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> |
1da177e4 | 78 | |
6e0534f2 | 79 | #include "sched_cpupri.h" |
21aa9af0 | 80 | #include "workqueue_sched.h" |
5091faa4 | 81 | #include "sched_autogroup.h" |
6e0534f2 | 82 | |
a8d154b0 | 83 | #define CREATE_TRACE_POINTS |
ad8d75ff | 84 | #include <trace/events/sched.h> |
a8d154b0 | 85 | |
1da177e4 LT |
86 | /* |
87 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
88 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
89 | * and back. | |
90 | */ | |
91 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
92 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
93 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
94 | ||
95 | /* | |
96 | * 'User priority' is the nice value converted to something we | |
97 | * can work with better when scaling various scheduler parameters, | |
98 | * it's a [ 0 ... 39 ] range. | |
99 | */ | |
100 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
101 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
102 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
103 | ||
104 | /* | |
d7876a08 | 105 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 106 | */ |
d6322faf | 107 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 108 | |
6aa645ea IM |
109 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
110 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
111 | ||
1da177e4 LT |
112 | /* |
113 | * These are the 'tuning knobs' of the scheduler: | |
114 | * | |
a4ec24b4 | 115 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
116 | * Timeslices get refilled after they expire. |
117 | */ | |
1da177e4 | 118 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 119 | |
d0b27fa7 PZ |
120 | /* |
121 | * single value that denotes runtime == period, ie unlimited time. | |
122 | */ | |
123 | #define RUNTIME_INF ((u64)~0ULL) | |
124 | ||
e05606d3 IM |
125 | static inline int rt_policy(int policy) |
126 | { | |
3f33a7ce | 127 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
128 | return 1; |
129 | return 0; | |
130 | } | |
131 | ||
132 | static inline int task_has_rt_policy(struct task_struct *p) | |
133 | { | |
134 | return rt_policy(p->policy); | |
135 | } | |
136 | ||
1da177e4 | 137 | /* |
6aa645ea | 138 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 139 | */ |
6aa645ea IM |
140 | struct rt_prio_array { |
141 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
142 | struct list_head queue[MAX_RT_PRIO]; | |
143 | }; | |
144 | ||
d0b27fa7 | 145 | struct rt_bandwidth { |
ea736ed5 | 146 | /* nests inside the rq lock: */ |
0986b11b | 147 | raw_spinlock_t rt_runtime_lock; |
ea736ed5 IM |
148 | ktime_t rt_period; |
149 | u64 rt_runtime; | |
150 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
151 | }; |
152 | ||
153 | static struct rt_bandwidth def_rt_bandwidth; | |
154 | ||
155 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
156 | ||
157 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
158 | { | |
159 | struct rt_bandwidth *rt_b = | |
160 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
161 | ktime_t now; | |
162 | int overrun; | |
163 | int idle = 0; | |
164 | ||
165 | for (;;) { | |
166 | now = hrtimer_cb_get_time(timer); | |
167 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
168 | ||
169 | if (!overrun) | |
170 | break; | |
171 | ||
172 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
173 | } | |
174 | ||
175 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
176 | } | |
177 | ||
178 | static | |
179 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
180 | { | |
181 | rt_b->rt_period = ns_to_ktime(period); | |
182 | rt_b->rt_runtime = runtime; | |
183 | ||
0986b11b | 184 | raw_spin_lock_init(&rt_b->rt_runtime_lock); |
ac086bc2 | 185 | |
d0b27fa7 PZ |
186 | hrtimer_init(&rt_b->rt_period_timer, |
187 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
188 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
189 | } |
190 | ||
c8bfff6d KH |
191 | static inline int rt_bandwidth_enabled(void) |
192 | { | |
193 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
194 | } |
195 | ||
196 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
197 | { | |
198 | ktime_t now; | |
199 | ||
cac64d00 | 200 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
201 | return; |
202 | ||
203 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
204 | return; | |
205 | ||
0986b11b | 206 | raw_spin_lock(&rt_b->rt_runtime_lock); |
d0b27fa7 | 207 | for (;;) { |
7f1e2ca9 PZ |
208 | unsigned long delta; |
209 | ktime_t soft, hard; | |
210 | ||
d0b27fa7 PZ |
211 | if (hrtimer_active(&rt_b->rt_period_timer)) |
212 | break; | |
213 | ||
214 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
215 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
216 | |
217 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
218 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
219 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
220 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
5c333864 | 221 | HRTIMER_MODE_ABS_PINNED, 0); |
d0b27fa7 | 222 | } |
0986b11b | 223 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
d0b27fa7 PZ |
224 | } |
225 | ||
226 | #ifdef CONFIG_RT_GROUP_SCHED | |
227 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
228 | { | |
229 | hrtimer_cancel(&rt_b->rt_period_timer); | |
230 | } | |
231 | #endif | |
232 | ||
712555ee HC |
233 | /* |
234 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
235 | * detach_destroy_domains and partition_sched_domains. | |
236 | */ | |
237 | static DEFINE_MUTEX(sched_domains_mutex); | |
238 | ||
7c941438 | 239 | #ifdef CONFIG_CGROUP_SCHED |
29f59db3 | 240 | |
68318b8e SV |
241 | #include <linux/cgroup.h> |
242 | ||
29f59db3 SV |
243 | struct cfs_rq; |
244 | ||
6f505b16 PZ |
245 | static LIST_HEAD(task_groups); |
246 | ||
29f59db3 | 247 | /* task group related information */ |
4cf86d77 | 248 | struct task_group { |
68318b8e | 249 | struct cgroup_subsys_state css; |
6c415b92 | 250 | |
052f1dc7 | 251 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
252 | /* schedulable entities of this group on each cpu */ |
253 | struct sched_entity **se; | |
254 | /* runqueue "owned" by this group on each cpu */ | |
255 | struct cfs_rq **cfs_rq; | |
256 | unsigned long shares; | |
2069dd75 PZ |
257 | |
258 | atomic_t load_weight; | |
052f1dc7 PZ |
259 | #endif |
260 | ||
261 | #ifdef CONFIG_RT_GROUP_SCHED | |
262 | struct sched_rt_entity **rt_se; | |
263 | struct rt_rq **rt_rq; | |
264 | ||
d0b27fa7 | 265 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 266 | #endif |
6b2d7700 | 267 | |
ae8393e5 | 268 | struct rcu_head rcu; |
6f505b16 | 269 | struct list_head list; |
f473aa5e PZ |
270 | |
271 | struct task_group *parent; | |
272 | struct list_head siblings; | |
273 | struct list_head children; | |
5091faa4 MG |
274 | |
275 | #ifdef CONFIG_SCHED_AUTOGROUP | |
276 | struct autogroup *autogroup; | |
277 | #endif | |
29f59db3 SV |
278 | }; |
279 | ||
3d4b47b4 | 280 | /* task_group_lock serializes the addition/removal of task groups */ |
8ed36996 | 281 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 282 | |
e9036b36 CG |
283 | #ifdef CONFIG_FAIR_GROUP_SCHED |
284 | ||
07e06b01 | 285 | # define ROOT_TASK_GROUP_LOAD NICE_0_LOAD |
052f1dc7 | 286 | |
cb4ad1ff | 287 | /* |
2e084786 LJ |
288 | * A weight of 0 or 1 can cause arithmetics problems. |
289 | * A weight of a cfs_rq is the sum of weights of which entities | |
290 | * are queued on this cfs_rq, so a weight of a entity should not be | |
291 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
292 | * (The default weight is 1024 - so there's no practical |
293 | * limitation from this.) | |
294 | */ | |
18d95a28 | 295 | #define MIN_SHARES 2 |
2e084786 | 296 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 297 | |
07e06b01 | 298 | static int root_task_group_load = ROOT_TASK_GROUP_LOAD; |
052f1dc7 PZ |
299 | #endif |
300 | ||
29f59db3 | 301 | /* Default task group. |
3a252015 | 302 | * Every task in system belong to this group at bootup. |
29f59db3 | 303 | */ |
07e06b01 | 304 | struct task_group root_task_group; |
29f59db3 | 305 | |
7c941438 | 306 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 307 | |
6aa645ea IM |
308 | /* CFS-related fields in a runqueue */ |
309 | struct cfs_rq { | |
310 | struct load_weight load; | |
311 | unsigned long nr_running; | |
312 | ||
6aa645ea | 313 | u64 exec_clock; |
e9acbff6 | 314 | u64 min_vruntime; |
6aa645ea IM |
315 | |
316 | struct rb_root tasks_timeline; | |
317 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
318 | |
319 | struct list_head tasks; | |
320 | struct list_head *balance_iterator; | |
321 | ||
322 | /* | |
323 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
324 | * It is set to NULL otherwise (i.e when none are currently running). |
325 | */ | |
ac53db59 | 326 | struct sched_entity *curr, *next, *last, *skip; |
ddc97297 | 327 | |
5ac5c4d6 | 328 | unsigned int nr_spread_over; |
ddc97297 | 329 | |
62160e3f | 330 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
331 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
332 | ||
41a2d6cf IM |
333 | /* |
334 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
335 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
336 | * (like users, containers etc.) | |
337 | * | |
338 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
339 | * list is used during load balance. | |
340 | */ | |
3d4b47b4 | 341 | int on_list; |
41a2d6cf IM |
342 | struct list_head leaf_cfs_rq_list; |
343 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
344 | |
345 | #ifdef CONFIG_SMP | |
c09595f6 | 346 | /* |
c8cba857 | 347 | * the part of load.weight contributed by tasks |
c09595f6 | 348 | */ |
c8cba857 | 349 | unsigned long task_weight; |
c09595f6 | 350 | |
c8cba857 PZ |
351 | /* |
352 | * h_load = weight * f(tg) | |
353 | * | |
354 | * Where f(tg) is the recursive weight fraction assigned to | |
355 | * this group. | |
356 | */ | |
357 | unsigned long h_load; | |
c09595f6 | 358 | |
c8cba857 | 359 | /* |
3b3d190e PT |
360 | * Maintaining per-cpu shares distribution for group scheduling |
361 | * | |
362 | * load_stamp is the last time we updated the load average | |
363 | * load_last is the last time we updated the load average and saw load | |
364 | * load_unacc_exec_time is currently unaccounted execution time | |
c8cba857 | 365 | */ |
2069dd75 PZ |
366 | u64 load_avg; |
367 | u64 load_period; | |
3b3d190e | 368 | u64 load_stamp, load_last, load_unacc_exec_time; |
f1d239f7 | 369 | |
2069dd75 | 370 | unsigned long load_contribution; |
c09595f6 | 371 | #endif |
6aa645ea IM |
372 | #endif |
373 | }; | |
1da177e4 | 374 | |
6aa645ea IM |
375 | /* Real-Time classes' related field in a runqueue: */ |
376 | struct rt_rq { | |
377 | struct rt_prio_array active; | |
63489e45 | 378 | unsigned long rt_nr_running; |
052f1dc7 | 379 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
380 | struct { |
381 | int curr; /* highest queued rt task prio */ | |
398a153b | 382 | #ifdef CONFIG_SMP |
e864c499 | 383 | int next; /* next highest */ |
398a153b | 384 | #endif |
e864c499 | 385 | } highest_prio; |
6f505b16 | 386 | #endif |
fa85ae24 | 387 | #ifdef CONFIG_SMP |
73fe6aae | 388 | unsigned long rt_nr_migratory; |
a1ba4d8b | 389 | unsigned long rt_nr_total; |
a22d7fc1 | 390 | int overloaded; |
917b627d | 391 | struct plist_head pushable_tasks; |
fa85ae24 | 392 | #endif |
6f505b16 | 393 | int rt_throttled; |
fa85ae24 | 394 | u64 rt_time; |
ac086bc2 | 395 | u64 rt_runtime; |
ea736ed5 | 396 | /* Nests inside the rq lock: */ |
0986b11b | 397 | raw_spinlock_t rt_runtime_lock; |
6f505b16 | 398 | |
052f1dc7 | 399 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
400 | unsigned long rt_nr_boosted; |
401 | ||
6f505b16 PZ |
402 | struct rq *rq; |
403 | struct list_head leaf_rt_rq_list; | |
404 | struct task_group *tg; | |
6f505b16 | 405 | #endif |
6aa645ea IM |
406 | }; |
407 | ||
57d885fe GH |
408 | #ifdef CONFIG_SMP |
409 | ||
410 | /* | |
411 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
412 | * variables. Each exclusive cpuset essentially defines an island domain by |
413 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
414 | * exclusive cpuset is created, we also create and attach a new root-domain |
415 | * object. | |
416 | * | |
57d885fe GH |
417 | */ |
418 | struct root_domain { | |
419 | atomic_t refcount; | |
c6c4927b RR |
420 | cpumask_var_t span; |
421 | cpumask_var_t online; | |
637f5085 | 422 | |
0eab9146 | 423 | /* |
637f5085 GH |
424 | * The "RT overload" flag: it gets set if a CPU has more than |
425 | * one runnable RT task. | |
426 | */ | |
c6c4927b | 427 | cpumask_var_t rto_mask; |
0eab9146 | 428 | atomic_t rto_count; |
6e0534f2 | 429 | struct cpupri cpupri; |
57d885fe GH |
430 | }; |
431 | ||
dc938520 GH |
432 | /* |
433 | * By default the system creates a single root-domain with all cpus as | |
434 | * members (mimicking the global state we have today). | |
435 | */ | |
57d885fe GH |
436 | static struct root_domain def_root_domain; |
437 | ||
ed2d372c | 438 | #endif /* CONFIG_SMP */ |
57d885fe | 439 | |
1da177e4 LT |
440 | /* |
441 | * This is the main, per-CPU runqueue data structure. | |
442 | * | |
443 | * Locking rule: those places that want to lock multiple runqueues | |
444 | * (such as the load balancing or the thread migration code), lock | |
445 | * acquire operations must be ordered by ascending &runqueue. | |
446 | */ | |
70b97a7f | 447 | struct rq { |
d8016491 | 448 | /* runqueue lock: */ |
05fa785c | 449 | raw_spinlock_t lock; |
1da177e4 LT |
450 | |
451 | /* | |
452 | * nr_running and cpu_load should be in the same cacheline because | |
453 | * remote CPUs use both these fields when doing load calculation. | |
454 | */ | |
455 | unsigned long nr_running; | |
6aa645ea IM |
456 | #define CPU_LOAD_IDX_MAX 5 |
457 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
fdf3e95d | 458 | unsigned long last_load_update_tick; |
46cb4b7c | 459 | #ifdef CONFIG_NO_HZ |
39c0cbe2 | 460 | u64 nohz_stamp; |
83cd4fe2 | 461 | unsigned char nohz_balance_kick; |
46cb4b7c | 462 | #endif |
a64692a3 MG |
463 | unsigned int skip_clock_update; |
464 | ||
d8016491 IM |
465 | /* capture load from *all* tasks on this cpu: */ |
466 | struct load_weight load; | |
6aa645ea IM |
467 | unsigned long nr_load_updates; |
468 | u64 nr_switches; | |
469 | ||
470 | struct cfs_rq cfs; | |
6f505b16 | 471 | struct rt_rq rt; |
6f505b16 | 472 | |
6aa645ea | 473 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
474 | /* list of leaf cfs_rq on this cpu: */ |
475 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
476 | #endif |
477 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 478 | struct list_head leaf_rt_rq_list; |
1da177e4 | 479 | #endif |
1da177e4 LT |
480 | |
481 | /* | |
482 | * This is part of a global counter where only the total sum | |
483 | * over all CPUs matters. A task can increase this counter on | |
484 | * one CPU and if it got migrated afterwards it may decrease | |
485 | * it on another CPU. Always updated under the runqueue lock: | |
486 | */ | |
487 | unsigned long nr_uninterruptible; | |
488 | ||
34f971f6 | 489 | struct task_struct *curr, *idle, *stop; |
c9819f45 | 490 | unsigned long next_balance; |
1da177e4 | 491 | struct mm_struct *prev_mm; |
6aa645ea | 492 | |
3e51f33f | 493 | u64 clock; |
305e6835 | 494 | u64 clock_task; |
6aa645ea | 495 | |
1da177e4 LT |
496 | atomic_t nr_iowait; |
497 | ||
498 | #ifdef CONFIG_SMP | |
0eab9146 | 499 | struct root_domain *rd; |
1da177e4 LT |
500 | struct sched_domain *sd; |
501 | ||
e51fd5e2 PZ |
502 | unsigned long cpu_power; |
503 | ||
a0a522ce | 504 | unsigned char idle_at_tick; |
1da177e4 | 505 | /* For active balancing */ |
3f029d3c | 506 | int post_schedule; |
1da177e4 LT |
507 | int active_balance; |
508 | int push_cpu; | |
969c7921 | 509 | struct cpu_stop_work active_balance_work; |
d8016491 IM |
510 | /* cpu of this runqueue: */ |
511 | int cpu; | |
1f11eb6a | 512 | int online; |
1da177e4 | 513 | |
a8a51d5e | 514 | unsigned long avg_load_per_task; |
1da177e4 | 515 | |
e9e9250b PZ |
516 | u64 rt_avg; |
517 | u64 age_stamp; | |
1b9508f6 MG |
518 | u64 idle_stamp; |
519 | u64 avg_idle; | |
1da177e4 LT |
520 | #endif |
521 | ||
aa483808 VP |
522 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
523 | u64 prev_irq_time; | |
524 | #endif | |
525 | ||
dce48a84 TG |
526 | /* calc_load related fields */ |
527 | unsigned long calc_load_update; | |
528 | long calc_load_active; | |
529 | ||
8f4d37ec | 530 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
531 | #ifdef CONFIG_SMP |
532 | int hrtick_csd_pending; | |
533 | struct call_single_data hrtick_csd; | |
534 | #endif | |
8f4d37ec PZ |
535 | struct hrtimer hrtick_timer; |
536 | #endif | |
537 | ||
1da177e4 LT |
538 | #ifdef CONFIG_SCHEDSTATS |
539 | /* latency stats */ | |
540 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
541 | unsigned long long rq_cpu_time; |
542 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
543 | |
544 | /* sys_sched_yield() stats */ | |
480b9434 | 545 | unsigned int yld_count; |
1da177e4 LT |
546 | |
547 | /* schedule() stats */ | |
480b9434 KC |
548 | unsigned int sched_switch; |
549 | unsigned int sched_count; | |
550 | unsigned int sched_goidle; | |
1da177e4 LT |
551 | |
552 | /* try_to_wake_up() stats */ | |
480b9434 KC |
553 | unsigned int ttwu_count; |
554 | unsigned int ttwu_local; | |
1da177e4 LT |
555 | #endif |
556 | }; | |
557 | ||
f34e3b61 | 558 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 559 | |
a64692a3 | 560 | |
1e5a7405 | 561 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); |
dd41f596 | 562 | |
0a2966b4 CL |
563 | static inline int cpu_of(struct rq *rq) |
564 | { | |
565 | #ifdef CONFIG_SMP | |
566 | return rq->cpu; | |
567 | #else | |
568 | return 0; | |
569 | #endif | |
570 | } | |
571 | ||
497f0ab3 | 572 | #define rcu_dereference_check_sched_domain(p) \ |
d11c563d PM |
573 | rcu_dereference_check((p), \ |
574 | rcu_read_lock_sched_held() || \ | |
575 | lockdep_is_held(&sched_domains_mutex)) | |
576 | ||
674311d5 NP |
577 | /* |
578 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 579 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
580 | * |
581 | * The domain tree of any CPU may only be accessed from within | |
582 | * preempt-disabled sections. | |
583 | */ | |
48f24c4d | 584 | #define for_each_domain(cpu, __sd) \ |
497f0ab3 | 585 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) |
1da177e4 LT |
586 | |
587 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
588 | #define this_rq() (&__get_cpu_var(runqueues)) | |
589 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
590 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 591 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 592 | |
dc61b1d6 PZ |
593 | #ifdef CONFIG_CGROUP_SCHED |
594 | ||
595 | /* | |
596 | * Return the group to which this tasks belongs. | |
597 | * | |
598 | * We use task_subsys_state_check() and extend the RCU verification | |
599 | * with lockdep_is_held(&task_rq(p)->lock) because cpu_cgroup_attach() | |
600 | * holds that lock for each task it moves into the cgroup. Therefore | |
601 | * by holding that lock, we pin the task to the current cgroup. | |
602 | */ | |
603 | static inline struct task_group *task_group(struct task_struct *p) | |
604 | { | |
5091faa4 | 605 | struct task_group *tg; |
dc61b1d6 PZ |
606 | struct cgroup_subsys_state *css; |
607 | ||
608 | css = task_subsys_state_check(p, cpu_cgroup_subsys_id, | |
609 | lockdep_is_held(&task_rq(p)->lock)); | |
5091faa4 MG |
610 | tg = container_of(css, struct task_group, css); |
611 | ||
612 | return autogroup_task_group(p, tg); | |
dc61b1d6 PZ |
613 | } |
614 | ||
615 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
616 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
617 | { | |
618 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
619 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; | |
620 | p->se.parent = task_group(p)->se[cpu]; | |
621 | #endif | |
622 | ||
623 | #ifdef CONFIG_RT_GROUP_SCHED | |
624 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; | |
625 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
626 | #endif | |
627 | } | |
628 | ||
629 | #else /* CONFIG_CGROUP_SCHED */ | |
630 | ||
631 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
632 | static inline struct task_group *task_group(struct task_struct *p) | |
633 | { | |
634 | return NULL; | |
635 | } | |
636 | ||
637 | #endif /* CONFIG_CGROUP_SCHED */ | |
638 | ||
fe44d621 | 639 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 640 | |
fe44d621 | 641 | static void update_rq_clock(struct rq *rq) |
3e51f33f | 642 | { |
fe44d621 | 643 | s64 delta; |
305e6835 | 644 | |
f26f9aff MG |
645 | if (rq->skip_clock_update) |
646 | return; | |
aa483808 | 647 | |
fe44d621 PZ |
648 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
649 | rq->clock += delta; | |
650 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
651 | } |
652 | ||
bf5c91ba IM |
653 | /* |
654 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
655 | */ | |
656 | #ifdef CONFIG_SCHED_DEBUG | |
657 | # define const_debug __read_mostly | |
658 | #else | |
659 | # define const_debug static const | |
660 | #endif | |
661 | ||
017730c1 | 662 | /** |
1fd06bb1 | 663 | * runqueue_is_locked - Returns true if the current cpu runqueue is locked |
e17b38bf | 664 | * @cpu: the processor in question. |
017730c1 | 665 | * |
017730c1 IM |
666 | * This interface allows printk to be called with the runqueue lock |
667 | * held and know whether or not it is OK to wake up the klogd. | |
668 | */ | |
89f19f04 | 669 | int runqueue_is_locked(int cpu) |
017730c1 | 670 | { |
05fa785c | 671 | return raw_spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
672 | } |
673 | ||
bf5c91ba IM |
674 | /* |
675 | * Debugging: various feature bits | |
676 | */ | |
f00b45c1 PZ |
677 | |
678 | #define SCHED_FEAT(name, enabled) \ | |
679 | __SCHED_FEAT_##name , | |
680 | ||
bf5c91ba | 681 | enum { |
f00b45c1 | 682 | #include "sched_features.h" |
bf5c91ba IM |
683 | }; |
684 | ||
f00b45c1 PZ |
685 | #undef SCHED_FEAT |
686 | ||
687 | #define SCHED_FEAT(name, enabled) \ | |
688 | (1UL << __SCHED_FEAT_##name) * enabled | | |
689 | ||
bf5c91ba | 690 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
691 | #include "sched_features.h" |
692 | 0; | |
693 | ||
694 | #undef SCHED_FEAT | |
695 | ||
696 | #ifdef CONFIG_SCHED_DEBUG | |
697 | #define SCHED_FEAT(name, enabled) \ | |
698 | #name , | |
699 | ||
983ed7a6 | 700 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
701 | #include "sched_features.h" |
702 | NULL | |
703 | }; | |
704 | ||
705 | #undef SCHED_FEAT | |
706 | ||
34f3a814 | 707 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 708 | { |
f00b45c1 PZ |
709 | int i; |
710 | ||
711 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
712 | if (!(sysctl_sched_features & (1UL << i))) |
713 | seq_puts(m, "NO_"); | |
714 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 715 | } |
34f3a814 | 716 | seq_puts(m, "\n"); |
f00b45c1 | 717 | |
34f3a814 | 718 | return 0; |
f00b45c1 PZ |
719 | } |
720 | ||
721 | static ssize_t | |
722 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
723 | size_t cnt, loff_t *ppos) | |
724 | { | |
725 | char buf[64]; | |
7740191c | 726 | char *cmp; |
f00b45c1 PZ |
727 | int neg = 0; |
728 | int i; | |
729 | ||
730 | if (cnt > 63) | |
731 | cnt = 63; | |
732 | ||
733 | if (copy_from_user(&buf, ubuf, cnt)) | |
734 | return -EFAULT; | |
735 | ||
736 | buf[cnt] = 0; | |
7740191c | 737 | cmp = strstrip(buf); |
f00b45c1 | 738 | |
524429c3 | 739 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
740 | neg = 1; |
741 | cmp += 3; | |
742 | } | |
743 | ||
744 | for (i = 0; sched_feat_names[i]; i++) { | |
7740191c | 745 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f00b45c1 PZ |
746 | if (neg) |
747 | sysctl_sched_features &= ~(1UL << i); | |
748 | else | |
749 | sysctl_sched_features |= (1UL << i); | |
750 | break; | |
751 | } | |
752 | } | |
753 | ||
754 | if (!sched_feat_names[i]) | |
755 | return -EINVAL; | |
756 | ||
42994724 | 757 | *ppos += cnt; |
f00b45c1 PZ |
758 | |
759 | return cnt; | |
760 | } | |
761 | ||
34f3a814 LZ |
762 | static int sched_feat_open(struct inode *inode, struct file *filp) |
763 | { | |
764 | return single_open(filp, sched_feat_show, NULL); | |
765 | } | |
766 | ||
828c0950 | 767 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
768 | .open = sched_feat_open, |
769 | .write = sched_feat_write, | |
770 | .read = seq_read, | |
771 | .llseek = seq_lseek, | |
772 | .release = single_release, | |
f00b45c1 PZ |
773 | }; |
774 | ||
775 | static __init int sched_init_debug(void) | |
776 | { | |
f00b45c1 PZ |
777 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
778 | &sched_feat_fops); | |
779 | ||
780 | return 0; | |
781 | } | |
782 | late_initcall(sched_init_debug); | |
783 | ||
784 | #endif | |
785 | ||
786 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 787 | |
b82d9fdd PZ |
788 | /* |
789 | * Number of tasks to iterate in a single balance run. | |
790 | * Limited because this is done with IRQs disabled. | |
791 | */ | |
792 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
793 | ||
e9e9250b PZ |
794 | /* |
795 | * period over which we average the RT time consumption, measured | |
796 | * in ms. | |
797 | * | |
798 | * default: 1s | |
799 | */ | |
800 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
801 | ||
fa85ae24 | 802 | /* |
9f0c1e56 | 803 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
804 | * default: 1s |
805 | */ | |
9f0c1e56 | 806 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 807 | |
6892b75e IM |
808 | static __read_mostly int scheduler_running; |
809 | ||
9f0c1e56 PZ |
810 | /* |
811 | * part of the period that we allow rt tasks to run in us. | |
812 | * default: 0.95s | |
813 | */ | |
814 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 815 | |
d0b27fa7 PZ |
816 | static inline u64 global_rt_period(void) |
817 | { | |
818 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
819 | } | |
820 | ||
821 | static inline u64 global_rt_runtime(void) | |
822 | { | |
e26873bb | 823 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
824 | return RUNTIME_INF; |
825 | ||
826 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
827 | } | |
fa85ae24 | 828 | |
1da177e4 | 829 | #ifndef prepare_arch_switch |
4866cde0 NP |
830 | # define prepare_arch_switch(next) do { } while (0) |
831 | #endif | |
832 | #ifndef finish_arch_switch | |
833 | # define finish_arch_switch(prev) do { } while (0) | |
834 | #endif | |
835 | ||
051a1d1a DA |
836 | static inline int task_current(struct rq *rq, struct task_struct *p) |
837 | { | |
838 | return rq->curr == p; | |
839 | } | |
840 | ||
70b97a7f | 841 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 842 | { |
3ca7a440 PZ |
843 | #ifdef CONFIG_SMP |
844 | return p->on_cpu; | |
845 | #else | |
051a1d1a | 846 | return task_current(rq, p); |
3ca7a440 | 847 | #endif |
4866cde0 NP |
848 | } |
849 | ||
3ca7a440 | 850 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 851 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 | 852 | { |
3ca7a440 PZ |
853 | #ifdef CONFIG_SMP |
854 | /* | |
855 | * We can optimise this out completely for !SMP, because the | |
856 | * SMP rebalancing from interrupt is the only thing that cares | |
857 | * here. | |
858 | */ | |
859 | next->on_cpu = 1; | |
860 | #endif | |
4866cde0 NP |
861 | } |
862 | ||
70b97a7f | 863 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 864 | { |
3ca7a440 PZ |
865 | #ifdef CONFIG_SMP |
866 | /* | |
867 | * After ->on_cpu is cleared, the task can be moved to a different CPU. | |
868 | * We must ensure this doesn't happen until the switch is completely | |
869 | * finished. | |
870 | */ | |
871 | smp_wmb(); | |
872 | prev->on_cpu = 0; | |
873 | #endif | |
da04c035 IM |
874 | #ifdef CONFIG_DEBUG_SPINLOCK |
875 | /* this is a valid case when another task releases the spinlock */ | |
876 | rq->lock.owner = current; | |
877 | #endif | |
8a25d5de IM |
878 | /* |
879 | * If we are tracking spinlock dependencies then we have to | |
880 | * fix up the runqueue lock - which gets 'carried over' from | |
881 | * prev into current: | |
882 | */ | |
883 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
884 | ||
05fa785c | 885 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 NP |
886 | } |
887 | ||
888 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 889 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
890 | { |
891 | #ifdef CONFIG_SMP | |
892 | /* | |
893 | * We can optimise this out completely for !SMP, because the | |
894 | * SMP rebalancing from interrupt is the only thing that cares | |
895 | * here. | |
896 | */ | |
3ca7a440 | 897 | next->on_cpu = 1; |
4866cde0 NP |
898 | #endif |
899 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
05fa785c | 900 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 | 901 | #else |
05fa785c | 902 | raw_spin_unlock(&rq->lock); |
4866cde0 NP |
903 | #endif |
904 | } | |
905 | ||
70b97a7f | 906 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
907 | { |
908 | #ifdef CONFIG_SMP | |
909 | /* | |
3ca7a440 | 910 | * After ->on_cpu is cleared, the task can be moved to a different CPU. |
4866cde0 NP |
911 | * We must ensure this doesn't happen until the switch is completely |
912 | * finished. | |
913 | */ | |
914 | smp_wmb(); | |
3ca7a440 | 915 | prev->on_cpu = 0; |
4866cde0 NP |
916 | #endif |
917 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
918 | local_irq_enable(); | |
1da177e4 | 919 | #endif |
4866cde0 NP |
920 | } |
921 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 922 | |
0970d299 | 923 | /* |
65cc8e48 PZ |
924 | * Check whether the task is waking, we use this to synchronize ->cpus_allowed |
925 | * against ttwu(). | |
0970d299 PZ |
926 | */ |
927 | static inline int task_is_waking(struct task_struct *p) | |
928 | { | |
0017d735 | 929 | return unlikely(p->state == TASK_WAKING); |
0970d299 PZ |
930 | } |
931 | ||
b29739f9 IM |
932 | /* |
933 | * __task_rq_lock - lock the runqueue a given task resides on. | |
934 | * Must be called interrupts disabled. | |
935 | */ | |
70b97a7f | 936 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
937 | __acquires(rq->lock) |
938 | { | |
0970d299 PZ |
939 | struct rq *rq; |
940 | ||
3a5c359a | 941 | for (;;) { |
0970d299 | 942 | rq = task_rq(p); |
05fa785c | 943 | raw_spin_lock(&rq->lock); |
65cc8e48 | 944 | if (likely(rq == task_rq(p))) |
3a5c359a | 945 | return rq; |
05fa785c | 946 | raw_spin_unlock(&rq->lock); |
b29739f9 | 947 | } |
b29739f9 IM |
948 | } |
949 | ||
1da177e4 LT |
950 | /* |
951 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 952 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
953 | * explicitly disabling preemption. |
954 | */ | |
70b97a7f | 955 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
956 | __acquires(rq->lock) |
957 | { | |
70b97a7f | 958 | struct rq *rq; |
1da177e4 | 959 | |
3a5c359a AK |
960 | for (;;) { |
961 | local_irq_save(*flags); | |
962 | rq = task_rq(p); | |
05fa785c | 963 | raw_spin_lock(&rq->lock); |
65cc8e48 | 964 | if (likely(rq == task_rq(p))) |
3a5c359a | 965 | return rq; |
05fa785c | 966 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 967 | } |
1da177e4 LT |
968 | } |
969 | ||
a9957449 | 970 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
971 | __releases(rq->lock) |
972 | { | |
05fa785c | 973 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
974 | } |
975 | ||
70b97a7f | 976 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
977 | __releases(rq->lock) |
978 | { | |
05fa785c | 979 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 LT |
980 | } |
981 | ||
1da177e4 | 982 | /* |
cc2a73b5 | 983 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 984 | */ |
a9957449 | 985 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
986 | __acquires(rq->lock) |
987 | { | |
70b97a7f | 988 | struct rq *rq; |
1da177e4 LT |
989 | |
990 | local_irq_disable(); | |
991 | rq = this_rq(); | |
05fa785c | 992 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
993 | |
994 | return rq; | |
995 | } | |
996 | ||
8f4d37ec PZ |
997 | #ifdef CONFIG_SCHED_HRTICK |
998 | /* | |
999 | * Use HR-timers to deliver accurate preemption points. | |
1000 | * | |
1001 | * Its all a bit involved since we cannot program an hrt while holding the | |
1002 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1003 | * reschedule event. | |
1004 | * | |
1005 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1006 | * rq->lock. | |
1007 | */ | |
8f4d37ec PZ |
1008 | |
1009 | /* | |
1010 | * Use hrtick when: | |
1011 | * - enabled by features | |
1012 | * - hrtimer is actually high res | |
1013 | */ | |
1014 | static inline int hrtick_enabled(struct rq *rq) | |
1015 | { | |
1016 | if (!sched_feat(HRTICK)) | |
1017 | return 0; | |
ba42059f | 1018 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1019 | return 0; |
8f4d37ec PZ |
1020 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1021 | } | |
1022 | ||
8f4d37ec PZ |
1023 | static void hrtick_clear(struct rq *rq) |
1024 | { | |
1025 | if (hrtimer_active(&rq->hrtick_timer)) | |
1026 | hrtimer_cancel(&rq->hrtick_timer); | |
1027 | } | |
1028 | ||
8f4d37ec PZ |
1029 | /* |
1030 | * High-resolution timer tick. | |
1031 | * Runs from hardirq context with interrupts disabled. | |
1032 | */ | |
1033 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1034 | { | |
1035 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1036 | ||
1037 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1038 | ||
05fa785c | 1039 | raw_spin_lock(&rq->lock); |
3e51f33f | 1040 | update_rq_clock(rq); |
8f4d37ec | 1041 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 1042 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
1043 | |
1044 | return HRTIMER_NORESTART; | |
1045 | } | |
1046 | ||
95e904c7 | 1047 | #ifdef CONFIG_SMP |
31656519 PZ |
1048 | /* |
1049 | * called from hardirq (IPI) context | |
1050 | */ | |
1051 | static void __hrtick_start(void *arg) | |
b328ca18 | 1052 | { |
31656519 | 1053 | struct rq *rq = arg; |
b328ca18 | 1054 | |
05fa785c | 1055 | raw_spin_lock(&rq->lock); |
31656519 PZ |
1056 | hrtimer_restart(&rq->hrtick_timer); |
1057 | rq->hrtick_csd_pending = 0; | |
05fa785c | 1058 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
1059 | } |
1060 | ||
31656519 PZ |
1061 | /* |
1062 | * Called to set the hrtick timer state. | |
1063 | * | |
1064 | * called with rq->lock held and irqs disabled | |
1065 | */ | |
1066 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1067 | { |
31656519 PZ |
1068 | struct hrtimer *timer = &rq->hrtick_timer; |
1069 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1070 | |
cc584b21 | 1071 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1072 | |
1073 | if (rq == this_rq()) { | |
1074 | hrtimer_restart(timer); | |
1075 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1076 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1077 | rq->hrtick_csd_pending = 1; |
1078 | } | |
b328ca18 PZ |
1079 | } |
1080 | ||
1081 | static int | |
1082 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1083 | { | |
1084 | int cpu = (int)(long)hcpu; | |
1085 | ||
1086 | switch (action) { | |
1087 | case CPU_UP_CANCELED: | |
1088 | case CPU_UP_CANCELED_FROZEN: | |
1089 | case CPU_DOWN_PREPARE: | |
1090 | case CPU_DOWN_PREPARE_FROZEN: | |
1091 | case CPU_DEAD: | |
1092 | case CPU_DEAD_FROZEN: | |
31656519 | 1093 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1094 | return NOTIFY_OK; |
1095 | } | |
1096 | ||
1097 | return NOTIFY_DONE; | |
1098 | } | |
1099 | ||
fa748203 | 1100 | static __init void init_hrtick(void) |
b328ca18 PZ |
1101 | { |
1102 | hotcpu_notifier(hotplug_hrtick, 0); | |
1103 | } | |
31656519 PZ |
1104 | #else |
1105 | /* | |
1106 | * Called to set the hrtick timer state. | |
1107 | * | |
1108 | * called with rq->lock held and irqs disabled | |
1109 | */ | |
1110 | static void hrtick_start(struct rq *rq, u64 delay) | |
1111 | { | |
7f1e2ca9 | 1112 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1113 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1114 | } |
b328ca18 | 1115 | |
006c75f1 | 1116 | static inline void init_hrtick(void) |
8f4d37ec | 1117 | { |
8f4d37ec | 1118 | } |
31656519 | 1119 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1120 | |
31656519 | 1121 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1122 | { |
31656519 PZ |
1123 | #ifdef CONFIG_SMP |
1124 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1125 | |
31656519 PZ |
1126 | rq->hrtick_csd.flags = 0; |
1127 | rq->hrtick_csd.func = __hrtick_start; | |
1128 | rq->hrtick_csd.info = rq; | |
1129 | #endif | |
8f4d37ec | 1130 | |
31656519 PZ |
1131 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1132 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1133 | } |
006c75f1 | 1134 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1135 | static inline void hrtick_clear(struct rq *rq) |
1136 | { | |
1137 | } | |
1138 | ||
8f4d37ec PZ |
1139 | static inline void init_rq_hrtick(struct rq *rq) |
1140 | { | |
1141 | } | |
1142 | ||
b328ca18 PZ |
1143 | static inline void init_hrtick(void) |
1144 | { | |
1145 | } | |
006c75f1 | 1146 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1147 | |
c24d20db IM |
1148 | /* |
1149 | * resched_task - mark a task 'to be rescheduled now'. | |
1150 | * | |
1151 | * On UP this means the setting of the need_resched flag, on SMP it | |
1152 | * might also involve a cross-CPU call to trigger the scheduler on | |
1153 | * the target CPU. | |
1154 | */ | |
1155 | #ifdef CONFIG_SMP | |
1156 | ||
1157 | #ifndef tsk_is_polling | |
1158 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1159 | #endif | |
1160 | ||
31656519 | 1161 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1162 | { |
1163 | int cpu; | |
1164 | ||
05fa785c | 1165 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 1166 | |
5ed0cec0 | 1167 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1168 | return; |
1169 | ||
5ed0cec0 | 1170 | set_tsk_need_resched(p); |
c24d20db IM |
1171 | |
1172 | cpu = task_cpu(p); | |
1173 | if (cpu == smp_processor_id()) | |
1174 | return; | |
1175 | ||
1176 | /* NEED_RESCHED must be visible before we test polling */ | |
1177 | smp_mb(); | |
1178 | if (!tsk_is_polling(p)) | |
1179 | smp_send_reschedule(cpu); | |
1180 | } | |
1181 | ||
1182 | static void resched_cpu(int cpu) | |
1183 | { | |
1184 | struct rq *rq = cpu_rq(cpu); | |
1185 | unsigned long flags; | |
1186 | ||
05fa785c | 1187 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
1188 | return; |
1189 | resched_task(cpu_curr(cpu)); | |
05fa785c | 1190 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 1191 | } |
06d8308c TG |
1192 | |
1193 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
1194 | /* |
1195 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
1196 | * from an idle cpu. This is good for power-savings. | |
1197 | * | |
1198 | * We don't do similar optimization for completely idle system, as | |
1199 | * selecting an idle cpu will add more delays to the timers than intended | |
1200 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
1201 | */ | |
1202 | int get_nohz_timer_target(void) | |
1203 | { | |
1204 | int cpu = smp_processor_id(); | |
1205 | int i; | |
1206 | struct sched_domain *sd; | |
1207 | ||
1208 | for_each_domain(cpu, sd) { | |
1209 | for_each_cpu(i, sched_domain_span(sd)) | |
1210 | if (!idle_cpu(i)) | |
1211 | return i; | |
1212 | } | |
1213 | return cpu; | |
1214 | } | |
06d8308c TG |
1215 | /* |
1216 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1217 | * idle CPU then this timer might expire before the next timer event | |
1218 | * which is scheduled to wake up that CPU. In case of a completely | |
1219 | * idle system the next event might even be infinite time into the | |
1220 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1221 | * leaves the inner idle loop so the newly added timer is taken into | |
1222 | * account when the CPU goes back to idle and evaluates the timer | |
1223 | * wheel for the next timer event. | |
1224 | */ | |
1225 | void wake_up_idle_cpu(int cpu) | |
1226 | { | |
1227 | struct rq *rq = cpu_rq(cpu); | |
1228 | ||
1229 | if (cpu == smp_processor_id()) | |
1230 | return; | |
1231 | ||
1232 | /* | |
1233 | * This is safe, as this function is called with the timer | |
1234 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1235 | * to idle and has not yet set rq->curr to idle then it will | |
1236 | * be serialized on the timer wheel base lock and take the new | |
1237 | * timer into account automatically. | |
1238 | */ | |
1239 | if (rq->curr != rq->idle) | |
1240 | return; | |
1241 | ||
1242 | /* | |
1243 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1244 | * lockless. The worst case is that the other CPU runs the | |
1245 | * idle task through an additional NOOP schedule() | |
1246 | */ | |
5ed0cec0 | 1247 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1248 | |
1249 | /* NEED_RESCHED must be visible before we test polling */ | |
1250 | smp_mb(); | |
1251 | if (!tsk_is_polling(rq->idle)) | |
1252 | smp_send_reschedule(cpu); | |
1253 | } | |
39c0cbe2 | 1254 | |
6d6bc0ad | 1255 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1256 | |
e9e9250b PZ |
1257 | static u64 sched_avg_period(void) |
1258 | { | |
1259 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1260 | } | |
1261 | ||
1262 | static void sched_avg_update(struct rq *rq) | |
1263 | { | |
1264 | s64 period = sched_avg_period(); | |
1265 | ||
1266 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
1267 | /* |
1268 | * Inline assembly required to prevent the compiler | |
1269 | * optimising this loop into a divmod call. | |
1270 | * See __iter_div_u64_rem() for another example of this. | |
1271 | */ | |
1272 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
1273 | rq->age_stamp += period; |
1274 | rq->rt_avg /= 2; | |
1275 | } | |
1276 | } | |
1277 | ||
1278 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1279 | { | |
1280 | rq->rt_avg += rt_delta; | |
1281 | sched_avg_update(rq); | |
1282 | } | |
1283 | ||
6d6bc0ad | 1284 | #else /* !CONFIG_SMP */ |
31656519 | 1285 | static void resched_task(struct task_struct *p) |
c24d20db | 1286 | { |
05fa785c | 1287 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 1288 | set_tsk_need_resched(p); |
c24d20db | 1289 | } |
e9e9250b PZ |
1290 | |
1291 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1292 | { | |
1293 | } | |
da2b71ed SS |
1294 | |
1295 | static void sched_avg_update(struct rq *rq) | |
1296 | { | |
1297 | } | |
6d6bc0ad | 1298 | #endif /* CONFIG_SMP */ |
c24d20db | 1299 | |
45bf76df IM |
1300 | #if BITS_PER_LONG == 32 |
1301 | # define WMULT_CONST (~0UL) | |
1302 | #else | |
1303 | # define WMULT_CONST (1UL << 32) | |
1304 | #endif | |
1305 | ||
1306 | #define WMULT_SHIFT 32 | |
1307 | ||
194081eb IM |
1308 | /* |
1309 | * Shift right and round: | |
1310 | */ | |
cf2ab469 | 1311 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1312 | |
a7be37ac PZ |
1313 | /* |
1314 | * delta *= weight / lw | |
1315 | */ | |
