Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/sched.c | |
3 | * | |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 LT |
34 | #include <linux/highmem.h> |
35 | #include <linux/smp_lock.h> | |
36 | #include <asm/mmu_context.h> | |
37 | #include <linux/interrupt.h> | |
c59ede7b | 38 | #include <linux/capability.h> |
1da177e4 LT |
39 | #include <linux/completion.h> |
40 | #include <linux/kernel_stat.h> | |
9a11b49a | 41 | #include <linux/debug_locks.h> |
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> | |
57 | #include <linux/kthread.h> | |
58 | #include <linux/seq_file.h> | |
e692ab53 | 59 | #include <linux/sysctl.h> |
1da177e4 LT |
60 | #include <linux/syscalls.h> |
61 | #include <linux/times.h> | |
8f0ab514 | 62 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 63 | #include <linux/kprobes.h> |
0ff92245 | 64 | #include <linux/delayacct.h> |
5517d86b | 65 | #include <linux/reciprocal_div.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> |
434d53b0 | 70 | #include <linux/bootmem.h> |
1da177e4 | 71 | |
5517d86b | 72 | #include <asm/tlb.h> |
838225b4 | 73 | #include <asm/irq_regs.h> |
1da177e4 | 74 | |
b035b6de AD |
75 | /* |
76 | * Scheduler clock - returns current time in nanosec units. | |
77 | * This is default implementation. | |
78 | * Architectures and sub-architectures can override this. | |
79 | */ | |
80 | unsigned long long __attribute__((weak)) sched_clock(void) | |
81 | { | |
d6322faf | 82 | return (unsigned long long)jiffies * (NSEC_PER_SEC / HZ); |
b035b6de AD |
83 | } |
84 | ||
1da177e4 LT |
85 | /* |
86 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
87 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
88 | * and back. | |
89 | */ | |
90 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
91 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
92 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
93 | ||
94 | /* | |
95 | * 'User priority' is the nice value converted to something we | |
96 | * can work with better when scaling various scheduler parameters, | |
97 | * it's a [ 0 ... 39 ] range. | |
98 | */ | |
99 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
100 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
101 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
102 | ||
103 | /* | |
d7876a08 | 104 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 105 | */ |
d6322faf | 106 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 107 | |
6aa645ea IM |
108 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
109 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
110 | ||
1da177e4 LT |
111 | /* |
112 | * These are the 'tuning knobs' of the scheduler: | |
113 | * | |
a4ec24b4 | 114 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
115 | * Timeslices get refilled after they expire. |
116 | */ | |
1da177e4 | 117 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 118 | |
d0b27fa7 PZ |
119 | /* |
120 | * single value that denotes runtime == period, ie unlimited time. | |
121 | */ | |
122 | #define RUNTIME_INF ((u64)~0ULL) | |
123 | ||
5517d86b ED |
124 | #ifdef CONFIG_SMP |
125 | /* | |
126 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
127 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
128 | */ | |
129 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
130 | { | |
131 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
132 | } | |
133 | ||
134 | /* | |
135 | * Each time a sched group cpu_power is changed, | |
136 | * we must compute its reciprocal value | |
137 | */ | |
138 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
139 | { | |
140 | sg->__cpu_power += val; | |
141 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
142 | } | |
143 | #endif | |
144 | ||
e05606d3 IM |
145 | static inline int rt_policy(int policy) |
146 | { | |
147 | if (unlikely(policy == SCHED_FIFO) || unlikely(policy == SCHED_RR)) | |
148 | return 1; | |
149 | return 0; | |
150 | } | |
151 | ||
152 | static inline int task_has_rt_policy(struct task_struct *p) | |
153 | { | |
154 | return rt_policy(p->policy); | |
155 | } | |
156 | ||
1da177e4 | 157 | /* |
6aa645ea | 158 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 159 | */ |
6aa645ea IM |
160 | struct rt_prio_array { |
161 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
162 | struct list_head queue[MAX_RT_PRIO]; | |
163 | }; | |
164 | ||
d0b27fa7 PZ |
165 | struct rt_bandwidth { |
166 | ktime_t rt_period; | |
167 | u64 rt_runtime; | |
ac086bc2 | 168 | spinlock_t rt_runtime_lock; |
d0b27fa7 PZ |
169 | struct hrtimer rt_period_timer; |
170 | }; | |
171 | ||
172 | static struct rt_bandwidth def_rt_bandwidth; | |
173 | ||
174 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
175 | ||
176 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
177 | { | |
178 | struct rt_bandwidth *rt_b = | |
179 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
180 | ktime_t now; | |
181 | int overrun; | |
182 | int idle = 0; | |
183 | ||
184 | for (;;) { | |
185 | now = hrtimer_cb_get_time(timer); | |
186 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
187 | ||
188 | if (!overrun) | |
189 | break; | |
190 | ||
191 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
192 | } | |
193 | ||
194 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
195 | } | |
196 | ||
197 | static | |
198 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
199 | { | |
200 | rt_b->rt_period = ns_to_ktime(period); | |
201 | rt_b->rt_runtime = runtime; | |
202 | ||
ac086bc2 PZ |
203 | spin_lock_init(&rt_b->rt_runtime_lock); |
204 | ||
d0b27fa7 PZ |
205 | hrtimer_init(&rt_b->rt_period_timer, |
206 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
207 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
208 | rt_b->rt_period_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ; | |
209 | } | |
210 | ||
211 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
212 | { | |
213 | ktime_t now; | |
214 | ||
215 | if (rt_b->rt_runtime == RUNTIME_INF) | |
216 | return; | |
217 | ||
218 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
219 | return; | |
220 | ||
221 | spin_lock(&rt_b->rt_runtime_lock); | |
222 | for (;;) { | |
223 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
224 | break; | |
225 | ||
226 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
227 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
228 | hrtimer_start(&rt_b->rt_period_timer, | |
229 | rt_b->rt_period_timer.expires, | |
230 | HRTIMER_MODE_ABS); | |
231 | } | |
232 | spin_unlock(&rt_b->rt_runtime_lock); | |
233 | } | |
234 | ||
235 | #ifdef CONFIG_RT_GROUP_SCHED | |
236 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
237 | { | |
238 | hrtimer_cancel(&rt_b->rt_period_timer); | |
239 | } | |
240 | #endif | |
241 | ||
052f1dc7 | 242 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 243 | |
68318b8e SV |
244 | #include <linux/cgroup.h> |
245 | ||
29f59db3 SV |
246 | struct cfs_rq; |
247 | ||
6f505b16 PZ |
248 | static LIST_HEAD(task_groups); |
249 | ||
29f59db3 | 250 | /* task group related information */ |
4cf86d77 | 251 | struct task_group { |
052f1dc7 | 252 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
253 | struct cgroup_subsys_state css; |
254 | #endif | |
052f1dc7 PZ |
255 | |
256 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
29f59db3 SV |
257 | /* schedulable entities of this group on each cpu */ |
258 | struct sched_entity **se; | |
259 | /* runqueue "owned" by this group on each cpu */ | |
260 | struct cfs_rq **cfs_rq; | |
261 | unsigned long shares; | |
052f1dc7 PZ |
262 | #endif |
263 | ||
264 | #ifdef CONFIG_RT_GROUP_SCHED | |
265 | struct sched_rt_entity **rt_se; | |
266 | struct rt_rq **rt_rq; | |
267 | ||
d0b27fa7 | 268 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 269 | #endif |
6b2d7700 | 270 | |
ae8393e5 | 271 | struct rcu_head rcu; |
6f505b16 | 272 | struct list_head list; |
29f59db3 SV |
273 | }; |
274 | ||
052f1dc7 | 275 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
276 | /* Default task group's sched entity on each cpu */ |
277 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
278 | /* Default task group's cfs_rq on each cpu */ | |
279 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
052f1dc7 PZ |
280 | #endif |
281 | ||
282 | #ifdef CONFIG_RT_GROUP_SCHED | |
283 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
284 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
052f1dc7 | 285 | #endif |
6f505b16 | 286 | |
8ed36996 | 287 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
288 | * a task group's cpu shares. |
289 | */ | |
8ed36996 | 290 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 291 | |
a1835615 SV |
292 | /* doms_cur_mutex serializes access to doms_cur[] array */ |
293 | static DEFINE_MUTEX(doms_cur_mutex); | |
294 | ||
052f1dc7 | 295 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
296 | #ifdef CONFIG_USER_SCHED |
297 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
298 | #else | |
299 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD | |
300 | #endif | |
301 | ||
052f1dc7 PZ |
302 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
303 | #endif | |
304 | ||
29f59db3 | 305 | /* Default task group. |
3a252015 | 306 | * Every task in system belong to this group at bootup. |
29f59db3 | 307 | */ |
434d53b0 | 308 | struct task_group init_task_group; |
29f59db3 SV |
309 | |
310 | /* return group to which a task belongs */ | |
4cf86d77 | 311 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 312 | { |
4cf86d77 | 313 | struct task_group *tg; |
9b5b7751 | 314 | |
052f1dc7 | 315 | #ifdef CONFIG_USER_SCHED |
24e377a8 | 316 | tg = p->user->tg; |
052f1dc7 | 317 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
318 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
319 | struct task_group, css); | |
24e377a8 | 320 | #else |
41a2d6cf | 321 | tg = &init_task_group; |
24e377a8 | 322 | #endif |
9b5b7751 | 323 | return tg; |
29f59db3 SV |
324 | } |
325 | ||
326 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 327 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 328 | { |
052f1dc7 | 329 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
330 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
331 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 332 | #endif |
6f505b16 | 333 | |
052f1dc7 | 334 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
335 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
336 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 337 | #endif |
29f59db3 SV |
338 | } |
339 | ||
a1835615 SV |
340 | static inline void lock_doms_cur(void) |
341 | { | |
342 | mutex_lock(&doms_cur_mutex); | |
343 | } | |
344 | ||
345 | static inline void unlock_doms_cur(void) | |
346 | { | |
347 | mutex_unlock(&doms_cur_mutex); | |
348 | } | |
349 | ||
29f59db3 SV |
350 | #else |
351 | ||
6f505b16 | 352 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
a1835615 SV |
353 | static inline void lock_doms_cur(void) { } |
354 | static inline void unlock_doms_cur(void) { } | |
29f59db3 | 355 | |
052f1dc7 | 356 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 357 | |
6aa645ea IM |
358 | /* CFS-related fields in a runqueue */ |
359 | struct cfs_rq { | |
360 | struct load_weight load; | |
361 | unsigned long nr_running; | |
362 | ||
6aa645ea | 363 | u64 exec_clock; |
e9acbff6 | 364 | u64 min_vruntime; |
6aa645ea IM |
365 | |
366 | struct rb_root tasks_timeline; | |
367 | struct rb_node *rb_leftmost; | |
368 | struct rb_node *rb_load_balance_curr; | |
6aa645ea IM |
369 | /* 'curr' points to currently running entity on this cfs_rq. |
370 | * It is set to NULL otherwise (i.e when none are currently running). | |
371 | */ | |
aa2ac252 | 372 | struct sched_entity *curr, *next; |
ddc97297 PZ |
373 | |
374 | unsigned long nr_spread_over; | |
375 | ||
62160e3f | 376 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
377 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
378 | ||
41a2d6cf IM |
379 | /* |
380 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
381 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
382 | * (like users, containers etc.) | |
383 | * | |
384 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
385 | * list is used during load balance. | |
386 | */ | |
41a2d6cf IM |
387 | struct list_head leaf_cfs_rq_list; |
388 | struct task_group *tg; /* group that "owns" this runqueue */ | |
6aa645ea IM |
389 | #endif |
390 | }; | |
1da177e4 | 391 | |
6aa645ea IM |
392 | /* Real-Time classes' related field in a runqueue: */ |
393 | struct rt_rq { | |
394 | struct rt_prio_array active; | |
63489e45 | 395 | unsigned long rt_nr_running; |
052f1dc7 | 396 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
397 | int highest_prio; /* highest queued rt task prio */ |
398 | #endif | |
fa85ae24 | 399 | #ifdef CONFIG_SMP |
73fe6aae | 400 | unsigned long rt_nr_migratory; |
a22d7fc1 | 401 | int overloaded; |
fa85ae24 | 402 | #endif |
6f505b16 | 403 | int rt_throttled; |
fa85ae24 | 404 | u64 rt_time; |
ac086bc2 PZ |
405 | u64 rt_runtime; |
406 | spinlock_t rt_runtime_lock; | |
6f505b16 | 407 | |
052f1dc7 | 408 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
409 | unsigned long rt_nr_boosted; |
410 | ||
6f505b16 PZ |
411 | struct rq *rq; |
412 | struct list_head leaf_rt_rq_list; | |
413 | struct task_group *tg; | |
414 | struct sched_rt_entity *rt_se; | |
415 | #endif | |
6aa645ea IM |
416 | }; |
417 | ||
57d885fe GH |
418 | #ifdef CONFIG_SMP |
419 | ||
420 | /* | |
421 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
422 | * variables. Each exclusive cpuset essentially defines an island domain by |
423 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
424 | * exclusive cpuset is created, we also create and attach a new root-domain |
425 | * object. | |
426 | * | |
57d885fe GH |
427 | */ |
428 | struct root_domain { | |
429 | atomic_t refcount; | |
430 | cpumask_t span; | |
431 | cpumask_t online; | |
637f5085 | 432 | |
0eab9146 | 433 | /* |
637f5085 GH |
434 | * The "RT overload" flag: it gets set if a CPU has more than |
435 | * one runnable RT task. | |
436 | */ | |
437 | cpumask_t rto_mask; | |
0eab9146 | 438 | atomic_t rto_count; |
57d885fe GH |
439 | }; |
440 | ||
dc938520 GH |
441 | /* |
442 | * By default the system creates a single root-domain with all cpus as | |
443 | * members (mimicking the global state we have today). | |
444 | */ | |
57d885fe GH |
445 | static struct root_domain def_root_domain; |
446 | ||
447 | #endif | |
448 | ||
1da177e4 LT |
449 | /* |
450 | * This is the main, per-CPU runqueue data structure. | |
451 | * | |
452 | * Locking rule: those places that want to lock multiple runqueues | |
453 | * (such as the load balancing or the thread migration code), lock | |
454 | * acquire operations must be ordered by ascending &runqueue. | |
455 | */ | |
70b97a7f | 456 | struct rq { |
d8016491 IM |
457 | /* runqueue lock: */ |
458 | spinlock_t lock; | |
1da177e4 LT |
459 | |
460 | /* | |
461 | * nr_running and cpu_load should be in the same cacheline because | |
462 | * remote CPUs use both these fields when doing load calculation. | |
463 | */ | |
464 | unsigned long nr_running; | |
6aa645ea IM |
465 | #define CPU_LOAD_IDX_MAX 5 |
466 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 467 | unsigned char idle_at_tick; |
46cb4b7c | 468 | #ifdef CONFIG_NO_HZ |
15934a37 | 469 | unsigned long last_tick_seen; |
46cb4b7c SS |
470 | unsigned char in_nohz_recently; |
471 | #endif | |
d8016491 IM |
472 | /* capture load from *all* tasks on this cpu: */ |
473 | struct load_weight load; | |
6aa645ea IM |
474 | unsigned long nr_load_updates; |
475 | u64 nr_switches; | |
476 | ||
477 | struct cfs_rq cfs; | |
6f505b16 | 478 | struct rt_rq rt; |
6f505b16 | 479 | |
6aa645ea | 480 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
481 | /* list of leaf cfs_rq on this cpu: */ |
482 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
483 | #endif |
484 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 485 | struct list_head leaf_rt_rq_list; |
1da177e4 | 486 | #endif |
1da177e4 LT |
487 | |
488 | /* | |
489 | * This is part of a global counter where only the total sum | |
490 | * over all CPUs matters. A task can increase this counter on | |
491 | * one CPU and if it got migrated afterwards it may decrease | |
492 | * it on another CPU. Always updated under the runqueue lock: | |
493 | */ | |
494 | unsigned long nr_uninterruptible; | |
495 | ||
36c8b586 | 496 | struct task_struct *curr, *idle; |
c9819f45 | 497 | unsigned long next_balance; |
1da177e4 | 498 | struct mm_struct *prev_mm; |
6aa645ea | 499 | |
6aa645ea IM |
500 | u64 clock, prev_clock_raw; |
501 | s64 clock_max_delta; | |
502 | ||
cc203d24 | 503 | unsigned int clock_warps, clock_overflows, clock_underflows; |
2aa44d05 IM |
504 | u64 idle_clock; |
505 | unsigned int clock_deep_idle_events; | |
529c7726 | 506 | u64 tick_timestamp; |
6aa645ea | 507 | |
1da177e4 LT |
508 | atomic_t nr_iowait; |
509 | ||
510 | #ifdef CONFIG_SMP | |
0eab9146 | 511 | struct root_domain *rd; |
1da177e4 LT |
512 | struct sched_domain *sd; |
513 | ||
514 | /* For active balancing */ | |
515 | int active_balance; | |
516 | int push_cpu; | |
d8016491 IM |
517 | /* cpu of this runqueue: */ |
518 | int cpu; | |
1da177e4 | 519 | |
36c8b586 | 520 | struct task_struct *migration_thread; |
1da177e4 LT |
521 | struct list_head migration_queue; |
522 | #endif | |
523 | ||
8f4d37ec PZ |
524 | #ifdef CONFIG_SCHED_HRTICK |
525 | unsigned long hrtick_flags; | |
526 | ktime_t hrtick_expire; | |
527 | struct hrtimer hrtick_timer; | |
528 | #endif | |
529 | ||
1da177e4 LT |
530 | #ifdef CONFIG_SCHEDSTATS |
531 | /* latency stats */ | |
532 | struct sched_info rq_sched_info; | |
533 | ||
534 | /* sys_sched_yield() stats */ | |
480b9434 KC |
535 | unsigned int yld_exp_empty; |
536 | unsigned int yld_act_empty; | |
537 | unsigned int yld_both_empty; | |
538 | unsigned int yld_count; | |
1da177e4 LT |
539 | |
540 | /* schedule() stats */ | |
480b9434 KC |
541 | unsigned int sched_switch; |
542 | unsigned int sched_count; | |
543 | unsigned int sched_goidle; | |
1da177e4 LT |
544 | |
545 | /* try_to_wake_up() stats */ | |
480b9434 KC |
546 | unsigned int ttwu_count; |
547 | unsigned int ttwu_local; | |
b8efb561 IM |
548 | |
549 | /* BKL stats */ | |
480b9434 | 550 | unsigned int bkl_count; |
1da177e4 | 551 | #endif |
fcb99371 | 552 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
553 | }; |
554 | ||
f34e3b61 | 555 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 556 | |
dd41f596 IM |
557 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
558 | { | |
559 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
560 | } | |
561 | ||
0a2966b4 CL |
562 | static inline int cpu_of(struct rq *rq) |
563 | { | |
564 | #ifdef CONFIG_SMP | |
565 | return rq->cpu; | |
566 | #else | |
567 | return 0; | |
568 | #endif | |
569 | } | |
570 | ||
15934a37 GC |
571 | #ifdef CONFIG_NO_HZ |
572 | static inline bool nohz_on(int cpu) | |
573 | { | |
574 | return tick_get_tick_sched(cpu)->nohz_mode != NOHZ_MODE_INACTIVE; | |
575 | } | |
576 | ||
577 | static inline u64 max_skipped_ticks(struct rq *rq) | |
578 | { | |
579 | return nohz_on(cpu_of(rq)) ? jiffies - rq->last_tick_seen + 2 : 1; | |
580 | } | |
581 | ||
582 | static inline void update_last_tick_seen(struct rq *rq) | |
583 | { | |
584 | rq->last_tick_seen = jiffies; | |
585 | } | |
586 | #else | |
587 | static inline u64 max_skipped_ticks(struct rq *rq) | |
588 | { | |
589 | return 1; | |
590 | } | |
591 | ||
592 | static inline void update_last_tick_seen(struct rq *rq) | |
593 | { | |
594 | } | |
595 | #endif | |
596 | ||
20d315d4 | 597 | /* |
b04a0f4c IM |
598 | * Update the per-runqueue clock, as finegrained as the platform can give |
599 | * us, but without assuming monotonicity, etc.: | |
20d315d4 | 600 | */ |
b04a0f4c | 601 | static void __update_rq_clock(struct rq *rq) |
20d315d4 IM |
602 | { |
603 | u64 prev_raw = rq->prev_clock_raw; | |
604 | u64 now = sched_clock(); | |
605 | s64 delta = now - prev_raw; | |
606 | u64 clock = rq->clock; | |
607 | ||
b04a0f4c IM |
608 | #ifdef CONFIG_SCHED_DEBUG |
609 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
610 | #endif | |
20d315d4 IM |
611 | /* |
612 | * Protect against sched_clock() occasionally going backwards: | |
613 | */ | |
614 | if (unlikely(delta < 0)) { | |
615 | clock++; | |
616 | rq->clock_warps++; | |
617 | } else { | |
618 | /* | |
619 | * Catch too large forward jumps too: | |
620 | */ | |
15934a37 GC |
621 | u64 max_jump = max_skipped_ticks(rq) * TICK_NSEC; |
622 | u64 max_time = rq->tick_timestamp + max_jump; | |
623 | ||
624 | if (unlikely(clock + delta > max_time)) { | |
625 | if (clock < max_time) | |
626 | clock = max_time; | |
529c7726 IM |
627 | else |
628 | clock++; | |
20d315d4 IM |
629 | rq->clock_overflows++; |
630 | } else { | |
631 | if (unlikely(delta > rq->clock_max_delta)) | |
632 | rq->clock_max_delta = delta; | |
633 | clock += delta; | |
634 | } | |
635 | } | |
636 | ||
637 | rq->prev_clock_raw = now; | |
638 | rq->clock = clock; | |
b04a0f4c | 639 | } |
20d315d4 | 640 | |
b04a0f4c IM |
641 | static void update_rq_clock(struct rq *rq) |
642 | { | |
643 | if (likely(smp_processor_id() == cpu_of(rq))) | |
644 | __update_rq_clock(rq); | |
20d315d4 IM |
645 | } |
646 | ||
674311d5 NP |
647 | /* |
648 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 649 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
650 | * |
651 | * The domain tree of any CPU may only be accessed from within | |
652 | * preempt-disabled sections. | |
653 | */ | |
48f24c4d IM |
654 | #define for_each_domain(cpu, __sd) \ |
655 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
656 | |
657 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
658 | #define this_rq() (&__get_cpu_var(runqueues)) | |
659 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
660 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
661 | ||
bf5c91ba IM |
662 | /* |
663 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
664 | */ | |
665 | #ifdef CONFIG_SCHED_DEBUG | |
666 | # define const_debug __read_mostly | |
667 | #else | |
668 | # define const_debug static const | |
669 | #endif | |
670 | ||
671 | /* | |
672 | * Debugging: various feature bits | |
673 | */ | |
674 | enum { | |
bbdba7c0 | 675 | SCHED_FEAT_NEW_FAIR_SLEEPERS = 1, |
9612633a IM |
676 | SCHED_FEAT_WAKEUP_PREEMPT = 2, |
677 | SCHED_FEAT_START_DEBIT = 4, | |
d25ce4cd IM |
678 | SCHED_FEAT_AFFINE_WAKEUPS = 8, |
679 | SCHED_FEAT_CACHE_HOT_BUDDY = 16, | |
02e2b83b IM |
680 | SCHED_FEAT_SYNC_WAKEUPS = 32, |
681 | SCHED_FEAT_HRTICK = 64, | |
682 | SCHED_FEAT_DOUBLE_TICK = 128, | |
112f53f5 | 683 | SCHED_FEAT_NORMALIZED_SLEEPER = 256, |
bf5c91ba IM |
684 | }; |
685 | ||
686 | const_debug unsigned int sysctl_sched_features = | |
8401f775 | 687 | SCHED_FEAT_NEW_FAIR_SLEEPERS * 1 | |
9612633a | 688 | SCHED_FEAT_WAKEUP_PREEMPT * 1 | |
8401f775 | 689 | SCHED_FEAT_START_DEBIT * 1 | |
d25ce4cd IM |
690 | SCHED_FEAT_AFFINE_WAKEUPS * 1 | |
691 | SCHED_FEAT_CACHE_HOT_BUDDY * 1 | | |
02e2b83b | 692 | SCHED_FEAT_SYNC_WAKEUPS * 1 | |
