get | |
2417 | * the runqueue lock when things look like they will | |
2418 | * work out! | |
2419 | */ | |
2420 | rq = task_rq(p); | |
fa490cfd | 2421 | |
3a5c359a AK |
2422 | /* |
2423 | * If the task is actively running on another CPU | |
2424 | * still, just relax and busy-wait without holding | |
2425 | * any locks. | |
2426 | * | |
2427 | * NOTE! Since we don't hold any locks, it's not | |
2428 | * even sure that "rq" stays as the right runqueue! | |
2429 | * But we don't care, since "task_running()" will | |
2430 | * return false if the runqueue has changed and p | |
2431 | * is actually now running somewhere else! | |
2432 | */ | |
85ba2d86 RM |
2433 | while (task_running(rq, p)) { |
2434 | if (match_state && unlikely(p->state != match_state)) | |
2435 | return 0; | |
3a5c359a | 2436 | cpu_relax(); |
85ba2d86 | 2437 | } |
fa490cfd | 2438 | |
3a5c359a AK |
2439 | /* |
2440 | * Ok, time to look more closely! We need the rq | |
2441 | * lock now, to be *sure*. If we're wrong, we'll | |
2442 | * just go back and repeat. | |
2443 | */ | |
2444 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 2445 | trace_sched_wait_task(p); |
3a5c359a | 2446 | running = task_running(rq, p); |
fd2f4419 | 2447 | on_rq = p->on_rq; |
85ba2d86 | 2448 | ncsw = 0; |
f31e11d8 | 2449 | if (!match_state || p->state == match_state) |
93dcf55f | 2450 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 2451 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 2452 | |
85ba2d86 RM |
2453 | /* |
2454 | * If it changed from the expected state, bail out now. | |
2455 | */ | |
2456 | if (unlikely(!ncsw)) | |
2457 | break; | |
2458 | ||
3a5c359a AK |
2459 | /* |
2460 | * Was it really running after all now that we | |
2461 | * checked with the proper locks actually held? | |
2462 | * | |
2463 | * Oops. Go back and try again.. | |
2464 | */ | |
2465 | if (unlikely(running)) { | |
2466 | cpu_relax(); | |
2467 | continue; | |
2468 | } | |
fa490cfd | 2469 | |
3a5c359a AK |
2470 | /* |
2471 | * It's not enough that it's not actively running, | |
2472 | * it must be off the runqueue _entirely_, and not | |
2473 | * preempted! | |
2474 | * | |
80dd99b3 | 2475 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2476 | * running right now), it's preempted, and we should |
2477 | * yield - it could be a while. | |
2478 | */ | |
2479 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
2480 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
2481 | ||
2482 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2483 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
2484 | continue; |
2485 | } | |
fa490cfd | 2486 | |
3a5c359a AK |
2487 | /* |
2488 | * Ahh, all good. It wasn't running, and it wasn't | |
2489 | * runnable, which means that it will never become | |
2490 | * running in the future either. We're all done! | |
2491 | */ | |
2492 | break; | |
2493 | } | |
85ba2d86 RM |
2494 | |
2495 | return ncsw; | |
1da177e4 LT |
2496 | } |
2497 | ||
2498 | /*** | |
2499 | * kick_process - kick a running thread to enter/exit the kernel | |
2500 | * @p: the to-be-kicked thread | |
2501 | * | |
2502 | * Cause a process which is running on another CPU to enter | |
2503 | * kernel-mode, without any delay. (to get signals handled.) | |
2504 | * | |
25985edc | 2505 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
2506 | * because all it wants to ensure is that the remote task enters |
2507 | * the kernel. If the IPI races and the task has been migrated | |
2508 | * to another CPU then no harm is done and the purpose has been | |
2509 | * achieved as well. | |
2510 | */ | |
36c8b586 | 2511 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2512 | { |
2513 | int cpu; | |
2514 | ||
2515 | preempt_disable(); | |
2516 | cpu = task_cpu(p); | |
2517 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2518 | smp_send_reschedule(cpu); | |
2519 | preempt_enable(); | |
2520 | } | |
b43e3521 | 2521 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2522 | #endif /* CONFIG_SMP */ |
1da177e4 | 2523 | |
970b13ba | 2524 | #ifdef CONFIG_SMP |
30da688e | 2525 | /* |
013fdb80 | 2526 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 2527 | */ |
5da9a0fb PZ |
2528 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2529 | { | |
2530 | int dest_cpu; | |
2531 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
2532 | ||
2533 | /* Look for allowed, online CPU in same node. */ | |
2534 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
2535 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
2536 | return dest_cpu; | |
2537 | ||
2538 | /* Any allowed, online CPU? */ | |
2539 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); | |
2540 | if (dest_cpu < nr_cpu_ids) | |
2541 | return dest_cpu; | |
2542 | ||
2543 | /* No more Mr. Nice Guy. */ | |
48c5ccae PZ |
2544 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
2545 | /* | |
2546 | * Don't tell them about moving exiting tasks or | |
2547 | * kernel threads (both mm NULL), since they never | |
2548 | * leave kernel. | |
2549 | */ | |
2550 | if (p->mm && printk_ratelimit()) { | |
2551 | printk(KERN_INFO "process %d (%s) no longer affine to cpu%d\n", | |
2552 | task_pid_nr(p), p->comm, cpu); | |
5da9a0fb PZ |
2553 | } |
2554 | ||
2555 | return dest_cpu; | |
2556 | } | |
2557 | ||
e2912009 | 2558 | /* |
013fdb80 | 2559 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 2560 | */ |
970b13ba | 2561 | static inline |
7608dec2 | 2562 | int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 2563 | { |
7608dec2 | 2564 | int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags); |
e2912009 PZ |
2565 | |
2566 | /* | |
2567 | * In order not to call set_task_cpu() on a blocking task we need | |
2568 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2569 | * cpu. | |
2570 | * | |
2571 | * Since this is common to all placement strategies, this lives here. | |
2572 | * | |
2573 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2574 | * not worry about this generic constraint ] | |
2575 | */ | |
2576 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
70f11205 | 2577 | !cpu_online(cpu))) |
5da9a0fb | 2578 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2579 | |
2580 | return cpu; | |
970b13ba | 2581 | } |
09a40af5 MG |
2582 | |
2583 | static void update_avg(u64 *avg, u64 sample) | |
2584 | { | |
2585 | s64 diff = sample - *avg; | |
2586 | *avg += diff >> 3; | |
2587 | } | |
970b13ba PZ |
2588 | #endif |
2589 | ||
d7c01d27 | 2590 | static void |
b84cb5df | 2591 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 2592 | { |
d7c01d27 | 2593 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
2594 | struct rq *rq = this_rq(); |
2595 | ||
d7c01d27 PZ |
2596 | #ifdef CONFIG_SMP |
2597 | int this_cpu = smp_processor_id(); | |
2598 | ||
2599 | if (cpu == this_cpu) { | |
2600 | schedstat_inc(rq, ttwu_local); | |
2601 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
2602 | } else { | |
2603 | struct sched_domain *sd; | |
2604 | ||
2605 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 2606 | rcu_read_lock(); |
d7c01d27 PZ |
2607 | for_each_domain(this_cpu, sd) { |
2608 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
2609 | schedstat_inc(sd, ttwu_wake_remote); | |
2610 | break; | |
2611 | } | |
2612 | } | |
057f3fad | 2613 | rcu_read_unlock(); |
d7c01d27 | 2614 | } |
f339b9dc PZ |
2615 | |
2616 | if (wake_flags & WF_MIGRATED) | |
2617 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
2618 | ||
d7c01d27 PZ |
2619 | #endif /* CONFIG_SMP */ |
2620 | ||
2621 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 2622 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
2623 | |
2624 | if (wake_flags & WF_SYNC) | |
9ed3811a | 2625 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 2626 | |
d7c01d27 PZ |
2627 | #endif /* CONFIG_SCHEDSTATS */ |
2628 | } | |
2629 | ||
2630 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
2631 | { | |
9ed3811a | 2632 | activate_task(rq, p, en_flags); |
fd2f4419 | 2633 | p->on_rq = 1; |
c2f7115e PZ |
2634 | |
2635 | /* if a worker is waking up, notify workqueue */ | |
2636 | if (p->flags & PF_WQ_WORKER) | |
2637 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
2638 | } |
2639 | ||
23f41eeb PZ |
2640 | /* |
2641 | * Mark the task runnable and perform wakeup-preemption. | |
2642 | */ | |
89363381 | 2643 | static void |
23f41eeb | 2644 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 2645 | { |
89363381 | 2646 | trace_sched_wakeup(p, true); |
9ed3811a TH |
2647 | check_preempt_curr(rq, p, wake_flags); |
2648 | ||
2649 | p->state = TASK_RUNNING; | |
2650 | #ifdef CONFIG_SMP | |
2651 | if (p->sched_class->task_woken) | |
2652 | p->sched_class->task_woken(rq, p); | |
2653 | ||
e69c6341 | 2654 | if (rq->idle_stamp) { |
9ed3811a TH |
2655 | u64 delta = rq->clock - rq->idle_stamp; |
2656 | u64 max = 2*sysctl_sched_migration_cost; | |
2657 | ||
2658 | if (delta > max) | |
2659 | rq->avg_idle = max; | |
2660 | else | |
2661 | update_avg(&rq->avg_idle, delta); | |
2662 | rq->idle_stamp = 0; | |
2663 | } | |
2664 | #endif | |
2665 | } | |
2666 | ||
c05fbafb PZ |
2667 | static void |
2668 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
2669 | { | |
2670 | #ifdef CONFIG_SMP | |
2671 | if (p->sched_contributes_to_load) | |
2672 | rq->nr_uninterruptible--; | |
2673 | #endif | |
2674 | ||
2675 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
2676 | ttwu_do_wakeup(rq, p, wake_flags); | |
2677 | } | |
2678 | ||
2679 | /* | |
2680 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
2681 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
2682 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
2683 | * the task is still ->on_rq. | |
2684 | */ | |
2685 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
2686 | { | |
2687 | struct rq *rq; | |
2688 | int ret = 0; | |
2689 | ||
2690 | rq = __task_rq_lock(p); | |
2691 | if (p->on_rq) { | |
2692 | ttwu_do_wakeup(rq, p, wake_flags); | |
2693 | ret = 1; | |
2694 | } | |
2695 | __task_rq_unlock(rq); | |
2696 | ||
2697 | return ret; | |
2698 | } | |
2699 | ||
317f3941 | 2700 | #ifdef CONFIG_SMP |
fa14ff4a | 2701 | static void sched_ttwu_pending(void) |
317f3941 PZ |
2702 | { |
2703 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
2704 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
2705 | struct task_struct *p; | |
317f3941 PZ |
2706 | |
2707 | raw_spin_lock(&rq->lock); | |
2708 | ||
fa14ff4a PZ |
2709 | while (llist) { |
2710 | p = llist_entry(llist, struct task_struct, wake_entry); | |
2711 | llist = llist_next(llist); | |
317f3941 PZ |
2712 | ttwu_do_activate(rq, p, 0); |
2713 | } | |
2714 | ||
2715 | raw_spin_unlock(&rq->lock); | |
2716 | } | |
2717 | ||
2718 | void scheduler_ipi(void) | |
2719 | { | |
fa14ff4a | 2720 | if (llist_empty(&this_rq()->wake_list)) |
c5d753a5 PZ |
2721 | return; |
2722 | ||
2723 | /* | |
2724 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
2725 | * traditionally all their work was done from the interrupt return | |
2726 | * path. Now that we actually do some work, we need to make sure | |
2727 | * we do call them. | |
2728 | * | |
2729 | * Some archs already do call them, luckily irq_enter/exit nest | |
2730 | * properly. | |
2731 | * | |
2732 | * Arguably we should visit all archs and update all handlers, | |
2733 | * however a fair share of IPIs are still resched only so this would | |
2734 | * somewhat pessimize the simple resched case. | |
2735 | */ | |
2736 | irq_enter(); | |
fa14ff4a | 2737 | sched_ttwu_pending(); |
c5d753a5 | 2738 | irq_exit(); |
317f3941 PZ |
2739 | } |
2740 | ||
2741 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
2742 | { | |
fa14ff4a | 2743 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) |
317f3941 PZ |
2744 | smp_send_reschedule(cpu); |
2745 | } | |
d6aa8f85 PZ |
2746 | |
2747 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
2748 | static int ttwu_activate_remote(struct task_struct *p, int wake_flags) | |
2749 | { | |
2750 | struct rq *rq; | |
2751 | int ret = 0; | |
2752 | ||
2753 | rq = __task_rq_lock(p); | |
2754 | if (p->on_cpu) { | |
2755 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); | |
2756 | ttwu_do_wakeup(rq, p, wake_flags); | |
2757 | ret = 1; | |
2758 | } | |
2759 | __task_rq_unlock(rq); | |
2760 | ||
2761 | return ret; | |
2762 | ||
2763 | } | |
2764 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
2765 | #endif /* CONFIG_SMP */ | |
317f3941 | 2766 | |
c05fbafb PZ |
2767 | static void ttwu_queue(struct task_struct *p, int cpu) |
2768 | { | |
2769 | struct rq *rq = cpu_rq(cpu); | |
2770 | ||
17d9f311 | 2771 | #if defined(CONFIG_SMP) |
317f3941 | 2772 | if (sched_feat(TTWU_QUEUE) && cpu != smp_processor_id()) { |
f01114cb | 2773 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
2774 | ttwu_queue_remote(p, cpu); |
2775 | return; | |
2776 | } | |
2777 | #endif | |
2778 | ||
c05fbafb PZ |
2779 | raw_spin_lock(&rq->lock); |
2780 | ttwu_do_activate(rq, p, 0); | |
2781 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
2782 | } |
2783 | ||
2784 | /** | |
1da177e4 | 2785 | * try_to_wake_up - wake up a thread |
9ed3811a | 2786 | * @p: the thread to be awakened |
1da177e4 | 2787 | * @state: the mask of task states that can be woken |
9ed3811a | 2788 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
2789 | * |
2790 | * Put it on the run-queue if it's not already there. The "current" | |
2791 | * thread is always on the run-queue (except when the actual | |
2792 | * re-schedule is in progress), and as such you're allowed to do | |
2793 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2794 | * runnable without the overhead of this. | |
2795 | * | |
9ed3811a TH |
2796 | * Returns %true if @p was woken up, %false if it was already running |
2797 | * or @state didn't match @p's state. | |
1da177e4 | 2798 | */ |
e4a52bcb PZ |
2799 | static int |
2800 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 2801 | { |
1da177e4 | 2802 | unsigned long flags; |
c05fbafb | 2803 | int cpu, success = 0; |
2398f2c6 | 2804 | |
04e2f174 | 2805 | smp_wmb(); |
013fdb80 | 2806 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 2807 | if (!(p->state & state)) |
1da177e4 LT |
2808 | goto out; |
2809 | ||
c05fbafb | 2810 | success = 1; /* we're going to change ->state */ |
1da177e4 | 2811 | cpu = task_cpu(p); |
1da177e4 | 2812 | |
c05fbafb PZ |
2813 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
2814 | goto stat; | |
1da177e4 | 2815 | |
1da177e4 | 2816 | #ifdef CONFIG_SMP |
e9c84311 | 2817 | /* |
c05fbafb PZ |
2818 | * If the owning (remote) cpu is still in the middle of schedule() with |
2819 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 2820 | */ |
e4a52bcb PZ |
2821 | while (p->on_cpu) { |
2822 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
2823 | /* | |
d6aa8f85 PZ |
2824 | * In case the architecture enables interrupts in |
2825 | * context_switch(), we cannot busy wait, since that | |
2826 | * would lead to deadlocks when an interrupt hits and | |
2827 | * tries to wake up @prev. So bail and do a complete | |
2828 | * remote wakeup. | |
e4a52bcb | 2829 | */ |
d6aa8f85 | 2830 | if (ttwu_activate_remote(p, wake_flags)) |
c05fbafb | 2831 | goto stat; |
d6aa8f85 | 2832 | #else |
e4a52bcb | 2833 | cpu_relax(); |
d6aa8f85 | 2834 | #endif |
371fd7e7 | 2835 | } |
0970d299 | 2836 | /* |
e4a52bcb | 2837 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 2838 | */ |
e4a52bcb | 2839 | smp_rmb(); |
1da177e4 | 2840 | |
a8e4f2ea | 2841 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 2842 | p->state = TASK_WAKING; |
e7693a36 | 2843 | |
e4a52bcb | 2844 | if (p->sched_class->task_waking) |
74f8e4b2 | 2845 | p->sched_class->task_waking(p); |
efbbd05a | 2846 | |
7608dec2 | 2847 | cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
2848 | if (task_cpu(p) != cpu) { |
2849 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 2850 | set_task_cpu(p, cpu); |
f339b9dc | 2851 | } |
1da177e4 | 2852 | #endif /* CONFIG_SMP */ |
1da177e4 | 2853 | |
c05fbafb PZ |
2854 | ttwu_queue(p, cpu); |
2855 | stat: | |
b84cb5df | 2856 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 2857 | out: |
013fdb80 | 2858 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
2859 | |
2860 | return success; | |
2861 | } | |
2862 | ||
21aa9af0 TH |
2863 | /** |
2864 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
2865 | * @p: the thread to be awakened | |
2866 | * | |
2acca55e | 2867 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 2868 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 2869 | * the current task. |
21aa9af0 TH |
2870 | */ |
2871 | static void try_to_wake_up_local(struct task_struct *p) | |
2872 | { | |
2873 | struct rq *rq = task_rq(p); | |
21aa9af0 TH |
2874 | |
2875 | BUG_ON(rq != this_rq()); | |
2876 | BUG_ON(p == current); | |
2877 | lockdep_assert_held(&rq->lock); | |
2878 | ||
2acca55e PZ |
2879 | if (!raw_spin_trylock(&p->pi_lock)) { |
2880 | raw_spin_unlock(&rq->lock); | |
2881 | raw_spin_lock(&p->pi_lock); | |
2882 | raw_spin_lock(&rq->lock); | |
2883 | } | |
2884 | ||
21aa9af0 | 2885 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 2886 | goto out; |
21aa9af0 | 2887 | |
fd2f4419 | 2888 | if (!p->on_rq) |
d7c01d27 PZ |
2889 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
2890 | ||
23f41eeb | 2891 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 2892 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
2893 | out: |
2894 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
2895 | } |
2896 | ||
50fa610a DH |
2897 | /** |
2898 | * wake_up_process - Wake up a specific process | |
2899 | * @p: The process to be woken up. | |
2900 | * | |
2901 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2902 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2903 | * running. | |
2904 | * | |
2905 | * It may be assumed that this function implies a write memory barrier before | |
2906 | * changing the task state if and only if any tasks are woken up. | |
2907 | */ | |
7ad5b3a5 | 2908 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2909 | { |
d9514f6c | 2910 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2911 | } |
1da177e4 LT |
2912 | EXPORT_SYMBOL(wake_up_process); |
2913 | ||
7ad5b3a5 | 2914 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2915 | { |
2916 | return try_to_wake_up(p, state, 0); | |
2917 | } | |
2918 | ||
1da177e4 LT |
2919 | /* |
2920 | * Perform scheduler related setup for a newly forked process p. | |
2921 | * p is forked by current. | |
dd41f596 IM |
2922 | * |
2923 | * __sched_fork() is basic setup used by init_idle() too: | |
2924 | */ | |
2925 | static void __sched_fork(struct task_struct *p) | |
2926 | { | |
fd2f4419 PZ |
2927 | p->on_rq = 0; |
2928 | ||
2929 | p->se.on_rq = 0; | |
dd41f596 IM |
2930 | p->se.exec_start = 0; |
2931 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2932 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2933 | p->se.nr_migrations = 0; |
da7a735e | 2934 | p->se.vruntime = 0; |
fd2f4419 | 2935 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
2936 | |
2937 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 2938 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2939 | #endif |
476d139c | 2940 | |
fa717060 | 2941 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 2942 | |
e107be36 AK |
2943 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2944 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2945 | #endif | |
dd41f596 IM |
2946 | } |
2947 | ||
2948 | /* | |
2949 | * fork()/clone()-time setup: | |
2950 | */ | |
3e51e3ed | 2951 | void sched_fork(struct task_struct *p) |
dd41f596 | 2952 | { |
0122ec5b | 2953 | unsigned long flags; |
dd41f596 IM |
2954 | int cpu = get_cpu(); |
2955 | ||
2956 | __sched_fork(p); | |
06b83b5f | 2957 | /* |
0017d735 | 2958 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2959 | * nobody will actually run it, and a signal or other external |
2960 | * event cannot wake it up and insert it on the runqueue either. | |
2961 | */ | |
0017d735 | 2962 | p->state = TASK_RUNNING; |
dd41f596 | 2963 | |
c350a04e MG |
2964 | /* |
2965 | * Make sure we do not leak PI boosting priority to the child. | |
2966 | */ | |
2967 | p->prio = current->normal_prio; | |
2968 | ||
b9dc29e7 MG |
2969 | /* |
2970 | * Revert to default priority/policy on fork if requested. | |
2971 | */ | |
2972 | if (unlikely(p->sched_reset_on_fork)) { | |
c350a04e | 2973 | if (task_has_rt_policy(p)) { |
b9dc29e7 | 2974 | p->policy = SCHED_NORMAL; |
6c697bdf | 2975 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
2976 | p->rt_priority = 0; |
2977 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
2978 | p->static_prio = NICE_TO_PRIO(0); | |
2979 | ||
2980 | p->prio = p->normal_prio = __normal_prio(p); | |
2981 | set_load_weight(p); | |
6c697bdf | 2982 | |
b9dc29e7 MG |
2983 | /* |
2984 | * We don't need the reset flag anymore after the fork. It has | |
2985 | * fulfilled its duty: | |
2986 | */ | |
2987 | p->sched_reset_on_fork = 0; | |
2988 | } | |
ca94c442 | 2989 | |
2ddbf952 HS |
2990 | if (!rt_prio(p->prio)) |
2991 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2992 | |
cd29fe6f PZ |
2993 | if (p->sched_class->task_fork) |
2994 | p->sched_class->task_fork(p); | |
2995 | ||
86951599 PZ |
2996 | /* |
2997 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2998 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2999 | * is ran before sched_fork(). | |
3000 | * | |
3001 | * Silence PROVE_RCU. | |
3002 | */ | |
0122ec5b | 3003 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 3004 | set_task_cpu(p, cpu); |
0122ec5b | 3005 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 3006 | |
52f17b6c | 3007 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 3008 | if (likely(sched_info_on())) |
52f17b6c | 3009 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 3010 | #endif |
3ca7a440 PZ |
3011 | #if defined(CONFIG_SMP) |
3012 | p->on_cpu = 0; | |
4866cde0 | 3013 | #endif |
bdd4e85d | 3014 | #ifdef CONFIG_PREEMPT_COUNT |
4866cde0 | 3015 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 3016 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 3017 | #endif |
806c09a7 | 3018 | #ifdef CONFIG_SMP |
917b627d | 3019 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
806c09a7 | 3020 | #endif |
917b627d | 3021 | |
476d139c | 3022 | put_cpu(); |
1da177e4 LT |
3023 | } |
3024 | ||
3025 | /* | |
3026 | * wake_up_new_task - wake up a newly created task for the first time. | |
3027 | * | |
3028 | * This function will do some initial scheduler statistics housekeeping | |
3029 | * that must be done for every newly created context, then puts the task | |
3030 | * on the runqueue and wakes it. | |
3031 | */ | |
3e51e3ed | 3032 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
3033 | { |
3034 | unsigned long flags; | |
dd41f596 | 3035 | struct rq *rq; |
fabf318e | 3036 | |
ab2515c4 | 3037 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
3038 | #ifdef CONFIG_SMP |
3039 | /* | |
3040 | * Fork balancing, do it here and not earlier because: | |
3041 | * - cpus_allowed can change in the fork path | |
3042 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 3043 | */ |
ab2515c4 | 3044 | set_task_cpu(p, select_task_rq(p, SD_BALANCE_FORK, 0)); |
0017d735 PZ |
3045 | #endif |
3046 | ||
ab2515c4 | 3047 | rq = __task_rq_lock(p); |
cd29fe6f | 3048 | activate_task(rq, p, 0); |
fd2f4419 | 3049 | p->on_rq = 1; |
89363381 | 3050 | trace_sched_wakeup_new(p, true); |
a7558e01 | 3051 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 3052 | #ifdef CONFIG_SMP |
efbbd05a PZ |
3053 | if (p->sched_class->task_woken) |
3054 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 3055 | #endif |
0122ec5b | 3056 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
3057 | } |
3058 | ||
e107be36 AK |
3059 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
3060 | ||
3061 | /** | |
80dd99b3 | 3062 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 3063 | * @notifier: notifier struct to register |
e107be36 AK |
3064 | */ |
3065 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
3066 | { | |
3067 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
3068 | } | |
3069 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
3070 | ||
3071 | /** | |
3072 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 3073 | * @notifier: notifier struct to unregister |
e107be36 AK |
3074 | * |
3075 | * This is safe to call from within a preemption notifier. | |
3076 | */ | |
3077 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
3078 | { | |
3079 | hlist_del(¬ifier->link); | |
3080 | } | |
3081 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
3082 | ||
3083 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
3084 | { | |
3085 | struct preempt_notifier *notifier; | |
3086 | struct hlist_node *node; | |
3087 | ||
3088 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
3089 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
3090 | } | |
3091 | ||
3092 | static void | |
3093 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
3094 | struct task_struct *next) | |
3095 | { | |
3096 | struct preempt_notifier *notifier; | |
3097 | struct hlist_node *node; | |
3098 | ||
3099 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
3100 | notifier->ops->sched_out(notifier, next); | |
3101 | } | |
3102 | ||
6d6bc0ad | 3103 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
3104 | |
3105 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
3106 | { | |
3107 | } | |
3108 | ||
3109 | static void | |
3110 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
3111 | struct task_struct *next) | |
3112 | { | |
3113 | } | |
3114 | ||
6d6bc0ad | 3115 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 3116 | |
4866cde0 NP |
3117 | /** |
3118 | * prepare_task_switch - prepare to switch tasks | |
3119 | * @rq: the runqueue preparing to switch | |
421cee29 | 3120 | * @prev: the current task that is being switched out |
4866cde0 NP |
3121 | * @next: the task we are going to switch to. |
3122 | * | |
3123 | * This is called with the rq lock held and interrupts off. It must | |
3124 | * be paired with a subsequent finish_task_switch after the context | |
3125 | * switch. | |
3126 | * | |
3127 | * prepare_task_switch sets up locking and calls architecture specific | |
3128 | * hooks. | |
3129 | */ | |
e107be36 AK |
3130 | static inline void |
3131 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
3132 | struct task_struct *next) | |
4866cde0 | 3133 | { |
fe4b04fa PZ |
3134 | sched_info_switch(prev, next); |
3135 | perf_event_task_sched_out(prev, next); | |
e107be36 | 3136 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
3137 | prepare_lock_switch(rq, next); |
3138 | prepare_arch_switch(next); | |
fe4b04fa | 3139 | trace_sched_switch(prev, next); |
4866cde0 NP |
3140 | } |
3141 | ||
1da177e4 LT |
3142 | /** |
3143 | * finish_task_switch - clean up after a task-switch | |
344babaa | 3144 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
3145 | * @prev: the thread we just switched away from. |
3146 | * | |
4866cde0 NP |
3147 | * finish_task_switch must be called after the context switch, paired |
3148 | * with a prepare_task_switch call before the context switch. | |
3149 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
3150 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
3151 | * |
3152 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 3153 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
3154 | * with the lock held can cause deadlocks; see schedule() for |
3155 | * details.) | |
3156 | */ | |
a9957449 | 3157 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
3158 | __releases(rq->lock) |
3159 | { | |
1da177e4 | 3160 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 3161 | long prev_state; |
1da177e4 LT |
3162 | |
3163 | rq->prev_mm = NULL; | |
3164 | ||
3165 | /* | |
3166 | * A task struct has one reference for the use as "current". | |
c394cc9f | 3167 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
3168 | * schedule one last time. The schedule call will never return, and |
3169 | * the scheduled task must drop that reference. | |
c394cc9f | 3170 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
3171 | * still held, otherwise prev could be scheduled on another cpu, die |
3172 | * there before we look at prev->state, and then the reference would | |
3173 | * be dropped twice. | |
3174 | * Manfred Spraul <manfred@colorfullife.com> | |
3175 | */ | |
55a101f8 | 3176 | prev_state = prev->state; |
4866cde0 | 3177 | finish_arch_switch(prev); |
8381f65d JI |
3178 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
3179 | local_irq_disable(); | |
3180 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
a8d757ef | 3181 | perf_event_task_sched_in(prev, current); |
8381f65d JI |
3182 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
3183 | local_irq_enable(); | |
3184 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 3185 | finish_lock_switch(rq, prev); |
e8fa1362 | 3186 | |
e107be36 | 3187 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
3188 | if (mm) |
3189 | mmdrop(mm); | |
c394cc9f | 3190 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 3191 | /* |
3192 | * Remove function-return probe instances associated with this | |
3193 | * task and put them back on the free list. | |
9761eea8 | 3194 | */ |
c6fd91f0 | 3195 | kprobe_flush_task(prev); |
1da177e4 | 3196 | put_task_struct(prev); |
c6fd91f0 | 3197 | } |
1da177e4 LT |
3198 | } |
3199 | ||
3f029d3c GH |
3200 | #ifdef CONFIG_SMP |
3201 | ||
3202 | /* assumes rq->lock is held */ | |
3203 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
3204 | { | |
3205 | if (prev->sched_class->pre_schedule) | |
3206 | prev->sched_class->pre_schedule(rq, prev); | |
3207 | } | |
3208 | ||
3209 | /* rq->lock is NOT held, but preemption is disabled */ | |
3210 | static inline void post_schedule(struct rq *rq) | |
3211 | { | |
3212 | if (rq->post_schedule) { | |
3213 | unsigned long flags; | |
3214 | ||
05fa785c | 3215 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
3216 | if (rq->curr->sched_class->post_schedule) |
3217 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 3218 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
3219 | |
3220 | rq->post_schedule = 0; | |
3221 | } | |
3222 | } | |
3223 | ||
3224 | #else | |
da19ab51 | 3225 | |
3f029d3c GH |
3226 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
3227 | { | |
3228 | } | |
3229 | ||
3230 | static inline void post_schedule(struct rq *rq) | |
3231 | { | |
1da177e4 LT |
3232 | } |
3233 | ||
3f029d3c GH |
3234 | #endif |
3235 | ||
1da177e4 LT |
3236 | /** |
3237 | * schedule_tail - first thing a freshly forked thread must call. | |
3238 | * @prev: the thread we just switched away from. | |
3239 | */ | |
36c8b586 | 3240 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
3241 | __releases(rq->lock) |
3242 | { | |
70b97a7f IM |
3243 | struct rq *rq = this_rq(); |
3244 | ||
4866cde0 | 3245 | finish_task_switch(rq, prev); |
da19ab51 | 3246 | |
3f029d3c GH |
3247 | /* |
3248 | * FIXME: do we need to worry about rq being invalidated by the | |
3249 | * task_switch? | |
3250 | */ | |
3251 | post_schedule(rq); | |
70b97a7f | 3252 | |
4866cde0 NP |
3253 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
3254 | /* In this case, finish_task_switch does not reenable preemption */ | |
3255 | preempt_enable(); | |
3256 | #endif | |
1da177e4 | 3257 | if (current->set_child_tid) |
b488893a | 3258 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
3259 | } |
3260 | ||
3261 | /* | |
3262 | * context_switch - switch to the new MM and the new | |
3263 | * thread's register state. | |
3264 | */ | |
dd41f596 | 3265 | static inline void |
70b97a7f | 3266 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 3267 | struct task_struct *next) |
1da177e4 | 3268 | { |
dd41f596 | 3269 | struct mm_struct *mm, *oldmm; |
1da177e4 | 3270 | |
e107be36 | 3271 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 3272 | |
dd41f596 IM |
3273 | mm = next->mm; |
3274 | oldmm = prev->active_mm; | |
9226d125 ZA |
3275 | /* |
3276 | * For paravirt, this is coupled with an exit in switch_to to | |
3277 | * combine the page table reload and the switch backend into | |
3278 | * one hypercall. | |
3279 | */ | |
224101ed | 3280 | arch_start_context_switch(prev); |
9226d125 | 3281 | |
31915ab4 | 3282 | if (!mm) { |
1da177e4 LT |
3283 | next->active_mm = oldmm; |
3284 | atomic_inc(&oldmm->mm_count); | |
3285 | enter_lazy_tlb(oldmm, next); | |
3286 | } else | |
3287 | switch_mm(oldmm, mm, next); | |
3288 | ||
31915ab4 | 3289 | if (!prev->mm) { |
1da177e4 | 3290 | prev->active_mm = NULL; |
1da177e4 LT |
3291 | rq->prev_mm = oldmm; |
3292 | } | |
3a5f5e48 IM |
3293 | /* |
3294 | * Since the runqueue lock will be released by the next | |
3295 | * task (which is an invalid locking op but in the case | |
3296 | * of the scheduler it's an obvious special-case), so we | |
3297 | * do an early lockdep release here: | |
3298 | */ | |
3299 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 3300 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 3301 | #endif |
1da177e4 LT |
3302 | |
3303 | /* Here we just switch the register state and the stack. */ | |
3304 | switch_to(prev, next, prev); | |
3305 | ||
dd41f596 IM |
3306 | barrier(); |
3307 | /* | |
3308 | * this_rq must be evaluated again because prev may have moved | |
3309 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
3310 | * frame will be invalid. | |
3311 | */ | |
3312 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
3313 | } |
3314 | ||
3315 | /* | |
3316 | * nr_running, nr_uninterruptible and nr_context_switches: | |