cb1c4fc9 | 1316 | static unsigned long |
45bf76df IM |
1317 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1318 | struct load_weight *lw) | |
1319 | { | |
1320 | u64 tmp; | |
1321 | ||
7a232e03 LJ |
1322 | if (!lw->inv_weight) { |
1323 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1324 | lw->inv_weight = 1; | |
1325 | else | |
1326 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1327 | / (lw->weight+1); | |
1328 | } | |
45bf76df IM |
1329 | |
1330 | tmp = (u64)delta_exec * weight; | |
1331 | /* | |
1332 | * Check whether we'd overflow the 64-bit multiplication: | |
1333 | */ | |
194081eb | 1334 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1335 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1336 | WMULT_SHIFT/2); |
1337 | else | |
cf2ab469 | 1338 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1339 | |
ecf691da | 1340 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1341 | } |
1342 | ||
1091985b | 1343 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1344 | { |
1345 | lw->weight += inc; | |
e89996ae | 1346 | lw->inv_weight = 0; |
45bf76df IM |
1347 | } |
1348 | ||
1091985b | 1349 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1350 | { |
1351 | lw->weight -= dec; | |
e89996ae | 1352 | lw->inv_weight = 0; |
45bf76df IM |
1353 | } |
1354 | ||
2069dd75 PZ |
1355 | static inline void update_load_set(struct load_weight *lw, unsigned long w) |
1356 | { | |
1357 | lw->weight = w; | |
1358 | lw->inv_weight = 0; | |
1359 | } | |
1360 | ||
2dd73a4f PW |
1361 | /* |
1362 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1363 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1364 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1365 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1366 | * scaled version of the new time slice allocation that they receive on time |
1367 | * slice expiry etc. | |
1368 | */ | |
1369 | ||
cce7ade8 PZ |
1370 | #define WEIGHT_IDLEPRIO 3 |
1371 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1372 | |
1373 | /* | |
1374 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1375 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1376 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1377 | * that remained on nice 0. | |
1378 | * | |
1379 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1380 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1381 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1382 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1383 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1384 | */ |
1385 | static const int prio_to_weight[40] = { | |
254753dc IM |
1386 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1387 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1388 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1389 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1390 | /* 0 */ 1024, 820, 655, 526, 423, | |
1391 | /* 5 */ 335, 272, 215, 172, 137, | |
1392 | /* 10 */ 110, 87, 70, 56, 45, | |
1393 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1394 | }; |
1395 | ||
5714d2de IM |
1396 | /* |
1397 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1398 | * | |
1399 | * In cases where the weight does not change often, we can use the | |
1400 | * precalculated inverse to speed up arithmetics by turning divisions | |
1401 | * into multiplications: | |
1402 | */ | |
dd41f596 | 1403 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1404 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1405 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1406 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1407 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1408 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1409 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1410 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1411 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1412 | }; |
2dd73a4f | 1413 | |
ef12fefa BR |
1414 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1415 | enum cpuacct_stat_index { | |
1416 | CPUACCT_STAT_USER, /* ... user mode */ | |
1417 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1418 | ||
1419 | CPUACCT_STAT_NSTATS, | |
1420 | }; | |
1421 | ||
d842de87 SV |
1422 | #ifdef CONFIG_CGROUP_CPUACCT |
1423 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1424 | static void cpuacct_update_stats(struct task_struct *tsk, |
1425 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1426 | #else |
1427 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1428 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1429 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1430 | #endif |
1431 | ||
18d95a28 PZ |
1432 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1433 | { | |
1434 | update_load_add(&rq->load, load); | |
1435 | } | |
1436 | ||
1437 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1438 | { | |
1439 | update_load_sub(&rq->load, load); | |
1440 | } | |
1441 | ||
7940ca36 | 1442 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1443 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1444 | |
1445 | /* | |
1446 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1447 | * leaving it for the final time. | |
1448 | */ | |
eb755805 | 1449 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1450 | { |
1451 | struct task_group *parent, *child; | |
eb755805 | 1452 | int ret; |
c09595f6 PZ |
1453 | |
1454 | rcu_read_lock(); | |
1455 | parent = &root_task_group; | |
1456 | down: | |
eb755805 PZ |
1457 | ret = (*down)(parent, data); |
1458 | if (ret) | |
1459 | goto out_unlock; | |
c09595f6 PZ |
1460 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1461 | parent = child; | |
1462 | goto down; | |
1463 | ||
1464 | up: | |
1465 | continue; | |
1466 | } | |
eb755805 PZ |
1467 | ret = (*up)(parent, data); |
1468 | if (ret) | |
1469 | goto out_unlock; | |
c09595f6 PZ |
1470 | |
1471 | child = parent; | |
1472 | parent = parent->parent; | |
1473 | if (parent) | |
1474 | goto up; | |
eb755805 | 1475 | out_unlock: |
c09595f6 | 1476 | rcu_read_unlock(); |
eb755805 PZ |
1477 | |
1478 | return ret; | |
c09595f6 PZ |
1479 | } |
1480 | ||
eb755805 PZ |
1481 | static int tg_nop(struct task_group *tg, void *data) |
1482 | { | |
1483 | return 0; | |
c09595f6 | 1484 | } |
eb755805 PZ |
1485 | #endif |
1486 | ||
1487 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1488 | /* Used instead of source_load when we know the type == 0 */ |
1489 | static unsigned long weighted_cpuload(const int cpu) | |
1490 | { | |
1491 | return cpu_rq(cpu)->load.weight; | |
1492 | } | |
1493 | ||
1494 | /* | |
1495 | * Return a low guess at the load of a migration-source cpu weighted | |
1496 | * according to the scheduling class and "nice" value. | |
1497 | * | |
1498 | * We want to under-estimate the load of migration sources, to | |
1499 | * balance conservatively. | |
1500 | */ | |
1501 | static unsigned long source_load(int cpu, int type) | |
1502 | { | |
1503 | struct rq *rq = cpu_rq(cpu); | |
1504 | unsigned long total = weighted_cpuload(cpu); | |
1505 | ||
1506 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1507 | return total; | |
1508 | ||
1509 | return min(rq->cpu_load[type-1], total); | |
1510 | } | |
1511 | ||
1512 | /* | |
1513 | * Return a high guess at the load of a migration-target cpu weighted | |
1514 | * according to the scheduling class and "nice" value. | |
1515 | */ | |
1516 | static unsigned long target_load(int cpu, int type) | |
1517 | { | |
1518 | struct rq *rq = cpu_rq(cpu); | |
1519 | unsigned long total = weighted_cpuload(cpu); | |
1520 | ||
1521 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1522 | return total; | |
1523 | ||
1524 | return max(rq->cpu_load[type-1], total); | |
1525 | } | |
1526 | ||
ae154be1 PZ |
1527 | static unsigned long power_of(int cpu) |
1528 | { | |
e51fd5e2 | 1529 | return cpu_rq(cpu)->cpu_power; |
ae154be1 PZ |
1530 | } |
1531 | ||
eb755805 PZ |
1532 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1533 | ||
1534 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1535 | { | |
1536 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1537 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1538 | |
4cd42620 SR |
1539 | if (nr_running) |
1540 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1541 | else |
1542 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1543 | |
1544 | return rq->avg_load_per_task; | |
1545 | } | |
1546 | ||
1547 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1548 | |
c09595f6 | 1549 | /* |
c8cba857 PZ |
1550 | * Compute the cpu's hierarchical load factor for each task group. |
1551 | * This needs to be done in a top-down fashion because the load of a child | |
1552 | * group is a fraction of its parents load. | |
c09595f6 | 1553 | */ |
eb755805 | 1554 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1555 | { |
c8cba857 | 1556 | unsigned long load; |
eb755805 | 1557 | long cpu = (long)data; |
c09595f6 | 1558 | |
c8cba857 PZ |
1559 | if (!tg->parent) { |
1560 | load = cpu_rq(cpu)->load.weight; | |
1561 | } else { | |
1562 | load = tg->parent->cfs_rq[cpu]->h_load; | |
2069dd75 | 1563 | load *= tg->se[cpu]->load.weight; |
c8cba857 PZ |
1564 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; |
1565 | } | |
c09595f6 | 1566 | |
c8cba857 | 1567 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1568 | |
eb755805 | 1569 | return 0; |
c09595f6 PZ |
1570 | } |
1571 | ||
eb755805 | 1572 | static void update_h_load(long cpu) |
c09595f6 | 1573 | { |
eb755805 | 1574 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1575 | } |
1576 | ||
18d95a28 PZ |
1577 | #endif |
1578 | ||
8f45e2b5 GH |
1579 | #ifdef CONFIG_PREEMPT |
1580 | ||
b78bb868 PZ |
1581 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1582 | ||
70574a99 | 1583 | /* |
8f45e2b5 GH |
1584 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1585 | * way at the expense of forcing extra atomic operations in all | |
1586 | * invocations. This assures that the double_lock is acquired using the | |
1587 | * same underlying policy as the spinlock_t on this architecture, which | |
1588 | * reduces latency compared to the unfair variant below. However, it | |
1589 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1590 | */ |
8f45e2b5 GH |
1591 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1592 | __releases(this_rq->lock) | |
1593 | __acquires(busiest->lock) | |
1594 | __acquires(this_rq->lock) | |
1595 | { | |
05fa785c | 1596 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1597 | double_rq_lock(this_rq, busiest); |
1598 | ||
1599 | return 1; | |
1600 | } | |
1601 | ||
1602 | #else | |
1603 | /* | |
1604 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1605 | * latency by eliminating extra atomic operations when the locks are | |
1606 | * already in proper order on entry. This favors lower cpu-ids and will | |
1607 | * grant the double lock to lower cpus over higher ids under contention, | |
1608 | * regardless of entry order into the function. | |
1609 | */ | |
1610 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1611 | __releases(this_rq->lock) |
1612 | __acquires(busiest->lock) | |
1613 | __acquires(this_rq->lock) | |
1614 | { | |
1615 | int ret = 0; | |
1616 | ||
05fa785c | 1617 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1618 | if (busiest < this_rq) { |
05fa785c TG |
1619 | raw_spin_unlock(&this_rq->lock); |
1620 | raw_spin_lock(&busiest->lock); | |
1621 | raw_spin_lock_nested(&this_rq->lock, | |
1622 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1623 | ret = 1; |
1624 | } else | |
05fa785c TG |
1625 | raw_spin_lock_nested(&busiest->lock, |
1626 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1627 | } |
1628 | return ret; | |
1629 | } | |
1630 | ||
8f45e2b5 GH |
1631 | #endif /* CONFIG_PREEMPT */ |
1632 | ||
1633 | /* | |
1634 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1635 | */ | |
1636 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1637 | { | |
1638 | if (unlikely(!irqs_disabled())) { | |
1639 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1640 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1641 | BUG_ON(1); |
1642 | } | |
1643 | ||
1644 | return _double_lock_balance(this_rq, busiest); | |
1645 | } | |
1646 | ||
70574a99 AD |
1647 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1648 | __releases(busiest->lock) | |
1649 | { | |
05fa785c | 1650 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1651 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1652 | } | |
1e3c88bd PZ |
1653 | |
1654 | /* | |
1655 | * double_rq_lock - safely lock two runqueues | |
1656 | * | |
1657 | * Note this does not disable interrupts like task_rq_lock, | |
1658 | * you need to do so manually before calling. | |
1659 | */ | |
1660 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1661 | __acquires(rq1->lock) | |
1662 | __acquires(rq2->lock) | |
1663 | { | |
1664 | BUG_ON(!irqs_disabled()); | |
1665 | if (rq1 == rq2) { | |
1666 | raw_spin_lock(&rq1->lock); | |
1667 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1668 | } else { | |
1669 | if (rq1 < rq2) { | |
1670 | raw_spin_lock(&rq1->lock); | |
1671 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1672 | } else { | |
1673 | raw_spin_lock(&rq2->lock); | |
1674 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1675 | } | |
1676 | } | |
1e3c88bd PZ |
1677 | } |
1678 | ||
1679 | /* | |
1680 | * double_rq_unlock - safely unlock two runqueues | |
1681 | * | |
1682 | * Note this does not restore interrupts like task_rq_unlock, | |
1683 | * you need to do so manually after calling. | |
1684 | */ | |
1685 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1686 | __releases(rq1->lock) | |
1687 | __releases(rq2->lock) | |
1688 | { | |
1689 | raw_spin_unlock(&rq1->lock); | |
1690 | if (rq1 != rq2) | |
1691 | raw_spin_unlock(&rq2->lock); | |
1692 | else | |
1693 | __release(rq2->lock); | |
1694 | } | |
1695 | ||
d95f4122 MG |
1696 | #else /* CONFIG_SMP */ |
1697 | ||
1698 | /* | |
1699 | * double_rq_lock - safely lock two runqueues | |
1700 | * | |
1701 | * Note this does not disable interrupts like task_rq_lock, | |
1702 | * you need to do so manually before calling. | |
1703 | */ | |
1704 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1705 | __acquires(rq1->lock) | |
1706 | __acquires(rq2->lock) | |
1707 | { | |
1708 | BUG_ON(!irqs_disabled()); | |
1709 | BUG_ON(rq1 != rq2); | |
1710 | raw_spin_lock(&rq1->lock); | |
1711 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1712 | } | |
1713 | ||
1714 | /* | |
1715 | * double_rq_unlock - safely unlock two runqueues | |
1716 | * | |
1717 | * Note this does not restore interrupts like task_rq_unlock, | |
1718 | * you need to do so manually after calling. | |
1719 | */ | |
1720 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1721 | __releases(rq1->lock) | |
1722 | __releases(rq2->lock) | |
1723 | { | |
1724 | BUG_ON(rq1 != rq2); | |
1725 | raw_spin_unlock(&rq1->lock); | |
1726 | __release(rq2->lock); | |
1727 | } | |
1728 | ||
18d95a28 PZ |
1729 | #endif |
1730 | ||
74f5187a | 1731 | static void calc_load_account_idle(struct rq *this_rq); |
0bcdcf28 | 1732 | static void update_sysctl(void); |
acb4a848 | 1733 | static int get_update_sysctl_factor(void); |
fdf3e95d | 1734 | static void update_cpu_load(struct rq *this_rq); |
dce48a84 | 1735 | |
cd29fe6f PZ |
1736 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1737 | { | |
1738 | set_task_rq(p, cpu); | |
1739 | #ifdef CONFIG_SMP | |
1740 | /* | |
1741 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1742 | * successfuly executed on another CPU. We must ensure that updates of | |
1743 | * per-task data have been completed by this moment. | |
1744 | */ | |
1745 | smp_wmb(); | |
1746 | task_thread_info(p)->cpu = cpu; | |
1747 | #endif | |
1748 | } | |
dce48a84 | 1749 | |
1e3c88bd | 1750 | static const struct sched_class rt_sched_class; |
dd41f596 | 1751 | |
34f971f6 | 1752 | #define sched_class_highest (&stop_sched_class) |
1f11eb6a GH |
1753 | #define for_each_class(class) \ |
1754 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1755 | |
1e3c88bd PZ |
1756 | #include "sched_stats.h" |
1757 | ||
c09595f6 | 1758 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1759 | { |
1760 | rq->nr_running++; | |
9c217245 IM |
1761 | } |
1762 | ||
c09595f6 | 1763 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1764 | { |
1765 | rq->nr_running--; | |
9c217245 IM |
1766 | } |
1767 | ||
45bf76df IM |
1768 | static void set_load_weight(struct task_struct *p) |
1769 | { | |
dd41f596 IM |
1770 | /* |
1771 | * SCHED_IDLE tasks get minimal weight: | |
1772 | */ | |
1773 | if (p->policy == SCHED_IDLE) { | |
1774 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1775 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1776 | return; | |
1777 | } | |
71f8bd46 | 1778 | |
dd41f596 IM |
1779 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1780 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1781 | } |
1782 | ||
371fd7e7 | 1783 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1784 | { |
a64692a3 | 1785 | update_rq_clock(rq); |
dd41f596 | 1786 | sched_info_queued(p); |
371fd7e7 | 1787 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
1788 | } |
1789 | ||
371fd7e7 | 1790 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1791 | { |
a64692a3 | 1792 | update_rq_clock(rq); |
46ac22ba | 1793 | sched_info_dequeued(p); |
371fd7e7 | 1794 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
1795 | } |
1796 | ||
1e3c88bd PZ |
1797 | /* |
1798 | * activate_task - move a task to the runqueue. | |
1799 | */ | |
371fd7e7 | 1800 | static void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1801 | { |
1802 | if (task_contributes_to_load(p)) | |
1803 | rq->nr_uninterruptible--; | |
1804 | ||
371fd7e7 | 1805 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
1806 | inc_nr_running(rq); |
1807 | } | |
1808 | ||
1809 | /* | |
1810 | * deactivate_task - remove a task from the runqueue. | |
1811 | */ | |
371fd7e7 | 1812 | static void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1813 | { |
1814 | if (task_contributes_to_load(p)) | |
1815 | rq->nr_uninterruptible++; | |
1816 | ||
371fd7e7 | 1817 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1818 | dec_nr_running(rq); |
1819 | } | |
1820 | ||
b52bfee4 VP |
1821 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
1822 | ||
305e6835 VP |
1823 | /* |
1824 | * There are no locks covering percpu hardirq/softirq time. | |
1825 | * They are only modified in account_system_vtime, on corresponding CPU | |
1826 | * with interrupts disabled. So, writes are safe. | |
1827 | * They are read and saved off onto struct rq in update_rq_clock(). | |
1828 | * This may result in other CPU reading this CPU's irq time and can | |
1829 | * race with irq/account_system_vtime on this CPU. We would either get old | |
8e92c201 PZ |
1830 | * or new value with a side effect of accounting a slice of irq time to wrong |
1831 | * task when irq is in progress while we read rq->clock. That is a worthy | |
1832 | * compromise in place of having locks on each irq in account_system_time. | |
305e6835 | 1833 | */ |
b52bfee4 VP |
1834 | static DEFINE_PER_CPU(u64, cpu_hardirq_time); |
1835 | static DEFINE_PER_CPU(u64, cpu_softirq_time); | |
1836 | ||
1837 | static DEFINE_PER_CPU(u64, irq_start_time); | |
1838 | static int sched_clock_irqtime; | |
1839 | ||
1840 | void enable_sched_clock_irqtime(void) | |
1841 | { | |
1842 | sched_clock_irqtime = 1; | |
1843 | } | |
1844 | ||
1845 | void disable_sched_clock_irqtime(void) | |
1846 | { | |
1847 | sched_clock_irqtime = 0; | |
1848 | } | |
1849 | ||
8e92c201 PZ |
1850 | #ifndef CONFIG_64BIT |
1851 | static DEFINE_PER_CPU(seqcount_t, irq_time_seq); | |
1852 | ||
1853 | static inline void irq_time_write_begin(void) | |
1854 | { | |
1855 | __this_cpu_inc(irq_time_seq.sequence); | |
1856 | smp_wmb(); | |
1857 | } | |
1858 | ||
1859 | static inline void irq_time_write_end(void) | |
1860 | { | |
1861 | smp_wmb(); | |
1862 | __this_cpu_inc(irq_time_seq.sequence); | |
1863 | } | |
1864 | ||
1865 | static inline u64 irq_time_read(int cpu) | |
1866 | { | |
1867 | u64 irq_time; | |
1868 | unsigned seq; | |
1869 | ||
1870 | do { | |
1871 | seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); | |
1872 | irq_time = per_cpu(cpu_softirq_time, cpu) + | |
1873 | per_cpu(cpu_hardirq_time, cpu); | |
1874 | } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); | |
1875 | ||
1876 | return irq_time; | |
1877 | } | |
1878 | #else /* CONFIG_64BIT */ | |
1879 | static inline void irq_time_write_begin(void) | |
1880 | { | |
1881 | } | |
1882 | ||
1883 | static inline void irq_time_write_end(void) | |
1884 | { | |
1885 | } | |
1886 | ||
1887 | static inline u64 irq_time_read(int cpu) | |
305e6835 | 1888 | { |
305e6835 VP |
1889 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); |
1890 | } | |
8e92c201 | 1891 | #endif /* CONFIG_64BIT */ |
305e6835 | 1892 | |
fe44d621 PZ |
1893 | /* |
1894 | * Called before incrementing preempt_count on {soft,}irq_enter | |
1895 | * and before decrementing preempt_count on {soft,}irq_exit. | |
1896 | */ | |
b52bfee4 VP |
1897 | void account_system_vtime(struct task_struct *curr) |
1898 | { | |
1899 | unsigned long flags; | |
fe44d621 | 1900 | s64 delta; |
b52bfee4 | 1901 | int cpu; |
b52bfee4 VP |
1902 | |
1903 | if (!sched_clock_irqtime) | |
1904 | return; | |
1905 | ||
1906 | local_irq_save(flags); | |
1907 | ||
b52bfee4 | 1908 | cpu = smp_processor_id(); |
fe44d621 PZ |
1909 | delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time); |
1910 | __this_cpu_add(irq_start_time, delta); | |
1911 | ||
8e92c201 | 1912 | irq_time_write_begin(); |
b52bfee4 VP |
1913 | /* |
1914 | * We do not account for softirq time from ksoftirqd here. | |
1915 | * We want to continue accounting softirq time to ksoftirqd thread | |
1916 | * in that case, so as not to confuse scheduler with a special task | |
1917 | * that do not consume any time, but still wants to run. | |
1918 | */ | |
1919 | if (hardirq_count()) | |
fe44d621 | 1920 | __this_cpu_add(cpu_hardirq_time, delta); |
4dd53d89 | 1921 | else if (in_serving_softirq() && curr != this_cpu_ksoftirqd()) |
fe44d621 | 1922 | __this_cpu_add(cpu_softirq_time, delta); |
b52bfee4 | 1923 | |
8e92c201 | 1924 | irq_time_write_end(); |
b52bfee4 VP |
1925 | local_irq_restore(flags); |
1926 | } | |
b7dadc38 | 1927 | EXPORT_SYMBOL_GPL(account_system_vtime); |
b52bfee4 | 1928 | |
fe44d621 | 1929 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 1930 | { |
fe44d621 PZ |
1931 | s64 irq_delta; |
1932 | ||
8e92c201 | 1933 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
1934 | |
1935 | /* | |
1936 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
1937 | * this case when a previous update_rq_clock() happened inside a | |
1938 | * {soft,}irq region. | |
1939 | * | |
1940 | * When this happens, we stop ->clock_task and only update the | |
1941 | * prev_irq_time stamp to account for the part that fit, so that a next | |
1942 | * update will consume the rest. This ensures ->clock_task is | |
1943 | * monotonic. | |
1944 | * | |
1945 | * It does however cause some slight miss-attribution of {soft,}irq | |
1946 | * time, a more accurate solution would be to update the irq_time using | |
1947 | * the current rq->clock timestamp, except that would require using | |
1948 | * atomic ops. | |
1949 | */ | |
1950 | if (irq_delta > delta) | |
1951 | irq_delta = delta; | |
1952 | ||
1953 | rq->prev_irq_time += irq_delta; | |
1954 | delta -= irq_delta; | |
1955 | rq->clock_task += delta; | |
1956 | ||
1957 | if (irq_delta && sched_feat(NONIRQ_POWER)) | |
1958 | sched_rt_avg_update(rq, irq_delta); | |
aa483808 VP |
1959 | } |
1960 | ||
abb74cef VP |
1961 | static int irqtime_account_hi_update(void) |
1962 | { | |
1963 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1964 | unsigned long flags; | |
1965 | u64 latest_ns; | |
1966 | int ret = 0; | |
1967 | ||
1968 | local_irq_save(flags); | |
1969 | latest_ns = this_cpu_read(cpu_hardirq_time); | |
1970 | if (cputime64_gt(nsecs_to_cputime64(latest_ns), cpustat->irq)) | |
1971 | ret = 1; | |
1972 | local_irq_restore(flags); | |
1973 | return ret; | |
1974 | } | |
1975 | ||
1976 | static int irqtime_account_si_update(void) | |
1977 | { | |
1978 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1979 | unsigned long flags; | |
1980 | u64 latest_ns; | |
1981 | int ret = 0; | |
1982 | ||
1983 | local_irq_save(flags); | |
1984 | latest_ns = this_cpu_read(cpu_softirq_time); | |
1985 | if (cputime64_gt(nsecs_to_cputime64(latest_ns), cpustat->softirq)) | |
1986 | ret = 1; | |
1987 | local_irq_restore(flags); | |
1988 | return ret; | |
1989 | } | |
1990 | ||
fe44d621 | 1991 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
305e6835 | 1992 | |
abb74cef VP |
1993 | #define sched_clock_irqtime (0) |
1994 | ||
fe44d621 | 1995 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
305e6835 | 1996 | { |
fe44d621 | 1997 | rq->clock_task += delta; |
305e6835 VP |
1998 | } |
1999 | ||
fe44d621 | 2000 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
b52bfee4 | 2001 | |
1e3c88bd PZ |
2002 | #include "sched_idletask.c" |
2003 | #include "sched_fair.c" | |
2004 | #include "sched_rt.c" | |
5091faa4 | 2005 | #include "sched_autogroup.c" |
34f971f6 | 2006 | #include "sched_stoptask.c" |
1e3c88bd PZ |
2007 | #ifdef CONFIG_SCHED_DEBUG |
2008 | # include "sched_debug.c" | |
2009 | #endif | |
2010 | ||
34f971f6 PZ |
2011 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
2012 | { | |
2013 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
2014 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
2015 | ||
2016 | if (stop) { | |
2017 | /* | |
2018 | * Make it appear like a SCHED_FIFO task, its something | |
2019 | * userspace knows about and won't get confused about. | |
2020 | * | |
2021 | * Also, it will make PI more or less work without too | |
2022 | * much confusion -- but then, stop work should not | |
2023 | * rely on PI working anyway. | |
2024 | */ | |
2025 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
2026 | ||
2027 | stop->sched_class = &stop_sched_class; | |
2028 | } | |
2029 | ||
2030 | cpu_rq(cpu)->stop = stop; | |
2031 | ||
2032 | if (old_stop) { | |
2033 | /* | |
2034 | * Reset it back to a normal scheduling class so that | |
2035 | * it can die in pieces. | |
2036 | */ | |
2037 | old_stop->sched_class = &rt_sched_class; | |
2038 | } | |
2039 | } | |
2040 | ||
14531189 | 2041 | /* |
dd41f596 | 2042 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 2043 | */ |
14531189 IM |
2044 | static inline int __normal_prio(struct task_struct *p) |
2045 | { | |
dd41f596 | 2046 | return p->static_prio; |
14531189 IM |
2047 | } |
2048 | ||
b29739f9 IM |
2049 | /* |
2050 | * Calculate the expected normal priority: i.e. priority | |
2051 | * without taking RT-inheritance into account. Might be | |
2052 | * boosted by interactivity modifiers. Changes upon fork, | |
2053 | * setprio syscalls, and whenever the interactivity | |
2054 | * estimator recalculates. | |
2055 | */ | |
36c8b586 | 2056 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
2057 | { |
2058 | int prio; | |
2059 | ||
e05606d3 | 2060 | if (task_has_rt_policy(p)) |
b29739f9 IM |
2061 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
2062 | else | |
2063 | prio = __normal_prio(p); | |
2064 | return prio; | |
2065 | } | |
2066 | ||
2067 | /* | |
2068 | * Calculate the current priority, i.e. the priority | |
2069 | * taken into account by the scheduler. This value might | |
2070 | * be boosted by RT tasks, or might be boosted by | |
2071 | * interactivity modifiers. Will be RT if the task got | |
2072 | * RT-boosted. If not then it returns p->normal_prio. | |
2073 | */ | |
36c8b586 | 2074 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
2075 | { |
2076 | p->normal_prio = normal_prio(p); | |
2077 | /* | |
2078 | * If we are RT tasks or we were boosted to RT priority, | |
2079 | * keep the priority unchanged. Otherwise, update priority | |
2080 | * to the normal priority: | |
2081 | */ | |
2082 | if (!rt_prio(p->prio)) | |
2083 | return p->normal_prio; | |
2084 | return p->prio; | |
2085 | } | |
2086 | ||
1da177e4 LT |
2087 | /** |
2088 | * task_curr - is this task currently executing on a CPU? | |
2089 | * @p: the task in question. | |
2090 | */ | |
36c8b586 | 2091 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
2092 | { |
2093 | return cpu_curr(task_cpu(p)) == p; | |
2094 | } | |
2095 | ||
cb469845 SR |
2096 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
2097 | const struct sched_class *prev_class, | |
da7a735e | 2098 | int oldprio) |
cb469845 SR |
2099 | { |
2100 | if (prev_class != p->sched_class) { | |
2101 | if (prev_class->switched_from) | |
da7a735e PZ |
2102 | prev_class->switched_from(rq, p); |
2103 | p->sched_class->switched_to(rq, p); | |
2104 | } else if (oldprio != p->prio) | |
2105 | p->sched_class->prio_changed(rq, p, oldprio); | |
cb469845 SR |
2106 | } |
2107 | ||
1e5a7405 PZ |
2108 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
2109 | { | |
2110 | const struct sched_class *class; | |
2111 | ||
2112 | if (p->sched_class == rq->curr->sched_class) { | |
2113 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
2114 | } else { | |
2115 | for_each_class(class) { | |
2116 | if (class == rq->curr->sched_class) | |
2117 | break; | |
2118 | if (class == p->sched_class) { | |
2119 | resched_task(rq->curr); | |
2120 | break; | |
2121 | } | |
2122 | } | |
2123 | } | |
2124 | ||
2125 | /* | |
2126 | * A queue event has occurred, and we're going to schedule. In | |
2127 | * this case, we can save a useless back to back clock update. | |
2128 | */ | |
fd2f4419 | 2129 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
2130 | rq->skip_clock_update = 1; |
2131 | } | |
2132 | ||
1da177e4 | 2133 | #ifdef CONFIG_SMP |
cc367732 IM |
2134 | /* |
2135 | * Is this task likely cache-hot: | |
2136 | */ | |
e7693a36 | 2137 | static int |
cc367732 IM |
2138 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
2139 | { | |
2140 | s64 delta; | |
2141 | ||
e6c8fba7 PZ |
2142 | if (p->sched_class != &fair_sched_class) |
2143 | return 0; | |
2144 | ||
ef8002f6 NR |
2145 | if (unlikely(p->policy == SCHED_IDLE)) |
2146 | return 0; | |
2147 | ||
f540a608 IM |
2148 | /* |
2149 | * Buddy candidates are cache hot: | |
2150 | */ | |
f685ceac | 2151 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
2152 | (&p->se == cfs_rq_of(&p->se)->next || |
2153 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2154 | return 1; |
2155 | ||
6bc1665b IM |
2156 | if (sysctl_sched_migration_cost == -1) |
2157 | return 1; | |
2158 | if (sysctl_sched_migration_cost == 0) | |
2159 | return 0; | |
2160 | ||
cc367732 IM |
2161 | delta = now - p->se.exec_start; |
2162 | ||
2163 | return delta < (s64)sysctl_sched_migration_cost; | |
2164 | } | |
2165 | ||
dd41f596 | 2166 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2167 | { |
e2912009 PZ |
2168 | #ifdef CONFIG_SCHED_DEBUG |
2169 | /* | |
2170 | * We should never call set_task_cpu() on a blocked task, | |
2171 | * ttwu() will sort out the placement. | |
2172 | */ | |
077614ee PZ |
2173 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
2174 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
e2912009 PZ |
2175 | #endif |
2176 | ||
de1d7286 | 2177 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2178 | |
0c69774e PZ |
2179 | if (task_cpu(p) != new_cpu) { |
2180 | p->se.nr_migrations++; | |
2181 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); | |
2182 | } | |
dd41f596 IM |
2183 | |
2184 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2185 | } |
2186 | ||
969c7921 | 2187 | struct migration_arg { |
36c8b586 | 2188 | struct task_struct *task; |
1da177e4 | 2189 | int dest_cpu; |
70b97a7f | 2190 | }; |
1da177e4 | 2191 | |
969c7921 TH |
2192 | static int migration_cpu_stop(void *data); |
2193 | ||
1da177e4 LT |
2194 | /* |
2195 | * The task's runqueue lock must be held. | |
2196 | * Returns true if you have to wait for migration thread. | |
2197 | */ | |
b7a2b39d | 2198 | static bool migrate_task(struct task_struct *p, struct rq *rq) |
1da177e4 | 2199 | { |
1da177e4 LT |
2200 | /* |
2201 | * If the task is not on a runqueue (and not running), then | |
e2912009 | 2202 | * the next wake-up will properly place the task. |
1da177e4 | 2203 | */ |
fd2f4419 | 2204 | return p->on_rq || task_running(rq, p); |
1da177e4 LT |
2205 | } |
2206 | ||
2207 | /* | |
2208 | * wait_task_inactive - wait for a thread to unschedule. | |
2209 | * | |
85ba2d86 RM |
2210 | * If @match_state is nonzero, it's the @p->state value just checked and |
2211 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2212 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2213 | * we return a positive number (its total switch count). If a second call | |
2214 | * a short while later returns the same number, the caller can be sure that | |
2215 | * @p has remained unscheduled the whole time. | |
2216 | * | |
1da177e4 LT |
2217 | * The caller must ensure that the task *will* unschedule sometime soon, |
2218 | * else this function might spin for a *long* time. This function can't | |
2219 | * be called with interrupts off, or it may introduce deadlock with | |
2220 | * smp_call_function() if an IPI is sent by the same process we are | |
2221 | * waiting to become inactive. | |
2222 | */ | |
85ba2d86 | 2223 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2224 | { |
2225 | unsigned long flags; | |
dd41f596 | 2226 | int running, on_rq; |
85ba2d86 | 2227 | unsigned long ncsw; |
70b97a7f | 2228 | struct rq *rq; |
1da177e4 | 2229 | |
3a5c359a AK |
2230 | for (;;) { |
2231 | /* | |
2232 | * We do the initial early heuristics without holding | |
2233 | * any task-queue locks at all. We'll only try to get | |
2234 | * the runqueue lock when things look like they will | |
2235 | * work out! | |
2236 | */ | |
2237 | rq = task_rq(p); | |
fa490cfd | 2238 | |
3a5c359a AK |
2239 | /* |
2240 | * If the task is actively running on another CPU | |
2241 | * still, just relax and busy-wait without holding | |
2242 | * any locks. | |
2243 | * | |
2244 | * NOTE! Since we don't hold any locks, it's not | |
2245 | * even sure that "rq" stays as the right runqueue! | |
2246 | * But we don't care, since "task_running()" will | |
2247 | * return false if the runqueue has changed and p | |
2248 | * is actually now running somewhere else! | |
2249 | */ | |
85ba2d86 RM |
2250 | while (task_running(rq, p)) { |
2251 | if (match_state && unlikely(p->state != match_state)) | |
2252 | return 0; | |
3a5c359a | 2253 | cpu_relax(); |
85ba2d86 | 2254 | } |
fa490cfd | 2255 | |
3a5c359a AK |
2256 | /* |
2257 | * Ok, time to look more closely! We need the rq | |
2258 | * lock now, to be *sure*. If we're wrong, we'll | |
2259 | * just go back and repeat. | |
2260 | */ | |
2261 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 2262 | trace_sched_wait_task(p); |
3a5c359a | 2263 | running = task_running(rq, p); |
fd2f4419 | 2264 | on_rq = p->on_rq; |
85ba2d86 | 2265 | ncsw = 0; |
f31e11d8 | 2266 | if (!match_state || p->state == match_state) |
93dcf55f | 2267 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2268 | task_rq_unlock(rq, &flags); |
fa490cfd | 2269 | |
85ba2d86 RM |
2270 | /* |
2271 | * If it changed from the expected state, bail out now. | |
2272 | */ | |
2273 | if (unlikely(!ncsw)) | |
2274 | break; | |
2275 | ||
3a5c359a AK |
2276 | /* |
2277 | * Was it really running after all now that we | |
2278 | * checked with the proper locks actually held? | |
2279 | * | |
2280 | * Oops. Go back and try again.. | |
2281 | */ | |
2282 | if (unlikely(running)) { | |
2283 | cpu_relax(); | |
2284 | continue; | |
2285 | } | |
fa490cfd | 2286 | |
3a5c359a AK |
2287 | /* |
2288 | * It's not enough that it's not actively running, | |
2289 | * it must be off the runqueue _entirely_, and not | |
2290 | * preempted! | |
2291 | * | |
80dd99b3 | 2292 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2293 | * running right now), it's preempted, and we should |
2294 | * yield - it could be a while. | |
2295 | */ | |
2296 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
2297 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
2298 | ||
2299 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2300 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
2301 | continue; |
2302 | } | |
fa490cfd | 2303 | |
3a5c359a AK |
2304 | /* |
2305 | * Ahh, all good. It wasn't running, and it wasn't | |
2306 | * runnable, which means that it will never become | |
2307 | * running in the future either. We're all done! | |
2308 | */ | |
2309 | break; | |
2310 | } | |
85ba2d86 RM |
2311 | |
2312 | return ncsw; | |
1da177e4 LT |
2313 | } |
2314 | ||
2315 | /*** | |
2316 | * kick_process - kick a running thread to enter/exit the kernel | |
2317 | * @p: the to-be-kicked thread | |
2318 | * | |
2319 | * Cause a process which is running on another CPU to enter | |
2320 | * kernel-mode, without any delay. (to get signals handled.) | |
2321 | * | |
25985edc | 2322 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
2323 | * because all it wants to ensure is that the remote task enters |
2324 | * the kernel. If the IPI races and the task has been migrated | |
2325 | * to another CPU then no harm is done and the purpose has been | |
2326 | * achieved as well. | |
2327 | */ | |
36c8b586 | 2328 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2329 | { |
2330 | int cpu; | |
2331 | ||
2332 | preempt_disable(); | |
2333 | cpu = task_cpu(p); | |
2334 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2335 | smp_send_reschedule(cpu); | |
2336 | preempt_enable(); | |
2337 | } | |
b43e3521 | 2338 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2339 | #endif /* CONFIG_SMP */ |
1da177e4 | 2340 | |
970b13ba | 2341 | #ifdef CONFIG_SMP |
30da688e ON |
2342 | /* |
2343 | * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held. | |
2344 | */ | |
5da9a0fb PZ |
2345 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2346 | { | |
2347 | int dest_cpu; | |
2348 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
2349 | ||
2350 | /* Look for allowed, online CPU in same node. */ | |
2351 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
2352 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
2353 | return dest_cpu; | |
2354 | ||
2355 | /* Any allowed, online CPU? */ | |
2356 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); | |
2357 | if (dest_cpu < nr_cpu_ids) | |
2358 | return dest_cpu; | |
2359 | ||
2360 | /* No more Mr. Nice Guy. */ | |
48c5ccae PZ |
2361 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
2362 | /* | |
2363 | * Don't tell them about moving exiting tasks or | |
2364 | * kernel threads (both mm NULL), since they never | |
2365 | * leave kernel. | |
2366 | */ | |
2367 | if (p->mm && printk_ratelimit()) { | |
2368 | printk(KERN_INFO "process %d (%s) no longer affine to cpu%d\n", | |
2369 | task_pid_nr(p), p->comm, cpu); | |
5da9a0fb PZ |
2370 | } |
2371 | ||
2372 | return dest_cpu; | |
2373 | } | |
2374 | ||
e2912009 | 2375 | /* |
30da688e | 2376 | * The caller (fork, wakeup) owns TASK_WAKING, ->cpus_allowed is stable. |
e2912009 | 2377 | */ |
970b13ba | 2378 | static inline |
0017d735 | 2379 | int select_task_rq(struct rq *rq, struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 2380 | { |
0017d735 | 2381 | int cpu = p->sched_class->select_task_rq(rq, p, sd_flags, wake_flags); |
e2912009 PZ |
2382 | |
2383 | /* | |
2384 | * In order not to call set_task_cpu() on a blocking task we need | |
2385 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2386 | * cpu. | |
2387 | * | |
2388 | * Since this is common to all placement strategies, this lives here. | |
2389 | * | |
2390 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2391 | * not worry about this generic constraint ] | |
2392 | */ | |
2393 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
70f11205 | 2394 | !cpu_online(cpu))) |
5da9a0fb | 2395 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2396 | |
2397 | return cpu; | |
970b13ba | 2398 | } |
09a40af5 MG |
2399 | |
2400 | static void update_avg(u64 *avg, u64 sample) | |
2401 | { | |
2402 | s64 diff = sample - *avg; | |
2403 | *avg += diff >> 3; | |
2404 | } | |
970b13ba PZ |
2405 | #endif |
2406 | ||
d7c01d27 PZ |
2407 | static void |
2408 | ttwu_stat(struct rq *rq, struct task_struct *p, int cpu, int wake_flags) | |
9ed3811a | 2409 | { |
d7c01d27 PZ |
2410 | #ifdef CONFIG_SCHEDSTATS |
2411 | #ifdef CONFIG_SMP | |
2412 | int this_cpu = smp_processor_id(); | |
2413 | ||
2414 | if (cpu == this_cpu) { | |
2415 | schedstat_inc(rq, ttwu_local); | |
2416 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
2417 | } else { | |
2418 | struct sched_domain *sd; | |
2419 | ||
2420 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
2421 | for_each_domain(this_cpu, sd) { | |
2422 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
2423 | schedstat_inc(sd, ttwu_wake_remote); | |
2424 | break; | |
2425 | } | |
2426 | } | |
2427 | } | |
2428 | #endif /* CONFIG_SMP */ | |
2429 | ||
2430 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 2431 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
2432 | |
2433 | if (wake_flags & WF_SYNC) | |
9ed3811a | 2434 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 PZ |
2435 | |
2436 | if (cpu != task_cpu(p)) | |
9ed3811a | 2437 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); |
9ed3811a | 2438 | |
d7c01d27 PZ |
2439 | #endif /* CONFIG_SCHEDSTATS */ |
2440 | } | |
2441 | ||
2442 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
2443 | { | |
9ed3811a | 2444 | activate_task(rq, p, en_flags); |
fd2f4419 | 2445 | p->on_rq = 1; |
c2f7115e PZ |
2446 | |
2447 | /* if a worker is waking up, notify workqueue */ | |
2448 | if (p->flags & PF_WQ_WORKER) | |
2449 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
2450 | } |
2451 | ||
89363381 PZ |
2452 | static void |
2453 | ttwu_post_activation(struct task_struct *p, struct rq *rq, int wake_flags) | |
9ed3811a | 2454 | { |
89363381 | 2455 | trace_sched_wakeup(p, true); |
9ed3811a TH |
2456 | check_preempt_curr(rq, p, wake_flags); |
2457 | ||
2458 | p->state = TASK_RUNNING; | |
2459 | #ifdef CONFIG_SMP | |
2460 | if (p->sched_class->task_woken) | |
2461 | p->sched_class->task_woken(rq, p); | |
2462 | ||
2463 | if (unlikely(rq->idle_stamp)) { | |
2464 | u64 delta = rq->clock - rq->idle_stamp; | |
2465 | u64 max = 2*sysctl_sched_migration_cost; | |
2466 | ||
2467 | if (delta > max) | |
2468 | rq->avg_idle = max; | |
2469 | else | |
2470 | update_avg(&rq->avg_idle, delta); | |
2471 | rq->idle_stamp = 0; | |
2472 | } | |
2473 | #endif | |
2474 | } | |
2475 | ||
2476 | /** | |
1da177e4 | 2477 | * try_to_wake_up - wake up a thread |
9ed3811a | 2478 | * @p: the thread to be awakened |
1da177e4 | 2479 | * @state: the mask of task states that can be woken |
9ed3811a | 2480 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
2481 | * |
2482 | * Put it on the run-queue if it's not already there. The "current" | |
2483 | * thread is always on the run-queue (except when the actual | |