8f4d37ec | 693 | SCHED_FEAT_HRTICK * 1 | |
112f53f5 PZ |
694 | SCHED_FEAT_DOUBLE_TICK * 0 | |
695 | SCHED_FEAT_NORMALIZED_SLEEPER * 1; | |
bf5c91ba IM |
696 | |
697 | #define sched_feat(x) (sysctl_sched_features & SCHED_FEAT_##x) | |
698 | ||
b82d9fdd PZ |
699 | /* |
700 | * Number of tasks to iterate in a single balance run. | |
701 | * Limited because this is done with IRQs disabled. | |
702 | */ | |
703 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
704 | ||
fa85ae24 | 705 | /* |
9f0c1e56 | 706 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
707 | * default: 1s |
708 | */ | |
9f0c1e56 | 709 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 710 | |
6892b75e IM |
711 | static __read_mostly int scheduler_running; |
712 | ||
9f0c1e56 PZ |
713 | /* |
714 | * part of the period that we allow rt tasks to run in us. | |
715 | * default: 0.95s | |
716 | */ | |
717 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 718 | |
d0b27fa7 PZ |
719 | static inline u64 global_rt_period(void) |
720 | { | |
721 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
722 | } | |
723 | ||
724 | static inline u64 global_rt_runtime(void) | |
725 | { | |
726 | if (sysctl_sched_rt_period < 0) | |
727 | return RUNTIME_INF; | |
728 | ||
729 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
730 | } | |
fa85ae24 | 731 | |
27ec4407 IM |
732 | static const unsigned long long time_sync_thresh = 100000; |
733 | ||
734 | static DEFINE_PER_CPU(unsigned long long, time_offset); | |
735 | static DEFINE_PER_CPU(unsigned long long, prev_cpu_time); | |
736 | ||
e436d800 | 737 | /* |
27ec4407 IM |
738 | * Global lock which we take every now and then to synchronize |
739 | * the CPUs time. This method is not warp-safe, but it's good | |
740 | * enough to synchronize slowly diverging time sources and thus | |
741 | * it's good enough for tracing: | |
e436d800 | 742 | */ |
27ec4407 IM |
743 | static DEFINE_SPINLOCK(time_sync_lock); |
744 | static unsigned long long prev_global_time; | |
745 | ||
746 | static unsigned long long __sync_cpu_clock(cycles_t time, int cpu) | |
747 | { | |
748 | unsigned long flags; | |
749 | ||
750 | spin_lock_irqsave(&time_sync_lock, flags); | |
751 | ||
752 | if (time < prev_global_time) { | |
753 | per_cpu(time_offset, cpu) += prev_global_time - time; | |
754 | time = prev_global_time; | |
755 | } else { | |
756 | prev_global_time = time; | |
757 | } | |
758 | ||
759 | spin_unlock_irqrestore(&time_sync_lock, flags); | |
760 | ||
761 | return time; | |
762 | } | |
763 | ||
764 | static unsigned long long __cpu_clock(int cpu) | |
e436d800 | 765 | { |
e436d800 IM |
766 | unsigned long long now; |
767 | unsigned long flags; | |
b04a0f4c | 768 | struct rq *rq; |
e436d800 | 769 | |
8ced5f69 IM |
770 | /* |
771 | * Only call sched_clock() if the scheduler has already been | |
772 | * initialized (some code might call cpu_clock() very early): | |
773 | */ | |
6892b75e IM |
774 | if (unlikely(!scheduler_running)) |
775 | return 0; | |
776 | ||
777 | local_irq_save(flags); | |
778 | rq = cpu_rq(cpu); | |
779 | update_rq_clock(rq); | |
b04a0f4c | 780 | now = rq->clock; |
2cd4d0ea | 781 | local_irq_restore(flags); |
e436d800 IM |
782 | |
783 | return now; | |
784 | } | |
27ec4407 IM |
785 | |
786 | /* | |
787 | * For kernel-internal use: high-speed (but slightly incorrect) per-cpu | |
788 | * clock constructed from sched_clock(): | |
789 | */ | |
790 | unsigned long long cpu_clock(int cpu) | |
791 | { | |
792 | unsigned long long prev_cpu_time, time, delta_time; | |
793 | ||
794 | prev_cpu_time = per_cpu(prev_cpu_time, cpu); | |
795 | time = __cpu_clock(cpu) + per_cpu(time_offset, cpu); | |
796 | delta_time = time-prev_cpu_time; | |
797 | ||
798 | if (unlikely(delta_time > time_sync_thresh)) | |
799 | time = __sync_cpu_clock(time, cpu); | |
800 | ||
801 | return time; | |
802 | } | |
a58f6f25 | 803 | EXPORT_SYMBOL_GPL(cpu_clock); |
e436d800 | 804 | |
1da177e4 | 805 | #ifndef prepare_arch_switch |
4866cde0 NP |
806 | # define prepare_arch_switch(next) do { } while (0) |
807 | #endif | |
808 | #ifndef finish_arch_switch | |
809 | # define finish_arch_switch(prev) do { } while (0) | |
810 | #endif | |
811 | ||
051a1d1a DA |
812 | static inline int task_current(struct rq *rq, struct task_struct *p) |
813 | { | |
814 | return rq->curr == p; | |
815 | } | |
816 | ||
4866cde0 | 817 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 818 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 819 | { |
051a1d1a | 820 | return task_current(rq, p); |
4866cde0 NP |
821 | } |
822 | ||
70b97a7f | 823 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
824 | { |
825 | } | |
826 | ||
70b97a7f | 827 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 828 | { |
da04c035 IM |
829 | #ifdef CONFIG_DEBUG_SPINLOCK |
830 | /* this is a valid case when another task releases the spinlock */ | |
831 | rq->lock.owner = current; | |
832 | #endif | |
8a25d5de IM |
833 | /* |
834 | * If we are tracking spinlock dependencies then we have to | |
835 | * fix up the runqueue lock - which gets 'carried over' from | |
836 | * prev into current: | |
837 | */ | |
838 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
839 | ||
4866cde0 NP |
840 | spin_unlock_irq(&rq->lock); |
841 | } | |
842 | ||
843 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 844 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
845 | { |
846 | #ifdef CONFIG_SMP | |
847 | return p->oncpu; | |
848 | #else | |
051a1d1a | 849 | return task_current(rq, p); |
4866cde0 NP |
850 | #endif |
851 | } | |
852 | ||
70b97a7f | 853 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
854 | { |
855 | #ifdef CONFIG_SMP | |
856 | /* | |
857 | * We can optimise this out completely for !SMP, because the | |
858 | * SMP rebalancing from interrupt is the only thing that cares | |
859 | * here. | |
860 | */ | |
861 | next->oncpu = 1; | |
862 | #endif | |
863 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
864 | spin_unlock_irq(&rq->lock); | |
865 | #else | |
866 | spin_unlock(&rq->lock); | |
867 | #endif | |
868 | } | |
869 | ||
70b97a7f | 870 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
871 | { |
872 | #ifdef CONFIG_SMP | |
873 | /* | |
874 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
875 | * We must ensure this doesn't happen until the switch is completely | |
876 | * finished. | |
877 | */ | |
878 | smp_wmb(); | |
879 | prev->oncpu = 0; | |
880 | #endif | |
881 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
882 | local_irq_enable(); | |
1da177e4 | 883 | #endif |
4866cde0 NP |
884 | } |
885 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 886 | |
b29739f9 IM |
887 | /* |
888 | * __task_rq_lock - lock the runqueue a given task resides on. | |
889 | * Must be called interrupts disabled. | |
890 | */ | |
70b97a7f | 891 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
892 | __acquires(rq->lock) |
893 | { | |
3a5c359a AK |
894 | for (;;) { |
895 | struct rq *rq = task_rq(p); | |
896 | spin_lock(&rq->lock); | |
897 | if (likely(rq == task_rq(p))) | |
898 | return rq; | |
b29739f9 | 899 | spin_unlock(&rq->lock); |
b29739f9 | 900 | } |
b29739f9 IM |
901 | } |
902 | ||
1da177e4 LT |
903 | /* |
904 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 905 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
906 | * explicitly disabling preemption. |
907 | */ | |
70b97a7f | 908 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
909 | __acquires(rq->lock) |
910 | { | |
70b97a7f | 911 | struct rq *rq; |
1da177e4 | 912 | |
3a5c359a AK |
913 | for (;;) { |
914 | local_irq_save(*flags); | |
915 | rq = task_rq(p); | |
916 | spin_lock(&rq->lock); | |
917 | if (likely(rq == task_rq(p))) | |
918 | return rq; | |
1da177e4 | 919 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 920 | } |
1da177e4 LT |
921 | } |
922 | ||
a9957449 | 923 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
924 | __releases(rq->lock) |
925 | { | |
926 | spin_unlock(&rq->lock); | |
927 | } | |
928 | ||
70b97a7f | 929 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
930 | __releases(rq->lock) |
931 | { | |
932 | spin_unlock_irqrestore(&rq->lock, *flags); | |
933 | } | |
934 | ||
1da177e4 | 935 | /* |
cc2a73b5 | 936 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 937 | */ |
a9957449 | 938 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
939 | __acquires(rq->lock) |
940 | { | |
70b97a7f | 941 | struct rq *rq; |
1da177e4 LT |
942 | |
943 | local_irq_disable(); | |
944 | rq = this_rq(); | |
945 | spin_lock(&rq->lock); | |
946 | ||
947 | return rq; | |
948 | } | |
949 | ||
1b9f19c2 | 950 | /* |
2aa44d05 | 951 | * We are going deep-idle (irqs are disabled): |
1b9f19c2 | 952 | */ |
2aa44d05 | 953 | void sched_clock_idle_sleep_event(void) |
1b9f19c2 | 954 | { |
2aa44d05 IM |
955 | struct rq *rq = cpu_rq(smp_processor_id()); |
956 | ||
957 | spin_lock(&rq->lock); | |
958 | __update_rq_clock(rq); | |
959 | spin_unlock(&rq->lock); | |
960 | rq->clock_deep_idle_events++; | |
961 | } | |
962 | EXPORT_SYMBOL_GPL(sched_clock_idle_sleep_event); | |
963 | ||
964 | /* | |
965 | * We just idled delta nanoseconds (called with irqs disabled): | |
966 | */ | |
967 | void sched_clock_idle_wakeup_event(u64 delta_ns) | |
968 | { | |
969 | struct rq *rq = cpu_rq(smp_processor_id()); | |
970 | u64 now = sched_clock(); | |
1b9f19c2 | 971 | |
2aa44d05 IM |
972 | rq->idle_clock += delta_ns; |
973 | /* | |
974 | * Override the previous timestamp and ignore all | |
975 | * sched_clock() deltas that occured while we idled, | |
976 | * and use the PM-provided delta_ns to advance the | |
977 | * rq clock: | |
978 | */ | |
979 | spin_lock(&rq->lock); | |
980 | rq->prev_clock_raw = now; | |
981 | rq->clock += delta_ns; | |
982 | spin_unlock(&rq->lock); | |
782daeee | 983 | touch_softlockup_watchdog(); |
1b9f19c2 | 984 | } |
2aa44d05 | 985 | EXPORT_SYMBOL_GPL(sched_clock_idle_wakeup_event); |
1b9f19c2 | 986 | |
8f4d37ec PZ |
987 | static void __resched_task(struct task_struct *p, int tif_bit); |
988 | ||
989 | static inline void resched_task(struct task_struct *p) | |
990 | { | |
991 | __resched_task(p, TIF_NEED_RESCHED); | |
992 | } | |
993 | ||
994 | #ifdef CONFIG_SCHED_HRTICK | |
995 | /* | |
996 | * Use HR-timers to deliver accurate preemption points. | |
997 | * | |
998 | * Its all a bit involved since we cannot program an hrt while holding the | |
999 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1000 | * reschedule event. | |
1001 | * | |
1002 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1003 | * rq->lock. | |
1004 | */ | |
1005 | static inline void resched_hrt(struct task_struct *p) | |
1006 | { | |
1007 | __resched_task(p, TIF_HRTICK_RESCHED); | |
1008 | } | |
1009 | ||
1010 | static inline void resched_rq(struct rq *rq) | |
1011 | { | |
1012 | unsigned long flags; | |
1013 | ||
1014 | spin_lock_irqsave(&rq->lock, flags); | |
1015 | resched_task(rq->curr); | |
1016 | spin_unlock_irqrestore(&rq->lock, flags); | |
1017 | } | |
1018 | ||
1019 | enum { | |
1020 | HRTICK_SET, /* re-programm hrtick_timer */ | |
1021 | HRTICK_RESET, /* not a new slice */ | |
1022 | }; | |
1023 | ||
1024 | /* | |
1025 | * Use hrtick when: | |
1026 | * - enabled by features | |
1027 | * - hrtimer is actually high res | |
1028 | */ | |
1029 | static inline int hrtick_enabled(struct rq *rq) | |
1030 | { | |
1031 | if (!sched_feat(HRTICK)) | |
1032 | return 0; | |
1033 | return hrtimer_is_hres_active(&rq->hrtick_timer); | |
1034 | } | |
1035 | ||
1036 | /* | |
1037 | * Called to set the hrtick timer state. | |
1038 | * | |
1039 | * called with rq->lock held and irqs disabled | |
1040 | */ | |
1041 | static void hrtick_start(struct rq *rq, u64 delay, int reset) | |
1042 | { | |
1043 | assert_spin_locked(&rq->lock); | |
1044 | ||
1045 | /* | |
1046 | * preempt at: now + delay | |
1047 | */ | |
1048 | rq->hrtick_expire = | |
1049 | ktime_add_ns(rq->hrtick_timer.base->get_time(), delay); | |
1050 | /* | |
1051 | * indicate we need to program the timer | |
1052 | */ | |
1053 | __set_bit(HRTICK_SET, &rq->hrtick_flags); | |
1054 | if (reset) | |
1055 | __set_bit(HRTICK_RESET, &rq->hrtick_flags); | |
1056 | ||
1057 | /* | |
1058 | * New slices are called from the schedule path and don't need a | |
1059 | * forced reschedule. | |
1060 | */ | |
1061 | if (reset) | |
1062 | resched_hrt(rq->curr); | |
1063 | } | |
1064 | ||
1065 | static void hrtick_clear(struct rq *rq) | |
1066 | { | |
1067 | if (hrtimer_active(&rq->hrtick_timer)) | |
1068 | hrtimer_cancel(&rq->hrtick_timer); | |
1069 | } | |
1070 | ||
1071 | /* | |
1072 | * Update the timer from the possible pending state. | |
1073 | */ | |
1074 | static void hrtick_set(struct rq *rq) | |
1075 | { | |
1076 | ktime_t time; | |
1077 | int set, reset; | |
1078 | unsigned long flags; | |
1079 | ||
1080 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1081 | ||
1082 | spin_lock_irqsave(&rq->lock, flags); | |
1083 | set = __test_and_clear_bit(HRTICK_SET, &rq->hrtick_flags); | |
1084 | reset = __test_and_clear_bit(HRTICK_RESET, &rq->hrtick_flags); | |
1085 | time = rq->hrtick_expire; | |
1086 | clear_thread_flag(TIF_HRTICK_RESCHED); | |
1087 | spin_unlock_irqrestore(&rq->lock, flags); | |
1088 | ||
1089 | if (set) { | |
1090 | hrtimer_start(&rq->hrtick_timer, time, HRTIMER_MODE_ABS); | |
1091 | if (reset && !hrtimer_active(&rq->hrtick_timer)) | |
1092 | resched_rq(rq); | |
1093 | } else | |
1094 | hrtick_clear(rq); | |
1095 | } | |
1096 | ||
1097 | /* | |
1098 | * High-resolution timer tick. | |
1099 | * Runs from hardirq context with interrupts disabled. | |
1100 | */ | |
1101 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1102 | { | |
1103 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1104 | ||
1105 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1106 | ||
1107 | spin_lock(&rq->lock); | |
1108 | __update_rq_clock(rq); | |
1109 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); | |
1110 | spin_unlock(&rq->lock); | |
1111 | ||
1112 | return HRTIMER_NORESTART; | |
1113 | } | |
1114 | ||
1115 | static inline void init_rq_hrtick(struct rq *rq) | |
1116 | { | |
1117 | rq->hrtick_flags = 0; | |
1118 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
1119 | rq->hrtick_timer.function = hrtick; | |
1120 | rq->hrtick_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ; | |
1121 | } | |
1122 | ||
1123 | void hrtick_resched(void) | |
1124 | { | |
1125 | struct rq *rq; | |
1126 | unsigned long flags; | |
1127 | ||
1128 | if (!test_thread_flag(TIF_HRTICK_RESCHED)) | |
1129 | return; | |
1130 | ||
1131 | local_irq_save(flags); | |
1132 | rq = cpu_rq(smp_processor_id()); | |
1133 | hrtick_set(rq); | |
1134 | local_irq_restore(flags); | |
1135 | } | |
1136 | #else | |
1137 | static inline void hrtick_clear(struct rq *rq) | |
1138 | { | |
1139 | } | |
1140 | ||
1141 | static inline void hrtick_set(struct rq *rq) | |
1142 | { | |
1143 | } | |
1144 | ||
1145 | static inline void init_rq_hrtick(struct rq *rq) | |
1146 | { | |
1147 | } | |
1148 | ||
1149 | void hrtick_resched(void) | |
1150 | { | |
1151 | } | |
1152 | #endif | |
1153 | ||
c24d20db IM |
1154 | /* |
1155 | * resched_task - mark a task 'to be rescheduled now'. | |
1156 | * | |
1157 | * On UP this means the setting of the need_resched flag, on SMP it | |
1158 | * might also involve a cross-CPU call to trigger the scheduler on | |
1159 | * the target CPU. | |
1160 | */ | |
1161 | #ifdef CONFIG_SMP | |
1162 | ||
1163 | #ifndef tsk_is_polling | |
1164 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1165 | #endif | |
1166 | ||
8f4d37ec | 1167 | static void __resched_task(struct task_struct *p, int tif_bit) |
c24d20db IM |
1168 | { |
1169 | int cpu; | |
1170 | ||
1171 | assert_spin_locked(&task_rq(p)->lock); | |
1172 | ||
8f4d37ec | 1173 | if (unlikely(test_tsk_thread_flag(p, tif_bit))) |
c24d20db IM |
1174 | return; |
1175 | ||
8f4d37ec | 1176 | set_tsk_thread_flag(p, tif_bit); |
c24d20db IM |
1177 | |
1178 | cpu = task_cpu(p); | |
1179 | if (cpu == smp_processor_id()) | |
1180 | return; | |
1181 | ||
1182 | /* NEED_RESCHED must be visible before we test polling */ | |
1183 | smp_mb(); | |
1184 | if (!tsk_is_polling(p)) | |
1185 | smp_send_reschedule(cpu); | |
1186 | } | |
1187 | ||
1188 | static void resched_cpu(int cpu) | |
1189 | { | |
1190 | struct rq *rq = cpu_rq(cpu); | |
1191 | unsigned long flags; | |
1192 | ||
1193 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1194 | return; | |
1195 | resched_task(cpu_curr(cpu)); | |
1196 | spin_unlock_irqrestore(&rq->lock, flags); | |
1197 | } | |
06d8308c TG |
1198 | |
1199 | #ifdef CONFIG_NO_HZ | |
1200 | /* | |
1201 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1202 | * idle CPU then this timer might expire before the next timer event | |
1203 | * which is scheduled to wake up that CPU. In case of a completely | |
1204 | * idle system the next event might even be infinite time into the | |
1205 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1206 | * leaves the inner idle loop so the newly added timer is taken into | |
1207 | * account when the CPU goes back to idle and evaluates the timer | |
1208 | * wheel for the next timer event. | |
1209 | */ | |
1210 | void wake_up_idle_cpu(int cpu) | |
1211 | { | |
1212 | struct rq *rq = cpu_rq(cpu); | |
1213 | ||
1214 | if (cpu == smp_processor_id()) | |
1215 | return; | |
1216 | ||
1217 | /* | |
1218 | * This is safe, as this function is called with the timer | |
1219 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1220 | * to idle and has not yet set rq->curr to idle then it will | |
1221 | * be serialized on the timer wheel base lock and take the new | |
1222 | * timer into account automatically. | |
1223 | */ | |
1224 | if (rq->curr != rq->idle) | |
1225 | return; | |
1226 | ||
1227 | /* | |
1228 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1229 | * lockless. The worst case is that the other CPU runs the | |
1230 | * idle task through an additional NOOP schedule() | |
1231 | */ | |
1232 | set_tsk_thread_flag(rq->idle, TIF_NEED_RESCHED); | |
1233 | ||
1234 | /* NEED_RESCHED must be visible before we test polling */ | |
1235 | smp_mb(); | |
1236 | if (!tsk_is_polling(rq->idle)) | |
1237 | smp_send_reschedule(cpu); | |
1238 | } | |
1239 | #endif | |
1240 | ||
c24d20db | 1241 | #else |
8f4d37ec | 1242 | static void __resched_task(struct task_struct *p, int tif_bit) |
c24d20db IM |
1243 | { |
1244 | assert_spin_locked(&task_rq(p)->lock); | |
8f4d37ec | 1245 | set_tsk_thread_flag(p, tif_bit); |
c24d20db IM |
1246 | } |
1247 | #endif | |
1248 | ||
45bf76df IM |
1249 | #if BITS_PER_LONG == 32 |
1250 | # define WMULT_CONST (~0UL) | |
1251 | #else | |
1252 | # define WMULT_CONST (1UL << 32) | |
1253 | #endif | |
1254 | ||
1255 | #define WMULT_SHIFT 32 | |
1256 | ||
194081eb IM |
1257 | /* |
1258 | * Shift right and round: | |
1259 | */ | |
cf2ab469 | 1260 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1261 | |
cb1c4fc9 | 1262 | static unsigned long |
45bf76df IM |
1263 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1264 | struct load_weight *lw) | |
1265 | { | |
1266 | u64 tmp; | |
1267 | ||
1268 | if (unlikely(!lw->inv_weight)) | |
27d11726 | 1269 | lw->inv_weight = (WMULT_CONST-lw->weight/2) / (lw->weight+1); |
45bf76df IM |
1270 | |
1271 | tmp = (u64)delta_exec * weight; | |
1272 | /* | |
1273 | * Check whether we'd overflow the 64-bit multiplication: | |
1274 | */ | |
194081eb | 1275 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1276 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1277 | WMULT_SHIFT/2); |
1278 | else | |
cf2ab469 | 1279 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1280 | |
ecf691da | 1281 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1282 | } |
1283 | ||
1284 | static inline unsigned long | |
1285 | calc_delta_fair(unsigned long delta_exec, struct load_weight *lw) | |
1286 | { | |
1287 | return calc_delta_mine(delta_exec, NICE_0_LOAD, lw); | |
1288 | } | |
1289 | ||
1091985b | 1290 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1291 | { |
1292 | lw->weight += inc; | |
e89996ae | 1293 | lw->inv_weight = 0; |
45bf76df IM |
1294 | } |
1295 | ||
1091985b | 1296 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1297 | { |
1298 | lw->weight -= dec; | |
e89996ae | 1299 | lw->inv_weight = 0; |
45bf76df IM |
1300 | } |
1301 | ||
2dd73a4f PW |
1302 | /* |
1303 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1304 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1305 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1306 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1307 | * scaled version of the new time slice allocation that they receive on time |
1308 | * slice expiry etc. | |
1309 | */ | |
1310 | ||
dd41f596 IM |
1311 | #define WEIGHT_IDLEPRIO 2 |
1312 | #define WMULT_IDLEPRIO (1 << 31) | |
1313 | ||
1314 | /* | |
1315 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1316 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1317 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1318 | * that remained on nice 0. | |
1319 | * | |
1320 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1321 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1322 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1323 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1324 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1325 | */ |
1326 | static const int prio_to_weight[40] = { | |
254753dc IM |
1327 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1328 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1329 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1330 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1331 | /* 0 */ 1024, 820, 655, 526, 423, | |
1332 | /* 5 */ 335, 272, 215, 172, 137, | |
1333 | /* 10 */ 110, 87, 70, 56, 45, | |
1334 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1335 | }; |
1336 | ||
5714d2de IM |
1337 | /* |
1338 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1339 | * | |
1340 | * In cases where the weight does not change often, we can use the | |
1341 | * precalculated inverse to speed up arithmetics by turning divisions | |
1342 | * into multiplications: | |
1343 | */ | |
dd41f596 | 1344 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1345 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1346 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1347 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1348 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1349 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1350 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1351 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1352 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1353 | }; |
2dd73a4f | 1354 | |
dd41f596 IM |
1355 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1356 | ||
1357 | /* | |
1358 | * runqueue iterator, to support SMP load-balancing between different | |
1359 | * scheduling classes, without having to expose their internal data | |
1360 | * structures to the load-balancing proper: | |
1361 | */ | |
1362 | struct rq_iterator { | |
1363 | void *arg; | |