3317 | * | |
3318 | * externally visible scheduler statistics: current number of runnable | |
3319 | * threads, current number of uninterruptible-sleeping threads, total | |
3320 | * number of context switches performed since bootup. | |
3321 | */ | |
3322 | unsigned long nr_running(void) | |
3323 | { | |
3324 | unsigned long i, sum = 0; | |
3325 | ||
3326 | for_each_online_cpu(i) | |
3327 | sum += cpu_rq(i)->nr_running; | |
3328 | ||
3329 | return sum; | |
f711f609 | 3330 | } |
1da177e4 LT |
3331 | |
3332 | unsigned long nr_uninterruptible(void) | |
f711f609 | 3333 | { |
1da177e4 | 3334 | unsigned long i, sum = 0; |
f711f609 | 3335 | |
0a945022 | 3336 | for_each_possible_cpu(i) |
1da177e4 | 3337 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
3338 | |
3339 | /* | |
1da177e4 LT |
3340 | * Since we read the counters lockless, it might be slightly |
3341 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 3342 | */ |
1da177e4 LT |
3343 | if (unlikely((long)sum < 0)) |
3344 | sum = 0; | |
f711f609 | 3345 | |
1da177e4 | 3346 | return sum; |
f711f609 | 3347 | } |
f711f609 | 3348 | |
1da177e4 | 3349 | unsigned long long nr_context_switches(void) |
46cb4b7c | 3350 | { |
cc94abfc SR |
3351 | int i; |
3352 | unsigned long long sum = 0; | |
46cb4b7c | 3353 | |
0a945022 | 3354 | for_each_possible_cpu(i) |
1da177e4 | 3355 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 3356 | |
1da177e4 LT |
3357 | return sum; |
3358 | } | |
483b4ee6 | 3359 | |
1da177e4 LT |
3360 | unsigned long nr_iowait(void) |
3361 | { | |
3362 | unsigned long i, sum = 0; | |
483b4ee6 | 3363 | |
0a945022 | 3364 | for_each_possible_cpu(i) |
1da177e4 | 3365 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 3366 | |
1da177e4 LT |
3367 | return sum; |
3368 | } | |
483b4ee6 | 3369 | |
8c215bd3 | 3370 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 3371 | { |
8c215bd3 | 3372 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
3373 | return atomic_read(&this->nr_iowait); |
3374 | } | |
46cb4b7c | 3375 | |
69d25870 AV |
3376 | unsigned long this_cpu_load(void) |
3377 | { | |
3378 | struct rq *this = this_rq(); | |
3379 | return this->cpu_load[0]; | |
3380 | } | |
e790fb0b | 3381 | |
46cb4b7c | 3382 | |
dce48a84 TG |
3383 | /* Variables and functions for calc_load */ |
3384 | static atomic_long_t calc_load_tasks; | |
3385 | static unsigned long calc_load_update; | |
3386 | unsigned long avenrun[3]; | |
3387 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 3388 | |
74f5187a PZ |
3389 | static long calc_load_fold_active(struct rq *this_rq) |
3390 | { | |
3391 | long nr_active, delta = 0; | |
3392 | ||
3393 | nr_active = this_rq->nr_running; | |
3394 | nr_active += (long) this_rq->nr_uninterruptible; | |
3395 | ||
3396 | if (nr_active != this_rq->calc_load_active) { | |
3397 | delta = nr_active - this_rq->calc_load_active; | |
3398 | this_rq->calc_load_active = nr_active; | |
3399 | } | |
3400 | ||
3401 | return delta; | |
3402 | } | |
3403 | ||
0f004f5a PZ |
3404 | static unsigned long |
3405 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
3406 | { | |
3407 | load *= exp; | |
3408 | load += active * (FIXED_1 - exp); | |
3409 | load += 1UL << (FSHIFT - 1); | |
3410 | return load >> FSHIFT; | |
3411 | } | |
3412 | ||
74f5187a PZ |
3413 | #ifdef CONFIG_NO_HZ |
3414 | /* | |
3415 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | |
3416 | * | |
3417 | * When making the ILB scale, we should try to pull this in as well. | |
3418 | */ | |
3419 | static atomic_long_t calc_load_tasks_idle; | |
3420 | ||
3421 | static void calc_load_account_idle(struct rq *this_rq) | |
3422 | { | |
3423 | long delta; | |
3424 | ||
3425 | delta = calc_load_fold_active(this_rq); | |
3426 | if (delta) | |
3427 | atomic_long_add(delta, &calc_load_tasks_idle); | |
3428 | } | |
3429 | ||
3430 | static long calc_load_fold_idle(void) | |
3431 | { | |
3432 | long delta = 0; | |
3433 | ||
3434 | /* | |
3435 | * Its got a race, we don't care... | |
3436 | */ | |
3437 | if (atomic_long_read(&calc_load_tasks_idle)) | |
3438 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | |
3439 | ||
3440 | return delta; | |
3441 | } | |
0f004f5a PZ |
3442 | |
3443 | /** | |
3444 | * fixed_power_int - compute: x^n, in O(log n) time | |
3445 | * | |
3446 | * @x: base of the power | |
3447 | * @frac_bits: fractional bits of @x | |
3448 | * @n: power to raise @x to. | |
3449 | * | |
3450 | * By exploiting the relation between the definition of the natural power | |
3451 | * function: x^n := x*x*...*x (x multiplied by itself for n times), and | |
3452 | * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i, | |
3453 | * (where: n_i \elem {0, 1}, the binary vector representing n), | |
3454 | * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is | |
3455 | * of course trivially computable in O(log_2 n), the length of our binary | |
3456 | * vector. | |
3457 | */ | |
3458 | static unsigned long | |
3459 | fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n) | |
3460 | { | |
3461 | unsigned long result = 1UL << frac_bits; | |
3462 | ||
3463 | if (n) for (;;) { | |
3464 | if (n & 1) { | |
3465 | result *= x; | |
3466 | result += 1UL << (frac_bits - 1); | |
3467 | result >>= frac_bits; | |
3468 | } | |
3469 | n >>= 1; | |
3470 | if (!n) | |
3471 | break; | |
3472 | x *= x; | |
3473 | x += 1UL << (frac_bits - 1); | |
3474 | x >>= frac_bits; | |
3475 | } | |
3476 | ||
3477 | return result; | |
3478 | } | |
3479 | ||
3480 | /* | |
3481 | * a1 = a0 * e + a * (1 - e) | |
3482 | * | |
3483 | * a2 = a1 * e + a * (1 - e) | |
3484 | * = (a0 * e + a * (1 - e)) * e + a * (1 - e) | |
3485 | * = a0 * e^2 + a * (1 - e) * (1 + e) | |
3486 | * | |
3487 | * a3 = a2 * e + a * (1 - e) | |
3488 | * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e) | |
3489 | * = a0 * e^3 + a * (1 - e) * (1 + e + e^2) | |
3490 | * | |
3491 | * ... | |
3492 | * | |
3493 | * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1] | |
3494 | * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e) | |
3495 | * = a0 * e^n + a * (1 - e^n) | |
3496 | * | |
3497 | * [1] application of the geometric series: | |
3498 | * | |
3499 | * n 1 - x^(n+1) | |
3500 | * S_n := \Sum x^i = ------------- | |
3501 | * i=0 1 - x | |
3502 | */ | |
3503 | static unsigned long | |
3504 | calc_load_n(unsigned long load, unsigned long exp, | |
3505 | unsigned long active, unsigned int n) | |
3506 | { | |
3507 | ||
3508 | return calc_load(load, fixed_power_int(exp, FSHIFT, n), active); | |
3509 | } | |
3510 | ||
3511 | /* | |
3512 | * NO_HZ can leave us missing all per-cpu ticks calling | |
3513 | * calc_load_account_active(), but since an idle CPU folds its delta into | |
3514 | * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold | |
3515 | * in the pending idle delta if our idle period crossed a load cycle boundary. | |
3516 | * | |
3517 | * Once we've updated the global active value, we need to apply the exponential | |
3518 | * weights adjusted to the number of cycles missed. | |
3519 | */ | |
3520 | static void calc_global_nohz(unsigned long ticks) | |
3521 | { | |
3522 | long delta, active, n; | |
3523 | ||
3524 | if (time_before(jiffies, calc_load_update)) | |
3525 | return; | |
3526 | ||
3527 | /* | |
3528 | * If we crossed a calc_load_update boundary, make sure to fold | |
3529 | * any pending idle changes, the respective CPUs might have | |
3530 | * missed the tick driven calc_load_account_active() update | |
3531 | * due to NO_HZ. | |
3532 | */ | |
3533 | delta = calc_load_fold_idle(); | |
3534 | if (delta) | |
3535 | atomic_long_add(delta, &calc_load_tasks); | |
3536 | ||
3537 | /* | |
3538 | * If we were idle for multiple load cycles, apply them. | |
3539 | */ | |
3540 | if (ticks >= LOAD_FREQ) { | |
3541 | n = ticks / LOAD_FREQ; | |
3542 | ||
3543 | active = atomic_long_read(&calc_load_tasks); | |
3544 | active = active > 0 ? active * FIXED_1 : 0; | |
3545 | ||
3546 | avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); | |
3547 | avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); | |
3548 | avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); | |
3549 | ||
3550 | calc_load_update += n * LOAD_FREQ; | |
3551 | } | |
3552 | ||
3553 | /* | |
3554 | * Its possible the remainder of the above division also crosses | |
3555 | * a LOAD_FREQ period, the regular check in calc_global_load() | |
3556 | * which comes after this will take care of that. | |
3557 | * | |
3558 | * Consider us being 11 ticks before a cycle completion, and us | |
3559 | * sleeping for 4*LOAD_FREQ + 22 ticks, then the above code will | |
3560 | * age us 4 cycles, and the test in calc_global_load() will | |
3561 | * pick up the final one. | |
3562 | */ | |
3563 | } | |
74f5187a PZ |
3564 | #else |
3565 | static void calc_load_account_idle(struct rq *this_rq) | |
3566 | { | |
3567 | } | |
3568 | ||
3569 | static inline long calc_load_fold_idle(void) | |
3570 | { | |
3571 | return 0; | |
3572 | } | |
0f004f5a PZ |
3573 | |
3574 | static void calc_global_nohz(unsigned long ticks) | |
3575 | { | |
3576 | } | |
74f5187a PZ |
3577 | #endif |
3578 | ||
2d02494f TG |
3579 | /** |
3580 | * get_avenrun - get the load average array | |
3581 | * @loads: pointer to dest load array | |
3582 | * @offset: offset to add | |
3583 | * @shift: shift count to shift the result left | |
3584 | * | |
3585 | * These values are estimates at best, so no need for locking. | |
3586 | */ | |
3587 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
3588 | { | |
3589 | loads[0] = (avenrun[0] + offset) << shift; | |
3590 | loads[1] = (avenrun[1] + offset) << shift; | |
3591 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 3592 | } |
46cb4b7c | 3593 | |
46cb4b7c | 3594 | /* |
dce48a84 TG |
3595 | * calc_load - update the avenrun load estimates 10 ticks after the |
3596 | * CPUs have updated calc_load_tasks. | |
7835b98b | 3597 | */ |
0f004f5a | 3598 | void calc_global_load(unsigned long ticks) |
7835b98b | 3599 | { |
dce48a84 | 3600 | long active; |
1da177e4 | 3601 | |
0f004f5a PZ |
3602 | calc_global_nohz(ticks); |
3603 | ||
3604 | if (time_before(jiffies, calc_load_update + 10)) | |
dce48a84 | 3605 | return; |
1da177e4 | 3606 | |
dce48a84 TG |
3607 | active = atomic_long_read(&calc_load_tasks); |
3608 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 3609 | |
dce48a84 TG |
3610 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3611 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3612 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 3613 | |
dce48a84 TG |
3614 | calc_load_update += LOAD_FREQ; |
3615 | } | |
1da177e4 | 3616 | |
dce48a84 | 3617 | /* |
74f5187a PZ |
3618 | * Called from update_cpu_load() to periodically update this CPU's |
3619 | * active count. | |
dce48a84 TG |
3620 | */ |
3621 | static void calc_load_account_active(struct rq *this_rq) | |
3622 | { | |
74f5187a | 3623 | long delta; |
08c183f3 | 3624 | |
74f5187a PZ |
3625 | if (time_before(jiffies, this_rq->calc_load_update)) |
3626 | return; | |
783609c6 | 3627 | |
74f5187a PZ |
3628 | delta = calc_load_fold_active(this_rq); |
3629 | delta += calc_load_fold_idle(); | |
3630 | if (delta) | |
dce48a84 | 3631 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
3632 | |
3633 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
3634 | } |
3635 | ||
fdf3e95d VP |
3636 | /* |
3637 | * The exact cpuload at various idx values, calculated at every tick would be | |
3638 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | |
3639 | * | |
3640 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | |
3641 | * on nth tick when cpu may be busy, then we have: | |
3642 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3643 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | |
3644 | * | |
3645 | * decay_load_missed() below does efficient calculation of | |
3646 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3647 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | |
3648 | * | |
3649 | * The calculation is approximated on a 128 point scale. | |
3650 | * degrade_zero_ticks is the number of ticks after which load at any | |
3651 | * particular idx is approximated to be zero. | |
3652 | * degrade_factor is a precomputed table, a row for each load idx. | |
3653 | * Each column corresponds to degradation factor for a power of two ticks, | |
3654 | * based on 128 point scale. | |
3655 | * Example: | |
3656 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | |
3657 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | |
3658 | * | |
3659 | * With this power of 2 load factors, we can degrade the load n times | |
3660 | * by looking at 1 bits in n and doing as many mult/shift instead of | |
3661 | * n mult/shifts needed by the exact degradation. | |
3662 | */ | |
3663 | #define DEGRADE_SHIFT 7 | |
3664 | static const unsigned char | |
3665 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | |
3666 | static const unsigned char | |
3667 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | |
3668 | {0, 0, 0, 0, 0, 0, 0, 0}, | |
3669 | {64, 32, 8, 0, 0, 0, 0, 0}, | |
3670 | {96, 72, 40, 12, 1, 0, 0}, | |
3671 | {112, 98, 75, 43, 15, 1, 0}, | |
3672 | {120, 112, 98, 76, 45, 16, 2} }; | |
3673 | ||
3674 | /* | |
3675 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | |
3676 | * would be when CPU is idle and so we just decay the old load without | |
3677 | * adding any new load. | |
3678 | */ | |
3679 | static unsigned long | |
3680 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | |
3681 | { | |
3682 | int j = 0; | |
3683 | ||
3684 | if (!missed_updates) | |
3685 | return load; | |
3686 | ||
3687 | if (missed_updates >= degrade_zero_ticks[idx]) | |
3688 | return 0; | |
3689 | ||
3690 | if (idx == 1) | |
3691 | return load >> missed_updates; | |
3692 | ||
3693 | while (missed_updates) { | |
3694 | if (missed_updates % 2) | |
3695 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | |
3696 | ||
3697 | missed_updates >>= 1; | |
3698 | j++; | |
3699 | } | |
3700 | return load; | |
3701 | } | |
3702 | ||
46cb4b7c | 3703 | /* |
dd41f596 | 3704 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
3705 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
3706 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 3707 | */ |
dd41f596 | 3708 | static void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 3709 | { |
495eca49 | 3710 | unsigned long this_load = this_rq->load.weight; |
fdf3e95d VP |
3711 | unsigned long curr_jiffies = jiffies; |
3712 | unsigned long pending_updates; | |
dd41f596 | 3713 | int i, scale; |
46cb4b7c | 3714 | |
dd41f596 | 3715 | this_rq->nr_load_updates++; |
46cb4b7c | 3716 | |
fdf3e95d VP |
3717 | /* Avoid repeated calls on same jiffy, when moving in and out of idle */ |
3718 | if (curr_jiffies == this_rq->last_load_update_tick) | |
3719 | return; | |
3720 | ||
3721 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
3722 | this_rq->last_load_update_tick = curr_jiffies; | |
3723 | ||
dd41f596 | 3724 | /* Update our load: */ |
fdf3e95d VP |
3725 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
3726 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 3727 | unsigned long old_load, new_load; |
7d1e6a9b | 3728 | |
dd41f596 | 3729 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 3730 | |
dd41f596 | 3731 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 3732 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 3733 | new_load = this_load; |
a25707f3 IM |
3734 | /* |
3735 | * Round up the averaging division if load is increasing. This | |
3736 | * prevents us from getting stuck on 9 if the load is 10, for | |
3737 | * example. | |
3738 | */ | |
3739 | if (new_load > old_load) | |
fdf3e95d VP |
3740 | new_load += scale - 1; |
3741 | ||
3742 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 3743 | } |
da2b71ed SS |
3744 | |
3745 | sched_avg_update(this_rq); | |
fdf3e95d VP |
3746 | } |
3747 | ||
3748 | static void update_cpu_load_active(struct rq *this_rq) | |
3749 | { | |
3750 | update_cpu_load(this_rq); | |
46cb4b7c | 3751 | |
74f5187a | 3752 | calc_load_account_active(this_rq); |
46cb4b7c SS |
3753 | } |
3754 | ||
dd41f596 | 3755 | #ifdef CONFIG_SMP |
8a0be9ef | 3756 | |
46cb4b7c | 3757 | /* |
38022906 PZ |
3758 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3759 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3760 | */ |
38022906 | 3761 | void sched_exec(void) |
46cb4b7c | 3762 | { |
38022906 | 3763 | struct task_struct *p = current; |
1da177e4 | 3764 | unsigned long flags; |
0017d735 | 3765 | int dest_cpu; |
46cb4b7c | 3766 | |
8f42ced9 | 3767 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
7608dec2 | 3768 | dest_cpu = p->sched_class->select_task_rq(p, SD_BALANCE_EXEC, 0); |
0017d735 PZ |
3769 | if (dest_cpu == smp_processor_id()) |
3770 | goto unlock; | |
38022906 | 3771 | |
8f42ced9 | 3772 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 3773 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 3774 | |
8f42ced9 PZ |
3775 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
3776 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
3777 | return; |
3778 | } | |
0017d735 | 3779 | unlock: |
8f42ced9 | 3780 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 3781 | } |
dd41f596 | 3782 | |
1da177e4 LT |
3783 | #endif |
3784 | ||
1da177e4 LT |
3785 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3786 | ||
3787 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3788 | ||
3789 | /* | |
c5f8d995 | 3790 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 3791 | * @p in case that task is currently running. |
c5f8d995 HS |
3792 | * |
3793 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 3794 | */ |
c5f8d995 HS |
3795 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
3796 | { | |
3797 | u64 ns = 0; | |
3798 | ||
3799 | if (task_current(rq, p)) { | |
3800 | update_rq_clock(rq); | |
305e6835 | 3801 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
3802 | if ((s64)ns < 0) |
3803 | ns = 0; | |
3804 | } | |
3805 | ||
3806 | return ns; | |
3807 | } | |
3808 | ||
bb34d92f | 3809 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 3810 | { |
1da177e4 | 3811 | unsigned long flags; |
41b86e9c | 3812 | struct rq *rq; |
bb34d92f | 3813 | u64 ns = 0; |
48f24c4d | 3814 | |
41b86e9c | 3815 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 3816 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 3817 | task_rq_unlock(rq, p, &flags); |
1508487e | 3818 | |
c5f8d995 HS |
3819 | return ns; |
3820 | } | |
f06febc9 | 3821 | |
c5f8d995 HS |
3822 | /* |
3823 | * Return accounted runtime for the task. | |
3824 | * In case the task is currently running, return the runtime plus current's | |
3825 | * pending runtime that have not been accounted yet. | |
3826 | */ | |
3827 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3828 | { | |
3829 | unsigned long flags; | |
3830 | struct rq *rq; | |
3831 | u64 ns = 0; | |
3832 | ||
3833 | rq = task_rq_lock(p, &flags); | |
3834 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 3835 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
3836 | |
3837 | return ns; | |
3838 | } | |
48f24c4d | 3839 | |
1da177e4 LT |
3840 | /* |
3841 | * Account user cpu time to a process. | |
3842 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 3843 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 3844 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 3845 | */ |
457533a7 MS |
3846 | void account_user_time(struct task_struct *p, cputime_t cputime, |
3847 | cputime_t cputime_scaled) | |
1da177e4 LT |
3848 | { |
3849 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3850 | cputime64_t tmp; | |
3851 | ||
457533a7 | 3852 | /* Add user time to process. */ |
1da177e4 | 3853 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3854 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3855 | account_group_user_time(p, cputime); |
1da177e4 LT |
3856 | |
3857 | /* Add user time to cpustat. */ | |
3858 | tmp = cputime_to_cputime64(cputime); | |
3859 | if (TASK_NICE(p) > 0) | |
3860 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3861 | else | |
3862 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
3863 | |
3864 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
3865 | /* Account for user time used */ |
3866 | acct_update_integrals(p); | |
1da177e4 LT |
3867 | } |
3868 | ||
94886b84 LV |
3869 | /* |
3870 | * Account guest cpu time to a process. | |
3871 | * @p: the process that the cpu time gets accounted to | |
3872 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 3873 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 3874 | */ |
457533a7 MS |
3875 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
3876 | cputime_t cputime_scaled) | |
94886b84 LV |
3877 | { |
3878 | cputime64_t tmp; | |
3879 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3880 | ||
3881 | tmp = cputime_to_cputime64(cputime); | |
3882 | ||
457533a7 | 3883 | /* Add guest time to process. */ |
94886b84 | 3884 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3885 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3886 | account_group_user_time(p, cputime); |
94886b84 LV |
3887 | p->gtime = cputime_add(p->gtime, cputime); |
3888 | ||
457533a7 | 3889 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
3890 | if (TASK_NICE(p) > 0) { |
3891 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3892 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
3893 | } else { | |
3894 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3895 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3896 | } | |
94886b84 LV |
3897 | } |
3898 | ||
70a89a66 VP |
3899 | /* |
3900 | * Account system cpu time to a process and desired cpustat field | |
3901 | * @p: the process that the cpu time gets accounted to | |
3902 | * @cputime: the cpu time spent in kernel space since the last update | |
3903 | * @cputime_scaled: cputime scaled by cpu frequency | |
3904 | * @target_cputime64: pointer to cpustat field that has to be updated | |
3905 | */ | |
3906 | static inline | |
3907 | void __account_system_time(struct task_struct *p, cputime_t cputime, | |
3908 | cputime_t cputime_scaled, cputime64_t *target_cputime64) | |
3909 | { | |
3910 | cputime64_t tmp = cputime_to_cputime64(cputime); | |
3911 | ||
3912 | /* Add system time to process. */ | |
3913 | p->stime = cputime_add(p->stime, cputime); | |
3914 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); | |
3915 | account_group_system_time(p, cputime); | |
3916 | ||
3917 | /* Add system time to cpustat. */ | |
3918 | *target_cputime64 = cputime64_add(*target_cputime64, tmp); | |
3919 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); | |
3920 | ||
3921 | /* Account for system time used */ | |
3922 | acct_update_integrals(p); | |
3923 | } | |
3924 | ||
1da177e4 LT |
3925 | /* |
3926 | * Account system cpu time to a process. | |
3927 | * @p: the process that the cpu time gets accounted to | |
3928 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3929 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 3930 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
3931 | */ |
3932 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 3933 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
3934 | { |
3935 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70a89a66 | 3936 | cputime64_t *target_cputime64; |
1da177e4 | 3937 | |
983ed7a6 | 3938 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 3939 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
3940 | return; |
3941 | } | |
94886b84 | 3942 | |
1da177e4 | 3943 | if (hardirq_count() - hardirq_offset) |
70a89a66 | 3944 | target_cputime64 = &cpustat->irq; |
75e1056f | 3945 | else if (in_serving_softirq()) |
70a89a66 | 3946 | target_cputime64 = &cpustat->softirq; |
1da177e4 | 3947 | else |
70a89a66 | 3948 | target_cputime64 = &cpustat->system; |
ef12fefa | 3949 | |
70a89a66 | 3950 | __account_system_time(p, cputime, cputime_scaled, target_cputime64); |
1da177e4 LT |
3951 | } |
3952 | ||
c66f08be | 3953 | /* |
1da177e4 | 3954 | * Account for involuntary wait time. |
544b4a1f | 3955 | * @cputime: the cpu time spent in involuntary wait |
c66f08be | 3956 | */ |
79741dd3 | 3957 | void account_steal_time(cputime_t cputime) |
c66f08be | 3958 | { |
79741dd3 MS |
3959 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
3960 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
3961 | ||
3962 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
3963 | } |
3964 | ||
1da177e4 | 3965 | /* |
79741dd3 MS |
3966 | * Account for idle time. |
3967 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 3968 | */ |
79741dd3 | 3969 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
3970 | { |
3971 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 3972 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 3973 | struct rq *rq = this_rq(); |
1da177e4 | 3974 | |
79741dd3 MS |
3975 | if (atomic_read(&rq->nr_iowait) > 0) |
3976 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
3977 | else | |
3978 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
3979 | } |
3980 | ||
e6e6685a GC |
3981 | static __always_inline bool steal_account_process_tick(void) |
3982 | { | |
3983 | #ifdef CONFIG_PARAVIRT | |
3984 | if (static_branch(¶virt_steal_enabled)) { | |
3985 | u64 steal, st = 0; | |
3986 | ||
3987 | steal = paravirt_steal_clock(smp_processor_id()); | |
3988 | steal -= this_rq()->prev_steal_time; | |
3989 | ||
3990 | st = steal_ticks(steal); | |
3991 | this_rq()->prev_steal_time += st * TICK_NSEC; | |
3992 | ||
3993 | account_steal_time(st); | |
3994 | return st; | |
3995 | } | |
3996 | #endif | |
3997 | return false; | |
3998 | } | |
3999 | ||
79741dd3 MS |
4000 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
4001 | ||
abb74cef VP |
4002 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
4003 | /* | |
4004 | * Account a tick to a process and cpustat | |
4005 | * @p: the process that the cpu time gets accounted to | |
4006 | * @user_tick: is the tick from userspace | |
4007 | * @rq: the pointer to rq | |
4008 | * | |
4009 | * Tick demultiplexing follows the order | |
4010 | * - pending hardirq update | |
4011 | * - pending softirq update | |
4012 | * - user_time | |
4013 | * - idle_time | |
4014 | * - system time | |
4015 | * - check for guest_time | |
4016 | * - else account as system_time | |
4017 | * | |
4018 | * Check for hardirq is done both for system and user time as there is | |
4019 | * no timer going off while we are on hardirq and hence we may never get an | |
4020 | * opportunity to update it solely in system time. | |
4021 | * p->stime and friends are only updated on system time and not on irq | |
4022 | * softirq as those do not count in task exec_runtime any more. | |
4023 | */ | |
4024 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
4025 | struct rq *rq) | |
4026 | { | |
4027 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); | |
4028 | cputime64_t tmp = cputime_to_cputime64(cputime_one_jiffy); | |
4029 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4030 | ||
e6e6685a GC |
4031 | if (steal_account_process_tick()) |
4032 | return; | |
4033 | ||
abb74cef VP |
4034 | if (irqtime_account_hi_update()) { |
4035 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4036 | } else if (irqtime_account_si_update()) { | |
4037 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
414bee9b VP |
4038 | } else if (this_cpu_ksoftirqd() == p) { |
4039 | /* | |
4040 | * ksoftirqd time do not get accounted in cpu_softirq_time. | |
4041 | * So, we have to handle it separately here. | |
4042 | * Also, p->stime needs to be updated for ksoftirqd. | |
4043 | */ | |
4044 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
4045 | &cpustat->softirq); | |
abb74cef VP |
4046 | } else if (user_tick) { |
4047 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
4048 | } else if (p == rq->idle) { | |
4049 | account_idle_time(cputime_one_jiffy); | |
4050 | } else if (p->flags & PF_VCPU) { /* System time or guest time */ | |
4051 | account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
4052 | } else { | |
4053 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
4054 | &cpustat->system); | |
4055 | } | |
4056 | } | |
4057 | ||
4058 | static void irqtime_account_idle_ticks(int ticks) | |
4059 | { | |
4060 | int i; | |
4061 | struct rq *rq = this_rq(); | |
4062 | ||
4063 | for (i = 0; i < ticks; i++) | |
4064 | irqtime_account_process_tick(current, 0, rq); | |
4065 | } | |
544b4a1f | 4066 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
abb74cef VP |
4067 | static void irqtime_account_idle_ticks(int ticks) {} |
4068 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
4069 | struct rq *rq) {} | |
544b4a1f | 4070 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
79741dd3 MS |
4071 | |
4072 | /* | |
4073 | * Account a single tick of cpu time. | |
4074 | * @p: the process that the cpu time gets accounted to | |
4075 | * @user_tick: indicates if the tick is a user or a system tick | |
4076 | */ | |
4077 | void account_process_tick(struct task_struct *p, int user_tick) | |
4078 | { | |
a42548a1 | 4079 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
4080 | struct rq *rq = this_rq(); |
4081 | ||
abb74cef VP |
4082 | if (sched_clock_irqtime) { |
4083 | irqtime_account_process_tick(p, user_tick, rq); | |
4084 | return; | |
4085 | } | |
4086 | ||
e6e6685a GC |
4087 | if (steal_account_process_tick()) |
4088 | return; | |
4089 | ||
79741dd3 | 4090 | if (user_tick) |
a42548a1 | 4091 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 4092 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 4093 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
4094 | one_jiffy_scaled); |
4095 | else | |
a42548a1 | 4096 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
4097 | } |
4098 | ||
4099 | /* | |
4100 | * Account multiple ticks of steal time. | |
4101 | * @p: the process from which the cpu time has been stolen | |
4102 | * @ticks: number of stolen ticks | |
4103 | */ | |
4104 | void account_steal_ticks(unsigned long ticks) | |
4105 | { | |
4106 | account_steal_time(jiffies_to_cputime(ticks)); | |
4107 | } | |
4108 | ||
4109 | /* | |
4110 | * Account multiple ticks of idle time. | |
4111 | * @ticks: number of stolen ticks | |
4112 | */ | |
4113 | void account_idle_ticks(unsigned long ticks) | |
4114 | { | |
abb74cef VP |
4115 | |
4116 | if (sched_clock_irqtime) { | |
4117 | irqtime_account_idle_ticks(ticks); | |
4118 | return; | |
4119 | } | |
4120 | ||
79741dd3 | 4121 | account_idle_time(jiffies_to_cputime(ticks)); |
1da177e4 LT |
4122 | } |
4123 | ||
79741dd3 MS |
4124 | #endif |
4125 | ||
49048622 BS |
4126 | /* |
4127 | * Use precise platform statistics if available: | |
4128 | */ | |
4129 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 4130 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 4131 | { |
d99ca3b9 HS |
4132 | *ut = p->utime; |
4133 | *st = p->stime; | |
49048622 BS |
4134 | } |
4135 | ||
0cf55e1e | 4136 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 4137 | { |
0cf55e1e HS |
4138 | struct task_cputime cputime; |
4139 | ||
4140 | thread_group_cputime(p, &cputime); | |
4141 | ||
4142 | *ut = cputime.utime; | |
4143 | *st = cputime.stime; | |
49048622 BS |
4144 | } |
4145 | #else | |
761b1d26 HS |
4146 | |
4147 | #ifndef nsecs_to_cputime | |
b7b20df9 | 4148 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
4149 | #endif |
4150 | ||
d180c5bc | 4151 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 4152 | { |
d99ca3b9 | 4153 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
4154 | |
4155 | /* | |
4156 | * Use CFS's precise accounting: | |
4157 | */ | |
d180c5bc | 4158 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
4159 | |
4160 | if (total) { | |
e75e863d | 4161 | u64 temp = rtime; |
d180c5bc | 4162 | |
e75e863d | 4163 | temp *= utime; |
49048622 | 4164 | do_div(temp, total); |
d180c5bc HS |
4165 | utime = (cputime_t)temp; |
4166 | } else | |
4167 | utime = rtime; | |
49048622 | 4168 | |
d180c5bc HS |
4169 | /* |
4170 | * Compare with previous values, to keep monotonicity: | |
4171 | */ | |
761b1d26 | 4172 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 4173 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 4174 | |
d99ca3b9 HS |
4175 | *ut = p->prev_utime; |
4176 | *st = p->prev_stime; | |
49048622 BS |
4177 | } |
4178 | ||
0cf55e1e HS |
4179 | /* |
4180 | * Must be called with siglock held. | |
4181 | */ | |
4182 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 4183 | { |
0cf55e1e HS |
4184 | struct signal_struct *sig = p->signal; |
4185 | struct task_cputime cputime; | |
4186 | cputime_t rtime, utime, total; | |
49048622 | 4187 | |
0cf55e1e | 4188 | thread_group_cputime(p, &cputime); |
49048622 | 4189 | |
0cf55e1e HS |
4190 | total = cputime_add(cputime.utime, cputime.stime); |
4191 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 4192 | |
0cf55e1e | 4193 | if (total) { |
e75e863d | 4194 | u64 temp = rtime; |
49048622 | 4195 | |
e75e863d | 4196 | temp *= cputime.utime; |
0cf55e1e HS |
4197 | do_div(temp, total); |
4198 | utime = (cputime_t)temp; | |
4199 | } else | |
4200 | utime = rtime; | |
4201 | ||
4202 | sig->prev_utime = max(sig->prev_utime, utime); | |
4203 | sig->prev_stime = max(sig->prev_stime, | |
4204 | cputime_sub(rtime, sig->prev_utime)); | |
4205 | ||
4206 | *ut = sig->prev_utime; | |
4207 | *st = sig->prev_stime; | |
49048622 | 4208 | } |
49048622 | 4209 | #endif |
49048622 | 4210 | |
7835b98b CL |
4211 | /* |
4212 | * This function gets called by the timer code, with HZ frequency. | |
4213 | * We call it with interrupts disabled. | |
7835b98b CL |
4214 | */ |