2484 | * re-schedule is in progress), and as such you're allowed to do | |
2485 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2486 | * runnable without the overhead of this. | |
2487 | * | |
9ed3811a TH |
2488 | * Returns %true if @p was woken up, %false if it was already running |
2489 | * or @state didn't match @p's state. | |
1da177e4 | 2490 | */ |
7d478721 PZ |
2491 | static int try_to_wake_up(struct task_struct *p, unsigned int state, |
2492 | int wake_flags) | |
1da177e4 | 2493 | { |
cc367732 | 2494 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 | 2495 | unsigned long flags; |
371fd7e7 | 2496 | unsigned long en_flags = ENQUEUE_WAKEUP; |
ab3b3aa5 | 2497 | struct rq *rq; |
1da177e4 | 2498 | |
e9c84311 | 2499 | this_cpu = get_cpu(); |
2398f2c6 | 2500 | |
04e2f174 | 2501 | smp_wmb(); |
ab3b3aa5 | 2502 | rq = task_rq_lock(p, &flags); |
e9c84311 | 2503 | if (!(p->state & state)) |
1da177e4 LT |
2504 | goto out; |
2505 | ||
d7c01d27 PZ |
2506 | cpu = task_cpu(p); |
2507 | ||
fd2f4419 | 2508 | if (p->on_rq) |
1da177e4 LT |
2509 | goto out_running; |
2510 | ||
cc367732 | 2511 | orig_cpu = cpu; |
1da177e4 LT |
2512 | #ifdef CONFIG_SMP |
2513 | if (unlikely(task_running(rq, p))) | |
2514 | goto out_activate; | |
2515 | ||
e9c84311 PZ |
2516 | /* |
2517 | * In order to handle concurrent wakeups and release the rq->lock | |
2518 | * we put the task in TASK_WAKING state. | |
eb24073b IM |
2519 | * |
2520 | * First fix up the nr_uninterruptible count: | |
e9c84311 | 2521 | */ |
cc87f76a PZ |
2522 | if (task_contributes_to_load(p)) { |
2523 | if (likely(cpu_online(orig_cpu))) | |
2524 | rq->nr_uninterruptible--; | |
2525 | else | |
2526 | this_rq()->nr_uninterruptible--; | |
2527 | } | |
e9c84311 | 2528 | p->state = TASK_WAKING; |
efbbd05a | 2529 | |
371fd7e7 | 2530 | if (p->sched_class->task_waking) { |
efbbd05a | 2531 | p->sched_class->task_waking(rq, p); |
371fd7e7 PZ |
2532 | en_flags |= ENQUEUE_WAKING; |
2533 | } | |
efbbd05a | 2534 | |
0017d735 PZ |
2535 | cpu = select_task_rq(rq, p, SD_BALANCE_WAKE, wake_flags); |
2536 | if (cpu != orig_cpu) | |
5d2f5a61 | 2537 | set_task_cpu(p, cpu); |
0017d735 | 2538 | __task_rq_unlock(rq); |
ab19cb23 | 2539 | |
0970d299 PZ |
2540 | rq = cpu_rq(cpu); |
2541 | raw_spin_lock(&rq->lock); | |
f5dc3753 | 2542 | |
0970d299 PZ |
2543 | /* |
2544 | * We migrated the task without holding either rq->lock, however | |
2545 | * since the task is not on the task list itself, nobody else | |
2546 | * will try and migrate the task, hence the rq should match the | |
2547 | * cpu we just moved it to. | |
2548 | */ | |
2549 | WARN_ON(task_cpu(p) != cpu); | |
e9c84311 | 2550 | WARN_ON(p->state != TASK_WAKING); |
1da177e4 LT |
2551 | |
2552 | out_activate: | |
2553 | #endif /* CONFIG_SMP */ | |
d7c01d27 | 2554 | ttwu_activate(rq, p, en_flags); |
1da177e4 | 2555 | out_running: |
89363381 | 2556 | ttwu_post_activation(p, rq, wake_flags); |
d7c01d27 | 2557 | ttwu_stat(rq, p, cpu, wake_flags); |
89363381 | 2558 | success = 1; |
1da177e4 LT |
2559 | out: |
2560 | task_rq_unlock(rq, &flags); | |
e9c84311 | 2561 | put_cpu(); |
1da177e4 LT |
2562 | |
2563 | return success; | |
2564 | } | |
2565 | ||
21aa9af0 TH |
2566 | /** |
2567 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
2568 | * @p: the thread to be awakened | |
2569 | * | |
b595076a | 2570 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 TH |
2571 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2572 | * the current task. this_rq() stays locked over invocation. | |
2573 | */ | |
2574 | static void try_to_wake_up_local(struct task_struct *p) | |
2575 | { | |
2576 | struct rq *rq = task_rq(p); | |
21aa9af0 TH |
2577 | |
2578 | BUG_ON(rq != this_rq()); | |
2579 | BUG_ON(p == current); | |
2580 | lockdep_assert_held(&rq->lock); | |
2581 | ||
2582 | if (!(p->state & TASK_NORMAL)) | |
2583 | return; | |
2584 | ||
fd2f4419 | 2585 | if (!p->on_rq) |
d7c01d27 PZ |
2586 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
2587 | ||
89363381 | 2588 | ttwu_post_activation(p, rq, 0); |
d7c01d27 | 2589 | ttwu_stat(rq, p, smp_processor_id(), 0); |
21aa9af0 TH |
2590 | } |
2591 | ||
50fa610a DH |
2592 | /** |
2593 | * wake_up_process - Wake up a specific process | |
2594 | * @p: The process to be woken up. | |
2595 | * | |
2596 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2597 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2598 | * running. | |
2599 | * | |
2600 | * It may be assumed that this function implies a write memory barrier before | |
2601 | * changing the task state if and only if any tasks are woken up. | |
2602 | */ | |
7ad5b3a5 | 2603 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2604 | { |
d9514f6c | 2605 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2606 | } |
1da177e4 LT |
2607 | EXPORT_SYMBOL(wake_up_process); |
2608 | ||
7ad5b3a5 | 2609 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2610 | { |
2611 | return try_to_wake_up(p, state, 0); | |
2612 | } | |
2613 | ||
1da177e4 LT |
2614 | /* |
2615 | * Perform scheduler related setup for a newly forked process p. | |
2616 | * p is forked by current. | |
dd41f596 IM |
2617 | * |
2618 | * __sched_fork() is basic setup used by init_idle() too: | |
2619 | */ | |
2620 | static void __sched_fork(struct task_struct *p) | |
2621 | { | |
fd2f4419 PZ |
2622 | p->on_rq = 0; |
2623 | ||
2624 | p->se.on_rq = 0; | |
dd41f596 IM |
2625 | p->se.exec_start = 0; |
2626 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2627 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2628 | p->se.nr_migrations = 0; |
da7a735e | 2629 | p->se.vruntime = 0; |
fd2f4419 | 2630 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
2631 | |
2632 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 2633 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2634 | #endif |
476d139c | 2635 | |
fa717060 | 2636 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 2637 | |
e107be36 AK |
2638 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2639 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2640 | #endif | |
dd41f596 IM |
2641 | } |
2642 | ||
2643 | /* | |
2644 | * fork()/clone()-time setup: | |
2645 | */ | |
2646 | void sched_fork(struct task_struct *p, int clone_flags) | |
2647 | { | |
2648 | int cpu = get_cpu(); | |
2649 | ||
2650 | __sched_fork(p); | |
06b83b5f | 2651 | /* |
0017d735 | 2652 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2653 | * nobody will actually run it, and a signal or other external |
2654 | * event cannot wake it up and insert it on the runqueue either. | |
2655 | */ | |
0017d735 | 2656 | p->state = TASK_RUNNING; |
dd41f596 | 2657 | |
b9dc29e7 MG |
2658 | /* |
2659 | * Revert to default priority/policy on fork if requested. | |
2660 | */ | |
2661 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2662 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2663 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2664 | p->normal_prio = p->static_prio; |
2665 | } | |
b9dc29e7 | 2666 | |
6c697bdf MG |
2667 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2668 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2669 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2670 | set_load_weight(p); |
2671 | } | |
2672 | ||
b9dc29e7 MG |
2673 | /* |
2674 | * We don't need the reset flag anymore after the fork. It has | |
2675 | * fulfilled its duty: | |
2676 | */ | |
2677 | p->sched_reset_on_fork = 0; | |
2678 | } | |
ca94c442 | 2679 | |
f83f9ac2 PW |
2680 | /* |
2681 | * Make sure we do not leak PI boosting priority to the child. | |
2682 | */ | |
2683 | p->prio = current->normal_prio; | |
2684 | ||
2ddbf952 HS |
2685 | if (!rt_prio(p->prio)) |
2686 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2687 | |
cd29fe6f PZ |
2688 | if (p->sched_class->task_fork) |
2689 | p->sched_class->task_fork(p); | |
2690 | ||
86951599 PZ |
2691 | /* |
2692 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2693 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2694 | * is ran before sched_fork(). | |
2695 | * | |
2696 | * Silence PROVE_RCU. | |
2697 | */ | |
2698 | rcu_read_lock(); | |
5f3edc1b | 2699 | set_task_cpu(p, cpu); |
86951599 | 2700 | rcu_read_unlock(); |
5f3edc1b | 2701 | |
52f17b6c | 2702 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2703 | if (likely(sched_info_on())) |
52f17b6c | 2704 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2705 | #endif |
3ca7a440 PZ |
2706 | #if defined(CONFIG_SMP) |
2707 | p->on_cpu = 0; | |
4866cde0 | 2708 | #endif |
1da177e4 | 2709 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2710 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2711 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2712 | #endif |
806c09a7 | 2713 | #ifdef CONFIG_SMP |
917b627d | 2714 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
806c09a7 | 2715 | #endif |
917b627d | 2716 | |
476d139c | 2717 | put_cpu(); |
1da177e4 LT |
2718 | } |
2719 | ||
2720 | /* | |
2721 | * wake_up_new_task - wake up a newly created task for the first time. | |
2722 | * | |
2723 | * This function will do some initial scheduler statistics housekeeping | |
2724 | * that must be done for every newly created context, then puts the task | |
2725 | * on the runqueue and wakes it. | |
2726 | */ | |
7ad5b3a5 | 2727 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2728 | { |
2729 | unsigned long flags; | |
dd41f596 | 2730 | struct rq *rq; |
c890692b | 2731 | int cpu __maybe_unused = get_cpu(); |
fabf318e PZ |
2732 | |
2733 | #ifdef CONFIG_SMP | |
0017d735 PZ |
2734 | rq = task_rq_lock(p, &flags); |
2735 | p->state = TASK_WAKING; | |
2736 | ||
fabf318e PZ |
2737 | /* |
2738 | * Fork balancing, do it here and not earlier because: | |
2739 | * - cpus_allowed can change in the fork path | |
2740 | * - any previously selected cpu might disappear through hotplug | |
2741 | * | |
0017d735 PZ |
2742 | * We set TASK_WAKING so that select_task_rq() can drop rq->lock |
2743 | * without people poking at ->cpus_allowed. | |
fabf318e | 2744 | */ |
0017d735 | 2745 | cpu = select_task_rq(rq, p, SD_BALANCE_FORK, 0); |
fabf318e | 2746 | set_task_cpu(p, cpu); |
1da177e4 | 2747 | |
06b83b5f | 2748 | p->state = TASK_RUNNING; |
0017d735 PZ |
2749 | task_rq_unlock(rq, &flags); |
2750 | #endif | |
2751 | ||
2752 | rq = task_rq_lock(p, &flags); | |
cd29fe6f | 2753 | activate_task(rq, p, 0); |
fd2f4419 | 2754 | p->on_rq = 1; |
89363381 | 2755 | trace_sched_wakeup_new(p, true); |
a7558e01 | 2756 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2757 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2758 | if (p->sched_class->task_woken) |
2759 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2760 | #endif |
dd41f596 | 2761 | task_rq_unlock(rq, &flags); |
fabf318e | 2762 | put_cpu(); |
1da177e4 LT |
2763 | } |
2764 | ||
e107be36 AK |
2765 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2766 | ||
2767 | /** | |
80dd99b3 | 2768 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2769 | * @notifier: notifier struct to register |
e107be36 AK |
2770 | */ |
2771 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2772 | { | |
2773 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2774 | } | |
2775 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2776 | ||
2777 | /** | |
2778 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2779 | * @notifier: notifier struct to unregister |
e107be36 AK |
2780 | * |
2781 | * This is safe to call from within a preemption notifier. | |
2782 | */ | |
2783 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2784 | { | |
2785 | hlist_del(¬ifier->link); | |
2786 | } | |
2787 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2788 | ||
2789 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2790 | { | |
2791 | struct preempt_notifier *notifier; | |
2792 | struct hlist_node *node; | |
2793 | ||
2794 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2795 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2796 | } | |
2797 | ||
2798 | static void | |
2799 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2800 | struct task_struct *next) | |
2801 | { | |
2802 | struct preempt_notifier *notifier; | |
2803 | struct hlist_node *node; | |
2804 | ||
2805 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2806 | notifier->ops->sched_out(notifier, next); | |
2807 | } | |
2808 | ||
6d6bc0ad | 2809 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2810 | |
2811 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2812 | { | |
2813 | } | |
2814 | ||
2815 | static void | |
2816 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2817 | struct task_struct *next) | |
2818 | { | |
2819 | } | |
2820 | ||
6d6bc0ad | 2821 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2822 | |
4866cde0 NP |
2823 | /** |
2824 | * prepare_task_switch - prepare to switch tasks | |
2825 | * @rq: the runqueue preparing to switch | |
421cee29 | 2826 | * @prev: the current task that is being switched out |
4866cde0 NP |
2827 | * @next: the task we are going to switch to. |
2828 | * | |
2829 | * This is called with the rq lock held and interrupts off. It must | |
2830 | * be paired with a subsequent finish_task_switch after the context | |
2831 | * switch. | |
2832 | * | |
2833 | * prepare_task_switch sets up locking and calls architecture specific | |
2834 | * hooks. | |
2835 | */ | |
e107be36 AK |
2836 | static inline void |
2837 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2838 | struct task_struct *next) | |
4866cde0 | 2839 | { |
fe4b04fa PZ |
2840 | sched_info_switch(prev, next); |
2841 | perf_event_task_sched_out(prev, next); | |
e107be36 | 2842 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2843 | prepare_lock_switch(rq, next); |
2844 | prepare_arch_switch(next); | |
fe4b04fa | 2845 | trace_sched_switch(prev, next); |
4866cde0 NP |
2846 | } |
2847 | ||
1da177e4 LT |
2848 | /** |
2849 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2850 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2851 | * @prev: the thread we just switched away from. |
2852 | * | |
4866cde0 NP |
2853 | * finish_task_switch must be called after the context switch, paired |
2854 | * with a prepare_task_switch call before the context switch. | |
2855 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2856 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2857 | * |
2858 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2859 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2860 | * with the lock held can cause deadlocks; see schedule() for |
2861 | * details.) | |
2862 | */ | |
a9957449 | 2863 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2864 | __releases(rq->lock) |
2865 | { | |
1da177e4 | 2866 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2867 | long prev_state; |
1da177e4 LT |
2868 | |
2869 | rq->prev_mm = NULL; | |
2870 | ||
2871 | /* | |
2872 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2873 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2874 | * schedule one last time. The schedule call will never return, and |
2875 | * the scheduled task must drop that reference. | |
c394cc9f | 2876 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2877 | * still held, otherwise prev could be scheduled on another cpu, die |
2878 | * there before we look at prev->state, and then the reference would | |
2879 | * be dropped twice. | |
2880 | * Manfred Spraul <manfred@colorfullife.com> | |
2881 | */ | |
55a101f8 | 2882 | prev_state = prev->state; |
4866cde0 | 2883 | finish_arch_switch(prev); |
8381f65d JI |
2884 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2885 | local_irq_disable(); | |
2886 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
49f47433 | 2887 | perf_event_task_sched_in(current); |
8381f65d JI |
2888 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2889 | local_irq_enable(); | |
2890 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 2891 | finish_lock_switch(rq, prev); |
e8fa1362 | 2892 | |
e107be36 | 2893 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2894 | if (mm) |
2895 | mmdrop(mm); | |
c394cc9f | 2896 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2897 | /* |
2898 | * Remove function-return probe instances associated with this | |
2899 | * task and put them back on the free list. | |
9761eea8 | 2900 | */ |
c6fd91f0 | 2901 | kprobe_flush_task(prev); |
1da177e4 | 2902 | put_task_struct(prev); |
c6fd91f0 | 2903 | } |
1da177e4 LT |
2904 | } |
2905 | ||
3f029d3c GH |
2906 | #ifdef CONFIG_SMP |
2907 | ||
2908 | /* assumes rq->lock is held */ | |
2909 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2910 | { | |
2911 | if (prev->sched_class->pre_schedule) | |
2912 | prev->sched_class->pre_schedule(rq, prev); | |
2913 | } | |
2914 | ||
2915 | /* rq->lock is NOT held, but preemption is disabled */ | |
2916 | static inline void post_schedule(struct rq *rq) | |
2917 | { | |
2918 | if (rq->post_schedule) { | |
2919 | unsigned long flags; | |
2920 | ||
05fa785c | 2921 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2922 | if (rq->curr->sched_class->post_schedule) |
2923 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2924 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2925 | |
2926 | rq->post_schedule = 0; | |
2927 | } | |
2928 | } | |
2929 | ||
2930 | #else | |
da19ab51 | 2931 | |
3f029d3c GH |
2932 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2933 | { | |
2934 | } | |
2935 | ||
2936 | static inline void post_schedule(struct rq *rq) | |
2937 | { | |
1da177e4 LT |
2938 | } |
2939 | ||
3f029d3c GH |
2940 | #endif |
2941 | ||
1da177e4 LT |
2942 | /** |
2943 | * schedule_tail - first thing a freshly forked thread must call. | |
2944 | * @prev: the thread we just switched away from. | |
2945 | */ | |
36c8b586 | 2946 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2947 | __releases(rq->lock) |
2948 | { | |
70b97a7f IM |
2949 | struct rq *rq = this_rq(); |
2950 | ||
4866cde0 | 2951 | finish_task_switch(rq, prev); |
da19ab51 | 2952 | |
3f029d3c GH |
2953 | /* |
2954 | * FIXME: do we need to worry about rq being invalidated by the | |
2955 | * task_switch? | |
2956 | */ | |
2957 | post_schedule(rq); | |
70b97a7f | 2958 | |
4866cde0 NP |
2959 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2960 | /* In this case, finish_task_switch does not reenable preemption */ | |
2961 | preempt_enable(); | |
2962 | #endif | |
1da177e4 | 2963 | if (current->set_child_tid) |
b488893a | 2964 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2965 | } |
2966 | ||
2967 | /* | |
2968 | * context_switch - switch to the new MM and the new | |
2969 | * thread's register state. | |
2970 | */ | |
dd41f596 | 2971 | static inline void |
70b97a7f | 2972 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2973 | struct task_struct *next) |
1da177e4 | 2974 | { |
dd41f596 | 2975 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2976 | |
e107be36 | 2977 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2978 | |
dd41f596 IM |
2979 | mm = next->mm; |
2980 | oldmm = prev->active_mm; | |
9226d125 ZA |
2981 | /* |
2982 | * For paravirt, this is coupled with an exit in switch_to to | |
2983 | * combine the page table reload and the switch backend into | |
2984 | * one hypercall. | |
2985 | */ | |
224101ed | 2986 | arch_start_context_switch(prev); |
9226d125 | 2987 | |
31915ab4 | 2988 | if (!mm) { |
1da177e4 LT |
2989 | next->active_mm = oldmm; |
2990 | atomic_inc(&oldmm->mm_count); | |
2991 | enter_lazy_tlb(oldmm, next); | |
2992 | } else | |
2993 | switch_mm(oldmm, mm, next); | |
2994 | ||
31915ab4 | 2995 | if (!prev->mm) { |
1da177e4 | 2996 | prev->active_mm = NULL; |
1da177e4 LT |
2997 | rq->prev_mm = oldmm; |
2998 | } | |
3a5f5e48 IM |
2999 | /* |
3000 | * Since the runqueue lock will be released by the next | |
3001 | * task (which is an invalid locking op but in the case | |
3002 | * of the scheduler it's an obvious special-case), so we | |
3003 | * do an early lockdep release here: | |
3004 | */ | |
3005 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 3006 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 3007 | #endif |
1da177e4 LT |
3008 | |
3009 | /* Here we just switch the register state and the stack. */ | |
3010 | switch_to(prev, next, prev); | |
3011 | ||
dd41f596 IM |
3012 | barrier(); |
3013 | /* | |
3014 | * this_rq must be evaluated again because prev may have moved | |
3015 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
3016 | * frame will be invalid. | |
3017 | */ | |
3018 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
3019 | } |
3020 | ||
3021 | /* | |
3022 | * nr_running, nr_uninterruptible and nr_context_switches: | |
3023 | * | |
3024 | * externally visible scheduler statistics: current number of runnable | |
3025 | * threads, current number of uninterruptible-sleeping threads, total | |
3026 | * number of context switches performed since bootup. | |
3027 | */ | |
3028 | unsigned long nr_running(void) | |
3029 | { | |
3030 | unsigned long i, sum = 0; | |
3031 | ||
3032 | for_each_online_cpu(i) | |
3033 | sum += cpu_rq(i)->nr_running; | |
3034 | ||
3035 | return sum; | |
f711f609 | 3036 | } |
1da177e4 LT |
3037 | |
3038 | unsigned long nr_uninterruptible(void) | |
f711f609 | 3039 | { |
1da177e4 | 3040 | unsigned long i, sum = 0; |
f711f609 | 3041 | |
0a945022 | 3042 | for_each_possible_cpu(i) |
1da177e4 | 3043 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
3044 | |
3045 | /* | |
1da177e4 LT |
3046 | * Since we read the counters lockless, it might be slightly |
3047 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 3048 | */ |
1da177e4 LT |
3049 | if (unlikely((long)sum < 0)) |
3050 | sum = 0; | |
f711f609 | 3051 | |
1da177e4 | 3052 | return sum; |
f711f609 | 3053 | } |
f711f609 | 3054 | |
1da177e4 | 3055 | unsigned long long nr_context_switches(void) |
46cb4b7c | 3056 | { |
cc94abfc SR |
3057 | int i; |
3058 | unsigned long long sum = 0; | |
46cb4b7c | 3059 | |
0a945022 | 3060 | for_each_possible_cpu(i) |
1da177e4 | 3061 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 3062 | |
1da177e4 LT |
3063 | return sum; |
3064 | } | |
483b4ee6 | 3065 | |
1da177e4 LT |
3066 | unsigned long nr_iowait(void) |
3067 | { | |
3068 | unsigned long i, sum = 0; | |
483b4ee6 | 3069 | |
0a945022 | 3070 | for_each_possible_cpu(i) |
1da177e4 | 3071 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 3072 | |
1da177e4 LT |
3073 | return sum; |
3074 | } | |
483b4ee6 | 3075 | |
8c215bd3 | 3076 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 3077 | { |
8c215bd3 | 3078 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
3079 | return atomic_read(&this->nr_iowait); |
3080 | } | |
46cb4b7c | 3081 | |
69d25870 AV |
3082 | unsigned long this_cpu_load(void) |
3083 | { | |
3084 | struct rq *this = this_rq(); | |
3085 | return this->cpu_load[0]; | |
3086 | } | |
e790fb0b | 3087 | |
46cb4b7c | 3088 | |
dce48a84 TG |
3089 | /* Variables and functions for calc_load */ |
3090 | static atomic_long_t calc_load_tasks; | |
3091 | static unsigned long calc_load_update; | |
3092 | unsigned long avenrun[3]; | |
3093 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 3094 | |
74f5187a PZ |
3095 | static long calc_load_fold_active(struct rq *this_rq) |
3096 | { | |
3097 | long nr_active, delta = 0; | |
3098 | ||
3099 | nr_active = this_rq->nr_running; | |
3100 | nr_active += (long) this_rq->nr_uninterruptible; | |
3101 | ||
3102 | if (nr_active != this_rq->calc_load_active) { | |
3103 | delta = nr_active - this_rq->calc_load_active; | |
3104 | this_rq->calc_load_active = nr_active; | |
3105 | } | |
3106 | ||
3107 | return delta; | |
3108 | } | |
3109 | ||
0f004f5a PZ |
3110 | static unsigned long |
3111 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
3112 | { | |
3113 | load *= exp; | |
3114 | load += active * (FIXED_1 - exp); | |
3115 | load += 1UL << (FSHIFT - 1); | |
3116 | return load >> FSHIFT; | |
3117 | } | |
3118 | ||
74f5187a PZ |
3119 | #ifdef CONFIG_NO_HZ |
3120 | /* | |
3121 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | |
3122 | * | |
3123 | * When making the ILB scale, we should try to pull this in as well. | |
3124 | */ | |
3125 | static atomic_long_t calc_load_tasks_idle; | |
3126 | ||
3127 | static void calc_load_account_idle(struct rq *this_rq) | |
3128 | { | |
3129 | long delta; | |
3130 | ||
3131 | delta = calc_load_fold_active(this_rq); | |
3132 | if (delta) | |
3133 | atomic_long_add(delta, &calc_load_tasks_idle); | |
3134 | } | |
3135 | ||
3136 | static long calc_load_fold_idle(void) | |
3137 | { | |
3138 | long delta = 0; | |
3139 | ||
3140 | /* | |
3141 | * Its got a race, we don't care... | |
3142 | */ | |
3143 | if (atomic_long_read(&calc_load_tasks_idle)) | |
3144 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | |
3145 | ||
3146 | return delta; | |
3147 | } | |
0f004f5a PZ |
3148 | |
3149 | /** | |
3150 | * fixed_power_int - compute: x^n, in O(log n) time | |
3151 | * | |
3152 | * @x: base of the power | |
3153 | * @frac_bits: fractional bits of @x | |
3154 | * @n: power to raise @x to. | |
3155 | * | |
3156 | * By exploiting the relation between the definition of the natural power | |
3157 | * function: x^n := x*x*...*x (x multiplied by itself for n times), and | |
3158 | * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i, | |
3159 | * (where: n_i \elem {0, 1}, the binary vector representing n), | |
3160 | * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is | |
3161 | * of course trivially computable in O(log_2 n), the length of our binary | |
3162 | * vector. | |
3163 | */ | |
3164 | static unsigned long | |
3165 | fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n) | |
3166 | { | |
3167 | unsigned long result = 1UL << frac_bits; | |
3168 | ||
3169 | if (n) for (;;) { | |
3170 | if (n & 1) { | |
3171 | result *= x; | |
3172 | result += 1UL << (frac_bits - 1); | |
3173 | result >>= frac_bits; | |
3174 | } | |
3175 | n >>= 1; | |
3176 | if (!n) | |
3177 | break; | |
3178 | x *= x; | |
3179 | x += 1UL << (frac_bits - 1); | |
3180 | x >>= frac_bits; | |
3181 | } | |
3182 | ||
3183 | return result; | |
3184 | } | |
3185 | ||
3186 | /* | |
3187 | * a1 = a0 * e + a * (1 - e) | |
3188 | * | |
3189 | * a2 = a1 * e + a * (1 - e) | |
3190 | * = (a0 * e + a * (1 - e)) * e + a * (1 - e) | |
3191 | * = a0 * e^2 + a * (1 - e) * (1 + e) | |
3192 | * | |
3193 | * a3 = a2 * e + a * (1 - e) | |
3194 | * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e) | |
3195 | * = a0 * e^3 + a * (1 - e) * (1 + e + e^2) | |
3196 | * | |
3197 | * ... | |
3198 | * | |
3199 | * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1] | |
3200 | * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e) | |
3201 | * = a0 * e^n + a * (1 - e^n) | |
3202 | * | |
3203 | * [1] application of the geometric series: | |
3204 | * | |
3205 | * n 1 - x^(n+1) | |
3206 | * S_n := \Sum x^i = ------------- | |
3207 | * i=0 1 - x | |
3208 | */ | |
3209 | static unsigned long | |
3210 | calc_load_n(unsigned long load, unsigned long exp, | |
3211 | unsigned long active, unsigned int n) | |
3212 | { | |
3213 | ||
3214 | return calc_load(load, fixed_power_int(exp, FSHIFT, n), active); | |
3215 | } | |
3216 | ||
3217 | /* | |
3218 | * NO_HZ can leave us missing all per-cpu ticks calling | |
3219 | * calc_load_account_active(), but since an idle CPU folds its delta into | |
3220 | * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold | |
3221 | * in the pending idle delta if our idle period crossed a load cycle boundary. | |
3222 | * | |
3223 | * Once we've updated the global active value, we need to apply the exponential | |
3224 | * weights adjusted to the number of cycles missed. | |
3225 | */ | |
3226 | static void calc_global_nohz(unsigned long ticks) | |
3227 | { | |
3228 | long delta, active, n; | |
3229 | ||
3230 | if (time_before(jiffies, calc_load_update)) | |
3231 | return; | |
3232 | ||
3233 | /* | |
3234 | * If we crossed a calc_load_update boundary, make sure to fold | |
3235 | * any pending idle changes, the respective CPUs might have | |
3236 | * missed the tick driven calc_load_account_active() update | |
3237 | * due to NO_HZ. | |
3238 | */ | |
3239 | delta = calc_load_fold_idle(); | |
3240 | if (delta) | |
3241 | atomic_long_add(delta, &calc_load_tasks); | |
3242 | ||
3243 | /* | |
3244 | * If we were idle for multiple load cycles, apply them. | |
3245 | */ | |
3246 | if (ticks >= LOAD_FREQ) { | |
3247 | n = ticks / LOAD_FREQ; | |
3248 | ||
3249 | active = atomic_long_read(&calc_load_tasks); | |
3250 | active = active > 0 ? active * FIXED_1 : 0; | |
3251 | ||
3252 | avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); | |
3253 | avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); | |
3254 | avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); | |
3255 | ||
3256 | calc_load_update += n * LOAD_FREQ; | |
3257 | } | |
3258 | ||
3259 | /* | |
3260 | * Its possible the remainder of the above division also crosses | |
3261 | * a LOAD_FREQ period, the regular check in calc_global_load() | |
3262 | * which comes after this will take care of that. | |
3263 | * | |
3264 | * Consider us being 11 ticks before a cycle completion, and us | |
3265 | * sleeping for 4*LOAD_FREQ + 22 ticks, then the above code will | |
3266 | * age us 4 cycles, and the test in calc_global_load() will | |
3267 | * pick up the final one. | |
3268 | */ | |
3269 | } | |
74f5187a PZ |
3270 | #else |
3271 | static void calc_load_account_idle(struct rq *this_rq) | |
3272 | { | |
3273 | } | |
3274 | ||
3275 | static inline long calc_load_fold_idle(void) | |
3276 | { | |
3277 | return 0; | |
3278 | } | |
0f004f5a PZ |
3279 | |
3280 | static void calc_global_nohz(unsigned long ticks) | |
3281 | { | |
3282 | } | |
74f5187a PZ |
3283 | #endif |
3284 | ||
2d02494f TG |
3285 | /** |
3286 | * get_avenrun - get the load average array | |
3287 | * @loads: pointer to dest load array | |
3288 | * @offset: offset to add | |
3289 | * @shift: shift count to shift the result left | |
3290 | * | |
3291 | * These values are estimates at best, so no need for locking. | |
3292 | */ | |
3293 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
3294 | { | |
3295 | loads[0] = (avenrun[0] + offset) << shift; | |
3296 | loads[1] = (avenrun[1] + offset) << shift; | |
3297 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 3298 | } |
46cb4b7c | 3299 | |
46cb4b7c | 3300 | /* |
dce48a84 TG |
3301 | * calc_load - update the avenrun load estimates 10 ticks after the |
3302 | * CPUs have updated calc_load_tasks. | |
7835b98b | 3303 | */ |
0f004f5a | 3304 | void calc_global_load(unsigned long ticks) |
7835b98b | 3305 | { |
dce48a84 | 3306 | long active; |
1da177e4 | 3307 | |
0f004f5a PZ |
3308 | calc_global_nohz(ticks); |
3309 | ||
3310 | if (time_before(jiffies, calc_load_update + 10)) | |
dce48a84 | 3311 | return; |
1da177e4 | 3312 | |
dce48a84 TG |
3313 | active = atomic_long_read(&calc_load_tasks); |
3314 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 3315 | |
dce48a84 TG |
3316 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3317 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3318 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 3319 | |
dce48a84 TG |
3320 | calc_load_update += LOAD_FREQ; |
3321 | } | |
1da177e4 | 3322 | |
dce48a84 | 3323 | /* |
74f5187a PZ |
3324 | * Called from update_cpu_load() to periodically update this CPU's |
3325 | * active count. | |
dce48a84 TG |
3326 | */ |
3327 | static void calc_load_account_active(struct rq *this_rq) | |
3328 | { | |
74f5187a | 3329 | long delta; |
08c183f3 | 3330 | |
74f5187a PZ |
3331 | if (time_before(jiffies, this_rq->calc_load_update)) |
3332 | return; | |
783609c6 | 3333 | |
74f5187a PZ |
3334 | delta = calc_load_fold_active(this_rq); |
3335 | delta += calc_load_fold_idle(); | |
3336 | if (delta) | |
dce48a84 | 3337 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
3338 | |
3339 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
3340 | } |
3341 | ||
fdf3e95d VP |
3342 | /* |
3343 | * The exact cpuload at various idx values, calculated at every tick would be | |
3344 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | |
3345 | * | |
3346 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | |
3347 | * on nth tick when cpu may be busy, then we have: | |
3348 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3349 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | |
3350 | * | |
3351 | * decay_load_missed() below does efficient calculation of | |
3352 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3353 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | |
3354 | * | |
3355 | * The calculation is approximated on a 128 point scale. | |
3356 | * degrade_zero_ticks is the number of ticks after which load at any | |
3357 | * particular idx is approximated to be zero. | |
3358 | * degrade_factor is a precomputed table, a row for each load idx. | |
3359 | * Each column corresponds to degradation factor for a power of two ticks, | |
3360 | * based on 128 point scale. | |
3361 | * Example: | |
3362 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | |
3363 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | |
3364 | * | |
3365 | * With this power of 2 load factors, we can degrade the load n times | |
3366 | * by looking at 1 bits in n and doing as many mult/shift instead of | |
3367 | * n mult/shifts needed by the exact degradation. | |
3368 | */ | |
3369 | #define DEGRADE_SHIFT 7 | |
3370 | static const unsigned char | |
3371 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | |
3372 | static const unsigned char | |
3373 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | |
3374 | {0, 0, 0, 0, 0, 0, 0, 0}, | |
3375 | {64, 32, 8, 0, 0, 0, 0, 0}, | |
3376 | {96, 72, 40, 12, 1, 0, 0}, | |
3377 | {112, 98, 75, 43, 15, 1, 0}, | |
3378 | {120, 112, 98, 76, 45, 16, 2} }; | |
3379 | ||
3380 | /* | |
3381 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | |
3382 | * would be when CPU is idle and so we just decay the old load without | |
3383 | * adding any new load. | |
3384 | */ | |
3385 | static unsigned long | |
3386 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | |
3387 | { | |
3388 | int j = 0; | |
3389 | ||
3390 | if (!missed_updates) | |
3391 | return load; | |
3392 | ||
3393 | if (missed_updates >= degrade_zero_ticks[idx]) | |
3394 | return 0; | |
3395 | ||
3396 | if (idx == 1) | |
3397 | return load >> missed_updates; | |
3398 | ||
3399 | while (missed_updates) { | |
3400 | if (missed_updates % 2) | |
3401 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | |
3402 | ||
3403 | missed_updates >>= 1; | |
3404 | j++; | |
3405 | } | |
3406 | return load; | |
3407 | } | |
3408 | ||
46cb4b7c | 3409 | /* |
dd41f596 | 3410 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
3411 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
3412 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 3413 | */ |
dd41f596 | 3414 | static void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 3415 | { |
495eca49 | 3416 | unsigned long this_load = this_rq->load.weight; |
fdf3e95d VP |
3417 | unsigned long curr_jiffies = jiffies; |
3418 | unsigned long pending_updates; | |
dd41f596 | 3419 | int i, scale; |
46cb4b7c | 3420 | |
dd41f596 | 3421 | this_rq->nr_load_updates++; |
46cb4b7c | 3422 | |
fdf3e95d VP |
3423 | /* Avoid repeated calls on same jiffy, when moving in and out of idle */ |
3424 | if (curr_jiffies == this_rq->last_load_update_tick) | |
3425 | return; | |
3426 | ||
3427 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
3428 | this_rq->last_load_update_tick = curr_jiffies; | |
3429 | ||
dd41f596 | 3430 | /* Update our load: */ |
fdf3e95d VP |
3431 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
3432 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 3433 | unsigned long old_load, new_load; |
7d1e6a9b | 3434 | |
dd41f596 | 3435 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 3436 | |
dd41f596 | 3437 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 3438 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 3439 | new_load = this_load; |
a25707f3 IM |
3440 | /* |
3441 | * Round up the averaging division if load is increasing. This | |
3442 | * prevents us from getting stuck on 9 if the load is 10, for | |
3443 | * example. | |
3444 | */ | |
3445 | if (new_load > old_load) | |
fdf3e95d VP |
3446 | new_load += scale - 1; |
3447 | ||
3448 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 3449 | } |
da2b71ed SS |
3450 | |
3451 | sched_avg_update(this_rq); | |
fdf3e95d VP |
3452 | } |
3453 | ||
3454 | static void update_cpu_load_active(struct rq *this_rq) | |
3455 | { | |
3456 | update_cpu_load(this_rq); | |
46cb4b7c | 3457 | |
74f5187a | 3458 | calc_load_account_active(this_rq); |
46cb4b7c SS |
3459 | } |
3460 | ||
dd41f596 | 3461 | #ifdef CONFIG_SMP |
8a0be9ef | 3462 | |
46cb4b7c | 3463 | /* |
38022906 PZ |
3464 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3465 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3466 | */ |
38022906 | 3467 | void sched_exec(void) |
46cb4b7c | 3468 | { |
38022906 | 3469 | struct task_struct *p = current; |
1da177e4 | 3470 | unsigned long flags; |
70b97a7f | 3471 | struct rq *rq; |
0017d735 | 3472 | int dest_cpu; |
46cb4b7c | 3473 | |
1da177e4 | 3474 | rq = task_rq_lock(p, &flags); |
0017d735 PZ |
3475 | dest_cpu = p->sched_class->select_task_rq(rq, p, SD_BALANCE_EXEC, 0); |
3476 | if (dest_cpu == smp_processor_id()) | |
3477 | goto unlock; | |
38022906 | 3478 | |
46cb4b7c | 3479 | /* |
38022906 | 3480 | * select_task_rq() can race against ->cpus_allowed |
46cb4b7c | 3481 | */ |
30da688e | 3482 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed) && |
b7a2b39d | 3483 | likely(cpu_active(dest_cpu)) && migrate_task(p, rq)) { |
969c7921 | 3484 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 3485 | |
1da177e4 | 3486 | task_rq_unlock(rq, &flags); |
969c7921 | 3487 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
3488 | return; |
3489 | } | |
0017d735 | 3490 | unlock: |
1da177e4 | 3491 | task_rq_unlock(rq, &flags); |
1da177e4 | 3492 | } |
dd41f596 | 3493 | |
1da177e4 LT |
3494 | #endif |
3495 | ||
1da177e4 LT |
3496 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3497 | ||
3498 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3499 | ||
3500 | /* | |
c5f8d995 | 3501 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 3502 | * @p in case that task is currently running. |
c5f8d995 HS |
3503 | * |
3504 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 3505 | */ |
c5f8d995 HS |
3506 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
3507 | { | |
3508 | u64 ns = 0; | |
3509 | ||
3510 | if (task_current(rq, p)) { | |
3511 | update_rq_clock(rq); | |
305e6835 | 3512 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
3513 | if ((s64)ns < 0) |
3514 | ns = 0; | |
3515 | } | |
3516 | ||
3517 | return ns; | |
3518 | } | |
3519 | ||
bb34d92f | 3520 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 3521 | { |
1da177e4 | 3522 | unsigned long flags; |
41b86e9c | 3523 | struct rq *rq; |
bb34d92f | 3524 | u64 ns = 0; |
48f24c4d | 3525 | |
41b86e9c | 3526 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
3527 | ns = do_task_delta_exec(p, rq); |
3528 | task_rq_unlock(rq, &flags); | |
1508487e | 3529 | |
c5f8d995 HS |
3530 | return ns; |
3531 | } | |
f06febc9 | 3532 | |
c5f8d995 HS |
3533 | /* |
3534 | * Return accounted runtime for the task. | |
3535 | * In case the task is currently running, return the runtime plus current's | |
3536 | * pending runtime that have not been accounted yet. | |
3537 | */ | |
3538 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3539 | { | |
3540 | unsigned long flags; | |
3541 | struct rq *rq; | |
3542 | u64 ns = 0; | |
3543 | ||
3544 | rq = task_rq_lock(p, &flags); | |
3545 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
3546 | task_rq_unlock(rq, &flags); | |
3547 | ||
3548 | return ns; | |
3549 | } | |
48f24c4d | 3550 | |
c5f8d995 HS |
3551 | /* |
3552 | * Return sum_exec_runtime for the thread group. | |
3553 | * In case the task is currently running, return the sum plus current's | |
3554 | * pending runtime that have not been accounted yet. | |
3555 | * | |
3556 | * Note that the thread group might have other running tasks as well, | |
3557 | * so the return value not includes other pending runtime that other | |
3558 | * running tasks might have. | |
3559 | */ | |
3560 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
3561 | { | |
3562 | struct task_cputime totals; | |
3563 | unsigned long flags; | |
3564 | struct rq *rq; | |
3565 | u64 ns; | |
3566 | ||
3567 | rq = task_rq_lock(p, &flags); | |
3568 | thread_group_cputime(p, &totals); | |
3569 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 3570 | task_rq_unlock(rq, &flags); |
48f24c4d | 3571 | |
1da177e4 LT |
3572 | return ns; |
3573 | } | |
3574 | ||
1da177e4 LT |
3575 | /* |
3576 | * Account user cpu time to a process. | |
3577 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 3578 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 3579 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 3580 | */ |
457533a7 MS |
3581 | void account_user_time(struct task_struct *p, cputime_t cputime, |
3582 | cputime_t cputime_scaled) | |
1da177e4 LT |
3583 | { |
3584 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3585 | cputime64_t tmp; | |
3586 | ||
457533a7 | 3587 | /* Add user time to process. */ |
1da177e4 | 3588 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3589 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3590 | account_group_user_time(p, cputime); |
1da177e4 LT |
3591 | |
3592 | /* Add user time to cpustat. */ | |
3593 | tmp = cputime_to_cputime64(cputime); | |
3594 | if (TASK_NICE(p) > 0) | |
3595 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3596 | else | |
3597 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
3598 | |
3599 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
3600 | /* Account for user time used */ |
3601 | acct_update_integrals(p); | |
1da177e4 LT |
3602 | } |
3603 | ||
94886b84 LV |
3604 | /* |
3605 | * Account guest cpu time to a process. | |
3606 | * @p: the process that the cpu time gets accounted to | |
3607 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 3608 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 3609 | */ |
457533a7 MS |
3610 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
3611 | cputime_t cputime_scaled) | |
94886b84 LV |
3612 | { |
3613 | cputime64_t tmp; | |
3614 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3615 | ||
3616 | tmp = cputime_to_cputime64(cputime); | |
3617 | ||
457533a7 | 3618 | /* Add guest time to process. */ |
94886b84 | 3619 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3620 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3621 | account_group_user_time(p, cputime); |
94886b84 LV |
3622 | p->gtime = cputime_add(p->gtime, cputime); |
3623 | ||
457533a7 | 3624 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
3625 | if (TASK_NICE(p) > 0) { |
3626 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3627 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
3628 | } else { | |
3629 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3630 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3631 | } | |
94886b84 LV |
3632 | } |