1364 | struct task_struct *(*start)(void *); | |
1365 | struct task_struct *(*next)(void *); | |
1366 | }; | |
1367 | ||
e1d1484f PW |
1368 | #ifdef CONFIG_SMP |
1369 | static unsigned long | |
1370 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1371 | unsigned long max_load_move, struct sched_domain *sd, | |
1372 | enum cpu_idle_type idle, int *all_pinned, | |
1373 | int *this_best_prio, struct rq_iterator *iterator); | |
1374 | ||
1375 | static int | |
1376 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1377 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1378 | struct rq_iterator *iterator); | |
e1d1484f | 1379 | #endif |
dd41f596 | 1380 | |
d842de87 SV |
1381 | #ifdef CONFIG_CGROUP_CPUACCT |
1382 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
1383 | #else | |
1384 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
1385 | #endif | |
1386 | ||
e7693a36 GH |
1387 | #ifdef CONFIG_SMP |
1388 | static unsigned long source_load(int cpu, int type); | |
1389 | static unsigned long target_load(int cpu, int type); | |
1390 | static unsigned long cpu_avg_load_per_task(int cpu); | |
1391 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1392 | #endif /* CONFIG_SMP */ | |
1393 | ||
dd41f596 | 1394 | #include "sched_stats.h" |
dd41f596 | 1395 | #include "sched_idletask.c" |
5522d5d5 IM |
1396 | #include "sched_fair.c" |
1397 | #include "sched_rt.c" | |
dd41f596 IM |
1398 | #ifdef CONFIG_SCHED_DEBUG |
1399 | # include "sched_debug.c" | |
1400 | #endif | |
1401 | ||
1402 | #define sched_class_highest (&rt_sched_class) | |
1403 | ||
62fb1851 PZ |
1404 | static inline void inc_load(struct rq *rq, const struct task_struct *p) |
1405 | { | |
1406 | update_load_add(&rq->load, p->se.load.weight); | |
1407 | } | |
1408 | ||
1409 | static inline void dec_load(struct rq *rq, const struct task_struct *p) | |
1410 | { | |
1411 | update_load_sub(&rq->load, p->se.load.weight); | |
1412 | } | |
1413 | ||
1414 | static void inc_nr_running(struct task_struct *p, struct rq *rq) | |
9c217245 IM |
1415 | { |
1416 | rq->nr_running++; | |
62fb1851 | 1417 | inc_load(rq, p); |
9c217245 IM |
1418 | } |
1419 | ||
62fb1851 | 1420 | static void dec_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
1421 | { |
1422 | rq->nr_running--; | |
62fb1851 | 1423 | dec_load(rq, p); |
9c217245 IM |
1424 | } |
1425 | ||
45bf76df IM |
1426 | static void set_load_weight(struct task_struct *p) |
1427 | { | |
1428 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1429 | p->se.load.weight = prio_to_weight[0] * 2; |
1430 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1431 | return; | |
1432 | } | |
45bf76df | 1433 | |
dd41f596 IM |
1434 | /* |
1435 | * SCHED_IDLE tasks get minimal weight: | |
1436 | */ | |
1437 | if (p->policy == SCHED_IDLE) { | |
1438 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1439 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1440 | return; | |
1441 | } | |
71f8bd46 | 1442 | |
dd41f596 IM |
1443 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1444 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1445 | } |
1446 | ||
8159f87e | 1447 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1448 | { |
dd41f596 | 1449 | sched_info_queued(p); |
fd390f6a | 1450 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1451 | p->se.on_rq = 1; |
71f8bd46 IM |
1452 | } |
1453 | ||
69be72c1 | 1454 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1455 | { |
f02231e5 | 1456 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1457 | p->se.on_rq = 0; |
71f8bd46 IM |
1458 | } |
1459 | ||
14531189 | 1460 | /* |
dd41f596 | 1461 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1462 | */ |
14531189 IM |
1463 | static inline int __normal_prio(struct task_struct *p) |
1464 | { | |
dd41f596 | 1465 | return p->static_prio; |
14531189 IM |
1466 | } |
1467 | ||
b29739f9 IM |
1468 | /* |
1469 | * Calculate the expected normal priority: i.e. priority | |
1470 | * without taking RT-inheritance into account. Might be | |
1471 | * boosted by interactivity modifiers. Changes upon fork, | |
1472 | * setprio syscalls, and whenever the interactivity | |
1473 | * estimator recalculates. | |
1474 | */ | |
36c8b586 | 1475 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1476 | { |
1477 | int prio; | |
1478 | ||
e05606d3 | 1479 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1480 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1481 | else | |
1482 | prio = __normal_prio(p); | |
1483 | return prio; | |
1484 | } | |
1485 | ||
1486 | /* | |
1487 | * Calculate the current priority, i.e. the priority | |
1488 | * taken into account by the scheduler. This value might | |
1489 | * be boosted by RT tasks, or might be boosted by | |
1490 | * interactivity modifiers. Will be RT if the task got | |
1491 | * RT-boosted. If not then it returns p->normal_prio. | |
1492 | */ | |
36c8b586 | 1493 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1494 | { |
1495 | p->normal_prio = normal_prio(p); | |
1496 | /* | |
1497 | * If we are RT tasks or we were boosted to RT priority, | |
1498 | * keep the priority unchanged. Otherwise, update priority | |
1499 | * to the normal priority: | |
1500 | */ | |
1501 | if (!rt_prio(p->prio)) | |
1502 | return p->normal_prio; | |
1503 | return p->prio; | |
1504 | } | |
1505 | ||
1da177e4 | 1506 | /* |
dd41f596 | 1507 | * activate_task - move a task to the runqueue. |
1da177e4 | 1508 | */ |
dd41f596 | 1509 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1510 | { |
d9514f6c | 1511 | if (task_contributes_to_load(p)) |
dd41f596 | 1512 | rq->nr_uninterruptible--; |
1da177e4 | 1513 | |
8159f87e | 1514 | enqueue_task(rq, p, wakeup); |
62fb1851 | 1515 | inc_nr_running(p, rq); |
1da177e4 LT |
1516 | } |
1517 | ||
1da177e4 LT |
1518 | /* |
1519 | * deactivate_task - remove a task from the runqueue. | |
1520 | */ | |
2e1cb74a | 1521 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1522 | { |
d9514f6c | 1523 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1524 | rq->nr_uninterruptible++; |
1525 | ||
69be72c1 | 1526 | dequeue_task(rq, p, sleep); |
62fb1851 | 1527 | dec_nr_running(p, rq); |
1da177e4 LT |
1528 | } |
1529 | ||
1da177e4 LT |
1530 | /** |
1531 | * task_curr - is this task currently executing on a CPU? | |
1532 | * @p: the task in question. | |
1533 | */ | |
36c8b586 | 1534 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1535 | { |
1536 | return cpu_curr(task_cpu(p)) == p; | |
1537 | } | |
1538 | ||
2dd73a4f PW |
1539 | /* Used instead of source_load when we know the type == 0 */ |
1540 | unsigned long weighted_cpuload(const int cpu) | |
1541 | { | |
495eca49 | 1542 | return cpu_rq(cpu)->load.weight; |
dd41f596 IM |
1543 | } |
1544 | ||
1545 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1546 | { | |
6f505b16 | 1547 | set_task_rq(p, cpu); |
dd41f596 | 1548 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1549 | /* |
1550 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1551 | * successfuly executed on another CPU. We must ensure that updates of | |
1552 | * per-task data have been completed by this moment. | |
1553 | */ | |
1554 | smp_wmb(); | |
dd41f596 | 1555 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1556 | #endif |
2dd73a4f PW |
1557 | } |
1558 | ||
cb469845 SR |
1559 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1560 | const struct sched_class *prev_class, | |
1561 | int oldprio, int running) | |
1562 | { | |
1563 | if (prev_class != p->sched_class) { | |
1564 | if (prev_class->switched_from) | |
1565 | prev_class->switched_from(rq, p, running); | |
1566 | p->sched_class->switched_to(rq, p, running); | |
1567 | } else | |
1568 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1569 | } | |
1570 | ||
1da177e4 | 1571 | #ifdef CONFIG_SMP |
c65cc870 | 1572 | |
cc367732 IM |
1573 | /* |
1574 | * Is this task likely cache-hot: | |
1575 | */ | |
e7693a36 | 1576 | static int |
cc367732 IM |
1577 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1578 | { | |
1579 | s64 delta; | |
1580 | ||
f540a608 IM |
1581 | /* |
1582 | * Buddy candidates are cache hot: | |
1583 | */ | |
d25ce4cd | 1584 | if (sched_feat(CACHE_HOT_BUDDY) && (&p->se == cfs_rq_of(&p->se)->next)) |
f540a608 IM |
1585 | return 1; |
1586 | ||
cc367732 IM |
1587 | if (p->sched_class != &fair_sched_class) |
1588 | return 0; | |
1589 | ||
6bc1665b IM |
1590 | if (sysctl_sched_migration_cost == -1) |
1591 | return 1; | |
1592 | if (sysctl_sched_migration_cost == 0) | |
1593 | return 0; | |
1594 | ||
cc367732 IM |
1595 | delta = now - p->se.exec_start; |
1596 | ||
1597 | return delta < (s64)sysctl_sched_migration_cost; | |
1598 | } | |
1599 | ||
1600 | ||
dd41f596 | 1601 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1602 | { |
dd41f596 IM |
1603 | int old_cpu = task_cpu(p); |
1604 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1605 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1606 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1607 | u64 clock_offset; |
dd41f596 IM |
1608 | |
1609 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1610 | |
1611 | #ifdef CONFIG_SCHEDSTATS | |
1612 | if (p->se.wait_start) | |
1613 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1614 | if (p->se.sleep_start) |
1615 | p->se.sleep_start -= clock_offset; | |
1616 | if (p->se.block_start) | |
1617 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1618 | if (old_cpu != new_cpu) { |
1619 | schedstat_inc(p, se.nr_migrations); | |
1620 | if (task_hot(p, old_rq->clock, NULL)) | |
1621 | schedstat_inc(p, se.nr_forced2_migrations); | |
1622 | } | |
6cfb0d5d | 1623 | #endif |
2830cf8c SV |
1624 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1625 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1626 | |
1627 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1628 | } |
1629 | ||
70b97a7f | 1630 | struct migration_req { |
1da177e4 | 1631 | struct list_head list; |
1da177e4 | 1632 | |
36c8b586 | 1633 | struct task_struct *task; |
1da177e4 LT |
1634 | int dest_cpu; |
1635 | ||
1da177e4 | 1636 | struct completion done; |
70b97a7f | 1637 | }; |
1da177e4 LT |
1638 | |
1639 | /* | |
1640 | * The task's runqueue lock must be held. | |
1641 | * Returns true if you have to wait for migration thread. | |
1642 | */ | |
36c8b586 | 1643 | static int |
70b97a7f | 1644 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1645 | { |
70b97a7f | 1646 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1647 | |
1648 | /* | |
1649 | * If the task is not on a runqueue (and not running), then | |
1650 | * it is sufficient to simply update the task's cpu field. | |
1651 | */ | |
dd41f596 | 1652 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1653 | set_task_cpu(p, dest_cpu); |
1654 | return 0; | |
1655 | } | |
1656 | ||
1657 | init_completion(&req->done); | |
1da177e4 LT |
1658 | req->task = p; |
1659 | req->dest_cpu = dest_cpu; | |
1660 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1661 | |
1da177e4 LT |
1662 | return 1; |
1663 | } | |
1664 | ||
1665 | /* | |
1666 | * wait_task_inactive - wait for a thread to unschedule. | |
1667 | * | |
1668 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1669 | * else this function might spin for a *long* time. This function can't | |
1670 | * be called with interrupts off, or it may introduce deadlock with | |
1671 | * smp_call_function() if an IPI is sent by the same process we are | |
1672 | * waiting to become inactive. | |
1673 | */ | |
36c8b586 | 1674 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1675 | { |
1676 | unsigned long flags; | |
dd41f596 | 1677 | int running, on_rq; |
70b97a7f | 1678 | struct rq *rq; |
1da177e4 | 1679 | |
3a5c359a AK |
1680 | for (;;) { |
1681 | /* | |
1682 | * We do the initial early heuristics without holding | |
1683 | * any task-queue locks at all. We'll only try to get | |
1684 | * the runqueue lock when things look like they will | |
1685 | * work out! | |
1686 | */ | |
1687 | rq = task_rq(p); | |
fa490cfd | 1688 | |
3a5c359a AK |
1689 | /* |
1690 | * If the task is actively running on another CPU | |
1691 | * still, just relax and busy-wait without holding | |
1692 | * any locks. | |
1693 | * | |
1694 | * NOTE! Since we don't hold any locks, it's not | |
1695 | * even sure that "rq" stays as the right runqueue! | |
1696 | * But we don't care, since "task_running()" will | |
1697 | * return false if the runqueue has changed and p | |
1698 | * is actually now running somewhere else! | |
1699 | */ | |
1700 | while (task_running(rq, p)) | |
1701 | cpu_relax(); | |
fa490cfd | 1702 | |
3a5c359a AK |
1703 | /* |
1704 | * Ok, time to look more closely! We need the rq | |
1705 | * lock now, to be *sure*. If we're wrong, we'll | |
1706 | * just go back and repeat. | |
1707 | */ | |
1708 | rq = task_rq_lock(p, &flags); | |
1709 | running = task_running(rq, p); | |
1710 | on_rq = p->se.on_rq; | |
1711 | task_rq_unlock(rq, &flags); | |
fa490cfd | 1712 | |
3a5c359a AK |
1713 | /* |
1714 | * Was it really running after all now that we | |
1715 | * checked with the proper locks actually held? | |
1716 | * | |
1717 | * Oops. Go back and try again.. | |
1718 | */ | |
1719 | if (unlikely(running)) { | |
1720 | cpu_relax(); | |
1721 | continue; | |
1722 | } | |
fa490cfd | 1723 | |
3a5c359a AK |
1724 | /* |
1725 | * It's not enough that it's not actively running, | |
1726 | * it must be off the runqueue _entirely_, and not | |
1727 | * preempted! | |
1728 | * | |
1729 | * So if it wa still runnable (but just not actively | |
1730 | * running right now), it's preempted, and we should | |
1731 | * yield - it could be a while. | |
1732 | */ | |
1733 | if (unlikely(on_rq)) { | |
1734 | schedule_timeout_uninterruptible(1); | |
1735 | continue; | |
1736 | } | |
fa490cfd | 1737 | |
3a5c359a AK |
1738 | /* |
1739 | * Ahh, all good. It wasn't running, and it wasn't | |
1740 | * runnable, which means that it will never become | |
1741 | * running in the future either. We're all done! | |
1742 | */ | |
1743 | break; | |
1744 | } | |
1da177e4 LT |
1745 | } |
1746 | ||
1747 | /*** | |
1748 | * kick_process - kick a running thread to enter/exit the kernel | |
1749 | * @p: the to-be-kicked thread | |
1750 | * | |
1751 | * Cause a process which is running on another CPU to enter | |
1752 | * kernel-mode, without any delay. (to get signals handled.) | |
1753 | * | |
1754 | * NOTE: this function doesnt have to take the runqueue lock, | |
1755 | * because all it wants to ensure is that the remote task enters | |
1756 | * the kernel. If the IPI races and the task has been migrated | |
1757 | * to another CPU then no harm is done and the purpose has been | |
1758 | * achieved as well. | |
1759 | */ | |
36c8b586 | 1760 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1761 | { |
1762 | int cpu; | |
1763 | ||
1764 | preempt_disable(); | |
1765 | cpu = task_cpu(p); | |
1766 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1767 | smp_send_reschedule(cpu); | |
1768 | preempt_enable(); | |
1769 | } | |
1770 | ||
1771 | /* | |
2dd73a4f PW |
1772 | * Return a low guess at the load of a migration-source cpu weighted |
1773 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1774 | * |
1775 | * We want to under-estimate the load of migration sources, to | |
1776 | * balance conservatively. | |
1777 | */ | |
a9957449 | 1778 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 1779 | { |
70b97a7f | 1780 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1781 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1782 | |
3b0bd9bc | 1783 | if (type == 0) |
dd41f596 | 1784 | return total; |
b910472d | 1785 | |
dd41f596 | 1786 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
1787 | } |
1788 | ||
1789 | /* | |
2dd73a4f PW |
1790 | * Return a high guess at the load of a migration-target cpu weighted |
1791 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1792 | */ |
a9957449 | 1793 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 1794 | { |
70b97a7f | 1795 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1796 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1797 | |
7897986b | 1798 | if (type == 0) |
dd41f596 | 1799 | return total; |
3b0bd9bc | 1800 | |
dd41f596 | 1801 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
1802 | } |
1803 | ||
1804 | /* | |
1805 | * Return the average load per task on the cpu's run queue | |
1806 | */ | |
e7693a36 | 1807 | static unsigned long cpu_avg_load_per_task(int cpu) |
2dd73a4f | 1808 | { |
70b97a7f | 1809 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1810 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f PW |
1811 | unsigned long n = rq->nr_running; |
1812 | ||
dd41f596 | 1813 | return n ? total / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1814 | } |
1815 | ||
147cbb4b NP |
1816 | /* |
1817 | * find_idlest_group finds and returns the least busy CPU group within the | |
1818 | * domain. | |
1819 | */ | |
1820 | static struct sched_group * | |
1821 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1822 | { | |
1823 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1824 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1825 | int load_idx = sd->forkexec_idx; | |
1826 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1827 | ||
1828 | do { | |
1829 | unsigned long load, avg_load; | |
1830 | int local_group; | |
1831 | int i; | |
1832 | ||
da5a5522 BD |
1833 | /* Skip over this group if it has no CPUs allowed */ |
1834 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 1835 | continue; |
da5a5522 | 1836 | |
147cbb4b | 1837 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1838 | |
1839 | /* Tally up the load of all CPUs in the group */ | |
1840 | avg_load = 0; | |
1841 | ||
1842 | for_each_cpu_mask(i, group->cpumask) { | |
1843 | /* Bias balancing toward cpus of our domain */ | |
1844 | if (local_group) | |
1845 | load = source_load(i, load_idx); | |
1846 | else | |
1847 | load = target_load(i, load_idx); | |
1848 | ||
1849 | avg_load += load; | |
1850 | } | |
1851 | ||
1852 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1853 | avg_load = sg_div_cpu_power(group, |
1854 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1855 | |
1856 | if (local_group) { | |
1857 | this_load = avg_load; | |
1858 | this = group; | |
1859 | } else if (avg_load < min_load) { | |
1860 | min_load = avg_load; | |
1861 | idlest = group; | |
1862 | } | |
3a5c359a | 1863 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
1864 | |
1865 | if (!idlest || 100*this_load < imbalance*min_load) | |
1866 | return NULL; | |
1867 | return idlest; | |
1868 | } | |
1869 | ||
1870 | /* | |
0feaece9 | 1871 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1872 | */ |
95cdf3b7 | 1873 | static int |
7c16ec58 MT |
1874 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu, |
1875 | cpumask_t *tmp) | |
147cbb4b NP |
1876 | { |
1877 | unsigned long load, min_load = ULONG_MAX; | |
1878 | int idlest = -1; | |
1879 | int i; | |
1880 | ||
da5a5522 | 1881 | /* Traverse only the allowed CPUs */ |
7c16ec58 | 1882 | cpus_and(*tmp, group->cpumask, p->cpus_allowed); |
da5a5522 | 1883 | |
7c16ec58 | 1884 | for_each_cpu_mask(i, *tmp) { |
2dd73a4f | 1885 | load = weighted_cpuload(i); |
147cbb4b NP |
1886 | |
1887 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1888 | min_load = load; | |
1889 | idlest = i; | |
1890 | } | |
1891 | } | |
1892 | ||
1893 | return idlest; | |
1894 | } | |
1895 | ||
476d139c NP |
1896 | /* |
1897 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1898 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1899 | * SD_BALANCE_EXEC. | |
1900 | * | |
1901 | * Balance, ie. select the least loaded group. | |
1902 | * | |
1903 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1904 | * | |
1905 | * preempt must be disabled. | |
1906 | */ | |
1907 | static int sched_balance_self(int cpu, int flag) | |
1908 | { | |
1909 | struct task_struct *t = current; | |
1910 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1911 | |
c96d145e | 1912 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
1913 | /* |
1914 | * If power savings logic is enabled for a domain, stop there. | |
1915 | */ | |
5c45bf27 SS |
1916 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
1917 | break; | |
476d139c NP |
1918 | if (tmp->flags & flag) |
1919 | sd = tmp; | |
c96d145e | 1920 | } |
476d139c NP |
1921 | |
1922 | while (sd) { | |
7c16ec58 | 1923 | cpumask_t span, tmpmask; |
476d139c | 1924 | struct sched_group *group; |
1a848870 SS |
1925 | int new_cpu, weight; |
1926 | ||
1927 | if (!(sd->flags & flag)) { | |
1928 | sd = sd->child; | |
1929 | continue; | |
1930 | } | |
476d139c NP |
1931 | |
1932 | span = sd->span; | |
1933 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1934 | if (!group) { |
1935 | sd = sd->child; | |
1936 | continue; | |
1937 | } | |
476d139c | 1938 | |
7c16ec58 | 1939 | new_cpu = find_idlest_cpu(group, t, cpu, &tmpmask); |
1a848870 SS |
1940 | if (new_cpu == -1 || new_cpu == cpu) { |
1941 | /* Now try balancing at a lower domain level of cpu */ | |
1942 | sd = sd->child; | |
1943 | continue; | |
1944 | } | |
476d139c | 1945 | |
1a848870 | 1946 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1947 | cpu = new_cpu; |
476d139c NP |
1948 | sd = NULL; |
1949 | weight = cpus_weight(span); | |
1950 | for_each_domain(cpu, tmp) { | |
1951 | if (weight <= cpus_weight(tmp->span)) | |
1952 | break; | |
1953 | if (tmp->flags & flag) | |
1954 | sd = tmp; | |
1955 | } | |
1956 | /* while loop will break here if sd == NULL */ | |
1957 | } | |
1958 | ||
1959 | return cpu; | |
1960 | } | |
1961 | ||
1962 | #endif /* CONFIG_SMP */ | |
1da177e4 | 1963 | |
1da177e4 LT |
1964 | /*** |
1965 | * try_to_wake_up - wake up a thread | |
1966 | * @p: the to-be-woken-up thread | |
1967 | * @state: the mask of task states that can be woken | |
1968 | * @sync: do a synchronous wakeup? | |
1969 | * | |
1970 | * Put it on the run-queue if it's not already there. The "current" | |
1971 | * thread is always on the run-queue (except when the actual | |
1972 | * re-schedule is in progress), and as such you're allowed to do | |
1973 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1974 | * runnable without the overhead of this. | |
1975 | * | |
1976 | * returns failure only if the task is already active. | |
1977 | */ | |
36c8b586 | 1978 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 1979 | { |
cc367732 | 1980 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
1981 | unsigned long flags; |
1982 | long old_state; | |
70b97a7f | 1983 | struct rq *rq; |
1da177e4 | 1984 | |
b85d0667 IM |
1985 | if (!sched_feat(SYNC_WAKEUPS)) |
1986 | sync = 0; | |
1987 | ||
04e2f174 | 1988 | smp_wmb(); |
1da177e4 LT |
1989 | rq = task_rq_lock(p, &flags); |
1990 | old_state = p->state; | |
1991 | if (!(old_state & state)) | |
1992 | goto out; | |
1993 | ||
dd41f596 | 1994 | if (p->se.on_rq) |
1da177e4 LT |
1995 | goto out_running; |
1996 | ||
1997 | cpu = task_cpu(p); | |
cc367732 | 1998 | orig_cpu = cpu; |
1da177e4 LT |
1999 | this_cpu = smp_processor_id(); |
2000 | ||
2001 | #ifdef CONFIG_SMP | |
2002 | if (unlikely(task_running(rq, p))) | |