4215 | void scheduler_tick(void) | |
4216 | { | |
7835b98b CL |
4217 | int cpu = smp_processor_id(); |
4218 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4219 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4220 | |
4221 | sched_clock_tick(); | |
dd41f596 | 4222 | |
05fa785c | 4223 | raw_spin_lock(&rq->lock); |
3e51f33f | 4224 | update_rq_clock(rq); |
fdf3e95d | 4225 | update_cpu_load_active(rq); |
fa85ae24 | 4226 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 4227 | raw_spin_unlock(&rq->lock); |
7835b98b | 4228 | |
e9d2b064 | 4229 | perf_event_task_tick(); |
e220d2dc | 4230 | |
e418e1c2 | 4231 | #ifdef CONFIG_SMP |
dd41f596 IM |
4232 | rq->idle_at_tick = idle_cpu(cpu); |
4233 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4234 | #endif |
1da177e4 LT |
4235 | } |
4236 | ||
132380a0 | 4237 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
4238 | { |
4239 | if (in_lock_functions(addr)) { | |
4240 | addr = CALLER_ADDR2; | |
4241 | if (in_lock_functions(addr)) | |
4242 | addr = CALLER_ADDR3; | |
4243 | } | |
4244 | return addr; | |
4245 | } | |
1da177e4 | 4246 | |
7e49fcce SR |
4247 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
4248 | defined(CONFIG_PREEMPT_TRACER)) | |
4249 | ||
43627582 | 4250 | void __kprobes add_preempt_count(int val) |
1da177e4 | 4251 | { |
6cd8a4bb | 4252 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4253 | /* |
4254 | * Underflow? | |
4255 | */ | |
9a11b49a IM |
4256 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4257 | return; | |
6cd8a4bb | 4258 | #endif |
1da177e4 | 4259 | preempt_count() += val; |
6cd8a4bb | 4260 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4261 | /* |
4262 | * Spinlock count overflowing soon? | |
4263 | */ | |
33859f7f MOS |
4264 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4265 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
4266 | #endif |
4267 | if (preempt_count() == val) | |
4268 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4269 | } |
4270 | EXPORT_SYMBOL(add_preempt_count); | |
4271 | ||
43627582 | 4272 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 4273 | { |
6cd8a4bb | 4274 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4275 | /* |
4276 | * Underflow? | |
4277 | */ | |
01e3eb82 | 4278 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 4279 | return; |
1da177e4 LT |
4280 | /* |
4281 | * Is the spinlock portion underflowing? | |
4282 | */ | |
9a11b49a IM |
4283 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4284 | !(preempt_count() & PREEMPT_MASK))) | |
4285 | return; | |
6cd8a4bb | 4286 | #endif |
9a11b49a | 4287 | |
6cd8a4bb SR |
4288 | if (preempt_count() == val) |
4289 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4290 | preempt_count() -= val; |
4291 | } | |
4292 | EXPORT_SYMBOL(sub_preempt_count); | |
4293 | ||
4294 | #endif | |
4295 | ||
4296 | /* | |
dd41f596 | 4297 | * Print scheduling while atomic bug: |
1da177e4 | 4298 | */ |
dd41f596 | 4299 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4300 | { |
838225b4 SS |
4301 | struct pt_regs *regs = get_irq_regs(); |
4302 | ||
3df0fc5b PZ |
4303 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
4304 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 4305 | |
dd41f596 | 4306 | debug_show_held_locks(prev); |
e21f5b15 | 4307 | print_modules(); |
dd41f596 IM |
4308 | if (irqs_disabled()) |
4309 | print_irqtrace_events(prev); | |
838225b4 SS |
4310 | |
4311 | if (regs) | |
4312 | show_regs(regs); | |
4313 | else | |
4314 | dump_stack(); | |
dd41f596 | 4315 | } |
1da177e4 | 4316 | |
dd41f596 IM |
4317 | /* |
4318 | * Various schedule()-time debugging checks and statistics: | |
4319 | */ | |
4320 | static inline void schedule_debug(struct task_struct *prev) | |
4321 | { | |
1da177e4 | 4322 | /* |
41a2d6cf | 4323 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4324 | * schedule() atomically, we ignore that path for now. |
4325 | * Otherwise, whine if we are scheduling when we should not be. | |
4326 | */ | |
3f33a7ce | 4327 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4328 | __schedule_bug(prev); |
4329 | ||
1da177e4 LT |
4330 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4331 | ||
2d72376b | 4332 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
4333 | } |
4334 | ||
6cecd084 | 4335 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 4336 | { |
61eadef6 | 4337 | if (prev->on_rq || rq->skip_clock_update < 0) |
a64692a3 | 4338 | update_rq_clock(rq); |
6cecd084 | 4339 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
4340 | } |
4341 | ||
dd41f596 IM |
4342 | /* |
4343 | * Pick up the highest-prio task: | |
4344 | */ | |
4345 | static inline struct task_struct * | |
b67802ea | 4346 | pick_next_task(struct rq *rq) |
dd41f596 | 4347 | { |
5522d5d5 | 4348 | const struct sched_class *class; |
dd41f596 | 4349 | struct task_struct *p; |
1da177e4 LT |
4350 | |
4351 | /* | |
dd41f596 IM |
4352 | * Optimization: we know that if all tasks are in |
4353 | * the fair class we can call that function directly: | |
1da177e4 | 4354 | */ |
953bfcd1 | 4355 | if (likely(rq->nr_running == rq->cfs.h_nr_running)) { |
fb8d4724 | 4356 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4357 | if (likely(p)) |
4358 | return p; | |
1da177e4 LT |
4359 | } |
4360 | ||
34f971f6 | 4361 | for_each_class(class) { |
fb8d4724 | 4362 | p = class->pick_next_task(rq); |
dd41f596 IM |
4363 | if (p) |
4364 | return p; | |
dd41f596 | 4365 | } |
34f971f6 PZ |
4366 | |
4367 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 4368 | } |
1da177e4 | 4369 | |
dd41f596 | 4370 | /* |
c259e01a | 4371 | * __schedule() is the main scheduler function. |
dd41f596 | 4372 | */ |
c259e01a | 4373 | static void __sched __schedule(void) |
dd41f596 IM |
4374 | { |
4375 | struct task_struct *prev, *next; | |
67ca7bde | 4376 | unsigned long *switch_count; |
dd41f596 | 4377 | struct rq *rq; |
31656519 | 4378 | int cpu; |
dd41f596 | 4379 | |
ff743345 PZ |
4380 | need_resched: |
4381 | preempt_disable(); | |
dd41f596 IM |
4382 | cpu = smp_processor_id(); |
4383 | rq = cpu_rq(cpu); | |
25502a6c | 4384 | rcu_note_context_switch(cpu); |
dd41f596 | 4385 | prev = rq->curr; |
dd41f596 | 4386 | |
dd41f596 | 4387 | schedule_debug(prev); |
1da177e4 | 4388 | |
31656519 | 4389 | if (sched_feat(HRTICK)) |
f333fdc9 | 4390 | hrtick_clear(rq); |
8f4d37ec | 4391 | |
05fa785c | 4392 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 4393 | |
246d86b5 | 4394 | switch_count = &prev->nivcsw; |
1da177e4 | 4395 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 4396 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 4397 | prev->state = TASK_RUNNING; |
21aa9af0 | 4398 | } else { |
2acca55e PZ |
4399 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
4400 | prev->on_rq = 0; | |
4401 | ||
21aa9af0 | 4402 | /* |
2acca55e PZ |
4403 | * If a worker went to sleep, notify and ask workqueue |
4404 | * whether it wants to wake up a task to maintain | |
4405 | * concurrency. | |
21aa9af0 TH |
4406 | */ |
4407 | if (prev->flags & PF_WQ_WORKER) { | |
4408 | struct task_struct *to_wakeup; | |
4409 | ||
4410 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
4411 | if (to_wakeup) | |
4412 | try_to_wake_up_local(to_wakeup); | |
4413 | } | |
21aa9af0 | 4414 | } |
dd41f596 | 4415 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4416 | } |
4417 | ||
3f029d3c | 4418 | pre_schedule(rq, prev); |
f65eda4f | 4419 | |
dd41f596 | 4420 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4421 | idle_balance(cpu, rq); |
1da177e4 | 4422 | |
df1c99d4 | 4423 | put_prev_task(rq, prev); |
b67802ea | 4424 | next = pick_next_task(rq); |
f26f9aff MG |
4425 | clear_tsk_need_resched(prev); |
4426 | rq->skip_clock_update = 0; | |
1da177e4 | 4427 | |
1da177e4 | 4428 | if (likely(prev != next)) { |
1da177e4 LT |
4429 | rq->nr_switches++; |
4430 | rq->curr = next; | |
4431 | ++*switch_count; | |
4432 | ||
dd41f596 | 4433 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 4434 | /* |
246d86b5 ON |
4435 | * The context switch have flipped the stack from under us |
4436 | * and restored the local variables which were saved when | |
4437 | * this task called schedule() in the past. prev == current | |
4438 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
4439 | */ |
4440 | cpu = smp_processor_id(); | |
4441 | rq = cpu_rq(cpu); | |
1da177e4 | 4442 | } else |
05fa785c | 4443 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 4444 | |
3f029d3c | 4445 | post_schedule(rq); |
1da177e4 | 4446 | |
1da177e4 | 4447 | preempt_enable_no_resched(); |
ff743345 | 4448 | if (need_resched()) |
1da177e4 LT |
4449 | goto need_resched; |
4450 | } | |
c259e01a | 4451 | |
9c40cef2 TG |
4452 | static inline void sched_submit_work(struct task_struct *tsk) |
4453 | { | |
4454 | if (!tsk->state) | |
4455 | return; | |
4456 | /* | |
4457 | * If we are going to sleep and we have plugged IO queued, | |
4458 | * make sure to submit it to avoid deadlocks. | |
4459 | */ | |
4460 | if (blk_needs_flush_plug(tsk)) | |
4461 | blk_schedule_flush_plug(tsk); | |
4462 | } | |
4463 | ||
6ebbe7a0 | 4464 | asmlinkage void __sched schedule(void) |
c259e01a | 4465 | { |
9c40cef2 TG |
4466 | struct task_struct *tsk = current; |
4467 | ||
4468 | sched_submit_work(tsk); | |
c259e01a TG |
4469 | __schedule(); |
4470 | } | |
1da177e4 LT |
4471 | EXPORT_SYMBOL(schedule); |
4472 | ||
c08f7829 | 4473 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d | 4474 | |
c6eb3dda PZ |
4475 | static inline bool owner_running(struct mutex *lock, struct task_struct *owner) |
4476 | { | |
c6eb3dda | 4477 | if (lock->owner != owner) |
307bf980 | 4478 | return false; |
0d66bf6d PZ |
4479 | |
4480 | /* | |
c6eb3dda PZ |
4481 | * Ensure we emit the owner->on_cpu, dereference _after_ checking |
4482 | * lock->owner still matches owner, if that fails, owner might | |
4483 | * point to free()d memory, if it still matches, the rcu_read_lock() | |
4484 | * ensures the memory stays valid. | |
0d66bf6d | 4485 | */ |
c6eb3dda | 4486 | barrier(); |
0d66bf6d | 4487 | |
307bf980 | 4488 | return owner->on_cpu; |
c6eb3dda | 4489 | } |
0d66bf6d | 4490 | |
c6eb3dda PZ |
4491 | /* |
4492 | * Look out! "owner" is an entirely speculative pointer | |
4493 | * access and not reliable. | |
4494 | */ | |
4495 | int mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner) | |
4496 | { | |
4497 | if (!sched_feat(OWNER_SPIN)) | |
4498 | return 0; | |
0d66bf6d | 4499 | |
307bf980 | 4500 | rcu_read_lock(); |
c6eb3dda PZ |
4501 | while (owner_running(lock, owner)) { |
4502 | if (need_resched()) | |
307bf980 | 4503 | break; |
0d66bf6d | 4504 | |
335d7afb | 4505 | arch_mutex_cpu_relax(); |
0d66bf6d | 4506 | } |
307bf980 | 4507 | rcu_read_unlock(); |
4b402210 | 4508 | |
c6eb3dda | 4509 | /* |
307bf980 TG |
4510 | * We break out the loop above on need_resched() and when the |
4511 | * owner changed, which is a sign for heavy contention. Return | |
4512 | * success only when lock->owner is NULL. | |
c6eb3dda | 4513 | */ |
307bf980 | 4514 | return lock->owner == NULL; |
0d66bf6d PZ |
4515 | } |
4516 | #endif | |
4517 | ||
1da177e4 LT |
4518 | #ifdef CONFIG_PREEMPT |
4519 | /* | |
2ed6e34f | 4520 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4521 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4522 | * occur there and call schedule directly. |
4523 | */ | |
d1f74e20 | 4524 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
4525 | { |
4526 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4527 | |
1da177e4 LT |
4528 | /* |
4529 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4530 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4531 | */ |
beed33a8 | 4532 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4533 | return; |
4534 | ||
3a5c359a | 4535 | do { |
d1f74e20 | 4536 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
c259e01a | 4537 | __schedule(); |
d1f74e20 | 4538 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 4539 | |
3a5c359a AK |
4540 | /* |
4541 | * Check again in case we missed a preemption opportunity | |
4542 | * between schedule and now. | |
4543 | */ | |
4544 | barrier(); | |
5ed0cec0 | 4545 | } while (need_resched()); |
1da177e4 | 4546 | } |
1da177e4 LT |
4547 | EXPORT_SYMBOL(preempt_schedule); |
4548 | ||
4549 | /* | |
2ed6e34f | 4550 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4551 | * off of irq context. |
4552 | * Note, that this is called and return with irqs disabled. This will | |
4553 | * protect us against recursive calling from irq. | |
4554 | */ | |
4555 | asmlinkage void __sched preempt_schedule_irq(void) | |
4556 | { | |
4557 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4558 | |
2ed6e34f | 4559 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4560 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4561 | ||
3a5c359a AK |
4562 | do { |
4563 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 4564 | local_irq_enable(); |
c259e01a | 4565 | __schedule(); |
3a5c359a | 4566 | local_irq_disable(); |
3a5c359a | 4567 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4568 | |
3a5c359a AK |
4569 | /* |
4570 | * Check again in case we missed a preemption opportunity | |
4571 | * between schedule and now. | |
4572 | */ | |
4573 | barrier(); | |
5ed0cec0 | 4574 | } while (need_resched()); |
1da177e4 LT |
4575 | } |
4576 | ||
4577 | #endif /* CONFIG_PREEMPT */ | |
4578 | ||
63859d4f | 4579 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 4580 | void *key) |
1da177e4 | 4581 | { |
63859d4f | 4582 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 4583 | } |
1da177e4 LT |
4584 | EXPORT_SYMBOL(default_wake_function); |
4585 | ||
4586 | /* | |
41a2d6cf IM |
4587 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4588 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4589 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4590 | * | |
4591 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4592 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4593 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4594 | */ | |
78ddb08f | 4595 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 4596 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 4597 | { |
2e45874c | 4598 | wait_queue_t *curr, *next; |
1da177e4 | 4599 | |
2e45874c | 4600 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4601 | unsigned flags = curr->flags; |
4602 | ||
63859d4f | 4603 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 4604 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4605 | break; |
4606 | } | |
4607 | } | |
4608 | ||
4609 | /** | |
4610 | * __wake_up - wake up threads blocked on a waitqueue. | |
4611 | * @q: the waitqueue | |
4612 | * @mode: which threads | |
4613 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4614 | * @key: is directly passed to the wakeup function |
50fa610a DH |
4615 | * |
4616 | * It may be assumed that this function implies a write memory barrier before | |
4617 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4618 | */ |
7ad5b3a5 | 4619 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4620 | int nr_exclusive, void *key) |
1da177e4 LT |
4621 | { |
4622 | unsigned long flags; | |
4623 | ||
4624 | spin_lock_irqsave(&q->lock, flags); | |
4625 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4626 | spin_unlock_irqrestore(&q->lock, flags); | |
4627 | } | |
1da177e4 LT |
4628 | EXPORT_SYMBOL(__wake_up); |
4629 | ||
4630 | /* | |
4631 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4632 | */ | |
7ad5b3a5 | 4633 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4634 | { |
4635 | __wake_up_common(q, mode, 1, 0, NULL); | |
4636 | } | |
22c43c81 | 4637 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 4638 | |
4ede816a DL |
4639 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
4640 | { | |
4641 | __wake_up_common(q, mode, 1, 0, key); | |
4642 | } | |
bf294b41 | 4643 | EXPORT_SYMBOL_GPL(__wake_up_locked_key); |
4ede816a | 4644 | |
1da177e4 | 4645 | /** |
4ede816a | 4646 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4647 | * @q: the waitqueue |
4648 | * @mode: which threads | |
4649 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 4650 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
4651 | * |
4652 | * The sync wakeup differs that the waker knows that it will schedule | |
4653 | * away soon, so while the target thread will be woken up, it will not | |
4654 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4655 | * with each other. This can prevent needless bouncing between CPUs. | |
4656 | * | |
4657 | * On UP it can prevent extra preemption. | |
50fa610a DH |
4658 | * |
4659 | * It may be assumed that this function implies a write memory barrier before | |
4660 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4661 | */ |
4ede816a DL |
4662 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
4663 | int nr_exclusive, void *key) | |
1da177e4 LT |
4664 | { |
4665 | unsigned long flags; | |
7d478721 | 4666 | int wake_flags = WF_SYNC; |
1da177e4 LT |
4667 | |
4668 | if (unlikely(!q)) | |
4669 | return; | |
4670 | ||
4671 | if (unlikely(!nr_exclusive)) | |
7d478721 | 4672 | wake_flags = 0; |
1da177e4 LT |
4673 | |
4674 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 4675 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
4676 | spin_unlock_irqrestore(&q->lock, flags); |
4677 | } | |
4ede816a DL |
4678 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
4679 | ||
4680 | /* | |
4681 | * __wake_up_sync - see __wake_up_sync_key() | |
4682 | */ | |
4683 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
4684 | { | |
4685 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
4686 | } | |
1da177e4 LT |
4687 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
4688 | ||
65eb3dc6 KD |
4689 | /** |
4690 | * complete: - signals a single thread waiting on this completion | |
4691 | * @x: holds the state of this particular completion | |
4692 | * | |
4693 | * This will wake up a single thread waiting on this completion. Threads will be | |
4694 | * awakened in the same order in which they were queued. | |
4695 | * | |
4696 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
4697 | * |
4698 | * It may be assumed that this function implies a write memory barrier before | |
4699 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4700 | */ |
b15136e9 | 4701 | void complete(struct completion *x) |
1da177e4 LT |
4702 | { |
4703 | unsigned long flags; | |
4704 | ||
4705 | spin_lock_irqsave(&x->wait.lock, flags); | |
4706 | x->done++; | |
d9514f6c | 4707 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4708 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4709 | } | |
4710 | EXPORT_SYMBOL(complete); | |
4711 | ||
65eb3dc6 KD |
4712 | /** |
4713 | * complete_all: - signals all threads waiting on this completion | |
4714 | * @x: holds the state of this particular completion | |
4715 | * | |
4716 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
4717 | * |
4718 | * It may be assumed that this function implies a write memory barrier before | |
4719 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4720 | */ |
b15136e9 | 4721 | void complete_all(struct completion *x) |
1da177e4 LT |
4722 | { |
4723 | unsigned long flags; | |
4724 | ||
4725 | spin_lock_irqsave(&x->wait.lock, flags); | |
4726 | x->done += UINT_MAX/2; | |
d9514f6c | 4727 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4728 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4729 | } | |
4730 | EXPORT_SYMBOL(complete_all); | |
4731 | ||
8cbbe86d AK |
4732 | static inline long __sched |
4733 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4734 | { |
1da177e4 LT |
4735 | if (!x->done) { |
4736 | DECLARE_WAITQUEUE(wait, current); | |
4737 | ||
a93d2f17 | 4738 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 4739 | do { |
94d3d824 | 4740 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
4741 | timeout = -ERESTARTSYS; |
4742 | break; | |
8cbbe86d AK |
4743 | } |
4744 | __set_current_state(state); | |
1da177e4 LT |
4745 | spin_unlock_irq(&x->wait.lock); |
4746 | timeout = schedule_timeout(timeout); | |
4747 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4748 | } while (!x->done && timeout); |
1da177e4 | 4749 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4750 | if (!x->done) |
4751 | return timeout; | |
1da177e4 LT |
4752 | } |
4753 | x->done--; | |
ea71a546 | 4754 | return timeout ?: 1; |
1da177e4 | 4755 | } |
1da177e4 | 4756 | |
8cbbe86d AK |
4757 | static long __sched |
4758 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4759 | { |
1da177e4 LT |
4760 | might_sleep(); |
4761 | ||
4762 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4763 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4764 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4765 | return timeout; |
4766 | } | |
1da177e4 | 4767 | |
65eb3dc6 KD |
4768 | /** |
4769 | * wait_for_completion: - waits for completion of a task | |
4770 | * @x: holds the state of this particular completion | |
4771 | * | |
4772 | * This waits to be signaled for completion of a specific task. It is NOT | |
4773 | * interruptible and there is no timeout. | |
4774 | * | |
4775 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4776 | * and interrupt capability. Also see complete(). | |
4777 | */ | |
b15136e9 | 4778 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4779 | { |
4780 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4781 | } |
8cbbe86d | 4782 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4783 | |
65eb3dc6 KD |
4784 | /** |
4785 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4786 | * @x: holds the state of this particular completion | |
4787 | * @timeout: timeout value in jiffies | |
4788 | * | |
4789 | * This waits for either a completion of a specific task to be signaled or for a | |
4790 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4791 | * interruptible. | |
4792 | */ | |
b15136e9 | 4793 | unsigned long __sched |
8cbbe86d | 4794 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4795 | { |
8cbbe86d | 4796 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4797 | } |
8cbbe86d | 4798 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4799 | |
65eb3dc6 KD |
4800 | /** |
4801 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4802 | * @x: holds the state of this particular completion | |
4803 | * | |
4804 | * This waits for completion of a specific task to be signaled. It is | |
4805 | * interruptible. | |
4806 | */ | |
8cbbe86d | 4807 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4808 | { |
51e97990 AK |
4809 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4810 | if (t == -ERESTARTSYS) | |
4811 | return t; | |
4812 | return 0; | |
0fec171c | 4813 | } |
8cbbe86d | 4814 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4815 | |
65eb3dc6 KD |
4816 | /** |
4817 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4818 | * @x: holds the state of this particular completion | |
4819 | * @timeout: timeout value in jiffies | |
4820 | * | |
4821 | * This waits for either a completion of a specific task to be signaled or for a | |
4822 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4823 | */ | |
6bf41237 | 4824 | long __sched |
8cbbe86d AK |
4825 | wait_for_completion_interruptible_timeout(struct completion *x, |
4826 | unsigned long timeout) | |
0fec171c | 4827 | { |
8cbbe86d | 4828 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4829 | } |
8cbbe86d | 4830 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4831 | |
65eb3dc6 KD |
4832 | /** |
4833 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4834 | * @x: holds the state of this particular completion | |
4835 | * | |
4836 | * This waits to be signaled for completion of a specific task. It can be | |
4837 | * interrupted by a kill signal. | |
4838 | */ | |
009e577e MW |
4839 | int __sched wait_for_completion_killable(struct completion *x) |
4840 | { | |
4841 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4842 | if (t == -ERESTARTSYS) | |
4843 | return t; | |
4844 | return 0; | |
4845 | } | |
4846 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4847 | ||
0aa12fb4 SW |
4848 | /** |
4849 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
4850 | * @x: holds the state of this particular completion | |
4851 | * @timeout: timeout value in jiffies | |
4852 | * | |
4853 | * This waits for either a completion of a specific task to be | |
4854 | * signaled or for a specified timeout to expire. It can be | |
4855 | * interrupted by a kill signal. The timeout is in jiffies. | |
4856 | */ | |
6bf41237 | 4857 | long __sched |
0aa12fb4 SW |
4858 | wait_for_completion_killable_timeout(struct completion *x, |
4859 | unsigned long timeout) | |
4860 | { | |
4861 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
4862 | } | |
4863 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
4864 | ||
be4de352 DC |
4865 | /** |
4866 | * try_wait_for_completion - try to decrement a completion without blocking | |
4867 | * @x: completion structure | |
4868 | * | |
4869 | * Returns: 0 if a decrement cannot be done without blocking | |
4870 | * 1 if a decrement succeeded. | |
4871 | * | |
4872 | * If a completion is being used as a counting completion, | |
4873 | * attempt to decrement the counter without blocking. This | |
4874 | * enables us to avoid waiting if the resource the completion | |
4875 | * is protecting is not available. | |
4876 | */ | |
4877 | bool try_wait_for_completion(struct completion *x) | |
4878 | { | |
7539a3b3 | 4879 | unsigned long flags; |
be4de352 DC |
4880 | int ret = 1; |
4881 | ||
7539a3b3 | 4882 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4883 | if (!x->done) |
4884 | ret = 0; | |
4885 | else | |
4886 | x->done--; | |
7539a3b3 | 4887 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4888 | return ret; |
4889 | } | |
4890 | EXPORT_SYMBOL(try_wait_for_completion); | |
4891 | ||
4892 | /** | |
4893 | * completion_done - Test to see if a completion has any waiters | |
4894 | * @x: completion structure | |
4895 | * | |
4896 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4897 | * 1 if there are no waiters. | |
4898 | * | |
4899 | */ | |
4900 | bool completion_done(struct completion *x) | |
4901 | { | |
7539a3b3 | 4902 | unsigned long flags; |
be4de352 DC |
4903 | int ret = 1; |
4904 | ||
7539a3b3 | 4905 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4906 | if (!x->done) |
4907 | ret = 0; | |
7539a3b3 | 4908 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4909 | return ret; |
4910 | } | |
4911 | EXPORT_SYMBOL(completion_done); | |
4912 | ||
8cbbe86d AK |
4913 | static long __sched |
4914 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4915 | { |
0fec171c IM |
4916 | unsigned long flags; |
4917 | wait_queue_t wait; | |
4918 | ||
4919 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4920 | |
8cbbe86d | 4921 | __set_current_state(state); |
1da177e4 | 4922 | |
8cbbe86d AK |
4923 | spin_lock_irqsave(&q->lock, flags); |
4924 | __add_wait_queue(q, &wait); | |
4925 | spin_unlock(&q->lock); | |
4926 | timeout = schedule_timeout(timeout); | |
4927 | spin_lock_irq(&q->lock); | |
4928 | __remove_wait_queue(q, &wait); | |
4929 | spin_unlock_irqrestore(&q->lock, flags); | |
4930 | ||
4931 | return timeout; | |
4932 | } | |
4933 | ||
4934 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4935 | { | |
4936 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4937 | } |
1da177e4 LT |
4938 | EXPORT_SYMBOL(interruptible_sleep_on); |
4939 | ||
0fec171c | 4940 | long __sched |
95cdf3b7 | 4941 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4942 | { |
8cbbe86d | 4943 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4944 | } |
1da177e4 LT |
4945 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4946 | ||
0fec171c | 4947 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4948 | { |
8cbbe86d | 4949 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4950 | } |
1da177e4 LT |
4951 | EXPORT_SYMBOL(sleep_on); |
4952 | ||
0fec171c | 4953 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4954 | { |
8cbbe86d | 4955 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4956 | } |
1da177e4 LT |
4957 | EXPORT_SYMBOL(sleep_on_timeout); |
4958 | ||
b29739f9 IM |
4959 | #ifdef CONFIG_RT_MUTEXES |
4960 | ||
4961 | /* | |
4962 | * rt_mutex_setprio - set the current priority of a task | |
4963 | * @p: task | |
4964 | * @prio: prio value (kernel-internal form) | |
4965 | * | |
4966 | * This function changes the 'effective' priority of a task. It does | |
4967 | * not touch ->normal_prio like __setscheduler(). | |
4968 | * | |
4969 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4970 | */ | |
36c8b586 | 4971 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 4972 | { |
83b699ed | 4973 | int oldprio, on_rq, running; |
70b97a7f | 4974 | struct rq *rq; |
83ab0aa0 | 4975 | const struct sched_class *prev_class; |
b29739f9 IM |
4976 | |
4977 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4978 | ||
0122ec5b | 4979 | rq = __task_rq_lock(p); |
b29739f9 | 4980 | |
a8027073 | 4981 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 4982 | oldprio = p->prio; |
83ab0aa0 | 4983 | prev_class = p->sched_class; |
fd2f4419 | 4984 | on_rq = p->on_rq; |
051a1d1a | 4985 | running = task_current(rq, p); |
0e1f3483 | 4986 | if (on_rq) |
69be72c1 | 4987 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4988 | if (running) |
4989 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4990 | |
4991 | if (rt_prio(prio)) | |
4992 | p->sched_class = &rt_sched_class; | |
4993 | else | |
4994 | p->sched_class = &fair_sched_class; | |
4995 | ||
b29739f9 IM |
4996 | p->prio = prio; |
4997 | ||
0e1f3483 HS |
4998 | if (running) |
4999 | p->sched_class->set_curr_task(rq); | |
da7a735e | 5000 | if (on_rq) |
371fd7e7 | 5001 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 | 5002 | |
da7a735e | 5003 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 5004 | __task_rq_unlock(rq); |
b29739f9 IM |
5005 | } |
5006 | ||
5007 | #endif | |
5008 | ||
36c8b586 | 5009 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 5010 | { |
dd41f596 | 5011 | int old_prio, delta, on_rq; |
1da177e4 | 5012 | unsigned long flags; |
70b97a7f | 5013 | struct rq *rq; |
1da177e4 LT |
5014 | |
5015 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
5016 | return; | |
5017 | /* | |
5018 | * We have to be careful, if called from sys_setpriority(), | |
5019 | * the task might be in the middle of scheduling on another CPU. | |
5020 | */ | |
5021 | rq = task_rq_lock(p, &flags); | |
5022 | /* | |
5023 | * The RT priorities are set via sched_setscheduler(), but we still | |
5024 | * allow the 'normal' nice value to be set - but as expected | |
5025 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 5026 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 5027 | */ |
e05606d3 | 5028 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
5029 | p->static_prio = NICE_TO_PRIO(nice); |
5030 | goto out_unlock; | |
5031 | } | |
fd2f4419 | 5032 | on_rq = p->on_rq; |
c09595f6 | 5033 | if (on_rq) |
69be72c1 | 5034 | dequeue_task(rq, p, 0); |
1da177e4 | 5035 | |
1da177e4 | 5036 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 5037 | set_load_weight(p); |
b29739f9 IM |
5038 | old_prio = p->prio; |
5039 | p->prio = effective_prio(p); | |
5040 | delta = p->prio - old_prio; | |
1da177e4 | 5041 | |
dd41f596 | 5042 | if (on_rq) { |
371fd7e7 | 5043 | enqueue_task(rq, p, 0); |
1da177e4 | 5044 | /* |
d5f9f942 AM |
5045 | * If the task increased its priority or is running and |
5046 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 5047 | */ |
d5f9f942 | 5048 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
5049 | resched_task(rq->curr); |
5050 | } | |
5051 | out_unlock: | |
0122ec5b | 5052 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 5053 | } |
1da177e4 LT |
5054 | EXPORT_SYMBOL(set_user_nice); |
5055 | ||
e43379f1 MM |
5056 | /* |
5057 | * can_nice - check if a task can reduce its nice value | |
5058 | * @p: task | |
5059 | * @nice: nice value | |
5060 | */ | |
36c8b586 | 5061 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 5062 | { |
024f4747 MM |
5063 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
5064 | int nice_rlim = 20 - nice; | |
48f24c4d | 5065 | |
78d7d407 | 5066 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
5067 | capable(CAP_SYS_NICE)); |
5068 | } | |
5069 | ||
1da177e4 LT |
5070 | #ifdef __ARCH_WANT_SYS_NICE |
5071 | ||
5072 | /* | |
5073 | * sys_nice - change the priority of the current process. | |
5074 | * @increment: priority increment | |
5075 | * | |
5076 | * sys_setpriority is a more generic, but much slower function that | |
5077 | * does similar things. | |
5078 | */ | |
5add95d4 | 5079 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 5080 | { |
48f24c4d | 5081 | long nice, retval; |
1da177e4 LT |
5082 | |
5083 | /* | |
5084 | * Setpriority might change our priority at the same moment. | |
5085 | * We don't have to worry. Conceptually one call occurs first | |
5086 | * and we have a single winner. | |
5087 | */ | |
e43379f1 MM |
5088 | if (increment < -40) |
5089 | increment = -40; | |
1da177e4 LT |
5090 | if (increment > 40) |
5091 | increment = 40; | |
5092 | ||
2b8f836f | 5093 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
5094 | if (nice < -20) |
5095 | nice = -20; | |
5096 | if (nice > 19) | |
5097 | nice = 19; | |
5098 | ||
e43379f1 MM |
5099 | if (increment < 0 && !can_nice(current, nice)) |
5100 | return -EPERM; | |
5101 | ||
1da177e4 LT |
5102 | retval = security_task_setnice(current, nice); |
5103 | if (retval) | |
5104 | return retval; | |
5105 | ||
5106 | set_user_nice(current, nice); | |