3633 | ||
70a89a66 VP |
3634 | /* |
3635 | * Account system cpu time to a process and desired cpustat field | |
3636 | * @p: the process that the cpu time gets accounted to | |
3637 | * @cputime: the cpu time spent in kernel space since the last update | |
3638 | * @cputime_scaled: cputime scaled by cpu frequency | |
3639 | * @target_cputime64: pointer to cpustat field that has to be updated | |
3640 | */ | |
3641 | static inline | |
3642 | void __account_system_time(struct task_struct *p, cputime_t cputime, | |
3643 | cputime_t cputime_scaled, cputime64_t *target_cputime64) | |
3644 | { | |
3645 | cputime64_t tmp = cputime_to_cputime64(cputime); | |
3646 | ||
3647 | /* Add system time to process. */ | |
3648 | p->stime = cputime_add(p->stime, cputime); | |
3649 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); | |
3650 | account_group_system_time(p, cputime); | |
3651 | ||
3652 | /* Add system time to cpustat. */ | |
3653 | *target_cputime64 = cputime64_add(*target_cputime64, tmp); | |
3654 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); | |
3655 | ||
3656 | /* Account for system time used */ | |
3657 | acct_update_integrals(p); | |
3658 | } | |
3659 | ||
1da177e4 LT |
3660 | /* |
3661 | * Account system cpu time to a process. | |
3662 | * @p: the process that the cpu time gets accounted to | |
3663 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3664 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 3665 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
3666 | */ |
3667 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 3668 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
3669 | { |
3670 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70a89a66 | 3671 | cputime64_t *target_cputime64; |
1da177e4 | 3672 | |
983ed7a6 | 3673 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 3674 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
3675 | return; |
3676 | } | |
94886b84 | 3677 | |
1da177e4 | 3678 | if (hardirq_count() - hardirq_offset) |
70a89a66 | 3679 | target_cputime64 = &cpustat->irq; |
75e1056f | 3680 | else if (in_serving_softirq()) |
70a89a66 | 3681 | target_cputime64 = &cpustat->softirq; |
1da177e4 | 3682 | else |
70a89a66 | 3683 | target_cputime64 = &cpustat->system; |
ef12fefa | 3684 | |
70a89a66 | 3685 | __account_system_time(p, cputime, cputime_scaled, target_cputime64); |
1da177e4 LT |
3686 | } |
3687 | ||
c66f08be | 3688 | /* |
1da177e4 | 3689 | * Account for involuntary wait time. |
544b4a1f | 3690 | * @cputime: the cpu time spent in involuntary wait |
c66f08be | 3691 | */ |
79741dd3 | 3692 | void account_steal_time(cputime_t cputime) |
c66f08be | 3693 | { |
79741dd3 MS |
3694 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
3695 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
3696 | ||
3697 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
3698 | } |
3699 | ||
1da177e4 | 3700 | /* |
79741dd3 MS |
3701 | * Account for idle time. |
3702 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 3703 | */ |
79741dd3 | 3704 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
3705 | { |
3706 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 3707 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 3708 | struct rq *rq = this_rq(); |
1da177e4 | 3709 | |
79741dd3 MS |
3710 | if (atomic_read(&rq->nr_iowait) > 0) |
3711 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
3712 | else | |
3713 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
3714 | } |
3715 | ||
79741dd3 MS |
3716 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
3717 | ||
abb74cef VP |
3718 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
3719 | /* | |
3720 | * Account a tick to a process and cpustat | |
3721 | * @p: the process that the cpu time gets accounted to | |
3722 | * @user_tick: is the tick from userspace | |
3723 | * @rq: the pointer to rq | |
3724 | * | |
3725 | * Tick demultiplexing follows the order | |
3726 | * - pending hardirq update | |
3727 | * - pending softirq update | |
3728 | * - user_time | |
3729 | * - idle_time | |
3730 | * - system time | |
3731 | * - check for guest_time | |
3732 | * - else account as system_time | |
3733 | * | |
3734 | * Check for hardirq is done both for system and user time as there is | |
3735 | * no timer going off while we are on hardirq and hence we may never get an | |
3736 | * opportunity to update it solely in system time. | |
3737 | * p->stime and friends are only updated on system time and not on irq | |
3738 | * softirq as those do not count in task exec_runtime any more. | |
3739 | */ | |
3740 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
3741 | struct rq *rq) | |
3742 | { | |
3743 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); | |
3744 | cputime64_t tmp = cputime_to_cputime64(cputime_one_jiffy); | |
3745 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3746 | ||
3747 | if (irqtime_account_hi_update()) { | |
3748 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3749 | } else if (irqtime_account_si_update()) { | |
3750 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
414bee9b VP |
3751 | } else if (this_cpu_ksoftirqd() == p) { |
3752 | /* | |
3753 | * ksoftirqd time do not get accounted in cpu_softirq_time. | |
3754 | * So, we have to handle it separately here. | |
3755 | * Also, p->stime needs to be updated for ksoftirqd. | |
3756 | */ | |
3757 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
3758 | &cpustat->softirq); | |
abb74cef VP |
3759 | } else if (user_tick) { |
3760 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
3761 | } else if (p == rq->idle) { | |
3762 | account_idle_time(cputime_one_jiffy); | |
3763 | } else if (p->flags & PF_VCPU) { /* System time or guest time */ | |
3764 | account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
3765 | } else { | |
3766 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
3767 | &cpustat->system); | |
3768 | } | |
3769 | } | |
3770 | ||
3771 | static void irqtime_account_idle_ticks(int ticks) | |
3772 | { | |
3773 | int i; | |
3774 | struct rq *rq = this_rq(); | |
3775 | ||
3776 | for (i = 0; i < ticks; i++) | |
3777 | irqtime_account_process_tick(current, 0, rq); | |
3778 | } | |
544b4a1f | 3779 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
abb74cef VP |
3780 | static void irqtime_account_idle_ticks(int ticks) {} |
3781 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
3782 | struct rq *rq) {} | |
544b4a1f | 3783 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
79741dd3 MS |
3784 | |
3785 | /* | |
3786 | * Account a single tick of cpu time. | |
3787 | * @p: the process that the cpu time gets accounted to | |
3788 | * @user_tick: indicates if the tick is a user or a system tick | |
3789 | */ | |
3790 | void account_process_tick(struct task_struct *p, int user_tick) | |
3791 | { | |
a42548a1 | 3792 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
3793 | struct rq *rq = this_rq(); |
3794 | ||
abb74cef VP |
3795 | if (sched_clock_irqtime) { |
3796 | irqtime_account_process_tick(p, user_tick, rq); | |
3797 | return; | |
3798 | } | |
3799 | ||
79741dd3 | 3800 | if (user_tick) |
a42548a1 | 3801 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 3802 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 3803 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
3804 | one_jiffy_scaled); |
3805 | else | |
a42548a1 | 3806 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
3807 | } |
3808 | ||
3809 | /* | |
3810 | * Account multiple ticks of steal time. | |
3811 | * @p: the process from which the cpu time has been stolen | |
3812 | * @ticks: number of stolen ticks | |
3813 | */ | |
3814 | void account_steal_ticks(unsigned long ticks) | |
3815 | { | |
3816 | account_steal_time(jiffies_to_cputime(ticks)); | |
3817 | } | |
3818 | ||
3819 | /* | |
3820 | * Account multiple ticks of idle time. | |
3821 | * @ticks: number of stolen ticks | |
3822 | */ | |
3823 | void account_idle_ticks(unsigned long ticks) | |
3824 | { | |
abb74cef VP |
3825 | |
3826 | if (sched_clock_irqtime) { | |
3827 | irqtime_account_idle_ticks(ticks); | |
3828 | return; | |
3829 | } | |
3830 | ||
79741dd3 | 3831 | account_idle_time(jiffies_to_cputime(ticks)); |
1da177e4 LT |
3832 | } |
3833 | ||
79741dd3 MS |
3834 | #endif |
3835 | ||
49048622 BS |
3836 | /* |
3837 | * Use precise platform statistics if available: | |
3838 | */ | |
3839 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 3840 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3841 | { |
d99ca3b9 HS |
3842 | *ut = p->utime; |
3843 | *st = p->stime; | |
49048622 BS |
3844 | } |
3845 | ||
0cf55e1e | 3846 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3847 | { |
0cf55e1e HS |
3848 | struct task_cputime cputime; |
3849 | ||
3850 | thread_group_cputime(p, &cputime); | |
3851 | ||
3852 | *ut = cputime.utime; | |
3853 | *st = cputime.stime; | |
49048622 BS |
3854 | } |
3855 | #else | |
761b1d26 HS |
3856 | |
3857 | #ifndef nsecs_to_cputime | |
b7b20df9 | 3858 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
3859 | #endif |
3860 | ||
d180c5bc | 3861 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3862 | { |
d99ca3b9 | 3863 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
3864 | |
3865 | /* | |
3866 | * Use CFS's precise accounting: | |
3867 | */ | |
d180c5bc | 3868 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
3869 | |
3870 | if (total) { | |
e75e863d | 3871 | u64 temp = rtime; |
d180c5bc | 3872 | |
e75e863d | 3873 | temp *= utime; |
49048622 | 3874 | do_div(temp, total); |
d180c5bc HS |
3875 | utime = (cputime_t)temp; |
3876 | } else | |
3877 | utime = rtime; | |
49048622 | 3878 | |
d180c5bc HS |
3879 | /* |
3880 | * Compare with previous values, to keep monotonicity: | |
3881 | */ | |
761b1d26 | 3882 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 3883 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 3884 | |
d99ca3b9 HS |
3885 | *ut = p->prev_utime; |
3886 | *st = p->prev_stime; | |
49048622 BS |
3887 | } |
3888 | ||
0cf55e1e HS |
3889 | /* |
3890 | * Must be called with siglock held. | |
3891 | */ | |
3892 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 3893 | { |
0cf55e1e HS |
3894 | struct signal_struct *sig = p->signal; |
3895 | struct task_cputime cputime; | |
3896 | cputime_t rtime, utime, total; | |
49048622 | 3897 | |
0cf55e1e | 3898 | thread_group_cputime(p, &cputime); |
49048622 | 3899 | |
0cf55e1e HS |
3900 | total = cputime_add(cputime.utime, cputime.stime); |
3901 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 3902 | |
0cf55e1e | 3903 | if (total) { |
e75e863d | 3904 | u64 temp = rtime; |
49048622 | 3905 | |
e75e863d | 3906 | temp *= cputime.utime; |
0cf55e1e HS |
3907 | do_div(temp, total); |
3908 | utime = (cputime_t)temp; | |
3909 | } else | |
3910 | utime = rtime; | |
3911 | ||
3912 | sig->prev_utime = max(sig->prev_utime, utime); | |
3913 | sig->prev_stime = max(sig->prev_stime, | |
3914 | cputime_sub(rtime, sig->prev_utime)); | |
3915 | ||
3916 | *ut = sig->prev_utime; | |
3917 | *st = sig->prev_stime; | |
49048622 | 3918 | } |
49048622 | 3919 | #endif |
49048622 | 3920 | |
7835b98b CL |
3921 | /* |
3922 | * This function gets called by the timer code, with HZ frequency. | |
3923 | * We call it with interrupts disabled. | |
3924 | * | |
3925 | * It also gets called by the fork code, when changing the parent's | |
3926 | * timeslices. | |
3927 | */ | |
3928 | void scheduler_tick(void) | |
3929 | { | |
7835b98b CL |
3930 | int cpu = smp_processor_id(); |
3931 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3932 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3933 | |
3934 | sched_clock_tick(); | |
dd41f596 | 3935 | |
05fa785c | 3936 | raw_spin_lock(&rq->lock); |
3e51f33f | 3937 | update_rq_clock(rq); |
fdf3e95d | 3938 | update_cpu_load_active(rq); |
fa85ae24 | 3939 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 3940 | raw_spin_unlock(&rq->lock); |
7835b98b | 3941 | |
e9d2b064 | 3942 | perf_event_task_tick(); |
e220d2dc | 3943 | |
e418e1c2 | 3944 | #ifdef CONFIG_SMP |
dd41f596 IM |
3945 | rq->idle_at_tick = idle_cpu(cpu); |
3946 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3947 | #endif |
1da177e4 LT |
3948 | } |
3949 | ||
132380a0 | 3950 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
3951 | { |
3952 | if (in_lock_functions(addr)) { | |
3953 | addr = CALLER_ADDR2; | |
3954 | if (in_lock_functions(addr)) | |
3955 | addr = CALLER_ADDR3; | |
3956 | } | |
3957 | return addr; | |
3958 | } | |
1da177e4 | 3959 | |
7e49fcce SR |
3960 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3961 | defined(CONFIG_PREEMPT_TRACER)) | |
3962 | ||
43627582 | 3963 | void __kprobes add_preempt_count(int val) |
1da177e4 | 3964 | { |
6cd8a4bb | 3965 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3966 | /* |
3967 | * Underflow? | |
3968 | */ | |
9a11b49a IM |
3969 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3970 | return; | |
6cd8a4bb | 3971 | #endif |
1da177e4 | 3972 | preempt_count() += val; |
6cd8a4bb | 3973 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3974 | /* |
3975 | * Spinlock count overflowing soon? | |
3976 | */ | |
33859f7f MOS |
3977 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3978 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
3979 | #endif |
3980 | if (preempt_count() == val) | |
3981 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3982 | } |
3983 | EXPORT_SYMBOL(add_preempt_count); | |
3984 | ||
43627582 | 3985 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 3986 | { |
6cd8a4bb | 3987 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3988 | /* |
3989 | * Underflow? | |
3990 | */ | |
01e3eb82 | 3991 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3992 | return; |
1da177e4 LT |
3993 | /* |
3994 | * Is the spinlock portion underflowing? | |
3995 | */ | |
9a11b49a IM |
3996 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3997 | !(preempt_count() & PREEMPT_MASK))) | |
3998 | return; | |
6cd8a4bb | 3999 | #endif |
9a11b49a | 4000 | |
6cd8a4bb SR |
4001 | if (preempt_count() == val) |
4002 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4003 | preempt_count() -= val; |
4004 | } | |
4005 | EXPORT_SYMBOL(sub_preempt_count); | |
4006 | ||
4007 | #endif | |
4008 | ||
4009 | /* | |
dd41f596 | 4010 | * Print scheduling while atomic bug: |
1da177e4 | 4011 | */ |
dd41f596 | 4012 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4013 | { |
838225b4 SS |
4014 | struct pt_regs *regs = get_irq_regs(); |
4015 | ||
3df0fc5b PZ |
4016 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
4017 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 4018 | |
dd41f596 | 4019 | debug_show_held_locks(prev); |
e21f5b15 | 4020 | print_modules(); |
dd41f596 IM |
4021 | if (irqs_disabled()) |
4022 | print_irqtrace_events(prev); | |
838225b4 SS |
4023 | |
4024 | if (regs) | |
4025 | show_regs(regs); | |
4026 | else | |
4027 | dump_stack(); | |
dd41f596 | 4028 | } |
1da177e4 | 4029 | |
dd41f596 IM |
4030 | /* |
4031 | * Various schedule()-time debugging checks and statistics: | |
4032 | */ | |
4033 | static inline void schedule_debug(struct task_struct *prev) | |
4034 | { | |
1da177e4 | 4035 | /* |
41a2d6cf | 4036 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4037 | * schedule() atomically, we ignore that path for now. |
4038 | * Otherwise, whine if we are scheduling when we should not be. | |
4039 | */ | |
3f33a7ce | 4040 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4041 | __schedule_bug(prev); |
4042 | ||
1da177e4 LT |
4043 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4044 | ||
2d72376b | 4045 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4046 | #ifdef CONFIG_SCHEDSTATS |
4047 | if (unlikely(prev->lock_depth >= 0)) { | |
fce20979 | 4048 | schedstat_inc(this_rq(), rq_sched_info.bkl_count); |
2d72376b | 4049 | schedstat_inc(prev, sched_info.bkl_count); |
b8efb561 IM |
4050 | } |
4051 | #endif | |
dd41f596 IM |
4052 | } |
4053 | ||
6cecd084 | 4054 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 4055 | { |
fd2f4419 | 4056 | if (prev->on_rq) |
a64692a3 | 4057 | update_rq_clock(rq); |
6cecd084 | 4058 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
4059 | } |
4060 | ||
dd41f596 IM |
4061 | /* |
4062 | * Pick up the highest-prio task: | |
4063 | */ | |
4064 | static inline struct task_struct * | |
b67802ea | 4065 | pick_next_task(struct rq *rq) |
dd41f596 | 4066 | { |
5522d5d5 | 4067 | const struct sched_class *class; |
dd41f596 | 4068 | struct task_struct *p; |
1da177e4 LT |
4069 | |
4070 | /* | |
dd41f596 IM |
4071 | * Optimization: we know that if all tasks are in |
4072 | * the fair class we can call that function directly: | |
1da177e4 | 4073 | */ |
dd41f596 | 4074 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4075 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4076 | if (likely(p)) |
4077 | return p; | |
1da177e4 LT |
4078 | } |
4079 | ||
34f971f6 | 4080 | for_each_class(class) { |
fb8d4724 | 4081 | p = class->pick_next_task(rq); |
dd41f596 IM |
4082 | if (p) |
4083 | return p; | |
dd41f596 | 4084 | } |
34f971f6 PZ |
4085 | |
4086 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 4087 | } |
1da177e4 | 4088 | |
dd41f596 IM |
4089 | /* |
4090 | * schedule() is the main scheduler function. | |
4091 | */ | |
ff743345 | 4092 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
4093 | { |
4094 | struct task_struct *prev, *next; | |
67ca7bde | 4095 | unsigned long *switch_count; |
dd41f596 | 4096 | struct rq *rq; |
31656519 | 4097 | int cpu; |
dd41f596 | 4098 | |
ff743345 PZ |
4099 | need_resched: |
4100 | preempt_disable(); | |
dd41f596 IM |
4101 | cpu = smp_processor_id(); |
4102 | rq = cpu_rq(cpu); | |
25502a6c | 4103 | rcu_note_context_switch(cpu); |
dd41f596 | 4104 | prev = rq->curr; |
dd41f596 | 4105 | |
dd41f596 | 4106 | schedule_debug(prev); |
1da177e4 | 4107 | |
31656519 | 4108 | if (sched_feat(HRTICK)) |
f333fdc9 | 4109 | hrtick_clear(rq); |
8f4d37ec | 4110 | |
05fa785c | 4111 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 4112 | |
246d86b5 | 4113 | switch_count = &prev->nivcsw; |
1da177e4 | 4114 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 4115 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 4116 | prev->state = TASK_RUNNING; |
21aa9af0 TH |
4117 | } else { |
4118 | /* | |
4119 | * If a worker is going to sleep, notify and | |
4120 | * ask workqueue whether it wants to wake up a | |
4121 | * task to maintain concurrency. If so, wake | |
4122 | * up the task. | |
4123 | */ | |
4124 | if (prev->flags & PF_WQ_WORKER) { | |
4125 | struct task_struct *to_wakeup; | |
4126 | ||
4127 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
4128 | if (to_wakeup) | |
4129 | try_to_wake_up_local(to_wakeup); | |
4130 | } | |
fd2f4419 | 4131 | |
371fd7e7 | 4132 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
fd2f4419 | 4133 | prev->on_rq = 0; |
6631e635 LT |
4134 | |
4135 | /* | |
4136 | * If we are going to sleep and we have plugged IO queued, make | |
4137 | * sure to submit it to avoid deadlocks. | |
4138 | */ | |
4139 | if (blk_needs_flush_plug(prev)) { | |
4140 | raw_spin_unlock(&rq->lock); | |
4141 | blk_flush_plug(prev); | |
4142 | raw_spin_lock(&rq->lock); | |
4143 | } | |
21aa9af0 | 4144 | } |
dd41f596 | 4145 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4146 | } |
4147 | ||
3f029d3c | 4148 | pre_schedule(rq, prev); |
f65eda4f | 4149 | |
dd41f596 | 4150 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4151 | idle_balance(cpu, rq); |
1da177e4 | 4152 | |
df1c99d4 | 4153 | put_prev_task(rq, prev); |
b67802ea | 4154 | next = pick_next_task(rq); |
f26f9aff MG |
4155 | clear_tsk_need_resched(prev); |
4156 | rq->skip_clock_update = 0; | |
1da177e4 | 4157 | |
1da177e4 | 4158 | if (likely(prev != next)) { |
1da177e4 LT |
4159 | rq->nr_switches++; |
4160 | rq->curr = next; | |
4161 | ++*switch_count; | |
4162 | ||
dd41f596 | 4163 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 4164 | /* |
246d86b5 ON |
4165 | * The context switch have flipped the stack from under us |
4166 | * and restored the local variables which were saved when | |
4167 | * this task called schedule() in the past. prev == current | |
4168 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
4169 | */ |
4170 | cpu = smp_processor_id(); | |
4171 | rq = cpu_rq(cpu); | |
1da177e4 | 4172 | } else |
05fa785c | 4173 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 4174 | |
3f029d3c | 4175 | post_schedule(rq); |
1da177e4 | 4176 | |
1da177e4 | 4177 | preempt_enable_no_resched(); |
ff743345 | 4178 | if (need_resched()) |
1da177e4 LT |
4179 | goto need_resched; |
4180 | } | |
1da177e4 LT |
4181 | EXPORT_SYMBOL(schedule); |
4182 | ||
c08f7829 | 4183 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d | 4184 | |
c6eb3dda PZ |
4185 | static inline bool owner_running(struct mutex *lock, struct task_struct *owner) |
4186 | { | |
4187 | bool ret = false; | |
0d66bf6d | 4188 | |
c6eb3dda PZ |
4189 | rcu_read_lock(); |
4190 | if (lock->owner != owner) | |
4191 | goto fail; | |
0d66bf6d PZ |
4192 | |
4193 | /* | |
c6eb3dda PZ |
4194 | * Ensure we emit the owner->on_cpu, dereference _after_ checking |
4195 | * lock->owner still matches owner, if that fails, owner might | |
4196 | * point to free()d memory, if it still matches, the rcu_read_lock() | |
4197 | * ensures the memory stays valid. | |
0d66bf6d | 4198 | */ |
c6eb3dda | 4199 | barrier(); |
0d66bf6d | 4200 | |
c6eb3dda PZ |
4201 | ret = owner->on_cpu; |
4202 | fail: | |
4203 | rcu_read_unlock(); | |
0d66bf6d | 4204 | |
c6eb3dda PZ |
4205 | return ret; |
4206 | } | |
0d66bf6d | 4207 | |
c6eb3dda PZ |
4208 | /* |
4209 | * Look out! "owner" is an entirely speculative pointer | |
4210 | * access and not reliable. | |
4211 | */ | |
4212 | int mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner) | |
4213 | { | |
4214 | if (!sched_feat(OWNER_SPIN)) | |
4215 | return 0; | |
0d66bf6d | 4216 | |
c6eb3dda PZ |
4217 | while (owner_running(lock, owner)) { |
4218 | if (need_resched()) | |
0d66bf6d PZ |
4219 | return 0; |
4220 | ||
335d7afb | 4221 | arch_mutex_cpu_relax(); |
0d66bf6d | 4222 | } |
4b402210 | 4223 | |
c6eb3dda PZ |
4224 | /* |
4225 | * If the owner changed to another task there is likely | |
4226 | * heavy contention, stop spinning. | |
4227 | */ | |
4228 | if (lock->owner) | |
4229 | return 0; | |
4230 | ||
0d66bf6d PZ |
4231 | return 1; |
4232 | } | |
4233 | #endif | |
4234 | ||
1da177e4 LT |
4235 | #ifdef CONFIG_PREEMPT |
4236 | /* | |
2ed6e34f | 4237 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4238 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4239 | * occur there and call schedule directly. |
4240 | */ | |
d1f74e20 | 4241 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
4242 | { |
4243 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4244 | |
1da177e4 LT |
4245 | /* |
4246 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4247 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4248 | */ |
beed33a8 | 4249 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4250 | return; |
4251 | ||
3a5c359a | 4252 | do { |
d1f74e20 | 4253 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
3a5c359a | 4254 | schedule(); |
d1f74e20 | 4255 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 4256 | |
3a5c359a AK |
4257 | /* |
4258 | * Check again in case we missed a preemption opportunity | |
4259 | * between schedule and now. | |
4260 | */ | |
4261 | barrier(); | |
5ed0cec0 | 4262 | } while (need_resched()); |
1da177e4 | 4263 | } |
1da177e4 LT |
4264 | EXPORT_SYMBOL(preempt_schedule); |
4265 | ||
4266 | /* | |
2ed6e34f | 4267 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4268 | * off of irq context. |
4269 | * Note, that this is called and return with irqs disabled. This will | |
4270 | * protect us against recursive calling from irq. | |
4271 | */ | |
4272 | asmlinkage void __sched preempt_schedule_irq(void) | |
4273 | { | |
4274 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4275 | |
2ed6e34f | 4276 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4277 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4278 | ||
3a5c359a AK |
4279 | do { |
4280 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4281 | local_irq_enable(); |
4282 | schedule(); | |
4283 | local_irq_disable(); | |
3a5c359a | 4284 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4285 | |
3a5c359a AK |
4286 | /* |
4287 | * Check again in case we missed a preemption opportunity | |
4288 | * between schedule and now. | |
4289 | */ | |
4290 | barrier(); | |
5ed0cec0 | 4291 | } while (need_resched()); |
1da177e4 LT |
4292 | } |
4293 | ||
4294 | #endif /* CONFIG_PREEMPT */ | |
4295 | ||
63859d4f | 4296 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 4297 | void *key) |
1da177e4 | 4298 | { |
63859d4f | 4299 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 4300 | } |
1da177e4 LT |
4301 | EXPORT_SYMBOL(default_wake_function); |
4302 | ||
4303 | /* | |
41a2d6cf IM |
4304 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4305 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4306 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4307 | * | |
4308 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4309 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4310 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4311 | */ | |
78ddb08f | 4312 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 4313 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 4314 | { |
2e45874c | 4315 | wait_queue_t *curr, *next; |
1da177e4 | 4316 | |
2e45874c | 4317 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4318 | unsigned flags = curr->flags; |
4319 | ||
63859d4f | 4320 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 4321 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4322 | break; |
4323 | } | |
4324 | } | |
4325 | ||
4326 | /** | |
4327 | * __wake_up - wake up threads blocked on a waitqueue. | |
4328 | * @q: the waitqueue | |
4329 | * @mode: which threads | |
4330 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4331 | * @key: is directly passed to the wakeup function |
50fa610a DH |
4332 | * |
4333 | * It may be assumed that this function implies a write memory barrier before | |
4334 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4335 | */ |
7ad5b3a5 | 4336 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4337 | int nr_exclusive, void *key) |
1da177e4 LT |
4338 | { |
4339 | unsigned long flags; | |
4340 | ||
4341 | spin_lock_irqsave(&q->lock, flags); | |
4342 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4343 | spin_unlock_irqrestore(&q->lock, flags); | |
4344 | } | |
1da177e4 LT |
4345 | EXPORT_SYMBOL(__wake_up); |
4346 | ||
4347 | /* | |
4348 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4349 | */ | |
7ad5b3a5 | 4350 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4351 | { |
4352 | __wake_up_common(q, mode, 1, 0, NULL); | |
4353 | } | |
22c43c81 | 4354 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 4355 | |
4ede816a DL |
4356 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
4357 | { | |
4358 | __wake_up_common(q, mode, 1, 0, key); | |
4359 | } | |
bf294b41 | 4360 | EXPORT_SYMBOL_GPL(__wake_up_locked_key); |
4ede816a | 4361 | |
1da177e4 | 4362 | /** |
4ede816a | 4363 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4364 | * @q: the waitqueue |
4365 | * @mode: which threads | |
4366 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 4367 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
4368 | * |
4369 | * The sync wakeup differs that the waker knows that it will schedule | |
4370 | * away soon, so while the target thread will be woken up, it will not | |
4371 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4372 | * with each other. This can prevent needless bouncing between CPUs. | |
4373 | * | |
4374 | * On UP it can prevent extra preemption. | |
50fa610a DH |
4375 | * |
4376 | * It may be assumed that this function implies a write memory barrier before | |
4377 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4378 | */ |
4ede816a DL |
4379 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
4380 | int nr_exclusive, void *key) | |
1da177e4 LT |
4381 | { |
4382 | unsigned long flags; | |
7d478721 | 4383 | int wake_flags = WF_SYNC; |
1da177e4 LT |
4384 | |
4385 | if (unlikely(!q)) | |
4386 | return; | |
4387 | ||
4388 | if (unlikely(!nr_exclusive)) | |
7d478721 | 4389 | wake_flags = 0; |
1da177e4 LT |
4390 | |
4391 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 4392 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
4393 | spin_unlock_irqrestore(&q->lock, flags); |
4394 | } | |
4ede816a DL |
4395 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
4396 | ||
4397 | /* | |
4398 | * __wake_up_sync - see __wake_up_sync_key() | |
4399 | */ | |
4400 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
4401 | { | |
4402 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
4403 | } | |
1da177e4 LT |
4404 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
4405 | ||
65eb3dc6 KD |
4406 | /** |
4407 | * complete: - signals a single thread waiting on this completion | |
4408 | * @x: holds the state of this particular completion | |
4409 | * | |
4410 | * This will wake up a single thread waiting on this completion. Threads will be | |
4411 | * awakened in the same order in which they were queued. | |
4412 | * | |
4413 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
4414 | * |
4415 | * It may be assumed that this function implies a write memory barrier before | |
4416 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4417 | */ |
b15136e9 | 4418 | void complete(struct completion *x) |
1da177e4 LT |
4419 | { |
4420 | unsigned long flags; | |
4421 | ||
4422 | spin_lock_irqsave(&x->wait.lock, flags); | |
4423 | x->done++; | |
d9514f6c | 4424 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4425 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4426 | } | |
4427 | EXPORT_SYMBOL(complete); | |
4428 | ||
65eb3dc6 KD |
4429 | /** |
4430 | * complete_all: - signals all threads waiting on this completion | |
4431 | * @x: holds the state of this particular completion | |
4432 | * | |
4433 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
4434 | * |
4435 | * It may be assumed that this function implies a write memory barrier before | |
4436 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4437 | */ |
b15136e9 | 4438 | void complete_all(struct completion *x) |
1da177e4 LT |
4439 | { |
4440 | unsigned long flags; | |
4441 | ||
4442 | spin_lock_irqsave(&x->wait.lock, flags); | |
4443 | x->done += UINT_MAX/2; | |
d9514f6c | 4444 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4445 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4446 | } | |
4447 | EXPORT_SYMBOL(complete_all); | |
4448 | ||
8cbbe86d AK |
4449 | static inline long __sched |
4450 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4451 | { |
1da177e4 LT |
4452 | if (!x->done) { |
4453 | DECLARE_WAITQUEUE(wait, current); | |
4454 | ||
a93d2f17 | 4455 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 4456 | do { |
94d3d824 | 4457 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
4458 | timeout = -ERESTARTSYS; |
4459 | break; | |
8cbbe86d AK |
4460 | } |
4461 | __set_current_state(state); | |
1da177e4 LT |
4462 | spin_unlock_irq(&x->wait.lock); |
4463 | timeout = schedule_timeout(timeout); | |
4464 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4465 | } while (!x->done && timeout); |
1da177e4 | 4466 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4467 | if (!x->done) |
4468 | return timeout; | |
1da177e4 LT |
4469 | } |
4470 | x->done--; | |
ea71a546 | 4471 | return timeout ?: 1; |
1da177e4 | 4472 | } |
1da177e4 | 4473 | |
8cbbe86d AK |
4474 | static long __sched |
4475 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4476 | { |
1da177e4 LT |
4477 | might_sleep(); |
4478 | ||
4479 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4480 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4481 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4482 | return timeout; |
4483 | } | |
1da177e4 | 4484 | |
65eb3dc6 KD |
4485 | /** |
4486 | * wait_for_completion: - waits for completion of a task | |
4487 | * @x: holds the state of this particular completion | |
4488 | * | |
4489 | * This waits to be signaled for completion of a specific task. It is NOT | |
4490 | * interruptible and there is no timeout. | |
4491 | * | |
4492 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4493 | * and interrupt capability. Also see complete(). | |
4494 | */ | |
b15136e9 | 4495 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4496 | { |
4497 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4498 | } |
8cbbe86d | 4499 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4500 | |
65eb3dc6 KD |
4501 | /** |
4502 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4503 | * @x: holds the state of this particular completion | |
4504 | * @timeout: timeout value in jiffies | |
4505 | * | |
4506 | * This waits for either a completion of a specific task to be signaled or for a | |
4507 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4508 | * interruptible. | |
4509 | */ | |
b15136e9 | 4510 | unsigned long __sched |
8cbbe86d | 4511 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4512 | { |
8cbbe86d | 4513 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4514 | } |
8cbbe86d | 4515 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4516 | |
65eb3dc6 KD |
4517 | /** |
4518 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4519 | * @x: holds the state of this particular completion | |
4520 | * | |
4521 | * This waits for completion of a specific task to be signaled. It is | |
4522 | * interruptible. | |
4523 | */ | |
8cbbe86d | 4524 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4525 | { |
51e97990 AK |
4526 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4527 | if (t == -ERESTARTSYS) | |
4528 | return t; | |
4529 | return 0; | |
0fec171c | 4530 | } |
8cbbe86d | 4531 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4532 | |
65eb3dc6 KD |
4533 | /** |
4534 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4535 | * @x: holds the state of this particular completion | |
4536 | * @timeout: timeout value in jiffies | |
4537 | * | |
4538 | * This waits for either a completion of a specific task to be signaled or for a | |
4539 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4540 | */ | |
6bf41237 | 4541 | long __sched |
8cbbe86d AK |
4542 | wait_for_completion_interruptible_timeout(struct completion *x, |
4543 | unsigned long timeout) | |
0fec171c | 4544 | { |
8cbbe86d | 4545 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4546 | } |
8cbbe86d | 4547 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4548 | |
65eb3dc6 KD |
4549 | /** |
4550 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4551 | * @x: holds the state of this particular completion | |
4552 | * | |
4553 | * This waits to be signaled for completion of a specific task. It can be | |
4554 | * interrupted by a kill signal. | |
4555 | */ | |
009e577e MW |
4556 | int __sched wait_for_completion_killable(struct completion *x) |
4557 | { | |
4558 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4559 | if (t == -ERESTARTSYS) | |
4560 | return t; | |
4561 | return 0; | |
4562 | } | |
4563 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4564 | ||
0aa12fb4 SW |
4565 | /** |
4566 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
4567 | * @x: holds the state of this particular completion | |
4568 | * @timeout: timeout value in jiffies | |
4569 | * | |
4570 | * This waits for either a completion of a specific task to be | |
4571 | * signaled or for a specified timeout to expire. It can be | |
4572 | * interrupted by a kill signal. The timeout is in jiffies. | |
4573 | */ | |
6bf41237 | 4574 | long __sched |
0aa12fb4 SW |
4575 | wait_for_completion_killable_timeout(struct completion *x, |
4576 | unsigned long timeout) | |
4577 | { | |
4578 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
4579 | } | |
4580 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
4581 | ||
be4de352 DC |
4582 | /** |
4583 | * try_wait_for_completion - try to decrement a completion without blocking | |
4584 | * @x: completion structure | |
4585 | * | |
4586 | * Returns: 0 if a decrement cannot be done without blocking | |
4587 | * 1 if a decrement succeeded. | |
4588 | * | |
4589 | * If a completion is being used as a counting completion, | |
4590 | * attempt to decrement the counter without blocking. This | |
4591 | * enables us to avoid waiting if the resource the completion | |
4592 | * is protecting is not available. | |
4593 | */ | |
4594 | bool try_wait_for_completion(struct completion *x) | |
4595 | { | |
7539a3b3 | 4596 | unsigned long flags; |
be4de352 DC |
4597 | int ret = 1; |
4598 | ||
7539a3b3 | 4599 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4600 | if (!x->done) |
4601 | ret = 0; | |
4602 | else | |
4603 | x->done--; | |
7539a3b3 | 4604 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4605 | return ret; |
4606 | } | |
4607 | EXPORT_SYMBOL(try_wait_for_completion); | |
4608 | ||
4609 | /** | |
4610 | * completion_done - Test to see if a completion has any waiters | |
4611 | * @x: completion structure | |
4612 | * | |
4613 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4614 | * 1 if there are no waiters. | |
4615 | * | |
4616 | */ | |
4617 | bool completion_done(struct completion *x) | |
4618 | { | |
7539a3b3 | 4619 | unsigned long flags; |
be4de352 DC |
4620 | int ret = 1; |
4621 | ||
7539a3b3 | 4622 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4623 | if (!x->done) |
4624 | ret = 0; | |
7539a3b3 | 4625 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4626 | return ret; |
4627 | } | |
4628 | EXPORT_SYMBOL(completion_done); | |
4629 | ||
8cbbe86d AK |
4630 | static long __sched |
4631 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4632 | { |
0fec171c IM |
4633 | unsigned long flags; |
4634 | wait_queue_t wait; | |
4635 | ||
4636 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4637 | |
8cbbe86d | 4638 | __set_current_state(state); |
1da177e4 | 4639 | |
8cbbe86d AK |
4640 | spin_lock_irqsave(&q->lock, flags); |
4641 | __add_wait_queue(q, &wait); | |
4642 | spin_unlock(&q->lock); | |
4643 | timeout = schedule_timeout(timeout); | |
4644 | spin_lock_irq(&q->lock); | |
4645 | __remove_wait_queue(q, &wait); | |
4646 | spin_unlock_irqrestore(&q->lock, flags); | |
4647 | ||
4648 | return timeout; | |
4649 | } | |
4650 | ||
4651 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4652 | { | |
4653 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4654 | } |
1da177e4 LT |
4655 | EXPORT_SYMBOL(interruptible_sleep_on); |
4656 | ||
0fec171c | 4657 | long __sched |
95cdf3b7 | 4658 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4659 | { |
8cbbe86d | 4660 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4661 | } |
1da177e4 LT |
4662 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4663 | ||
0fec171c | 4664 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4665 | { |
8cbbe86d | 4666 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4667 | } |
1da177e4 LT |
4668 | EXPORT_SYMBOL(sleep_on); |
4669 | ||
0fec171c | 4670 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4671 | { |
8cbbe86d | 4672 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4673 | } |
1da177e4 LT |
4674 | EXPORT_SYMBOL(sleep_on_timeout); |
4675 | ||
b29739f9 IM |
4676 | #ifdef CONFIG_RT_MUTEXES |
4677 | ||
4678 | /* | |
4679 | * rt_mutex_setprio - set the current priority of a task | |
4680 | * @p: task | |
4681 | * @prio: prio value (kernel-internal form) | |
4682 | * | |
4683 | * This function changes the 'effective' priority of a task. It does | |
4684 | * not touch ->normal_prio like __setscheduler(). | |
4685 | * | |
4686 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4687 | */ | |
36c8b586 | 4688 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4689 | { |
4690 | unsigned long flags; | |
83b699ed | 4691 | int oldprio, on_rq, running; |
70b97a7f | 4692 | struct rq *rq; |
83ab0aa0 | 4693 | const struct sched_class *prev_class; |
b29739f9 IM |
4694 | |
4695 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4696 | ||
4697 | rq = task_rq_lock(p, &flags); | |
4698 | ||
a8027073 | 4699 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 4700 | oldprio = p->prio; |
83ab0aa0 | 4701 | prev_class = p->sched_class; |
fd2f4419 | 4702 | on_rq = p->on_rq; |
051a1d1a | 4703 | running = task_current(rq, p); |
0e1f3483 | 4704 | if (on_rq) |
69be72c1 | 4705 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4706 | if (running) |
4707 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4708 | |
4709 | if (rt_prio(prio)) | |
4710 | p->sched_class = &rt_sched_class; | |
4711 | else | |
4712 | p->sched_class = &fair_sched_class; | |
4713 | ||
b29739f9 IM |
4714 | p->prio = prio; |
4715 | ||
0e1f3483 HS |
4716 | if (running) |
4717 | p->sched_class->set_curr_task(rq); | |
da7a735e | 4718 | if (on_rq) |
371fd7e7 | 4719 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 | 4720 | |
da7a735e | 4721 | check_class_changed(rq, p, prev_class, oldprio); |
b29739f9 IM |
4722 | task_rq_unlock(rq, &flags); |
4723 | } | |
4724 | ||
4725 | #endif | |
4726 | ||
36c8b586 | 4727 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4728 | { |
dd41f596 | 4729 | int old_prio, delta, on_rq; |
1da177e4 | 4730 | unsigned long flags; |
70b97a7f | 4731 | struct rq *rq; |
1da177e4 LT |
4732 | |
4733 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4734 | return; | |
4735 | /* | |
4736 | * We have to be careful, if called from sys_setpriority(), | |
4737 | * the task might be in the middle of scheduling on another CPU. | |
4738 | */ | |
4739 | rq = task_rq_lock(p, &flags); | |
4740 | /* | |
4741 | * The RT priorities are set via sched_setscheduler(), but we still | |
4742 | * allow the 'normal' nice value to be set - but as expected | |
4743 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4744 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4745 | */ |
e05606d3 | 4746 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4747 | p->static_prio = NICE_TO_PRIO(nice); |
4748 | goto out_unlock; | |
4749 | } | |
fd2f4419 | 4750 | on_rq = p->on_rq; |
c09595f6 | 4751 | if (on_rq) |
69be72c1 | 4752 | dequeue_task(rq, p, 0); |
1da177e4 | 4753 | |
1da177e4 | 4754 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4755 | set_load_weight(p); |
b29739f9 IM |
4756 | old_prio = p->prio; |
4757 | p->prio = effective_prio(p); | |
4758 | delta = p->prio - old_prio; | |
1da177e4 | 4759 | |
dd41f596 | 4760 | if (on_rq) { |
371fd7e7 | 4761 | enqueue_task(rq, p, 0); |
1da177e4 | 4762 | /* |
d5f9f942 AM |
4763 | * If the task increased its priority or is running and |
4764 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4765 | */ |
d5f9f942 | 4766 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4767 | resched_task(rq->curr); |
4768 | } | |
4769 | out_unlock: | |
4770 | task_rq_unlock(rq, &flags); | |
4771 | } | |
1da177e4 LT |
4772 | EXPORT_SYMBOL(set_user_nice); |
4773 | ||
e43379f1 MM |
4774 | /* |