2003 | goto out_activate; | |
2004 | ||
5d2f5a61 DA |
2005 | cpu = p->sched_class->select_task_rq(p, sync); |
2006 | if (cpu != orig_cpu) { | |
2007 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2008 | task_rq_unlock(rq, &flags); |
2009 | /* might preempt at this point */ | |
2010 | rq = task_rq_lock(p, &flags); | |
2011 | old_state = p->state; | |
2012 | if (!(old_state & state)) | |
2013 | goto out; | |
dd41f596 | 2014 | if (p->se.on_rq) |
1da177e4 LT |
2015 | goto out_running; |
2016 | ||
2017 | this_cpu = smp_processor_id(); | |
2018 | cpu = task_cpu(p); | |
2019 | } | |
2020 | ||
e7693a36 GH |
2021 | #ifdef CONFIG_SCHEDSTATS |
2022 | schedstat_inc(rq, ttwu_count); | |
2023 | if (cpu == this_cpu) | |
2024 | schedstat_inc(rq, ttwu_local); | |
2025 | else { | |
2026 | struct sched_domain *sd; | |
2027 | for_each_domain(this_cpu, sd) { | |
2028 | if (cpu_isset(cpu, sd->span)) { | |
2029 | schedstat_inc(sd, ttwu_wake_remote); | |
2030 | break; | |
2031 | } | |
2032 | } | |
2033 | } | |
e7693a36 GH |
2034 | #endif |
2035 | ||
1da177e4 LT |
2036 | out_activate: |
2037 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2038 | schedstat_inc(p, se.nr_wakeups); |
2039 | if (sync) | |
2040 | schedstat_inc(p, se.nr_wakeups_sync); | |
2041 | if (orig_cpu != cpu) | |
2042 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2043 | if (cpu == this_cpu) | |
2044 | schedstat_inc(p, se.nr_wakeups_local); | |
2045 | else | |
2046 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 2047 | update_rq_clock(rq); |
dd41f596 | 2048 | activate_task(rq, p, 1); |
1da177e4 LT |
2049 | success = 1; |
2050 | ||
2051 | out_running: | |
4ae7d5ce IM |
2052 | check_preempt_curr(rq, p); |
2053 | ||
1da177e4 | 2054 | p->state = TASK_RUNNING; |
9a897c5a SR |
2055 | #ifdef CONFIG_SMP |
2056 | if (p->sched_class->task_wake_up) | |
2057 | p->sched_class->task_wake_up(rq, p); | |
2058 | #endif | |
1da177e4 LT |
2059 | out: |
2060 | task_rq_unlock(rq, &flags); | |
2061 | ||
2062 | return success; | |
2063 | } | |
2064 | ||
7ad5b3a5 | 2065 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2066 | { |
d9514f6c | 2067 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2068 | } |
1da177e4 LT |
2069 | EXPORT_SYMBOL(wake_up_process); |
2070 | ||
7ad5b3a5 | 2071 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2072 | { |
2073 | return try_to_wake_up(p, state, 0); | |
2074 | } | |
2075 | ||
1da177e4 LT |
2076 | /* |
2077 | * Perform scheduler related setup for a newly forked process p. | |
2078 | * p is forked by current. | |
dd41f596 IM |
2079 | * |
2080 | * __sched_fork() is basic setup used by init_idle() too: | |
2081 | */ | |
2082 | static void __sched_fork(struct task_struct *p) | |
2083 | { | |
dd41f596 IM |
2084 | p->se.exec_start = 0; |
2085 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2086 | p->se.prev_sum_exec_runtime = 0; |
4ae7d5ce IM |
2087 | p->se.last_wakeup = 0; |
2088 | p->se.avg_overlap = 0; | |
6cfb0d5d IM |
2089 | |
2090 | #ifdef CONFIG_SCHEDSTATS | |
2091 | p->se.wait_start = 0; | |
dd41f596 IM |
2092 | p->se.sum_sleep_runtime = 0; |
2093 | p->se.sleep_start = 0; | |
dd41f596 IM |
2094 | p->se.block_start = 0; |
2095 | p->se.sleep_max = 0; | |
2096 | p->se.block_max = 0; | |
2097 | p->se.exec_max = 0; | |
eba1ed4b | 2098 | p->se.slice_max = 0; |
dd41f596 | 2099 | p->se.wait_max = 0; |
6cfb0d5d | 2100 | #endif |
476d139c | 2101 | |
fa717060 | 2102 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2103 | p->se.on_rq = 0; |
476d139c | 2104 | |
e107be36 AK |
2105 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2106 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2107 | #endif | |
2108 | ||
1da177e4 LT |
2109 | /* |
2110 | * We mark the process as running here, but have not actually | |
2111 | * inserted it onto the runqueue yet. This guarantees that | |
2112 | * nobody will actually run it, and a signal or other external | |
2113 | * event cannot wake it up and insert it on the runqueue either. | |
2114 | */ | |
2115 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2116 | } |
2117 | ||
2118 | /* | |
2119 | * fork()/clone()-time setup: | |
2120 | */ | |
2121 | void sched_fork(struct task_struct *p, int clone_flags) | |
2122 | { | |
2123 | int cpu = get_cpu(); | |
2124 | ||
2125 | __sched_fork(p); | |
2126 | ||
2127 | #ifdef CONFIG_SMP | |
2128 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2129 | #endif | |
02e4bac2 | 2130 | set_task_cpu(p, cpu); |
b29739f9 IM |
2131 | |
2132 | /* | |
2133 | * Make sure we do not leak PI boosting priority to the child: | |
2134 | */ | |
2135 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2136 | if (!rt_prio(p->prio)) |
2137 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2138 | |
52f17b6c | 2139 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2140 | if (likely(sched_info_on())) |
52f17b6c | 2141 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2142 | #endif |
d6077cb8 | 2143 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2144 | p->oncpu = 0; |
2145 | #endif | |
1da177e4 | 2146 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2147 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2148 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2149 | #endif |
476d139c | 2150 | put_cpu(); |
1da177e4 LT |
2151 | } |
2152 | ||
2153 | /* | |
2154 | * wake_up_new_task - wake up a newly created task for the first time. | |
2155 | * | |
2156 | * This function will do some initial scheduler statistics housekeeping | |
2157 | * that must be done for every newly created context, then puts the task | |
2158 | * on the runqueue and wakes it. | |
2159 | */ | |
7ad5b3a5 | 2160 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2161 | { |
2162 | unsigned long flags; | |
dd41f596 | 2163 | struct rq *rq; |
1da177e4 LT |
2164 | |
2165 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2166 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2167 | update_rq_clock(rq); |
1da177e4 LT |
2168 | |
2169 | p->prio = effective_prio(p); | |
2170 | ||
b9dca1e0 | 2171 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2172 | activate_task(rq, p, 0); |
1da177e4 | 2173 | } else { |
1da177e4 | 2174 | /* |
dd41f596 IM |
2175 | * Let the scheduling class do new task startup |
2176 | * management (if any): | |
1da177e4 | 2177 | */ |
ee0827d8 | 2178 | p->sched_class->task_new(rq, p); |
62fb1851 | 2179 | inc_nr_running(p, rq); |
1da177e4 | 2180 | } |
dd41f596 | 2181 | check_preempt_curr(rq, p); |
9a897c5a SR |
2182 | #ifdef CONFIG_SMP |
2183 | if (p->sched_class->task_wake_up) | |
2184 | p->sched_class->task_wake_up(rq, p); | |
2185 | #endif | |
dd41f596 | 2186 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2187 | } |
2188 | ||
e107be36 AK |
2189 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2190 | ||
2191 | /** | |
421cee29 RD |
2192 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
2193 | * @notifier: notifier struct to register | |
e107be36 AK |
2194 | */ |
2195 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2196 | { | |
2197 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2198 | } | |
2199 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2200 | ||
2201 | /** | |
2202 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2203 | * @notifier: notifier struct to unregister |
e107be36 AK |
2204 | * |
2205 | * This is safe to call from within a preemption notifier. | |
2206 | */ | |
2207 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2208 | { | |
2209 | hlist_del(¬ifier->link); | |
2210 | } | |
2211 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2212 | ||
2213 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2214 | { | |
2215 | struct preempt_notifier *notifier; | |
2216 | struct hlist_node *node; | |
2217 | ||
2218 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2219 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2220 | } | |
2221 | ||
2222 | static void | |
2223 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2224 | struct task_struct *next) | |
2225 | { | |
2226 | struct preempt_notifier *notifier; | |
2227 | struct hlist_node *node; | |
2228 | ||
2229 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2230 | notifier->ops->sched_out(notifier, next); | |
2231 | } | |
2232 | ||
2233 | #else | |
2234 | ||
2235 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2236 | { | |
2237 | } | |
2238 | ||
2239 | static void | |
2240 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2241 | struct task_struct *next) | |
2242 | { | |
2243 | } | |
2244 | ||
2245 | #endif | |
2246 | ||
4866cde0 NP |
2247 | /** |
2248 | * prepare_task_switch - prepare to switch tasks | |
2249 | * @rq: the runqueue preparing to switch | |
421cee29 | 2250 | * @prev: the current task that is being switched out |
4866cde0 NP |
2251 | * @next: the task we are going to switch to. |
2252 | * | |
2253 | * This is called with the rq lock held and interrupts off. It must | |
2254 | * be paired with a subsequent finish_task_switch after the context | |
2255 | * switch. | |
2256 | * | |
2257 | * prepare_task_switch sets up locking and calls architecture specific | |
2258 | * hooks. | |
2259 | */ | |
e107be36 AK |
2260 | static inline void |
2261 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2262 | struct task_struct *next) | |
4866cde0 | 2263 | { |
e107be36 | 2264 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2265 | prepare_lock_switch(rq, next); |
2266 | prepare_arch_switch(next); | |
2267 | } | |
2268 | ||
1da177e4 LT |
2269 | /** |
2270 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2271 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2272 | * @prev: the thread we just switched away from. |
2273 | * | |
4866cde0 NP |
2274 | * finish_task_switch must be called after the context switch, paired |
2275 | * with a prepare_task_switch call before the context switch. | |
2276 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2277 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2278 | * |
2279 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2280 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2281 | * with the lock held can cause deadlocks; see schedule() for |
2282 | * details.) | |
2283 | */ | |
a9957449 | 2284 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2285 | __releases(rq->lock) |
2286 | { | |
1da177e4 | 2287 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2288 | long prev_state; |
1da177e4 LT |
2289 | |
2290 | rq->prev_mm = NULL; | |
2291 | ||
2292 | /* | |
2293 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2294 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2295 | * schedule one last time. The schedule call will never return, and |
2296 | * the scheduled task must drop that reference. | |
c394cc9f | 2297 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2298 | * still held, otherwise prev could be scheduled on another cpu, die |
2299 | * there before we look at prev->state, and then the reference would | |
2300 | * be dropped twice. | |
2301 | * Manfred Spraul <manfred@colorfullife.com> | |
2302 | */ | |
55a101f8 | 2303 | prev_state = prev->state; |
4866cde0 NP |
2304 | finish_arch_switch(prev); |
2305 | finish_lock_switch(rq, prev); | |
9a897c5a SR |
2306 | #ifdef CONFIG_SMP |
2307 | if (current->sched_class->post_schedule) | |
2308 | current->sched_class->post_schedule(rq); | |
2309 | #endif | |
e8fa1362 | 2310 | |
e107be36 | 2311 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2312 | if (mm) |
2313 | mmdrop(mm); | |
c394cc9f | 2314 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2315 | /* |
2316 | * Remove function-return probe instances associated with this | |
2317 | * task and put them back on the free list. | |
9761eea8 | 2318 | */ |
c6fd91f0 | 2319 | kprobe_flush_task(prev); |
1da177e4 | 2320 | put_task_struct(prev); |
c6fd91f0 | 2321 | } |
1da177e4 LT |
2322 | } |
2323 | ||
2324 | /** | |
2325 | * schedule_tail - first thing a freshly forked thread must call. | |
2326 | * @prev: the thread we just switched away from. | |
2327 | */ | |
36c8b586 | 2328 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2329 | __releases(rq->lock) |
2330 | { | |
70b97a7f IM |
2331 | struct rq *rq = this_rq(); |
2332 | ||
4866cde0 NP |
2333 | finish_task_switch(rq, prev); |
2334 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2335 | /* In this case, finish_task_switch does not reenable preemption */ | |
2336 | preempt_enable(); | |
2337 | #endif | |
1da177e4 | 2338 | if (current->set_child_tid) |
b488893a | 2339 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2340 | } |
2341 | ||
2342 | /* | |
2343 | * context_switch - switch to the new MM and the new | |
2344 | * thread's register state. | |
2345 | */ | |
dd41f596 | 2346 | static inline void |
70b97a7f | 2347 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2348 | struct task_struct *next) |
1da177e4 | 2349 | { |
dd41f596 | 2350 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2351 | |
e107be36 | 2352 | prepare_task_switch(rq, prev, next); |
dd41f596 IM |
2353 | mm = next->mm; |
2354 | oldmm = prev->active_mm; | |
9226d125 ZA |
2355 | /* |
2356 | * For paravirt, this is coupled with an exit in switch_to to | |
2357 | * combine the page table reload and the switch backend into | |
2358 | * one hypercall. | |
2359 | */ | |
2360 | arch_enter_lazy_cpu_mode(); | |
2361 | ||
dd41f596 | 2362 | if (unlikely(!mm)) { |
1da177e4 LT |
2363 | next->active_mm = oldmm; |
2364 | atomic_inc(&oldmm->mm_count); | |
2365 | enter_lazy_tlb(oldmm, next); | |
2366 | } else | |
2367 | switch_mm(oldmm, mm, next); | |
2368 | ||
dd41f596 | 2369 | if (unlikely(!prev->mm)) { |
1da177e4 | 2370 | prev->active_mm = NULL; |
1da177e4 LT |
2371 | rq->prev_mm = oldmm; |
2372 | } | |
3a5f5e48 IM |
2373 | /* |
2374 | * Since the runqueue lock will be released by the next | |
2375 | * task (which is an invalid locking op but in the case | |
2376 | * of the scheduler it's an obvious special-case), so we | |
2377 | * do an early lockdep release here: | |
2378 | */ | |
2379 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2380 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2381 | #endif |
1da177e4 LT |
2382 | |
2383 | /* Here we just switch the register state and the stack. */ | |
2384 | switch_to(prev, next, prev); | |
2385 | ||
dd41f596 IM |
2386 | barrier(); |
2387 | /* | |
2388 | * this_rq must be evaluated again because prev may have moved | |
2389 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2390 | * frame will be invalid. | |
2391 | */ | |
2392 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2393 | } |
2394 | ||
2395 | /* | |
2396 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2397 | * | |
2398 | * externally visible scheduler statistics: current number of runnable | |
2399 | * threads, current number of uninterruptible-sleeping threads, total | |
2400 | * number of context switches performed since bootup. | |
2401 | */ | |
2402 | unsigned long nr_running(void) | |
2403 | { | |
2404 | unsigned long i, sum = 0; | |
2405 | ||
2406 | for_each_online_cpu(i) | |
2407 | sum += cpu_rq(i)->nr_running; | |
2408 | ||
2409 | return sum; | |
2410 | } | |
2411 | ||
2412 | unsigned long nr_uninterruptible(void) | |
2413 | { | |
2414 | unsigned long i, sum = 0; | |
2415 | ||
0a945022 | 2416 | for_each_possible_cpu(i) |
1da177e4 LT |
2417 | sum += cpu_rq(i)->nr_uninterruptible; |
2418 | ||
2419 | /* | |
2420 | * Since we read the counters lockless, it might be slightly | |
2421 | * inaccurate. Do not allow it to go below zero though: | |
2422 | */ | |
2423 | if (unlikely((long)sum < 0)) | |
2424 | sum = 0; | |
2425 | ||
2426 | return sum; | |
2427 | } | |
2428 | ||
2429 | unsigned long long nr_context_switches(void) | |
2430 | { | |
cc94abfc SR |
2431 | int i; |
2432 | unsigned long long sum = 0; | |
1da177e4 | 2433 | |
0a945022 | 2434 | for_each_possible_cpu(i) |
1da177e4 LT |
2435 | sum += cpu_rq(i)->nr_switches; |
2436 | ||
2437 | return sum; | |
2438 | } | |
2439 | ||
2440 | unsigned long nr_iowait(void) | |
2441 | { | |
2442 | unsigned long i, sum = 0; | |
2443 | ||
0a945022 | 2444 | for_each_possible_cpu(i) |
1da177e4 LT |
2445 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2446 | ||
2447 | return sum; | |
2448 | } | |
2449 | ||
db1b1fef JS |
2450 | unsigned long nr_active(void) |
2451 | { | |
2452 | unsigned long i, running = 0, uninterruptible = 0; | |
2453 | ||
2454 | for_each_online_cpu(i) { | |
2455 | running += cpu_rq(i)->nr_running; | |
2456 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2457 | } | |
2458 | ||
2459 | if (unlikely((long)uninterruptible < 0)) | |
2460 | uninterruptible = 0; | |
2461 | ||
2462 | return running + uninterruptible; | |
2463 | } | |
2464 | ||
48f24c4d | 2465 | /* |
dd41f596 IM |
2466 | * Update rq->cpu_load[] statistics. This function is usually called every |
2467 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2468 | */ |
dd41f596 | 2469 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2470 | { |
495eca49 | 2471 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2472 | int i, scale; |
2473 | ||
2474 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2475 | |
2476 | /* Update our load: */ | |
2477 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2478 | unsigned long old_load, new_load; | |
2479 | ||
2480 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2481 | ||
2482 | old_load = this_rq->cpu_load[i]; | |
2483 | new_load = this_load; | |
a25707f3 IM |
2484 | /* |
2485 | * Round up the averaging division if load is increasing. This | |
2486 | * prevents us from getting stuck on 9 if the load is 10, for | |
2487 | * example. | |
2488 | */ | |
2489 | if (new_load > old_load) | |
2490 | new_load += scale-1; | |
dd41f596 IM |
2491 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2492 | } | |
48f24c4d IM |
2493 | } |
2494 | ||
dd41f596 IM |
2495 | #ifdef CONFIG_SMP |
2496 | ||
1da177e4 LT |
2497 | /* |
2498 | * double_rq_lock - safely lock two runqueues | |
2499 | * | |
2500 | * Note this does not disable interrupts like task_rq_lock, | |
2501 | * you need to do so manually before calling. | |
2502 | */ | |
70b97a7f | 2503 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2504 | __acquires(rq1->lock) |
2505 | __acquires(rq2->lock) | |
2506 | { | |
054b9108 | 2507 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2508 | if (rq1 == rq2) { |
2509 | spin_lock(&rq1->lock); | |
2510 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2511 | } else { | |
c96d145e | 2512 | if (rq1 < rq2) { |
1da177e4 LT |
2513 | spin_lock(&rq1->lock); |
2514 | spin_lock(&rq2->lock); | |
2515 | } else { | |
2516 | spin_lock(&rq2->lock); | |
2517 | spin_lock(&rq1->lock); | |
2518 | } | |
2519 | } | |
6e82a3be IM |
2520 | update_rq_clock(rq1); |
2521 | update_rq_clock(rq2); | |
1da177e4 LT |
2522 | } |
2523 | ||
2524 | /* | |
2525 | * double_rq_unlock - safely unlock two runqueues | |
2526 | * | |
2527 | * Note this does not restore interrupts like task_rq_unlock, | |
2528 | * you need to do so manually after calling. | |
2529 | */ | |
70b97a7f | 2530 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2531 | __releases(rq1->lock) |
2532 | __releases(rq2->lock) | |
2533 | { | |
2534 | spin_unlock(&rq1->lock); | |
2535 | if (rq1 != rq2) | |
2536 | spin_unlock(&rq2->lock); | |
2537 | else | |
2538 | __release(rq2->lock); | |
2539 | } | |
2540 | ||
2541 | /* | |
2542 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2543 | */ | |
e8fa1362 | 2544 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2545 | __releases(this_rq->lock) |
2546 | __acquires(busiest->lock) | |
2547 | __acquires(this_rq->lock) | |
2548 | { | |
e8fa1362 SR |
2549 | int ret = 0; |
2550 | ||
054b9108 KK |
2551 | if (unlikely(!irqs_disabled())) { |
2552 | /* printk() doesn't work good under rq->lock */ | |
2553 | spin_unlock(&this_rq->lock); | |
2554 | BUG_ON(1); | |
2555 | } | |
1da177e4 | 2556 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2557 | if (busiest < this_rq) { |
1da177e4 LT |
2558 | spin_unlock(&this_rq->lock); |
2559 | spin_lock(&busiest->lock); | |
2560 | spin_lock(&this_rq->lock); | |
e8fa1362 | 2561 | ret = 1; |
1da177e4 LT |
2562 | } else |
2563 | spin_lock(&busiest->lock); | |
2564 | } | |
e8fa1362 | 2565 | return ret; |
1da177e4 LT |
2566 | } |
2567 | ||
1da177e4 LT |
2568 | /* |
2569 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2570 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2571 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2572 | * the cpu_allowed mask is restored. |
2573 | */ | |
36c8b586 | 2574 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2575 | { |
70b97a7f | 2576 | struct migration_req req; |
1da177e4 | 2577 | unsigned long flags; |
70b97a7f | 2578 | struct rq *rq; |
1da177e4 LT |
2579 | |
2580 | rq = task_rq_lock(p, &flags); | |
2581 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2582 | || unlikely(cpu_is_offline(dest_cpu))) | |
2583 | goto out; | |
2584 | ||
2585 | /* force the process onto the specified CPU */ | |
2586 | if (migrate_task(p, dest_cpu, &req)) { | |
2587 | /* Need to wait for migration thread (might exit: take ref). */ | |
2588 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2589 | |
1da177e4 LT |
2590 | get_task_struct(mt); |
2591 | task_rq_unlock(rq, &flags); | |
2592 | wake_up_process(mt); | |
2593 | put_task_struct(mt); | |
2594 | wait_for_completion(&req.done); | |
36c8b586 | 2595 | |
1da177e4 LT |
2596 | return; |
2597 | } | |
2598 | out: | |
2599 | task_rq_unlock(rq, &flags); | |
2600 | } | |
2601 | ||
2602 | /* | |
476d139c NP |
2603 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2604 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2605 | */ |
2606 | void sched_exec(void) | |
2607 | { | |
1da177e4 | 2608 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2609 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2610 | put_cpu(); |
476d139c NP |
2611 | if (new_cpu != this_cpu) |
2612 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2613 | } |
2614 | ||
2615 | /* | |
2616 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2617 | * Both runqueues must be locked. | |
2618 | */ | |
dd41f596 IM |
2619 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2620 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2621 | { |
2e1cb74a | 2622 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2623 | set_task_cpu(p, this_cpu); |
dd41f596 | 2624 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2625 | /* |
2626 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2627 | * to be always true for them. | |
2628 | */ | |
dd41f596 | 2629 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2630 | } |
2631 | ||
2632 | /* | |
2633 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2634 | */ | |
858119e1 | 2635 | static |
70b97a7f | 2636 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2637 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2638 | int *all_pinned) |
1da177e4 LT |
2639 | { |
2640 | /* | |
2641 | * We do not migrate tasks that are: | |
2642 | * 1) running (obviously), or | |