5107 | return 0; | |
5108 | } | |
5109 | ||
5110 | #endif | |
5111 | ||
5112 | /** | |
5113 | * task_prio - return the priority value of a given task. | |
5114 | * @p: the task in question. | |
5115 | * | |
5116 | * This is the priority value as seen by users in /proc. | |
5117 | * RT tasks are offset by -200. Normal tasks are centered | |
5118 | * around 0, value goes from -16 to +15. | |
5119 | */ | |
36c8b586 | 5120 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
5121 | { |
5122 | return p->prio - MAX_RT_PRIO; | |
5123 | } | |
5124 | ||
5125 | /** | |
5126 | * task_nice - return the nice value of a given task. | |
5127 | * @p: the task in question. | |
5128 | */ | |
36c8b586 | 5129 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
5130 | { |
5131 | return TASK_NICE(p); | |
5132 | } | |
150d8bed | 5133 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
5134 | |
5135 | /** | |
5136 | * idle_cpu - is a given cpu idle currently? | |
5137 | * @cpu: the processor in question. | |
5138 | */ | |
5139 | int idle_cpu(int cpu) | |
5140 | { | |
908a3283 TG |
5141 | struct rq *rq = cpu_rq(cpu); |
5142 | ||
5143 | if (rq->curr != rq->idle) | |
5144 | return 0; | |
5145 | ||
5146 | if (rq->nr_running) | |
5147 | return 0; | |
5148 | ||
5149 | #ifdef CONFIG_SMP | |
5150 | if (!llist_empty(&rq->wake_list)) | |
5151 | return 0; | |
5152 | #endif | |
5153 | ||
5154 | return 1; | |
1da177e4 LT |
5155 | } |
5156 | ||
1da177e4 LT |
5157 | /** |
5158 | * idle_task - return the idle task for a given cpu. | |
5159 | * @cpu: the processor in question. | |
5160 | */ | |
36c8b586 | 5161 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
5162 | { |
5163 | return cpu_rq(cpu)->idle; | |
5164 | } | |
5165 | ||
5166 | /** | |
5167 | * find_process_by_pid - find a process with a matching PID value. | |
5168 | * @pid: the pid in question. | |
5169 | */ | |
a9957449 | 5170 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 5171 | { |
228ebcbe | 5172 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
5173 | } |
5174 | ||
5175 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
5176 | static void |
5177 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 5178 | { |
1da177e4 LT |
5179 | p->policy = policy; |
5180 | p->rt_priority = prio; | |
b29739f9 IM |
5181 | p->normal_prio = normal_prio(p); |
5182 | /* we are holding p->pi_lock already */ | |
5183 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
5184 | if (rt_prio(p->prio)) |
5185 | p->sched_class = &rt_sched_class; | |
5186 | else | |
5187 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 5188 | set_load_weight(p); |
1da177e4 LT |
5189 | } |
5190 | ||
c69e8d9c DH |
5191 | /* |
5192 | * check the target process has a UID that matches the current process's | |
5193 | */ | |
5194 | static bool check_same_owner(struct task_struct *p) | |
5195 | { | |
5196 | const struct cred *cred = current_cred(), *pcred; | |
5197 | bool match; | |
5198 | ||
5199 | rcu_read_lock(); | |
5200 | pcred = __task_cred(p); | |
b0e77598 SH |
5201 | if (cred->user->user_ns == pcred->user->user_ns) |
5202 | match = (cred->euid == pcred->euid || | |
5203 | cred->euid == pcred->uid); | |
5204 | else | |
5205 | match = false; | |
c69e8d9c DH |
5206 | rcu_read_unlock(); |
5207 | return match; | |
5208 | } | |
5209 | ||
961ccddd | 5210 | static int __sched_setscheduler(struct task_struct *p, int policy, |
fe7de49f | 5211 | const struct sched_param *param, bool user) |
1da177e4 | 5212 | { |
83b699ed | 5213 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 5214 | unsigned long flags; |
83ab0aa0 | 5215 | const struct sched_class *prev_class; |
70b97a7f | 5216 | struct rq *rq; |
ca94c442 | 5217 | int reset_on_fork; |
1da177e4 | 5218 | |
66e5393a SR |
5219 | /* may grab non-irq protected spin_locks */ |
5220 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
5221 | recheck: |
5222 | /* double check policy once rq lock held */ | |
ca94c442 LP |
5223 | if (policy < 0) { |
5224 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 5225 | policy = oldpolicy = p->policy; |
ca94c442 LP |
5226 | } else { |
5227 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
5228 | policy &= ~SCHED_RESET_ON_FORK; | |
5229 | ||
5230 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
5231 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
5232 | policy != SCHED_IDLE) | |
5233 | return -EINVAL; | |
5234 | } | |
5235 | ||
1da177e4 LT |
5236 | /* |
5237 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
5238 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
5239 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
5240 | */ |
5241 | if (param->sched_priority < 0 || | |
95cdf3b7 | 5242 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 5243 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 5244 | return -EINVAL; |
e05606d3 | 5245 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
5246 | return -EINVAL; |
5247 | ||
37e4ab3f OC |
5248 | /* |
5249 | * Allow unprivileged RT tasks to decrease priority: | |
5250 | */ | |
961ccddd | 5251 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 5252 | if (rt_policy(policy)) { |
a44702e8 ON |
5253 | unsigned long rlim_rtprio = |
5254 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
5255 | |
5256 | /* can't set/change the rt policy */ | |
5257 | if (policy != p->policy && !rlim_rtprio) | |
5258 | return -EPERM; | |
5259 | ||
5260 | /* can't increase priority */ | |
5261 | if (param->sched_priority > p->rt_priority && | |
5262 | param->sched_priority > rlim_rtprio) | |
5263 | return -EPERM; | |
5264 | } | |
c02aa73b | 5265 | |
dd41f596 | 5266 | /* |
c02aa73b DH |
5267 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
5268 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 5269 | */ |
c02aa73b DH |
5270 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
5271 | if (!can_nice(p, TASK_NICE(p))) | |
5272 | return -EPERM; | |
5273 | } | |
5fe1d75f | 5274 | |
37e4ab3f | 5275 | /* can't change other user's priorities */ |
c69e8d9c | 5276 | if (!check_same_owner(p)) |
37e4ab3f | 5277 | return -EPERM; |
ca94c442 LP |
5278 | |
5279 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
5280 | if (p->sched_reset_on_fork && !reset_on_fork) | |
5281 | return -EPERM; | |
37e4ab3f | 5282 | } |
1da177e4 | 5283 | |
725aad24 | 5284 | if (user) { |
b0ae1981 | 5285 | retval = security_task_setscheduler(p); |
725aad24 JF |
5286 | if (retval) |
5287 | return retval; | |
5288 | } | |
5289 | ||
b29739f9 IM |
5290 | /* |
5291 | * make sure no PI-waiters arrive (or leave) while we are | |
5292 | * changing the priority of the task: | |
0122ec5b | 5293 | * |
25985edc | 5294 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
5295 | * runqueue lock must be held. |
5296 | */ | |
0122ec5b | 5297 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 5298 | |
34f971f6 PZ |
5299 | /* |
5300 | * Changing the policy of the stop threads its a very bad idea | |
5301 | */ | |
5302 | if (p == rq->stop) { | |
0122ec5b | 5303 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
5304 | return -EINVAL; |
5305 | } | |
5306 | ||
a51e9198 DF |
5307 | /* |
5308 | * If not changing anything there's no need to proceed further: | |
5309 | */ | |
5310 | if (unlikely(policy == p->policy && (!rt_policy(policy) || | |
5311 | param->sched_priority == p->rt_priority))) { | |
5312 | ||
5313 | __task_rq_unlock(rq); | |
5314 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
5315 | return 0; | |
5316 | } | |
5317 | ||
dc61b1d6 PZ |
5318 | #ifdef CONFIG_RT_GROUP_SCHED |
5319 | if (user) { | |
5320 | /* | |
5321 | * Do not allow realtime tasks into groups that have no runtime | |
5322 | * assigned. | |
5323 | */ | |
5324 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
5325 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
5326 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 5327 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
5328 | return -EPERM; |
5329 | } | |
5330 | } | |
5331 | #endif | |
5332 | ||
1da177e4 LT |
5333 | /* recheck policy now with rq lock held */ |
5334 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5335 | policy = oldpolicy = -1; | |
0122ec5b | 5336 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
5337 | goto recheck; |
5338 | } | |
fd2f4419 | 5339 | on_rq = p->on_rq; |
051a1d1a | 5340 | running = task_current(rq, p); |
0e1f3483 | 5341 | if (on_rq) |
2e1cb74a | 5342 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5343 | if (running) |
5344 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5345 | |
ca94c442 LP |
5346 | p->sched_reset_on_fork = reset_on_fork; |
5347 | ||
1da177e4 | 5348 | oldprio = p->prio; |
83ab0aa0 | 5349 | prev_class = p->sched_class; |
dd41f596 | 5350 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5351 | |
0e1f3483 HS |
5352 | if (running) |
5353 | p->sched_class->set_curr_task(rq); | |
da7a735e | 5354 | if (on_rq) |
dd41f596 | 5355 | activate_task(rq, p, 0); |
cb469845 | 5356 | |
da7a735e | 5357 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 5358 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 5359 | |
95e02ca9 TG |
5360 | rt_mutex_adjust_pi(p); |
5361 | ||
1da177e4 LT |
5362 | return 0; |
5363 | } | |
961ccddd RR |
5364 | |
5365 | /** | |
5366 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5367 | * @p: the task in question. | |
5368 | * @policy: new policy. | |
5369 | * @param: structure containing the new RT priority. | |
5370 | * | |
5371 | * NOTE that the task may be already dead. | |
5372 | */ | |
5373 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 5374 | const struct sched_param *param) |
961ccddd RR |
5375 | { |
5376 | return __sched_setscheduler(p, policy, param, true); | |
5377 | } | |
1da177e4 LT |
5378 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5379 | ||
961ccddd RR |
5380 | /** |
5381 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5382 | * @p: the task in question. | |
5383 | * @policy: new policy. | |
5384 | * @param: structure containing the new RT priority. | |
5385 | * | |
5386 | * Just like sched_setscheduler, only don't bother checking if the | |
5387 | * current context has permission. For example, this is needed in | |
5388 | * stop_machine(): we create temporary high priority worker threads, | |
5389 | * but our caller might not have that capability. | |
5390 | */ | |
5391 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 5392 | const struct sched_param *param) |
961ccddd RR |
5393 | { |
5394 | return __sched_setscheduler(p, policy, param, false); | |
5395 | } | |
5396 | ||
95cdf3b7 IM |
5397 | static int |
5398 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5399 | { |
1da177e4 LT |
5400 | struct sched_param lparam; |
5401 | struct task_struct *p; | |
36c8b586 | 5402 | int retval; |
1da177e4 LT |
5403 | |
5404 | if (!param || pid < 0) | |
5405 | return -EINVAL; | |
5406 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5407 | return -EFAULT; | |
5fe1d75f ON |
5408 | |
5409 | rcu_read_lock(); | |
5410 | retval = -ESRCH; | |
1da177e4 | 5411 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5412 | if (p != NULL) |
5413 | retval = sched_setscheduler(p, policy, &lparam); | |
5414 | rcu_read_unlock(); | |
36c8b586 | 5415 | |
1da177e4 LT |
5416 | return retval; |
5417 | } | |
5418 | ||
5419 | /** | |
5420 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5421 | * @pid: the pid in question. | |
5422 | * @policy: new policy. | |
5423 | * @param: structure containing the new RT priority. | |
5424 | */ | |
5add95d4 HC |
5425 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
5426 | struct sched_param __user *, param) | |
1da177e4 | 5427 | { |
c21761f1 JB |
5428 | /* negative values for policy are not valid */ |
5429 | if (policy < 0) | |
5430 | return -EINVAL; | |
5431 | ||
1da177e4 LT |
5432 | return do_sched_setscheduler(pid, policy, param); |
5433 | } | |
5434 | ||
5435 | /** | |
5436 | * sys_sched_setparam - set/change the RT priority of a thread | |
5437 | * @pid: the pid in question. | |
5438 | * @param: structure containing the new RT priority. | |
5439 | */ | |
5add95d4 | 5440 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5441 | { |
5442 | return do_sched_setscheduler(pid, -1, param); | |
5443 | } | |
5444 | ||
5445 | /** | |
5446 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5447 | * @pid: the pid in question. | |
5448 | */ | |
5add95d4 | 5449 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 5450 | { |
36c8b586 | 5451 | struct task_struct *p; |
3a5c359a | 5452 | int retval; |
1da177e4 LT |
5453 | |
5454 | if (pid < 0) | |
3a5c359a | 5455 | return -EINVAL; |
1da177e4 LT |
5456 | |
5457 | retval = -ESRCH; | |
5fe85be0 | 5458 | rcu_read_lock(); |
1da177e4 LT |
5459 | p = find_process_by_pid(pid); |
5460 | if (p) { | |
5461 | retval = security_task_getscheduler(p); | |
5462 | if (!retval) | |
ca94c442 LP |
5463 | retval = p->policy |
5464 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 5465 | } |
5fe85be0 | 5466 | rcu_read_unlock(); |
1da177e4 LT |
5467 | return retval; |
5468 | } | |
5469 | ||
5470 | /** | |
ca94c442 | 5471 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
5472 | * @pid: the pid in question. |
5473 | * @param: structure containing the RT priority. | |
5474 | */ | |
5add95d4 | 5475 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5476 | { |
5477 | struct sched_param lp; | |
36c8b586 | 5478 | struct task_struct *p; |
3a5c359a | 5479 | int retval; |
1da177e4 LT |
5480 | |
5481 | if (!param || pid < 0) | |
3a5c359a | 5482 | return -EINVAL; |
1da177e4 | 5483 | |
5fe85be0 | 5484 | rcu_read_lock(); |
1da177e4 LT |
5485 | p = find_process_by_pid(pid); |
5486 | retval = -ESRCH; | |
5487 | if (!p) | |
5488 | goto out_unlock; | |
5489 | ||
5490 | retval = security_task_getscheduler(p); | |
5491 | if (retval) | |
5492 | goto out_unlock; | |
5493 | ||
5494 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 5495 | rcu_read_unlock(); |
1da177e4 LT |
5496 | |
5497 | /* | |
5498 | * This one might sleep, we cannot do it with a spinlock held ... | |
5499 | */ | |
5500 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5501 | ||
1da177e4 LT |
5502 | return retval; |
5503 | ||
5504 | out_unlock: | |
5fe85be0 | 5505 | rcu_read_unlock(); |
1da177e4 LT |
5506 | return retval; |
5507 | } | |
5508 | ||
96f874e2 | 5509 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 5510 | { |
5a16f3d3 | 5511 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
5512 | struct task_struct *p; |
5513 | int retval; | |
1da177e4 | 5514 | |
95402b38 | 5515 | get_online_cpus(); |
23f5d142 | 5516 | rcu_read_lock(); |
1da177e4 LT |
5517 | |
5518 | p = find_process_by_pid(pid); | |
5519 | if (!p) { | |
23f5d142 | 5520 | rcu_read_unlock(); |
95402b38 | 5521 | put_online_cpus(); |
1da177e4 LT |
5522 | return -ESRCH; |
5523 | } | |
5524 | ||
23f5d142 | 5525 | /* Prevent p going away */ |
1da177e4 | 5526 | get_task_struct(p); |
23f5d142 | 5527 | rcu_read_unlock(); |
1da177e4 | 5528 | |
5a16f3d3 RR |
5529 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
5530 | retval = -ENOMEM; | |
5531 | goto out_put_task; | |
5532 | } | |
5533 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
5534 | retval = -ENOMEM; | |
5535 | goto out_free_cpus_allowed; | |
5536 | } | |
1da177e4 | 5537 | retval = -EPERM; |
b0e77598 | 5538 | if (!check_same_owner(p) && !task_ns_capable(p, CAP_SYS_NICE)) |
1da177e4 LT |
5539 | goto out_unlock; |
5540 | ||
b0ae1981 | 5541 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
5542 | if (retval) |
5543 | goto out_unlock; | |
5544 | ||
5a16f3d3 RR |
5545 | cpuset_cpus_allowed(p, cpus_allowed); |
5546 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 5547 | again: |
5a16f3d3 | 5548 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 5549 | |
8707d8b8 | 5550 | if (!retval) { |
5a16f3d3 RR |
5551 | cpuset_cpus_allowed(p, cpus_allowed); |
5552 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
5553 | /* |
5554 | * We must have raced with a concurrent cpuset | |
5555 | * update. Just reset the cpus_allowed to the | |
5556 | * cpuset's cpus_allowed | |
5557 | */ | |
5a16f3d3 | 5558 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
5559 | goto again; |
5560 | } | |
5561 | } | |
1da177e4 | 5562 | out_unlock: |
5a16f3d3 RR |
5563 | free_cpumask_var(new_mask); |
5564 | out_free_cpus_allowed: | |
5565 | free_cpumask_var(cpus_allowed); | |
5566 | out_put_task: | |
1da177e4 | 5567 | put_task_struct(p); |
95402b38 | 5568 | put_online_cpus(); |
1da177e4 LT |
5569 | return retval; |
5570 | } | |
5571 | ||
5572 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 5573 | struct cpumask *new_mask) |
1da177e4 | 5574 | { |
96f874e2 RR |
5575 | if (len < cpumask_size()) |
5576 | cpumask_clear(new_mask); | |
5577 | else if (len > cpumask_size()) | |
5578 | len = cpumask_size(); | |
5579 | ||
1da177e4 LT |
5580 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
5581 | } | |
5582 | ||
5583 | /** | |
5584 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5585 | * @pid: pid of the process | |
5586 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5587 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5588 | */ | |
5add95d4 HC |
5589 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
5590 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 5591 | { |
5a16f3d3 | 5592 | cpumask_var_t new_mask; |
1da177e4 LT |
5593 | int retval; |
5594 | ||
5a16f3d3 RR |
5595 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
5596 | return -ENOMEM; | |
1da177e4 | 5597 | |
5a16f3d3 RR |
5598 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
5599 | if (retval == 0) | |
5600 | retval = sched_setaffinity(pid, new_mask); | |
5601 | free_cpumask_var(new_mask); | |
5602 | return retval; | |
1da177e4 LT |
5603 | } |
5604 | ||
96f874e2 | 5605 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 5606 | { |
36c8b586 | 5607 | struct task_struct *p; |
31605683 | 5608 | unsigned long flags; |
1da177e4 | 5609 | int retval; |
1da177e4 | 5610 | |
95402b38 | 5611 | get_online_cpus(); |
23f5d142 | 5612 | rcu_read_lock(); |
1da177e4 LT |
5613 | |
5614 | retval = -ESRCH; | |
5615 | p = find_process_by_pid(pid); | |
5616 | if (!p) | |
5617 | goto out_unlock; | |
5618 | ||
e7834f8f DQ |
5619 | retval = security_task_getscheduler(p); |
5620 | if (retval) | |
5621 | goto out_unlock; | |
5622 | ||
013fdb80 | 5623 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
96f874e2 | 5624 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
013fdb80 | 5625 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
5626 | |
5627 | out_unlock: | |
23f5d142 | 5628 | rcu_read_unlock(); |
95402b38 | 5629 | put_online_cpus(); |
1da177e4 | 5630 | |
9531b62f | 5631 | return retval; |
1da177e4 LT |
5632 | } |
5633 | ||
5634 | /** | |
5635 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5636 | * @pid: pid of the process | |
5637 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5638 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5639 | */ | |
5add95d4 HC |
5640 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
5641 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
5642 | { |
5643 | int ret; | |
f17c8607 | 5644 | cpumask_var_t mask; |
1da177e4 | 5645 | |
84fba5ec | 5646 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
5647 | return -EINVAL; |
5648 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
5649 | return -EINVAL; |
5650 | ||
f17c8607 RR |
5651 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
5652 | return -ENOMEM; | |
1da177e4 | 5653 | |
f17c8607 RR |
5654 | ret = sched_getaffinity(pid, mask); |
5655 | if (ret == 0) { | |
8bc037fb | 5656 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
5657 | |
5658 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
5659 | ret = -EFAULT; |
5660 | else | |
cd3d8031 | 5661 | ret = retlen; |
f17c8607 RR |
5662 | } |
5663 | free_cpumask_var(mask); | |
1da177e4 | 5664 | |
f17c8607 | 5665 | return ret; |
1da177e4 LT |
5666 | } |
5667 | ||
5668 | /** | |
5669 | * sys_sched_yield - yield the current processor to other threads. | |
5670 | * | |
dd41f596 IM |
5671 | * This function yields the current CPU to other tasks. If there are no |
5672 | * other threads running on this CPU then this function will return. | |
1da177e4 | 5673 | */ |
5add95d4 | 5674 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 5675 | { |
70b97a7f | 5676 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5677 | |
2d72376b | 5678 | schedstat_inc(rq, yld_count); |
4530d7ab | 5679 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5680 | |
5681 | /* | |
5682 | * Since we are going to call schedule() anyway, there's | |
5683 | * no need to preempt or enable interrupts: | |
5684 | */ | |
5685 | __release(rq->lock); | |
8a25d5de | 5686 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 5687 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
5688 | preempt_enable_no_resched(); |
5689 | ||
5690 | schedule(); | |
5691 | ||
5692 | return 0; | |
5693 | } | |
5694 | ||
d86ee480 PZ |
5695 | static inline int should_resched(void) |
5696 | { | |
5697 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
5698 | } | |
5699 | ||
e7b38404 | 5700 | static void __cond_resched(void) |
1da177e4 | 5701 | { |
e7aaaa69 | 5702 | add_preempt_count(PREEMPT_ACTIVE); |
c259e01a | 5703 | __schedule(); |
e7aaaa69 | 5704 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 LT |
5705 | } |
5706 | ||
02b67cc3 | 5707 | int __sched _cond_resched(void) |
1da177e4 | 5708 | { |
d86ee480 | 5709 | if (should_resched()) { |
1da177e4 LT |
5710 | __cond_resched(); |
5711 | return 1; | |
5712 | } | |
5713 | return 0; | |
5714 | } | |
02b67cc3 | 5715 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5716 | |
5717 | /* | |
613afbf8 | 5718 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
5719 | * call schedule, and on return reacquire the lock. |
5720 | * | |
41a2d6cf | 5721 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5722 | * operations here to prevent schedule() from being called twice (once via |
5723 | * spin_unlock(), once by hand). | |
5724 | */ | |
613afbf8 | 5725 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5726 | { |
d86ee480 | 5727 | int resched = should_resched(); |
6df3cecb JK |
5728 | int ret = 0; |
5729 | ||
f607c668 PZ |
5730 | lockdep_assert_held(lock); |
5731 | ||
95c354fe | 5732 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5733 | spin_unlock(lock); |
d86ee480 | 5734 | if (resched) |
95c354fe NP |
5735 | __cond_resched(); |
5736 | else | |
5737 | cpu_relax(); | |
6df3cecb | 5738 | ret = 1; |
1da177e4 | 5739 | spin_lock(lock); |
1da177e4 | 5740 | } |
6df3cecb | 5741 | return ret; |
1da177e4 | 5742 | } |
613afbf8 | 5743 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 5744 | |
613afbf8 | 5745 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
5746 | { |
5747 | BUG_ON(!in_softirq()); | |
5748 | ||
d86ee480 | 5749 | if (should_resched()) { |
98d82567 | 5750 | local_bh_enable(); |
1da177e4 LT |
5751 | __cond_resched(); |
5752 | local_bh_disable(); | |
5753 | return 1; | |
5754 | } | |
5755 | return 0; | |
5756 | } | |
613afbf8 | 5757 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 5758 | |
1da177e4 LT |
5759 | /** |
5760 | * yield - yield the current processor to other threads. | |
5761 | * | |
72fd4a35 | 5762 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5763 | * thread runnable and calls sys_sched_yield(). |
5764 | */ | |
5765 | void __sched yield(void) | |
5766 | { | |
5767 | set_current_state(TASK_RUNNING); | |
5768 | sys_sched_yield(); | |
5769 | } | |
1da177e4 LT |
5770 | EXPORT_SYMBOL(yield); |
5771 | ||
d95f4122 MG |
5772 | /** |
5773 | * yield_to - yield the current processor to another thread in | |
5774 | * your thread group, or accelerate that thread toward the | |
5775 | * processor it's on. | |
16addf95 RD |
5776 | * @p: target task |
5777 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
5778 | * |
5779 | * It's the caller's job to ensure that the target task struct | |
5780 | * can't go away on us before we can do any checks. | |
5781 | * | |
5782 | * Returns true if we indeed boosted the target task. | |
5783 | */ | |
5784 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
5785 | { | |
5786 | struct task_struct *curr = current; | |
5787 | struct rq *rq, *p_rq; | |
5788 | unsigned long flags; | |
5789 | bool yielded = 0; | |
5790 | ||
5791 | local_irq_save(flags); | |
5792 | rq = this_rq(); | |
5793 | ||
5794 | again: | |
5795 | p_rq = task_rq(p); | |
5796 | double_rq_lock(rq, p_rq); | |
5797 | while (task_rq(p) != p_rq) { | |
5798 | double_rq_unlock(rq, p_rq); | |
5799 | goto again; | |
5800 | } | |
5801 | ||
5802 | if (!curr->sched_class->yield_to_task) | |
5803 | goto out; | |
5804 | ||
5805 | if (curr->sched_class != p->sched_class) | |
5806 | goto out; | |
5807 | ||
5808 | if (task_running(p_rq, p) || p->state) | |
5809 | goto out; | |
5810 | ||
5811 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 5812 | if (yielded) { |
d95f4122 | 5813 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
5814 | /* |
5815 | * Make p's CPU reschedule; pick_next_entity takes care of | |
5816 | * fairness. | |
5817 | */ | |
5818 | if (preempt && rq != p_rq) | |
5819 | resched_task(p_rq->curr); | |
5820 | } | |
d95f4122 MG |
5821 | |
5822 | out: | |
5823 | double_rq_unlock(rq, p_rq); | |
5824 | local_irq_restore(flags); | |
5825 | ||
5826 | if (yielded) | |
5827 | schedule(); | |
5828 | ||
5829 | return yielded; | |
5830 | } | |
5831 | EXPORT_SYMBOL_GPL(yield_to); | |
5832 | ||
1da177e4 | 5833 | /* |
41a2d6cf | 5834 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5835 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
5836 | */ |
5837 | void __sched io_schedule(void) | |
5838 | { | |
54d35f29 | 5839 | struct rq *rq = raw_rq(); |
1da177e4 | 5840 | |
0ff92245 | 5841 | delayacct_blkio_start(); |
1da177e4 | 5842 | atomic_inc(&rq->nr_iowait); |
73c10101 | 5843 | blk_flush_plug(current); |
8f0dfc34 | 5844 | current->in_iowait = 1; |
1da177e4 | 5845 | schedule(); |
8f0dfc34 | 5846 | current->in_iowait = 0; |
1da177e4 | 5847 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5848 | delayacct_blkio_end(); |
1da177e4 | 5849 | } |
1da177e4 LT |
5850 | EXPORT_SYMBOL(io_schedule); |
5851 | ||
5852 | long __sched io_schedule_timeout(long timeout) | |
5853 | { | |
54d35f29 | 5854 | struct rq *rq = raw_rq(); |
1da177e4 LT |
5855 | long ret; |
5856 | ||
0ff92245 | 5857 | delayacct_blkio_start(); |
1da177e4 | 5858 | atomic_inc(&rq->nr_iowait); |
73c10101 | 5859 | blk_flush_plug(current); |
8f0dfc34 | 5860 | current->in_iowait = 1; |
1da177e4 | 5861 | ret = schedule_timeout(timeout); |
8f0dfc34 | 5862 | current->in_iowait = 0; |
1da177e4 | 5863 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5864 | delayacct_blkio_end(); |
1da177e4 LT |
5865 | return ret; |
5866 | } | |
5867 | ||
5868 | /** | |
5869 | * sys_sched_get_priority_max - return maximum RT priority. | |
5870 | * @policy: scheduling class. | |
5871 | * | |
5872 | * this syscall returns the maximum rt_priority that can be used | |
5873 | * by a given scheduling class. | |
5874 | */ | |
5add95d4 | 5875 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5876 | { |
5877 | int ret = -EINVAL; | |
5878 | ||
5879 | switch (policy) { | |
5880 | case SCHED_FIFO: | |
5881 | case SCHED_RR: | |
5882 | ret = MAX_USER_RT_PRIO-1; | |
5883 | break; | |
5884 | case SCHED_NORMAL: | |
b0a9499c | 5885 | case SCHED_BATCH: |
dd41f596 | 5886 | case SCHED_IDLE: |
1da177e4 LT |
5887 | ret = 0; |
5888 | break; | |
5889 | } | |
5890 | return ret; | |
5891 | } | |
5892 | ||
5893 | /** | |
5894 | * sys_sched_get_priority_min - return minimum RT priority. | |
5895 | * @policy: scheduling class. | |
5896 | * | |
5897 | * this syscall returns the minimum rt_priority that can be used | |
5898 | * by a given scheduling class. | |
5899 | */ | |
5add95d4 | 5900 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5901 | { |
5902 | int ret = -EINVAL; | |
5903 | ||
5904 | switch (policy) { | |
5905 | case SCHED_FIFO: | |
5906 | case SCHED_RR: | |
5907 | ret = 1; | |
5908 | break; | |
5909 | case SCHED_NORMAL: | |
b0a9499c | 5910 | case SCHED_BATCH: |
dd41f596 | 5911 | case SCHED_IDLE: |
1da177e4 LT |
5912 | ret = 0; |
5913 | } | |
5914 | return ret; | |
5915 | } | |
5916 | ||
5917 | /** | |
5918 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5919 | * @pid: pid of the process. | |
5920 | * @interval: userspace pointer to the timeslice value. | |
5921 | * | |
5922 | * this syscall writes the default timeslice value of a given process | |
5923 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5924 | */ | |
17da2bd9 | 5925 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5926 | struct timespec __user *, interval) |
1da177e4 | 5927 | { |
36c8b586 | 5928 | struct task_struct *p; |
a4ec24b4 | 5929 | unsigned int time_slice; |
dba091b9 TG |
5930 | unsigned long flags; |
5931 | struct rq *rq; | |
3a5c359a | 5932 | int retval; |
1da177e4 | 5933 | struct timespec t; |
1da177e4 LT |
5934 | |
5935 | if (pid < 0) | |
3a5c359a | 5936 | return -EINVAL; |
1da177e4 LT |
5937 | |
5938 | retval = -ESRCH; | |
1a551ae7 | 5939 | rcu_read_lock(); |
1da177e4 LT |
5940 | p = find_process_by_pid(pid); |
5941 | if (!p) | |
5942 | goto out_unlock; | |
5943 | ||
5944 | retval = security_task_getscheduler(p); | |
5945 | if (retval) | |
5946 | goto out_unlock; | |
5947 | ||
dba091b9 TG |
5948 | rq = task_rq_lock(p, &flags); |
5949 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 5950 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 5951 | |
1a551ae7 | 5952 | rcu_read_unlock(); |
a4ec24b4 | 5953 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5954 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5955 | return retval; |
3a5c359a | 5956 | |
1da177e4 | 5957 | out_unlock: |
1a551ae7 | 5958 | rcu_read_unlock(); |
1da177e4 LT |
5959 | return retval; |
5960 | } | |
5961 | ||
7c731e0a | 5962 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5963 | |
82a1fcb9 | 5964 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5965 | { |
1da177e4 | 5966 | unsigned long free = 0; |
36c8b586 | 5967 | unsigned state; |
1da177e4 | 5968 | |
1da177e4 | 5969 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 5970 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 5971 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5972 | #if BITS_PER_LONG == 32 |
1da177e4 | 5973 | if (state == TASK_RUNNING) |
3df0fc5b | 5974 | printk(KERN_CONT " running "); |
1da177e4 | 5975 | else |
3df0fc5b | 5976 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5977 | #else |
5978 | if (state == TASK_RUNNING) | |
3df0fc5b | 5979 | printk(KERN_CONT " running task "); |
1da177e4 | 5980 | else |
3df0fc5b | 5981 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5982 | #endif |
5983 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5984 | free = stack_not_used(p); |
1da177e4 | 5985 | #endif |
3df0fc5b | 5986 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
5987 | task_pid_nr(p), task_pid_nr(p->real_parent), |
5988 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 5989 | |
5fb5e6de | 5990 | show_stack(p, NULL); |
1da177e4 LT |
5991 | } |
5992 | ||
e59e2ae2 | 5993 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5994 | { |
36c8b586 | 5995 | struct task_struct *g, *p; |
1da177e4 | 5996 | |
4bd77321 | 5997 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5998 | printk(KERN_INFO |
5999 | " task PC stack pid father\n"); | |
1da177e4 | 6000 | #else |
3df0fc5b PZ |
6001 | printk(KERN_INFO |
6002 | " task PC stack pid father\n"); | |
1da177e4 LT |
6003 | #endif |
6004 | read_lock(&tasklist_lock); | |
6005 | do_each_thread(g, p) { | |
6006 | /* | |
6007 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 6008 | * console might take a lot of time: |
1da177e4 LT |
6009 | */ |
6010 | touch_nmi_watchdog(); | |
39bc89fd | 6011 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 6012 | sched_show_task(p); |
1da177e4 LT |
6013 | } while_each_thread(g, p); |
6014 | ||
04c9167f JF |
6015 | touch_all_softlockup_watchdogs(); |
6016 | ||
dd41f596 IM |
6017 | #ifdef CONFIG_SCHED_DEBUG |
6018 | sysrq_sched_debug_show(); | |
6019 | #endif | |
1da177e4 | 6020 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
6021 | /* |
6022 | * Only show locks if all tasks are dumped: | |
6023 | */ | |
93335a21 | 6024 | if (!state_filter) |
e59e2ae2 | 6025 | debug_show_all_locks(); |
1da177e4 LT |
6026 | } |
6027 | ||
1df21055 IM |
6028 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
6029 | { | |
dd41f596 | 6030 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
6031 | } |
6032 | ||
f340c0d1 IM |
6033 | /** |
6034 | * init_idle - set up an idle thread for a given CPU | |
6035 | * @idle: task in question | |
6036 | * @cpu: cpu the idle task belongs to | |
6037 | * | |
6038 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
6039 | * flag, to make booting more robust. | |
6040 | */ | |
5c1e1767 | 6041 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 6042 | { |
70b97a7f | 6043 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
6044 | unsigned long flags; |
6045 | ||
05fa785c | 6046 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 6047 | |
dd41f596 | 6048 | __sched_fork(idle); |
06b83b5f | 6049 | idle->state = TASK_RUNNING; |
dd41f596 IM |
6050 | idle->se.exec_start = sched_clock(); |
6051 | ||
1e1b6c51 | 6052 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
6053 | /* |
6054 | * We're having a chicken and egg problem, even though we are | |
6055 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
6056 | * lockdep check in task_group() will fail. | |