4775 | * can_nice - check if a task can reduce its nice value | |
4776 | * @p: task | |
4777 | * @nice: nice value | |
4778 | */ | |
36c8b586 | 4779 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4780 | { |
024f4747 MM |
4781 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4782 | int nice_rlim = 20 - nice; | |
48f24c4d | 4783 | |
78d7d407 | 4784 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
4785 | capable(CAP_SYS_NICE)); |
4786 | } | |
4787 | ||
1da177e4 LT |
4788 | #ifdef __ARCH_WANT_SYS_NICE |
4789 | ||
4790 | /* | |
4791 | * sys_nice - change the priority of the current process. | |
4792 | * @increment: priority increment | |
4793 | * | |
4794 | * sys_setpriority is a more generic, but much slower function that | |
4795 | * does similar things. | |
4796 | */ | |
5add95d4 | 4797 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 4798 | { |
48f24c4d | 4799 | long nice, retval; |
1da177e4 LT |
4800 | |
4801 | /* | |
4802 | * Setpriority might change our priority at the same moment. | |
4803 | * We don't have to worry. Conceptually one call occurs first | |
4804 | * and we have a single winner. | |
4805 | */ | |
e43379f1 MM |
4806 | if (increment < -40) |
4807 | increment = -40; | |
1da177e4 LT |
4808 | if (increment > 40) |
4809 | increment = 40; | |
4810 | ||
2b8f836f | 4811 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
4812 | if (nice < -20) |
4813 | nice = -20; | |
4814 | if (nice > 19) | |
4815 | nice = 19; | |
4816 | ||
e43379f1 MM |
4817 | if (increment < 0 && !can_nice(current, nice)) |
4818 | return -EPERM; | |
4819 | ||
1da177e4 LT |
4820 | retval = security_task_setnice(current, nice); |
4821 | if (retval) | |
4822 | return retval; | |
4823 | ||
4824 | set_user_nice(current, nice); | |
4825 | return 0; | |
4826 | } | |
4827 | ||
4828 | #endif | |
4829 | ||
4830 | /** | |
4831 | * task_prio - return the priority value of a given task. | |
4832 | * @p: the task in question. | |
4833 | * | |
4834 | * This is the priority value as seen by users in /proc. | |
4835 | * RT tasks are offset by -200. Normal tasks are centered | |
4836 | * around 0, value goes from -16 to +15. | |
4837 | */ | |
36c8b586 | 4838 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4839 | { |
4840 | return p->prio - MAX_RT_PRIO; | |
4841 | } | |
4842 | ||
4843 | /** | |
4844 | * task_nice - return the nice value of a given task. | |
4845 | * @p: the task in question. | |
4846 | */ | |
36c8b586 | 4847 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4848 | { |
4849 | return TASK_NICE(p); | |
4850 | } | |
150d8bed | 4851 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4852 | |
4853 | /** | |
4854 | * idle_cpu - is a given cpu idle currently? | |
4855 | * @cpu: the processor in question. | |
4856 | */ | |
4857 | int idle_cpu(int cpu) | |
4858 | { | |
4859 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4860 | } | |
4861 | ||
1da177e4 LT |
4862 | /** |
4863 | * idle_task - return the idle task for a given cpu. | |
4864 | * @cpu: the processor in question. | |
4865 | */ | |
36c8b586 | 4866 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4867 | { |
4868 | return cpu_rq(cpu)->idle; | |
4869 | } | |
4870 | ||
4871 | /** | |
4872 | * find_process_by_pid - find a process with a matching PID value. | |
4873 | * @pid: the pid in question. | |
4874 | */ | |
a9957449 | 4875 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4876 | { |
228ebcbe | 4877 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4878 | } |
4879 | ||
4880 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4881 | static void |
4882 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4883 | { |
1da177e4 LT |
4884 | p->policy = policy; |
4885 | p->rt_priority = prio; | |
b29739f9 IM |
4886 | p->normal_prio = normal_prio(p); |
4887 | /* we are holding p->pi_lock already */ | |
4888 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4889 | if (rt_prio(p->prio)) |
4890 | p->sched_class = &rt_sched_class; | |
4891 | else | |
4892 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 4893 | set_load_weight(p); |
1da177e4 LT |
4894 | } |
4895 | ||
c69e8d9c DH |
4896 | /* |
4897 | * check the target process has a UID that matches the current process's | |
4898 | */ | |
4899 | static bool check_same_owner(struct task_struct *p) | |
4900 | { | |
4901 | const struct cred *cred = current_cred(), *pcred; | |
4902 | bool match; | |
4903 | ||
4904 | rcu_read_lock(); | |
4905 | pcred = __task_cred(p); | |
b0e77598 SH |
4906 | if (cred->user->user_ns == pcred->user->user_ns) |
4907 | match = (cred->euid == pcred->euid || | |
4908 | cred->euid == pcred->uid); | |
4909 | else | |
4910 | match = false; | |
c69e8d9c DH |
4911 | rcu_read_unlock(); |
4912 | return match; | |
4913 | } | |
4914 | ||
961ccddd | 4915 | static int __sched_setscheduler(struct task_struct *p, int policy, |
fe7de49f | 4916 | const struct sched_param *param, bool user) |
1da177e4 | 4917 | { |
83b699ed | 4918 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4919 | unsigned long flags; |
83ab0aa0 | 4920 | const struct sched_class *prev_class; |
70b97a7f | 4921 | struct rq *rq; |
ca94c442 | 4922 | int reset_on_fork; |
1da177e4 | 4923 | |
66e5393a SR |
4924 | /* may grab non-irq protected spin_locks */ |
4925 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4926 | recheck: |
4927 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4928 | if (policy < 0) { |
4929 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4930 | policy = oldpolicy = p->policy; |
ca94c442 LP |
4931 | } else { |
4932 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
4933 | policy &= ~SCHED_RESET_ON_FORK; | |
4934 | ||
4935 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
4936 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
4937 | policy != SCHED_IDLE) | |
4938 | return -EINVAL; | |
4939 | } | |
4940 | ||
1da177e4 LT |
4941 | /* |
4942 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4943 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4944 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4945 | */ |
4946 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4947 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4948 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4949 | return -EINVAL; |
e05606d3 | 4950 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4951 | return -EINVAL; |
4952 | ||
37e4ab3f OC |
4953 | /* |
4954 | * Allow unprivileged RT tasks to decrease priority: | |
4955 | */ | |
961ccddd | 4956 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4957 | if (rt_policy(policy)) { |
a44702e8 ON |
4958 | unsigned long rlim_rtprio = |
4959 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
4960 | |
4961 | /* can't set/change the rt policy */ | |
4962 | if (policy != p->policy && !rlim_rtprio) | |
4963 | return -EPERM; | |
4964 | ||
4965 | /* can't increase priority */ | |
4966 | if (param->sched_priority > p->rt_priority && | |
4967 | param->sched_priority > rlim_rtprio) | |
4968 | return -EPERM; | |
4969 | } | |
c02aa73b | 4970 | |
dd41f596 | 4971 | /* |
c02aa73b DH |
4972 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
4973 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 4974 | */ |
c02aa73b DH |
4975 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
4976 | if (!can_nice(p, TASK_NICE(p))) | |
4977 | return -EPERM; | |
4978 | } | |
5fe1d75f | 4979 | |
37e4ab3f | 4980 | /* can't change other user's priorities */ |
c69e8d9c | 4981 | if (!check_same_owner(p)) |
37e4ab3f | 4982 | return -EPERM; |
ca94c442 LP |
4983 | |
4984 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4985 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4986 | return -EPERM; | |
37e4ab3f | 4987 | } |
1da177e4 | 4988 | |
725aad24 | 4989 | if (user) { |
b0ae1981 | 4990 | retval = security_task_setscheduler(p); |
725aad24 JF |
4991 | if (retval) |
4992 | return retval; | |
4993 | } | |
4994 | ||
b29739f9 IM |
4995 | /* |
4996 | * make sure no PI-waiters arrive (or leave) while we are | |
4997 | * changing the priority of the task: | |
4998 | */ | |
1d615482 | 4999 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
1da177e4 | 5000 | /* |
25985edc | 5001 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
5002 | * runqueue lock must be held. |
5003 | */ | |
b29739f9 | 5004 | rq = __task_rq_lock(p); |
dc61b1d6 | 5005 | |
34f971f6 PZ |
5006 | /* |
5007 | * Changing the policy of the stop threads its a very bad idea | |
5008 | */ | |
5009 | if (p == rq->stop) { | |
5010 | __task_rq_unlock(rq); | |
5011 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
5012 | return -EINVAL; | |
5013 | } | |
5014 | ||
a51e9198 DF |
5015 | /* |
5016 | * If not changing anything there's no need to proceed further: | |
5017 | */ | |
5018 | if (unlikely(policy == p->policy && (!rt_policy(policy) || | |
5019 | param->sched_priority == p->rt_priority))) { | |
5020 | ||
5021 | __task_rq_unlock(rq); | |
5022 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
5023 | return 0; | |
5024 | } | |
5025 | ||
dc61b1d6 PZ |
5026 | #ifdef CONFIG_RT_GROUP_SCHED |
5027 | if (user) { | |
5028 | /* | |
5029 | * Do not allow realtime tasks into groups that have no runtime | |
5030 | * assigned. | |
5031 | */ | |
5032 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
5033 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
5034 | !task_group_is_autogroup(task_group(p))) { | |
dc61b1d6 PZ |
5035 | __task_rq_unlock(rq); |
5036 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
5037 | return -EPERM; | |
5038 | } | |
5039 | } | |
5040 | #endif | |
5041 | ||
1da177e4 LT |
5042 | /* recheck policy now with rq lock held */ |
5043 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5044 | policy = oldpolicy = -1; | |
b29739f9 | 5045 | __task_rq_unlock(rq); |
1d615482 | 5046 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
5047 | goto recheck; |
5048 | } | |
fd2f4419 | 5049 | on_rq = p->on_rq; |
051a1d1a | 5050 | running = task_current(rq, p); |
0e1f3483 | 5051 | if (on_rq) |
2e1cb74a | 5052 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5053 | if (running) |
5054 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5055 | |
ca94c442 LP |
5056 | p->sched_reset_on_fork = reset_on_fork; |
5057 | ||
1da177e4 | 5058 | oldprio = p->prio; |
83ab0aa0 | 5059 | prev_class = p->sched_class; |
dd41f596 | 5060 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5061 | |
0e1f3483 HS |
5062 | if (running) |
5063 | p->sched_class->set_curr_task(rq); | |
da7a735e | 5064 | if (on_rq) |
dd41f596 | 5065 | activate_task(rq, p, 0); |
cb469845 | 5066 | |
da7a735e | 5067 | check_class_changed(rq, p, prev_class, oldprio); |
b29739f9 | 5068 | __task_rq_unlock(rq); |
1d615482 | 5069 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
b29739f9 | 5070 | |
95e02ca9 TG |
5071 | rt_mutex_adjust_pi(p); |
5072 | ||
1da177e4 LT |
5073 | return 0; |
5074 | } | |
961ccddd RR |
5075 | |
5076 | /** | |
5077 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5078 | * @p: the task in question. | |
5079 | * @policy: new policy. | |
5080 | * @param: structure containing the new RT priority. | |
5081 | * | |
5082 | * NOTE that the task may be already dead. | |
5083 | */ | |
5084 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 5085 | const struct sched_param *param) |
961ccddd RR |
5086 | { |
5087 | return __sched_setscheduler(p, policy, param, true); | |
5088 | } | |
1da177e4 LT |
5089 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5090 | ||
961ccddd RR |
5091 | /** |
5092 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5093 | * @p: the task in question. | |
5094 | * @policy: new policy. | |
5095 | * @param: structure containing the new RT priority. | |
5096 | * | |
5097 | * Just like sched_setscheduler, only don't bother checking if the | |
5098 | * current context has permission. For example, this is needed in | |
5099 | * stop_machine(): we create temporary high priority worker threads, | |
5100 | * but our caller might not have that capability. | |
5101 | */ | |
5102 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 5103 | const struct sched_param *param) |
961ccddd RR |
5104 | { |
5105 | return __sched_setscheduler(p, policy, param, false); | |
5106 | } | |
5107 | ||
95cdf3b7 IM |
5108 | static int |
5109 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5110 | { |
1da177e4 LT |
5111 | struct sched_param lparam; |
5112 | struct task_struct *p; | |
36c8b586 | 5113 | int retval; |
1da177e4 LT |
5114 | |
5115 | if (!param || pid < 0) | |
5116 | return -EINVAL; | |
5117 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5118 | return -EFAULT; | |
5fe1d75f ON |
5119 | |
5120 | rcu_read_lock(); | |
5121 | retval = -ESRCH; | |
1da177e4 | 5122 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5123 | if (p != NULL) |
5124 | retval = sched_setscheduler(p, policy, &lparam); | |
5125 | rcu_read_unlock(); | |
36c8b586 | 5126 | |
1da177e4 LT |
5127 | return retval; |
5128 | } | |
5129 | ||
5130 | /** | |
5131 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5132 | * @pid: the pid in question. | |
5133 | * @policy: new policy. | |
5134 | * @param: structure containing the new RT priority. | |
5135 | */ | |
5add95d4 HC |
5136 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
5137 | struct sched_param __user *, param) | |
1da177e4 | 5138 | { |
c21761f1 JB |
5139 | /* negative values for policy are not valid */ |
5140 | if (policy < 0) | |
5141 | return -EINVAL; | |
5142 | ||
1da177e4 LT |
5143 | return do_sched_setscheduler(pid, policy, param); |
5144 | } | |
5145 | ||
5146 | /** | |
5147 | * sys_sched_setparam - set/change the RT priority of a thread | |
5148 | * @pid: the pid in question. | |
5149 | * @param: structure containing the new RT priority. | |
5150 | */ | |
5add95d4 | 5151 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5152 | { |
5153 | return do_sched_setscheduler(pid, -1, param); | |
5154 | } | |
5155 | ||
5156 | /** | |
5157 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5158 | * @pid: the pid in question. | |
5159 | */ | |
5add95d4 | 5160 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 5161 | { |
36c8b586 | 5162 | struct task_struct *p; |
3a5c359a | 5163 | int retval; |
1da177e4 LT |
5164 | |
5165 | if (pid < 0) | |
3a5c359a | 5166 | return -EINVAL; |
1da177e4 LT |
5167 | |
5168 | retval = -ESRCH; | |
5fe85be0 | 5169 | rcu_read_lock(); |
1da177e4 LT |
5170 | p = find_process_by_pid(pid); |
5171 | if (p) { | |
5172 | retval = security_task_getscheduler(p); | |
5173 | if (!retval) | |
ca94c442 LP |
5174 | retval = p->policy |
5175 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 5176 | } |
5fe85be0 | 5177 | rcu_read_unlock(); |
1da177e4 LT |
5178 | return retval; |
5179 | } | |
5180 | ||
5181 | /** | |
ca94c442 | 5182 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
5183 | * @pid: the pid in question. |
5184 | * @param: structure containing the RT priority. | |
5185 | */ | |
5add95d4 | 5186 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5187 | { |
5188 | struct sched_param lp; | |
36c8b586 | 5189 | struct task_struct *p; |
3a5c359a | 5190 | int retval; |
1da177e4 LT |
5191 | |
5192 | if (!param || pid < 0) | |
3a5c359a | 5193 | return -EINVAL; |
1da177e4 | 5194 | |
5fe85be0 | 5195 | rcu_read_lock(); |
1da177e4 LT |
5196 | p = find_process_by_pid(pid); |
5197 | retval = -ESRCH; | |
5198 | if (!p) | |
5199 | goto out_unlock; | |
5200 | ||
5201 | retval = security_task_getscheduler(p); | |
5202 | if (retval) | |
5203 | goto out_unlock; | |
5204 | ||
5205 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 5206 | rcu_read_unlock(); |
1da177e4 LT |
5207 | |
5208 | /* | |
5209 | * This one might sleep, we cannot do it with a spinlock held ... | |
5210 | */ | |
5211 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5212 | ||
1da177e4 LT |
5213 | return retval; |
5214 | ||
5215 | out_unlock: | |
5fe85be0 | 5216 | rcu_read_unlock(); |
1da177e4 LT |
5217 | return retval; |
5218 | } | |
5219 | ||
96f874e2 | 5220 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 5221 | { |
5a16f3d3 | 5222 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
5223 | struct task_struct *p; |
5224 | int retval; | |
1da177e4 | 5225 | |
95402b38 | 5226 | get_online_cpus(); |
23f5d142 | 5227 | rcu_read_lock(); |
1da177e4 LT |
5228 | |
5229 | p = find_process_by_pid(pid); | |
5230 | if (!p) { | |
23f5d142 | 5231 | rcu_read_unlock(); |
95402b38 | 5232 | put_online_cpus(); |
1da177e4 LT |
5233 | return -ESRCH; |
5234 | } | |
5235 | ||
23f5d142 | 5236 | /* Prevent p going away */ |
1da177e4 | 5237 | get_task_struct(p); |
23f5d142 | 5238 | rcu_read_unlock(); |
1da177e4 | 5239 | |
5a16f3d3 RR |
5240 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
5241 | retval = -ENOMEM; | |
5242 | goto out_put_task; | |
5243 | } | |
5244 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
5245 | retval = -ENOMEM; | |
5246 | goto out_free_cpus_allowed; | |
5247 | } | |
1da177e4 | 5248 | retval = -EPERM; |
b0e77598 | 5249 | if (!check_same_owner(p) && !task_ns_capable(p, CAP_SYS_NICE)) |
1da177e4 LT |
5250 | goto out_unlock; |
5251 | ||
b0ae1981 | 5252 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
5253 | if (retval) |
5254 | goto out_unlock; | |
5255 | ||
5a16f3d3 RR |
5256 | cpuset_cpus_allowed(p, cpus_allowed); |
5257 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 5258 | again: |
5a16f3d3 | 5259 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 5260 | |
8707d8b8 | 5261 | if (!retval) { |
5a16f3d3 RR |
5262 | cpuset_cpus_allowed(p, cpus_allowed); |
5263 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
5264 | /* |
5265 | * We must have raced with a concurrent cpuset | |
5266 | * update. Just reset the cpus_allowed to the | |
5267 | * cpuset's cpus_allowed | |
5268 | */ | |
5a16f3d3 | 5269 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
5270 | goto again; |
5271 | } | |
5272 | } | |
1da177e4 | 5273 | out_unlock: |
5a16f3d3 RR |
5274 | free_cpumask_var(new_mask); |
5275 | out_free_cpus_allowed: | |
5276 | free_cpumask_var(cpus_allowed); | |
5277 | out_put_task: | |
1da177e4 | 5278 | put_task_struct(p); |
95402b38 | 5279 | put_online_cpus(); |
1da177e4 LT |
5280 | return retval; |
5281 | } | |
5282 | ||
5283 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 5284 | struct cpumask *new_mask) |
1da177e4 | 5285 | { |
96f874e2 RR |
5286 | if (len < cpumask_size()) |
5287 | cpumask_clear(new_mask); | |
5288 | else if (len > cpumask_size()) | |
5289 | len = cpumask_size(); | |
5290 | ||
1da177e4 LT |
5291 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
5292 | } | |
5293 | ||
5294 | /** | |
5295 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5296 | * @pid: pid of the process | |
5297 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5298 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5299 | */ | |
5add95d4 HC |
5300 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
5301 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 5302 | { |
5a16f3d3 | 5303 | cpumask_var_t new_mask; |
1da177e4 LT |
5304 | int retval; |
5305 | ||
5a16f3d3 RR |
5306 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
5307 | return -ENOMEM; | |
1da177e4 | 5308 | |
5a16f3d3 RR |
5309 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
5310 | if (retval == 0) | |
5311 | retval = sched_setaffinity(pid, new_mask); | |
5312 | free_cpumask_var(new_mask); | |
5313 | return retval; | |
1da177e4 LT |
5314 | } |
5315 | ||
96f874e2 | 5316 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 5317 | { |
36c8b586 | 5318 | struct task_struct *p; |
31605683 TG |
5319 | unsigned long flags; |
5320 | struct rq *rq; | |
1da177e4 | 5321 | int retval; |
1da177e4 | 5322 | |
95402b38 | 5323 | get_online_cpus(); |
23f5d142 | 5324 | rcu_read_lock(); |
1da177e4 LT |
5325 | |
5326 | retval = -ESRCH; | |
5327 | p = find_process_by_pid(pid); | |
5328 | if (!p) | |
5329 | goto out_unlock; | |
5330 | ||
e7834f8f DQ |
5331 | retval = security_task_getscheduler(p); |
5332 | if (retval) | |
5333 | goto out_unlock; | |
5334 | ||
31605683 | 5335 | rq = task_rq_lock(p, &flags); |
96f874e2 | 5336 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
31605683 | 5337 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
5338 | |
5339 | out_unlock: | |
23f5d142 | 5340 | rcu_read_unlock(); |
95402b38 | 5341 | put_online_cpus(); |
1da177e4 | 5342 | |
9531b62f | 5343 | return retval; |
1da177e4 LT |
5344 | } |
5345 | ||
5346 | /** | |
5347 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5348 | * @pid: pid of the process | |
5349 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5350 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5351 | */ | |
5add95d4 HC |
5352 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
5353 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
5354 | { |
5355 | int ret; | |
f17c8607 | 5356 | cpumask_var_t mask; |
1da177e4 | 5357 | |
84fba5ec | 5358 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
5359 | return -EINVAL; |
5360 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
5361 | return -EINVAL; |
5362 | ||
f17c8607 RR |
5363 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
5364 | return -ENOMEM; | |
1da177e4 | 5365 | |
f17c8607 RR |
5366 | ret = sched_getaffinity(pid, mask); |
5367 | if (ret == 0) { | |
8bc037fb | 5368 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
5369 | |
5370 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
5371 | ret = -EFAULT; |
5372 | else | |
cd3d8031 | 5373 | ret = retlen; |
f17c8607 RR |
5374 | } |
5375 | free_cpumask_var(mask); | |
1da177e4 | 5376 | |
f17c8607 | 5377 | return ret; |
1da177e4 LT |
5378 | } |
5379 | ||
5380 | /** | |
5381 | * sys_sched_yield - yield the current processor to other threads. | |
5382 | * | |
dd41f596 IM |
5383 | * This function yields the current CPU to other tasks. If there are no |
5384 | * other threads running on this CPU then this function will return. | |
1da177e4 | 5385 | */ |
5add95d4 | 5386 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 5387 | { |
70b97a7f | 5388 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5389 | |
2d72376b | 5390 | schedstat_inc(rq, yld_count); |
4530d7ab | 5391 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5392 | |
5393 | /* | |
5394 | * Since we are going to call schedule() anyway, there's | |
5395 | * no need to preempt or enable interrupts: | |
5396 | */ | |
5397 | __release(rq->lock); | |
8a25d5de | 5398 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 5399 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
5400 | preempt_enable_no_resched(); |
5401 | ||
5402 | schedule(); | |
5403 | ||
5404 | return 0; | |
5405 | } | |
5406 | ||
d86ee480 PZ |
5407 | static inline int should_resched(void) |
5408 | { | |
5409 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
5410 | } | |
5411 | ||
e7b38404 | 5412 | static void __cond_resched(void) |
1da177e4 | 5413 | { |
e7aaaa69 FW |
5414 | add_preempt_count(PREEMPT_ACTIVE); |
5415 | schedule(); | |
5416 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
5417 | } |
5418 | ||
02b67cc3 | 5419 | int __sched _cond_resched(void) |
1da177e4 | 5420 | { |
d86ee480 | 5421 | if (should_resched()) { |
1da177e4 LT |
5422 | __cond_resched(); |
5423 | return 1; | |
5424 | } | |
5425 | return 0; | |
5426 | } | |
02b67cc3 | 5427 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5428 | |
5429 | /* | |
613afbf8 | 5430 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
5431 | * call schedule, and on return reacquire the lock. |
5432 | * | |
41a2d6cf | 5433 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5434 | * operations here to prevent schedule() from being called twice (once via |
5435 | * spin_unlock(), once by hand). | |
5436 | */ | |
613afbf8 | 5437 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5438 | { |
d86ee480 | 5439 | int resched = should_resched(); |
6df3cecb JK |
5440 | int ret = 0; |
5441 | ||
f607c668 PZ |
5442 | lockdep_assert_held(lock); |
5443 | ||
95c354fe | 5444 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5445 | spin_unlock(lock); |
d86ee480 | 5446 | if (resched) |
95c354fe NP |
5447 | __cond_resched(); |
5448 | else | |
5449 | cpu_relax(); | |
6df3cecb | 5450 | ret = 1; |
1da177e4 | 5451 | spin_lock(lock); |
1da177e4 | 5452 | } |
6df3cecb | 5453 | return ret; |
1da177e4 | 5454 | } |
613afbf8 | 5455 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 5456 | |
613afbf8 | 5457 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
5458 | { |
5459 | BUG_ON(!in_softirq()); | |
5460 | ||
d86ee480 | 5461 | if (should_resched()) { |
98d82567 | 5462 | local_bh_enable(); |
1da177e4 LT |
5463 | __cond_resched(); |
5464 | local_bh_disable(); | |
5465 | return 1; | |
5466 | } | |
5467 | return 0; | |
5468 | } | |
613afbf8 | 5469 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 5470 | |
1da177e4 LT |
5471 | /** |
5472 | * yield - yield the current processor to other threads. | |
5473 | * | |
72fd4a35 | 5474 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5475 | * thread runnable and calls sys_sched_yield(). |
5476 | */ | |
5477 | void __sched yield(void) | |
5478 | { | |
5479 | set_current_state(TASK_RUNNING); | |
5480 | sys_sched_yield(); | |
5481 | } | |
1da177e4 LT |
5482 | EXPORT_SYMBOL(yield); |
5483 | ||
d95f4122 MG |
5484 | /** |
5485 | * yield_to - yield the current processor to another thread in | |
5486 | * your thread group, or accelerate that thread toward the | |
5487 | * processor it's on. | |
16addf95 RD |
5488 | * @p: target task |
5489 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
5490 | * |
5491 | * It's the caller's job to ensure that the target task struct | |
5492 | * can't go away on us before we can do any checks. | |
5493 | * | |
5494 | * Returns true if we indeed boosted the target task. | |
5495 | */ | |
5496 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
5497 | { | |
5498 | struct task_struct *curr = current; | |
5499 | struct rq *rq, *p_rq; | |
5500 | unsigned long flags; | |
5501 | bool yielded = 0; | |
5502 | ||
5503 | local_irq_save(flags); | |
5504 | rq = this_rq(); | |
5505 | ||
5506 | again: | |
5507 | p_rq = task_rq(p); | |
5508 | double_rq_lock(rq, p_rq); | |
5509 | while (task_rq(p) != p_rq) { | |
5510 | double_rq_unlock(rq, p_rq); | |
5511 | goto again; | |
5512 | } | |
5513 | ||
5514 | if (!curr->sched_class->yield_to_task) | |
5515 | goto out; | |
5516 | ||
5517 | if (curr->sched_class != p->sched_class) | |
5518 | goto out; | |
5519 | ||
5520 | if (task_running(p_rq, p) || p->state) | |
5521 | goto out; | |
5522 | ||
5523 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 5524 | if (yielded) { |
d95f4122 | 5525 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
5526 | /* |
5527 | * Make p's CPU reschedule; pick_next_entity takes care of | |
5528 | * fairness. | |
5529 | */ | |
5530 | if (preempt && rq != p_rq) | |
5531 | resched_task(p_rq->curr); | |
5532 | } | |
d95f4122 MG |
5533 | |
5534 | out: | |
5535 | double_rq_unlock(rq, p_rq); | |
5536 | local_irq_restore(flags); | |
5537 | ||
5538 | if (yielded) | |
5539 | schedule(); | |
5540 | ||
5541 | return yielded; | |
5542 | } | |
5543 | EXPORT_SYMBOL_GPL(yield_to); | |
5544 | ||
1da177e4 | 5545 | /* |
41a2d6cf | 5546 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5547 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
5548 | */ |
5549 | void __sched io_schedule(void) | |
5550 | { | |
54d35f29 | 5551 | struct rq *rq = raw_rq(); |
1da177e4 | 5552 | |
0ff92245 | 5553 | delayacct_blkio_start(); |
1da177e4 | 5554 | atomic_inc(&rq->nr_iowait); |
73c10101 | 5555 | blk_flush_plug(current); |
8f0dfc34 | 5556 | current->in_iowait = 1; |
1da177e4 | 5557 | schedule(); |
8f0dfc34 | 5558 | current->in_iowait = 0; |
1da177e4 | 5559 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5560 | delayacct_blkio_end(); |
1da177e4 | 5561 | } |
1da177e4 LT |
5562 | EXPORT_SYMBOL(io_schedule); |
5563 | ||
5564 | long __sched io_schedule_timeout(long timeout) | |
5565 | { | |
54d35f29 | 5566 | struct rq *rq = raw_rq(); |
1da177e4 LT |
5567 | long ret; |
5568 | ||
0ff92245 | 5569 | delayacct_blkio_start(); |
1da177e4 | 5570 | atomic_inc(&rq->nr_iowait); |
73c10101 | 5571 | blk_flush_plug(current); |
8f0dfc34 | 5572 | current->in_iowait = 1; |
1da177e4 | 5573 | ret = schedule_timeout(timeout); |
8f0dfc34 | 5574 | current->in_iowait = 0; |
1da177e4 | 5575 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5576 | delayacct_blkio_end(); |
1da177e4 LT |
5577 | return ret; |
5578 | } | |
5579 | ||
5580 | /** | |
5581 | * sys_sched_get_priority_max - return maximum RT priority. | |
5582 | * @policy: scheduling class. | |
5583 | * | |
5584 | * this syscall returns the maximum rt_priority that can be used | |
5585 | * by a given scheduling class. | |
5586 | */ | |
5add95d4 | 5587 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5588 | { |
5589 | int ret = -EINVAL; | |
5590 | ||
5591 | switch (policy) { | |
5592 | case SCHED_FIFO: | |
5593 | case SCHED_RR: | |
5594 | ret = MAX_USER_RT_PRIO-1; | |
5595 | break; | |
5596 | case SCHED_NORMAL: | |
b0a9499c | 5597 | case SCHED_BATCH: |
dd41f596 | 5598 | case SCHED_IDLE: |
1da177e4 LT |
5599 | ret = 0; |
5600 | break; | |
5601 | } | |
5602 | return ret; | |
5603 | } | |
5604 | ||
5605 | /** | |
5606 | * sys_sched_get_priority_min - return minimum RT priority. | |
5607 | * @policy: scheduling class. | |
5608 | * | |
5609 | * this syscall returns the minimum rt_priority that can be used | |
5610 | * by a given scheduling class. | |
5611 | */ | |
5add95d4 | 5612 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5613 | { |
5614 | int ret = -EINVAL; | |
5615 | ||
5616 | switch (policy) { | |
5617 | case SCHED_FIFO: | |
5618 | case SCHED_RR: | |
5619 | ret = 1; | |
5620 | break; | |
5621 | case SCHED_NORMAL: | |
b0a9499c | 5622 | case SCHED_BATCH: |
dd41f596 | 5623 | case SCHED_IDLE: |
1da177e4 LT |
5624 | ret = 0; |
5625 | } | |
5626 | return ret; | |
5627 | } | |
5628 | ||
5629 | /** | |
5630 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5631 | * @pid: pid of the process. | |
5632 | * @interval: userspace pointer to the timeslice value. | |
5633 | * | |
5634 | * this syscall writes the default timeslice value of a given process | |
5635 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5636 | */ | |
17da2bd9 | 5637 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5638 | struct timespec __user *, interval) |
1da177e4 | 5639 | { |
36c8b586 | 5640 | struct task_struct *p; |
a4ec24b4 | 5641 | unsigned int time_slice; |
dba091b9 TG |
5642 | unsigned long flags; |
5643 | struct rq *rq; | |
3a5c359a | 5644 | int retval; |
1da177e4 | 5645 | struct timespec t; |
1da177e4 LT |
5646 | |
5647 | if (pid < 0) | |
3a5c359a | 5648 | return -EINVAL; |
1da177e4 LT |
5649 | |
5650 | retval = -ESRCH; | |
1a551ae7 | 5651 | rcu_read_lock(); |
1da177e4 LT |
5652 | p = find_process_by_pid(pid); |
5653 | if (!p) | |
5654 | goto out_unlock; | |
5655 | ||
5656 | retval = security_task_getscheduler(p); | |
5657 | if (retval) | |
5658 | goto out_unlock; | |
5659 | ||
dba091b9 TG |
5660 | rq = task_rq_lock(p, &flags); |
5661 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
5662 | task_rq_unlock(rq, &flags); | |
a4ec24b4 | 5663 | |
1a551ae7 | 5664 | rcu_read_unlock(); |
a4ec24b4 | 5665 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5666 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5667 | return retval; |
3a5c359a | 5668 | |
1da177e4 | 5669 | out_unlock: |
1a551ae7 | 5670 | rcu_read_unlock(); |
1da177e4 LT |
5671 | return retval; |
5672 | } | |
5673 | ||
7c731e0a | 5674 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5675 | |
82a1fcb9 | 5676 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5677 | { |
1da177e4 | 5678 | unsigned long free = 0; |
36c8b586 | 5679 | unsigned state; |
1da177e4 | 5680 | |
1da177e4 | 5681 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 5682 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 5683 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5684 | #if BITS_PER_LONG == 32 |
1da177e4 | 5685 | if (state == TASK_RUNNING) |
3df0fc5b | 5686 | printk(KERN_CONT " running "); |
1da177e4 | 5687 | else |
3df0fc5b | 5688 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5689 | #else |
5690 | if (state == TASK_RUNNING) | |
3df0fc5b | 5691 | printk(KERN_CONT " running task "); |
1da177e4 | 5692 | else |
3df0fc5b | 5693 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5694 | #endif |
5695 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5696 | free = stack_not_used(p); |
1da177e4 | 5697 | #endif |
3df0fc5b | 5698 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
5699 | task_pid_nr(p), task_pid_nr(p->real_parent), |
5700 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 5701 | |
5fb5e6de | 5702 | show_stack(p, NULL); |
1da177e4 LT |
5703 | } |
5704 | ||
e59e2ae2 | 5705 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5706 | { |
36c8b586 | 5707 | struct task_struct *g, *p; |
1da177e4 | 5708 | |
4bd77321 | 5709 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5710 | printk(KERN_INFO |
5711 | " task PC stack pid father\n"); | |
1da177e4 | 5712 | #else |
3df0fc5b PZ |
5713 | printk(KERN_INFO |
5714 | " task PC stack pid father\n"); | |
1da177e4 LT |
5715 | #endif |
5716 | read_lock(&tasklist_lock); | |
5717 | do_each_thread(g, p) { | |
5718 | /* | |
5719 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 5720 | * console might take a lot of time: |
1da177e4 LT |
5721 | */ |
5722 | touch_nmi_watchdog(); | |
39bc89fd | 5723 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5724 | sched_show_task(p); |
1da177e4 LT |
5725 | } while_each_thread(g, p); |
5726 | ||
04c9167f JF |
5727 | touch_all_softlockup_watchdogs(); |
5728 | ||
dd41f596 IM |
5729 | #ifdef CONFIG_SCHED_DEBUG |
5730 | sysrq_sched_debug_show(); | |
5731 | #endif | |
1da177e4 | 5732 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5733 | /* |
5734 | * Only show locks if all tasks are dumped: | |
5735 | */ | |
93335a21 | 5736 | if (!state_filter) |
e59e2ae2 | 5737 | debug_show_all_locks(); |
1da177e4 LT |
5738 | } |
5739 | ||
1df21055 IM |
5740 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5741 | { | |
dd41f596 | 5742 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5743 | } |
5744 | ||
f340c0d1 IM |
5745 | /** |
5746 | * init_idle - set up an idle thread for a given CPU | |
5747 | * @idle: task in question | |
5748 | * @cpu: cpu the idle task belongs to | |
5749 | * | |
5750 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5751 | * flag, to make booting more robust. | |
5752 | */ | |
5c1e1767 | 5753 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5754 | { |
70b97a7f | 5755 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5756 | unsigned long flags; |
5757 | ||
05fa785c | 5758 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 5759 | |
dd41f596 | 5760 | __sched_fork(idle); |
06b83b5f | 5761 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5762 | idle->se.exec_start = sched_clock(); |
5763 | ||
96f874e2 | 5764 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
6506cf6c PZ |
5765 | /* |
5766 | * We're having a chicken and egg problem, even though we are | |
5767 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
5768 | * lockdep check in task_group() will fail. | |
5769 | * | |
5770 | * Similar case to sched_fork(). / Alternatively we could | |
5771 | * use task_rq_lock() here and obtain the other rq->lock. | |
5772 | * | |
5773 | * Silence PROVE_RCU | |
5774 | */ | |
5775 | rcu_read_lock(); | |
dd41f596 | 5776 | __set_task_cpu(idle, cpu); |
6506cf6c | 5777 | rcu_read_unlock(); |
1da177e4 | 5778 | |
1da177e4 | 5779 | rq->curr = rq->idle = idle; |
3ca7a440 PZ |
5780 | #if defined(CONFIG_SMP) |
5781 | idle->on_cpu = 1; | |
4866cde0 | 5782 | #endif |
05fa785c | 5783 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5784 | |
5785 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5786 | #if defined(CONFIG_PREEMPT) |
5787 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5788 | #else | |
a1261f54 | 5789 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5790 | #endif |
dd41f596 IM |
5791 | /* |
5792 | * The idle tasks have their own, simple scheduling class: | |
5793 | */ | |
5794 | idle->sched_class = &idle_sched_class; | |
868baf07 | 5795 | ftrace_graph_init_idle_task(idle, cpu); |
1da177e4 LT |
5796 | } |
5797 | ||
5798 | /* | |
5799 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5800 | * indicates which cpus entered this state. This is used | |
5801 | * in the rcu update to wait only for active cpus. For system | |
5802 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 5803 | * always be CPU_BITS_NONE. |
1da177e4 | 5804 | */ |
6a7b3dc3 | 5805 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 5806 | |
19978ca6 IM |
5807 | /* |
5808 | * Increase the granularity value when there are more CPUs, | |
5809 | * because with more CPUs the 'effective latency' as visible | |
5810 | * to users decreases. But the relationship is not linear, | |
5811 | * so pick a second-best guess by going with the log2 of the | |
5812 | * number of CPUs. | |
5813 | * | |
5814 | * This idea comes from the SD scheduler of Con Kolivas: | |
5815 | */ | |
acb4a848 | 5816 | static int get_update_sysctl_factor(void) |
19978ca6 | 5817 | { |
4ca3ef71 | 5818 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
5819 | unsigned int factor; |
5820 | ||
5821 | switch (sysctl_sched_tunable_scaling) { | |
5822 | case SCHED_TUNABLESCALING_NONE: | |
5823 | factor = 1; | |
5824 | break; | |
5825 | case SCHED_TUNABLESCALING_LINEAR: | |
5826 | factor = cpus; | |
5827 | break; | |
5828 | case SCHED_TUNABLESCALING_LOG: | |
5829 | default: | |
5830 | factor = 1 + ilog2(cpus); | |
5831 | break; | |
5832 | } | |
19978ca6 | 5833 | |
acb4a848 CE |
5834 | return factor; |
5835 | } | |
19978ca6 | 5836 | |
acb4a848 CE |
5837 | static void update_sysctl(void) |
5838 | { | |
5839 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 5840 | |
0bcdcf28 CE |
5841 | #define SET_SYSCTL(name) \ |
5842 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
5843 | SET_SYSCTL(sched_min_granularity); | |
5844 | SET_SYSCTL(sched_latency); | |
5845 | SET_SYSCTL(sched_wakeup_granularity); | |
0bcdcf28 CE |
5846 | #undef SET_SYSCTL |
5847 | } | |
55cd5340 | 5848 | |
0bcdcf28 CE |
5849 | static inline void sched_init_granularity(void) |
5850 | { | |
5851 | update_sysctl(); | |
19978ca6 IM |
5852 | } |
5853 | ||
1da177e4 LT |
5854 | #ifdef CONFIG_SMP |
5855 | /* | |
5856 | * This is how migration works: | |
5857 | * | |
969c7921 TH |
5858 | * 1) we invoke migration_cpu_stop() on the target CPU using |
5859 | * stop_one_cpu(). | |
5860 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
5861 | * off the CPU) | |
5862 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
5863 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 5864 | * it and puts it into the right queue. |
969c7921 TH |
5865 | * 5) stopper completes and stop_one_cpu() returns and the migration |
5866 | * is done. | |
1da177e4 LT |
5867 | */ |
5868 | ||
5869 | /* | |
5870 | * Change a given task's CPU affinity. Migrate the thread to a | |
5871 | * proper CPU and schedule it away if the CPU it's executing on | |
5872 | * is removed from the allowed bitmask. | |
5873 | * | |
5874 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5875 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5876 | * call is not atomic; no spinlocks may be held. |
5877 | */ | |
96f874e2 | 5878 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
5879 | { |
5880 | unsigned long flags; | |
70b97a7f | 5881 | struct rq *rq; |
969c7921 | 5882 | unsigned int dest_cpu; |
48f24c4d | 5883 | int ret = 0; |
1da177e4 | 5884 | |
65cc8e48 PZ |
5885 | /* |
5886 | * Serialize against TASK_WAKING so that ttwu() and wunt() can | |
5887 | * drop the rq->lock and still rely on ->cpus_allowed. | |
5888 | */ | |
5889 | again: | |
5890 | while (task_is_waking(p)) | |
5891 | cpu_relax(); | |
1da177e4 | 5892 | rq = task_rq_lock(p, &flags); |
65cc8e48 PZ |
5893 | if (task_is_waking(p)) { |
5894 | task_rq_unlock(rq, &flags); | |
5895 | goto again; | |
5896 | } | |
e2912009 | 5897 | |
6ad4c188 | 5898 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
5899 | ret = -EINVAL; |
5900 | goto out; | |
5901 | } | |
5902 | ||
9985b0ba | 5903 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 5904 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
5905 | ret = -EINVAL; |
5906 | goto out; | |
5907 | } | |
5908 | ||
73fe6aae | 5909 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5910 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5911 | else { |
96f874e2 RR |
5912 | cpumask_copy(&p->cpus_allowed, new_mask); |
5913 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
5914 | } |
5915 | ||
1da177e4 | 5916 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 5917 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
5918 | goto out; |
5919 | ||
969c7921 | 5920 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
b7a2b39d | 5921 | if (migrate_task(p, rq)) { |
969c7921 | 5922 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 LT |
5923 | /* Need help from migration thread: drop lock and wait. */ |
5924 | task_rq_unlock(rq, &flags); | |
969c7921 | 5925 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
5926 | tlb_migrate_finish(p->mm); |
5927 | return 0; | |
5928 | } | |
5929 | out: | |
5930 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5931 | |
1da177e4 LT |
5932 | return ret; |
5933 | } | |
cd8ba7cd | 5934 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5935 | |
5936 | /* | |
41a2d6cf | 5937 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5938 | * this because either it can't run here any more (set_cpus_allowed() |
5939 | * away from this CPU, or CPU going down), or because we're | |
5940 | * attempting to rebalance this task on exec (sched_exec). | |
5941 | * | |
5942 | * So we race with normal scheduler movements, but that's OK, as long | |
5943 | * as the task is no longer on this CPU. | |
efc30814 KK |
5944 | * |
5945 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5946 | */ |
efc30814 | 5947 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5948 | { |
70b97a7f | 5949 | struct rq *rq_dest, *rq_src; |
e2912009 | 5950 | int ret = 0; |
1da177e4 | 5951 | |
e761b772 | 5952 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5953 | return ret; |
1da177e4 LT |
5954 | |
5955 | rq_src = cpu_rq(src_cpu); | |
5956 | rq_dest = cpu_rq(dest_cpu); | |
5957 | ||
5958 | double_rq_lock(rq_src, rq_dest); | |
5959 | /* Already moved. */ | |
5960 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5961 | goto done; |
1da177e4 | 5962 | /* Affinity changed (again). */ |
96f874e2 | 5963 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 5964 | goto fail; |
1da177e4 | 5965 | |
e2912009 PZ |
5966 | /* |
5967 | * If we're not on a rq, the next wake-up will ensure we're | |
5968 | * placed properly. | |
5969 | */ | |
fd2f4419 | 5970 | if (p->on_rq) { |
2e1cb74a | 5971 | deactivate_task(rq_src, p, 0); |
e2912009 | 5972 | set_task_cpu(p, dest_cpu); |
dd41f596 | 5973 | activate_task(rq_dest, p, 0); |
15afe09b | 5974 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 5975 | } |