2643 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2644 | * 3) are cache-hot on their current CPU. | |
2645 | */ | |
cc367732 IM |
2646 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2647 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2648 | return 0; |
cc367732 | 2649 | } |
81026794 NP |
2650 | *all_pinned = 0; |
2651 | ||
cc367732 IM |
2652 | if (task_running(rq, p)) { |
2653 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2654 | return 0; |
cc367732 | 2655 | } |
1da177e4 | 2656 | |
da84d961 IM |
2657 | /* |
2658 | * Aggressive migration if: | |
2659 | * 1) task is cache cold, or | |
2660 | * 2) too many balance attempts have failed. | |
2661 | */ | |
2662 | ||
6bc1665b IM |
2663 | if (!task_hot(p, rq->clock, sd) || |
2664 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2665 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2666 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2667 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2668 | schedstat_inc(p, se.nr_forced_migrations); |
2669 | } | |
da84d961 IM |
2670 | #endif |
2671 | return 1; | |
2672 | } | |
2673 | ||
cc367732 IM |
2674 | if (task_hot(p, rq->clock, sd)) { |
2675 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2676 | return 0; |
cc367732 | 2677 | } |
1da177e4 LT |
2678 | return 1; |
2679 | } | |
2680 | ||
e1d1484f PW |
2681 | static unsigned long |
2682 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2683 | unsigned long max_load_move, struct sched_domain *sd, | |
2684 | enum cpu_idle_type idle, int *all_pinned, | |
2685 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2686 | { |
b82d9fdd | 2687 | int loops = 0, pulled = 0, pinned = 0, skip_for_load; |
dd41f596 IM |
2688 | struct task_struct *p; |
2689 | long rem_load_move = max_load_move; | |
1da177e4 | 2690 | |
e1d1484f | 2691 | if (max_load_move == 0) |
1da177e4 LT |
2692 | goto out; |
2693 | ||
81026794 NP |
2694 | pinned = 1; |
2695 | ||
1da177e4 | 2696 | /* |
dd41f596 | 2697 | * Start the load-balancing iterator: |
1da177e4 | 2698 | */ |
dd41f596 IM |
2699 | p = iterator->start(iterator->arg); |
2700 | next: | |
b82d9fdd | 2701 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2702 | goto out; |
50ddd969 | 2703 | /* |
b82d9fdd | 2704 | * To help distribute high priority tasks across CPUs we don't |
50ddd969 PW |
2705 | * skip a task if it will be the highest priority task (i.e. smallest |
2706 | * prio value) on its new queue regardless of its load weight | |
2707 | */ | |
dd41f596 IM |
2708 | skip_for_load = (p->se.load.weight >> 1) > rem_load_move + |
2709 | SCHED_LOAD_SCALE_FUZZ; | |
a4ac01c3 | 2710 | if ((skip_for_load && p->prio >= *this_best_prio) || |
dd41f596 | 2711 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2712 | p = iterator->next(iterator->arg); |
2713 | goto next; | |
1da177e4 LT |
2714 | } |
2715 | ||
dd41f596 | 2716 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2717 | pulled++; |
dd41f596 | 2718 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2719 | |
2dd73a4f | 2720 | /* |
b82d9fdd | 2721 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 2722 | */ |
e1d1484f | 2723 | if (rem_load_move > 0) { |
a4ac01c3 PW |
2724 | if (p->prio < *this_best_prio) |
2725 | *this_best_prio = p->prio; | |
dd41f596 IM |
2726 | p = iterator->next(iterator->arg); |
2727 | goto next; | |
1da177e4 LT |
2728 | } |
2729 | out: | |
2730 | /* | |
e1d1484f | 2731 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
2732 | * so we can safely collect pull_task() stats here rather than |
2733 | * inside pull_task(). | |
2734 | */ | |
2735 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2736 | |
2737 | if (all_pinned) | |
2738 | *all_pinned = pinned; | |
e1d1484f PW |
2739 | |
2740 | return max_load_move - rem_load_move; | |
1da177e4 LT |
2741 | } |
2742 | ||
dd41f596 | 2743 | /* |
43010659 PW |
2744 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2745 | * this_rq, as part of a balancing operation within domain "sd". | |
2746 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2747 | * |
2748 | * Called with both runqueues locked. | |
2749 | */ | |
2750 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2751 | unsigned long max_load_move, |
dd41f596 IM |
2752 | struct sched_domain *sd, enum cpu_idle_type idle, |
2753 | int *all_pinned) | |
2754 | { | |
5522d5d5 | 2755 | const struct sched_class *class = sched_class_highest; |
43010659 | 2756 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2757 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
2758 | |
2759 | do { | |
43010659 PW |
2760 | total_load_moved += |
2761 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 2762 | max_load_move - total_load_moved, |
a4ac01c3 | 2763 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 2764 | class = class->next; |
43010659 | 2765 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 2766 | |
43010659 PW |
2767 | return total_load_moved > 0; |
2768 | } | |
2769 | ||
e1d1484f PW |
2770 | static int |
2771 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2772 | struct sched_domain *sd, enum cpu_idle_type idle, | |
2773 | struct rq_iterator *iterator) | |
2774 | { | |
2775 | struct task_struct *p = iterator->start(iterator->arg); | |
2776 | int pinned = 0; | |
2777 | ||
2778 | while (p) { | |
2779 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
2780 | pull_task(busiest, p, this_rq, this_cpu); | |
2781 | /* | |
2782 | * Right now, this is only the second place pull_task() | |
2783 | * is called, so we can safely collect pull_task() | |
2784 | * stats here rather than inside pull_task(). | |
2785 | */ | |
2786 | schedstat_inc(sd, lb_gained[idle]); | |
2787 | ||
2788 | return 1; | |
2789 | } | |
2790 | p = iterator->next(iterator->arg); | |
2791 | } | |
2792 | ||
2793 | return 0; | |
2794 | } | |
2795 | ||
43010659 PW |
2796 | /* |
2797 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
2798 | * part of active balancing operations within "domain". | |
2799 | * Returns 1 if successful and 0 otherwise. | |
2800 | * | |
2801 | * Called with both runqueues locked. | |
2802 | */ | |
2803 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2804 | struct sched_domain *sd, enum cpu_idle_type idle) | |
2805 | { | |
5522d5d5 | 2806 | const struct sched_class *class; |
43010659 PW |
2807 | |
2808 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 2809 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
2810 | return 1; |
2811 | ||
2812 | return 0; | |
dd41f596 IM |
2813 | } |
2814 | ||
1da177e4 LT |
2815 | /* |
2816 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2817 | * domain. It calculates and returns the amount of weighted load which |
2818 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2819 | */ |
2820 | static struct sched_group * | |
2821 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 | 2822 | unsigned long *imbalance, enum cpu_idle_type idle, |
7c16ec58 | 2823 | int *sd_idle, const cpumask_t *cpus, int *balance) |
1da177e4 LT |
2824 | { |
2825 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2826 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2827 | unsigned long max_pull; |
2dd73a4f PW |
2828 | unsigned long busiest_load_per_task, busiest_nr_running; |
2829 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 2830 | int load_idx, group_imb = 0; |
5c45bf27 SS |
2831 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2832 | int power_savings_balance = 1; | |
2833 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2834 | unsigned long min_nr_running = ULONG_MAX; | |
2835 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2836 | #endif | |
1da177e4 LT |
2837 | |
2838 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2839 | busiest_load_per_task = busiest_nr_running = 0; |
2840 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2841 | if (idle == CPU_NOT_IDLE) |
7897986b | 2842 | load_idx = sd->busy_idx; |
d15bcfdb | 2843 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2844 | load_idx = sd->newidle_idx; |
2845 | else | |
2846 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2847 | |
2848 | do { | |
908a7c1b | 2849 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
2850 | int local_group; |
2851 | int i; | |
908a7c1b | 2852 | int __group_imb = 0; |
783609c6 | 2853 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2854 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2855 | |
2856 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2857 | ||
783609c6 SS |
2858 | if (local_group) |
2859 | balance_cpu = first_cpu(group->cpumask); | |
2860 | ||
1da177e4 | 2861 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2862 | sum_weighted_load = sum_nr_running = avg_load = 0; |
908a7c1b KC |
2863 | max_cpu_load = 0; |
2864 | min_cpu_load = ~0UL; | |
1da177e4 LT |
2865 | |
2866 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2867 | struct rq *rq; |
2868 | ||
2869 | if (!cpu_isset(i, *cpus)) | |
2870 | continue; | |
2871 | ||
2872 | rq = cpu_rq(i); | |
2dd73a4f | 2873 | |
9439aab8 | 2874 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
2875 | *sd_idle = 0; |
2876 | ||
1da177e4 | 2877 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2878 | if (local_group) { |
2879 | if (idle_cpu(i) && !first_idle_cpu) { | |
2880 | first_idle_cpu = 1; | |
2881 | balance_cpu = i; | |
2882 | } | |
2883 | ||
a2000572 | 2884 | load = target_load(i, load_idx); |
908a7c1b | 2885 | } else { |
a2000572 | 2886 | load = source_load(i, load_idx); |
908a7c1b KC |
2887 | if (load > max_cpu_load) |
2888 | max_cpu_load = load; | |
2889 | if (min_cpu_load > load) | |
2890 | min_cpu_load = load; | |
2891 | } | |
1da177e4 LT |
2892 | |
2893 | avg_load += load; | |
2dd73a4f | 2894 | sum_nr_running += rq->nr_running; |
dd41f596 | 2895 | sum_weighted_load += weighted_cpuload(i); |
1da177e4 LT |
2896 | } |
2897 | ||
783609c6 SS |
2898 | /* |
2899 | * First idle cpu or the first cpu(busiest) in this sched group | |
2900 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
2901 | * domains. In the newly idle case, we will allow all the cpu's |
2902 | * to do the newly idle load balance. | |
783609c6 | 2903 | */ |
9439aab8 SS |
2904 | if (idle != CPU_NEWLY_IDLE && local_group && |
2905 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
2906 | *balance = 0; |
2907 | goto ret; | |
2908 | } | |
2909 | ||
1da177e4 | 2910 | total_load += avg_load; |
5517d86b | 2911 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2912 | |
2913 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2914 | avg_load = sg_div_cpu_power(group, |
2915 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2916 | |
908a7c1b KC |
2917 | if ((max_cpu_load - min_cpu_load) > SCHED_LOAD_SCALE) |
2918 | __group_imb = 1; | |
2919 | ||
5517d86b | 2920 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2921 | |
1da177e4 LT |
2922 | if (local_group) { |
2923 | this_load = avg_load; | |
2924 | this = group; | |
2dd73a4f PW |
2925 | this_nr_running = sum_nr_running; |
2926 | this_load_per_task = sum_weighted_load; | |
2927 | } else if (avg_load > max_load && | |
908a7c1b | 2928 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
2929 | max_load = avg_load; |
2930 | busiest = group; | |
2dd73a4f PW |
2931 | busiest_nr_running = sum_nr_running; |
2932 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 2933 | group_imb = __group_imb; |
1da177e4 | 2934 | } |
5c45bf27 SS |
2935 | |
2936 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2937 | /* | |
2938 | * Busy processors will not participate in power savings | |
2939 | * balance. | |
2940 | */ | |
dd41f596 IM |
2941 | if (idle == CPU_NOT_IDLE || |
2942 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
2943 | goto group_next; | |
5c45bf27 SS |
2944 | |
2945 | /* | |
2946 | * If the local group is idle or completely loaded | |
2947 | * no need to do power savings balance at this domain | |
2948 | */ | |
2949 | if (local_group && (this_nr_running >= group_capacity || | |
2950 | !this_nr_running)) | |
2951 | power_savings_balance = 0; | |
2952 | ||
dd41f596 | 2953 | /* |
5c45bf27 SS |
2954 | * If a group is already running at full capacity or idle, |
2955 | * don't include that group in power savings calculations | |
dd41f596 IM |
2956 | */ |
2957 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 2958 | || !sum_nr_running) |
dd41f596 | 2959 | goto group_next; |
5c45bf27 | 2960 | |
dd41f596 | 2961 | /* |
5c45bf27 | 2962 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
2963 | * This is the group from where we need to pick up the load |
2964 | * for saving power | |
2965 | */ | |
2966 | if ((sum_nr_running < min_nr_running) || | |
2967 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
2968 | first_cpu(group->cpumask) < |
2969 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
2970 | group_min = group; |
2971 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
2972 | min_load_per_task = sum_weighted_load / |
2973 | sum_nr_running; | |
dd41f596 | 2974 | } |
5c45bf27 | 2975 | |
dd41f596 | 2976 | /* |
5c45bf27 | 2977 | * Calculate the group which is almost near its |
dd41f596 IM |
2978 | * capacity but still has some space to pick up some load |
2979 | * from other group and save more power | |
2980 | */ | |
2981 | if (sum_nr_running <= group_capacity - 1) { | |
2982 | if (sum_nr_running > leader_nr_running || | |
2983 | (sum_nr_running == leader_nr_running && | |
2984 | first_cpu(group->cpumask) > | |
2985 | first_cpu(group_leader->cpumask))) { | |
2986 | group_leader = group; | |
2987 | leader_nr_running = sum_nr_running; | |
2988 | } | |
48f24c4d | 2989 | } |
5c45bf27 SS |
2990 | group_next: |
2991 | #endif | |
1da177e4 LT |
2992 | group = group->next; |
2993 | } while (group != sd->groups); | |
2994 | ||
2dd73a4f | 2995 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
2996 | goto out_balanced; |
2997 | ||
2998 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
2999 | ||
3000 | if (this_load >= avg_load || | |
3001 | 100*max_load <= sd->imbalance_pct*this_load) | |
3002 | goto out_balanced; | |
3003 | ||
2dd73a4f | 3004 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
3005 | if (group_imb) |
3006 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
3007 | ||
1da177e4 LT |
3008 | /* |
3009 | * We're trying to get all the cpus to the average_load, so we don't | |
3010 | * want to push ourselves above the average load, nor do we wish to | |
3011 | * reduce the max loaded cpu below the average load, as either of these | |
3012 | * actions would just result in more rebalancing later, and ping-pong | |
3013 | * tasks around. Thus we look for the minimum possible imbalance. | |
3014 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3015 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3016 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3017 | * appear as very large values with unsigned longs. |
3018 | */ | |
2dd73a4f PW |
3019 | if (max_load <= busiest_load_per_task) |
3020 | goto out_balanced; | |
3021 | ||
3022 | /* | |
3023 | * In the presence of smp nice balancing, certain scenarios can have | |
3024 | * max load less than avg load(as we skip the groups at or below | |
3025 | * its cpu_power, while calculating max_load..) | |
3026 | */ | |
3027 | if (max_load < avg_load) { | |
3028 | *imbalance = 0; | |
3029 | goto small_imbalance; | |
3030 | } | |
0c117f1b SS |
3031 | |
3032 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 3033 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 3034 | |
1da177e4 | 3035 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
3036 | *imbalance = min(max_pull * busiest->__cpu_power, |
3037 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
3038 | / SCHED_LOAD_SCALE; |
3039 | ||
2dd73a4f PW |
3040 | /* |
3041 | * if *imbalance is less than the average load per runnable task | |
3042 | * there is no gaurantee that any tasks will be moved so we'll have | |
3043 | * a think about bumping its value to force at least one task to be | |
3044 | * moved | |
3045 | */ | |
7fd0d2dd | 3046 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 3047 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
3048 | unsigned int imbn; |
3049 | ||
3050 | small_imbalance: | |
3051 | pwr_move = pwr_now = 0; | |
3052 | imbn = 2; | |
3053 | if (this_nr_running) { | |
3054 | this_load_per_task /= this_nr_running; | |
3055 | if (busiest_load_per_task > this_load_per_task) | |
3056 | imbn = 1; | |
3057 | } else | |
3058 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 3059 | |
dd41f596 IM |
3060 | if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >= |
3061 | busiest_load_per_task * imbn) { | |
2dd73a4f | 3062 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
3063 | return busiest; |
3064 | } | |
3065 | ||
3066 | /* | |
3067 | * OK, we don't have enough imbalance to justify moving tasks, | |
3068 | * however we may be able to increase total CPU power used by | |
3069 | * moving them. | |
3070 | */ | |
3071 | ||
5517d86b ED |
3072 | pwr_now += busiest->__cpu_power * |
3073 | min(busiest_load_per_task, max_load); | |
3074 | pwr_now += this->__cpu_power * | |
3075 | min(this_load_per_task, this_load); | |
1da177e4 LT |
3076 | pwr_now /= SCHED_LOAD_SCALE; |
3077 | ||
3078 | /* Amount of load we'd subtract */ | |
5517d86b ED |
3079 | tmp = sg_div_cpu_power(busiest, |
3080 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 3081 | if (max_load > tmp) |
5517d86b | 3082 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 3083 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
3084 | |
3085 | /* Amount of load we'd add */ | |
5517d86b | 3086 | if (max_load * busiest->__cpu_power < |
33859f7f | 3087 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
3088 | tmp = sg_div_cpu_power(this, |
3089 | max_load * busiest->__cpu_power); | |
1da177e4 | 3090 | else |
5517d86b ED |
3091 | tmp = sg_div_cpu_power(this, |
3092 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
3093 | pwr_move += this->__cpu_power * | |
3094 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
3095 | pwr_move /= SCHED_LOAD_SCALE; |
3096 | ||
3097 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
3098 | if (pwr_move > pwr_now) |
3099 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
3100 | } |
3101 | ||
1da177e4 LT |
3102 | return busiest; |
3103 | ||
3104 | out_balanced: | |
5c45bf27 | 3105 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 3106 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 3107 | goto ret; |
1da177e4 | 3108 | |
5c45bf27 SS |
3109 | if (this == group_leader && group_leader != group_min) { |
3110 | *imbalance = min_load_per_task; | |
3111 | return group_min; | |
3112 | } | |
5c45bf27 | 3113 | #endif |
783609c6 | 3114 | ret: |
1da177e4 LT |
3115 | *imbalance = 0; |
3116 | return NULL; | |
3117 | } | |
3118 | ||
3119 | /* | |
3120 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3121 | */ | |
70b97a7f | 3122 | static struct rq * |
d15bcfdb | 3123 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
7c16ec58 | 3124 | unsigned long imbalance, const cpumask_t *cpus) |
1da177e4 | 3125 | { |
70b97a7f | 3126 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3127 | unsigned long max_load = 0; |
1da177e4 LT |
3128 | int i; |
3129 | ||
3130 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 3131 | unsigned long wl; |
0a2966b4 CL |
3132 | |
3133 | if (!cpu_isset(i, *cpus)) | |
3134 | continue; | |
3135 | ||
48f24c4d | 3136 | rq = cpu_rq(i); |
dd41f596 | 3137 | wl = weighted_cpuload(i); |
2dd73a4f | 3138 | |
dd41f596 | 3139 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3140 | continue; |
1da177e4 | 3141 | |
dd41f596 IM |
3142 | if (wl > max_load) { |
3143 | max_load = wl; | |
48f24c4d | 3144 | busiest = rq; |
1da177e4 LT |
3145 | } |
3146 | } | |
3147 | ||
3148 | return busiest; | |
3149 | } | |
3150 | ||
77391d71 NP |
3151 | /* |
3152 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3153 | * so long as it is large enough. | |
3154 | */ | |
3155 | #define MAX_PINNED_INTERVAL 512 | |
3156 | ||
1da177e4 LT |
3157 | /* |
3158 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3159 | * tasks if there is an imbalance. | |
1da177e4 | 3160 | */ |
70b97a7f | 3161 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3162 | struct sched_domain *sd, enum cpu_idle_type idle, |
7c16ec58 | 3163 | int *balance, cpumask_t *cpus) |
1da177e4 | 3164 | { |
43010659 | 3165 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3166 | struct sched_group *group; |
1da177e4 | 3167 | unsigned long imbalance; |
70b97a7f | 3168 | struct rq *busiest; |
fe2eea3f | 3169 | unsigned long flags; |
5969fe06 | 3170 | |
7c16ec58 MT |
3171 | cpus_setall(*cpus); |
3172 | ||
89c4710e SS |
3173 | /* |
3174 | * When power savings policy is enabled for the parent domain, idle | |
3175 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3176 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3177 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3178 | */ |
d15bcfdb | 3179 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3180 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3181 | sd_idle = 1; |
1da177e4 | 3182 | |
2d72376b | 3183 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3184 | |
0a2966b4 CL |
3185 | redo: |
3186 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
7c16ec58 | 3187 | cpus, balance); |
783609c6 | 3188 | |
06066714 | 3189 | if (*balance == 0) |
783609c6 | 3190 | goto out_balanced; |
783609c6 | 3191 | |
1da177e4 LT |
3192 | if (!group) { |
3193 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3194 | goto out_balanced; | |
3195 | } | |
3196 | ||
7c16ec58 | 3197 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
3198 | if (!busiest) { |
3199 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3200 | goto out_balanced; | |
3201 | } | |
3202 | ||
db935dbd | 3203 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3204 | |
3205 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3206 | ||
43010659 | 3207 | ld_moved = 0; |
1da177e4 LT |
3208 | if (busiest->nr_running > 1) { |
3209 | /* | |
3210 | * Attempt to move tasks. If find_busiest_group has found | |
3211 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3212 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3213 | * correctly treated as an imbalance. |
3214 | */ | |
fe2eea3f | 3215 | local_irq_save(flags); |
e17224bf | 3216 | double_rq_lock(this_rq, busiest); |
43010659 | 3217 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3218 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3219 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3220 | local_irq_restore(flags); |
81026794 | 3221 | |
46cb4b7c SS |
3222 | /* |
3223 | * some other cpu did the load balance for us. | |
3224 | */ | |
43010659 | 3225 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3226 | resched_cpu(this_cpu); |
3227 | ||
81026794 | 3228 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 3229 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3230 | cpu_clear(cpu_of(busiest), *cpus); |
3231 | if (!cpus_empty(*cpus)) | |
0a2966b4 | 3232 | goto redo; |
81026794 | 3233 | goto out_balanced; |
0a2966b4 | 3234 | } |