6057 | * | |
6058 | * Similar case to sched_fork(). / Alternatively we could | |
6059 | * use task_rq_lock() here and obtain the other rq->lock. | |
6060 | * | |
6061 | * Silence PROVE_RCU | |
6062 | */ | |
6063 | rcu_read_lock(); | |
dd41f596 | 6064 | __set_task_cpu(idle, cpu); |
6506cf6c | 6065 | rcu_read_unlock(); |
1da177e4 | 6066 | |
1da177e4 | 6067 | rq->curr = rq->idle = idle; |
3ca7a440 PZ |
6068 | #if defined(CONFIG_SMP) |
6069 | idle->on_cpu = 1; | |
4866cde0 | 6070 | #endif |
05fa785c | 6071 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
6072 | |
6073 | /* Set the preempt count _outside_ the spinlocks! */ | |
a1261f54 | 6074 | task_thread_info(idle)->preempt_count = 0; |
625f2a37 | 6075 | |
dd41f596 IM |
6076 | /* |
6077 | * The idle tasks have their own, simple scheduling class: | |
6078 | */ | |
6079 | idle->sched_class = &idle_sched_class; | |
868baf07 | 6080 | ftrace_graph_init_idle_task(idle, cpu); |
1da177e4 LT |
6081 | } |
6082 | ||
6083 | /* | |
6084 | * In a system that switches off the HZ timer nohz_cpu_mask | |
6085 | * indicates which cpus entered this state. This is used | |
6086 | * in the rcu update to wait only for active cpus. For system | |
6087 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 6088 | * always be CPU_BITS_NONE. |
1da177e4 | 6089 | */ |
6a7b3dc3 | 6090 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 6091 | |
19978ca6 IM |
6092 | /* |
6093 | * Increase the granularity value when there are more CPUs, | |
6094 | * because with more CPUs the 'effective latency' as visible | |
6095 | * to users decreases. But the relationship is not linear, | |
6096 | * so pick a second-best guess by going with the log2 of the | |
6097 | * number of CPUs. | |
6098 | * | |
6099 | * This idea comes from the SD scheduler of Con Kolivas: | |
6100 | */ | |
acb4a848 | 6101 | static int get_update_sysctl_factor(void) |
19978ca6 | 6102 | { |
4ca3ef71 | 6103 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
6104 | unsigned int factor; |
6105 | ||
6106 | switch (sysctl_sched_tunable_scaling) { | |
6107 | case SCHED_TUNABLESCALING_NONE: | |
6108 | factor = 1; | |
6109 | break; | |
6110 | case SCHED_TUNABLESCALING_LINEAR: | |
6111 | factor = cpus; | |
6112 | break; | |
6113 | case SCHED_TUNABLESCALING_LOG: | |
6114 | default: | |
6115 | factor = 1 + ilog2(cpus); | |
6116 | break; | |
6117 | } | |
19978ca6 | 6118 | |
acb4a848 CE |
6119 | return factor; |
6120 | } | |
19978ca6 | 6121 | |
acb4a848 CE |
6122 | static void update_sysctl(void) |
6123 | { | |
6124 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 6125 | |
0bcdcf28 CE |
6126 | #define SET_SYSCTL(name) \ |
6127 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
6128 | SET_SYSCTL(sched_min_granularity); | |
6129 | SET_SYSCTL(sched_latency); | |
6130 | SET_SYSCTL(sched_wakeup_granularity); | |
0bcdcf28 CE |
6131 | #undef SET_SYSCTL |
6132 | } | |
55cd5340 | 6133 | |
0bcdcf28 CE |
6134 | static inline void sched_init_granularity(void) |
6135 | { | |
6136 | update_sysctl(); | |
19978ca6 IM |
6137 | } |
6138 | ||
1da177e4 | 6139 | #ifdef CONFIG_SMP |
1e1b6c51 KM |
6140 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
6141 | { | |
6142 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
6143 | p->sched_class->set_cpus_allowed(p, new_mask); | |
6144 | else { | |
6145 | cpumask_copy(&p->cpus_allowed, new_mask); | |
6146 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
6147 | } | |
6148 | } | |
6149 | ||
1da177e4 LT |
6150 | /* |
6151 | * This is how migration works: | |
6152 | * | |
969c7921 TH |
6153 | * 1) we invoke migration_cpu_stop() on the target CPU using |
6154 | * stop_one_cpu(). | |
6155 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
6156 | * off the CPU) | |
6157 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
6158 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 6159 | * it and puts it into the right queue. |
969c7921 TH |
6160 | * 5) stopper completes and stop_one_cpu() returns and the migration |
6161 | * is done. | |
1da177e4 LT |
6162 | */ |
6163 | ||
6164 | /* | |
6165 | * Change a given task's CPU affinity. Migrate the thread to a | |
6166 | * proper CPU and schedule it away if the CPU it's executing on | |
6167 | * is removed from the allowed bitmask. | |
6168 | * | |
6169 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 6170 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
6171 | * call is not atomic; no spinlocks may be held. |
6172 | */ | |
96f874e2 | 6173 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
6174 | { |
6175 | unsigned long flags; | |
70b97a7f | 6176 | struct rq *rq; |
969c7921 | 6177 | unsigned int dest_cpu; |
48f24c4d | 6178 | int ret = 0; |
1da177e4 LT |
6179 | |
6180 | rq = task_rq_lock(p, &flags); | |
e2912009 | 6181 | |
db44fc01 YZ |
6182 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
6183 | goto out; | |
6184 | ||
6ad4c188 | 6185 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
6186 | ret = -EINVAL; |
6187 | goto out; | |
6188 | } | |
6189 | ||
db44fc01 | 6190 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current)) { |
9985b0ba DR |
6191 | ret = -EINVAL; |
6192 | goto out; | |
6193 | } | |
6194 | ||
1e1b6c51 | 6195 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 6196 | |
1da177e4 | 6197 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 6198 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
6199 | goto out; |
6200 | ||
969c7921 | 6201 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
bd8e7dde | 6202 | if (p->on_rq) { |
969c7921 | 6203 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 6204 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 6205 | task_rq_unlock(rq, p, &flags); |
969c7921 | 6206 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
6207 | tlb_migrate_finish(p->mm); |
6208 | return 0; | |
6209 | } | |
6210 | out: | |
0122ec5b | 6211 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 6212 | |
1da177e4 LT |
6213 | return ret; |
6214 | } | |
cd8ba7cd | 6215 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
6216 | |
6217 | /* | |
41a2d6cf | 6218 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
6219 | * this because either it can't run here any more (set_cpus_allowed() |
6220 | * away from this CPU, or CPU going down), or because we're | |
6221 | * attempting to rebalance this task on exec (sched_exec). | |
6222 | * | |
6223 | * So we race with normal scheduler movements, but that's OK, as long | |
6224 | * as the task is no longer on this CPU. | |
efc30814 KK |
6225 | * |
6226 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 6227 | */ |
efc30814 | 6228 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 6229 | { |
70b97a7f | 6230 | struct rq *rq_dest, *rq_src; |
e2912009 | 6231 | int ret = 0; |
1da177e4 | 6232 | |
e761b772 | 6233 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 6234 | return ret; |
1da177e4 LT |
6235 | |
6236 | rq_src = cpu_rq(src_cpu); | |
6237 | rq_dest = cpu_rq(dest_cpu); | |
6238 | ||
0122ec5b | 6239 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
6240 | double_rq_lock(rq_src, rq_dest); |
6241 | /* Already moved. */ | |
6242 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 6243 | goto done; |
1da177e4 | 6244 | /* Affinity changed (again). */ |
96f874e2 | 6245 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 6246 | goto fail; |
1da177e4 | 6247 | |
e2912009 PZ |
6248 | /* |
6249 | * If we're not on a rq, the next wake-up will ensure we're | |
6250 | * placed properly. | |
6251 | */ | |
fd2f4419 | 6252 | if (p->on_rq) { |
2e1cb74a | 6253 | deactivate_task(rq_src, p, 0); |
e2912009 | 6254 | set_task_cpu(p, dest_cpu); |
dd41f596 | 6255 | activate_task(rq_dest, p, 0); |
15afe09b | 6256 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 6257 | } |
b1e38734 | 6258 | done: |
efc30814 | 6259 | ret = 1; |
b1e38734 | 6260 | fail: |
1da177e4 | 6261 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 6262 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 6263 | return ret; |
1da177e4 LT |
6264 | } |
6265 | ||
6266 | /* | |
969c7921 TH |
6267 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
6268 | * and performs thread migration by bumping thread off CPU then | |
6269 | * 'pushing' onto another runqueue. | |
1da177e4 | 6270 | */ |
969c7921 | 6271 | static int migration_cpu_stop(void *data) |
1da177e4 | 6272 | { |
969c7921 | 6273 | struct migration_arg *arg = data; |
f7b4cddc | 6274 | |
969c7921 TH |
6275 | /* |
6276 | * The original target cpu might have gone down and we might | |
6277 | * be on another cpu but it doesn't matter. | |
6278 | */ | |
f7b4cddc | 6279 | local_irq_disable(); |
969c7921 | 6280 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 6281 | local_irq_enable(); |
1da177e4 | 6282 | return 0; |
f7b4cddc ON |
6283 | } |
6284 | ||
1da177e4 | 6285 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 6286 | |
054b9108 | 6287 | /* |
48c5ccae PZ |
6288 | * Ensures that the idle task is using init_mm right before its cpu goes |
6289 | * offline. | |
054b9108 | 6290 | */ |
48c5ccae | 6291 | void idle_task_exit(void) |
1da177e4 | 6292 | { |
48c5ccae | 6293 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 6294 | |
48c5ccae | 6295 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 6296 | |
48c5ccae PZ |
6297 | if (mm != &init_mm) |
6298 | switch_mm(mm, &init_mm, current); | |
6299 | mmdrop(mm); | |
1da177e4 LT |
6300 | } |
6301 | ||
6302 | /* | |
6303 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
6304 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
6305 | * for performance reasons the counter is not stricly tracking tasks to | |
6306 | * their home CPUs. So we just add the counter to another CPU's counter, | |
6307 | * to keep the global sum constant after CPU-down: | |
6308 | */ | |
70b97a7f | 6309 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 6310 | { |
6ad4c188 | 6311 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 | 6312 | |
1da177e4 LT |
6313 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; |
6314 | rq_src->nr_uninterruptible = 0; | |
1da177e4 LT |
6315 | } |
6316 | ||
dd41f596 | 6317 | /* |
48c5ccae | 6318 | * remove the tasks which were accounted by rq from calc_load_tasks. |
1da177e4 | 6319 | */ |
48c5ccae | 6320 | static void calc_global_load_remove(struct rq *rq) |
1da177e4 | 6321 | { |
48c5ccae PZ |
6322 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); |
6323 | rq->calc_load_active = 0; | |
1da177e4 LT |
6324 | } |
6325 | ||
8cb120d3 PT |
6326 | #ifdef CONFIG_CFS_BANDWIDTH |
6327 | static void unthrottle_offline_cfs_rqs(struct rq *rq) | |
6328 | { | |
6329 | struct cfs_rq *cfs_rq; | |
6330 | ||
6331 | for_each_leaf_cfs_rq(rq, cfs_rq) { | |
6332 | struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg); | |
6333 | ||
6334 | if (!cfs_rq->runtime_enabled) | |
6335 | continue; | |
6336 | ||
6337 | /* | |
6338 | * clock_task is not advancing so we just need to make sure | |
6339 | * there's some valid quota amount | |
6340 | */ | |
6341 | cfs_rq->runtime_remaining = cfs_b->quota; | |
6342 | if (cfs_rq_throttled(cfs_rq)) | |
6343 | unthrottle_cfs_rq(cfs_rq); | |
6344 | } | |
6345 | } | |
6346 | #else | |
6347 | static void unthrottle_offline_cfs_rqs(struct rq *rq) {} | |
6348 | #endif | |
6349 | ||
48f24c4d | 6350 | /* |
48c5ccae PZ |
6351 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
6352 | * try_to_wake_up()->select_task_rq(). | |
6353 | * | |
6354 | * Called with rq->lock held even though we'er in stop_machine() and | |
6355 | * there's no concurrency possible, we hold the required locks anyway | |
6356 | * because of lock validation efforts. | |
1da177e4 | 6357 | */ |
48c5ccae | 6358 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 6359 | { |
70b97a7f | 6360 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
6361 | struct task_struct *next, *stop = rq->stop; |
6362 | int dest_cpu; | |
1da177e4 LT |
6363 | |
6364 | /* | |
48c5ccae PZ |
6365 | * Fudge the rq selection such that the below task selection loop |
6366 | * doesn't get stuck on the currently eligible stop task. | |
6367 | * | |
6368 | * We're currently inside stop_machine() and the rq is either stuck | |
6369 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
6370 | * either way we should never end up calling schedule() until we're | |
6371 | * done here. | |
1da177e4 | 6372 | */ |
48c5ccae | 6373 | rq->stop = NULL; |
48f24c4d | 6374 | |
8cb120d3 PT |
6375 | /* Ensure any throttled groups are reachable by pick_next_task */ |
6376 | unthrottle_offline_cfs_rqs(rq); | |
6377 | ||
dd41f596 | 6378 | for ( ; ; ) { |
48c5ccae PZ |
6379 | /* |
6380 | * There's this thread running, bail when that's the only | |
6381 | * remaining thread. | |
6382 | */ | |
6383 | if (rq->nr_running == 1) | |
dd41f596 | 6384 | break; |
48c5ccae | 6385 | |
b67802ea | 6386 | next = pick_next_task(rq); |
48c5ccae | 6387 | BUG_ON(!next); |
79c53799 | 6388 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 6389 | |
48c5ccae PZ |
6390 | /* Find suitable destination for @next, with force if needed. */ |
6391 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
6392 | raw_spin_unlock(&rq->lock); | |
6393 | ||
6394 | __migrate_task(next, dead_cpu, dest_cpu); | |
6395 | ||
6396 | raw_spin_lock(&rq->lock); | |
1da177e4 | 6397 | } |
dce48a84 | 6398 | |
48c5ccae | 6399 | rq->stop = stop; |
dce48a84 | 6400 | } |
48c5ccae | 6401 | |
1da177e4 LT |
6402 | #endif /* CONFIG_HOTPLUG_CPU */ |
6403 | ||
e692ab53 NP |
6404 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6405 | ||
6406 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6407 | { |
6408 | .procname = "sched_domain", | |
c57baf1e | 6409 | .mode = 0555, |
e0361851 | 6410 | }, |
56992309 | 6411 | {} |
e692ab53 NP |
6412 | }; |
6413 | ||
6414 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
6415 | { |
6416 | .procname = "kernel", | |
c57baf1e | 6417 | .mode = 0555, |
e0361851 AD |
6418 | .child = sd_ctl_dir, |
6419 | }, | |
56992309 | 6420 | {} |
e692ab53 NP |
6421 | }; |
6422 | ||
6423 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6424 | { | |
6425 | struct ctl_table *entry = | |
5cf9f062 | 6426 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6427 | |
e692ab53 NP |
6428 | return entry; |
6429 | } | |
6430 | ||
6382bc90 MM |
6431 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6432 | { | |
cd790076 | 6433 | struct ctl_table *entry; |
6382bc90 | 6434 | |
cd790076 MM |
6435 | /* |
6436 | * In the intermediate directories, both the child directory and | |
6437 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6438 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6439 | * static strings and all have proc handlers. |
6440 | */ | |
6441 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6442 | if (entry->child) |
6443 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6444 | if (entry->proc_handler == NULL) |
6445 | kfree(entry->procname); | |
6446 | } | |
6382bc90 MM |
6447 | |
6448 | kfree(*tablep); | |
6449 | *tablep = NULL; | |
6450 | } | |
6451 | ||
e692ab53 | 6452 | static void |
e0361851 | 6453 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6454 | const char *procname, void *data, int maxlen, |
6455 | mode_t mode, proc_handler *proc_handler) | |
6456 | { | |
e692ab53 NP |
6457 | entry->procname = procname; |
6458 | entry->data = data; | |
6459 | entry->maxlen = maxlen; | |
6460 | entry->mode = mode; | |
6461 | entry->proc_handler = proc_handler; | |
6462 | } | |
6463 | ||
6464 | static struct ctl_table * | |
6465 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6466 | { | |
a5d8c348 | 6467 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 6468 | |
ad1cdc1d MM |
6469 | if (table == NULL) |
6470 | return NULL; | |
6471 | ||
e0361851 | 6472 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6473 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6474 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6475 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6476 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6477 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6478 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6479 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6480 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6481 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6482 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6483 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6484 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6485 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6486 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6487 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6488 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6489 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6490 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6491 | &sd->cache_nice_tries, |
6492 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6493 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6494 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
6495 | set_table_entry(&table[11], "name", sd->name, |
6496 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
6497 | /* &table[12] is terminator */ | |
e692ab53 NP |
6498 | |
6499 | return table; | |
6500 | } | |
6501 | ||
9a4e7159 | 6502 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6503 | { |
6504 | struct ctl_table *entry, *table; | |
6505 | struct sched_domain *sd; | |
6506 | int domain_num = 0, i; | |
6507 | char buf[32]; | |
6508 | ||
6509 | for_each_domain(cpu, sd) | |
6510 | domain_num++; | |
6511 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6512 | if (table == NULL) |
6513 | return NULL; | |
e692ab53 NP |
6514 | |
6515 | i = 0; | |
6516 | for_each_domain(cpu, sd) { | |
6517 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6518 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6519 | entry->mode = 0555; |
e692ab53 NP |
6520 | entry->child = sd_alloc_ctl_domain_table(sd); |
6521 | entry++; | |
6522 | i++; | |
6523 | } | |
6524 | return table; | |
6525 | } | |
6526 | ||
6527 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6528 | static void register_sched_domain_sysctl(void) |
e692ab53 | 6529 | { |
6ad4c188 | 6530 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
6531 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
6532 | char buf[32]; | |
6533 | ||
7378547f MM |
6534 | WARN_ON(sd_ctl_dir[0].child); |
6535 | sd_ctl_dir[0].child = entry; | |
6536 | ||
ad1cdc1d MM |
6537 | if (entry == NULL) |
6538 | return; | |
6539 | ||
6ad4c188 | 6540 | for_each_possible_cpu(i) { |
e692ab53 | 6541 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6542 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6543 | entry->mode = 0555; |
e692ab53 | 6544 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6545 | entry++; |
e692ab53 | 6546 | } |
7378547f MM |
6547 | |
6548 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6549 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6550 | } | |
6382bc90 | 6551 | |
7378547f | 6552 | /* may be called multiple times per register */ |
6382bc90 MM |
6553 | static void unregister_sched_domain_sysctl(void) |
6554 | { | |
7378547f MM |
6555 | if (sd_sysctl_header) |
6556 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6557 | sd_sysctl_header = NULL; |
7378547f MM |
6558 | if (sd_ctl_dir[0].child) |
6559 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6560 | } |
e692ab53 | 6561 | #else |
6382bc90 MM |
6562 | static void register_sched_domain_sysctl(void) |
6563 | { | |
6564 | } | |
6565 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6566 | { |
6567 | } | |
6568 | #endif | |
6569 | ||
1f11eb6a GH |
6570 | static void set_rq_online(struct rq *rq) |
6571 | { | |
6572 | if (!rq->online) { | |
6573 | const struct sched_class *class; | |
6574 | ||
c6c4927b | 6575 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6576 | rq->online = 1; |
6577 | ||
6578 | for_each_class(class) { | |
6579 | if (class->rq_online) | |
6580 | class->rq_online(rq); | |
6581 | } | |
6582 | } | |
6583 | } | |
6584 | ||
6585 | static void set_rq_offline(struct rq *rq) | |
6586 | { | |
6587 | if (rq->online) { | |
6588 | const struct sched_class *class; | |
6589 | ||
6590 | for_each_class(class) { | |
6591 | if (class->rq_offline) | |
6592 | class->rq_offline(rq); | |
6593 | } | |
6594 | ||
c6c4927b | 6595 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6596 | rq->online = 0; |
6597 | } | |
6598 | } | |
6599 | ||
1da177e4 LT |
6600 | /* |
6601 | * migration_call - callback that gets triggered when a CPU is added. | |
6602 | * Here we can start up the necessary migration thread for the new CPU. | |
6603 | */ | |
48f24c4d IM |
6604 | static int __cpuinit |
6605 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6606 | { |
48f24c4d | 6607 | int cpu = (long)hcpu; |
1da177e4 | 6608 | unsigned long flags; |
969c7921 | 6609 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 6610 | |
48c5ccae | 6611 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 6612 | |
1da177e4 | 6613 | case CPU_UP_PREPARE: |
a468d389 | 6614 | rq->calc_load_update = calc_load_update; |
1da177e4 | 6615 | break; |
48f24c4d | 6616 | |
1da177e4 | 6617 | case CPU_ONLINE: |
1f94ef59 | 6618 | /* Update our root-domain */ |
05fa785c | 6619 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 6620 | if (rq->rd) { |
c6c4927b | 6621 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
6622 | |
6623 | set_rq_online(rq); | |
1f94ef59 | 6624 | } |
05fa785c | 6625 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 6626 | break; |
48f24c4d | 6627 | |
1da177e4 | 6628 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 6629 | case CPU_DYING: |
317f3941 | 6630 | sched_ttwu_pending(); |
57d885fe | 6631 | /* Update our root-domain */ |
05fa785c | 6632 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 6633 | if (rq->rd) { |
c6c4927b | 6634 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 6635 | set_rq_offline(rq); |
57d885fe | 6636 | } |
48c5ccae PZ |
6637 | migrate_tasks(cpu); |
6638 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 6639 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
48c5ccae PZ |
6640 | |
6641 | migrate_nr_uninterruptible(rq); | |
6642 | calc_global_load_remove(rq); | |
57d885fe | 6643 | break; |
1da177e4 LT |
6644 | #endif |
6645 | } | |
49c022e6 PZ |
6646 | |
6647 | update_max_interval(); | |
6648 | ||
1da177e4 LT |
6649 | return NOTIFY_OK; |
6650 | } | |
6651 | ||
f38b0820 PM |
6652 | /* |
6653 | * Register at high priority so that task migration (migrate_all_tasks) | |
6654 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 6655 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 6656 | */ |
26c2143b | 6657 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 6658 | .notifier_call = migration_call, |
50a323b7 | 6659 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
6660 | }; |
6661 | ||
3a101d05 TH |
6662 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
6663 | unsigned long action, void *hcpu) | |
6664 | { | |
6665 | switch (action & ~CPU_TASKS_FROZEN) { | |
6666 | case CPU_ONLINE: | |
6667 | case CPU_DOWN_FAILED: | |
6668 | set_cpu_active((long)hcpu, true); | |
6669 | return NOTIFY_OK; | |
6670 | default: | |
6671 | return NOTIFY_DONE; | |
6672 | } | |
6673 | } | |
6674 | ||
6675 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
6676 | unsigned long action, void *hcpu) | |
6677 | { | |
6678 | switch (action & ~CPU_TASKS_FROZEN) { | |
6679 | case CPU_DOWN_PREPARE: | |
6680 | set_cpu_active((long)hcpu, false); | |
6681 | return NOTIFY_OK; | |
6682 | default: | |
6683 | return NOTIFY_DONE; | |
6684 | } | |
6685 | } | |
6686 | ||
7babe8db | 6687 | static int __init migration_init(void) |
1da177e4 LT |
6688 | { |
6689 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6690 | int err; |
48f24c4d | 6691 | |
3a101d05 | 6692 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
6693 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6694 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6695 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6696 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 6697 | |
3a101d05 TH |
6698 | /* Register cpu active notifiers */ |
6699 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
6700 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
6701 | ||
a004cd42 | 6702 | return 0; |
1da177e4 | 6703 | } |
7babe8db | 6704 | early_initcall(migration_init); |
1da177e4 LT |
6705 | #endif |
6706 | ||
6707 | #ifdef CONFIG_SMP | |
476f3534 | 6708 | |
4cb98839 PZ |
6709 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
6710 | ||
3e9830dc | 6711 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6712 | |
f6630114 MT |
6713 | static __read_mostly int sched_domain_debug_enabled; |
6714 | ||
6715 | static int __init sched_domain_debug_setup(char *str) | |
6716 | { | |
6717 | sched_domain_debug_enabled = 1; | |
6718 | ||
6719 | return 0; | |
6720 | } | |
6721 | early_param("sched_debug", sched_domain_debug_setup); | |
6722 | ||
7c16ec58 | 6723 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 6724 | struct cpumask *groupmask) |
1da177e4 | 6725 | { |
4dcf6aff | 6726 | struct sched_group *group = sd->groups; |
434d53b0 | 6727 | char str[256]; |
1da177e4 | 6728 | |
968ea6d8 | 6729 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 6730 | cpumask_clear(groupmask); |
4dcf6aff IM |
6731 | |
6732 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6733 | ||
6734 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 6735 | printk("does not load-balance\n"); |
4dcf6aff | 6736 | if (sd->parent) |
3df0fc5b PZ |
6737 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
6738 | " has parent"); | |
4dcf6aff | 6739 | return -1; |
41c7ce9a NP |
6740 | } |
6741 | ||
3df0fc5b | 6742 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 6743 | |
758b2cdc | 6744 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
6745 | printk(KERN_ERR "ERROR: domain->span does not contain " |
6746 | "CPU%d\n", cpu); | |
4dcf6aff | 6747 | } |
758b2cdc | 6748 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
6749 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
6750 | " CPU%d\n", cpu); | |
4dcf6aff | 6751 | } |
1da177e4 | 6752 | |
4dcf6aff | 6753 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6754 | do { |
4dcf6aff | 6755 | if (!group) { |
3df0fc5b PZ |
6756 | printk("\n"); |
6757 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6758 | break; |
6759 | } | |
6760 | ||
9c3f75cb | 6761 | if (!group->sgp->power) { |
3df0fc5b PZ |
6762 | printk(KERN_CONT "\n"); |
6763 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6764 | "set\n"); | |
4dcf6aff IM |
6765 | break; |
6766 | } | |
1da177e4 | 6767 | |
758b2cdc | 6768 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
6769 | printk(KERN_CONT "\n"); |
6770 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
6771 | break; |
6772 | } | |
1da177e4 | 6773 | |
758b2cdc | 6774 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
6775 | printk(KERN_CONT "\n"); |
6776 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
6777 | break; |
6778 | } | |
1da177e4 | 6779 | |
758b2cdc | 6780 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 6781 | |
968ea6d8 | 6782 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 6783 | |
3df0fc5b | 6784 | printk(KERN_CONT " %s", str); |
9c3f75cb | 6785 | if (group->sgp->power != SCHED_POWER_SCALE) { |
3df0fc5b | 6786 | printk(KERN_CONT " (cpu_power = %d)", |
9c3f75cb | 6787 | group->sgp->power); |
381512cf | 6788 | } |
1da177e4 | 6789 | |
4dcf6aff IM |
6790 | group = group->next; |
6791 | } while (group != sd->groups); | |
3df0fc5b | 6792 | printk(KERN_CONT "\n"); |
1da177e4 | 6793 | |
758b2cdc | 6794 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 6795 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6796 | |
758b2cdc RR |
6797 | if (sd->parent && |
6798 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
6799 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6800 | "of domain->span\n"); | |
4dcf6aff IM |
6801 | return 0; |
6802 | } | |
1da177e4 | 6803 | |
4dcf6aff IM |
6804 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6805 | { | |
6806 | int level = 0; | |
1da177e4 | 6807 | |
f6630114 MT |
6808 | if (!sched_domain_debug_enabled) |
6809 | return; | |
6810 | ||
4dcf6aff IM |
6811 | if (!sd) { |
6812 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6813 | return; | |
6814 | } | |
1da177e4 | 6815 | |
4dcf6aff IM |
6816 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6817 | ||
6818 | for (;;) { | |
4cb98839 | 6819 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 6820 | break; |
1da177e4 LT |
6821 | level++; |
6822 | sd = sd->parent; | |
33859f7f | 6823 | if (!sd) |
4dcf6aff IM |
6824 | break; |
6825 | } | |
1da177e4 | 6826 | } |
6d6bc0ad | 6827 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6828 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6829 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6830 | |
1a20ff27 | 6831 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 6832 | { |
758b2cdc | 6833 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
6834 | return 1; |
6835 | ||
6836 | /* Following flags need at least 2 groups */ | |
6837 | if (sd->flags & (SD_LOAD_BALANCE | | |
6838 | SD_BALANCE_NEWIDLE | | |
6839 | SD_BALANCE_FORK | | |
89c4710e SS |
6840 | SD_BALANCE_EXEC | |
6841 | SD_SHARE_CPUPOWER | | |
6842 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6843 | if (sd->groups != sd->groups->next) |
6844 | return 0; | |
6845 | } | |
6846 | ||
6847 | /* Following flags don't use groups */ | |
c88d5910 | 6848 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
6849 | return 0; |
6850 | ||
6851 | return 1; | |
6852 | } | |
6853 | ||
48f24c4d IM |
6854 | static int |
6855 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6856 | { |
6857 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6858 | ||
6859 | if (sd_degenerate(parent)) | |
6860 | return 1; | |
6861 | ||
758b2cdc | 6862 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
6863 | return 0; |
6864 | ||
245af2c7 SS |
6865 | /* Flags needing groups don't count if only 1 group in parent */ |
6866 | if (parent->groups == parent->groups->next) { | |
6867 | pflags &= ~(SD_LOAD_BALANCE | | |
6868 | SD_BALANCE_NEWIDLE | | |
6869 | SD_BALANCE_FORK | | |
89c4710e SS |
6870 | SD_BALANCE_EXEC | |
6871 | SD_SHARE_CPUPOWER | | |
6872 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6873 | if (nr_node_ids == 1) |
6874 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6875 | } |
6876 | if (~cflags & pflags) | |
6877 | return 0; | |
6878 | ||
6879 | return 1; | |
6880 | } | |
6881 | ||
dce840a0 | 6882 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 6883 | { |
dce840a0 | 6884 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 6885 | |
68e74568 | 6886 | cpupri_cleanup(&rd->cpupri); |
c6c4927b RR |
6887 | free_cpumask_var(rd->rto_mask); |
6888 | free_cpumask_var(rd->online); | |
6889 | free_cpumask_var(rd->span); | |
6890 | kfree(rd); | |
6891 | } | |
6892 | ||
57d885fe GH |
6893 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6894 | { | |
a0490fa3 | 6895 | struct root_domain *old_rd = NULL; |
57d885fe | 6896 | unsigned long flags; |
57d885fe | 6897 | |
05fa785c | 6898 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
6899 | |
6900 | if (rq->rd) { | |
a0490fa3 | 6901 | old_rd = rq->rd; |
57d885fe | 6902 | |
c6c4927b | 6903 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6904 | set_rq_offline(rq); |
57d885fe | 6905 | |
c6c4927b | 6906 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6907 | |
a0490fa3 IM |
6908 | /* |
6909 | * If we dont want to free the old_rt yet then | |
6910 | * set old_rd to NULL to skip the freeing later | |
6911 | * in this function: | |
6912 | */ | |
6913 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
6914 | old_rd = NULL; | |
57d885fe GH |
6915 | } |
6916 | ||
6917 | atomic_inc(&rd->refcount); | |
6918 | rq->rd = rd; | |
6919 | ||
c6c4927b | 6920 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 6921 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 6922 | set_rq_online(rq); |
57d885fe | 6923 | |
05fa785c | 6924 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
6925 | |
6926 | if (old_rd) | |
dce840a0 | 6927 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
6928 | } |
6929 | ||
68c38fc3 | 6930 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6931 | { |
6932 | memset(rd, 0, sizeof(*rd)); | |
6933 | ||
68c38fc3 | 6934 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 6935 | goto out; |
68c38fc3 | 6936 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 6937 | goto free_span; |
68c38fc3 | 6938 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 6939 | goto free_online; |
6e0534f2 | 6940 | |
68c38fc3 | 6941 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 6942 | goto free_rto_mask; |
c6c4927b | 6943 | return 0; |
6e0534f2 | 6944 | |
68e74568 RR |
6945 | free_rto_mask: |
6946 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6947 | free_online: |
6948 | free_cpumask_var(rd->online); | |
6949 | free_span: | |
6950 | free_cpumask_var(rd->span); | |
0c910d28 | 6951 | out: |
c6c4927b | 6952 | return -ENOMEM; |
57d885fe GH |
6953 | } |
6954 | ||
6955 | static void init_defrootdomain(void) | |
6956 | { | |
68c38fc3 | 6957 | init_rootdomain(&def_root_domain); |
c6c4927b | 6958 | |
57d885fe GH |
6959 | atomic_set(&def_root_domain.refcount, 1); |
6960 | } | |
6961 | ||
dc938520 | 6962 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6963 | { |
6964 | struct root_domain *rd; | |
6965 | ||