b1e38734 | 5976 | done: |
efc30814 | 5977 | ret = 1; |
b1e38734 | 5978 | fail: |
1da177e4 | 5979 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 5980 | return ret; |
1da177e4 LT |
5981 | } |
5982 | ||
5983 | /* | |
969c7921 TH |
5984 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
5985 | * and performs thread migration by bumping thread off CPU then | |
5986 | * 'pushing' onto another runqueue. | |
1da177e4 | 5987 | */ |
969c7921 | 5988 | static int migration_cpu_stop(void *data) |
1da177e4 | 5989 | { |
969c7921 | 5990 | struct migration_arg *arg = data; |
f7b4cddc | 5991 | |
969c7921 TH |
5992 | /* |
5993 | * The original target cpu might have gone down and we might | |
5994 | * be on another cpu but it doesn't matter. | |
5995 | */ | |
f7b4cddc | 5996 | local_irq_disable(); |
969c7921 | 5997 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 5998 | local_irq_enable(); |
1da177e4 | 5999 | return 0; |
f7b4cddc ON |
6000 | } |
6001 | ||
1da177e4 | 6002 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 6003 | |
054b9108 | 6004 | /* |
48c5ccae PZ |
6005 | * Ensures that the idle task is using init_mm right before its cpu goes |
6006 | * offline. | |
054b9108 | 6007 | */ |
48c5ccae | 6008 | void idle_task_exit(void) |
1da177e4 | 6009 | { |
48c5ccae | 6010 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 6011 | |
48c5ccae | 6012 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 6013 | |
48c5ccae PZ |
6014 | if (mm != &init_mm) |
6015 | switch_mm(mm, &init_mm, current); | |
6016 | mmdrop(mm); | |
1da177e4 LT |
6017 | } |
6018 | ||
6019 | /* | |
6020 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
6021 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
6022 | * for performance reasons the counter is not stricly tracking tasks to | |
6023 | * their home CPUs. So we just add the counter to another CPU's counter, | |
6024 | * to keep the global sum constant after CPU-down: | |
6025 | */ | |
70b97a7f | 6026 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 6027 | { |
6ad4c188 | 6028 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 | 6029 | |
1da177e4 LT |
6030 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; |
6031 | rq_src->nr_uninterruptible = 0; | |
1da177e4 LT |
6032 | } |
6033 | ||
dd41f596 | 6034 | /* |
48c5ccae | 6035 | * remove the tasks which were accounted by rq from calc_load_tasks. |
1da177e4 | 6036 | */ |
48c5ccae | 6037 | static void calc_global_load_remove(struct rq *rq) |
1da177e4 | 6038 | { |
48c5ccae PZ |
6039 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); |
6040 | rq->calc_load_active = 0; | |
1da177e4 LT |
6041 | } |
6042 | ||
48f24c4d | 6043 | /* |
48c5ccae PZ |
6044 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
6045 | * try_to_wake_up()->select_task_rq(). | |
6046 | * | |
6047 | * Called with rq->lock held even though we'er in stop_machine() and | |
6048 | * there's no concurrency possible, we hold the required locks anyway | |
6049 | * because of lock validation efforts. | |
1da177e4 | 6050 | */ |
48c5ccae | 6051 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 6052 | { |
70b97a7f | 6053 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
6054 | struct task_struct *next, *stop = rq->stop; |
6055 | int dest_cpu; | |
1da177e4 LT |
6056 | |
6057 | /* | |
48c5ccae PZ |
6058 | * Fudge the rq selection such that the below task selection loop |
6059 | * doesn't get stuck on the currently eligible stop task. | |
6060 | * | |
6061 | * We're currently inside stop_machine() and the rq is either stuck | |
6062 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
6063 | * either way we should never end up calling schedule() until we're | |
6064 | * done here. | |
1da177e4 | 6065 | */ |
48c5ccae | 6066 | rq->stop = NULL; |
48f24c4d | 6067 | |
dd41f596 | 6068 | for ( ; ; ) { |
48c5ccae PZ |
6069 | /* |
6070 | * There's this thread running, bail when that's the only | |
6071 | * remaining thread. | |
6072 | */ | |
6073 | if (rq->nr_running == 1) | |
dd41f596 | 6074 | break; |
48c5ccae | 6075 | |
b67802ea | 6076 | next = pick_next_task(rq); |
48c5ccae | 6077 | BUG_ON(!next); |
79c53799 | 6078 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 6079 | |
48c5ccae PZ |
6080 | /* Find suitable destination for @next, with force if needed. */ |
6081 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
6082 | raw_spin_unlock(&rq->lock); | |
6083 | ||
6084 | __migrate_task(next, dead_cpu, dest_cpu); | |
6085 | ||
6086 | raw_spin_lock(&rq->lock); | |
1da177e4 | 6087 | } |
dce48a84 | 6088 | |
48c5ccae | 6089 | rq->stop = stop; |
dce48a84 | 6090 | } |
48c5ccae | 6091 | |
1da177e4 LT |
6092 | #endif /* CONFIG_HOTPLUG_CPU */ |
6093 | ||
e692ab53 NP |
6094 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6095 | ||
6096 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6097 | { |
6098 | .procname = "sched_domain", | |
c57baf1e | 6099 | .mode = 0555, |
e0361851 | 6100 | }, |
56992309 | 6101 | {} |
e692ab53 NP |
6102 | }; |
6103 | ||
6104 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
6105 | { |
6106 | .procname = "kernel", | |
c57baf1e | 6107 | .mode = 0555, |
e0361851 AD |
6108 | .child = sd_ctl_dir, |
6109 | }, | |
56992309 | 6110 | {} |
e692ab53 NP |
6111 | }; |
6112 | ||
6113 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6114 | { | |
6115 | struct ctl_table *entry = | |
5cf9f062 | 6116 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6117 | |
e692ab53 NP |
6118 | return entry; |
6119 | } | |
6120 | ||
6382bc90 MM |
6121 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6122 | { | |
cd790076 | 6123 | struct ctl_table *entry; |
6382bc90 | 6124 | |
cd790076 MM |
6125 | /* |
6126 | * In the intermediate directories, both the child directory and | |
6127 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6128 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6129 | * static strings and all have proc handlers. |
6130 | */ | |
6131 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6132 | if (entry->child) |
6133 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6134 | if (entry->proc_handler == NULL) |
6135 | kfree(entry->procname); | |
6136 | } | |
6382bc90 MM |
6137 | |
6138 | kfree(*tablep); | |
6139 | *tablep = NULL; | |
6140 | } | |
6141 | ||
e692ab53 | 6142 | static void |
e0361851 | 6143 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6144 | const char *procname, void *data, int maxlen, |
6145 | mode_t mode, proc_handler *proc_handler) | |
6146 | { | |
e692ab53 NP |
6147 | entry->procname = procname; |
6148 | entry->data = data; | |
6149 | entry->maxlen = maxlen; | |
6150 | entry->mode = mode; | |
6151 | entry->proc_handler = proc_handler; | |
6152 | } | |
6153 | ||
6154 | static struct ctl_table * | |
6155 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6156 | { | |
a5d8c348 | 6157 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 6158 | |
ad1cdc1d MM |
6159 | if (table == NULL) |
6160 | return NULL; | |
6161 | ||
e0361851 | 6162 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6163 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6164 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6165 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6166 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6167 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6168 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6169 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6170 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6171 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6172 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6173 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6174 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6175 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6176 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6177 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6178 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6179 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6180 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6181 | &sd->cache_nice_tries, |
6182 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6183 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6184 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
6185 | set_table_entry(&table[11], "name", sd->name, |
6186 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
6187 | /* &table[12] is terminator */ | |
e692ab53 NP |
6188 | |
6189 | return table; | |
6190 | } | |
6191 | ||
9a4e7159 | 6192 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6193 | { |
6194 | struct ctl_table *entry, *table; | |
6195 | struct sched_domain *sd; | |
6196 | int domain_num = 0, i; | |
6197 | char buf[32]; | |
6198 | ||
6199 | for_each_domain(cpu, sd) | |
6200 | domain_num++; | |
6201 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6202 | if (table == NULL) |
6203 | return NULL; | |
e692ab53 NP |
6204 | |
6205 | i = 0; | |
6206 | for_each_domain(cpu, sd) { | |
6207 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6208 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6209 | entry->mode = 0555; |
e692ab53 NP |
6210 | entry->child = sd_alloc_ctl_domain_table(sd); |
6211 | entry++; | |
6212 | i++; | |
6213 | } | |
6214 | return table; | |
6215 | } | |
6216 | ||
6217 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6218 | static void register_sched_domain_sysctl(void) |
e692ab53 | 6219 | { |
6ad4c188 | 6220 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
6221 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
6222 | char buf[32]; | |
6223 | ||
7378547f MM |
6224 | WARN_ON(sd_ctl_dir[0].child); |
6225 | sd_ctl_dir[0].child = entry; | |
6226 | ||
ad1cdc1d MM |
6227 | if (entry == NULL) |
6228 | return; | |
6229 | ||
6ad4c188 | 6230 | for_each_possible_cpu(i) { |
e692ab53 | 6231 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6232 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6233 | entry->mode = 0555; |
e692ab53 | 6234 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6235 | entry++; |
e692ab53 | 6236 | } |
7378547f MM |
6237 | |
6238 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6239 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6240 | } | |
6382bc90 | 6241 | |
7378547f | 6242 | /* may be called multiple times per register */ |
6382bc90 MM |
6243 | static void unregister_sched_domain_sysctl(void) |
6244 | { | |
7378547f MM |
6245 | if (sd_sysctl_header) |
6246 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6247 | sd_sysctl_header = NULL; |
7378547f MM |
6248 | if (sd_ctl_dir[0].child) |
6249 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6250 | } |
e692ab53 | 6251 | #else |
6382bc90 MM |
6252 | static void register_sched_domain_sysctl(void) |
6253 | { | |
6254 | } | |
6255 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6256 | { |
6257 | } | |
6258 | #endif | |
6259 | ||
1f11eb6a GH |
6260 | static void set_rq_online(struct rq *rq) |
6261 | { | |
6262 | if (!rq->online) { | |
6263 | const struct sched_class *class; | |
6264 | ||
c6c4927b | 6265 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6266 | rq->online = 1; |
6267 | ||
6268 | for_each_class(class) { | |
6269 | if (class->rq_online) | |
6270 | class->rq_online(rq); | |
6271 | } | |
6272 | } | |
6273 | } | |
6274 | ||
6275 | static void set_rq_offline(struct rq *rq) | |
6276 | { | |
6277 | if (rq->online) { | |
6278 | const struct sched_class *class; | |
6279 | ||
6280 | for_each_class(class) { | |
6281 | if (class->rq_offline) | |
6282 | class->rq_offline(rq); | |
6283 | } | |
6284 | ||
c6c4927b | 6285 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6286 | rq->online = 0; |
6287 | } | |
6288 | } | |
6289 | ||
1da177e4 LT |
6290 | /* |
6291 | * migration_call - callback that gets triggered when a CPU is added. | |
6292 | * Here we can start up the necessary migration thread for the new CPU. | |
6293 | */ | |
48f24c4d IM |
6294 | static int __cpuinit |
6295 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6296 | { |
48f24c4d | 6297 | int cpu = (long)hcpu; |
1da177e4 | 6298 | unsigned long flags; |
969c7921 | 6299 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 6300 | |
48c5ccae | 6301 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 6302 | |
1da177e4 | 6303 | case CPU_UP_PREPARE: |
a468d389 | 6304 | rq->calc_load_update = calc_load_update; |
1da177e4 | 6305 | break; |
48f24c4d | 6306 | |
1da177e4 | 6307 | case CPU_ONLINE: |
1f94ef59 | 6308 | /* Update our root-domain */ |
05fa785c | 6309 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 6310 | if (rq->rd) { |
c6c4927b | 6311 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
6312 | |
6313 | set_rq_online(rq); | |
1f94ef59 | 6314 | } |
05fa785c | 6315 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 6316 | break; |
48f24c4d | 6317 | |
1da177e4 | 6318 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 6319 | case CPU_DYING: |
57d885fe | 6320 | /* Update our root-domain */ |
05fa785c | 6321 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 6322 | if (rq->rd) { |
c6c4927b | 6323 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 6324 | set_rq_offline(rq); |
57d885fe | 6325 | } |
48c5ccae PZ |
6326 | migrate_tasks(cpu); |
6327 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 6328 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
48c5ccae PZ |
6329 | |
6330 | migrate_nr_uninterruptible(rq); | |
6331 | calc_global_load_remove(rq); | |
57d885fe | 6332 | break; |
1da177e4 LT |
6333 | #endif |
6334 | } | |
49c022e6 PZ |
6335 | |
6336 | update_max_interval(); | |
6337 | ||
1da177e4 LT |
6338 | return NOTIFY_OK; |
6339 | } | |
6340 | ||
f38b0820 PM |
6341 | /* |
6342 | * Register at high priority so that task migration (migrate_all_tasks) | |
6343 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 6344 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 6345 | */ |
26c2143b | 6346 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 6347 | .notifier_call = migration_call, |
50a323b7 | 6348 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
6349 | }; |
6350 | ||
3a101d05 TH |
6351 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
6352 | unsigned long action, void *hcpu) | |
6353 | { | |
6354 | switch (action & ~CPU_TASKS_FROZEN) { | |
6355 | case CPU_ONLINE: | |
6356 | case CPU_DOWN_FAILED: | |
6357 | set_cpu_active((long)hcpu, true); | |
6358 | return NOTIFY_OK; | |
6359 | default: | |
6360 | return NOTIFY_DONE; | |
6361 | } | |
6362 | } | |
6363 | ||
6364 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
6365 | unsigned long action, void *hcpu) | |
6366 | { | |
6367 | switch (action & ~CPU_TASKS_FROZEN) { | |
6368 | case CPU_DOWN_PREPARE: | |
6369 | set_cpu_active((long)hcpu, false); | |
6370 | return NOTIFY_OK; | |
6371 | default: | |
6372 | return NOTIFY_DONE; | |
6373 | } | |
6374 | } | |
6375 | ||
7babe8db | 6376 | static int __init migration_init(void) |
1da177e4 LT |
6377 | { |
6378 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6379 | int err; |
48f24c4d | 6380 | |
3a101d05 | 6381 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
6382 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6383 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6384 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6385 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 6386 | |
3a101d05 TH |
6387 | /* Register cpu active notifiers */ |
6388 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
6389 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
6390 | ||
a004cd42 | 6391 | return 0; |
1da177e4 | 6392 | } |
7babe8db | 6393 | early_initcall(migration_init); |
1da177e4 LT |
6394 | #endif |
6395 | ||
6396 | #ifdef CONFIG_SMP | |
476f3534 | 6397 | |
3e9830dc | 6398 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6399 | |
f6630114 MT |
6400 | static __read_mostly int sched_domain_debug_enabled; |
6401 | ||
6402 | static int __init sched_domain_debug_setup(char *str) | |
6403 | { | |
6404 | sched_domain_debug_enabled = 1; | |
6405 | ||
6406 | return 0; | |
6407 | } | |
6408 | early_param("sched_debug", sched_domain_debug_setup); | |
6409 | ||
7c16ec58 | 6410 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 6411 | struct cpumask *groupmask) |
1da177e4 | 6412 | { |
4dcf6aff | 6413 | struct sched_group *group = sd->groups; |
434d53b0 | 6414 | char str[256]; |
1da177e4 | 6415 | |
968ea6d8 | 6416 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 6417 | cpumask_clear(groupmask); |
4dcf6aff IM |
6418 | |
6419 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6420 | ||
6421 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 6422 | printk("does not load-balance\n"); |
4dcf6aff | 6423 | if (sd->parent) |
3df0fc5b PZ |
6424 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
6425 | " has parent"); | |
4dcf6aff | 6426 | return -1; |
41c7ce9a NP |
6427 | } |
6428 | ||
3df0fc5b | 6429 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 6430 | |
758b2cdc | 6431 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
6432 | printk(KERN_ERR "ERROR: domain->span does not contain " |
6433 | "CPU%d\n", cpu); | |
4dcf6aff | 6434 | } |
758b2cdc | 6435 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
6436 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
6437 | " CPU%d\n", cpu); | |
4dcf6aff | 6438 | } |
1da177e4 | 6439 | |
4dcf6aff | 6440 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6441 | do { |
4dcf6aff | 6442 | if (!group) { |
3df0fc5b PZ |
6443 | printk("\n"); |
6444 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6445 | break; |
6446 | } | |
6447 | ||
18a3885f | 6448 | if (!group->cpu_power) { |
3df0fc5b PZ |
6449 | printk(KERN_CONT "\n"); |
6450 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6451 | "set\n"); | |
4dcf6aff IM |
6452 | break; |
6453 | } | |
1da177e4 | 6454 | |
758b2cdc | 6455 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
6456 | printk(KERN_CONT "\n"); |
6457 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
6458 | break; |
6459 | } | |
1da177e4 | 6460 | |
758b2cdc | 6461 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
6462 | printk(KERN_CONT "\n"); |
6463 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
6464 | break; |
6465 | } | |
1da177e4 | 6466 | |
758b2cdc | 6467 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 6468 | |
968ea6d8 | 6469 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 6470 | |
3df0fc5b | 6471 | printk(KERN_CONT " %s", str); |
18a3885f | 6472 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
3df0fc5b PZ |
6473 | printk(KERN_CONT " (cpu_power = %d)", |
6474 | group->cpu_power); | |
381512cf | 6475 | } |
1da177e4 | 6476 | |
4dcf6aff IM |
6477 | group = group->next; |
6478 | } while (group != sd->groups); | |
3df0fc5b | 6479 | printk(KERN_CONT "\n"); |
1da177e4 | 6480 | |
758b2cdc | 6481 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 6482 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6483 | |
758b2cdc RR |
6484 | if (sd->parent && |
6485 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
6486 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6487 | "of domain->span\n"); | |
4dcf6aff IM |
6488 | return 0; |
6489 | } | |
1da177e4 | 6490 | |
4dcf6aff IM |
6491 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6492 | { | |
d5dd3db1 | 6493 | cpumask_var_t groupmask; |
4dcf6aff | 6494 | int level = 0; |
1da177e4 | 6495 | |
f6630114 MT |
6496 | if (!sched_domain_debug_enabled) |
6497 | return; | |
6498 | ||
4dcf6aff IM |
6499 | if (!sd) { |
6500 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6501 | return; | |
6502 | } | |
1da177e4 | 6503 | |
4dcf6aff IM |
6504 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6505 | ||
d5dd3db1 | 6506 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
6507 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
6508 | return; | |
6509 | } | |
6510 | ||
4dcf6aff | 6511 | for (;;) { |
7c16ec58 | 6512 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6513 | break; |
1da177e4 LT |
6514 | level++; |
6515 | sd = sd->parent; | |
33859f7f | 6516 | if (!sd) |
4dcf6aff IM |
6517 | break; |
6518 | } | |
d5dd3db1 | 6519 | free_cpumask_var(groupmask); |
1da177e4 | 6520 | } |
6d6bc0ad | 6521 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6522 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6523 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6524 | |
1a20ff27 | 6525 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 6526 | { |
758b2cdc | 6527 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
6528 | return 1; |
6529 | ||
6530 | /* Following flags need at least 2 groups */ | |
6531 | if (sd->flags & (SD_LOAD_BALANCE | | |
6532 | SD_BALANCE_NEWIDLE | | |
6533 | SD_BALANCE_FORK | | |
89c4710e SS |
6534 | SD_BALANCE_EXEC | |
6535 | SD_SHARE_CPUPOWER | | |
6536 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6537 | if (sd->groups != sd->groups->next) |
6538 | return 0; | |
6539 | } | |
6540 | ||
6541 | /* Following flags don't use groups */ | |
c88d5910 | 6542 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
6543 | return 0; |
6544 | ||
6545 | return 1; | |
6546 | } | |
6547 | ||
48f24c4d IM |
6548 | static int |
6549 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6550 | { |
6551 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6552 | ||
6553 | if (sd_degenerate(parent)) | |
6554 | return 1; | |
6555 | ||
758b2cdc | 6556 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
6557 | return 0; |
6558 | ||
245af2c7 SS |
6559 | /* Flags needing groups don't count if only 1 group in parent */ |
6560 | if (parent->groups == parent->groups->next) { | |
6561 | pflags &= ~(SD_LOAD_BALANCE | | |
6562 | SD_BALANCE_NEWIDLE | | |
6563 | SD_BALANCE_FORK | | |
89c4710e SS |
6564 | SD_BALANCE_EXEC | |
6565 | SD_SHARE_CPUPOWER | | |
6566 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6567 | if (nr_node_ids == 1) |
6568 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6569 | } |
6570 | if (~cflags & pflags) | |
6571 | return 0; | |
6572 | ||
6573 | return 1; | |
6574 | } | |
6575 | ||
c6c4927b RR |
6576 | static void free_rootdomain(struct root_domain *rd) |
6577 | { | |
047106ad PZ |
6578 | synchronize_sched(); |
6579 | ||
68e74568 RR |
6580 | cpupri_cleanup(&rd->cpupri); |
6581 | ||
c6c4927b RR |
6582 | free_cpumask_var(rd->rto_mask); |
6583 | free_cpumask_var(rd->online); | |
6584 | free_cpumask_var(rd->span); | |
6585 | kfree(rd); | |
6586 | } | |
6587 | ||
57d885fe GH |
6588 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6589 | { | |
a0490fa3 | 6590 | struct root_domain *old_rd = NULL; |
57d885fe | 6591 | unsigned long flags; |
57d885fe | 6592 | |
05fa785c | 6593 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
6594 | |
6595 | if (rq->rd) { | |
a0490fa3 | 6596 | old_rd = rq->rd; |
57d885fe | 6597 | |
c6c4927b | 6598 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6599 | set_rq_offline(rq); |
57d885fe | 6600 | |
c6c4927b | 6601 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6602 | |
a0490fa3 IM |
6603 | /* |
6604 | * If we dont want to free the old_rt yet then | |
6605 | * set old_rd to NULL to skip the freeing later | |
6606 | * in this function: | |
6607 | */ | |
6608 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
6609 | old_rd = NULL; | |
57d885fe GH |
6610 | } |
6611 | ||
6612 | atomic_inc(&rd->refcount); | |
6613 | rq->rd = rd; | |
6614 | ||
c6c4927b | 6615 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 6616 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 6617 | set_rq_online(rq); |
57d885fe | 6618 | |
05fa785c | 6619 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
6620 | |
6621 | if (old_rd) | |
6622 | free_rootdomain(old_rd); | |
57d885fe GH |
6623 | } |
6624 | ||
68c38fc3 | 6625 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6626 | { |
6627 | memset(rd, 0, sizeof(*rd)); | |
6628 | ||
68c38fc3 | 6629 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 6630 | goto out; |
68c38fc3 | 6631 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 6632 | goto free_span; |
68c38fc3 | 6633 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 6634 | goto free_online; |
6e0534f2 | 6635 | |
68c38fc3 | 6636 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 6637 | goto free_rto_mask; |
c6c4927b | 6638 | return 0; |
6e0534f2 | 6639 | |
68e74568 RR |
6640 | free_rto_mask: |
6641 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6642 | free_online: |
6643 | free_cpumask_var(rd->online); | |
6644 | free_span: | |
6645 | free_cpumask_var(rd->span); | |
0c910d28 | 6646 | out: |
c6c4927b | 6647 | return -ENOMEM; |
57d885fe GH |
6648 | } |
6649 | ||
6650 | static void init_defrootdomain(void) | |
6651 | { | |
68c38fc3 | 6652 | init_rootdomain(&def_root_domain); |
c6c4927b | 6653 | |
57d885fe GH |
6654 | atomic_set(&def_root_domain.refcount, 1); |
6655 | } | |
6656 | ||
dc938520 | 6657 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6658 | { |
6659 | struct root_domain *rd; | |
6660 | ||
6661 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6662 | if (!rd) | |
6663 | return NULL; | |
6664 | ||
68c38fc3 | 6665 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
6666 | kfree(rd); |
6667 | return NULL; | |
6668 | } | |
57d885fe GH |
6669 | |
6670 | return rd; | |
6671 | } | |
6672 | ||
1da177e4 | 6673 | /* |
0eab9146 | 6674 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6675 | * hold the hotplug lock. |
6676 | */ | |
0eab9146 IM |
6677 | static void |
6678 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6679 | { |
70b97a7f | 6680 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6681 | struct sched_domain *tmp; |
6682 | ||
669c55e9 PZ |
6683 | for (tmp = sd; tmp; tmp = tmp->parent) |
6684 | tmp->span_weight = cpumask_weight(sched_domain_span(tmp)); | |
6685 | ||
245af2c7 | 6686 | /* Remove the sched domains which do not contribute to scheduling. */ |
f29c9b1c | 6687 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6688 | struct sched_domain *parent = tmp->parent; |
6689 | if (!parent) | |
6690 | break; | |
f29c9b1c | 6691 | |
1a848870 | 6692 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6693 | tmp->parent = parent->parent; |
1a848870 SS |
6694 | if (parent->parent) |
6695 | parent->parent->child = tmp; | |
f29c9b1c LZ |
6696 | } else |
6697 | tmp = tmp->parent; | |
245af2c7 SS |
6698 | } |
6699 | ||
1a848870 | 6700 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6701 | sd = sd->parent; |
1a848870 SS |
6702 | if (sd) |
6703 | sd->child = NULL; | |
6704 | } | |
1da177e4 LT |
6705 | |
6706 | sched_domain_debug(sd, cpu); | |
6707 | ||
57d885fe | 6708 | rq_attach_root(rq, rd); |
674311d5 | 6709 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6710 | } |
6711 | ||
6712 | /* cpus with isolated domains */ | |
dcc30a35 | 6713 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
6714 | |
6715 | /* Setup the mask of cpus configured for isolated domains */ | |
6716 | static int __init isolated_cpu_setup(char *str) | |
6717 | { | |
bdddd296 | 6718 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 6719 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
6720 | return 1; |
6721 | } | |
6722 | ||
8927f494 | 6723 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6724 | |
6725 | /* | |
6711cab4 SS |
6726 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6727 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
6728 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
6729 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
6730 | * |
6731 | * init_sched_build_groups will build a circular linked list of the groups | |
6732 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6733 | * and ->cpu_power to 0. | |
6734 | */ | |
a616058b | 6735 | static void |
96f874e2 RR |
6736 | init_sched_build_groups(const struct cpumask *span, |
6737 | const struct cpumask *cpu_map, | |
6738 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 6739 | struct sched_group **sg, |
96f874e2 RR |
6740 | struct cpumask *tmpmask), |
6741 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
6742 | { |
6743 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6744 | int i; |
6745 | ||
96f874e2 | 6746 | cpumask_clear(covered); |
7c16ec58 | 6747 | |
abcd083a | 6748 | for_each_cpu(i, span) { |
6711cab4 | 6749 | struct sched_group *sg; |
7c16ec58 | 6750 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6751 | int j; |
6752 | ||
758b2cdc | 6753 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
6754 | continue; |
6755 | ||
758b2cdc | 6756 | cpumask_clear(sched_group_cpus(sg)); |
18a3885f | 6757 | sg->cpu_power = 0; |
1da177e4 | 6758 | |
abcd083a | 6759 | for_each_cpu(j, span) { |
7c16ec58 | 6760 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6761 | continue; |
6762 | ||
96f874e2 | 6763 | cpumask_set_cpu(j, covered); |
758b2cdc | 6764 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
6765 | } |
6766 | if (!first) | |
6767 | first = sg; | |
6768 | if (last) | |
6769 | last->next = sg; | |
6770 | last = sg; | |
6771 | } | |
6772 | last->next = first; | |
6773 | } | |
6774 | ||
9c1cfda2 | 6775 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6776 | |
9c1cfda2 | 6777 | #ifdef CONFIG_NUMA |
198e2f18 | 6778 | |
9c1cfda2 JH |
6779 | /** |
6780 | * find_next_best_node - find the next node to include in a sched_domain | |
6781 | * @node: node whose sched_domain we're building | |
6782 | * @used_nodes: nodes already in the sched_domain | |
6783 | * | |
41a2d6cf | 6784 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6785 | * finds the closest node not already in the @used_nodes map. |
6786 | * | |
6787 | * Should use nodemask_t. | |
6788 | */ | |
c5f59f08 | 6789 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6790 | { |
6791 | int i, n, val, min_val, best_node = 0; | |
6792 | ||
6793 | min_val = INT_MAX; | |
6794 | ||
076ac2af | 6795 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6796 | /* Start at @node */ |
076ac2af | 6797 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6798 | |
6799 | if (!nr_cpus_node(n)) | |
6800 | continue; | |
6801 | ||
6802 | /* Skip already used nodes */ | |
c5f59f08 | 6803 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6804 | continue; |
6805 | ||
6806 | /* Simple min distance search */ | |
6807 | val = node_distance(node, n); | |
6808 | ||
6809 | if (val < min_val) { | |
6810 | min_val = val; | |
6811 | best_node = n; | |
6812 | } | |
6813 | } | |
6814 | ||
c5f59f08 | 6815 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6816 | return best_node; |
6817 | } | |
6818 | ||
6819 | /** | |
6820 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6821 | * @node: node whose cpumask we're constructing | |
73486722 | 6822 | * @span: resulting cpumask |
9c1cfda2 | 6823 | * |
41a2d6cf | 6824 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6825 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6826 | * out optimally. | |
6827 | */ | |
96f874e2 | 6828 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 6829 | { |
c5f59f08 | 6830 | nodemask_t used_nodes; |
48f24c4d | 6831 | int i; |
9c1cfda2 | 6832 | |
6ca09dfc | 6833 | cpumask_clear(span); |
c5f59f08 | 6834 | nodes_clear(used_nodes); |
9c1cfda2 | 6835 | |
6ca09dfc | 6836 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 6837 | node_set(node, used_nodes); |
9c1cfda2 JH |
6838 | |
6839 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6840 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6841 | |
6ca09dfc | 6842 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 6843 | } |
9c1cfda2 | 6844 | } |
6d6bc0ad | 6845 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6846 | |
5c45bf27 | 6847 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6848 | |
6c99e9ad RR |
6849 | /* |
6850 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
6851 | * |
6852 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
6853 | * and struct sched_domain. ) | |
6c99e9ad RR |
6854 | */ |
6855 | struct static_sched_group { | |
6856 | struct sched_group sg; | |
6857 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
6858 | }; | |
6859 | ||
6860 | struct static_sched_domain { | |
6861 | struct sched_domain sd; | |
6862 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
6863 | }; | |
6864 | ||
49a02c51 AH |
6865 | struct s_data { |
6866 | #ifdef CONFIG_NUMA | |
6867 | int sd_allnodes; | |
6868 | cpumask_var_t domainspan; | |
6869 | cpumask_var_t covered; | |
6870 | cpumask_var_t notcovered; | |
6871 | #endif | |
6872 | cpumask_var_t nodemask; | |
6873 | cpumask_var_t this_sibling_map; | |
6874 | cpumask_var_t this_core_map; | |
01a08546 | 6875 | cpumask_var_t this_book_map; |
49a02c51 AH |
6876 | cpumask_var_t send_covered; |
6877 | cpumask_var_t tmpmask; | |
6878 | struct sched_group **sched_group_nodes; | |
6879 | struct root_domain *rd; | |
6880 | }; | |
6881 | ||
2109b99e AH |
6882 | enum s_alloc { |
6883 | sa_sched_groups = 0, | |
6884 | sa_rootdomain, | |
6885 | sa_tmpmask, | |
6886 | sa_send_covered, | |
01a08546 | 6887 | sa_this_book_map, |
2109b99e AH |
6888 | sa_this_core_map, |
6889 | sa_this_sibling_map, | |
6890 | sa_nodemask, | |
6891 | sa_sched_group_nodes, | |
6892 | #ifdef CONFIG_NUMA | |
6893 | sa_notcovered, | |
6894 | sa_covered, | |
6895 | sa_domainspan, | |
6896 | #endif | |
6897 | sa_none, | |
6898 | }; | |
6899 | ||
9c1cfda2 | 6900 | /* |
48f24c4d | 6901 | * SMT sched-domains: |
9c1cfda2 | 6902 | */ |
1da177e4 | 6903 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 6904 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
1871e52c | 6905 | static DEFINE_PER_CPU(struct static_sched_group, sched_groups); |
48f24c4d | 6906 | |
41a2d6cf | 6907 | static int |
96f874e2 RR |
6908 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
6909 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 6910 | { |
6711cab4 | 6911 | if (sg) |
1871e52c | 6912 | *sg = &per_cpu(sched_groups, cpu).sg; |
1da177e4 LT |
6913 | return cpu; |
6914 | } | |
6d6bc0ad | 6915 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6916 | |
48f24c4d IM |
6917 | /* |
6918 | * multi-core sched-domains: | |
6919 | */ | |
1e9f28fa | 6920 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
6921 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
6922 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
1e9f28fa | 6923 | |
41a2d6cf | 6924 | static int |
96f874e2 RR |
6925 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6926 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6927 | { |
6711cab4 | 6928 | int group; |
f269893c | 6929 | #ifdef CONFIG_SCHED_SMT |
c69fc56d | 6930 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6931 | group = cpumask_first(mask); |
f269893c HC |
6932 | #else |
6933 | group = cpu; | |
6934 | #endif | |
6711cab4 | 6935 | if (sg) |
6c99e9ad | 6936 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 6937 | return group; |
1e9f28fa | 6938 | } |
f269893c | 6939 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa | 6940 | |
01a08546 HC |
6941 | /* |
6942 | * book sched-domains: | |
6943 | */ | |
6944 | #ifdef CONFIG_SCHED_BOOK | |
6945 | static DEFINE_PER_CPU(struct static_sched_domain, book_domains); | |
6946 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_book); | |
6947 | ||
41a2d6cf | 6948 | static int |
01a08546 HC |
6949 | cpu_to_book_group(int cpu, const struct cpumask *cpu_map, |
6950 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6951 | { |
01a08546 HC |
6952 | int group = cpu; |
6953 | #ifdef CONFIG_SCHED_MC | |
6954 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); | |
6955 | group = cpumask_first(mask); | |
6956 | #elif defined(CONFIG_SCHED_SMT) | |
6957 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); | |
6958 | group = cpumask_first(mask); | |
6959 | #endif | |
6711cab4 | 6960 | if (sg) |
01a08546 HC |
6961 | *sg = &per_cpu(sched_group_book, group).sg; |
6962 | return group; | |
1e9f28fa | 6963 | } |
01a08546 | 6964 | #endif /* CONFIG_SCHED_BOOK */ |
1e9f28fa | 6965 | |
6c99e9ad RR |
6966 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
6967 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 6968 | |
41a2d6cf | 6969 | static int |
96f874e2 RR |
6970 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
6971 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 6972 | { |
6711cab4 | 6973 | int group; |
01a08546 HC |
6974 | #ifdef CONFIG_SCHED_BOOK |
6975 | cpumask_and(mask, cpu_book_mask(cpu), cpu_map); | |
6976 | group = cpumask_first(mask); | |
6977 | #elif defined(CONFIG_SCHED_MC) | |
6ca09dfc | 6978 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 6979 | group = cpumask_first(mask); |
1e9f28fa | 6980 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 6981 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6982 | group = cpumask_first(mask); |
1da177e4 | 6983 | #else |
6711cab4 | 6984 | group = cpu; |
1da177e4 | 6985 | #endif |
6711cab4 | 6986 | if (sg) |
6c99e9ad | 6987 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 6988 | return group; |
1da177e4 LT |
6989 | } |
6990 | ||
6991 | #ifdef CONFIG_NUMA | |
1da177e4 | 6992 | /* |
9c1cfda2 JH |
6993 | * The init_sched_build_groups can't handle what we want to do with node |
6994 | * groups, so roll our own. Now each node has its own list of groups which | |
6995 | * gets dynamically allocated. | |
1da177e4 | 6996 | */ |
62ea9ceb | 6997 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 6998 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6999 | |
62ea9ceb | 7000 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 7001 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 7002 | |
96f874e2 RR |
7003 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
7004 | struct sched_group **sg, | |
7005 | struct cpumask *nodemask) | |
9c1cfda2 | 7006 | { |
6711cab4 SS |
7007 | int group; |
7008 | ||
6ca09dfc | 7009 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 7010 | group = cpumask_first(nodemask); |
6711cab4 SS |
7011 | |
7012 | if (sg) | |
6c99e9ad | 7013 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 7014 | return group; |
1da177e4 | 7015 | } |
6711cab4 | 7016 | |
08069033 SS |
7017 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
7018 | { | |
7019 | struct sched_group *sg = group_head; | |
7020 | int j; | |
7021 | ||
7022 | if (!sg) | |
7023 | return; | |
3a5c359a | 7024 | do { |
758b2cdc | 7025 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 7026 | struct sched_domain *sd; |
08069033 | 7027 | |
6c99e9ad | 7028 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 7029 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
7030 | /* |
7031 | * Only add "power" once for each | |
7032 | * physical package. | |
7033 | */ | |
7034 | continue; | |
7035 | } | |
08069033 | 7036 | |
18a3885f | 7037 | sg->cpu_power += sd->groups->cpu_power; |
3a5c359a AK |
7038 | } |
7039 | sg = sg->next; | |
7040 | } while (sg != group_head); | |
08069033 | 7041 | } |
0601a88d AH |
7042 | |
7043 | static int build_numa_sched_groups(struct s_data *d, | |
7044 | const struct cpumask *cpu_map, int num) | |
7045 | { | |
7046 | struct sched_domain *sd; | |
7047 | struct sched_group *sg, *prev; | |
7048 | int n, j; | |
7049 | ||
7050 | cpumask_clear(d->covered); | |
7051 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
7052 | if (cpumask_empty(d->nodemask)) { | |
7053 | d->sched_group_nodes[num] = NULL; | |
7054 | goto out; | |
7055 | } | |
7056 | ||
7057 | sched_domain_node_span(num, d->domainspan); | |
7058 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
7059 | ||
7060 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
7061 | GFP_KERNEL, num); | |
7062 | if (!sg) { | |
3df0fc5b PZ |
7063 | printk(KERN_WARNING "Can not alloc domain group for node %d\n", |
7064 | num); | |
0601a88d AH |
7065 | return -ENOMEM; |
7066 | } | |
7067 | d->sched_group_nodes[num] = sg; | |
7068 | ||
7069 | for_each_cpu(j, d->nodemask) { | |
7070 | sd = &per_cpu(node_domains, j).sd; | |
7071 | sd->groups = sg; | |
7072 | } | |
7073 | ||
18a3885f | 7074 | sg->cpu_power = 0; |
0601a88d AH |
7075 | cpumask_copy(sched_group_cpus(sg), d->nodemask); |
7076 | sg->next = sg; | |
7077 | cpumask_or(d->covered, d->covered, d->nodemask); | |
7078 | ||
7079 | prev = sg; | |
7080 | for (j = 0; j < nr_node_ids; j++) { | |
7081 | n = (num + j) % nr_node_ids; | |
7082 | cpumask_complement(d->notcovered, d->covered); | |
7083 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
7084 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
7085 | if (cpumask_empty(d->tmpmask)) | |
7086 | break; | |
7087 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
7088 | if (cpumask_empty(d->tmpmask)) | |
7089 | continue; | |
7090 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
7091 | GFP_KERNEL, num); | |
7092 | if (!sg) { | |
3df0fc5b PZ |
7093 | printk(KERN_WARNING |
7094 | "Can not alloc domain group for node %d\n", j); | |
0601a88d AH |
7095 | return -ENOMEM; |
7096 | } | |
18a3885f | 7097 | sg->cpu_power = 0; |
0601a88d AH |
7098 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); |
7099 | sg->next = prev->next; | |
7100 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
7101 | prev->next = sg; | |
7102 | prev = sg; | |
7103 | } | |
7104 | out: | |
7105 | return 0; | |
7106 | } | |
6d6bc0ad | 7107 | #endif /* CONFIG_NUMA */ |
1da177e4 | 7108 | |
a616058b | 7109 | #ifdef CONFIG_NUMA |
51888ca2 | 7110 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
7111 | static void free_sched_groups(const struct cpumask *cpu_map, |
7112 | struct cpumask *nodemask) | |
51888ca2 | 7113 | { |
a616058b | 7114 | int cpu, i; |
51888ca2 | 7115 | |
abcd083a | 7116 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
7117 | struct sched_group **sched_group_nodes |
7118 | = sched_group_nodes_bycpu[cpu]; | |
7119 | ||
51888ca2 SV |
7120 | if (!sched_group_nodes) |
7121 | continue; | |
7122 | ||
076ac2af | 7123 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
7124 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
7125 | ||
6ca09dfc | 7126 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 7127 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
7128 | continue; |
7129 | ||
7130 | if (sg == NULL) | |
7131 | continue; | |
7132 | sg = sg->next; | |
7133 | next_sg: | |
7134 | oldsg = sg; | |
7135 | sg = sg->next; | |
7136 | kfree(oldsg); | |
7137 | if (oldsg != sched_group_nodes[i]) | |
7138 | goto next_sg; | |
7139 | } | |
7140 | kfree(sched_group_nodes); | |
7141 | sched_group_nodes_bycpu[cpu] = NULL; | |
7142 | } | |
51888ca2 | 7143 | } |
6d6bc0ad | 7144 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
7145 | static void free_sched_groups(const struct cpumask *cpu_map, |
7146 | struct cpumask *nodemask) | |
a616058b SS |
7147 | { |
7148 | } | |
6d6bc0ad | 7149 | #endif /* CONFIG_NUMA */ |
51888ca2 | 7150 | |
89c4710e SS |
7151 | /* |
7152 | * Initialize sched groups cpu_power. | |
7153 | * | |
7154 | * cpu_power indicates the capacity of sched group, which is used while | |
7155 | * distributing the load between different sched groups in a sched domain. | |
7156 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
7157 | * there are asymmetries in the topology. If there are asymmetries, group | |