1da177e4 | 3235 | } |
81026794 | 3236 | |
43010659 | 3237 | if (!ld_moved) { |
1da177e4 LT |
3238 | schedstat_inc(sd, lb_failed[idle]); |
3239 | sd->nr_balance_failed++; | |
3240 | ||
3241 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3242 | |
fe2eea3f | 3243 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3244 | |
3245 | /* don't kick the migration_thread, if the curr | |
3246 | * task on busiest cpu can't be moved to this_cpu | |
3247 | */ | |
3248 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3249 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3250 | all_pinned = 1; |
3251 | goto out_one_pinned; | |
3252 | } | |
3253 | ||
1da177e4 LT |
3254 | if (!busiest->active_balance) { |
3255 | busiest->active_balance = 1; | |
3256 | busiest->push_cpu = this_cpu; | |
81026794 | 3257 | active_balance = 1; |
1da177e4 | 3258 | } |
fe2eea3f | 3259 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3260 | if (active_balance) |
1da177e4 LT |
3261 | wake_up_process(busiest->migration_thread); |
3262 | ||
3263 | /* | |
3264 | * We've kicked active balancing, reset the failure | |
3265 | * counter. | |
3266 | */ | |
39507451 | 3267 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3268 | } |
81026794 | 3269 | } else |
1da177e4 LT |
3270 | sd->nr_balance_failed = 0; |
3271 | ||
81026794 | 3272 | if (likely(!active_balance)) { |
1da177e4 LT |
3273 | /* We were unbalanced, so reset the balancing interval */ |
3274 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
3275 | } else { |
3276 | /* | |
3277 | * If we've begun active balancing, start to back off. This | |
3278 | * case may not be covered by the all_pinned logic if there | |
3279 | * is only 1 task on the busy runqueue (because we don't call | |
3280 | * move_tasks). | |
3281 | */ | |
3282 | if (sd->balance_interval < sd->max_interval) | |
3283 | sd->balance_interval *= 2; | |
1da177e4 LT |
3284 | } |
3285 | ||
43010659 | 3286 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3287 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3288 | return -1; |
43010659 | 3289 | return ld_moved; |
1da177e4 LT |
3290 | |
3291 | out_balanced: | |
1da177e4 LT |
3292 | schedstat_inc(sd, lb_balanced[idle]); |
3293 | ||
16cfb1c0 | 3294 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
3295 | |
3296 | out_one_pinned: | |
1da177e4 | 3297 | /* tune up the balancing interval */ |
77391d71 NP |
3298 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
3299 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
3300 | sd->balance_interval *= 2; |
3301 | ||
48f24c4d | 3302 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3303 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3304 | return -1; |
1da177e4 LT |
3305 | return 0; |
3306 | } | |
3307 | ||
3308 | /* | |
3309 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3310 | * tasks if there is an imbalance. | |
3311 | * | |
d15bcfdb | 3312 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
3313 | * this_rq is locked. |
3314 | */ | |
48f24c4d | 3315 | static int |
7c16ec58 MT |
3316 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, |
3317 | cpumask_t *cpus) | |
1da177e4 LT |
3318 | { |
3319 | struct sched_group *group; | |
70b97a7f | 3320 | struct rq *busiest = NULL; |
1da177e4 | 3321 | unsigned long imbalance; |
43010659 | 3322 | int ld_moved = 0; |
5969fe06 | 3323 | int sd_idle = 0; |
969bb4e4 | 3324 | int all_pinned = 0; |
7c16ec58 MT |
3325 | |
3326 | cpus_setall(*cpus); | |
5969fe06 | 3327 | |
89c4710e SS |
3328 | /* |
3329 | * When power savings policy is enabled for the parent domain, idle | |
3330 | * sibling can pick up load irrespective of busy siblings. In this case, | |
3331 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 3332 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
3333 | */ |
3334 | if (sd->flags & SD_SHARE_CPUPOWER && | |
3335 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3336 | sd_idle = 1; |
1da177e4 | 3337 | |
2d72376b | 3338 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 3339 | redo: |
d15bcfdb | 3340 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 3341 | &sd_idle, cpus, NULL); |
1da177e4 | 3342 | if (!group) { |
d15bcfdb | 3343 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3344 | goto out_balanced; |
1da177e4 LT |
3345 | } |
3346 | ||
7c16ec58 | 3347 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 3348 | if (!busiest) { |
d15bcfdb | 3349 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3350 | goto out_balanced; |
1da177e4 LT |
3351 | } |
3352 | ||
db935dbd NP |
3353 | BUG_ON(busiest == this_rq); |
3354 | ||
d15bcfdb | 3355 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 3356 | |
43010659 | 3357 | ld_moved = 0; |
d6d5cfaf NP |
3358 | if (busiest->nr_running > 1) { |
3359 | /* Attempt to move tasks */ | |
3360 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
3361 | /* this_rq->clock is already updated */ |
3362 | update_rq_clock(busiest); | |
43010659 | 3363 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
3364 | imbalance, sd, CPU_NEWLY_IDLE, |
3365 | &all_pinned); | |
d6d5cfaf | 3366 | spin_unlock(&busiest->lock); |
0a2966b4 | 3367 | |
969bb4e4 | 3368 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3369 | cpu_clear(cpu_of(busiest), *cpus); |
3370 | if (!cpus_empty(*cpus)) | |
0a2966b4 CL |
3371 | goto redo; |
3372 | } | |
d6d5cfaf NP |
3373 | } |
3374 | ||
43010659 | 3375 | if (!ld_moved) { |
d15bcfdb | 3376 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
3377 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
3378 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
3379 | return -1; |
3380 | } else | |
16cfb1c0 | 3381 | sd->nr_balance_failed = 0; |
1da177e4 | 3382 | |
43010659 | 3383 | return ld_moved; |
16cfb1c0 NP |
3384 | |
3385 | out_balanced: | |
d15bcfdb | 3386 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 3387 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3388 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3389 | return -1; |
16cfb1c0 | 3390 | sd->nr_balance_failed = 0; |
48f24c4d | 3391 | |
16cfb1c0 | 3392 | return 0; |
1da177e4 LT |
3393 | } |
3394 | ||
3395 | /* | |
3396 | * idle_balance is called by schedule() if this_cpu is about to become | |
3397 | * idle. Attempts to pull tasks from other CPUs. | |
3398 | */ | |
70b97a7f | 3399 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
3400 | { |
3401 | struct sched_domain *sd; | |
dd41f596 IM |
3402 | int pulled_task = -1; |
3403 | unsigned long next_balance = jiffies + HZ; | |
7c16ec58 | 3404 | cpumask_t tmpmask; |
1da177e4 LT |
3405 | |
3406 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
3407 | unsigned long interval; |
3408 | ||
3409 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
3410 | continue; | |
3411 | ||
3412 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 3413 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 MT |
3414 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
3415 | sd, &tmpmask); | |
92c4ca5c CL |
3416 | |
3417 | interval = msecs_to_jiffies(sd->balance_interval); | |
3418 | if (time_after(next_balance, sd->last_balance + interval)) | |
3419 | next_balance = sd->last_balance + interval; | |
3420 | if (pulled_task) | |
3421 | break; | |
1da177e4 | 3422 | } |
dd41f596 | 3423 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
3424 | /* |
3425 | * We are going idle. next_balance may be set based on | |
3426 | * a busy processor. So reset next_balance. | |
3427 | */ | |
3428 | this_rq->next_balance = next_balance; | |
dd41f596 | 3429 | } |
1da177e4 LT |
3430 | } |
3431 | ||
3432 | /* | |
3433 | * active_load_balance is run by migration threads. It pushes running tasks | |
3434 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
3435 | * running on each physical CPU where possible, and avoids physical / | |
3436 | * logical imbalances. | |
3437 | * | |
3438 | * Called with busiest_rq locked. | |
3439 | */ | |
70b97a7f | 3440 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3441 | { |
39507451 | 3442 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3443 | struct sched_domain *sd; |
3444 | struct rq *target_rq; | |
39507451 | 3445 | |
48f24c4d | 3446 | /* Is there any task to move? */ |
39507451 | 3447 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3448 | return; |
3449 | ||
3450 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3451 | |
3452 | /* | |
39507451 | 3453 | * This condition is "impossible", if it occurs |
41a2d6cf | 3454 | * we need to fix it. Originally reported by |
39507451 | 3455 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3456 | */ |
39507451 | 3457 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3458 | |
39507451 NP |
3459 | /* move a task from busiest_rq to target_rq */ |
3460 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3461 | update_rq_clock(busiest_rq); |
3462 | update_rq_clock(target_rq); | |
39507451 NP |
3463 | |
3464 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3465 | for_each_domain(target_cpu, sd) { |
39507451 | 3466 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 3467 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 3468 | break; |
c96d145e | 3469 | } |
39507451 | 3470 | |
48f24c4d | 3471 | if (likely(sd)) { |
2d72376b | 3472 | schedstat_inc(sd, alb_count); |
39507451 | 3473 | |
43010659 PW |
3474 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3475 | sd, CPU_IDLE)) | |
48f24c4d IM |
3476 | schedstat_inc(sd, alb_pushed); |
3477 | else | |
3478 | schedstat_inc(sd, alb_failed); | |
3479 | } | |
39507451 | 3480 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
3481 | } |
3482 | ||
46cb4b7c SS |
3483 | #ifdef CONFIG_NO_HZ |
3484 | static struct { | |
3485 | atomic_t load_balancer; | |
41a2d6cf | 3486 | cpumask_t cpu_mask; |
46cb4b7c SS |
3487 | } nohz ____cacheline_aligned = { |
3488 | .load_balancer = ATOMIC_INIT(-1), | |
3489 | .cpu_mask = CPU_MASK_NONE, | |
3490 | }; | |
3491 | ||
7835b98b | 3492 | /* |
46cb4b7c SS |
3493 | * This routine will try to nominate the ilb (idle load balancing) |
3494 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3495 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3496 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3497 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3498 | * arrives... | |
3499 | * | |
3500 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3501 | * for idle load balancing. ilb owner will still be part of | |
3502 | * nohz.cpu_mask.. | |
7835b98b | 3503 | * |
46cb4b7c SS |
3504 | * While stopping the tick, this cpu will become the ilb owner if there |
3505 | * is no other owner. And will be the owner till that cpu becomes busy | |
3506 | * or if all cpus in the system stop their ticks at which point | |
3507 | * there is no need for ilb owner. | |
3508 | * | |
3509 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3510 | * next busy scheduler_tick() | |
3511 | */ | |
3512 | int select_nohz_load_balancer(int stop_tick) | |
3513 | { | |
3514 | int cpu = smp_processor_id(); | |
3515 | ||
3516 | if (stop_tick) { | |
3517 | cpu_set(cpu, nohz.cpu_mask); | |
3518 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3519 | ||
3520 | /* | |
3521 | * If we are going offline and still the leader, give up! | |
3522 | */ | |
3523 | if (cpu_is_offline(cpu) && | |
3524 | atomic_read(&nohz.load_balancer) == cpu) { | |
3525 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3526 | BUG(); | |
3527 | return 0; | |
3528 | } | |
3529 | ||
3530 | /* time for ilb owner also to sleep */ | |
3531 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3532 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3533 | atomic_set(&nohz.load_balancer, -1); | |
3534 | return 0; | |
3535 | } | |
3536 | ||
3537 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3538 | /* make me the ilb owner */ | |
3539 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3540 | return 1; | |
3541 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3542 | return 1; | |
3543 | } else { | |
3544 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3545 | return 0; | |
3546 | ||
3547 | cpu_clear(cpu, nohz.cpu_mask); | |
3548 | ||
3549 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3550 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3551 | BUG(); | |
3552 | } | |
3553 | return 0; | |
3554 | } | |
3555 | #endif | |
3556 | ||
3557 | static DEFINE_SPINLOCK(balancing); | |
3558 | ||
3559 | /* | |
7835b98b CL |
3560 | * It checks each scheduling domain to see if it is due to be balanced, |
3561 | * and initiates a balancing operation if so. | |
3562 | * | |
3563 | * Balancing parameters are set up in arch_init_sched_domains. | |
3564 | */ | |
a9957449 | 3565 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3566 | { |
46cb4b7c SS |
3567 | int balance = 1; |
3568 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3569 | unsigned long interval; |
3570 | struct sched_domain *sd; | |
46cb4b7c | 3571 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3572 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3573 | int update_next_balance = 0; |
7c16ec58 | 3574 | cpumask_t tmp; |
1da177e4 | 3575 | |
46cb4b7c | 3576 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3577 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3578 | continue; | |
3579 | ||
3580 | interval = sd->balance_interval; | |
d15bcfdb | 3581 | if (idle != CPU_IDLE) |
1da177e4 LT |
3582 | interval *= sd->busy_factor; |
3583 | ||
3584 | /* scale ms to jiffies */ | |
3585 | interval = msecs_to_jiffies(interval); | |
3586 | if (unlikely(!interval)) | |
3587 | interval = 1; | |
dd41f596 IM |
3588 | if (interval > HZ*NR_CPUS/10) |
3589 | interval = HZ*NR_CPUS/10; | |
3590 | ||
1da177e4 | 3591 | |
08c183f3 CL |
3592 | if (sd->flags & SD_SERIALIZE) { |
3593 | if (!spin_trylock(&balancing)) | |
3594 | goto out; | |
3595 | } | |
3596 | ||
c9819f45 | 3597 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
7c16ec58 | 3598 | if (load_balance(cpu, rq, sd, idle, &balance, &tmp)) { |
fa3b6ddc SS |
3599 | /* |
3600 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3601 | * longer idle, or one of our SMT siblings is |
3602 | * not idle. | |
3603 | */ | |
d15bcfdb | 3604 | idle = CPU_NOT_IDLE; |
1da177e4 | 3605 | } |
1bd77f2d | 3606 | sd->last_balance = jiffies; |
1da177e4 | 3607 | } |
08c183f3 CL |
3608 | if (sd->flags & SD_SERIALIZE) |
3609 | spin_unlock(&balancing); | |
3610 | out: | |
f549da84 | 3611 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3612 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3613 | update_next_balance = 1; |
3614 | } | |
783609c6 SS |
3615 | |
3616 | /* | |
3617 | * Stop the load balance at this level. There is another | |
3618 | * CPU in our sched group which is doing load balancing more | |
3619 | * actively. | |
3620 | */ | |
3621 | if (!balance) | |
3622 | break; | |
1da177e4 | 3623 | } |
f549da84 SS |
3624 | |
3625 | /* | |
3626 | * next_balance will be updated only when there is a need. | |
3627 | * When the cpu is attached to null domain for ex, it will not be | |
3628 | * updated. | |
3629 | */ | |
3630 | if (likely(update_next_balance)) | |
3631 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3632 | } |
3633 | ||
3634 | /* | |
3635 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3636 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3637 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3638 | */ | |
3639 | static void run_rebalance_domains(struct softirq_action *h) | |
3640 | { | |
dd41f596 IM |
3641 | int this_cpu = smp_processor_id(); |
3642 | struct rq *this_rq = cpu_rq(this_cpu); | |
3643 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3644 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3645 | |
dd41f596 | 3646 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3647 | |
3648 | #ifdef CONFIG_NO_HZ | |
3649 | /* | |
3650 | * If this cpu is the owner for idle load balancing, then do the | |
3651 | * balancing on behalf of the other idle cpus whose ticks are | |
3652 | * stopped. | |
3653 | */ | |
dd41f596 IM |
3654 | if (this_rq->idle_at_tick && |
3655 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3656 | cpumask_t cpus = nohz.cpu_mask; |
3657 | struct rq *rq; | |
3658 | int balance_cpu; | |
3659 | ||
dd41f596 | 3660 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3661 | for_each_cpu_mask(balance_cpu, cpus) { |
3662 | /* | |
3663 | * If this cpu gets work to do, stop the load balancing | |
3664 | * work being done for other cpus. Next load | |
3665 | * balancing owner will pick it up. | |
3666 | */ | |
3667 | if (need_resched()) | |
3668 | break; | |
3669 | ||
de0cf899 | 3670 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3671 | |
3672 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3673 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3674 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3675 | } |
3676 | } | |
3677 | #endif | |
3678 | } | |
3679 | ||
3680 | /* | |
3681 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3682 | * | |
3683 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3684 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3685 | * if the whole system is idle. | |
3686 | */ | |
dd41f596 | 3687 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3688 | { |
46cb4b7c SS |
3689 | #ifdef CONFIG_NO_HZ |
3690 | /* | |
3691 | * If we were in the nohz mode recently and busy at the current | |
3692 | * scheduler tick, then check if we need to nominate new idle | |
3693 | * load balancer. | |
3694 | */ | |
3695 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3696 | rq->in_nohz_recently = 0; | |
3697 | ||
3698 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3699 | cpu_clear(cpu, nohz.cpu_mask); | |
3700 | atomic_set(&nohz.load_balancer, -1); | |
3701 | } | |
3702 | ||
3703 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3704 | /* | |
3705 | * simple selection for now: Nominate the | |
3706 | * first cpu in the nohz list to be the next | |
3707 | * ilb owner. | |
3708 | * | |
3709 | * TBD: Traverse the sched domains and nominate | |
3710 | * the nearest cpu in the nohz.cpu_mask. | |
3711 | */ | |
3712 | int ilb = first_cpu(nohz.cpu_mask); | |
3713 | ||
434d53b0 | 3714 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
3715 | resched_cpu(ilb); |
3716 | } | |
3717 | } | |
3718 | ||
3719 | /* | |
3720 | * If this cpu is idle and doing idle load balancing for all the | |
3721 | * cpus with ticks stopped, is it time for that to stop? | |
3722 | */ | |
3723 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3724 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3725 | resched_cpu(cpu); | |
3726 | return; | |
3727 | } | |
3728 | ||
3729 | /* | |
3730 | * If this cpu is idle and the idle load balancing is done by | |
3731 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3732 | */ | |
3733 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3734 | cpu_isset(cpu, nohz.cpu_mask)) | |
3735 | return; | |
3736 | #endif | |
3737 | if (time_after_eq(jiffies, rq->next_balance)) | |
3738 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3739 | } |
dd41f596 IM |
3740 | |
3741 | #else /* CONFIG_SMP */ | |
3742 | ||
1da177e4 LT |
3743 | /* |
3744 | * on UP we do not need to balance between CPUs: | |
3745 | */ | |
70b97a7f | 3746 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3747 | { |
3748 | } | |
dd41f596 | 3749 | |
1da177e4 LT |
3750 | #endif |
3751 | ||
1da177e4 LT |
3752 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3753 | ||
3754 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3755 | ||
3756 | /* | |
41b86e9c IM |
3757 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
3758 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 3759 | */ |
41b86e9c | 3760 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 3761 | { |
1da177e4 | 3762 | unsigned long flags; |
41b86e9c IM |
3763 | u64 ns, delta_exec; |
3764 | struct rq *rq; | |
48f24c4d | 3765 | |
41b86e9c IM |
3766 | rq = task_rq_lock(p, &flags); |
3767 | ns = p->se.sum_exec_runtime; | |
051a1d1a | 3768 | if (task_current(rq, p)) { |
a8e504d2 IM |
3769 | update_rq_clock(rq); |
3770 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
3771 | if ((s64)delta_exec > 0) |
3772 | ns += delta_exec; | |
3773 | } | |
3774 | task_rq_unlock(rq, &flags); | |
48f24c4d | 3775 | |
1da177e4 LT |
3776 | return ns; |
3777 | } | |
3778 | ||
1da177e4 LT |
3779 | /* |
3780 | * Account user cpu time to a process. | |
3781 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
3782 | * @cputime: the cpu time spent in user space since the last update |
3783 | */ | |
3784 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3785 | { | |
3786 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3787 | cputime64_t tmp; | |
3788 | ||
3789 | p->utime = cputime_add(p->utime, cputime); | |
3790 | ||
3791 | /* Add user time to cpustat. */ | |
3792 | tmp = cputime_to_cputime64(cputime); | |
3793 | if (TASK_NICE(p) > 0) | |
3794 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3795 | else | |
3796 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3797 | } | |
3798 | ||
94886b84 LV |
3799 | /* |
3800 | * Account guest cpu time to a process. | |
3801 | * @p: the process that the cpu time gets accounted to | |
3802 | * @cputime: the cpu time spent in virtual machine since the last update | |
3803 | */ | |
f7402e03 | 3804 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
3805 | { |
3806 | cputime64_t tmp; | |
3807 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3808 | ||
3809 | tmp = cputime_to_cputime64(cputime); | |
3810 | ||
3811 | p->utime = cputime_add(p->utime, cputime); | |
3812 | p->gtime = cputime_add(p->gtime, cputime); | |
3813 | ||
3814 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3815 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3816 | } | |
3817 | ||
c66f08be MN |
3818 | /* |
3819 | * Account scaled user cpu time to a process. | |
3820 | * @p: the process that the cpu time gets accounted to | |
3821 | * @cputime: the cpu time spent in user space since the last update | |
3822 | */ | |
3823 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
3824 | { | |
3825 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
3826 | } | |
3827 | ||
1da177e4 LT |
3828 | /* |
3829 | * Account system cpu time to a process. | |
3830 | * @p: the process that the cpu time gets accounted to | |
3831 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3832 | * @cputime: the cpu time spent in kernel space since the last update | |
3833 | */ | |
3834 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3835 | cputime_t cputime) | |
3836 | { | |
3837 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3838 | struct rq *rq = this_rq(); |
1da177e4 LT |
3839 | cputime64_t tmp; |
3840 | ||
9778385d CB |
3841 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) |
3842 | return account_guest_time(p, cputime); | |
94886b84 | 3843 | |
1da177e4 LT |
3844 | p->stime = cputime_add(p->stime, cputime); |
3845 | ||
3846 | /* Add system time to cpustat. */ | |
3847 | tmp = cputime_to_cputime64(cputime); | |
3848 | if (hardirq_count() - hardirq_offset) | |