6966 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6967 | if (!rd) | |
6968 | return NULL; | |
6969 | ||
68c38fc3 | 6970 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
6971 | kfree(rd); |
6972 | return NULL; | |
6973 | } | |
57d885fe GH |
6974 | |
6975 | return rd; | |
6976 | } | |
6977 | ||
e3589f6c PZ |
6978 | static void free_sched_groups(struct sched_group *sg, int free_sgp) |
6979 | { | |
6980 | struct sched_group *tmp, *first; | |
6981 | ||
6982 | if (!sg) | |
6983 | return; | |
6984 | ||
6985 | first = sg; | |
6986 | do { | |
6987 | tmp = sg->next; | |
6988 | ||
6989 | if (free_sgp && atomic_dec_and_test(&sg->sgp->ref)) | |
6990 | kfree(sg->sgp); | |
6991 | ||
6992 | kfree(sg); | |
6993 | sg = tmp; | |
6994 | } while (sg != first); | |
6995 | } | |
6996 | ||
dce840a0 PZ |
6997 | static void free_sched_domain(struct rcu_head *rcu) |
6998 | { | |
6999 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
7000 | |
7001 | /* | |
7002 | * If its an overlapping domain it has private groups, iterate and | |
7003 | * nuke them all. | |
7004 | */ | |
7005 | if (sd->flags & SD_OVERLAP) { | |
7006 | free_sched_groups(sd->groups, 1); | |
7007 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
9c3f75cb | 7008 | kfree(sd->groups->sgp); |
dce840a0 | 7009 | kfree(sd->groups); |
9c3f75cb | 7010 | } |
dce840a0 PZ |
7011 | kfree(sd); |
7012 | } | |
7013 | ||
7014 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
7015 | { | |
7016 | call_rcu(&sd->rcu, free_sched_domain); | |
7017 | } | |
7018 | ||
7019 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
7020 | { | |
7021 | for (; sd; sd = sd->parent) | |
7022 | destroy_sched_domain(sd, cpu); | |
7023 | } | |
7024 | ||
1da177e4 | 7025 | /* |
0eab9146 | 7026 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
7027 | * hold the hotplug lock. |
7028 | */ | |
0eab9146 IM |
7029 | static void |
7030 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 7031 | { |
70b97a7f | 7032 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
7033 | struct sched_domain *tmp; |
7034 | ||
7035 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 7036 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
7037 | struct sched_domain *parent = tmp->parent; |
7038 | if (!parent) | |
7039 | break; | |
f29c9b1c | 7040 | |
1a848870 | 7041 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 7042 | tmp->parent = parent->parent; |
1a848870 SS |
7043 | if (parent->parent) |
7044 | parent->parent->child = tmp; | |
dce840a0 | 7045 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
7046 | } else |
7047 | tmp = tmp->parent; | |
245af2c7 SS |
7048 | } |
7049 | ||
1a848870 | 7050 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 7051 | tmp = sd; |
245af2c7 | 7052 | sd = sd->parent; |
dce840a0 | 7053 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
7054 | if (sd) |
7055 | sd->child = NULL; | |
7056 | } | |
1da177e4 | 7057 | |
4cb98839 | 7058 | sched_domain_debug(sd, cpu); |
1da177e4 | 7059 | |
57d885fe | 7060 | rq_attach_root(rq, rd); |
dce840a0 | 7061 | tmp = rq->sd; |
674311d5 | 7062 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 7063 | destroy_sched_domains(tmp, cpu); |
1da177e4 LT |
7064 | } |
7065 | ||
7066 | /* cpus with isolated domains */ | |
dcc30a35 | 7067 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
7068 | |
7069 | /* Setup the mask of cpus configured for isolated domains */ | |
7070 | static int __init isolated_cpu_setup(char *str) | |
7071 | { | |
bdddd296 | 7072 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 7073 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
7074 | return 1; |
7075 | } | |
7076 | ||
8927f494 | 7077 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 7078 | |
9c1cfda2 | 7079 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 7080 | |
9c1cfda2 | 7081 | #ifdef CONFIG_NUMA |
198e2f18 | 7082 | |
9c1cfda2 JH |
7083 | /** |
7084 | * find_next_best_node - find the next node to include in a sched_domain | |
7085 | * @node: node whose sched_domain we're building | |
7086 | * @used_nodes: nodes already in the sched_domain | |
7087 | * | |
41a2d6cf | 7088 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
7089 | * finds the closest node not already in the @used_nodes map. |
7090 | * | |
7091 | * Should use nodemask_t. | |
7092 | */ | |
c5f59f08 | 7093 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 | 7094 | { |
7142d17e | 7095 | int i, n, val, min_val, best_node = -1; |
9c1cfda2 JH |
7096 | |
7097 | min_val = INT_MAX; | |
7098 | ||
076ac2af | 7099 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 7100 | /* Start at @node */ |
076ac2af | 7101 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
7102 | |
7103 | if (!nr_cpus_node(n)) | |
7104 | continue; | |
7105 | ||
7106 | /* Skip already used nodes */ | |
c5f59f08 | 7107 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
7108 | continue; |
7109 | ||
7110 | /* Simple min distance search */ | |
7111 | val = node_distance(node, n); | |
7112 | ||
7113 | if (val < min_val) { | |
7114 | min_val = val; | |
7115 | best_node = n; | |
7116 | } | |
7117 | } | |
7118 | ||
7142d17e HD |
7119 | if (best_node != -1) |
7120 | node_set(best_node, *used_nodes); | |
9c1cfda2 JH |
7121 | return best_node; |
7122 | } | |
7123 | ||
7124 | /** | |
7125 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
7126 | * @node: node whose cpumask we're constructing | |
73486722 | 7127 | * @span: resulting cpumask |
9c1cfda2 | 7128 | * |
41a2d6cf | 7129 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
7130 | * should be one that prevents unnecessary balancing, but also spreads tasks |
7131 | * out optimally. | |
7132 | */ | |
96f874e2 | 7133 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 7134 | { |
c5f59f08 | 7135 | nodemask_t used_nodes; |
48f24c4d | 7136 | int i; |
9c1cfda2 | 7137 | |
6ca09dfc | 7138 | cpumask_clear(span); |
c5f59f08 | 7139 | nodes_clear(used_nodes); |
9c1cfda2 | 7140 | |
6ca09dfc | 7141 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 7142 | node_set(node, used_nodes); |
9c1cfda2 JH |
7143 | |
7144 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 7145 | int next_node = find_next_best_node(node, &used_nodes); |
7142d17e HD |
7146 | if (next_node < 0) |
7147 | break; | |
6ca09dfc | 7148 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 7149 | } |
9c1cfda2 | 7150 | } |
d3081f52 PZ |
7151 | |
7152 | static const struct cpumask *cpu_node_mask(int cpu) | |
7153 | { | |
7154 | lockdep_assert_held(&sched_domains_mutex); | |
7155 | ||
7156 | sched_domain_node_span(cpu_to_node(cpu), sched_domains_tmpmask); | |
7157 | ||
7158 | return sched_domains_tmpmask; | |
7159 | } | |
2c402dc3 PZ |
7160 | |
7161 | static const struct cpumask *cpu_allnodes_mask(int cpu) | |
7162 | { | |
7163 | return cpu_possible_mask; | |
7164 | } | |
6d6bc0ad | 7165 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 7166 | |
d3081f52 PZ |
7167 | static const struct cpumask *cpu_cpu_mask(int cpu) |
7168 | { | |
7169 | return cpumask_of_node(cpu_to_node(cpu)); | |
7170 | } | |
7171 | ||
5c45bf27 | 7172 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 7173 | |
dce840a0 PZ |
7174 | struct sd_data { |
7175 | struct sched_domain **__percpu sd; | |
7176 | struct sched_group **__percpu sg; | |
9c3f75cb | 7177 | struct sched_group_power **__percpu sgp; |
dce840a0 PZ |
7178 | }; |
7179 | ||
49a02c51 | 7180 | struct s_data { |
21d42ccf | 7181 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
7182 | struct root_domain *rd; |
7183 | }; | |
7184 | ||
2109b99e | 7185 | enum s_alloc { |
2109b99e | 7186 | sa_rootdomain, |
21d42ccf | 7187 | sa_sd, |
dce840a0 | 7188 | sa_sd_storage, |
2109b99e AH |
7189 | sa_none, |
7190 | }; | |
7191 | ||
54ab4ff4 PZ |
7192 | struct sched_domain_topology_level; |
7193 | ||
7194 | typedef struct sched_domain *(*sched_domain_init_f)(struct sched_domain_topology_level *tl, int cpu); | |
eb7a74e6 PZ |
7195 | typedef const struct cpumask *(*sched_domain_mask_f)(int cpu); |
7196 | ||
e3589f6c PZ |
7197 | #define SDTL_OVERLAP 0x01 |
7198 | ||
eb7a74e6 | 7199 | struct sched_domain_topology_level { |
2c402dc3 PZ |
7200 | sched_domain_init_f init; |
7201 | sched_domain_mask_f mask; | |
e3589f6c | 7202 | int flags; |
54ab4ff4 | 7203 | struct sd_data data; |
eb7a74e6 PZ |
7204 | }; |
7205 | ||
e3589f6c PZ |
7206 | static int |
7207 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
7208 | { | |
7209 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
7210 | const struct cpumask *span = sched_domain_span(sd); | |
7211 | struct cpumask *covered = sched_domains_tmpmask; | |
7212 | struct sd_data *sdd = sd->private; | |
7213 | struct sched_domain *child; | |
7214 | int i; | |
7215 | ||
7216 | cpumask_clear(covered); | |
7217 | ||
7218 | for_each_cpu(i, span) { | |
7219 | struct cpumask *sg_span; | |
7220 | ||
7221 | if (cpumask_test_cpu(i, covered)) | |
7222 | continue; | |
7223 | ||
7224 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
7225 | GFP_KERNEL, cpu_to_node(i)); | |
7226 | ||
7227 | if (!sg) | |
7228 | goto fail; | |
7229 | ||
7230 | sg_span = sched_group_cpus(sg); | |
7231 | ||
7232 | child = *per_cpu_ptr(sdd->sd, i); | |
7233 | if (child->child) { | |
7234 | child = child->child; | |
7235 | cpumask_copy(sg_span, sched_domain_span(child)); | |
7236 | } else | |
7237 | cpumask_set_cpu(i, sg_span); | |
7238 | ||
7239 | cpumask_or(covered, covered, sg_span); | |
7240 | ||
7241 | sg->sgp = *per_cpu_ptr(sdd->sgp, cpumask_first(sg_span)); | |
7242 | atomic_inc(&sg->sgp->ref); | |
7243 | ||
7244 | if (cpumask_test_cpu(cpu, sg_span)) | |
7245 | groups = sg; | |
7246 | ||
7247 | if (!first) | |
7248 | first = sg; | |
7249 | if (last) | |
7250 | last->next = sg; | |
7251 | last = sg; | |
7252 | last->next = first; | |
7253 | } | |
7254 | sd->groups = groups; | |
7255 | ||
7256 | return 0; | |
7257 | ||
7258 | fail: | |
7259 | free_sched_groups(first, 0); | |
7260 | ||
7261 | return -ENOMEM; | |
7262 | } | |
7263 | ||
dce840a0 | 7264 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 7265 | { |
dce840a0 PZ |
7266 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
7267 | struct sched_domain *child = sd->child; | |
1da177e4 | 7268 | |
dce840a0 PZ |
7269 | if (child) |
7270 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 7271 | |
9c3f75cb | 7272 | if (sg) { |
dce840a0 | 7273 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
9c3f75cb | 7274 | (*sg)->sgp = *per_cpu_ptr(sdd->sgp, cpu); |
e3589f6c | 7275 | atomic_set(&(*sg)->sgp->ref, 1); /* for claim_allocations */ |
9c3f75cb | 7276 | } |
dce840a0 PZ |
7277 | |
7278 | return cpu; | |
1e9f28fa | 7279 | } |
1e9f28fa | 7280 | |
01a08546 | 7281 | /* |
dce840a0 PZ |
7282 | * build_sched_groups will build a circular linked list of the groups |
7283 | * covered by the given span, and will set each group's ->cpumask correctly, | |
7284 | * and ->cpu_power to 0. | |
e3589f6c PZ |
7285 | * |
7286 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 7287 | */ |
e3589f6c PZ |
7288 | static int |
7289 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 7290 | { |
dce840a0 PZ |
7291 | struct sched_group *first = NULL, *last = NULL; |
7292 | struct sd_data *sdd = sd->private; | |
7293 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 7294 | struct cpumask *covered; |
dce840a0 | 7295 | int i; |
9c1cfda2 | 7296 | |
e3589f6c PZ |
7297 | get_group(cpu, sdd, &sd->groups); |
7298 | atomic_inc(&sd->groups->ref); | |
7299 | ||
7300 | if (cpu != cpumask_first(sched_domain_span(sd))) | |
7301 | return 0; | |
7302 | ||
f96225fd PZ |
7303 | lockdep_assert_held(&sched_domains_mutex); |
7304 | covered = sched_domains_tmpmask; | |
7305 | ||
dce840a0 | 7306 | cpumask_clear(covered); |
6711cab4 | 7307 | |
dce840a0 PZ |
7308 | for_each_cpu(i, span) { |
7309 | struct sched_group *sg; | |
7310 | int group = get_group(i, sdd, &sg); | |
7311 | int j; | |
6711cab4 | 7312 | |
dce840a0 PZ |
7313 | if (cpumask_test_cpu(i, covered)) |
7314 | continue; | |
6711cab4 | 7315 | |
dce840a0 | 7316 | cpumask_clear(sched_group_cpus(sg)); |
9c3f75cb | 7317 | sg->sgp->power = 0; |
0601a88d | 7318 | |
dce840a0 PZ |
7319 | for_each_cpu(j, span) { |
7320 | if (get_group(j, sdd, NULL) != group) | |
7321 | continue; | |
0601a88d | 7322 | |
dce840a0 PZ |
7323 | cpumask_set_cpu(j, covered); |
7324 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
7325 | } | |
0601a88d | 7326 | |
dce840a0 PZ |
7327 | if (!first) |
7328 | first = sg; | |
7329 | if (last) | |
7330 | last->next = sg; | |
7331 | last = sg; | |
7332 | } | |
7333 | last->next = first; | |
e3589f6c PZ |
7334 | |
7335 | return 0; | |
0601a88d | 7336 | } |
51888ca2 | 7337 | |
89c4710e SS |
7338 | /* |
7339 | * Initialize sched groups cpu_power. | |
7340 | * | |
7341 | * cpu_power indicates the capacity of sched group, which is used while | |
7342 | * distributing the load between different sched groups in a sched domain. | |
7343 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
7344 | * there are asymmetries in the topology. If there are asymmetries, group | |
7345 | * having more cpu_power will pickup more load compared to the group having | |
7346 | * less cpu_power. | |
89c4710e SS |
7347 | */ |
7348 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
7349 | { | |
e3589f6c | 7350 | struct sched_group *sg = sd->groups; |
89c4710e | 7351 | |
e3589f6c PZ |
7352 | WARN_ON(!sd || !sg); |
7353 | ||
7354 | do { | |
7355 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
7356 | sg = sg->next; | |
7357 | } while (sg != sd->groups); | |
89c4710e | 7358 | |
e3589f6c PZ |
7359 | if (cpu != group_first_cpu(sg)) |
7360 | return; | |
aae6d3dd | 7361 | |
d274cb30 | 7362 | update_group_power(sd, cpu); |
89c4710e SS |
7363 | } |
7364 | ||
7c16ec58 MT |
7365 | /* |
7366 | * Initializers for schedule domains | |
7367 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7368 | */ | |
7369 | ||
a5d8c348 IM |
7370 | #ifdef CONFIG_SCHED_DEBUG |
7371 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
7372 | #else | |
7373 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
7374 | #endif | |
7375 | ||
54ab4ff4 PZ |
7376 | #define SD_INIT_FUNC(type) \ |
7377 | static noinline struct sched_domain * \ | |
7378 | sd_init_##type(struct sched_domain_topology_level *tl, int cpu) \ | |
7379 | { \ | |
7380 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); \ | |
7381 | *sd = SD_##type##_INIT; \ | |
54ab4ff4 PZ |
7382 | SD_INIT_NAME(sd, type); \ |
7383 | sd->private = &tl->data; \ | |
7384 | return sd; \ | |
7c16ec58 MT |
7385 | } |
7386 | ||
7387 | SD_INIT_FUNC(CPU) | |
7388 | #ifdef CONFIG_NUMA | |
7389 | SD_INIT_FUNC(ALLNODES) | |
7390 | SD_INIT_FUNC(NODE) | |
7391 | #endif | |
7392 | #ifdef CONFIG_SCHED_SMT | |
7393 | SD_INIT_FUNC(SIBLING) | |
7394 | #endif | |
7395 | #ifdef CONFIG_SCHED_MC | |
7396 | SD_INIT_FUNC(MC) | |
7397 | #endif | |
01a08546 HC |
7398 | #ifdef CONFIG_SCHED_BOOK |
7399 | SD_INIT_FUNC(BOOK) | |
7400 | #endif | |
7c16ec58 | 7401 | |
1d3504fc | 7402 | static int default_relax_domain_level = -1; |
60495e77 | 7403 | int sched_domain_level_max; |
1d3504fc HS |
7404 | |
7405 | static int __init setup_relax_domain_level(char *str) | |
7406 | { | |
30e0e178 LZ |
7407 | unsigned long val; |
7408 | ||
7409 | val = simple_strtoul(str, NULL, 0); | |
60495e77 | 7410 | if (val < sched_domain_level_max) |
30e0e178 LZ |
7411 | default_relax_domain_level = val; |
7412 | ||
1d3504fc HS |
7413 | return 1; |
7414 | } | |
7415 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7416 | ||
7417 | static void set_domain_attribute(struct sched_domain *sd, | |
7418 | struct sched_domain_attr *attr) | |
7419 | { | |
7420 | int request; | |
7421 | ||
7422 | if (!attr || attr->relax_domain_level < 0) { | |
7423 | if (default_relax_domain_level < 0) | |
7424 | return; | |
7425 | else | |
7426 | request = default_relax_domain_level; | |
7427 | } else | |
7428 | request = attr->relax_domain_level; | |
7429 | if (request < sd->level) { | |
7430 | /* turn off idle balance on this domain */ | |
c88d5910 | 7431 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7432 | } else { |
7433 | /* turn on idle balance on this domain */ | |
c88d5910 | 7434 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7435 | } |
7436 | } | |
7437 | ||
54ab4ff4 PZ |
7438 | static void __sdt_free(const struct cpumask *cpu_map); |
7439 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
7440 | ||
2109b99e AH |
7441 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
7442 | const struct cpumask *cpu_map) | |
7443 | { | |
7444 | switch (what) { | |
2109b99e | 7445 | case sa_rootdomain: |
822ff793 PZ |
7446 | if (!atomic_read(&d->rd->refcount)) |
7447 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
7448 | case sa_sd: |
7449 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 7450 | case sa_sd_storage: |
54ab4ff4 | 7451 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
7452 | case sa_none: |
7453 | break; | |
7454 | } | |
7455 | } | |
3404c8d9 | 7456 | |
2109b99e AH |
7457 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
7458 | const struct cpumask *cpu_map) | |
7459 | { | |
dce840a0 PZ |
7460 | memset(d, 0, sizeof(*d)); |
7461 | ||
54ab4ff4 PZ |
7462 | if (__sdt_alloc(cpu_map)) |
7463 | return sa_sd_storage; | |
dce840a0 PZ |
7464 | d->sd = alloc_percpu(struct sched_domain *); |
7465 | if (!d->sd) | |
7466 | return sa_sd_storage; | |
2109b99e | 7467 | d->rd = alloc_rootdomain(); |
dce840a0 | 7468 | if (!d->rd) |
21d42ccf | 7469 | return sa_sd; |
2109b99e AH |
7470 | return sa_rootdomain; |
7471 | } | |
57d885fe | 7472 | |
dce840a0 PZ |
7473 | /* |
7474 | * NULL the sd_data elements we've used to build the sched_domain and | |
7475 | * sched_group structure so that the subsequent __free_domain_allocs() | |
7476 | * will not free the data we're using. | |
7477 | */ | |
7478 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
7479 | { | |
7480 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
7481 | |
7482 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
7483 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
7484 | ||
e3589f6c | 7485 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 7486 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c PZ |
7487 | |
7488 | if (atomic_read(&(*per_cpu_ptr(sdd->sgp, cpu))->ref)) | |
9c3f75cb | 7489 | *per_cpu_ptr(sdd->sgp, cpu) = NULL; |
dce840a0 PZ |
7490 | } |
7491 | ||
2c402dc3 PZ |
7492 | #ifdef CONFIG_SCHED_SMT |
7493 | static const struct cpumask *cpu_smt_mask(int cpu) | |
7f4588f3 | 7494 | { |
2c402dc3 | 7495 | return topology_thread_cpumask(cpu); |
3bd65a80 | 7496 | } |
2c402dc3 | 7497 | #endif |
7f4588f3 | 7498 | |
d069b916 PZ |
7499 | /* |
7500 | * Topology list, bottom-up. | |
7501 | */ | |
2c402dc3 | 7502 | static struct sched_domain_topology_level default_topology[] = { |
d069b916 PZ |
7503 | #ifdef CONFIG_SCHED_SMT |
7504 | { sd_init_SIBLING, cpu_smt_mask, }, | |
01a08546 | 7505 | #endif |
1e9f28fa | 7506 | #ifdef CONFIG_SCHED_MC |
2c402dc3 | 7507 | { sd_init_MC, cpu_coregroup_mask, }, |
1e9f28fa | 7508 | #endif |
d069b916 PZ |
7509 | #ifdef CONFIG_SCHED_BOOK |
7510 | { sd_init_BOOK, cpu_book_mask, }, | |
7511 | #endif | |
7512 | { sd_init_CPU, cpu_cpu_mask, }, | |
7513 | #ifdef CONFIG_NUMA | |
e3589f6c | 7514 | { sd_init_NODE, cpu_node_mask, SDTL_OVERLAP, }, |
d069b916 | 7515 | { sd_init_ALLNODES, cpu_allnodes_mask, }, |
1da177e4 | 7516 | #endif |
eb7a74e6 PZ |
7517 | { NULL, }, |
7518 | }; | |
7519 | ||
7520 | static struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
7521 | ||
54ab4ff4 PZ |
7522 | static int __sdt_alloc(const struct cpumask *cpu_map) |
7523 | { | |
7524 | struct sched_domain_topology_level *tl; | |
7525 | int j; | |
7526 | ||
7527 | for (tl = sched_domain_topology; tl->init; tl++) { | |
7528 | struct sd_data *sdd = &tl->data; | |
7529 | ||
7530 | sdd->sd = alloc_percpu(struct sched_domain *); | |
7531 | if (!sdd->sd) | |
7532 | return -ENOMEM; | |
7533 | ||
7534 | sdd->sg = alloc_percpu(struct sched_group *); | |
7535 | if (!sdd->sg) | |
7536 | return -ENOMEM; | |
7537 | ||
9c3f75cb PZ |
7538 | sdd->sgp = alloc_percpu(struct sched_group_power *); |
7539 | if (!sdd->sgp) | |
7540 | return -ENOMEM; | |
7541 | ||
54ab4ff4 PZ |
7542 | for_each_cpu(j, cpu_map) { |
7543 | struct sched_domain *sd; | |
7544 | struct sched_group *sg; | |
9c3f75cb | 7545 | struct sched_group_power *sgp; |
54ab4ff4 PZ |
7546 | |
7547 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
7548 | GFP_KERNEL, cpu_to_node(j)); | |
7549 | if (!sd) | |
7550 | return -ENOMEM; | |
7551 | ||
7552 | *per_cpu_ptr(sdd->sd, j) = sd; | |
7553 | ||
7554 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
7555 | GFP_KERNEL, cpu_to_node(j)); | |
7556 | if (!sg) | |
7557 | return -ENOMEM; | |
7558 | ||
7559 | *per_cpu_ptr(sdd->sg, j) = sg; | |
9c3f75cb PZ |
7560 | |
7561 | sgp = kzalloc_node(sizeof(struct sched_group_power), | |
7562 | GFP_KERNEL, cpu_to_node(j)); | |
7563 | if (!sgp) | |
7564 | return -ENOMEM; | |
7565 | ||
7566 | *per_cpu_ptr(sdd->sgp, j) = sgp; | |
54ab4ff4 PZ |
7567 | } |
7568 | } | |
7569 | ||
7570 | return 0; | |
7571 | } | |
7572 | ||
7573 | static void __sdt_free(const struct cpumask *cpu_map) | |
7574 | { | |
7575 | struct sched_domain_topology_level *tl; | |
7576 | int j; | |
7577 | ||
7578 | for (tl = sched_domain_topology; tl->init; tl++) { | |
7579 | struct sd_data *sdd = &tl->data; | |
7580 | ||
7581 | for_each_cpu(j, cpu_map) { | |
e3589f6c PZ |
7582 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, j); |
7583 | if (sd && (sd->flags & SD_OVERLAP)) | |
7584 | free_sched_groups(sd->groups, 0); | |
feff8fa0 | 7585 | kfree(*per_cpu_ptr(sdd->sd, j)); |
54ab4ff4 | 7586 | kfree(*per_cpu_ptr(sdd->sg, j)); |
9c3f75cb | 7587 | kfree(*per_cpu_ptr(sdd->sgp, j)); |
54ab4ff4 PZ |
7588 | } |
7589 | free_percpu(sdd->sd); | |
7590 | free_percpu(sdd->sg); | |
9c3f75cb | 7591 | free_percpu(sdd->sgp); |
54ab4ff4 PZ |
7592 | } |
7593 | } | |
7594 | ||
2c402dc3 PZ |
7595 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
7596 | struct s_data *d, const struct cpumask *cpu_map, | |
d069b916 | 7597 | struct sched_domain_attr *attr, struct sched_domain *child, |
2c402dc3 PZ |
7598 | int cpu) |
7599 | { | |
54ab4ff4 | 7600 | struct sched_domain *sd = tl->init(tl, cpu); |
2c402dc3 | 7601 | if (!sd) |
d069b916 | 7602 | return child; |
2c402dc3 PZ |
7603 | |
7604 | set_domain_attribute(sd, attr); | |
7605 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); | |
60495e77 PZ |
7606 | if (child) { |
7607 | sd->level = child->level + 1; | |
7608 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 7609 | child->parent = sd; |
60495e77 | 7610 | } |
d069b916 | 7611 | sd->child = child; |
2c402dc3 PZ |
7612 | |
7613 | return sd; | |
7614 | } | |
7615 | ||
2109b99e AH |
7616 | /* |
7617 | * Build sched domains for a given set of cpus and attach the sched domains | |
7618 | * to the individual cpus | |
7619 | */ | |
dce840a0 PZ |
7620 | static int build_sched_domains(const struct cpumask *cpu_map, |
7621 | struct sched_domain_attr *attr) | |
2109b99e AH |
7622 | { |
7623 | enum s_alloc alloc_state = sa_none; | |
dce840a0 | 7624 | struct sched_domain *sd; |
2109b99e | 7625 | struct s_data d; |
822ff793 | 7626 | int i, ret = -ENOMEM; |
9c1cfda2 | 7627 | |
2109b99e AH |
7628 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
7629 | if (alloc_state != sa_rootdomain) | |
7630 | goto error; | |
9c1cfda2 | 7631 | |
dce840a0 | 7632 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 7633 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
7634 | struct sched_domain_topology_level *tl; |
7635 | ||
3bd65a80 | 7636 | sd = NULL; |
e3589f6c | 7637 | for (tl = sched_domain_topology; tl->init; tl++) { |
2c402dc3 | 7638 | sd = build_sched_domain(tl, &d, cpu_map, attr, sd, i); |
e3589f6c PZ |
7639 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
7640 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
7641 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
7642 | break; | |
e3589f6c | 7643 | } |
d274cb30 | 7644 | |
d069b916 PZ |
7645 | while (sd->child) |
7646 | sd = sd->child; | |
7647 | ||
21d42ccf | 7648 | *per_cpu_ptr(d.sd, i) = sd; |
dce840a0 PZ |
7649 | } |
7650 | ||
7651 | /* Build the groups for the domains */ | |
7652 | for_each_cpu(i, cpu_map) { | |
7653 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
7654 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
7655 | if (sd->flags & SD_OVERLAP) { |
7656 | if (build_overlap_sched_groups(sd, i)) | |
7657 | goto error; | |
7658 | } else { | |
7659 | if (build_sched_groups(sd, i)) | |
7660 | goto error; | |
7661 | } | |
1cf51902 | 7662 | } |
a06dadbe | 7663 | } |
9c1cfda2 | 7664 | |
1da177e4 | 7665 | /* Calculate CPU power for physical packages and nodes */ |
a9c9a9b6 PZ |
7666 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
7667 | if (!cpumask_test_cpu(i, cpu_map)) | |
7668 | continue; | |
9c1cfda2 | 7669 | |
dce840a0 PZ |
7670 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
7671 | claim_allocations(i, sd); | |
cd4ea6ae | 7672 | init_sched_groups_power(i, sd); |
dce840a0 | 7673 | } |
f712c0c7 | 7674 | } |
9c1cfda2 | 7675 | |
1da177e4 | 7676 | /* Attach the domains */ |
dce840a0 | 7677 | rcu_read_lock(); |
abcd083a | 7678 | for_each_cpu(i, cpu_map) { |
21d42ccf | 7679 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 7680 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 7681 | } |
dce840a0 | 7682 | rcu_read_unlock(); |
51888ca2 | 7683 | |
822ff793 | 7684 | ret = 0; |
51888ca2 | 7685 | error: |
2109b99e | 7686 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 7687 | return ret; |
1da177e4 | 7688 | } |
029190c5 | 7689 | |
acc3f5d7 | 7690 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7691 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7692 | static struct sched_domain_attr *dattr_cur; |
7693 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7694 | |
7695 | /* | |
7696 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7697 | * cpumask) fails, then fallback to a single sched domain, |
7698 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7699 | */ |
4212823f | 7700 | static cpumask_var_t fallback_doms; |
029190c5 | 7701 | |
ee79d1bd HC |
7702 | /* |
7703 | * arch_update_cpu_topology lets virtualized architectures update the | |
7704 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7705 | * or 0 if it stayed the same. | |
7706 | */ | |
7707 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7708 | { |
ee79d1bd | 7709 | return 0; |
22e52b07 HC |
7710 | } |
7711 | ||
acc3f5d7 RR |
7712 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7713 | { | |
7714 | int i; | |
7715 | cpumask_var_t *doms; | |
7716 | ||
7717 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7718 | if (!doms) | |
7719 | return NULL; | |
7720 | for (i = 0; i < ndoms; i++) { | |
7721 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7722 | free_sched_domains(doms, i); | |
7723 | return NULL; | |
7724 | } | |
7725 | } | |
7726 | return doms; | |
7727 | } | |
7728 | ||
7729 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7730 | { | |
7731 | unsigned int i; | |
7732 | for (i = 0; i < ndoms; i++) | |
7733 | free_cpumask_var(doms[i]); | |
7734 | kfree(doms); | |
7735 | } | |
7736 | ||
1a20ff27 | 7737 | /* |
41a2d6cf | 7738 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7739 | * For now this just excludes isolated cpus, but could be used to |
7740 | * exclude other special cases in the future. | |
1a20ff27 | 7741 | */ |
c4a8849a | 7742 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7743 | { |
7378547f MM |
7744 | int err; |
7745 | ||
22e52b07 | 7746 | arch_update_cpu_topology(); |
029190c5 | 7747 | ndoms_cur = 1; |
acc3f5d7 | 7748 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7749 | if (!doms_cur) |
acc3f5d7 RR |
7750 | doms_cur = &fallback_doms; |
7751 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 7752 | dattr_cur = NULL; |
dce840a0 | 7753 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 7754 | register_sched_domain_sysctl(); |
7378547f MM |
7755 | |
7756 | return err; | |
1a20ff27 DG |
7757 | } |
7758 | ||
1a20ff27 DG |
7759 | /* |
7760 | * Detach sched domains from a group of cpus specified in cpu_map | |
7761 | * These cpus will now be attached to the NULL domain | |
7762 | */ | |
96f874e2 | 7763 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
7764 | { |
7765 | int i; | |
7766 | ||
dce840a0 | 7767 | rcu_read_lock(); |
abcd083a | 7768 | for_each_cpu(i, cpu_map) |
57d885fe | 7769 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 7770 | rcu_read_unlock(); |
1a20ff27 DG |
7771 | } |
7772 | ||
1d3504fc HS |
7773 | /* handle null as "default" */ |
7774 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7775 | struct sched_domain_attr *new, int idx_new) | |
7776 | { | |
7777 | struct sched_domain_attr tmp; | |
7778 | ||
7779 | /* fast path */ | |
7780 | if (!new && !cur) | |
7781 | return 1; | |
7782 | ||
7783 | tmp = SD_ATTR_INIT; | |
7784 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7785 | new ? (new + idx_new) : &tmp, | |
7786 | sizeof(struct sched_domain_attr)); | |
7787 | } | |
7788 | ||
029190c5 PJ |
7789 | /* |
7790 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7791 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7792 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7793 | * It destroys each deleted domain and builds each new domain. | |
7794 | * | |
acc3f5d7 | 7795 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7796 | * The masks don't intersect (don't overlap.) We should setup one |
7797 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7798 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7799 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7800 | * it as it is. | |
7801 | * | |
acc3f5d7 RR |
7802 | * The passed in 'doms_new' should be allocated using |
7803 | * alloc_sched_domains. This routine takes ownership of it and will | |
7804 | * free_sched_domains it when done with it. If the caller failed the | |
7805 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7806 | * and partition_sched_domains() will fallback to the single partition | |
7807 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7808 | * |
96f874e2 | 7809 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7810 | * ndoms_new == 0 is a special case for destroying existing domains, |
7811 | * and it will not create the default domain. | |
dfb512ec | 7812 | * |
029190c5 PJ |
7813 | * Call with hotplug lock held |
7814 | */ | |
acc3f5d7 | 7815 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7816 | struct sched_domain_attr *dattr_new) |
029190c5 | 7817 | { |
dfb512ec | 7818 | int i, j, n; |
d65bd5ec | 7819 | int new_topology; |
029190c5 | 7820 | |
712555ee | 7821 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7822 | |
7378547f MM |
7823 | /* always unregister in case we don't destroy any domains */ |
7824 | unregister_sched_domain_sysctl(); | |
7825 | ||
d65bd5ec HC |
7826 | /* Let architecture update cpu core mappings. */ |
7827 | new_topology = arch_update_cpu_topology(); | |
7828 | ||
dfb512ec | 7829 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7830 | |
7831 | /* Destroy deleted domains */ | |
7832 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7833 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7834 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7835 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7836 | goto match1; |
7837 | } | |
7838 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7839 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7840 | match1: |
7841 | ; | |
7842 | } | |
7843 | ||
e761b772 MK |
7844 | if (doms_new == NULL) { |
7845 | ndoms_cur = 0; | |
acc3f5d7 | 7846 | doms_new = &fallback_doms; |
6ad4c188 | 7847 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7848 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7849 | } |
7850 | ||
029190c5 PJ |
7851 | /* Build new domains */ |
7852 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7853 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7854 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7855 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7856 | goto match2; |
7857 | } | |
7858 | /* no match - add a new doms_new */ | |
dce840a0 | 7859 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7860 | match2: |
7861 | ; | |
7862 | } | |
7863 | ||
7864 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7865 | if (doms_cur != &fallback_doms) |
7866 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7867 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7868 | doms_cur = doms_new; |
1d3504fc | 7869 | dattr_cur = dattr_new; |
029190c5 | 7870 | ndoms_cur = ndoms_new; |
7378547f MM |
7871 | |
7872 | register_sched_domain_sysctl(); | |
a1835615 | 7873 | |
712555ee | 7874 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7875 | } |
7876 | ||
5c45bf27 | 7877 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c4a8849a | 7878 | static void reinit_sched_domains(void) |
5c45bf27 | 7879 | { |
95402b38 | 7880 | get_online_cpus(); |
dfb512ec MK |
7881 | |
7882 | /* Destroy domains first to force the rebuild */ | |
7883 | partition_sched_domains(0, NULL, NULL); | |
7884 | ||
e761b772 | 7885 | rebuild_sched_domains(); |
95402b38 | 7886 | put_online_cpus(); |
5c45bf27 SS |
7887 | } |
7888 | ||
7889 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7890 | { | |
afb8a9b7 | 7891 | unsigned int level = 0; |
5c45bf27 | 7892 | |
afb8a9b7 GS |
7893 | if (sscanf(buf, "%u", &level) != 1) |
7894 | return -EINVAL; | |
7895 | ||
7896 | /* | |
7897 | * level is always be positive so don't check for | |