7158 | * having more cpu_power will pickup more load compared to the group having | |
7159 | * less cpu_power. | |
89c4710e SS |
7160 | */ |
7161 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
7162 | { | |
7163 | struct sched_domain *child; | |
7164 | struct sched_group *group; | |
f93e65c1 PZ |
7165 | long power; |
7166 | int weight; | |
89c4710e SS |
7167 | |
7168 | WARN_ON(!sd || !sd->groups); | |
7169 | ||
13318a71 | 7170 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
7171 | return; |
7172 | ||
aae6d3dd SS |
7173 | sd->groups->group_weight = cpumask_weight(sched_group_cpus(sd->groups)); |
7174 | ||
89c4710e SS |
7175 | child = sd->child; |
7176 | ||
18a3885f | 7177 | sd->groups->cpu_power = 0; |
5517d86b | 7178 | |
f93e65c1 PZ |
7179 | if (!child) { |
7180 | power = SCHED_LOAD_SCALE; | |
7181 | weight = cpumask_weight(sched_domain_span(sd)); | |
7182 | /* | |
7183 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
7184 | * Usually multiple threads get a better yield out of |
7185 | * that one core than a single thread would have, | |
7186 | * reflect that in sd->smt_gain. | |
f93e65c1 | 7187 | */ |
a52bfd73 PZ |
7188 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
7189 | power *= sd->smt_gain; | |
f93e65c1 | 7190 | power /= weight; |
a52bfd73 PZ |
7191 | power >>= SCHED_LOAD_SHIFT; |
7192 | } | |
18a3885f | 7193 | sd->groups->cpu_power += power; |
89c4710e SS |
7194 | return; |
7195 | } | |
7196 | ||
89c4710e | 7197 | /* |
f93e65c1 | 7198 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
7199 | */ |
7200 | group = child->groups; | |
7201 | do { | |
18a3885f | 7202 | sd->groups->cpu_power += group->cpu_power; |
89c4710e SS |
7203 | group = group->next; |
7204 | } while (group != child->groups); | |
7205 | } | |
7206 | ||
7c16ec58 MT |
7207 | /* |
7208 | * Initializers for schedule domains | |
7209 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7210 | */ | |
7211 | ||
a5d8c348 IM |
7212 | #ifdef CONFIG_SCHED_DEBUG |
7213 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
7214 | #else | |
7215 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
7216 | #endif | |
7217 | ||
7c16ec58 | 7218 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 7219 | |
7c16ec58 MT |
7220 | #define SD_INIT_FUNC(type) \ |
7221 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
7222 | { \ | |
7223 | memset(sd, 0, sizeof(*sd)); \ | |
7224 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 7225 | sd->level = SD_LV_##type; \ |
a5d8c348 | 7226 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
7227 | } |
7228 | ||
7229 | SD_INIT_FUNC(CPU) | |
7230 | #ifdef CONFIG_NUMA | |
7231 | SD_INIT_FUNC(ALLNODES) | |
7232 | SD_INIT_FUNC(NODE) | |
7233 | #endif | |
7234 | #ifdef CONFIG_SCHED_SMT | |
7235 | SD_INIT_FUNC(SIBLING) | |
7236 | #endif | |
7237 | #ifdef CONFIG_SCHED_MC | |
7238 | SD_INIT_FUNC(MC) | |
7239 | #endif | |
01a08546 HC |
7240 | #ifdef CONFIG_SCHED_BOOK |
7241 | SD_INIT_FUNC(BOOK) | |
7242 | #endif | |
7c16ec58 | 7243 | |
1d3504fc HS |
7244 | static int default_relax_domain_level = -1; |
7245 | ||
7246 | static int __init setup_relax_domain_level(char *str) | |
7247 | { | |
30e0e178 LZ |
7248 | unsigned long val; |
7249 | ||
7250 | val = simple_strtoul(str, NULL, 0); | |
7251 | if (val < SD_LV_MAX) | |
7252 | default_relax_domain_level = val; | |
7253 | ||
1d3504fc HS |
7254 | return 1; |
7255 | } | |
7256 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7257 | ||
7258 | static void set_domain_attribute(struct sched_domain *sd, | |
7259 | struct sched_domain_attr *attr) | |
7260 | { | |
7261 | int request; | |
7262 | ||
7263 | if (!attr || attr->relax_domain_level < 0) { | |
7264 | if (default_relax_domain_level < 0) | |
7265 | return; | |
7266 | else | |
7267 | request = default_relax_domain_level; | |
7268 | } else | |
7269 | request = attr->relax_domain_level; | |
7270 | if (request < sd->level) { | |
7271 | /* turn off idle balance on this domain */ | |
c88d5910 | 7272 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7273 | } else { |
7274 | /* turn on idle balance on this domain */ | |
c88d5910 | 7275 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7276 | } |
7277 | } | |
7278 | ||
2109b99e AH |
7279 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
7280 | const struct cpumask *cpu_map) | |
7281 | { | |
7282 | switch (what) { | |
7283 | case sa_sched_groups: | |
7284 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
7285 | d->sched_group_nodes = NULL; | |
7286 | case sa_rootdomain: | |
7287 | free_rootdomain(d->rd); /* fall through */ | |
7288 | case sa_tmpmask: | |
7289 | free_cpumask_var(d->tmpmask); /* fall through */ | |
7290 | case sa_send_covered: | |
7291 | free_cpumask_var(d->send_covered); /* fall through */ | |
01a08546 HC |
7292 | case sa_this_book_map: |
7293 | free_cpumask_var(d->this_book_map); /* fall through */ | |
2109b99e AH |
7294 | case sa_this_core_map: |
7295 | free_cpumask_var(d->this_core_map); /* fall through */ | |
7296 | case sa_this_sibling_map: | |
7297 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
7298 | case sa_nodemask: | |
7299 | free_cpumask_var(d->nodemask); /* fall through */ | |
7300 | case sa_sched_group_nodes: | |
d1b55138 | 7301 | #ifdef CONFIG_NUMA |
2109b99e AH |
7302 | kfree(d->sched_group_nodes); /* fall through */ |
7303 | case sa_notcovered: | |
7304 | free_cpumask_var(d->notcovered); /* fall through */ | |
7305 | case sa_covered: | |
7306 | free_cpumask_var(d->covered); /* fall through */ | |
7307 | case sa_domainspan: | |
7308 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 7309 | #endif |
2109b99e AH |
7310 | case sa_none: |
7311 | break; | |
7312 | } | |
7313 | } | |
3404c8d9 | 7314 | |
2109b99e AH |
7315 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
7316 | const struct cpumask *cpu_map) | |
7317 | { | |
3404c8d9 | 7318 | #ifdef CONFIG_NUMA |
2109b99e AH |
7319 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
7320 | return sa_none; | |
7321 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
7322 | return sa_domainspan; | |
7323 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
7324 | return sa_covered; | |
7325 | /* Allocate the per-node list of sched groups */ | |
7326 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
7327 | sizeof(struct sched_group *), GFP_KERNEL); | |
7328 | if (!d->sched_group_nodes) { | |
3df0fc5b | 7329 | printk(KERN_WARNING "Can not alloc sched group node list\n"); |
2109b99e | 7330 | return sa_notcovered; |
d1b55138 | 7331 | } |
2109b99e | 7332 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 7333 | #endif |
2109b99e AH |
7334 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
7335 | return sa_sched_group_nodes; | |
7336 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
7337 | return sa_nodemask; | |
7338 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
7339 | return sa_this_sibling_map; | |
01a08546 | 7340 | if (!alloc_cpumask_var(&d->this_book_map, GFP_KERNEL)) |
2109b99e | 7341 | return sa_this_core_map; |
01a08546 HC |
7342 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) |
7343 | return sa_this_book_map; | |
2109b99e AH |
7344 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) |
7345 | return sa_send_covered; | |
7346 | d->rd = alloc_rootdomain(); | |
7347 | if (!d->rd) { | |
3df0fc5b | 7348 | printk(KERN_WARNING "Cannot alloc root domain\n"); |
2109b99e | 7349 | return sa_tmpmask; |
57d885fe | 7350 | } |
2109b99e AH |
7351 | return sa_rootdomain; |
7352 | } | |
57d885fe | 7353 | |
7f4588f3 AH |
7354 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
7355 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
7356 | { | |
7357 | struct sched_domain *sd = NULL; | |
7c16ec58 | 7358 | #ifdef CONFIG_NUMA |
7f4588f3 | 7359 | struct sched_domain *parent; |
1da177e4 | 7360 | |
7f4588f3 AH |
7361 | d->sd_allnodes = 0; |
7362 | if (cpumask_weight(cpu_map) > | |
7363 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
7364 | sd = &per_cpu(allnodes_domains, i).sd; | |
7365 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 7366 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
7367 | cpumask_copy(sched_domain_span(sd), cpu_map); |
7368 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
7369 | d->sd_allnodes = 1; | |
7370 | } | |
7371 | parent = sd; | |
7372 | ||
7373 | sd = &per_cpu(node_domains, i).sd; | |
7374 | SD_INIT(sd, NODE); | |
7375 | set_domain_attribute(sd, attr); | |
7376 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
7377 | sd->parent = parent; | |
7378 | if (parent) | |
7379 | parent->child = sd; | |
7380 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 7381 | #endif |
7f4588f3 AH |
7382 | return sd; |
7383 | } | |
1da177e4 | 7384 | |
87cce662 AH |
7385 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
7386 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7387 | struct sched_domain *parent, int i) | |
7388 | { | |
7389 | struct sched_domain *sd; | |
7390 | sd = &per_cpu(phys_domains, i).sd; | |
7391 | SD_INIT(sd, CPU); | |
7392 | set_domain_attribute(sd, attr); | |
7393 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
7394 | sd->parent = parent; | |
7395 | if (parent) | |
7396 | parent->child = sd; | |
7397 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
7398 | return sd; | |
7399 | } | |
1da177e4 | 7400 | |
01a08546 HC |
7401 | static struct sched_domain *__build_book_sched_domain(struct s_data *d, |
7402 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7403 | struct sched_domain *parent, int i) | |
7404 | { | |
7405 | struct sched_domain *sd = parent; | |
7406 | #ifdef CONFIG_SCHED_BOOK | |
7407 | sd = &per_cpu(book_domains, i).sd; | |
7408 | SD_INIT(sd, BOOK); | |
7409 | set_domain_attribute(sd, attr); | |
7410 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_book_mask(i)); | |
7411 | sd->parent = parent; | |
7412 | parent->child = sd; | |
7413 | cpu_to_book_group(i, cpu_map, &sd->groups, d->tmpmask); | |
7414 | #endif | |
7415 | return sd; | |
7416 | } | |
7417 | ||
410c4081 AH |
7418 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
7419 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7420 | struct sched_domain *parent, int i) | |
7421 | { | |
7422 | struct sched_domain *sd = parent; | |
1e9f28fa | 7423 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
7424 | sd = &per_cpu(core_domains, i).sd; |
7425 | SD_INIT(sd, MC); | |
7426 | set_domain_attribute(sd, attr); | |
7427 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
7428 | sd->parent = parent; | |
7429 | parent->child = sd; | |
7430 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 7431 | #endif |
410c4081 AH |
7432 | return sd; |
7433 | } | |
1e9f28fa | 7434 | |
d8173535 AH |
7435 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
7436 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7437 | struct sched_domain *parent, int i) | |
7438 | { | |
7439 | struct sched_domain *sd = parent; | |
1da177e4 | 7440 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
7441 | sd = &per_cpu(cpu_domains, i).sd; |
7442 | SD_INIT(sd, SIBLING); | |
7443 | set_domain_attribute(sd, attr); | |
7444 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
7445 | sd->parent = parent; | |
7446 | parent->child = sd; | |
7447 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 7448 | #endif |
d8173535 AH |
7449 | return sd; |
7450 | } | |
1da177e4 | 7451 | |
0e8e85c9 AH |
7452 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
7453 | const struct cpumask *cpu_map, int cpu) | |
7454 | { | |
7455 | switch (l) { | |
1da177e4 | 7456 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
7457 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
7458 | cpumask_and(d->this_sibling_map, cpu_map, | |
7459 | topology_thread_cpumask(cpu)); | |
7460 | if (cpu == cpumask_first(d->this_sibling_map)) | |
7461 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
7462 | &cpu_to_cpu_group, | |
7463 | d->send_covered, d->tmpmask); | |
7464 | break; | |
1da177e4 | 7465 | #endif |
1e9f28fa | 7466 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
7467 | case SD_LV_MC: /* set up multi-core groups */ |
7468 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
7469 | if (cpu == cpumask_first(d->this_core_map)) | |
7470 | init_sched_build_groups(d->this_core_map, cpu_map, | |
7471 | &cpu_to_core_group, | |
7472 | d->send_covered, d->tmpmask); | |
7473 | break; | |
01a08546 HC |
7474 | #endif |
7475 | #ifdef CONFIG_SCHED_BOOK | |
7476 | case SD_LV_BOOK: /* set up book groups */ | |
7477 | cpumask_and(d->this_book_map, cpu_map, cpu_book_mask(cpu)); | |
7478 | if (cpu == cpumask_first(d->this_book_map)) | |
7479 | init_sched_build_groups(d->this_book_map, cpu_map, | |
7480 | &cpu_to_book_group, | |
7481 | d->send_covered, d->tmpmask); | |
7482 | break; | |
1e9f28fa | 7483 | #endif |
86548096 AH |
7484 | case SD_LV_CPU: /* set up physical groups */ |
7485 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
7486 | if (!cpumask_empty(d->nodemask)) | |
7487 | init_sched_build_groups(d->nodemask, cpu_map, | |
7488 | &cpu_to_phys_group, | |
7489 | d->send_covered, d->tmpmask); | |
7490 | break; | |
1da177e4 | 7491 | #ifdef CONFIG_NUMA |
de616e36 AH |
7492 | case SD_LV_ALLNODES: |
7493 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
7494 | d->send_covered, d->tmpmask); | |
7495 | break; | |
7496 | #endif | |
0e8e85c9 AH |
7497 | default: |
7498 | break; | |
7c16ec58 | 7499 | } |
0e8e85c9 | 7500 | } |
9c1cfda2 | 7501 | |
2109b99e AH |
7502 | /* |
7503 | * Build sched domains for a given set of cpus and attach the sched domains | |
7504 | * to the individual cpus | |
7505 | */ | |
7506 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
7507 | struct sched_domain_attr *attr) | |
7508 | { | |
7509 | enum s_alloc alloc_state = sa_none; | |
7510 | struct s_data d; | |
294b0c96 | 7511 | struct sched_domain *sd; |
2109b99e | 7512 | int i; |
7c16ec58 | 7513 | #ifdef CONFIG_NUMA |
2109b99e | 7514 | d.sd_allnodes = 0; |
7c16ec58 | 7515 | #endif |
9c1cfda2 | 7516 | |
2109b99e AH |
7517 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
7518 | if (alloc_state != sa_rootdomain) | |
7519 | goto error; | |
7520 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 7521 | |
1da177e4 | 7522 | /* |
1a20ff27 | 7523 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7524 | */ |
abcd083a | 7525 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
7526 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
7527 | cpu_map); | |
9761eea8 | 7528 | |
7f4588f3 | 7529 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 7530 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
01a08546 | 7531 | sd = __build_book_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 7532 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 7533 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 7534 | } |
9c1cfda2 | 7535 | |
abcd083a | 7536 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 7537 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
01a08546 | 7538 | build_sched_groups(&d, SD_LV_BOOK, cpu_map, i); |
a2af04cd | 7539 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 7540 | } |
9c1cfda2 | 7541 | |
1da177e4 | 7542 | /* Set up physical groups */ |
86548096 AH |
7543 | for (i = 0; i < nr_node_ids; i++) |
7544 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 7545 | |
1da177e4 LT |
7546 | #ifdef CONFIG_NUMA |
7547 | /* Set up node groups */ | |
de616e36 AH |
7548 | if (d.sd_allnodes) |
7549 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 7550 | |
0601a88d AH |
7551 | for (i = 0; i < nr_node_ids; i++) |
7552 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 7553 | goto error; |
1da177e4 LT |
7554 | #endif |
7555 | ||
7556 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7557 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 7558 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7559 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 7560 | init_sched_groups_power(i, sd); |
5c45bf27 | 7561 | } |
1da177e4 | 7562 | #endif |
1e9f28fa | 7563 | #ifdef CONFIG_SCHED_MC |
abcd083a | 7564 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7565 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 7566 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7567 | } |
7568 | #endif | |
01a08546 HC |
7569 | #ifdef CONFIG_SCHED_BOOK |
7570 | for_each_cpu(i, cpu_map) { | |
7571 | sd = &per_cpu(book_domains, i).sd; | |
7572 | init_sched_groups_power(i, sd); | |
7573 | } | |
7574 | #endif | |
1e9f28fa | 7575 | |
abcd083a | 7576 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7577 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 7578 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7579 | } |
7580 | ||
9c1cfda2 | 7581 | #ifdef CONFIG_NUMA |
076ac2af | 7582 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 7583 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 7584 | |
49a02c51 | 7585 | if (d.sd_allnodes) { |
6711cab4 | 7586 | struct sched_group *sg; |
f712c0c7 | 7587 | |
96f874e2 | 7588 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 7589 | d.tmpmask); |
f712c0c7 SS |
7590 | init_numa_sched_groups_power(sg); |
7591 | } | |
9c1cfda2 JH |
7592 | #endif |
7593 | ||
1da177e4 | 7594 | /* Attach the domains */ |
abcd083a | 7595 | for_each_cpu(i, cpu_map) { |
1da177e4 | 7596 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 7597 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 7598 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 7599 | sd = &per_cpu(core_domains, i).sd; |
01a08546 HC |
7600 | #elif defined(CONFIG_SCHED_BOOK) |
7601 | sd = &per_cpu(book_domains, i).sd; | |
1da177e4 | 7602 | #else |
6c99e9ad | 7603 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 7604 | #endif |
49a02c51 | 7605 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 7606 | } |
51888ca2 | 7607 | |
2109b99e AH |
7608 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
7609 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
7610 | return 0; | |
51888ca2 | 7611 | |
51888ca2 | 7612 | error: |
2109b99e AH |
7613 | __free_domain_allocs(&d, alloc_state, cpu_map); |
7614 | return -ENOMEM; | |
1da177e4 | 7615 | } |
029190c5 | 7616 | |
96f874e2 | 7617 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
7618 | { |
7619 | return __build_sched_domains(cpu_map, NULL); | |
7620 | } | |
7621 | ||
acc3f5d7 | 7622 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7623 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7624 | static struct sched_domain_attr *dattr_cur; |
7625 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7626 | |
7627 | /* | |
7628 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7629 | * cpumask) fails, then fallback to a single sched domain, |
7630 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7631 | */ |
4212823f | 7632 | static cpumask_var_t fallback_doms; |
029190c5 | 7633 | |
ee79d1bd HC |
7634 | /* |
7635 | * arch_update_cpu_topology lets virtualized architectures update the | |
7636 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7637 | * or 0 if it stayed the same. | |
7638 | */ | |
7639 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7640 | { |
ee79d1bd | 7641 | return 0; |
22e52b07 HC |
7642 | } |
7643 | ||
acc3f5d7 RR |
7644 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7645 | { | |
7646 | int i; | |
7647 | cpumask_var_t *doms; | |
7648 | ||
7649 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7650 | if (!doms) | |
7651 | return NULL; | |
7652 | for (i = 0; i < ndoms; i++) { | |
7653 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7654 | free_sched_domains(doms, i); | |
7655 | return NULL; | |
7656 | } | |
7657 | } | |
7658 | return doms; | |
7659 | } | |
7660 | ||
7661 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7662 | { | |
7663 | unsigned int i; | |
7664 | for (i = 0; i < ndoms; i++) | |
7665 | free_cpumask_var(doms[i]); | |
7666 | kfree(doms); | |
7667 | } | |
7668 | ||
1a20ff27 | 7669 | /* |
41a2d6cf | 7670 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7671 | * For now this just excludes isolated cpus, but could be used to |
7672 | * exclude other special cases in the future. | |
1a20ff27 | 7673 | */ |
96f874e2 | 7674 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7675 | { |
7378547f MM |
7676 | int err; |
7677 | ||
22e52b07 | 7678 | arch_update_cpu_topology(); |
029190c5 | 7679 | ndoms_cur = 1; |
acc3f5d7 | 7680 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7681 | if (!doms_cur) |
acc3f5d7 RR |
7682 | doms_cur = &fallback_doms; |
7683 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 7684 | dattr_cur = NULL; |
acc3f5d7 | 7685 | err = build_sched_domains(doms_cur[0]); |
6382bc90 | 7686 | register_sched_domain_sysctl(); |
7378547f MM |
7687 | |
7688 | return err; | |
1a20ff27 DG |
7689 | } |
7690 | ||
96f874e2 RR |
7691 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
7692 | struct cpumask *tmpmask) | |
1da177e4 | 7693 | { |
7c16ec58 | 7694 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7695 | } |
1da177e4 | 7696 | |
1a20ff27 DG |
7697 | /* |
7698 | * Detach sched domains from a group of cpus specified in cpu_map | |
7699 | * These cpus will now be attached to the NULL domain | |
7700 | */ | |
96f874e2 | 7701 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7702 | { |
96f874e2 RR |
7703 | /* Save because hotplug lock held. */ |
7704 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
7705 | int i; |
7706 | ||
abcd083a | 7707 | for_each_cpu(i, cpu_map) |
57d885fe | 7708 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7709 | synchronize_sched(); |
96f874e2 | 7710 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
7711 | } |
7712 | ||
1d3504fc HS |
7713 | /* handle null as "default" */ |
7714 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7715 | struct sched_domain_attr *new, int idx_new) | |
7716 | { | |
7717 | struct sched_domain_attr tmp; | |
7718 | ||
7719 | /* fast path */ | |
7720 | if (!new && !cur) | |
7721 | return 1; | |
7722 | ||
7723 | tmp = SD_ATTR_INIT; | |
7724 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7725 | new ? (new + idx_new) : &tmp, | |
7726 | sizeof(struct sched_domain_attr)); | |
7727 | } | |
7728 | ||
029190c5 PJ |
7729 | /* |
7730 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7731 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7732 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7733 | * It destroys each deleted domain and builds each new domain. | |
7734 | * | |
acc3f5d7 | 7735 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7736 | * The masks don't intersect (don't overlap.) We should setup one |
7737 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7738 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7739 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7740 | * it as it is. | |
7741 | * | |
acc3f5d7 RR |
7742 | * The passed in 'doms_new' should be allocated using |
7743 | * alloc_sched_domains. This routine takes ownership of it and will | |
7744 | * free_sched_domains it when done with it. If the caller failed the | |
7745 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7746 | * and partition_sched_domains() will fallback to the single partition | |
7747 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7748 | * |
96f874e2 | 7749 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7750 | * ndoms_new == 0 is a special case for destroying existing domains, |
7751 | * and it will not create the default domain. | |
dfb512ec | 7752 | * |
029190c5 PJ |
7753 | * Call with hotplug lock held |
7754 | */ | |
acc3f5d7 | 7755 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7756 | struct sched_domain_attr *dattr_new) |
029190c5 | 7757 | { |
dfb512ec | 7758 | int i, j, n; |
d65bd5ec | 7759 | int new_topology; |
029190c5 | 7760 | |
712555ee | 7761 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7762 | |
7378547f MM |
7763 | /* always unregister in case we don't destroy any domains */ |
7764 | unregister_sched_domain_sysctl(); | |
7765 | ||
d65bd5ec HC |
7766 | /* Let architecture update cpu core mappings. */ |
7767 | new_topology = arch_update_cpu_topology(); | |
7768 | ||
dfb512ec | 7769 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7770 | |
7771 | /* Destroy deleted domains */ | |
7772 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7773 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7774 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7775 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7776 | goto match1; |
7777 | } | |
7778 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7779 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7780 | match1: |
7781 | ; | |
7782 | } | |
7783 | ||
e761b772 MK |
7784 | if (doms_new == NULL) { |
7785 | ndoms_cur = 0; | |
acc3f5d7 | 7786 | doms_new = &fallback_doms; |
6ad4c188 | 7787 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7788 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7789 | } |
7790 | ||
029190c5 PJ |
7791 | /* Build new domains */ |
7792 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7793 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7794 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7795 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7796 | goto match2; |
7797 | } | |
7798 | /* no match - add a new doms_new */ | |
acc3f5d7 | 7799 | __build_sched_domains(doms_new[i], |
1d3504fc | 7800 | dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7801 | match2: |
7802 | ; | |
7803 | } | |
7804 | ||
7805 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7806 | if (doms_cur != &fallback_doms) |
7807 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7808 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7809 | doms_cur = doms_new; |
1d3504fc | 7810 | dattr_cur = dattr_new; |
029190c5 | 7811 | ndoms_cur = ndoms_new; |
7378547f MM |
7812 | |
7813 | register_sched_domain_sysctl(); | |
a1835615 | 7814 | |
712555ee | 7815 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7816 | } |
7817 | ||
5c45bf27 | 7818 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 7819 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 7820 | { |
95402b38 | 7821 | get_online_cpus(); |
dfb512ec MK |
7822 | |
7823 | /* Destroy domains first to force the rebuild */ | |
7824 | partition_sched_domains(0, NULL, NULL); | |
7825 | ||
e761b772 | 7826 | rebuild_sched_domains(); |
95402b38 | 7827 | put_online_cpus(); |
5c45bf27 SS |
7828 | } |
7829 | ||
7830 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7831 | { | |
afb8a9b7 | 7832 | unsigned int level = 0; |
5c45bf27 | 7833 | |
afb8a9b7 GS |
7834 | if (sscanf(buf, "%u", &level) != 1) |
7835 | return -EINVAL; | |
7836 | ||
7837 | /* | |
7838 | * level is always be positive so don't check for | |
7839 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7840 | * What happens on 0 or 1 byte write, | |
7841 | * need to check for count as well? | |
7842 | */ | |
7843 | ||
7844 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7845 | return -EINVAL; |
7846 | ||
7847 | if (smt) | |
afb8a9b7 | 7848 | sched_smt_power_savings = level; |
5c45bf27 | 7849 | else |
afb8a9b7 | 7850 | sched_mc_power_savings = level; |
5c45bf27 | 7851 | |
c70f22d2 | 7852 | arch_reinit_sched_domains(); |
5c45bf27 | 7853 | |
c70f22d2 | 7854 | return count; |
5c45bf27 SS |
7855 | } |
7856 | ||
5c45bf27 | 7857 | #ifdef CONFIG_SCHED_MC |
f718cd4a | 7858 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
c9be0a36 | 7859 | struct sysdev_class_attribute *attr, |
f718cd4a | 7860 | char *page) |
5c45bf27 SS |
7861 | { |
7862 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7863 | } | |
f718cd4a | 7864 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
c9be0a36 | 7865 | struct sysdev_class_attribute *attr, |
48f24c4d | 7866 | const char *buf, size_t count) |
5c45bf27 SS |
7867 | { |
7868 | return sched_power_savings_store(buf, count, 0); | |
7869 | } | |
f718cd4a AK |
7870 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7871 | sched_mc_power_savings_show, | |
7872 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7873 | #endif |
7874 | ||
7875 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a | 7876 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
c9be0a36 | 7877 | struct sysdev_class_attribute *attr, |
f718cd4a | 7878 | char *page) |
5c45bf27 SS |
7879 | { |
7880 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7881 | } | |
f718cd4a | 7882 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
c9be0a36 | 7883 | struct sysdev_class_attribute *attr, |
48f24c4d | 7884 | const char *buf, size_t count) |
5c45bf27 SS |
7885 | { |
7886 | return sched_power_savings_store(buf, count, 1); | |
7887 | } | |
f718cd4a AK |
7888 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7889 | sched_smt_power_savings_show, | |
6707de00 AB |
7890 | sched_smt_power_savings_store); |
7891 | #endif | |
7892 | ||
39aac648 | 7893 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
7894 | { |
7895 | int err = 0; | |
7896 | ||
7897 | #ifdef CONFIG_SCHED_SMT | |
7898 | if (smt_capable()) | |
7899 | err = sysfs_create_file(&cls->kset.kobj, | |
7900 | &attr_sched_smt_power_savings.attr); | |
7901 | #endif | |
7902 | #ifdef CONFIG_SCHED_MC | |
7903 | if (!err && mc_capable()) | |
7904 | err = sysfs_create_file(&cls->kset.kobj, | |
7905 | &attr_sched_mc_power_savings.attr); | |
7906 | #endif | |
7907 | return err; | |
7908 | } | |
6d6bc0ad | 7909 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7910 | |
1da177e4 | 7911 | /* |
3a101d05 TH |
7912 | * Update cpusets according to cpu_active mask. If cpusets are |
7913 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7914 | * around partition_sched_domains(). | |
1da177e4 | 7915 | */ |
0b2e918a TH |
7916 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
7917 | void *hcpu) | |
e761b772 | 7918 | { |
3a101d05 | 7919 | switch (action & ~CPU_TASKS_FROZEN) { |
e761b772 | 7920 | case CPU_ONLINE: |
6ad4c188 | 7921 | case CPU_DOWN_FAILED: |
3a101d05 | 7922 | cpuset_update_active_cpus(); |
e761b772 | 7923 | return NOTIFY_OK; |
3a101d05 TH |
7924 | default: |
7925 | return NOTIFY_DONE; | |
7926 | } | |
7927 | } | |
e761b772 | 7928 | |
0b2e918a TH |
7929 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
7930 | void *hcpu) | |
3a101d05 TH |
7931 | { |
7932 | switch (action & ~CPU_TASKS_FROZEN) { | |
7933 | case CPU_DOWN_PREPARE: | |
7934 | cpuset_update_active_cpus(); | |
7935 | return NOTIFY_OK; | |
e761b772 MK |
7936 | default: |
7937 | return NOTIFY_DONE; | |
7938 | } | |
7939 | } | |
e761b772 MK |
7940 | |
7941 | static int update_runtime(struct notifier_block *nfb, | |
7942 | unsigned long action, void *hcpu) | |
1da177e4 | 7943 | { |
7def2be1 PZ |
7944 | int cpu = (int)(long)hcpu; |
7945 | ||
1da177e4 | 7946 | switch (action) { |
1da177e4 | 7947 | case CPU_DOWN_PREPARE: |
8bb78442 | 7948 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7949 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7950 | return NOTIFY_OK; |
7951 | ||
1da177e4 | 7952 | case CPU_DOWN_FAILED: |
8bb78442 | 7953 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7954 | case CPU_ONLINE: |
8bb78442 | 7955 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7956 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7957 | return NOTIFY_OK; |
7958 | ||
1da177e4 LT |
7959 | default: |
7960 | return NOTIFY_DONE; | |
7961 | } | |
1da177e4 | 7962 | } |
1da177e4 LT |
7963 | |
7964 | void __init sched_init_smp(void) | |
7965 | { | |
dcc30a35 RR |
7966 | cpumask_var_t non_isolated_cpus; |
7967 | ||
7968 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7969 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7970 | |
434d53b0 MT |
7971 | #if defined(CONFIG_NUMA) |
7972 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7973 | GFP_KERNEL); | |
7974 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7975 | #endif | |
95402b38 | 7976 | get_online_cpus(); |
712555ee | 7977 | mutex_lock(&sched_domains_mutex); |
6ad4c188 | 7978 | arch_init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7979 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7980 | if (cpumask_empty(non_isolated_cpus)) | |
7981 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7982 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7983 | put_online_cpus(); |
e761b772 | 7984 | |
3a101d05 TH |
7985 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
7986 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
7987 | |
7988 | /* RT runtime code needs to handle some hotplug events */ | |
7989 | hotcpu_notifier(update_runtime, 0); | |
7990 | ||
b328ca18 | 7991 | init_hrtick(); |
5c1e1767 NP |
7992 | |
7993 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7994 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7995 | BUG(); |
19978ca6 | 7996 | sched_init_granularity(); |
dcc30a35 | 7997 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7998 | |
0e3900e6 | 7999 | init_sched_rt_class(); |
1da177e4 LT |
8000 | } |
8001 | #else | |
8002 | void __init sched_init_smp(void) | |
8003 | { | |
19978ca6 | 8004 | sched_init_granularity(); |
1da177e4 LT |
8005 | } |
8006 | #endif /* CONFIG_SMP */ | |
8007 | ||
cd1bb94b AB |
8008 | const_debug unsigned int sysctl_timer_migration = 1; |
8009 | ||
1da177e4 LT |
8010 | int in_sched_functions(unsigned long addr) |
8011 | { | |
1da177e4 LT |
8012 | return in_lock_functions(addr) || |
8013 | (addr >= (unsigned long)__sched_text_start | |
8014 | && addr < (unsigned long)__sched_text_end); | |
8015 | } | |
8016 | ||
a9957449 | 8017 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
8018 | { |
8019 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 8020 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
8021 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8022 | cfs_rq->rq = rq; | |
f07333bf | 8023 | /* allow initial update_cfs_load() to truncate */ |
6ea72f12 | 8024 | #ifdef CONFIG_SMP |
f07333bf | 8025 | cfs_rq->load_stamp = 1; |
6ea72f12 | 8026 | #endif |
dd41f596 | 8027 | #endif |
67e9fb2a | 8028 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
8029 | } |
8030 | ||
fa85ae24 PZ |
8031 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
8032 | { | |
8033 | struct rt_prio_array *array; | |
8034 | int i; | |
8035 | ||
8036 | array = &rt_rq->active; | |
8037 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
8038 | INIT_LIST_HEAD(array->queue + i); | |
8039 | __clear_bit(i, array->bitmap); | |
8040 | } | |
8041 | /* delimiter for bitsearch: */ | |
8042 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
8043 | ||
052f1dc7 | 8044 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 8045 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 8046 | #ifdef CONFIG_SMP |
e864c499 | 8047 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 8048 | #endif |
48d5e258 | 8049 | #endif |
fa85ae24 PZ |
8050 | #ifdef CONFIG_SMP |
8051 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 8052 | rt_rq->overloaded = 0; |
05fa785c | 8053 | plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
8054 | #endif |
8055 | ||
8056 | rt_rq->rt_time = 0; | |
8057 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 8058 | rt_rq->rt_runtime = 0; |
0986b11b | 8059 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
6f505b16 | 8060 | |
052f1dc7 | 8061 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 8062 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
8063 | rt_rq->rq = rq; |
8064 | #endif | |
fa85ae24 PZ |
8065 | } |
8066 | ||
6f505b16 | 8067 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac | 8068 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
3d4b47b4 | 8069 | struct sched_entity *se, int cpu, |
ec7dc8ac | 8070 | struct sched_entity *parent) |
6f505b16 | 8071 | { |
ec7dc8ac | 8072 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
8073 | tg->cfs_rq[cpu] = cfs_rq; |
8074 | init_cfs_rq(cfs_rq, rq); | |
8075 | cfs_rq->tg = tg; | |
6f505b16 PZ |
8076 | |
8077 | tg->se[cpu] = se; | |
07e06b01 | 8078 | /* se could be NULL for root_task_group */ |
354d60c2 DG |
8079 | if (!se) |
8080 | return; | |
8081 | ||
ec7dc8ac DG |
8082 | if (!parent) |
8083 | se->cfs_rq = &rq->cfs; | |
8084 | else | |
8085 | se->cfs_rq = parent->my_q; | |
8086 | ||
6f505b16 | 8087 | se->my_q = cfs_rq; |
9437178f | 8088 | update_load_set(&se->load, 0); |
ec7dc8ac | 8089 | se->parent = parent; |
6f505b16 | 8090 | } |
052f1dc7 | 8091 | #endif |
6f505b16 | 8092 | |
052f1dc7 | 8093 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac | 8094 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
3d4b47b4 | 8095 | struct sched_rt_entity *rt_se, int cpu, |
ec7dc8ac | 8096 | struct sched_rt_entity *parent) |
6f505b16 | 8097 | { |
ec7dc8ac DG |
8098 | struct rq *rq = cpu_rq(cpu); |
8099 | ||
6f505b16 PZ |
8100 | tg->rt_rq[cpu] = rt_rq; |
8101 | init_rt_rq(rt_rq, rq); | |
8102 | rt_rq->tg = tg; | |
ac086bc2 | 8103 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
8104 | |
8105 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
8106 | if (!rt_se) |
8107 | return; | |
8108 | ||
ec7dc8ac DG |
8109 | if (!parent) |
8110 | rt_se->rt_rq = &rq->rt; | |
8111 | else | |
8112 | rt_se->rt_rq = parent->my_q; | |
8113 | ||
6f505b16 | 8114 | rt_se->my_q = rt_rq; |
ec7dc8ac | 8115 | rt_se->parent = parent; |
6f505b16 PZ |
8116 | INIT_LIST_HEAD(&rt_se->run_list); |
8117 | } | |
8118 | #endif | |
8119 | ||
1da177e4 LT |
8120 | void __init sched_init(void) |
8121 | { | |
dd41f596 | 8122 | int i, j; |
434d53b0 MT |
8123 | unsigned long alloc_size = 0, ptr; |
8124 | ||
8125 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8126 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
8127 | #endif | |
8128 | #ifdef CONFIG_RT_GROUP_SCHED | |
8129 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 8130 | #endif |
df7c8e84 | 8131 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 8132 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 8133 | #endif |
434d53b0 | 8134 | if (alloc_size) { |
36b7b6d4 | 8135 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
8136 | |
8137 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 8138 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
8139 | ptr += nr_cpu_ids * sizeof(void **); |
8140 | ||
07e06b01 | 8141 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 8142 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 8143 | |
6d6bc0ad | 8144 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 8145 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 8146 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
8147 | ptr += nr_cpu_ids * sizeof(void **); |
8148 | ||
07e06b01 | 8149 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
8150 | ptr += nr_cpu_ids * sizeof(void **); |
8151 | ||
6d6bc0ad | 8152 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
8153 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8154 | for_each_possible_cpu(i) { | |
8155 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
8156 | ptr += cpumask_size(); | |
8157 | } | |
8158 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 8159 | } |
dd41f596 | 8160 | |
57d885fe GH |
8161 | #ifdef CONFIG_SMP |
8162 | init_defrootdomain(); | |
8163 | #endif | |
8164 | ||
d0b27fa7 PZ |
8165 | init_rt_bandwidth(&def_rt_bandwidth, |
8166 | global_rt_period(), global_rt_runtime()); | |
8167 | ||
8168 | #ifdef CONFIG_RT_GROUP_SCHED | |
07e06b01 | 8169 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 8170 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 8171 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 8172 | |
7c941438 | 8173 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
8174 | list_add(&root_task_group.list, &task_groups); |
8175 | INIT_LIST_HEAD(&root_task_group.children); | |
5091faa4 | 8176 | autogroup_init(&init_task); |
7c941438 | 8177 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 8178 | |
0a945022 | 8179 | for_each_possible_cpu(i) { |
70b97a7f | 8180 | struct rq *rq; |
1da177e4 LT |
8181 | |
8182 | rq = cpu_rq(i); | |
05fa785c | 8183 | raw_spin_lock_init(&rq->lock); |
7897986b | 8184 | rq->nr_running = 0; |
dce48a84 TG |
8185 | rq->calc_load_active = 0; |
8186 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 8187 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 8188 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 8189 | #ifdef CONFIG_FAIR_GROUP_SCHED |
07e06b01 | 8190 | root_task_group.shares = root_task_group_load; |
6f505b16 | 8191 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 8192 | /* |
07e06b01 | 8193 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
8194 | * |
8195 | * In case of task-groups formed thr' the cgroup filesystem, it | |
8196 | * gets 100% of the cpu resources in the system. This overall | |
8197 | * system cpu resource is divided among the tasks of | |
07e06b01 | 8198 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
8199 | * based on each entity's (task or task-group's) weight |
8200 | * (se->load.weight). | |
8201 | * | |
07e06b01 | 8202 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
8203 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
8204 | * then A0's share of the cpu resource is: | |
8205 | * | |
0d905bca | 8206 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 8207 | * |
07e06b01 YZ |
8208 | * We achieve this by letting root_task_group's tasks sit |
8209 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 8210 | */ |
07e06b01 | 8211 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
8212 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8213 | ||
8214 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8215 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8216 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
07e06b01 | 8217 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 8218 | #endif |
1da177e4 | 8219 | |
dd41f596 IM |
8220 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
8221 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
8222 | |
8223 | rq->last_load_update_tick = jiffies; | |
8224 | ||
1da177e4 | 8225 | #ifdef CONFIG_SMP |
41c7ce9a | 8226 | rq->sd = NULL; |
57d885fe | 8227 | rq->rd = NULL; |
e51fd5e2 | 8228 | rq->cpu_power = SCHED_LOAD_SCALE; |
3f029d3c | 8229 | rq->post_schedule = 0; |
1da177e4 | 8230 | rq->active_balance = 0; |
dd41f596 | 8231 | rq->next_balance = jiffies; |
1da177e4 | 8232 | rq->push_cpu = 0; |
0a2966b4 | 8233 | rq->cpu = i; |
1f11eb6a | 8234 | rq->online = 0; |
eae0c9df MG |
8235 | rq->idle_stamp = 0; |
8236 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
dc938520 | 8237 | rq_attach_root(rq, &def_root_domain); |
83cd4fe2 VP |
8238 | #ifdef CONFIG_NO_HZ |
8239 | rq->nohz_balance_kick = 0; | |
8240 | init_sched_softirq_csd(&per_cpu(remote_sched_softirq_cb, i)); | |
8241 | #endif | |
1da177e4 | 8242 | #endif |
8f4d37ec | 8243 | init_rq_hrtick(rq); |
1da177e4 | 8244 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8245 | } |
8246 | ||
2dd73a4f | 8247 | set_load_weight(&init_task); |
b50f60ce | 8248 | |
e107be36 AK |
8249 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8250 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8251 | #endif | |
8252 | ||
c9819f45 | 8253 | #ifdef CONFIG_SMP |
962cf36c | 8254 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