3849 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3850 | else if (softirq_count()) | |
3851 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 3852 | else if (p != rq->idle) |
1da177e4 | 3853 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 3854 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
3855 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
3856 | else | |
3857 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3858 | /* Account for system time used */ | |
3859 | acct_update_integrals(p); | |
1da177e4 LT |
3860 | } |
3861 | ||
c66f08be MN |
3862 | /* |
3863 | * Account scaled system cpu time to a process. | |
3864 | * @p: the process that the cpu time gets accounted to | |
3865 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3866 | * @cputime: the cpu time spent in kernel space since the last update | |
3867 | */ | |
3868 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
3869 | { | |
3870 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
3871 | } | |
3872 | ||
1da177e4 LT |
3873 | /* |
3874 | * Account for involuntary wait time. | |
3875 | * @p: the process from which the cpu time has been stolen | |
3876 | * @steal: the cpu time spent in involuntary wait | |
3877 | */ | |
3878 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3879 | { | |
3880 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3881 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3882 | struct rq *rq = this_rq(); |
1da177e4 LT |
3883 | |
3884 | if (p == rq->idle) { | |
3885 | p->stime = cputime_add(p->stime, steal); | |
3886 | if (atomic_read(&rq->nr_iowait) > 0) | |
3887 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3888 | else | |
3889 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 3890 | } else |
1da177e4 LT |
3891 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
3892 | } | |
3893 | ||
7835b98b CL |
3894 | /* |
3895 | * This function gets called by the timer code, with HZ frequency. | |
3896 | * We call it with interrupts disabled. | |
3897 | * | |
3898 | * It also gets called by the fork code, when changing the parent's | |
3899 | * timeslices. | |
3900 | */ | |
3901 | void scheduler_tick(void) | |
3902 | { | |
7835b98b CL |
3903 | int cpu = smp_processor_id(); |
3904 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3905 | struct task_struct *curr = rq->curr; |
529c7726 | 3906 | u64 next_tick = rq->tick_timestamp + TICK_NSEC; |
dd41f596 IM |
3907 | |
3908 | spin_lock(&rq->lock); | |
546fe3c9 | 3909 | __update_rq_clock(rq); |
529c7726 IM |
3910 | /* |
3911 | * Let rq->clock advance by at least TICK_NSEC: | |
3912 | */ | |
cc203d24 | 3913 | if (unlikely(rq->clock < next_tick)) { |
529c7726 | 3914 | rq->clock = next_tick; |
cc203d24 GC |
3915 | rq->clock_underflows++; |
3916 | } | |
529c7726 | 3917 | rq->tick_timestamp = rq->clock; |
15934a37 | 3918 | update_last_tick_seen(rq); |
f1a438d8 | 3919 | update_cpu_load(rq); |
fa85ae24 | 3920 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 3921 | spin_unlock(&rq->lock); |
7835b98b | 3922 | |
e418e1c2 | 3923 | #ifdef CONFIG_SMP |
dd41f596 IM |
3924 | rq->idle_at_tick = idle_cpu(cpu); |
3925 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3926 | #endif |
1da177e4 LT |
3927 | } |
3928 | ||
1da177e4 LT |
3929 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3930 | ||
43627582 | 3931 | void __kprobes add_preempt_count(int val) |
1da177e4 LT |
3932 | { |
3933 | /* | |
3934 | * Underflow? | |
3935 | */ | |
9a11b49a IM |
3936 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3937 | return; | |
1da177e4 LT |
3938 | preempt_count() += val; |
3939 | /* | |
3940 | * Spinlock count overflowing soon? | |
3941 | */ | |
33859f7f MOS |
3942 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3943 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3944 | } |
3945 | EXPORT_SYMBOL(add_preempt_count); | |
3946 | ||
43627582 | 3947 | void __kprobes sub_preempt_count(int val) |
1da177e4 LT |
3948 | { |
3949 | /* | |
3950 | * Underflow? | |
3951 | */ | |
9a11b49a IM |
3952 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3953 | return; | |
1da177e4 LT |
3954 | /* |
3955 | * Is the spinlock portion underflowing? | |
3956 | */ | |
9a11b49a IM |
3957 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3958 | !(preempt_count() & PREEMPT_MASK))) | |
3959 | return; | |
3960 | ||
1da177e4 LT |
3961 | preempt_count() -= val; |
3962 | } | |
3963 | EXPORT_SYMBOL(sub_preempt_count); | |
3964 | ||
3965 | #endif | |
3966 | ||
3967 | /* | |
dd41f596 | 3968 | * Print scheduling while atomic bug: |
1da177e4 | 3969 | */ |
dd41f596 | 3970 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3971 | { |
838225b4 SS |
3972 | struct pt_regs *regs = get_irq_regs(); |
3973 | ||
3974 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
3975 | prev->comm, prev->pid, preempt_count()); | |
3976 | ||
dd41f596 IM |
3977 | debug_show_held_locks(prev); |
3978 | if (irqs_disabled()) | |
3979 | print_irqtrace_events(prev); | |
838225b4 SS |
3980 | |
3981 | if (regs) | |
3982 | show_regs(regs); | |
3983 | else | |
3984 | dump_stack(); | |
dd41f596 | 3985 | } |
1da177e4 | 3986 | |
dd41f596 IM |
3987 | /* |
3988 | * Various schedule()-time debugging checks and statistics: | |
3989 | */ | |
3990 | static inline void schedule_debug(struct task_struct *prev) | |
3991 | { | |
1da177e4 | 3992 | /* |
41a2d6cf | 3993 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3994 | * schedule() atomically, we ignore that path for now. |
3995 | * Otherwise, whine if we are scheduling when we should not be. | |
3996 | */ | |
dd41f596 IM |
3997 | if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state)) |
3998 | __schedule_bug(prev); | |
3999 | ||
1da177e4 LT |
4000 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4001 | ||
2d72376b | 4002 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4003 | #ifdef CONFIG_SCHEDSTATS |
4004 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
4005 | schedstat_inc(this_rq(), bkl_count); |
4006 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
4007 | } |
4008 | #endif | |
dd41f596 IM |
4009 | } |
4010 | ||
4011 | /* | |
4012 | * Pick up the highest-prio task: | |
4013 | */ | |
4014 | static inline struct task_struct * | |
ff95f3df | 4015 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 4016 | { |
5522d5d5 | 4017 | const struct sched_class *class; |
dd41f596 | 4018 | struct task_struct *p; |
1da177e4 LT |
4019 | |
4020 | /* | |
dd41f596 IM |
4021 | * Optimization: we know that if all tasks are in |
4022 | * the fair class we can call that function directly: | |
1da177e4 | 4023 | */ |
dd41f596 | 4024 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4025 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4026 | if (likely(p)) |
4027 | return p; | |
1da177e4 LT |
4028 | } |
4029 | ||
dd41f596 IM |
4030 | class = sched_class_highest; |
4031 | for ( ; ; ) { | |
fb8d4724 | 4032 | p = class->pick_next_task(rq); |
dd41f596 IM |
4033 | if (p) |
4034 | return p; | |
4035 | /* | |
4036 | * Will never be NULL as the idle class always | |
4037 | * returns a non-NULL p: | |
4038 | */ | |
4039 | class = class->next; | |
4040 | } | |
4041 | } | |
1da177e4 | 4042 | |
dd41f596 IM |
4043 | /* |
4044 | * schedule() is the main scheduler function. | |
4045 | */ | |
4046 | asmlinkage void __sched schedule(void) | |
4047 | { | |
4048 | struct task_struct *prev, *next; | |
67ca7bde | 4049 | unsigned long *switch_count; |
dd41f596 | 4050 | struct rq *rq; |
dd41f596 IM |
4051 | int cpu; |
4052 | ||
4053 | need_resched: | |
4054 | preempt_disable(); | |
4055 | cpu = smp_processor_id(); | |
4056 | rq = cpu_rq(cpu); | |
4057 | rcu_qsctr_inc(cpu); | |
4058 | prev = rq->curr; | |
4059 | switch_count = &prev->nivcsw; | |
4060 | ||
4061 | release_kernel_lock(prev); | |
4062 | need_resched_nonpreemptible: | |
4063 | ||
4064 | schedule_debug(prev); | |
1da177e4 | 4065 | |
8f4d37ec PZ |
4066 | hrtick_clear(rq); |
4067 | ||
1e819950 IM |
4068 | /* |
4069 | * Do the rq-clock update outside the rq lock: | |
4070 | */ | |
4071 | local_irq_disable(); | |
c1b3da3e | 4072 | __update_rq_clock(rq); |
1e819950 IM |
4073 | spin_lock(&rq->lock); |
4074 | clear_tsk_need_resched(prev); | |
1da177e4 | 4075 | |
1da177e4 | 4076 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
1da177e4 | 4077 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && |
23e3c3cd | 4078 | signal_pending(prev))) { |
1da177e4 | 4079 | prev->state = TASK_RUNNING; |
dd41f596 | 4080 | } else { |
2e1cb74a | 4081 | deactivate_task(rq, prev, 1); |
1da177e4 | 4082 | } |
dd41f596 | 4083 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4084 | } |
4085 | ||
9a897c5a SR |
4086 | #ifdef CONFIG_SMP |
4087 | if (prev->sched_class->pre_schedule) | |
4088 | prev->sched_class->pre_schedule(rq, prev); | |
4089 | #endif | |
f65eda4f | 4090 | |
dd41f596 | 4091 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4092 | idle_balance(cpu, rq); |
1da177e4 | 4093 | |
31ee529c | 4094 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 4095 | next = pick_next_task(rq, prev); |
1da177e4 LT |
4096 | |
4097 | sched_info_switch(prev, next); | |
dd41f596 | 4098 | |
1da177e4 | 4099 | if (likely(prev != next)) { |
1da177e4 LT |
4100 | rq->nr_switches++; |
4101 | rq->curr = next; | |
4102 | ++*switch_count; | |
4103 | ||
dd41f596 | 4104 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
4105 | /* |
4106 | * the context switch might have flipped the stack from under | |
4107 | * us, hence refresh the local variables. | |
4108 | */ | |
4109 | cpu = smp_processor_id(); | |
4110 | rq = cpu_rq(cpu); | |
1da177e4 LT |
4111 | } else |
4112 | spin_unlock_irq(&rq->lock); | |
4113 | ||
8f4d37ec PZ |
4114 | hrtick_set(rq); |
4115 | ||
4116 | if (unlikely(reacquire_kernel_lock(current) < 0)) | |
1da177e4 | 4117 | goto need_resched_nonpreemptible; |
8f4d37ec | 4118 | |
1da177e4 LT |
4119 | preempt_enable_no_resched(); |
4120 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
4121 | goto need_resched; | |
4122 | } | |
1da177e4 LT |
4123 | EXPORT_SYMBOL(schedule); |
4124 | ||
4125 | #ifdef CONFIG_PREEMPT | |
4126 | /* | |
2ed6e34f | 4127 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4128 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4129 | * occur there and call schedule directly. |
4130 | */ | |
4131 | asmlinkage void __sched preempt_schedule(void) | |
4132 | { | |
4133 | struct thread_info *ti = current_thread_info(); | |
1da177e4 LT |
4134 | struct task_struct *task = current; |
4135 | int saved_lock_depth; | |
6478d880 | 4136 | |
1da177e4 LT |
4137 | /* |
4138 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4139 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4140 | */ |
beed33a8 | 4141 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4142 | return; |
4143 | ||
3a5c359a AK |
4144 | do { |
4145 | add_preempt_count(PREEMPT_ACTIVE); | |
4146 | ||
4147 | /* | |
4148 | * We keep the big kernel semaphore locked, but we | |
4149 | * clear ->lock_depth so that schedule() doesnt | |
4150 | * auto-release the semaphore: | |
4151 | */ | |
3a5c359a AK |
4152 | saved_lock_depth = task->lock_depth; |
4153 | task->lock_depth = -1; | |
3a5c359a | 4154 | schedule(); |
3a5c359a | 4155 | task->lock_depth = saved_lock_depth; |
3a5c359a | 4156 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4157 | |
3a5c359a AK |
4158 | /* |
4159 | * Check again in case we missed a preemption opportunity | |
4160 | * between schedule and now. | |
4161 | */ | |
4162 | barrier(); | |
4163 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 4164 | } |
1da177e4 LT |
4165 | EXPORT_SYMBOL(preempt_schedule); |
4166 | ||
4167 | /* | |
2ed6e34f | 4168 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4169 | * off of irq context. |
4170 | * Note, that this is called and return with irqs disabled. This will | |
4171 | * protect us against recursive calling from irq. | |
4172 | */ | |
4173 | asmlinkage void __sched preempt_schedule_irq(void) | |
4174 | { | |
4175 | struct thread_info *ti = current_thread_info(); | |
1da177e4 LT |
4176 | struct task_struct *task = current; |
4177 | int saved_lock_depth; | |
6478d880 | 4178 | |
2ed6e34f | 4179 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4180 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4181 | ||
3a5c359a AK |
4182 | do { |
4183 | add_preempt_count(PREEMPT_ACTIVE); | |
4184 | ||
4185 | /* | |
4186 | * We keep the big kernel semaphore locked, but we | |
4187 | * clear ->lock_depth so that schedule() doesnt | |
4188 | * auto-release the semaphore: | |
4189 | */ | |
3a5c359a AK |
4190 | saved_lock_depth = task->lock_depth; |
4191 | task->lock_depth = -1; | |
3a5c359a AK |
4192 | local_irq_enable(); |
4193 | schedule(); | |
4194 | local_irq_disable(); | |
3a5c359a | 4195 | task->lock_depth = saved_lock_depth; |
3a5c359a | 4196 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4197 | |
3a5c359a AK |
4198 | /* |
4199 | * Check again in case we missed a preemption opportunity | |
4200 | * between schedule and now. | |
4201 | */ | |
4202 | barrier(); | |
4203 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
4204 | } |
4205 | ||
4206 | #endif /* CONFIG_PREEMPT */ | |
4207 | ||
95cdf3b7 IM |
4208 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
4209 | void *key) | |
1da177e4 | 4210 | { |
48f24c4d | 4211 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 4212 | } |
1da177e4 LT |
4213 | EXPORT_SYMBOL(default_wake_function); |
4214 | ||
4215 | /* | |
41a2d6cf IM |
4216 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4217 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4218 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4219 | * | |
4220 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4221 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4222 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4223 | */ | |
4224 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
4225 | int nr_exclusive, int sync, void *key) | |
4226 | { | |
2e45874c | 4227 | wait_queue_t *curr, *next; |
1da177e4 | 4228 | |
2e45874c | 4229 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4230 | unsigned flags = curr->flags; |
4231 | ||
1da177e4 | 4232 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 4233 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4234 | break; |
4235 | } | |
4236 | } | |
4237 | ||
4238 | /** | |
4239 | * __wake_up - wake up threads blocked on a waitqueue. | |
4240 | * @q: the waitqueue | |
4241 | * @mode: which threads | |
4242 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4243 | * @key: is directly passed to the wakeup function |
1da177e4 | 4244 | */ |
7ad5b3a5 | 4245 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4246 | int nr_exclusive, void *key) |
1da177e4 LT |
4247 | { |
4248 | unsigned long flags; | |
4249 | ||
4250 | spin_lock_irqsave(&q->lock, flags); | |
4251 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4252 | spin_unlock_irqrestore(&q->lock, flags); | |
4253 | } | |
1da177e4 LT |
4254 | EXPORT_SYMBOL(__wake_up); |
4255 | ||
4256 | /* | |
4257 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4258 | */ | |
7ad5b3a5 | 4259 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4260 | { |
4261 | __wake_up_common(q, mode, 1, 0, NULL); | |
4262 | } | |
4263 | ||
4264 | /** | |
67be2dd1 | 4265 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4266 | * @q: the waitqueue |
4267 | * @mode: which threads | |
4268 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4269 | * | |
4270 | * The sync wakeup differs that the waker knows that it will schedule | |
4271 | * away soon, so while the target thread will be woken up, it will not | |
4272 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4273 | * with each other. This can prevent needless bouncing between CPUs. | |
4274 | * | |
4275 | * On UP it can prevent extra preemption. | |
4276 | */ | |
7ad5b3a5 | 4277 | void |
95cdf3b7 | 4278 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
1da177e4 LT |
4279 | { |
4280 | unsigned long flags; | |
4281 | int sync = 1; | |
4282 | ||
4283 | if (unlikely(!q)) | |
4284 | return; | |
4285 | ||
4286 | if (unlikely(!nr_exclusive)) | |
4287 | sync = 0; | |
4288 | ||
4289 | spin_lock_irqsave(&q->lock, flags); | |
4290 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
4291 | spin_unlock_irqrestore(&q->lock, flags); | |
4292 | } | |
4293 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
4294 | ||
b15136e9 | 4295 | void complete(struct completion *x) |
1da177e4 LT |
4296 | { |
4297 | unsigned long flags; | |
4298 | ||
4299 | spin_lock_irqsave(&x->wait.lock, flags); | |
4300 | x->done++; | |
d9514f6c | 4301 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4302 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4303 | } | |
4304 | EXPORT_SYMBOL(complete); | |
4305 | ||
b15136e9 | 4306 | void complete_all(struct completion *x) |
1da177e4 LT |
4307 | { |
4308 | unsigned long flags; | |
4309 | ||
4310 | spin_lock_irqsave(&x->wait.lock, flags); | |
4311 | x->done += UINT_MAX/2; | |
d9514f6c | 4312 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4313 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4314 | } | |
4315 | EXPORT_SYMBOL(complete_all); | |
4316 | ||
8cbbe86d AK |
4317 | static inline long __sched |
4318 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4319 | { |
1da177e4 LT |
4320 | if (!x->done) { |
4321 | DECLARE_WAITQUEUE(wait, current); | |
4322 | ||
4323 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
4324 | __add_wait_queue_tail(&x->wait, &wait); | |
4325 | do { | |
009e577e MW |
4326 | if ((state == TASK_INTERRUPTIBLE && |
4327 | signal_pending(current)) || | |
4328 | (state == TASK_KILLABLE && | |
4329 | fatal_signal_pending(current))) { | |
8cbbe86d AK |
4330 | __remove_wait_queue(&x->wait, &wait); |
4331 | return -ERESTARTSYS; | |
4332 | } | |
4333 | __set_current_state(state); | |
1da177e4 LT |
4334 | spin_unlock_irq(&x->wait.lock); |
4335 | timeout = schedule_timeout(timeout); | |
4336 | spin_lock_irq(&x->wait.lock); | |
4337 | if (!timeout) { | |
4338 | __remove_wait_queue(&x->wait, &wait); | |
8cbbe86d | 4339 | return timeout; |
1da177e4 LT |
4340 | } |
4341 | } while (!x->done); | |
4342 | __remove_wait_queue(&x->wait, &wait); | |
4343 | } | |
4344 | x->done--; | |
1da177e4 LT |
4345 | return timeout; |
4346 | } | |
1da177e4 | 4347 | |
8cbbe86d AK |
4348 | static long __sched |
4349 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4350 | { |
1da177e4 LT |
4351 | might_sleep(); |
4352 | ||
4353 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4354 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4355 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4356 | return timeout; |
4357 | } | |
1da177e4 | 4358 | |
b15136e9 | 4359 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4360 | { |
4361 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4362 | } |
8cbbe86d | 4363 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4364 | |
b15136e9 | 4365 | unsigned long __sched |
8cbbe86d | 4366 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4367 | { |
8cbbe86d | 4368 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4369 | } |
8cbbe86d | 4370 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4371 | |
8cbbe86d | 4372 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4373 | { |
51e97990 AK |
4374 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4375 | if (t == -ERESTARTSYS) | |
4376 | return t; | |
4377 | return 0; | |
0fec171c | 4378 | } |
8cbbe86d | 4379 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4380 | |
b15136e9 | 4381 | unsigned long __sched |
8cbbe86d AK |
4382 | wait_for_completion_interruptible_timeout(struct completion *x, |
4383 | unsigned long timeout) | |
0fec171c | 4384 | { |
8cbbe86d | 4385 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4386 | } |
8cbbe86d | 4387 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4388 | |
009e577e MW |
4389 | int __sched wait_for_completion_killable(struct completion *x) |
4390 | { | |
4391 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4392 | if (t == -ERESTARTSYS) | |
4393 | return t; | |
4394 | return 0; | |
4395 | } | |
4396 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4397 | ||
8cbbe86d AK |
4398 | static long __sched |
4399 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4400 | { |
0fec171c IM |
4401 | unsigned long flags; |
4402 | wait_queue_t wait; | |
4403 | ||
4404 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4405 | |
8cbbe86d | 4406 | __set_current_state(state); |
1da177e4 | 4407 | |
8cbbe86d AK |
4408 | spin_lock_irqsave(&q->lock, flags); |
4409 | __add_wait_queue(q, &wait); | |
4410 | spin_unlock(&q->lock); | |
4411 | timeout = schedule_timeout(timeout); | |
4412 | spin_lock_irq(&q->lock); | |
4413 | __remove_wait_queue(q, &wait); | |
4414 | spin_unlock_irqrestore(&q->lock, flags); | |
4415 | ||
4416 | return timeout; | |
4417 | } | |
4418 | ||
4419 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4420 | { | |
4421 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4422 | } |
1da177e4 LT |
4423 | EXPORT_SYMBOL(interruptible_sleep_on); |
4424 | ||
0fec171c | 4425 | long __sched |
95cdf3b7 | 4426 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4427 | { |
8cbbe86d | 4428 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4429 | } |
1da177e4 LT |
4430 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4431 | ||
0fec171c | 4432 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4433 | { |
8cbbe86d | 4434 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4435 | } |
1da177e4 LT |
4436 | EXPORT_SYMBOL(sleep_on); |
4437 | ||
0fec171c | 4438 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4439 | { |
8cbbe86d | 4440 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4441 | } |
1da177e4 LT |
4442 | EXPORT_SYMBOL(sleep_on_timeout); |
4443 | ||
b29739f9 IM |
4444 | #ifdef CONFIG_RT_MUTEXES |
4445 | ||
4446 | /* | |
4447 | * rt_mutex_setprio - set the current priority of a task | |
4448 | * @p: task | |
4449 | * @prio: prio value (kernel-internal form) | |
4450 | * | |
4451 | * This function changes the 'effective' priority of a task. It does | |
4452 | * not touch ->normal_prio like __setscheduler(). | |
4453 | * | |
4454 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4455 | */ | |
36c8b586 | 4456 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4457 | { |
4458 | unsigned long flags; | |
83b699ed | 4459 | int oldprio, on_rq, running; |
70b97a7f | 4460 | struct rq *rq; |
cb469845 | 4461 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
4462 | |
4463 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4464 | ||
4465 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4466 | update_rq_clock(rq); |
b29739f9 | 4467 | |
d5f9f942 | 4468 | oldprio = p->prio; |
dd41f596 | 4469 | on_rq = p->se.on_rq; |
051a1d1a | 4470 | running = task_current(rq, p); |
0e1f3483 | 4471 | if (on_rq) |
69be72c1 | 4472 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4473 | if (running) |
4474 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4475 | |