7898 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7899 | * What happens on 0 or 1 byte write, | |
7900 | * need to check for count as well? | |
7901 | */ | |
7902 | ||
7903 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7904 | return -EINVAL; |
7905 | ||
7906 | if (smt) | |
afb8a9b7 | 7907 | sched_smt_power_savings = level; |
5c45bf27 | 7908 | else |
afb8a9b7 | 7909 | sched_mc_power_savings = level; |
5c45bf27 | 7910 | |
c4a8849a | 7911 | reinit_sched_domains(); |
5c45bf27 | 7912 | |
c70f22d2 | 7913 | return count; |
5c45bf27 SS |
7914 | } |
7915 | ||
5c45bf27 | 7916 | #ifdef CONFIG_SCHED_MC |
f718cd4a | 7917 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
c9be0a36 | 7918 | struct sysdev_class_attribute *attr, |
f718cd4a | 7919 | char *page) |
5c45bf27 SS |
7920 | { |
7921 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7922 | } | |
f718cd4a | 7923 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
c9be0a36 | 7924 | struct sysdev_class_attribute *attr, |
48f24c4d | 7925 | const char *buf, size_t count) |
5c45bf27 SS |
7926 | { |
7927 | return sched_power_savings_store(buf, count, 0); | |
7928 | } | |
f718cd4a AK |
7929 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7930 | sched_mc_power_savings_show, | |
7931 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7932 | #endif |
7933 | ||
7934 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a | 7935 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
c9be0a36 | 7936 | struct sysdev_class_attribute *attr, |
f718cd4a | 7937 | char *page) |
5c45bf27 SS |
7938 | { |
7939 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7940 | } | |
f718cd4a | 7941 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
c9be0a36 | 7942 | struct sysdev_class_attribute *attr, |
48f24c4d | 7943 | const char *buf, size_t count) |
5c45bf27 SS |
7944 | { |
7945 | return sched_power_savings_store(buf, count, 1); | |
7946 | } | |
f718cd4a AK |
7947 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7948 | sched_smt_power_savings_show, | |
6707de00 AB |
7949 | sched_smt_power_savings_store); |
7950 | #endif | |
7951 | ||
39aac648 | 7952 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
7953 | { |
7954 | int err = 0; | |
7955 | ||
7956 | #ifdef CONFIG_SCHED_SMT | |
7957 | if (smt_capable()) | |
7958 | err = sysfs_create_file(&cls->kset.kobj, | |
7959 | &attr_sched_smt_power_savings.attr); | |
7960 | #endif | |
7961 | #ifdef CONFIG_SCHED_MC | |
7962 | if (!err && mc_capable()) | |
7963 | err = sysfs_create_file(&cls->kset.kobj, | |
7964 | &attr_sched_mc_power_savings.attr); | |
7965 | #endif | |
7966 | return err; | |
7967 | } | |
6d6bc0ad | 7968 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7969 | |
1da177e4 | 7970 | /* |
3a101d05 TH |
7971 | * Update cpusets according to cpu_active mask. If cpusets are |
7972 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7973 | * around partition_sched_domains(). | |
1da177e4 | 7974 | */ |
0b2e918a TH |
7975 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
7976 | void *hcpu) | |
e761b772 | 7977 | { |
3a101d05 | 7978 | switch (action & ~CPU_TASKS_FROZEN) { |
e761b772 | 7979 | case CPU_ONLINE: |
6ad4c188 | 7980 | case CPU_DOWN_FAILED: |
3a101d05 | 7981 | cpuset_update_active_cpus(); |
e761b772 | 7982 | return NOTIFY_OK; |
3a101d05 TH |
7983 | default: |
7984 | return NOTIFY_DONE; | |
7985 | } | |
7986 | } | |
e761b772 | 7987 | |
0b2e918a TH |
7988 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
7989 | void *hcpu) | |
3a101d05 TH |
7990 | { |
7991 | switch (action & ~CPU_TASKS_FROZEN) { | |
7992 | case CPU_DOWN_PREPARE: | |
7993 | cpuset_update_active_cpus(); | |
7994 | return NOTIFY_OK; | |
e761b772 MK |
7995 | default: |
7996 | return NOTIFY_DONE; | |
7997 | } | |
7998 | } | |
e761b772 MK |
7999 | |
8000 | static int update_runtime(struct notifier_block *nfb, | |
8001 | unsigned long action, void *hcpu) | |
1da177e4 | 8002 | { |
7def2be1 PZ |
8003 | int cpu = (int)(long)hcpu; |
8004 | ||
1da177e4 | 8005 | switch (action) { |
1da177e4 | 8006 | case CPU_DOWN_PREPARE: |
8bb78442 | 8007 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 8008 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
8009 | return NOTIFY_OK; |
8010 | ||
1da177e4 | 8011 | case CPU_DOWN_FAILED: |
8bb78442 | 8012 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 8013 | case CPU_ONLINE: |
8bb78442 | 8014 | case CPU_ONLINE_FROZEN: |
7def2be1 | 8015 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
8016 | return NOTIFY_OK; |
8017 | ||
1da177e4 LT |
8018 | default: |
8019 | return NOTIFY_DONE; | |
8020 | } | |
1da177e4 | 8021 | } |
1da177e4 LT |
8022 | |
8023 | void __init sched_init_smp(void) | |
8024 | { | |
dcc30a35 RR |
8025 | cpumask_var_t non_isolated_cpus; |
8026 | ||
8027 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 8028 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 8029 | |
95402b38 | 8030 | get_online_cpus(); |
712555ee | 8031 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 8032 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
8033 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
8034 | if (cpumask_empty(non_isolated_cpus)) | |
8035 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 8036 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 8037 | put_online_cpus(); |
e761b772 | 8038 | |
3a101d05 TH |
8039 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
8040 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
8041 | |
8042 | /* RT runtime code needs to handle some hotplug events */ | |
8043 | hotcpu_notifier(update_runtime, 0); | |
8044 | ||
b328ca18 | 8045 | init_hrtick(); |
5c1e1767 NP |
8046 | |
8047 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 8048 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 8049 | BUG(); |
19978ca6 | 8050 | sched_init_granularity(); |
dcc30a35 | 8051 | free_cpumask_var(non_isolated_cpus); |
4212823f | 8052 | |
0e3900e6 | 8053 | init_sched_rt_class(); |
1da177e4 LT |
8054 | } |
8055 | #else | |
8056 | void __init sched_init_smp(void) | |
8057 | { | |
19978ca6 | 8058 | sched_init_granularity(); |
1da177e4 LT |
8059 | } |
8060 | #endif /* CONFIG_SMP */ | |
8061 | ||
cd1bb94b AB |
8062 | const_debug unsigned int sysctl_timer_migration = 1; |
8063 | ||
1da177e4 LT |
8064 | int in_sched_functions(unsigned long addr) |
8065 | { | |
1da177e4 LT |
8066 | return in_lock_functions(addr) || |
8067 | (addr >= (unsigned long)__sched_text_start | |
8068 | && addr < (unsigned long)__sched_text_end); | |
8069 | } | |
8070 | ||
acb5a9ba | 8071 | static void init_cfs_rq(struct cfs_rq *cfs_rq) |
dd41f596 IM |
8072 | { |
8073 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 8074 | INIT_LIST_HEAD(&cfs_rq->tasks); |
67e9fb2a | 8075 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
c64be78f PZ |
8076 | #ifndef CONFIG_64BIT |
8077 | cfs_rq->min_vruntime_copy = cfs_rq->min_vruntime; | |
8078 | #endif | |
dd41f596 IM |
8079 | } |
8080 | ||
fa85ae24 PZ |
8081 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
8082 | { | |
8083 | struct rt_prio_array *array; | |
8084 | int i; | |
8085 | ||
8086 | array = &rt_rq->active; | |
8087 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
8088 | INIT_LIST_HEAD(array->queue + i); | |
8089 | __clear_bit(i, array->bitmap); | |
8090 | } | |
8091 | /* delimiter for bitsearch: */ | |
8092 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
8093 | ||
acb5a9ba | 8094 | #if defined CONFIG_SMP |
e864c499 GH |
8095 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
8096 | rt_rq->highest_prio.next = MAX_RT_PRIO; | |
fa85ae24 | 8097 | rt_rq->rt_nr_migratory = 0; |
fa85ae24 | 8098 | rt_rq->overloaded = 0; |
732375c6 | 8099 | plist_head_init(&rt_rq->pushable_tasks); |
fa85ae24 PZ |
8100 | #endif |
8101 | ||
8102 | rt_rq->rt_time = 0; | |
8103 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 8104 | rt_rq->rt_runtime = 0; |
0986b11b | 8105 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
fa85ae24 PZ |
8106 | } |
8107 | ||
6f505b16 | 8108 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac | 8109 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
3d4b47b4 | 8110 | struct sched_entity *se, int cpu, |
ec7dc8ac | 8111 | struct sched_entity *parent) |
6f505b16 | 8112 | { |
ec7dc8ac | 8113 | struct rq *rq = cpu_rq(cpu); |
acb5a9ba | 8114 | |
6f505b16 | 8115 | cfs_rq->tg = tg; |
acb5a9ba JS |
8116 | cfs_rq->rq = rq; |
8117 | #ifdef CONFIG_SMP | |
8118 | /* allow initial update_cfs_load() to truncate */ | |
8119 | cfs_rq->load_stamp = 1; | |
8120 | #endif | |
ab84d31e | 8121 | init_cfs_rq_runtime(cfs_rq); |
6f505b16 | 8122 | |
acb5a9ba | 8123 | tg->cfs_rq[cpu] = cfs_rq; |
6f505b16 | 8124 | tg->se[cpu] = se; |
acb5a9ba | 8125 | |
07e06b01 | 8126 | /* se could be NULL for root_task_group */ |
354d60c2 DG |
8127 | if (!se) |
8128 | return; | |
8129 | ||
ec7dc8ac DG |
8130 | if (!parent) |
8131 | se->cfs_rq = &rq->cfs; | |
8132 | else | |
8133 | se->cfs_rq = parent->my_q; | |
8134 | ||
6f505b16 | 8135 | se->my_q = cfs_rq; |
9437178f | 8136 | update_load_set(&se->load, 0); |
ec7dc8ac | 8137 | se->parent = parent; |
6f505b16 | 8138 | } |
052f1dc7 | 8139 | #endif |
6f505b16 | 8140 | |
052f1dc7 | 8141 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac | 8142 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
3d4b47b4 | 8143 | struct sched_rt_entity *rt_se, int cpu, |
ec7dc8ac | 8144 | struct sched_rt_entity *parent) |
6f505b16 | 8145 | { |
ec7dc8ac DG |
8146 | struct rq *rq = cpu_rq(cpu); |
8147 | ||
acb5a9ba JS |
8148 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
8149 | rt_rq->rt_nr_boosted = 0; | |
8150 | rt_rq->rq = rq; | |
6f505b16 | 8151 | rt_rq->tg = tg; |
6f505b16 | 8152 | |
acb5a9ba | 8153 | tg->rt_rq[cpu] = rt_rq; |
6f505b16 | 8154 | tg->rt_se[cpu] = rt_se; |
acb5a9ba | 8155 | |
354d60c2 DG |
8156 | if (!rt_se) |
8157 | return; | |
8158 | ||
ec7dc8ac DG |
8159 | if (!parent) |
8160 | rt_se->rt_rq = &rq->rt; | |
8161 | else | |
8162 | rt_se->rt_rq = parent->my_q; | |
8163 | ||
6f505b16 | 8164 | rt_se->my_q = rt_rq; |
ec7dc8ac | 8165 | rt_se->parent = parent; |
6f505b16 PZ |
8166 | INIT_LIST_HEAD(&rt_se->run_list); |
8167 | } | |
8168 | #endif | |
8169 | ||
1da177e4 LT |
8170 | void __init sched_init(void) |
8171 | { | |
dd41f596 | 8172 | int i, j; |
434d53b0 MT |
8173 | unsigned long alloc_size = 0, ptr; |
8174 | ||
8175 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8176 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
8177 | #endif | |
8178 | #ifdef CONFIG_RT_GROUP_SCHED | |
8179 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 8180 | #endif |
df7c8e84 | 8181 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 8182 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 8183 | #endif |
434d53b0 | 8184 | if (alloc_size) { |
36b7b6d4 | 8185 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
8186 | |
8187 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 8188 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
8189 | ptr += nr_cpu_ids * sizeof(void **); |
8190 | ||
07e06b01 | 8191 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 8192 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 8193 | |
6d6bc0ad | 8194 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 8195 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 8196 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
8197 | ptr += nr_cpu_ids * sizeof(void **); |
8198 | ||
07e06b01 | 8199 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
8200 | ptr += nr_cpu_ids * sizeof(void **); |
8201 | ||
6d6bc0ad | 8202 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
8203 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8204 | for_each_possible_cpu(i) { | |
8205 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
8206 | ptr += cpumask_size(); | |
8207 | } | |
8208 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 8209 | } |
dd41f596 | 8210 | |
57d885fe GH |
8211 | #ifdef CONFIG_SMP |
8212 | init_defrootdomain(); | |
8213 | #endif | |
8214 | ||
d0b27fa7 PZ |
8215 | init_rt_bandwidth(&def_rt_bandwidth, |
8216 | global_rt_period(), global_rt_runtime()); | |
8217 | ||
8218 | #ifdef CONFIG_RT_GROUP_SCHED | |
07e06b01 | 8219 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 8220 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 8221 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 8222 | |
7c941438 | 8223 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
8224 | list_add(&root_task_group.list, &task_groups); |
8225 | INIT_LIST_HEAD(&root_task_group.children); | |
5091faa4 | 8226 | autogroup_init(&init_task); |
7c941438 | 8227 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 8228 | |
0a945022 | 8229 | for_each_possible_cpu(i) { |
70b97a7f | 8230 | struct rq *rq; |
1da177e4 LT |
8231 | |
8232 | rq = cpu_rq(i); | |
05fa785c | 8233 | raw_spin_lock_init(&rq->lock); |
7897986b | 8234 | rq->nr_running = 0; |
dce48a84 TG |
8235 | rq->calc_load_active = 0; |
8236 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 8237 | init_cfs_rq(&rq->cfs); |
6f505b16 | 8238 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 8239 | #ifdef CONFIG_FAIR_GROUP_SCHED |
07e06b01 | 8240 | root_task_group.shares = root_task_group_load; |
6f505b16 | 8241 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 8242 | /* |
07e06b01 | 8243 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
8244 | * |
8245 | * In case of task-groups formed thr' the cgroup filesystem, it | |
8246 | * gets 100% of the cpu resources in the system. This overall | |
8247 | * system cpu resource is divided among the tasks of | |
07e06b01 | 8248 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
8249 | * based on each entity's (task or task-group's) weight |
8250 | * (se->load.weight). | |
8251 | * | |
07e06b01 | 8252 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
8253 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
8254 | * then A0's share of the cpu resource is: | |
8255 | * | |
0d905bca | 8256 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 8257 | * |
07e06b01 YZ |
8258 | * We achieve this by letting root_task_group's tasks sit |
8259 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 8260 | */ |
ab84d31e | 8261 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 8262 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
8263 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8264 | ||
8265 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8266 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8267 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
07e06b01 | 8268 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 8269 | #endif |
1da177e4 | 8270 | |
dd41f596 IM |
8271 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
8272 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
8273 | |
8274 | rq->last_load_update_tick = jiffies; | |
8275 | ||
1da177e4 | 8276 | #ifdef CONFIG_SMP |
41c7ce9a | 8277 | rq->sd = NULL; |
57d885fe | 8278 | rq->rd = NULL; |
1399fa78 | 8279 | rq->cpu_power = SCHED_POWER_SCALE; |
3f029d3c | 8280 | rq->post_schedule = 0; |
1da177e4 | 8281 | rq->active_balance = 0; |
dd41f596 | 8282 | rq->next_balance = jiffies; |
1da177e4 | 8283 | rq->push_cpu = 0; |
0a2966b4 | 8284 | rq->cpu = i; |
1f11eb6a | 8285 | rq->online = 0; |
eae0c9df MG |
8286 | rq->idle_stamp = 0; |
8287 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
dc938520 | 8288 | rq_attach_root(rq, &def_root_domain); |
83cd4fe2 VP |
8289 | #ifdef CONFIG_NO_HZ |
8290 | rq->nohz_balance_kick = 0; | |
8291 | init_sched_softirq_csd(&per_cpu(remote_sched_softirq_cb, i)); | |
8292 | #endif | |
1da177e4 | 8293 | #endif |
8f4d37ec | 8294 | init_rq_hrtick(rq); |
1da177e4 | 8295 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8296 | } |
8297 | ||
2dd73a4f | 8298 | set_load_weight(&init_task); |
b50f60ce | 8299 | |
e107be36 AK |
8300 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8301 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8302 | #endif | |
8303 | ||
c9819f45 | 8304 | #ifdef CONFIG_SMP |
962cf36c | 8305 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
8306 | #endif |
8307 | ||
b50f60ce | 8308 | #ifdef CONFIG_RT_MUTEXES |
732375c6 | 8309 | plist_head_init(&init_task.pi_waiters); |
b50f60ce HC |
8310 | #endif |
8311 | ||
1da177e4 LT |
8312 | /* |
8313 | * The boot idle thread does lazy MMU switching as well: | |
8314 | */ | |
8315 | atomic_inc(&init_mm.mm_count); | |
8316 | enter_lazy_tlb(&init_mm, current); | |
8317 | ||
8318 | /* | |
8319 | * Make us the idle thread. Technically, schedule() should not be | |
8320 | * called from this thread, however somewhere below it might be, | |
8321 | * but because we are the idle thread, we just pick up running again | |
8322 | * when this runqueue becomes "idle". | |
8323 | */ | |
8324 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
8325 | |
8326 | calc_load_update = jiffies + LOAD_FREQ; | |
8327 | ||
dd41f596 IM |
8328 | /* |
8329 | * During early bootup we pretend to be a normal task: | |
8330 | */ | |
8331 | current->sched_class = &fair_sched_class; | |
6892b75e | 8332 | |
6a7b3dc3 | 8333 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 8334 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 8335 | #ifdef CONFIG_SMP |
4cb98839 | 8336 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
7d1e6a9b | 8337 | #ifdef CONFIG_NO_HZ |
83cd4fe2 VP |
8338 | zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT); |
8339 | alloc_cpumask_var(&nohz.grp_idle_mask, GFP_NOWAIT); | |
8340 | atomic_set(&nohz.load_balancer, nr_cpu_ids); | |
8341 | atomic_set(&nohz.first_pick_cpu, nr_cpu_ids); | |
8342 | atomic_set(&nohz.second_pick_cpu, nr_cpu_ids); | |
7d1e6a9b | 8343 | #endif |
bdddd296 RR |
8344 | /* May be allocated at isolcpus cmdline parse time */ |
8345 | if (cpu_isolated_map == NULL) | |
8346 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 8347 | #endif /* SMP */ |
6a7b3dc3 | 8348 | |
6892b75e | 8349 | scheduler_running = 1; |
1da177e4 LT |
8350 | } |
8351 | ||
d902db1e | 8352 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
8353 | static inline int preempt_count_equals(int preempt_offset) |
8354 | { | |
234da7bc | 8355 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 8356 | |
4ba8216c | 8357 | return (nested == preempt_offset); |
e4aafea2 FW |
8358 | } |
8359 | ||
d894837f | 8360 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 8361 | { |
1da177e4 LT |
8362 | static unsigned long prev_jiffy; /* ratelimiting */ |
8363 | ||
e4aafea2 FW |
8364 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
8365 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
8366 | return; |
8367 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8368 | return; | |
8369 | prev_jiffy = jiffies; | |
8370 | ||
3df0fc5b PZ |
8371 | printk(KERN_ERR |
8372 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
8373 | file, line); | |
8374 | printk(KERN_ERR | |
8375 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
8376 | in_atomic(), irqs_disabled(), | |
8377 | current->pid, current->comm); | |
aef745fc IM |
8378 | |
8379 | debug_show_held_locks(current); | |
8380 | if (irqs_disabled()) | |
8381 | print_irqtrace_events(current); | |
8382 | dump_stack(); | |
1da177e4 LT |
8383 | } |
8384 | EXPORT_SYMBOL(__might_sleep); | |
8385 | #endif | |
8386 | ||
8387 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8388 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8389 | { | |
da7a735e PZ |
8390 | const struct sched_class *prev_class = p->sched_class; |
8391 | int old_prio = p->prio; | |
3a5e4dc1 | 8392 | int on_rq; |
3e51f33f | 8393 | |
fd2f4419 | 8394 | on_rq = p->on_rq; |
3a5e4dc1 AK |
8395 | if (on_rq) |
8396 | deactivate_task(rq, p, 0); | |
8397 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8398 | if (on_rq) { | |
8399 | activate_task(rq, p, 0); | |
8400 | resched_task(rq->curr); | |
8401 | } | |
da7a735e PZ |
8402 | |
8403 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
8404 | } |
8405 | ||
1da177e4 LT |
8406 | void normalize_rt_tasks(void) |
8407 | { | |
a0f98a1c | 8408 | struct task_struct *g, *p; |
1da177e4 | 8409 | unsigned long flags; |
70b97a7f | 8410 | struct rq *rq; |
1da177e4 | 8411 | |
4cf5d77a | 8412 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8413 | do_each_thread(g, p) { |
178be793 IM |
8414 | /* |
8415 | * Only normalize user tasks: | |
8416 | */ | |
8417 | if (!p->mm) | |
8418 | continue; | |
8419 | ||
6cfb0d5d | 8420 | p->se.exec_start = 0; |
6cfb0d5d | 8421 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
8422 | p->se.statistics.wait_start = 0; |
8423 | p->se.statistics.sleep_start = 0; | |
8424 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 8425 | #endif |
dd41f596 IM |
8426 | |
8427 | if (!rt_task(p)) { | |
8428 | /* | |
8429 | * Renice negative nice level userspace | |
8430 | * tasks back to 0: | |
8431 | */ | |
8432 | if (TASK_NICE(p) < 0 && p->mm) | |
8433 | set_user_nice(p, 0); | |
1da177e4 | 8434 | continue; |
dd41f596 | 8435 | } |
1da177e4 | 8436 | |
1d615482 | 8437 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 8438 | rq = __task_rq_lock(p); |
1da177e4 | 8439 | |
178be793 | 8440 | normalize_task(rq, p); |
3a5e4dc1 | 8441 | |
b29739f9 | 8442 | __task_rq_unlock(rq); |
1d615482 | 8443 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8444 | } while_each_thread(g, p); |
8445 | ||
4cf5d77a | 8446 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8447 | } |
8448 | ||
8449 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 8450 | |
67fc4e0c | 8451 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 8452 | /* |
67fc4e0c | 8453 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
8454 | * |
8455 | * They can only be called when the whole system has been | |
8456 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8457 | * activity can take place. Using them for anything else would | |
8458 | * be a serious bug, and as a result, they aren't even visible | |
8459 | * under any other configuration. | |
8460 | */ | |
8461 | ||
8462 | /** | |
8463 | * curr_task - return the current task for a given cpu. | |
8464 | * @cpu: the processor in question. | |
8465 | * | |
8466 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8467 | */ | |
36c8b586 | 8468 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8469 | { |
8470 | return cpu_curr(cpu); | |
8471 | } | |
8472 | ||
67fc4e0c JW |
8473 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
8474 | ||
8475 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
8476 | /** |
8477 | * set_curr_task - set the current task for a given cpu. | |
8478 | * @cpu: the processor in question. | |
8479 | * @p: the task pointer to set. | |
8480 | * | |
8481 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8482 | * are serviced on a separate stack. It allows the architecture to switch the |
8483 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8484 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8485 | * and caller must save the original value of the current task (see | |
8486 | * curr_task() above) and restore that value before reenabling interrupts and | |
8487 | * re-starting the system. | |
8488 | * | |
8489 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8490 | */ | |
36c8b586 | 8491 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8492 | { |
8493 | cpu_curr(cpu) = p; | |
8494 | } | |
8495 | ||
8496 | #endif | |
29f59db3 | 8497 | |
bccbe08a PZ |
8498 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8499 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8500 | { |
8501 | int i; | |
8502 | ||
ab84d31e PT |
8503 | destroy_cfs_bandwidth(tg_cfs_bandwidth(tg)); |
8504 | ||
6f505b16 PZ |
8505 | for_each_possible_cpu(i) { |
8506 | if (tg->cfs_rq) | |
8507 | kfree(tg->cfs_rq[i]); | |
8508 | if (tg->se) | |
8509 | kfree(tg->se[i]); | |
6f505b16 PZ |
8510 | } |
8511 | ||
8512 | kfree(tg->cfs_rq); | |
8513 | kfree(tg->se); | |
6f505b16 PZ |
8514 | } |
8515 | ||
ec7dc8ac DG |
8516 | static |
8517 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8518 | { |
29f59db3 | 8519 | struct cfs_rq *cfs_rq; |
eab17229 | 8520 | struct sched_entity *se; |
29f59db3 SV |
8521 | int i; |
8522 | ||
434d53b0 | 8523 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8524 | if (!tg->cfs_rq) |
8525 | goto err; | |
434d53b0 | 8526 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8527 | if (!tg->se) |
8528 | goto err; | |
052f1dc7 PZ |
8529 | |
8530 | tg->shares = NICE_0_LOAD; | |
29f59db3 | 8531 | |
ab84d31e PT |
8532 | init_cfs_bandwidth(tg_cfs_bandwidth(tg)); |
8533 | ||
29f59db3 | 8534 | for_each_possible_cpu(i) { |
eab17229 LZ |
8535 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
8536 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
8537 | if (!cfs_rq) |
8538 | goto err; | |
8539 | ||
eab17229 LZ |
8540 | se = kzalloc_node(sizeof(struct sched_entity), |
8541 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 8542 | if (!se) |
dfc12eb2 | 8543 | goto err_free_rq; |
29f59db3 | 8544 | |
acb5a9ba | 8545 | init_cfs_rq(cfs_rq); |
3d4b47b4 | 8546 | init_tg_cfs_entry(tg, cfs_rq, se, i, parent->se[i]); |
bccbe08a PZ |
8547 | } |
8548 | ||
8549 | return 1; | |
8550 | ||
49246274 | 8551 | err_free_rq: |
dfc12eb2 | 8552 | kfree(cfs_rq); |
49246274 | 8553 | err: |
bccbe08a PZ |
8554 | return 0; |
8555 | } | |
8556 | ||
bccbe08a PZ |
8557 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8558 | { | |
3d4b47b4 PZ |
8559 | struct rq *rq = cpu_rq(cpu); |
8560 | unsigned long flags; | |
3d4b47b4 PZ |
8561 | |
8562 | /* | |
8563 | * Only empty task groups can be destroyed; so we can speculatively | |
8564 | * check on_list without danger of it being re-added. | |
8565 | */ | |
8566 | if (!tg->cfs_rq[cpu]->on_list) | |
8567 | return; | |
8568 | ||
8569 | raw_spin_lock_irqsave(&rq->lock, flags); | |
822bc180 | 8570 | list_del_leaf_cfs_rq(tg->cfs_rq[cpu]); |
3d4b47b4 | 8571 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
bccbe08a | 8572 | } |
5f817d67 | 8573 | #else /* !CONFIG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8574 | static inline void free_fair_sched_group(struct task_group *tg) |
8575 | { | |
8576 | } | |
8577 | ||
ec7dc8ac DG |
8578 | static inline |
8579 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8580 | { |
8581 | return 1; | |
8582 | } | |
8583 | ||
bccbe08a PZ |
8584 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8585 | { | |
8586 | } | |
6d6bc0ad | 8587 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8588 | |
8589 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8590 | static void free_rt_sched_group(struct task_group *tg) |
8591 | { | |
8592 | int i; | |
8593 | ||
99bc5242 BL |
8594 | if (tg->rt_se) |
8595 | destroy_rt_bandwidth(&tg->rt_bandwidth); | |
d0b27fa7 | 8596 | |
bccbe08a PZ |
8597 | for_each_possible_cpu(i) { |
8598 | if (tg->rt_rq) | |
8599 | kfree(tg->rt_rq[i]); | |
8600 | if (tg->rt_se) | |
8601 | kfree(tg->rt_se[i]); | |
8602 | } | |
8603 | ||
8604 | kfree(tg->rt_rq); | |
8605 | kfree(tg->rt_se); | |
8606 | } | |
8607 | ||
ec7dc8ac DG |
8608 | static |
8609 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8610 | { |
8611 | struct rt_rq *rt_rq; | |
eab17229 | 8612 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8613 | int i; |
8614 | ||
434d53b0 | 8615 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8616 | if (!tg->rt_rq) |
8617 | goto err; | |
434d53b0 | 8618 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8619 | if (!tg->rt_se) |
8620 | goto err; | |
8621 | ||
d0b27fa7 PZ |
8622 | init_rt_bandwidth(&tg->rt_bandwidth, |
8623 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8624 | |
8625 | for_each_possible_cpu(i) { | |
eab17229 LZ |
8626 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8627 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8628 | if (!rt_rq) |
8629 | goto err; | |
29f59db3 | 8630 | |
eab17229 LZ |
8631 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8632 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 8633 | if (!rt_se) |
dfc12eb2 | 8634 | goto err_free_rq; |
29f59db3 | 8635 | |
acb5a9ba JS |
8636 | init_rt_rq(rt_rq, cpu_rq(i)); |
8637 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; | |
3d4b47b4 | 8638 | init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]); |
29f59db3 SV |
8639 | } |
8640 | ||
bccbe08a PZ |
8641 | return 1; |
8642 | ||
49246274 | 8643 | err_free_rq: |
dfc12eb2 | 8644 | kfree(rt_rq); |
49246274 | 8645 | err: |
bccbe08a PZ |
8646 | return 0; |
8647 | } | |
6d6bc0ad | 8648 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8649 | static inline void free_rt_sched_group(struct task_group *tg) |
8650 | { | |
8651 | } | |
8652 | ||
ec7dc8ac DG |
8653 | static inline |
8654 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8655 | { |
8656 | return 1; | |
8657 | } | |
6d6bc0ad | 8658 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8659 | |
7c941438 | 8660 | #ifdef CONFIG_CGROUP_SCHED |
bccbe08a PZ |
8661 | static void free_sched_group(struct task_group *tg) |
8662 | { | |
8663 | free_fair_sched_group(tg); | |
8664 | free_rt_sched_group(tg); | |
e9aa1dd1 | 8665 | autogroup_free(tg); |
bccbe08a PZ |
8666 | kfree(tg); |
8667 | } | |
8668 | ||
8669 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8670 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8671 | { |
8672 | struct task_group *tg; | |
8673 | unsigned long flags; | |
bccbe08a PZ |
8674 | |
8675 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8676 | if (!tg) | |
8677 | return ERR_PTR(-ENOMEM); | |
8678 | ||
ec7dc8ac | 8679 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8680 | goto err; |
8681 | ||
ec7dc8ac | 8682 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8683 | goto err; |
8684 | ||
8ed36996 | 8685 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 8686 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8687 | |
8688 | WARN_ON(!parent); /* root should already exist */ | |
8689 | ||
8690 | tg->parent = parent; | |
f473aa5e | 8691 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8692 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8693 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8694 | |
9b5b7751 | 8695 | return tg; |
29f59db3 SV |
8696 | |
8697 | err: | |
6f505b16 | 8698 | free_sched_group(tg); |
29f59db3 SV |
8699 | return ERR_PTR(-ENOMEM); |
8700 | } | |
8701 | ||
9b5b7751 | 8702 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8703 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8704 | { |
29f59db3 | 8705 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8706 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8707 | } |
8708 | ||
9b5b7751 | 8709 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8710 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8711 | { |
8ed36996 | 8712 | unsigned long flags; |
9b5b7751 | 8713 | int i; |
29f59db3 | 8714 | |
3d4b47b4 PZ |
8715 | /* end participation in shares distribution */ |
8716 | for_each_possible_cpu(i) | |
bccbe08a | 8717 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
8718 | |
8719 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 8720 | list_del_rcu(&tg->list); |
f473aa5e | 8721 | list_del_rcu(&tg->siblings); |
8ed36996 | 8722 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8723 | |
9b5b7751 | 8724 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8725 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8726 | } |
8727 | ||
9b5b7751 | 8728 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8729 | * The caller of this function should have put the task in its new group |
8730 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8731 | * reflect its new group. | |
9b5b7751 SV |
8732 | */ |
8733 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8734 | { |
8735 | int on_rq, running; | |
8736 | unsigned long flags; | |
8737 | struct rq *rq; | |
8738 | ||
8739 | rq = task_rq_lock(tsk, &flags); | |
8740 | ||
051a1d1a | 8741 | running = task_current(rq, tsk); |
fd2f4419 | 8742 | on_rq = tsk->on_rq; |
29f59db3 | 8743 | |
0e1f3483 | 8744 | if (on_rq) |
29f59db3 | 8745 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8746 | if (unlikely(running)) |
8747 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8748 | |
810b3817 | 8749 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
8750 | if (tsk->sched_class->task_move_group) |
8751 | tsk->sched_class->task_move_group(tsk, on_rq); | |
8752 | else | |
810b3817 | 8753 | #endif |
b2b5ce02 | 8754 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 8755 | |
0e1f3483 HS |
8756 | if (unlikely(running)) |
8757 | tsk->sched_class->set_curr_task(rq); | |
8758 | if (on_rq) | |
371fd7e7 | 8759 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8760 | |
0122ec5b | 8761 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 8762 | } |
7c941438 | 8763 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8764 | |
052f1dc7 | 8765 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8ed36996 PZ |
8766 | static DEFINE_MUTEX(shares_mutex); |
8767 | ||
4cf86d77 | 8768 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8769 | { |
8770 | int i; | |
8ed36996 | 8771 | unsigned long flags; |
c61935fd | 8772 | |
ec7dc8ac DG |
8773 | /* |
8774 | * We can't change the weight of the root cgroup. | |
8775 | */ | |
8776 | if (!tg->se[0]) | |
8777 | return -EINVAL; | |
8778 | ||
cd62287e | 8779 | shares = clamp(shares, scale_load(MIN_SHARES), scale_load(MAX_SHARES)); |
62fb1851 | 8780 | |
8ed36996 | 8781 | mutex_lock(&shares_mutex); |
9b5b7751 | 8782 | if (tg->shares == shares) |
5cb350ba | 8783 | goto done; |
29f59db3 | 8784 | |
9b5b7751 | 8785 | tg->shares = shares; |
c09595f6 | 8786 | for_each_possible_cpu(i) { |
9437178f PT |