8255 | #endif |
8256 | ||
b50f60ce | 8257 | #ifdef CONFIG_RT_MUTEXES |
1d615482 | 8258 | plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock); |
b50f60ce HC |
8259 | #endif |
8260 | ||
1da177e4 LT |
8261 | /* |
8262 | * The boot idle thread does lazy MMU switching as well: | |
8263 | */ | |
8264 | atomic_inc(&init_mm.mm_count); | |
8265 | enter_lazy_tlb(&init_mm, current); | |
8266 | ||
8267 | /* | |
8268 | * Make us the idle thread. Technically, schedule() should not be | |
8269 | * called from this thread, however somewhere below it might be, | |
8270 | * but because we are the idle thread, we just pick up running again | |
8271 | * when this runqueue becomes "idle". | |
8272 | */ | |
8273 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
8274 | |
8275 | calc_load_update = jiffies + LOAD_FREQ; | |
8276 | ||
dd41f596 IM |
8277 | /* |
8278 | * During early bootup we pretend to be a normal task: | |
8279 | */ | |
8280 | current->sched_class = &fair_sched_class; | |
6892b75e | 8281 | |
6a7b3dc3 | 8282 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 8283 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 8284 | #ifdef CONFIG_SMP |
7d1e6a9b | 8285 | #ifdef CONFIG_NO_HZ |
83cd4fe2 VP |
8286 | zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT); |
8287 | alloc_cpumask_var(&nohz.grp_idle_mask, GFP_NOWAIT); | |
8288 | atomic_set(&nohz.load_balancer, nr_cpu_ids); | |
8289 | atomic_set(&nohz.first_pick_cpu, nr_cpu_ids); | |
8290 | atomic_set(&nohz.second_pick_cpu, nr_cpu_ids); | |
7d1e6a9b | 8291 | #endif |
bdddd296 RR |
8292 | /* May be allocated at isolcpus cmdline parse time */ |
8293 | if (cpu_isolated_map == NULL) | |
8294 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 8295 | #endif /* SMP */ |
6a7b3dc3 | 8296 | |
6892b75e | 8297 | scheduler_running = 1; |
1da177e4 LT |
8298 | } |
8299 | ||
8300 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
8301 | static inline int preempt_count_equals(int preempt_offset) |
8302 | { | |
234da7bc | 8303 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 8304 | |
4ba8216c | 8305 | return (nested == preempt_offset); |
e4aafea2 FW |
8306 | } |
8307 | ||
d894837f | 8308 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 8309 | { |
48f24c4d | 8310 | #ifdef in_atomic |
1da177e4 LT |
8311 | static unsigned long prev_jiffy; /* ratelimiting */ |
8312 | ||
e4aafea2 FW |
8313 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
8314 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
8315 | return; |
8316 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8317 | return; | |
8318 | prev_jiffy = jiffies; | |
8319 | ||
3df0fc5b PZ |
8320 | printk(KERN_ERR |
8321 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
8322 | file, line); | |
8323 | printk(KERN_ERR | |
8324 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
8325 | in_atomic(), irqs_disabled(), | |
8326 | current->pid, current->comm); | |
aef745fc IM |
8327 | |
8328 | debug_show_held_locks(current); | |
8329 | if (irqs_disabled()) | |
8330 | print_irqtrace_events(current); | |
8331 | dump_stack(); | |
1da177e4 LT |
8332 | #endif |
8333 | } | |
8334 | EXPORT_SYMBOL(__might_sleep); | |
8335 | #endif | |
8336 | ||
8337 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8338 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8339 | { | |
da7a735e PZ |
8340 | const struct sched_class *prev_class = p->sched_class; |
8341 | int old_prio = p->prio; | |
3a5e4dc1 | 8342 | int on_rq; |
3e51f33f | 8343 | |
fd2f4419 | 8344 | on_rq = p->on_rq; |
3a5e4dc1 AK |
8345 | if (on_rq) |
8346 | deactivate_task(rq, p, 0); | |
8347 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8348 | if (on_rq) { | |
8349 | activate_task(rq, p, 0); | |
8350 | resched_task(rq->curr); | |
8351 | } | |
da7a735e PZ |
8352 | |
8353 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
8354 | } |
8355 | ||
1da177e4 LT |
8356 | void normalize_rt_tasks(void) |
8357 | { | |
a0f98a1c | 8358 | struct task_struct *g, *p; |
1da177e4 | 8359 | unsigned long flags; |
70b97a7f | 8360 | struct rq *rq; |
1da177e4 | 8361 | |
4cf5d77a | 8362 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8363 | do_each_thread(g, p) { |
178be793 IM |
8364 | /* |
8365 | * Only normalize user tasks: | |
8366 | */ | |
8367 | if (!p->mm) | |
8368 | continue; | |
8369 | ||
6cfb0d5d | 8370 | p->se.exec_start = 0; |
6cfb0d5d | 8371 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
8372 | p->se.statistics.wait_start = 0; |
8373 | p->se.statistics.sleep_start = 0; | |
8374 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 8375 | #endif |
dd41f596 IM |
8376 | |
8377 | if (!rt_task(p)) { | |
8378 | /* | |
8379 | * Renice negative nice level userspace | |
8380 | * tasks back to 0: | |
8381 | */ | |
8382 | if (TASK_NICE(p) < 0 && p->mm) | |
8383 | set_user_nice(p, 0); | |
1da177e4 | 8384 | continue; |
dd41f596 | 8385 | } |
1da177e4 | 8386 | |
1d615482 | 8387 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 8388 | rq = __task_rq_lock(p); |
1da177e4 | 8389 | |
178be793 | 8390 | normalize_task(rq, p); |
3a5e4dc1 | 8391 | |
b29739f9 | 8392 | __task_rq_unlock(rq); |
1d615482 | 8393 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8394 | } while_each_thread(g, p); |
8395 | ||
4cf5d77a | 8396 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8397 | } |
8398 | ||
8399 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 8400 | |
67fc4e0c | 8401 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 8402 | /* |
67fc4e0c | 8403 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
8404 | * |
8405 | * They can only be called when the whole system has been | |
8406 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8407 | * activity can take place. Using them for anything else would | |
8408 | * be a serious bug, and as a result, they aren't even visible | |
8409 | * under any other configuration. | |
8410 | */ | |
8411 | ||
8412 | /** | |
8413 | * curr_task - return the current task for a given cpu. | |
8414 | * @cpu: the processor in question. | |
8415 | * | |
8416 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8417 | */ | |
36c8b586 | 8418 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8419 | { |
8420 | return cpu_curr(cpu); | |
8421 | } | |
8422 | ||
67fc4e0c JW |
8423 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
8424 | ||
8425 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
8426 | /** |
8427 | * set_curr_task - set the current task for a given cpu. | |
8428 | * @cpu: the processor in question. | |
8429 | * @p: the task pointer to set. | |
8430 | * | |
8431 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8432 | * are serviced on a separate stack. It allows the architecture to switch the |
8433 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8434 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8435 | * and caller must save the original value of the current task (see | |
8436 | * curr_task() above) and restore that value before reenabling interrupts and | |
8437 | * re-starting the system. | |
8438 | * | |
8439 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8440 | */ | |
36c8b586 | 8441 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8442 | { |
8443 | cpu_curr(cpu) = p; | |
8444 | } | |
8445 | ||
8446 | #endif | |
29f59db3 | 8447 | |
bccbe08a PZ |
8448 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8449 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8450 | { |
8451 | int i; | |
8452 | ||
8453 | for_each_possible_cpu(i) { | |
8454 | if (tg->cfs_rq) | |
8455 | kfree(tg->cfs_rq[i]); | |
8456 | if (tg->se) | |
8457 | kfree(tg->se[i]); | |
6f505b16 PZ |
8458 | } |
8459 | ||
8460 | kfree(tg->cfs_rq); | |
8461 | kfree(tg->se); | |
6f505b16 PZ |
8462 | } |
8463 | ||
ec7dc8ac DG |
8464 | static |
8465 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8466 | { |
29f59db3 | 8467 | struct cfs_rq *cfs_rq; |
eab17229 | 8468 | struct sched_entity *se; |
29f59db3 SV |
8469 | int i; |
8470 | ||
434d53b0 | 8471 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8472 | if (!tg->cfs_rq) |
8473 | goto err; | |
434d53b0 | 8474 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8475 | if (!tg->se) |
8476 | goto err; | |
052f1dc7 PZ |
8477 | |
8478 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8479 | |
8480 | for_each_possible_cpu(i) { | |
eab17229 LZ |
8481 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
8482 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
8483 | if (!cfs_rq) |
8484 | goto err; | |
8485 | ||
eab17229 LZ |
8486 | se = kzalloc_node(sizeof(struct sched_entity), |
8487 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 8488 | if (!se) |
dfc12eb2 | 8489 | goto err_free_rq; |
29f59db3 | 8490 | |
3d4b47b4 | 8491 | init_tg_cfs_entry(tg, cfs_rq, se, i, parent->se[i]); |
bccbe08a PZ |
8492 | } |
8493 | ||
8494 | return 1; | |
8495 | ||
49246274 | 8496 | err_free_rq: |
dfc12eb2 | 8497 | kfree(cfs_rq); |
49246274 | 8498 | err: |
bccbe08a PZ |
8499 | return 0; |
8500 | } | |
8501 | ||
bccbe08a PZ |
8502 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8503 | { | |
3d4b47b4 PZ |
8504 | struct rq *rq = cpu_rq(cpu); |
8505 | unsigned long flags; | |
3d4b47b4 PZ |
8506 | |
8507 | /* | |
8508 | * Only empty task groups can be destroyed; so we can speculatively | |
8509 | * check on_list without danger of it being re-added. | |
8510 | */ | |
8511 | if (!tg->cfs_rq[cpu]->on_list) | |
8512 | return; | |
8513 | ||
8514 | raw_spin_lock_irqsave(&rq->lock, flags); | |
822bc180 | 8515 | list_del_leaf_cfs_rq(tg->cfs_rq[cpu]); |
3d4b47b4 | 8516 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
bccbe08a | 8517 | } |
6d6bc0ad | 8518 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8519 | static inline void free_fair_sched_group(struct task_group *tg) |
8520 | { | |
8521 | } | |
8522 | ||
ec7dc8ac DG |
8523 | static inline |
8524 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8525 | { |
8526 | return 1; | |
8527 | } | |
8528 | ||
bccbe08a PZ |
8529 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8530 | { | |
8531 | } | |
6d6bc0ad | 8532 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8533 | |
8534 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8535 | static void free_rt_sched_group(struct task_group *tg) |
8536 | { | |
8537 | int i; | |
8538 | ||
d0b27fa7 PZ |
8539 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8540 | ||
bccbe08a PZ |
8541 | for_each_possible_cpu(i) { |
8542 | if (tg->rt_rq) | |
8543 | kfree(tg->rt_rq[i]); | |
8544 | if (tg->rt_se) | |
8545 | kfree(tg->rt_se[i]); | |
8546 | } | |
8547 | ||
8548 | kfree(tg->rt_rq); | |
8549 | kfree(tg->rt_se); | |
8550 | } | |
8551 | ||
ec7dc8ac DG |
8552 | static |
8553 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8554 | { |
8555 | struct rt_rq *rt_rq; | |
eab17229 | 8556 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8557 | struct rq *rq; |
8558 | int i; | |
8559 | ||
434d53b0 | 8560 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8561 | if (!tg->rt_rq) |
8562 | goto err; | |
434d53b0 | 8563 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8564 | if (!tg->rt_se) |
8565 | goto err; | |
8566 | ||
d0b27fa7 PZ |
8567 | init_rt_bandwidth(&tg->rt_bandwidth, |
8568 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8569 | |
8570 | for_each_possible_cpu(i) { | |
8571 | rq = cpu_rq(i); | |
8572 | ||
eab17229 LZ |
8573 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8574 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8575 | if (!rt_rq) |
8576 | goto err; | |
29f59db3 | 8577 | |
eab17229 LZ |
8578 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8579 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 8580 | if (!rt_se) |
dfc12eb2 | 8581 | goto err_free_rq; |
29f59db3 | 8582 | |
3d4b47b4 | 8583 | init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]); |
29f59db3 SV |
8584 | } |
8585 | ||
bccbe08a PZ |
8586 | return 1; |
8587 | ||
49246274 | 8588 | err_free_rq: |
dfc12eb2 | 8589 | kfree(rt_rq); |
49246274 | 8590 | err: |
bccbe08a PZ |
8591 | return 0; |
8592 | } | |
6d6bc0ad | 8593 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8594 | static inline void free_rt_sched_group(struct task_group *tg) |
8595 | { | |
8596 | } | |
8597 | ||
ec7dc8ac DG |
8598 | static inline |
8599 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8600 | { |
8601 | return 1; | |
8602 | } | |
6d6bc0ad | 8603 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8604 | |
7c941438 | 8605 | #ifdef CONFIG_CGROUP_SCHED |
bccbe08a PZ |
8606 | static void free_sched_group(struct task_group *tg) |
8607 | { | |
8608 | free_fair_sched_group(tg); | |
8609 | free_rt_sched_group(tg); | |
e9aa1dd1 | 8610 | autogroup_free(tg); |
bccbe08a PZ |
8611 | kfree(tg); |
8612 | } | |
8613 | ||
8614 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8615 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8616 | { |
8617 | struct task_group *tg; | |
8618 | unsigned long flags; | |
bccbe08a PZ |
8619 | |
8620 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8621 | if (!tg) | |
8622 | return ERR_PTR(-ENOMEM); | |
8623 | ||
ec7dc8ac | 8624 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8625 | goto err; |
8626 | ||
ec7dc8ac | 8627 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8628 | goto err; |
8629 | ||
8ed36996 | 8630 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 8631 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8632 | |
8633 | WARN_ON(!parent); /* root should already exist */ | |
8634 | ||
8635 | tg->parent = parent; | |
f473aa5e | 8636 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8637 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8638 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8639 | |
9b5b7751 | 8640 | return tg; |
29f59db3 SV |
8641 | |
8642 | err: | |
6f505b16 | 8643 | free_sched_group(tg); |
29f59db3 SV |
8644 | return ERR_PTR(-ENOMEM); |
8645 | } | |
8646 | ||
9b5b7751 | 8647 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8648 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8649 | { |
29f59db3 | 8650 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8651 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8652 | } |
8653 | ||
9b5b7751 | 8654 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8655 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8656 | { |
8ed36996 | 8657 | unsigned long flags; |
9b5b7751 | 8658 | int i; |
29f59db3 | 8659 | |
3d4b47b4 PZ |
8660 | /* end participation in shares distribution */ |
8661 | for_each_possible_cpu(i) | |
bccbe08a | 8662 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
8663 | |
8664 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 8665 | list_del_rcu(&tg->list); |
f473aa5e | 8666 | list_del_rcu(&tg->siblings); |
8ed36996 | 8667 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8668 | |
9b5b7751 | 8669 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8670 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8671 | } |
8672 | ||
9b5b7751 | 8673 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8674 | * The caller of this function should have put the task in its new group |
8675 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8676 | * reflect its new group. | |
9b5b7751 SV |
8677 | */ |
8678 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8679 | { |
8680 | int on_rq, running; | |
8681 | unsigned long flags; | |
8682 | struct rq *rq; | |
8683 | ||
8684 | rq = task_rq_lock(tsk, &flags); | |
8685 | ||
051a1d1a | 8686 | running = task_current(rq, tsk); |
fd2f4419 | 8687 | on_rq = tsk->on_rq; |
29f59db3 | 8688 | |
0e1f3483 | 8689 | if (on_rq) |
29f59db3 | 8690 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8691 | if (unlikely(running)) |
8692 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8693 | |
810b3817 | 8694 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
8695 | if (tsk->sched_class->task_move_group) |
8696 | tsk->sched_class->task_move_group(tsk, on_rq); | |
8697 | else | |
810b3817 | 8698 | #endif |
b2b5ce02 | 8699 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 8700 | |
0e1f3483 HS |
8701 | if (unlikely(running)) |
8702 | tsk->sched_class->set_curr_task(rq); | |
8703 | if (on_rq) | |
371fd7e7 | 8704 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8705 | |
29f59db3 SV |
8706 | task_rq_unlock(rq, &flags); |
8707 | } | |
7c941438 | 8708 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8709 | |
052f1dc7 | 8710 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8ed36996 PZ |
8711 | static DEFINE_MUTEX(shares_mutex); |
8712 | ||
4cf86d77 | 8713 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8714 | { |
8715 | int i; | |
8ed36996 | 8716 | unsigned long flags; |
c61935fd | 8717 | |
ec7dc8ac DG |
8718 | /* |
8719 | * We can't change the weight of the root cgroup. | |
8720 | */ | |
8721 | if (!tg->se[0]) | |
8722 | return -EINVAL; | |
8723 | ||
18d95a28 PZ |
8724 | if (shares < MIN_SHARES) |
8725 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8726 | else if (shares > MAX_SHARES) |
8727 | shares = MAX_SHARES; | |
62fb1851 | 8728 | |
8ed36996 | 8729 | mutex_lock(&shares_mutex); |
9b5b7751 | 8730 | if (tg->shares == shares) |
5cb350ba | 8731 | goto done; |
29f59db3 | 8732 | |
9b5b7751 | 8733 | tg->shares = shares; |
c09595f6 | 8734 | for_each_possible_cpu(i) { |
9437178f PT |
8735 | struct rq *rq = cpu_rq(i); |
8736 | struct sched_entity *se; | |
8737 | ||
8738 | se = tg->se[i]; | |
8739 | /* Propagate contribution to hierarchy */ | |
8740 | raw_spin_lock_irqsave(&rq->lock, flags); | |
8741 | for_each_sched_entity(se) | |
6d5ab293 | 8742 | update_cfs_shares(group_cfs_rq(se)); |
9437178f | 8743 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c09595f6 | 8744 | } |
29f59db3 | 8745 | |
5cb350ba | 8746 | done: |
8ed36996 | 8747 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8748 | return 0; |
29f59db3 SV |
8749 | } |
8750 | ||
5cb350ba DG |
8751 | unsigned long sched_group_shares(struct task_group *tg) |
8752 | { | |
8753 | return tg->shares; | |
8754 | } | |
052f1dc7 | 8755 | #endif |
5cb350ba | 8756 | |
052f1dc7 | 8757 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8758 | /* |
9f0c1e56 | 8759 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8760 | */ |
9f0c1e56 PZ |
8761 | static DEFINE_MUTEX(rt_constraints_mutex); |
8762 | ||
8763 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8764 | { | |
8765 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8766 | return 1ULL << 20; |
9f0c1e56 | 8767 | |
9a7e0b18 | 8768 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8769 | } |
8770 | ||
9a7e0b18 PZ |
8771 | /* Must be called with tasklist_lock held */ |
8772 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8773 | { |
9a7e0b18 | 8774 | struct task_struct *g, *p; |
b40b2e8e | 8775 | |
9a7e0b18 PZ |
8776 | do_each_thread(g, p) { |
8777 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8778 | return 1; | |
8779 | } while_each_thread(g, p); | |
b40b2e8e | 8780 | |
9a7e0b18 PZ |
8781 | return 0; |
8782 | } | |
b40b2e8e | 8783 | |
9a7e0b18 PZ |
8784 | struct rt_schedulable_data { |
8785 | struct task_group *tg; | |
8786 | u64 rt_period; | |
8787 | u64 rt_runtime; | |
8788 | }; | |
b40b2e8e | 8789 | |
9a7e0b18 PZ |
8790 | static int tg_schedulable(struct task_group *tg, void *data) |
8791 | { | |
8792 | struct rt_schedulable_data *d = data; | |
8793 | struct task_group *child; | |
8794 | unsigned long total, sum = 0; | |
8795 | u64 period, runtime; | |
b40b2e8e | 8796 | |
9a7e0b18 PZ |
8797 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8798 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8799 | |
9a7e0b18 PZ |
8800 | if (tg == d->tg) { |
8801 | period = d->rt_period; | |
8802 | runtime = d->rt_runtime; | |
b40b2e8e | 8803 | } |
b40b2e8e | 8804 | |
4653f803 PZ |
8805 | /* |
8806 | * Cannot have more runtime than the period. | |
8807 | */ | |
8808 | if (runtime > period && runtime != RUNTIME_INF) | |
8809 | return -EINVAL; | |
6f505b16 | 8810 | |
4653f803 PZ |
8811 | /* |
8812 | * Ensure we don't starve existing RT tasks. | |
8813 | */ | |
9a7e0b18 PZ |
8814 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8815 | return -EBUSY; | |
6f505b16 | 8816 | |
9a7e0b18 | 8817 | total = to_ratio(period, runtime); |
6f505b16 | 8818 | |
4653f803 PZ |
8819 | /* |
8820 | * Nobody can have more than the global setting allows. | |
8821 | */ | |
8822 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8823 | return -EINVAL; | |
6f505b16 | 8824 | |
4653f803 PZ |
8825 | /* |
8826 | * The sum of our children's runtime should not exceed our own. | |
8827 | */ | |
9a7e0b18 PZ |
8828 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8829 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8830 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8831 | |
9a7e0b18 PZ |
8832 | if (child == d->tg) { |
8833 | period = d->rt_period; | |
8834 | runtime = d->rt_runtime; | |
8835 | } | |
6f505b16 | 8836 | |
9a7e0b18 | 8837 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8838 | } |
6f505b16 | 8839 | |
9a7e0b18 PZ |
8840 | if (sum > total) |
8841 | return -EINVAL; | |
8842 | ||
8843 | return 0; | |
6f505b16 PZ |
8844 | } |
8845 | ||
9a7e0b18 | 8846 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8847 | { |
9a7e0b18 PZ |
8848 | struct rt_schedulable_data data = { |
8849 | .tg = tg, | |
8850 | .rt_period = period, | |
8851 | .rt_runtime = runtime, | |
8852 | }; | |
8853 | ||
8854 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8855 | } |
8856 | ||
d0b27fa7 PZ |
8857 | static int tg_set_bandwidth(struct task_group *tg, |
8858 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8859 | { |
ac086bc2 | 8860 | int i, err = 0; |
9f0c1e56 | 8861 | |
9f0c1e56 | 8862 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8863 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8864 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8865 | if (err) | |
9f0c1e56 | 8866 | goto unlock; |
ac086bc2 | 8867 | |
0986b11b | 8868 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8869 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8870 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8871 | |
8872 | for_each_possible_cpu(i) { | |
8873 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8874 | ||
0986b11b | 8875 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8876 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8877 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8878 | } |
0986b11b | 8879 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 8880 | unlock: |
521f1a24 | 8881 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8882 | mutex_unlock(&rt_constraints_mutex); |
8883 | ||
8884 | return err; | |
6f505b16 PZ |
8885 | } |
8886 | ||
d0b27fa7 PZ |
8887 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8888 | { | |
8889 | u64 rt_runtime, rt_period; | |
8890 | ||
8891 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8892 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8893 | if (rt_runtime_us < 0) | |
8894 | rt_runtime = RUNTIME_INF; | |
8895 | ||
8896 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8897 | } | |
8898 | ||
9f0c1e56 PZ |
8899 | long sched_group_rt_runtime(struct task_group *tg) |
8900 | { | |
8901 | u64 rt_runtime_us; | |
8902 | ||
d0b27fa7 | 8903 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8904 | return -1; |
8905 | ||
d0b27fa7 | 8906 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8907 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8908 | return rt_runtime_us; | |
8909 | } | |
d0b27fa7 PZ |
8910 | |
8911 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8912 | { | |
8913 | u64 rt_runtime, rt_period; | |
8914 | ||
8915 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8916 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8917 | ||
619b0488 R |
8918 | if (rt_period == 0) |
8919 | return -EINVAL; | |
8920 | ||
d0b27fa7 PZ |
8921 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8922 | } | |
8923 | ||
8924 | long sched_group_rt_period(struct task_group *tg) | |
8925 | { | |
8926 | u64 rt_period_us; | |
8927 | ||
8928 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8929 | do_div(rt_period_us, NSEC_PER_USEC); | |
8930 | return rt_period_us; | |
8931 | } | |
8932 | ||
8933 | static int sched_rt_global_constraints(void) | |
8934 | { | |
4653f803 | 8935 | u64 runtime, period; |
d0b27fa7 PZ |
8936 | int ret = 0; |
8937 | ||
ec5d4989 HS |
8938 | if (sysctl_sched_rt_period <= 0) |
8939 | return -EINVAL; | |
8940 | ||
4653f803 PZ |
8941 | runtime = global_rt_runtime(); |
8942 | period = global_rt_period(); | |
8943 | ||
8944 | /* | |
8945 | * Sanity check on the sysctl variables. | |
8946 | */ | |
8947 | if (runtime > period && runtime != RUNTIME_INF) | |
8948 | return -EINVAL; | |
10b612f4 | 8949 | |
d0b27fa7 | 8950 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 8951 | read_lock(&tasklist_lock); |
4653f803 | 8952 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 8953 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
8954 | mutex_unlock(&rt_constraints_mutex); |
8955 | ||
8956 | return ret; | |
8957 | } | |
54e99124 DG |
8958 | |
8959 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
8960 | { | |
8961 | /* Don't accept realtime tasks when there is no way for them to run */ | |
8962 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
8963 | return 0; | |
8964 | ||
8965 | return 1; | |
8966 | } | |
8967 | ||
6d6bc0ad | 8968 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8969 | static int sched_rt_global_constraints(void) |
8970 | { | |
ac086bc2 PZ |
8971 | unsigned long flags; |
8972 | int i; | |
8973 | ||
ec5d4989 HS |
8974 | if (sysctl_sched_rt_period <= 0) |
8975 | return -EINVAL; | |
8976 | ||
60aa605d PZ |
8977 | /* |
8978 | * There's always some RT tasks in the root group | |
8979 | * -- migration, kstopmachine etc.. | |
8980 | */ | |
8981 | if (sysctl_sched_rt_runtime == 0) | |
8982 | return -EBUSY; | |
8983 | ||
0986b11b | 8984 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
8985 | for_each_possible_cpu(i) { |
8986 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8987 | ||
0986b11b | 8988 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8989 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 8990 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8991 | } |
0986b11b | 8992 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 8993 | |
d0b27fa7 PZ |
8994 | return 0; |
8995 | } | |
6d6bc0ad | 8996 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8997 | |
8998 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 8999 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
9000 | loff_t *ppos) |
9001 | { | |
9002 | int ret; | |
9003 | int old_period, old_runtime; | |
9004 | static DEFINE_MUTEX(mutex); | |
9005 | ||
9006 | mutex_lock(&mutex); | |
9007 | old_period = sysctl_sched_rt_period; | |
9008 | old_runtime = sysctl_sched_rt_runtime; | |
9009 | ||
8d65af78 | 9010 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
9011 | |
9012 | if (!ret && write) { | |
9013 | ret = sched_rt_global_constraints(); | |
9014 | if (ret) { | |
9015 | sysctl_sched_rt_period = old_period; | |
9016 | sysctl_sched_rt_runtime = old_runtime; | |
9017 | } else { | |
9018 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
9019 | def_rt_bandwidth.rt_period = | |
9020 | ns_to_ktime(global_rt_period()); | |
9021 | } | |
9022 | } | |
9023 | mutex_unlock(&mutex); | |
9024 | ||
9025 | return ret; | |
9026 | } | |
68318b8e | 9027 | |
052f1dc7 | 9028 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
9029 | |
9030 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 9031 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 9032 | { |
2b01dfe3 PM |
9033 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
9034 | struct task_group, css); | |
68318b8e SV |
9035 | } |
9036 | ||
9037 | static struct cgroup_subsys_state * | |
2b01dfe3 | 9038 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 9039 | { |
ec7dc8ac | 9040 | struct task_group *tg, *parent; |
68318b8e | 9041 | |
2b01dfe3 | 9042 | if (!cgrp->parent) { |
68318b8e | 9043 | /* This is early initialization for the top cgroup */ |
07e06b01 | 9044 | return &root_task_group.css; |
68318b8e SV |
9045 | } |
9046 | ||
ec7dc8ac DG |
9047 | parent = cgroup_tg(cgrp->parent); |
9048 | tg = sched_create_group(parent); | |
68318b8e SV |
9049 | if (IS_ERR(tg)) |
9050 | return ERR_PTR(-ENOMEM); | |
9051 | ||
68318b8e SV |
9052 | return &tg->css; |
9053 | } | |
9054 | ||
41a2d6cf IM |
9055 | static void |
9056 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 9057 | { |
2b01dfe3 | 9058 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9059 | |
9060 | sched_destroy_group(tg); | |
9061 | } | |
9062 | ||
41a2d6cf | 9063 | static int |
be367d09 | 9064 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 9065 | { |
b68aa230 | 9066 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 9067 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
9068 | return -EINVAL; |
9069 | #else | |
68318b8e SV |
9070 | /* We don't support RT-tasks being in separate groups */ |
9071 | if (tsk->sched_class != &fair_sched_class) | |
9072 | return -EINVAL; | |
b68aa230 | 9073 | #endif |
be367d09 BB |
9074 | return 0; |
9075 | } | |
68318b8e | 9076 | |
be367d09 BB |
9077 | static int |
9078 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
9079 | struct task_struct *tsk, bool threadgroup) | |
9080 | { | |
9081 | int retval = cpu_cgroup_can_attach_task(cgrp, tsk); | |
9082 | if (retval) | |
9083 | return retval; | |
9084 | if (threadgroup) { | |
9085 | struct task_struct *c; | |
9086 | rcu_read_lock(); | |
9087 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
9088 | retval = cpu_cgroup_can_attach_task(cgrp, c); | |
9089 | if (retval) { | |
9090 | rcu_read_unlock(); | |
9091 | return retval; | |
9092 | } | |
9093 | } | |
9094 | rcu_read_unlock(); | |
9095 | } | |
68318b8e SV |
9096 | return 0; |
9097 | } | |
9098 | ||
9099 | static void | |
2b01dfe3 | 9100 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
be367d09 BB |
9101 | struct cgroup *old_cont, struct task_struct *tsk, |
9102 | bool threadgroup) | |
68318b8e SV |
9103 | { |
9104 | sched_move_task(tsk); | |
be367d09 BB |
9105 | if (threadgroup) { |
9106 | struct task_struct *c; | |
9107 | rcu_read_lock(); | |
9108 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
9109 | sched_move_task(c); | |
9110 | } | |
9111 | rcu_read_unlock(); | |
9112 | } | |
68318b8e SV |
9113 | } |
9114 | ||
068c5cc5 | 9115 | static void |
d41d5a01 PZ |
9116 | cpu_cgroup_exit(struct cgroup_subsys *ss, struct cgroup *cgrp, |
9117 | struct cgroup *old_cgrp, struct task_struct *task) | |
068c5cc5 PZ |
9118 | { |
9119 | /* | |
9120 | * cgroup_exit() is called in the copy_process() failure path. | |
9121 | * Ignore this case since the task hasn't ran yet, this avoids | |
9122 | * trying to poke a half freed task state from generic code. | |
9123 | */ | |
9124 | if (!(task->flags & PF_EXITING)) | |
9125 | return; | |
9126 | ||
9127 | sched_move_task(task); | |
9128 | } | |
9129 | ||
052f1dc7 | 9130 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 9131 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 9132 | u64 shareval) |
68318b8e | 9133 | { |
2b01dfe3 | 9134 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
9135 | } |
9136 | ||
f4c753b7 | 9137 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 9138 | { |
2b01dfe3 | 9139 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9140 | |
9141 | return (u64) tg->shares; | |
9142 | } | |
6d6bc0ad | 9143 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 9144 | |
052f1dc7 | 9145 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 9146 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 9147 | s64 val) |
6f505b16 | 9148 | { |
06ecb27c | 9149 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
9150 | } |
9151 | ||
06ecb27c | 9152 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 9153 | { |
06ecb27c | 9154 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 9155 | } |
d0b27fa7 PZ |
9156 | |
9157 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
9158 | u64 rt_period_us) | |
9159 | { | |
9160 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
9161 | } | |
9162 | ||
9163 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
9164 | { | |
9165 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
9166 | } | |
6d6bc0ad | 9167 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 9168 | |
fe5c7cc2 | 9169 | static struct cftype cpu_files[] = { |
052f1dc7 | 9170 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
9171 | { |
9172 | .name = "shares", | |
f4c753b7 PM |
9173 | .read_u64 = cpu_shares_read_u64, |
9174 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 9175 | }, |
052f1dc7 PZ |
9176 | #endif |
9177 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 9178 | { |
9f0c1e56 | 9179 | .name = "rt_runtime_us", |
06ecb27c PM |
9180 | .read_s64 = cpu_rt_runtime_read, |
9181 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 9182 | }, |
d0b27fa7 PZ |
9183 | { |
9184 | .name = "rt_period_us", | |
f4c753b7 PM |
9185 | .read_u64 = cpu_rt_period_read_uint, |
9186 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 9187 | }, |
052f1dc7 | 9188 | #endif |
68318b8e SV |
9189 | }; |
9190 | ||
9191 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
9192 | { | |
fe5c7cc2 | 9193 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
9194 | } |
9195 | ||
9196 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
9197 | .name = "cpu", |
9198 | .create = cpu_cgroup_create, | |
9199 | .destroy = cpu_cgroup_destroy, | |
9200 | .can_attach = cpu_cgroup_can_attach, | |
9201 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 9202 | .exit = cpu_cgroup_exit, |
38605cae IM |
9203 | .populate = cpu_cgroup_populate, |
9204 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
9205 | .early_init = 1, |
9206 | }; | |
9207 | ||
052f1dc7 | 9208 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
9209 | |
9210 | #ifdef CONFIG_CGROUP_CPUACCT | |
9211 | ||
9212 | /* | |
9213 | * CPU accounting code for task groups. | |
9214 | * | |
9215 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
9216 | * (balbir@in.ibm.com). | |
9217 | */ | |
9218 | ||
934352f2 | 9219 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
9220 | struct cpuacct { |
9221 | struct cgroup_subsys_state css; | |
9222 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43cf38eb | 9223 | u64 __percpu *cpuusage; |
ef12fefa | 9224 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 9225 | struct cpuacct *parent; |
d842de87 SV |
9226 | }; |
9227 | ||
9228 | struct cgroup_subsys cpuacct_subsys; | |
9229 | ||
9230 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 9231 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 9232 | { |
32cd756a | 9233 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
9234 | struct cpuacct, css); |
9235 | } | |
9236 | ||
9237 | /* return cpu accounting group to which this task belongs */ | |
9238 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
9239 | { | |
9240 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
9241 | struct cpuacct, css); | |
9242 | } | |
9243 | ||
9244 | /* create a new cpu accounting group */ | |
9245 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 9246 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
9247 | { |
9248 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 9249 | int i; |
d842de87 SV |
9250 | |
9251 | if (!ca) | |
ef12fefa | 9252 | goto out; |
d842de87 SV |
9253 | |
9254 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
9255 | if (!ca->cpuusage) |
9256 | goto out_free_ca; | |
9257 | ||
9258 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
9259 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
9260 | goto out_free_counters; | |
d842de87 | 9261 | |
934352f2 BR |
9262 | if (cgrp->parent) |
9263 | ca->parent = cgroup_ca(cgrp->parent); | |
9264 | ||
d842de87 | 9265 | return &ca->css; |
ef12fefa BR |
9266 | |
9267 | out_free_counters: | |
9268 | while (--i >= 0) | |
9269 | percpu_counter_destroy(&ca->cpustat[i]); | |
9270 | free_percpu(ca->cpuusage); | |
9271 | out_free_ca: | |
9272 | kfree(ca); | |
9273 | out: | |
9274 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
9275 | } |
9276 | ||
9277 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9278 | static void |
32cd756a | 9279 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9280 | { |
32cd756a | 9281 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 9282 | int i; |
d842de87 | 9283 | |
ef12fefa BR |
9284 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
9285 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
9286 | free_percpu(ca->cpuusage); |
9287 | kfree(ca); | |
9288 | } | |
9289 | ||
720f5498 KC |
9290 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
9291 | { | |
b36128c8 | 9292 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
9293 | u64 data; |
9294 | ||
9295 | #ifndef CONFIG_64BIT | |
9296 | /* | |
9297 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
9298 | */ | |
05fa785c | 9299 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 9300 | data = *cpuusage; |
05fa785c | 9301 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
9302 | #else |
9303 | data = *cpuusage; | |
9304 | #endif | |
9305 | ||
9306 | return data; | |
9307 | } | |
9308 | ||
9309 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
9310 | { | |
b36128c8 | 9311 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
9312 | |
9313 | #ifndef CONFIG_64BIT | |
9314 | /* | |
9315 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
9316 | */ | |
05fa785c | 9317 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 9318 | *cpuusage = val; |
05fa785c | 9319 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
9320 | #else |
9321 | *cpuusage = val; | |
9322 | #endif | |
9323 | } | |
9324 | ||
d842de87 | 9325 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 9326 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9327 | { |
32cd756a | 9328 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9329 | u64 totalcpuusage = 0; |
9330 | int i; | |
9331 | ||
720f5498 KC |
9332 | for_each_present_cpu(i) |
9333 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
9334 | |
9335 | return totalcpuusage; | |
9336 | } | |
9337 | ||
0297b803 DG |
9338 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9339 | u64 reset) | |
9340 | { | |
9341 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9342 | int err = 0; | |
9343 | int i; | |
9344 | ||
9345 | if (reset) { | |
9346 | err = -EINVAL; | |
9347 | goto out; | |
9348 | } | |
9349 | ||
720f5498 KC |
9350 | for_each_present_cpu(i) |
9351 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 9352 | |
0297b803 DG |
9353 | out: |
9354 | return err; | |
9355 | } | |
9356 | ||
e9515c3c KC |
9357 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
9358 | struct seq_file *m) | |
9359 | { | |
9360 | struct cpuacct *ca = cgroup_ca(cgroup); | |
9361 | u64 percpu; | |
9362 | int i; | |
9363 | ||
9364 | for_each_present_cpu(i) { | |
9365 | percpu = cpuacct_cpuusage_read(ca, i); | |
9366 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
9367 | } | |
9368 | seq_printf(m, "\n"); | |
9369 | return 0; | |
9370 | } | |
9371 | ||
ef12fefa BR |
9372 | static const char *cpuacct_stat_desc[] = { |
9373 | [CPUACCT_STAT_USER] = "user", | |
9374 | [CPUACCT_STAT_SYSTEM] = "system", | |
9375 | }; | |
9376 | ||
9377 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
9378 | struct cgroup_map_cb *cb) | |
9379 | { | |
9380 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9381 | int i; | |
9382 | ||
9383 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
9384 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
9385 | val = cputime64_to_clock_t(val); | |
9386 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
9387 | } | |
9388 | return 0; | |
9389 | } | |
9390 | ||
d842de87 SV |
9391 | static struct cftype files[] = { |
9392 | { | |
9393 | .name = "usage", | |
f4c753b7 PM |
9394 | .read_u64 = cpuusage_read, |
9395 | .write_u64 = cpuusage_write, | |
d842de87 | 9396 | }, |
e9515c3c KC |
9397 | { |
9398 | .name = "usage_percpu", | |
9399 | .read_seq_string = cpuacct_percpu_seq_read, | |
9400 | }, | |
ef12fefa BR |
9401 | { |
9402 | .name = "stat", | |
9403 | .read_map = cpuacct_stats_show, | |
9404 | }, | |
d842de87 SV |
9405 | }; |
9406 | ||
32cd756a | 9407 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9408 | { |
32cd756a | 9409 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9410 | } |
9411 | ||
9412 | /* | |
9413 | * charge this task's execution time to its accounting group. | |
9414 | * | |
9415 | * called with rq->lock held. | |
9416 | */ | |
9417 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9418 | { | |
9419 | struct cpuacct *ca; | |
934352f2 | 9420 | int cpu; |
d842de87 | 9421 | |
c40c6f85 | 9422 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
9423 | return; |
9424 | ||
934352f2 | 9425 | cpu = task_cpu(tsk); |
a18b83b7 BR |
9426 | |
9427 | rcu_read_lock(); | |
9428 | ||
d842de87 | 9429 | ca = task_ca(tsk); |
d842de87 | 9430 | |
934352f2 | 9431 | for (; ca; ca = ca->parent) { |
b36128c8 | 9432 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
9433 | *cpuusage += cputime; |
9434 | } | |
a18b83b7 BR |
9435 | |
9436 | rcu_read_unlock(); | |
d842de87 SV |
9437 | } |
9438 | ||
fa535a77 AB |
9439 | /* |
9440 | * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large | |
9441 | * in cputime_t units. As a result, cpuacct_update_stats calls | |
9442 | * percpu_counter_add with values large enough to always overflow the | |
9443 | * per cpu batch limit causing bad SMP scalability. | |
9444 | * | |
9445 | * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we | |
9446 | * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled | |
9447 | * and enabled. We cap it at INT_MAX which is the largest allowed batch value. | |
9448 | */ | |
9449 | #ifdef CONFIG_SMP | |
9450 | #define CPUACCT_BATCH \ | |
9451 | min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX) | |
9452 | #else | |
9453 | #define CPUACCT_BATCH 0 | |
9454 | #endif | |
9455 | ||
ef12fefa BR |
9456 | /* |
9457 | * Charge the system/user time to the task's accounting group. | |
9458 | */ | |
9459 | static void cpuacct_update_stats(struct task_struct *tsk, | |
9460 | enum cpuacct_stat_index idx, cputime_t val) | |
9461 | { | |
9462 | struct cpuacct *ca; | |
fa535a77 | 9463 | int batch = CPUACCT_BATCH; |
ef12fefa BR |
9464 | |
9465 | if (unlikely(!cpuacct_subsys.active)) | |
9466 | return; | |
9467 | ||
9468 | rcu_read_lock(); | |
9469 | ca = task_ca(tsk); | |
9470 | ||
9471 | do { | |
fa535a77 | 9472 | __percpu_counter_add(&ca->cpustat[idx], val, batch); |
ef12fefa BR |
9473 | ca = ca->parent; |
9474 | } while (ca); | |
9475 | rcu_read_unlock(); | |
9476 | } | |
9477 | ||
d842de87 SV |
9478 | struct cgroup_subsys cpuacct_subsys = { |
9479 | .name = "cpuacct", | |
9480 | .create = cpuacct_create, | |
9481 | .destroy = cpuacct_destroy, | |
9482 | .populate = cpuacct_populate, | |
9483 | .subsys_id = cpuacct_subsys_id, | |
9484 | }; | |
9485 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf | 9486 |