8787 | struct rq *rq = cpu_rq(i); |
8788 | struct sched_entity *se; | |
8789 | ||
8790 | se = tg->se[i]; | |
8791 | /* Propagate contribution to hierarchy */ | |
8792 | raw_spin_lock_irqsave(&rq->lock, flags); | |
8793 | for_each_sched_entity(se) | |
6d5ab293 | 8794 | update_cfs_shares(group_cfs_rq(se)); |
9437178f | 8795 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c09595f6 | 8796 | } |
29f59db3 | 8797 | |
5cb350ba | 8798 | done: |
8ed36996 | 8799 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8800 | return 0; |
29f59db3 SV |
8801 | } |
8802 | ||
5cb350ba DG |
8803 | unsigned long sched_group_shares(struct task_group *tg) |
8804 | { | |
8805 | return tg->shares; | |
8806 | } | |
052f1dc7 | 8807 | #endif |
5cb350ba | 8808 | |
a790de99 | 8809 | #if defined(CONFIG_RT_GROUP_SCHED) || defined(CONFIG_CFS_BANDWIDTH) |
9f0c1e56 PZ |
8810 | static unsigned long to_ratio(u64 period, u64 runtime) |
8811 | { | |
8812 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8813 | return 1ULL << 20; |
9f0c1e56 | 8814 | |
9a7e0b18 | 8815 | return div64_u64(runtime << 20, period); |
9f0c1e56 | 8816 | } |
a790de99 PT |
8817 | #endif |
8818 | ||
8819 | #ifdef CONFIG_RT_GROUP_SCHED | |
8820 | /* | |
8821 | * Ensure that the real time constraints are schedulable. | |
8822 | */ | |
8823 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 8824 | |
9a7e0b18 PZ |
8825 | /* Must be called with tasklist_lock held */ |
8826 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8827 | { |
9a7e0b18 | 8828 | struct task_struct *g, *p; |
b40b2e8e | 8829 | |
9a7e0b18 PZ |
8830 | do_each_thread(g, p) { |
8831 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8832 | return 1; | |
8833 | } while_each_thread(g, p); | |
b40b2e8e | 8834 | |
9a7e0b18 PZ |
8835 | return 0; |
8836 | } | |
b40b2e8e | 8837 | |
9a7e0b18 PZ |
8838 | struct rt_schedulable_data { |
8839 | struct task_group *tg; | |
8840 | u64 rt_period; | |
8841 | u64 rt_runtime; | |
8842 | }; | |
b40b2e8e | 8843 | |
a790de99 | 8844 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
8845 | { |
8846 | struct rt_schedulable_data *d = data; | |
8847 | struct task_group *child; | |
8848 | unsigned long total, sum = 0; | |
8849 | u64 period, runtime; | |
b40b2e8e | 8850 | |
9a7e0b18 PZ |
8851 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8852 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8853 | |
9a7e0b18 PZ |
8854 | if (tg == d->tg) { |
8855 | period = d->rt_period; | |
8856 | runtime = d->rt_runtime; | |
b40b2e8e | 8857 | } |
b40b2e8e | 8858 | |
4653f803 PZ |
8859 | /* |
8860 | * Cannot have more runtime than the period. | |
8861 | */ | |
8862 | if (runtime > period && runtime != RUNTIME_INF) | |
8863 | return -EINVAL; | |
6f505b16 | 8864 | |
4653f803 PZ |
8865 | /* |
8866 | * Ensure we don't starve existing RT tasks. | |
8867 | */ | |
9a7e0b18 PZ |
8868 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8869 | return -EBUSY; | |
6f505b16 | 8870 | |
9a7e0b18 | 8871 | total = to_ratio(period, runtime); |
6f505b16 | 8872 | |
4653f803 PZ |
8873 | /* |
8874 | * Nobody can have more than the global setting allows. | |
8875 | */ | |
8876 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8877 | return -EINVAL; | |
6f505b16 | 8878 | |
4653f803 PZ |
8879 | /* |
8880 | * The sum of our children's runtime should not exceed our own. | |
8881 | */ | |
9a7e0b18 PZ |
8882 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8883 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8884 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8885 | |
9a7e0b18 PZ |
8886 | if (child == d->tg) { |
8887 | period = d->rt_period; | |
8888 | runtime = d->rt_runtime; | |
8889 | } | |
6f505b16 | 8890 | |
9a7e0b18 | 8891 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8892 | } |
6f505b16 | 8893 | |
9a7e0b18 PZ |
8894 | if (sum > total) |
8895 | return -EINVAL; | |
8896 | ||
8897 | return 0; | |
6f505b16 PZ |
8898 | } |
8899 | ||
9a7e0b18 | 8900 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8901 | { |
8277434e PT |
8902 | int ret; |
8903 | ||
9a7e0b18 PZ |
8904 | struct rt_schedulable_data data = { |
8905 | .tg = tg, | |
8906 | .rt_period = period, | |
8907 | .rt_runtime = runtime, | |
8908 | }; | |
8909 | ||
8277434e PT |
8910 | rcu_read_lock(); |
8911 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
8912 | rcu_read_unlock(); | |
8913 | ||
8914 | return ret; | |
521f1a24 DG |
8915 | } |
8916 | ||
ab84d31e | 8917 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 8918 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 8919 | { |
ac086bc2 | 8920 | int i, err = 0; |
9f0c1e56 | 8921 | |
9f0c1e56 | 8922 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8923 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8924 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8925 | if (err) | |
9f0c1e56 | 8926 | goto unlock; |
ac086bc2 | 8927 | |
0986b11b | 8928 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8929 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8930 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8931 | |
8932 | for_each_possible_cpu(i) { | |
8933 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8934 | ||
0986b11b | 8935 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8936 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8937 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8938 | } |
0986b11b | 8939 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 8940 | unlock: |
521f1a24 | 8941 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8942 | mutex_unlock(&rt_constraints_mutex); |
8943 | ||
8944 | return err; | |
6f505b16 PZ |
8945 | } |
8946 | ||
d0b27fa7 PZ |
8947 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8948 | { | |
8949 | u64 rt_runtime, rt_period; | |
8950 | ||
8951 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8952 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8953 | if (rt_runtime_us < 0) | |
8954 | rt_runtime = RUNTIME_INF; | |
8955 | ||
ab84d31e | 8956 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
8957 | } |
8958 | ||
9f0c1e56 PZ |
8959 | long sched_group_rt_runtime(struct task_group *tg) |
8960 | { | |
8961 | u64 rt_runtime_us; | |
8962 | ||
d0b27fa7 | 8963 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8964 | return -1; |
8965 | ||
d0b27fa7 | 8966 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8967 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8968 | return rt_runtime_us; | |
8969 | } | |
d0b27fa7 PZ |
8970 | |
8971 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8972 | { | |
8973 | u64 rt_runtime, rt_period; | |
8974 | ||
8975 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8976 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8977 | ||
619b0488 R |
8978 | if (rt_period == 0) |
8979 | return -EINVAL; | |
8980 | ||
ab84d31e | 8981 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
8982 | } |
8983 | ||
8984 | long sched_group_rt_period(struct task_group *tg) | |
8985 | { | |
8986 | u64 rt_period_us; | |
8987 | ||
8988 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8989 | do_div(rt_period_us, NSEC_PER_USEC); | |
8990 | return rt_period_us; | |
8991 | } | |
8992 | ||
8993 | static int sched_rt_global_constraints(void) | |
8994 | { | |
4653f803 | 8995 | u64 runtime, period; |
d0b27fa7 PZ |
8996 | int ret = 0; |
8997 | ||
ec5d4989 HS |
8998 | if (sysctl_sched_rt_period <= 0) |
8999 | return -EINVAL; | |
9000 | ||
4653f803 PZ |
9001 | runtime = global_rt_runtime(); |
9002 | period = global_rt_period(); | |
9003 | ||
9004 | /* | |
9005 | * Sanity check on the sysctl variables. | |
9006 | */ | |
9007 | if (runtime > period && runtime != RUNTIME_INF) | |
9008 | return -EINVAL; | |
10b612f4 | 9009 | |
d0b27fa7 | 9010 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 9011 | read_lock(&tasklist_lock); |
4653f803 | 9012 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 9013 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
9014 | mutex_unlock(&rt_constraints_mutex); |
9015 | ||
9016 | return ret; | |
9017 | } | |
54e99124 DG |
9018 | |
9019 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
9020 | { | |
9021 | /* Don't accept realtime tasks when there is no way for them to run */ | |
9022 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
9023 | return 0; | |
9024 | ||
9025 | return 1; | |
9026 | } | |
9027 | ||
6d6bc0ad | 9028 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9029 | static int sched_rt_global_constraints(void) |
9030 | { | |
ac086bc2 PZ |
9031 | unsigned long flags; |
9032 | int i; | |
9033 | ||
ec5d4989 HS |
9034 | if (sysctl_sched_rt_period <= 0) |
9035 | return -EINVAL; | |
9036 | ||
60aa605d PZ |
9037 | /* |
9038 | * There's always some RT tasks in the root group | |
9039 | * -- migration, kstopmachine etc.. | |
9040 | */ | |
9041 | if (sysctl_sched_rt_runtime == 0) | |
9042 | return -EBUSY; | |
9043 | ||
0986b11b | 9044 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
9045 | for_each_possible_cpu(i) { |
9046 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
9047 | ||
0986b11b | 9048 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 9049 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 9050 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 9051 | } |
0986b11b | 9052 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 9053 | |
d0b27fa7 PZ |
9054 | return 0; |
9055 | } | |
6d6bc0ad | 9056 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9057 | |
9058 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 9059 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
9060 | loff_t *ppos) |
9061 | { | |
9062 | int ret; | |
9063 | int old_period, old_runtime; | |
9064 | static DEFINE_MUTEX(mutex); | |
9065 | ||
9066 | mutex_lock(&mutex); | |
9067 | old_period = sysctl_sched_rt_period; | |
9068 | old_runtime = sysctl_sched_rt_runtime; | |
9069 | ||
8d65af78 | 9070 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
9071 | |
9072 | if (!ret && write) { | |
9073 | ret = sched_rt_global_constraints(); | |
9074 | if (ret) { | |
9075 | sysctl_sched_rt_period = old_period; | |
9076 | sysctl_sched_rt_runtime = old_runtime; | |
9077 | } else { | |
9078 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
9079 | def_rt_bandwidth.rt_period = | |
9080 | ns_to_ktime(global_rt_period()); | |
9081 | } | |
9082 | } | |
9083 | mutex_unlock(&mutex); | |
9084 | ||
9085 | return ret; | |
9086 | } | |
68318b8e | 9087 | |
052f1dc7 | 9088 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
9089 | |
9090 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 9091 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 9092 | { |
2b01dfe3 PM |
9093 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
9094 | struct task_group, css); | |
68318b8e SV |
9095 | } |
9096 | ||
9097 | static struct cgroup_subsys_state * | |
2b01dfe3 | 9098 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 9099 | { |
ec7dc8ac | 9100 | struct task_group *tg, *parent; |
68318b8e | 9101 | |
2b01dfe3 | 9102 | if (!cgrp->parent) { |
68318b8e | 9103 | /* This is early initialization for the top cgroup */ |
07e06b01 | 9104 | return &root_task_group.css; |
68318b8e SV |
9105 | } |
9106 | ||
ec7dc8ac DG |
9107 | parent = cgroup_tg(cgrp->parent); |
9108 | tg = sched_create_group(parent); | |
68318b8e SV |
9109 | if (IS_ERR(tg)) |
9110 | return ERR_PTR(-ENOMEM); | |
9111 | ||
68318b8e SV |
9112 | return &tg->css; |
9113 | } | |
9114 | ||
41a2d6cf IM |
9115 | static void |
9116 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 9117 | { |
2b01dfe3 | 9118 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9119 | |
9120 | sched_destroy_group(tg); | |
9121 | } | |
9122 | ||
41a2d6cf | 9123 | static int |
be367d09 | 9124 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 9125 | { |
b68aa230 | 9126 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 9127 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
9128 | return -EINVAL; |
9129 | #else | |
68318b8e SV |
9130 | /* We don't support RT-tasks being in separate groups */ |
9131 | if (tsk->sched_class != &fair_sched_class) | |
9132 | return -EINVAL; | |
b68aa230 | 9133 | #endif |
be367d09 BB |
9134 | return 0; |
9135 | } | |
68318b8e | 9136 | |
68318b8e | 9137 | static void |
f780bdb7 | 9138 | cpu_cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e SV |
9139 | { |
9140 | sched_move_task(tsk); | |
9141 | } | |
9142 | ||
068c5cc5 | 9143 | static void |
d41d5a01 PZ |
9144 | cpu_cgroup_exit(struct cgroup_subsys *ss, struct cgroup *cgrp, |
9145 | struct cgroup *old_cgrp, struct task_struct *task) | |
068c5cc5 PZ |
9146 | { |
9147 | /* | |
9148 | * cgroup_exit() is called in the copy_process() failure path. | |
9149 | * Ignore this case since the task hasn't ran yet, this avoids | |
9150 | * trying to poke a half freed task state from generic code. | |
9151 | */ | |
9152 | if (!(task->flags & PF_EXITING)) | |
9153 | return; | |
9154 | ||
9155 | sched_move_task(task); | |
9156 | } | |
9157 | ||
052f1dc7 | 9158 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 9159 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 9160 | u64 shareval) |
68318b8e | 9161 | { |
c8b28116 | 9162 | return sched_group_set_shares(cgroup_tg(cgrp), scale_load(shareval)); |
68318b8e SV |
9163 | } |
9164 | ||
f4c753b7 | 9165 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 9166 | { |
2b01dfe3 | 9167 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e | 9168 | |
c8b28116 | 9169 | return (u64) scale_load_down(tg->shares); |
68318b8e | 9170 | } |
ab84d31e PT |
9171 | |
9172 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
9173 | static DEFINE_MUTEX(cfs_constraints_mutex); |
9174 | ||
ab84d31e PT |
9175 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
9176 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
9177 | ||
a790de99 PT |
9178 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
9179 | ||
ab84d31e PT |
9180 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
9181 | { | |
58088ad0 | 9182 | int i, ret = 0, runtime_enabled; |
ab84d31e | 9183 | struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(tg); |
ab84d31e PT |
9184 | |
9185 | if (tg == &root_task_group) | |
9186 | return -EINVAL; | |
9187 | ||
9188 | /* | |
9189 | * Ensure we have at some amount of bandwidth every period. This is | |
9190 | * to prevent reaching a state of large arrears when throttled via | |
9191 | * entity_tick() resulting in prolonged exit starvation. | |
9192 | */ | |
9193 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
9194 | return -EINVAL; | |
9195 | ||
9196 | /* | |
9197 | * Likewise, bound things on the otherside by preventing insane quota | |
9198 | * periods. This also allows us to normalize in computing quota | |
9199 | * feasibility. | |
9200 | */ | |
9201 | if (period > max_cfs_quota_period) | |
9202 | return -EINVAL; | |
9203 | ||
a790de99 PT |
9204 | mutex_lock(&cfs_constraints_mutex); |
9205 | ret = __cfs_schedulable(tg, period, quota); | |
9206 | if (ret) | |
9207 | goto out_unlock; | |
9208 | ||
58088ad0 | 9209 | runtime_enabled = quota != RUNTIME_INF; |
ab84d31e PT |
9210 | raw_spin_lock_irq(&cfs_b->lock); |
9211 | cfs_b->period = ns_to_ktime(period); | |
9212 | cfs_b->quota = quota; | |
58088ad0 | 9213 | |
a9cf55b2 | 9214 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
9215 | /* restart the period timer (if active) to handle new period expiry */ |
9216 | if (runtime_enabled && cfs_b->timer_active) { | |
9217 | /* force a reprogram */ | |
9218 | cfs_b->timer_active = 0; | |
9219 | __start_cfs_bandwidth(cfs_b); | |
9220 | } | |
ab84d31e PT |
9221 | raw_spin_unlock_irq(&cfs_b->lock); |
9222 | ||
9223 | for_each_possible_cpu(i) { | |
9224 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; | |
9225 | struct rq *rq = rq_of(cfs_rq); | |
9226 | ||
9227 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 9228 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 9229 | cfs_rq->runtime_remaining = 0; |
671fd9da PT |
9230 | |
9231 | if (cfs_rq_throttled(cfs_rq)) | |
9232 | unthrottle_cfs_rq(cfs_rq); | |
ab84d31e PT |
9233 | raw_spin_unlock_irq(&rq->lock); |
9234 | } | |
a790de99 PT |
9235 | out_unlock: |
9236 | mutex_unlock(&cfs_constraints_mutex); | |
ab84d31e | 9237 | |
a790de99 | 9238 | return ret; |
ab84d31e PT |
9239 | } |
9240 | ||
9241 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
9242 | { | |
9243 | u64 quota, period; | |
9244 | ||
9245 | period = ktime_to_ns(tg_cfs_bandwidth(tg)->period); | |
9246 | if (cfs_quota_us < 0) | |
9247 | quota = RUNTIME_INF; | |
9248 | else | |
9249 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
9250 | ||
9251 | return tg_set_cfs_bandwidth(tg, period, quota); | |
9252 | } | |
9253 | ||
9254 | long tg_get_cfs_quota(struct task_group *tg) | |
9255 | { | |
9256 | u64 quota_us; | |
9257 | ||
9258 | if (tg_cfs_bandwidth(tg)->quota == RUNTIME_INF) | |
9259 | return -1; | |
9260 | ||
9261 | quota_us = tg_cfs_bandwidth(tg)->quota; | |
9262 | do_div(quota_us, NSEC_PER_USEC); | |
9263 | ||
9264 | return quota_us; | |
9265 | } | |
9266 | ||
9267 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
9268 | { | |
9269 | u64 quota, period; | |
9270 | ||
9271 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
9272 | quota = tg_cfs_bandwidth(tg)->quota; | |
9273 | ||
9274 | if (period <= 0) | |
9275 | return -EINVAL; | |
9276 | ||
9277 | return tg_set_cfs_bandwidth(tg, period, quota); | |
9278 | } | |
9279 | ||
9280 | long tg_get_cfs_period(struct task_group *tg) | |
9281 | { | |
9282 | u64 cfs_period_us; | |
9283 | ||
9284 | cfs_period_us = ktime_to_ns(tg_cfs_bandwidth(tg)->period); | |
9285 | do_div(cfs_period_us, NSEC_PER_USEC); | |
9286 | ||
9287 | return cfs_period_us; | |
9288 | } | |
9289 | ||
9290 | static s64 cpu_cfs_quota_read_s64(struct cgroup *cgrp, struct cftype *cft) | |
9291 | { | |
9292 | return tg_get_cfs_quota(cgroup_tg(cgrp)); | |
9293 | } | |
9294 | ||
9295 | static int cpu_cfs_quota_write_s64(struct cgroup *cgrp, struct cftype *cftype, | |
9296 | s64 cfs_quota_us) | |
9297 | { | |
9298 | return tg_set_cfs_quota(cgroup_tg(cgrp), cfs_quota_us); | |
9299 | } | |
9300 | ||
9301 | static u64 cpu_cfs_period_read_u64(struct cgroup *cgrp, struct cftype *cft) | |
9302 | { | |
9303 | return tg_get_cfs_period(cgroup_tg(cgrp)); | |
9304 | } | |
9305 | ||
9306 | static int cpu_cfs_period_write_u64(struct cgroup *cgrp, struct cftype *cftype, | |
9307 | u64 cfs_period_us) | |
9308 | { | |
9309 | return tg_set_cfs_period(cgroup_tg(cgrp), cfs_period_us); | |
9310 | } | |
9311 | ||
a790de99 PT |
9312 | struct cfs_schedulable_data { |
9313 | struct task_group *tg; | |
9314 | u64 period, quota; | |
9315 | }; | |
9316 | ||
9317 | /* | |
9318 | * normalize group quota/period to be quota/max_period | |
9319 | * note: units are usecs | |
9320 | */ | |
9321 | static u64 normalize_cfs_quota(struct task_group *tg, | |
9322 | struct cfs_schedulable_data *d) | |
9323 | { | |
9324 | u64 quota, period; | |
9325 | ||
9326 | if (tg == d->tg) { | |
9327 | period = d->period; | |
9328 | quota = d->quota; | |
9329 | } else { | |
9330 | period = tg_get_cfs_period(tg); | |
9331 | quota = tg_get_cfs_quota(tg); | |
9332 | } | |
9333 | ||
9334 | /* note: these should typically be equivalent */ | |
9335 | if (quota == RUNTIME_INF || quota == -1) | |
9336 | return RUNTIME_INF; | |
9337 | ||
9338 | return to_ratio(period, quota); | |
9339 | } | |
9340 | ||
9341 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
9342 | { | |
9343 | struct cfs_schedulable_data *d = data; | |
9344 | struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(tg); | |
9345 | s64 quota = 0, parent_quota = -1; | |
9346 | ||
9347 | if (!tg->parent) { | |
9348 | quota = RUNTIME_INF; | |
9349 | } else { | |
9350 | struct cfs_bandwidth *parent_b = tg_cfs_bandwidth(tg->parent); | |
9351 | ||
9352 | quota = normalize_cfs_quota(tg, d); | |
9353 | parent_quota = parent_b->hierarchal_quota; | |
9354 | ||
9355 | /* | |
9356 | * ensure max(child_quota) <= parent_quota, inherit when no | |
9357 | * limit is set | |
9358 | */ | |
9359 | if (quota == RUNTIME_INF) | |
9360 | quota = parent_quota; | |
9361 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
9362 | return -EINVAL; | |
9363 | } | |
9364 | cfs_b->hierarchal_quota = quota; | |
9365 | ||
9366 | return 0; | |
9367 | } | |
9368 | ||
9369 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
9370 | { | |
8277434e | 9371 | int ret; |
a790de99 PT |
9372 | struct cfs_schedulable_data data = { |
9373 | .tg = tg, | |
9374 | .period = period, | |
9375 | .quota = quota, | |
9376 | }; | |
9377 | ||
9378 | if (quota != RUNTIME_INF) { | |
9379 | do_div(data.period, NSEC_PER_USEC); | |
9380 | do_div(data.quota, NSEC_PER_USEC); | |
9381 | } | |
9382 | ||
8277434e PT |
9383 | rcu_read_lock(); |
9384 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
9385 | rcu_read_unlock(); | |
9386 | ||
9387 | return ret; | |
a790de99 | 9388 | } |
e8da1b18 NR |
9389 | |
9390 | static int cpu_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
9391 | struct cgroup_map_cb *cb) | |
9392 | { | |
9393 | struct task_group *tg = cgroup_tg(cgrp); | |
9394 | struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(tg); | |
9395 | ||
9396 | cb->fill(cb, "nr_periods", cfs_b->nr_periods); | |
9397 | cb->fill(cb, "nr_throttled", cfs_b->nr_throttled); | |
9398 | cb->fill(cb, "throttled_time", cfs_b->throttled_time); | |
9399 | ||
9400 | return 0; | |
9401 | } | |
ab84d31e | 9402 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 9403 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 9404 | |
052f1dc7 | 9405 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 9406 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 9407 | s64 val) |
6f505b16 | 9408 | { |
06ecb27c | 9409 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
9410 | } |
9411 | ||
06ecb27c | 9412 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 9413 | { |
06ecb27c | 9414 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 9415 | } |
d0b27fa7 PZ |
9416 | |
9417 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
9418 | u64 rt_period_us) | |
9419 | { | |
9420 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
9421 | } | |
9422 | ||
9423 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
9424 | { | |
9425 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
9426 | } | |
6d6bc0ad | 9427 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 9428 | |
fe5c7cc2 | 9429 | static struct cftype cpu_files[] = { |
052f1dc7 | 9430 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
9431 | { |
9432 | .name = "shares", | |
f4c753b7 PM |
9433 | .read_u64 = cpu_shares_read_u64, |
9434 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 9435 | }, |
052f1dc7 | 9436 | #endif |
ab84d31e PT |
9437 | #ifdef CONFIG_CFS_BANDWIDTH |
9438 | { | |
9439 | .name = "cfs_quota_us", | |
9440 | .read_s64 = cpu_cfs_quota_read_s64, | |
9441 | .write_s64 = cpu_cfs_quota_write_s64, | |
9442 | }, | |
9443 | { | |
9444 | .name = "cfs_period_us", | |
9445 | .read_u64 = cpu_cfs_period_read_u64, | |
9446 | .write_u64 = cpu_cfs_period_write_u64, | |
9447 | }, | |
e8da1b18 NR |
9448 | { |
9449 | .name = "stat", | |
9450 | .read_map = cpu_stats_show, | |
9451 | }, | |
ab84d31e | 9452 | #endif |
052f1dc7 | 9453 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9454 | { |
9f0c1e56 | 9455 | .name = "rt_runtime_us", |
06ecb27c PM |
9456 | .read_s64 = cpu_rt_runtime_read, |
9457 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 9458 | }, |
d0b27fa7 PZ |
9459 | { |
9460 | .name = "rt_period_us", | |
f4c753b7 PM |
9461 | .read_u64 = cpu_rt_period_read_uint, |
9462 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 9463 | }, |
052f1dc7 | 9464 | #endif |
68318b8e SV |
9465 | }; |
9466 | ||
9467 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
9468 | { | |
fe5c7cc2 | 9469 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
9470 | } |
9471 | ||
9472 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
9473 | .name = "cpu", |
9474 | .create = cpu_cgroup_create, | |
9475 | .destroy = cpu_cgroup_destroy, | |
f780bdb7 BB |
9476 | .can_attach_task = cpu_cgroup_can_attach_task, |
9477 | .attach_task = cpu_cgroup_attach_task, | |
068c5cc5 | 9478 | .exit = cpu_cgroup_exit, |
38605cae IM |
9479 | .populate = cpu_cgroup_populate, |
9480 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
9481 | .early_init = 1, |
9482 | }; | |
9483 | ||
052f1dc7 | 9484 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
9485 | |
9486 | #ifdef CONFIG_CGROUP_CPUACCT | |
9487 | ||
9488 | /* | |
9489 | * CPU accounting code for task groups. | |
9490 | * | |
9491 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
9492 | * (balbir@in.ibm.com). | |
9493 | */ | |
9494 | ||
934352f2 | 9495 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
9496 | struct cpuacct { |
9497 | struct cgroup_subsys_state css; | |
9498 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43cf38eb | 9499 | u64 __percpu *cpuusage; |
ef12fefa | 9500 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 9501 | struct cpuacct *parent; |
d842de87 SV |
9502 | }; |
9503 | ||
9504 | struct cgroup_subsys cpuacct_subsys; | |
9505 | ||
9506 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 9507 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 9508 | { |
32cd756a | 9509 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
9510 | struct cpuacct, css); |
9511 | } | |
9512 | ||
9513 | /* return cpu accounting group to which this task belongs */ | |
9514 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
9515 | { | |
9516 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
9517 | struct cpuacct, css); | |
9518 | } | |
9519 | ||
9520 | /* create a new cpu accounting group */ | |
9521 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 9522 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
9523 | { |
9524 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 9525 | int i; |
d842de87 SV |
9526 | |
9527 | if (!ca) | |
ef12fefa | 9528 | goto out; |
d842de87 SV |
9529 | |
9530 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
9531 | if (!ca->cpuusage) |
9532 | goto out_free_ca; | |
9533 | ||
9534 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
9535 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
9536 | goto out_free_counters; | |
d842de87 | 9537 | |
934352f2 BR |
9538 | if (cgrp->parent) |
9539 | ca->parent = cgroup_ca(cgrp->parent); | |
9540 | ||
d842de87 | 9541 | return &ca->css; |
ef12fefa BR |
9542 | |
9543 | out_free_counters: | |
9544 | while (--i >= 0) | |
9545 | percpu_counter_destroy(&ca->cpustat[i]); | |
9546 | free_percpu(ca->cpuusage); | |
9547 | out_free_ca: | |
9548 | kfree(ca); | |
9549 | out: | |
9550 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
9551 | } |
9552 | ||
9553 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9554 | static void |
32cd756a | 9555 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9556 | { |
32cd756a | 9557 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 9558 | int i; |
d842de87 | 9559 | |
ef12fefa BR |
9560 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
9561 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
9562 | free_percpu(ca->cpuusage); |
9563 | kfree(ca); | |
9564 | } | |
9565 | ||
720f5498 KC |
9566 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
9567 | { | |
b36128c8 | 9568 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
9569 | u64 data; |
9570 | ||
9571 | #ifndef CONFIG_64BIT | |
9572 | /* | |
9573 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
9574 | */ | |
05fa785c | 9575 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 9576 | data = *cpuusage; |
05fa785c | 9577 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
9578 | #else |
9579 | data = *cpuusage; | |
9580 | #endif | |
9581 | ||
9582 | return data; | |
9583 | } | |
9584 | ||
9585 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
9586 | { | |
b36128c8 | 9587 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
9588 | |
9589 | #ifndef CONFIG_64BIT | |
9590 | /* | |
9591 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
9592 | */ | |
05fa785c | 9593 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 9594 | *cpuusage = val; |
05fa785c | 9595 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
9596 | #else |
9597 | *cpuusage = val; | |
9598 | #endif | |
9599 | } | |
9600 | ||
d842de87 | 9601 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 9602 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9603 | { |
32cd756a | 9604 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9605 | u64 totalcpuusage = 0; |
9606 | int i; | |
9607 | ||
720f5498 KC |
9608 | for_each_present_cpu(i) |
9609 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
9610 | |
9611 | return totalcpuusage; | |
9612 | } | |
9613 | ||
0297b803 DG |
9614 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9615 | u64 reset) | |
9616 | { | |
9617 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9618 | int err = 0; | |
9619 | int i; | |
9620 | ||
9621 | if (reset) { | |
9622 | err = -EINVAL; | |
9623 | goto out; | |
9624 | } | |
9625 | ||
720f5498 KC |
9626 | for_each_present_cpu(i) |
9627 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 9628 | |
0297b803 DG |
9629 | out: |
9630 | return err; | |
9631 | } | |
9632 | ||
e9515c3c KC |
9633 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
9634 | struct seq_file *m) | |
9635 | { | |
9636 | struct cpuacct *ca = cgroup_ca(cgroup); | |
9637 | u64 percpu; | |
9638 | int i; | |
9639 | ||
9640 | for_each_present_cpu(i) { | |
9641 | percpu = cpuacct_cpuusage_read(ca, i); | |
9642 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
9643 | } | |
9644 | seq_printf(m, "\n"); | |
9645 | return 0; | |
9646 | } | |
9647 | ||
ef12fefa BR |
9648 | static const char *cpuacct_stat_desc[] = { |
9649 | [CPUACCT_STAT_USER] = "user", | |
9650 | [CPUACCT_STAT_SYSTEM] = "system", | |
9651 | }; | |
9652 | ||
9653 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
9654 | struct cgroup_map_cb *cb) | |
9655 | { | |
9656 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9657 | int i; | |
9658 | ||
9659 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
9660 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
9661 | val = cputime64_to_clock_t(val); | |
9662 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
9663 | } | |
9664 | return 0; | |
9665 | } | |
9666 | ||
d842de87 SV |
9667 | static struct cftype files[] = { |
9668 | { | |
9669 | .name = "usage", | |
f4c753b7 PM |
9670 | .read_u64 = cpuusage_read, |
9671 | .write_u64 = cpuusage_write, | |
d842de87 | 9672 | }, |
e9515c3c KC |
9673 | { |
9674 | .name = "usage_percpu", | |
9675 | .read_seq_string = cpuacct_percpu_seq_read, | |
9676 | }, | |
ef12fefa BR |
9677 | { |
9678 | .name = "stat", | |
9679 | .read_map = cpuacct_stats_show, | |
9680 | }, | |
d842de87 SV |
9681 | }; |
9682 | ||
32cd756a | 9683 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9684 | { |
32cd756a | 9685 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9686 | } |
9687 | ||
9688 | /* | |
9689 | * charge this task's execution time to its accounting group. | |
9690 | * | |
9691 | * called with rq->lock held. | |
9692 | */ | |
9693 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9694 | { | |
9695 | struct cpuacct *ca; | |
934352f2 | 9696 | int cpu; |
d842de87 | 9697 | |
c40c6f85 | 9698 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
9699 | return; |
9700 | ||
934352f2 | 9701 | cpu = task_cpu(tsk); |
a18b83b7 BR |
9702 | |
9703 | rcu_read_lock(); | |
9704 | ||
d842de87 | 9705 | ca = task_ca(tsk); |
d842de87 | 9706 | |
934352f2 | 9707 | for (; ca; ca = ca->parent) { |
b36128c8 | 9708 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
9709 | *cpuusage += cputime; |
9710 | } | |
a18b83b7 BR |
9711 | |
9712 | rcu_read_unlock(); | |
d842de87 SV |
9713 | } |
9714 | ||
fa535a77 AB |
9715 | /* |
9716 | * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large | |
9717 | * in cputime_t units. As a result, cpuacct_update_stats calls | |
9718 | * percpu_counter_add with values large enough to always overflow the | |
9719 | * per cpu batch limit causing bad SMP scalability. | |
9720 | * | |
9721 | * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we | |
9722 | * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled | |
9723 | * and enabled. We cap it at INT_MAX which is the largest allowed batch value. | |
9724 | */ | |
9725 | #ifdef CONFIG_SMP | |
9726 | #define CPUACCT_BATCH \ | |
9727 | min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX) | |
9728 | #else | |
9729 | #define CPUACCT_BATCH 0 | |
9730 | #endif | |
9731 | ||
ef12fefa BR |
9732 | /* |
9733 | * Charge the system/user time to the task's accounting group. | |
9734 | */ | |
9735 | static void cpuacct_update_stats(struct task_struct *tsk, | |
9736 | enum cpuacct_stat_index idx, cputime_t val) | |
9737 | { | |
9738 | struct cpuacct *ca; | |
fa535a77 | 9739 | int batch = CPUACCT_BATCH; |
ef12fefa BR |
9740 | |
9741 | if (unlikely(!cpuacct_subsys.active)) | |
9742 | return; | |
9743 | ||
9744 | rcu_read_lock(); | |
9745 | ca = task_ca(tsk); | |
9746 | ||
9747 | do { | |
fa535a77 | 9748 | __percpu_counter_add(&ca->cpustat[idx], val, batch); |
ef12fefa BR |
9749 | ca = ca->parent; |
9750 | } while (ca); | |
9751 | rcu_read_unlock(); | |
9752 | } | |
9753 | ||
d842de87 SV |
9754 | struct cgroup_subsys cpuacct_subsys = { |
9755 | .name = "cpuacct", | |
9756 | .create = cpuacct_create, | |
9757 | .destroy = cpuacct_destroy, | |
9758 | .populate = cpuacct_populate, | |
9759 | .subsys_id = cpuacct_subsys_id, | |
9760 | }; | |
9761 | #endif /* CONFIG_CGROUP_CPUACCT */ |