4476 | if (rt_prio(prio)) | |
4477 | p->sched_class = &rt_sched_class; | |
4478 | else | |
4479 | p->sched_class = &fair_sched_class; | |
4480 | ||
b29739f9 IM |
4481 | p->prio = prio; |
4482 | ||
0e1f3483 HS |
4483 | if (running) |
4484 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4485 | if (on_rq) { |
8159f87e | 4486 | enqueue_task(rq, p, 0); |
cb469845 SR |
4487 | |
4488 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4489 | } |
4490 | task_rq_unlock(rq, &flags); | |
4491 | } | |
4492 | ||
4493 | #endif | |
4494 | ||
36c8b586 | 4495 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4496 | { |
dd41f596 | 4497 | int old_prio, delta, on_rq; |
1da177e4 | 4498 | unsigned long flags; |
70b97a7f | 4499 | struct rq *rq; |
1da177e4 LT |
4500 | |
4501 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4502 | return; | |
4503 | /* | |
4504 | * We have to be careful, if called from sys_setpriority(), | |
4505 | * the task might be in the middle of scheduling on another CPU. | |
4506 | */ | |
4507 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4508 | update_rq_clock(rq); |
1da177e4 LT |
4509 | /* |
4510 | * The RT priorities are set via sched_setscheduler(), but we still | |
4511 | * allow the 'normal' nice value to be set - but as expected | |
4512 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4513 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4514 | */ |
e05606d3 | 4515 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4516 | p->static_prio = NICE_TO_PRIO(nice); |
4517 | goto out_unlock; | |
4518 | } | |
dd41f596 | 4519 | on_rq = p->se.on_rq; |
62fb1851 | 4520 | if (on_rq) { |
69be72c1 | 4521 | dequeue_task(rq, p, 0); |
62fb1851 PZ |
4522 | dec_load(rq, p); |
4523 | } | |
1da177e4 | 4524 | |
1da177e4 | 4525 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4526 | set_load_weight(p); |
b29739f9 IM |
4527 | old_prio = p->prio; |
4528 | p->prio = effective_prio(p); | |
4529 | delta = p->prio - old_prio; | |
1da177e4 | 4530 | |
dd41f596 | 4531 | if (on_rq) { |
8159f87e | 4532 | enqueue_task(rq, p, 0); |
62fb1851 | 4533 | inc_load(rq, p); |
1da177e4 | 4534 | /* |
d5f9f942 AM |
4535 | * If the task increased its priority or is running and |
4536 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4537 | */ |
d5f9f942 | 4538 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4539 | resched_task(rq->curr); |
4540 | } | |
4541 | out_unlock: | |
4542 | task_rq_unlock(rq, &flags); | |
4543 | } | |
1da177e4 LT |
4544 | EXPORT_SYMBOL(set_user_nice); |
4545 | ||
e43379f1 MM |
4546 | /* |
4547 | * can_nice - check if a task can reduce its nice value | |
4548 | * @p: task | |
4549 | * @nice: nice value | |
4550 | */ | |
36c8b586 | 4551 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4552 | { |
024f4747 MM |
4553 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4554 | int nice_rlim = 20 - nice; | |
48f24c4d | 4555 | |
e43379f1 MM |
4556 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4557 | capable(CAP_SYS_NICE)); | |
4558 | } | |
4559 | ||
1da177e4 LT |
4560 | #ifdef __ARCH_WANT_SYS_NICE |
4561 | ||
4562 | /* | |
4563 | * sys_nice - change the priority of the current process. | |
4564 | * @increment: priority increment | |
4565 | * | |
4566 | * sys_setpriority is a more generic, but much slower function that | |
4567 | * does similar things. | |
4568 | */ | |
4569 | asmlinkage long sys_nice(int increment) | |
4570 | { | |
48f24c4d | 4571 | long nice, retval; |
1da177e4 LT |
4572 | |
4573 | /* | |
4574 | * Setpriority might change our priority at the same moment. | |
4575 | * We don't have to worry. Conceptually one call occurs first | |
4576 | * and we have a single winner. | |
4577 | */ | |
e43379f1 MM |
4578 | if (increment < -40) |
4579 | increment = -40; | |
1da177e4 LT |
4580 | if (increment > 40) |
4581 | increment = 40; | |
4582 | ||
4583 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4584 | if (nice < -20) | |
4585 | nice = -20; | |
4586 | if (nice > 19) | |
4587 | nice = 19; | |
4588 | ||
e43379f1 MM |
4589 | if (increment < 0 && !can_nice(current, nice)) |
4590 | return -EPERM; | |
4591 | ||
1da177e4 LT |
4592 | retval = security_task_setnice(current, nice); |
4593 | if (retval) | |
4594 | return retval; | |
4595 | ||
4596 | set_user_nice(current, nice); | |
4597 | return 0; | |
4598 | } | |
4599 | ||
4600 | #endif | |
4601 | ||
4602 | /** | |
4603 | * task_prio - return the priority value of a given task. | |
4604 | * @p: the task in question. | |
4605 | * | |
4606 | * This is the priority value as seen by users in /proc. | |
4607 | * RT tasks are offset by -200. Normal tasks are centered | |
4608 | * around 0, value goes from -16 to +15. | |
4609 | */ | |
36c8b586 | 4610 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4611 | { |
4612 | return p->prio - MAX_RT_PRIO; | |
4613 | } | |
4614 | ||
4615 | /** | |
4616 | * task_nice - return the nice value of a given task. | |
4617 | * @p: the task in question. | |
4618 | */ | |
36c8b586 | 4619 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4620 | { |
4621 | return TASK_NICE(p); | |
4622 | } | |
150d8bed | 4623 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4624 | |
4625 | /** | |
4626 | * idle_cpu - is a given cpu idle currently? | |
4627 | * @cpu: the processor in question. | |
4628 | */ | |
4629 | int idle_cpu(int cpu) | |
4630 | { | |
4631 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4632 | } | |
4633 | ||
1da177e4 LT |
4634 | /** |
4635 | * idle_task - return the idle task for a given cpu. | |
4636 | * @cpu: the processor in question. | |
4637 | */ | |
36c8b586 | 4638 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4639 | { |
4640 | return cpu_rq(cpu)->idle; | |
4641 | } | |
4642 | ||
4643 | /** | |
4644 | * find_process_by_pid - find a process with a matching PID value. | |
4645 | * @pid: the pid in question. | |
4646 | */ | |
a9957449 | 4647 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4648 | { |
228ebcbe | 4649 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4650 | } |
4651 | ||
4652 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4653 | static void |
4654 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4655 | { |
dd41f596 | 4656 | BUG_ON(p->se.on_rq); |
48f24c4d | 4657 | |
1da177e4 | 4658 | p->policy = policy; |
dd41f596 IM |
4659 | switch (p->policy) { |
4660 | case SCHED_NORMAL: | |
4661 | case SCHED_BATCH: | |
4662 | case SCHED_IDLE: | |
4663 | p->sched_class = &fair_sched_class; | |
4664 | break; | |
4665 | case SCHED_FIFO: | |
4666 | case SCHED_RR: | |
4667 | p->sched_class = &rt_sched_class; | |
4668 | break; | |
4669 | } | |
4670 | ||
1da177e4 | 4671 | p->rt_priority = prio; |
b29739f9 IM |
4672 | p->normal_prio = normal_prio(p); |
4673 | /* we are holding p->pi_lock already */ | |
4674 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4675 | set_load_weight(p); |
1da177e4 LT |
4676 | } |
4677 | ||
4678 | /** | |
72fd4a35 | 4679 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4680 | * @p: the task in question. |
4681 | * @policy: new policy. | |
4682 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4683 | * |
72fd4a35 | 4684 | * NOTE that the task may be already dead. |
1da177e4 | 4685 | */ |
95cdf3b7 IM |
4686 | int sched_setscheduler(struct task_struct *p, int policy, |
4687 | struct sched_param *param) | |
1da177e4 | 4688 | { |
83b699ed | 4689 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4690 | unsigned long flags; |
cb469845 | 4691 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 4692 | struct rq *rq; |
1da177e4 | 4693 | |
66e5393a SR |
4694 | /* may grab non-irq protected spin_locks */ |
4695 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4696 | recheck: |
4697 | /* double check policy once rq lock held */ | |
4698 | if (policy < 0) | |
4699 | policy = oldpolicy = p->policy; | |
4700 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4701 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4702 | policy != SCHED_IDLE) | |
b0a9499c | 4703 | return -EINVAL; |
1da177e4 LT |
4704 | /* |
4705 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4706 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4707 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4708 | */ |
4709 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4710 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4711 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4712 | return -EINVAL; |
e05606d3 | 4713 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4714 | return -EINVAL; |
4715 | ||
37e4ab3f OC |
4716 | /* |
4717 | * Allow unprivileged RT tasks to decrease priority: | |
4718 | */ | |
4719 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4720 | if (rt_policy(policy)) { |
8dc3e909 | 4721 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4722 | |
4723 | if (!lock_task_sighand(p, &flags)) | |
4724 | return -ESRCH; | |
4725 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4726 | unlock_task_sighand(p, &flags); | |
4727 | ||
4728 | /* can't set/change the rt policy */ | |
4729 | if (policy != p->policy && !rlim_rtprio) | |
4730 | return -EPERM; | |
4731 | ||
4732 | /* can't increase priority */ | |
4733 | if (param->sched_priority > p->rt_priority && | |
4734 | param->sched_priority > rlim_rtprio) | |
4735 | return -EPERM; | |
4736 | } | |
dd41f596 IM |
4737 | /* |
4738 | * Like positive nice levels, dont allow tasks to | |
4739 | * move out of SCHED_IDLE either: | |
4740 | */ | |
4741 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4742 | return -EPERM; | |
5fe1d75f | 4743 | |
37e4ab3f OC |
4744 | /* can't change other user's priorities */ |
4745 | if ((current->euid != p->euid) && | |
4746 | (current->euid != p->uid)) | |
4747 | return -EPERM; | |
4748 | } | |
1da177e4 | 4749 | |
b68aa230 PZ |
4750 | #ifdef CONFIG_RT_GROUP_SCHED |
4751 | /* | |
4752 | * Do not allow realtime tasks into groups that have no runtime | |
4753 | * assigned. | |
4754 | */ | |
d0b27fa7 | 4755 | if (rt_policy(policy) && task_group(p)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
4756 | return -EPERM; |
4757 | #endif | |
4758 | ||
1da177e4 LT |
4759 | retval = security_task_setscheduler(p, policy, param); |
4760 | if (retval) | |
4761 | return retval; | |
b29739f9 IM |
4762 | /* |
4763 | * make sure no PI-waiters arrive (or leave) while we are | |
4764 | * changing the priority of the task: | |
4765 | */ | |
4766 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4767 | /* |
4768 | * To be able to change p->policy safely, the apropriate | |
4769 | * runqueue lock must be held. | |
4770 | */ | |
b29739f9 | 4771 | rq = __task_rq_lock(p); |
1da177e4 LT |
4772 | /* recheck policy now with rq lock held */ |
4773 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4774 | policy = oldpolicy = -1; | |
b29739f9 IM |
4775 | __task_rq_unlock(rq); |
4776 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4777 | goto recheck; |
4778 | } | |
2daa3577 | 4779 | update_rq_clock(rq); |
dd41f596 | 4780 | on_rq = p->se.on_rq; |
051a1d1a | 4781 | running = task_current(rq, p); |
0e1f3483 | 4782 | if (on_rq) |
2e1cb74a | 4783 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
4784 | if (running) |
4785 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4786 | |
1da177e4 | 4787 | oldprio = p->prio; |
dd41f596 | 4788 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4789 | |
0e1f3483 HS |
4790 | if (running) |
4791 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
4792 | if (on_rq) { |
4793 | activate_task(rq, p, 0); | |
cb469845 SR |
4794 | |
4795 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 4796 | } |
b29739f9 IM |
4797 | __task_rq_unlock(rq); |
4798 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4799 | ||
95e02ca9 TG |
4800 | rt_mutex_adjust_pi(p); |
4801 | ||
1da177e4 LT |
4802 | return 0; |
4803 | } | |
4804 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4805 | ||
95cdf3b7 IM |
4806 | static int |
4807 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4808 | { |
1da177e4 LT |
4809 | struct sched_param lparam; |
4810 | struct task_struct *p; | |
36c8b586 | 4811 | int retval; |
1da177e4 LT |
4812 | |
4813 | if (!param || pid < 0) | |
4814 | return -EINVAL; | |
4815 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4816 | return -EFAULT; | |
5fe1d75f ON |
4817 | |
4818 | rcu_read_lock(); | |
4819 | retval = -ESRCH; | |
1da177e4 | 4820 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4821 | if (p != NULL) |
4822 | retval = sched_setscheduler(p, policy, &lparam); | |
4823 | rcu_read_unlock(); | |
36c8b586 | 4824 | |
1da177e4 LT |
4825 | return retval; |
4826 | } | |
4827 | ||
4828 | /** | |
4829 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4830 | * @pid: the pid in question. | |
4831 | * @policy: new policy. | |
4832 | * @param: structure containing the new RT priority. | |
4833 | */ | |
41a2d6cf IM |
4834 | asmlinkage long |
4835 | sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4836 | { |
c21761f1 JB |
4837 | /* negative values for policy are not valid */ |
4838 | if (policy < 0) | |
4839 | return -EINVAL; | |
4840 | ||
1da177e4 LT |
4841 | return do_sched_setscheduler(pid, policy, param); |
4842 | } | |
4843 | ||
4844 | /** | |
4845 | * sys_sched_setparam - set/change the RT priority of a thread | |
4846 | * @pid: the pid in question. | |
4847 | * @param: structure containing the new RT priority. | |
4848 | */ | |
4849 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4850 | { | |
4851 | return do_sched_setscheduler(pid, -1, param); | |
4852 | } | |
4853 | ||
4854 | /** | |
4855 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4856 | * @pid: the pid in question. | |
4857 | */ | |
4858 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4859 | { | |
36c8b586 | 4860 | struct task_struct *p; |
3a5c359a | 4861 | int retval; |
1da177e4 LT |
4862 | |
4863 | if (pid < 0) | |
3a5c359a | 4864 | return -EINVAL; |
1da177e4 LT |
4865 | |
4866 | retval = -ESRCH; | |
4867 | read_lock(&tasklist_lock); | |
4868 | p = find_process_by_pid(pid); | |
4869 | if (p) { | |
4870 | retval = security_task_getscheduler(p); | |
4871 | if (!retval) | |
4872 | retval = p->policy; | |
4873 | } | |
4874 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
4875 | return retval; |
4876 | } | |
4877 | ||
4878 | /** | |
4879 | * sys_sched_getscheduler - get the RT priority of a thread | |
4880 | * @pid: the pid in question. | |
4881 | * @param: structure containing the RT priority. | |
4882 | */ | |
4883 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
4884 | { | |
4885 | struct sched_param lp; | |
36c8b586 | 4886 | struct task_struct *p; |
3a5c359a | 4887 | int retval; |
1da177e4 LT |
4888 | |
4889 | if (!param || pid < 0) | |
3a5c359a | 4890 | return -EINVAL; |
1da177e4 LT |
4891 | |
4892 | read_lock(&tasklist_lock); | |
4893 | p = find_process_by_pid(pid); | |
4894 | retval = -ESRCH; | |
4895 | if (!p) | |
4896 | goto out_unlock; | |
4897 | ||
4898 | retval = security_task_getscheduler(p); | |
4899 | if (retval) | |
4900 | goto out_unlock; | |
4901 | ||
4902 | lp.sched_priority = p->rt_priority; | |
4903 | read_unlock(&tasklist_lock); | |
4904 | ||
4905 | /* | |
4906 | * This one might sleep, we cannot do it with a spinlock held ... | |
4907 | */ | |
4908 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4909 | ||
1da177e4 LT |
4910 | return retval; |
4911 | ||
4912 | out_unlock: | |
4913 | read_unlock(&tasklist_lock); | |
4914 | return retval; | |
4915 | } | |
4916 | ||
b53e921b | 4917 | long sched_setaffinity(pid_t pid, const cpumask_t *in_mask) |
1da177e4 | 4918 | { |
1da177e4 | 4919 | cpumask_t cpus_allowed; |
b53e921b | 4920 | cpumask_t new_mask = *in_mask; |
36c8b586 IM |
4921 | struct task_struct *p; |
4922 | int retval; | |
1da177e4 | 4923 | |
95402b38 | 4924 | get_online_cpus(); |
1da177e4 LT |
4925 | read_lock(&tasklist_lock); |
4926 | ||
4927 | p = find_process_by_pid(pid); | |
4928 | if (!p) { | |
4929 | read_unlock(&tasklist_lock); | |
95402b38 | 4930 | put_online_cpus(); |
1da177e4 LT |
4931 | return -ESRCH; |
4932 | } | |
4933 | ||
4934 | /* | |
4935 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 4936 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
4937 | * usage count and then drop tasklist_lock. |
4938 | */ | |
4939 | get_task_struct(p); | |
4940 | read_unlock(&tasklist_lock); | |
4941 | ||
4942 | retval = -EPERM; | |
4943 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
4944 | !capable(CAP_SYS_NICE)) | |
4945 | goto out_unlock; | |
4946 | ||
e7834f8f DQ |
4947 | retval = security_task_setscheduler(p, 0, NULL); |
4948 | if (retval) | |
4949 | goto out_unlock; | |
4950 | ||
f9a86fcb | 4951 | cpuset_cpus_allowed(p, &cpus_allowed); |
1da177e4 | 4952 | cpus_and(new_mask, new_mask, cpus_allowed); |
8707d8b8 | 4953 | again: |
7c16ec58 | 4954 | retval = set_cpus_allowed_ptr(p, &new_mask); |
1da177e4 | 4955 | |
8707d8b8 | 4956 | if (!retval) { |
f9a86fcb | 4957 | cpuset_cpus_allowed(p, &cpus_allowed); |
8707d8b8 PM |
4958 | if (!cpus_subset(new_mask, cpus_allowed)) { |
4959 | /* | |
4960 | * We must have raced with a concurrent cpuset | |
4961 | * update. Just reset the cpus_allowed to the | |
4962 | * cpuset's cpus_allowed | |
4963 | */ | |
4964 | new_mask = cpus_allowed; | |
4965 | goto again; | |
4966 | } | |
4967 | } | |
1da177e4 LT |
4968 | out_unlock: |
4969 | put_task_struct(p); | |
95402b38 | 4970 | put_online_cpus(); |
1da177e4 LT |
4971 | return retval; |
4972 | } | |
4973 | ||
4974 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
4975 | cpumask_t *new_mask) | |
4976 | { | |
4977 | if (len < sizeof(cpumask_t)) { | |
4978 | memset(new_mask, 0, sizeof(cpumask_t)); | |
4979 | } else if (len > sizeof(cpumask_t)) { | |
4980 | len = sizeof(cpumask_t); | |
4981 | } | |
4982 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
4983 | } | |
4984 | ||
4985 | /** | |
4986 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4987 | * @pid: pid of the process | |
4988 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4989 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4990 | */ | |
4991 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
4992 | unsigned long __user *user_mask_ptr) | |
4993 | { | |
4994 | cpumask_t new_mask; | |
4995 | int retval; | |
4996 | ||
4997 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
4998 | if (retval) | |
4999 | return retval; | |
5000 | ||
b53e921b | 5001 | return sched_setaffinity(pid, &new_mask); |
1da177e4 LT |
5002 | } |
5003 | ||
5004 | /* | |
5005 | * Represents all cpu's present in the system | |
5006 | * In systems capable of hotplug, this map could dynamically grow | |
5007 | * as new cpu's are detected in the system via any platform specific | |
5008 | * method, such as ACPI for e.g. | |
5009 | */ | |
5010 | ||
4cef0c61 | 5011 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
5012 | EXPORT_SYMBOL(cpu_present_map); |
5013 | ||
5014 | #ifndef CONFIG_SMP | |
4cef0c61 | 5015 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
5016 | EXPORT_SYMBOL(cpu_online_map); |
5017 | ||
4cef0c61 | 5018 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 5019 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
5020 | #endif |
5021 | ||
5022 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
5023 | { | |
36c8b586 | 5024 | struct task_struct *p; |
1da177e4 | 5025 | int retval; |
1da177e4 | 5026 | |
95402b38 | 5027 | get_online_cpus(); |
1da177e4 LT |
5028 | read_lock(&tasklist_lock); |
5029 | ||
5030 | retval = -ESRCH; | |
5031 | p = find_process_by_pid(pid); | |
5032 | if (!p) | |
5033 | goto out_unlock; | |
5034 | ||
e7834f8f DQ |
5035 | retval = security_task_getscheduler(p); |
5036 | if (retval) | |
5037 | goto out_unlock; | |
5038 | ||
2f7016d9 | 5039 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
5040 | |
5041 | out_unlock: | |
5042 | read_unlock(&tasklist_lock); | |
95402b38 | 5043 | put_online_cpus(); |
1da177e4 | 5044 | |
9531b62f | 5045 | return retval; |
1da177e4 LT |
5046 | } |
5047 | ||
5048 | /** | |
5049 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5050 | * @pid: pid of the process | |
5051 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5052 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5053 | */ | |
5054 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
5055 | unsigned long __user *user_mask_ptr) | |
5056 | { | |
5057 | int ret; | |
5058 | cpumask_t mask; | |
5059 | ||
5060 | if (len < sizeof(cpumask_t)) | |
5061 | return -EINVAL; | |
5062 | ||
5063 | ret = sched_getaffinity(pid, &mask); | |
5064 | if (ret < 0) | |
5065 | return ret; | |
5066 | ||
5067 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
5068 | return -EFAULT; | |
5069 | ||
5070 | return sizeof(cpumask_t); | |
5071 | } | |
5072 | ||
5073 | /** | |
5074 | * sys_sched_yield - yield the current processor to other threads. | |
5075 | * | |
dd41f596 IM |
5076 | * This function yields the current CPU to other tasks. If there are no |
5077 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
5078 | */ |
5079 | asmlinkage long sys_sched_yield(void) | |
5080 | { | |
70b97a7f | 5081 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5082 | |
2d72376b | 5083 | schedstat_inc(rq, yld_count); |
4530d7ab | 5084 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5085 | |
5086 | /* | |
5087 | * Since we are going to call schedule() anyway, there's | |
5088 | * no need to preempt or enable interrupts: | |
5089 | */ | |
5090 | __release(rq->lock); | |
8a25d5de | 5091 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
5092 | _raw_spin_unlock(&rq->lock); |
5093 | preempt_enable_no_resched(); | |
5094 | ||
5095 | schedule(); | |
5096 | ||
5097 | return 0; | |
5098 | } | |
5099 | ||
e7b38404 | 5100 | static void __cond_resched(void) |
1da177e4 | 5101 | { |
8e0a43d8 IM |
5102 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
5103 | __might_sleep(__FILE__, __LINE__); | |
5104 | #endif | |
5bbcfd90 IM |
5105 | /* |
5106 | * The BKS might be reacquired before we have dropped | |
5107 | * PREEMPT_ACTIVE, which could trigger a second | |
5108 | * cond_resched() call. | |
5109 | */ | |
1da177e4 LT |
5110 | do { |
5111 | add_preempt_count(PREEMPT_ACTIVE); | |
5112 | schedule(); | |
5113 | sub_preempt_count(PREEMPT_ACTIVE); | |
5114 | } while (need_resched()); | |
5115 | } | |
5116 | ||
02b67cc3 HX |
5117 | #if !defined(CONFIG_PREEMPT) || defined(CONFIG_PREEMPT_VOLUNTARY) |
5118 | int __sched _cond_resched(void) | |
1da177e4 | 5119 | { |
9414232f IM |
5120 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
5121 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
5122 | __cond_resched(); |
5123 | return 1; | |
5124 | } | |
5125 | return 0; | |
5126 | } | |
02b67cc3 HX |
5127 | EXPORT_SYMBOL(_cond_resched); |
5128 | #endif | |
1da177e4 LT |
5129 | |
5130 | /* | |
5131 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
5132 | * call schedule, and on return reacquire the lock. | |
5133 | * | |
41a2d6cf | 5134 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5135 | * operations here to prevent schedule() from being called twice (once via |
5136 | * spin_unlock(), once by hand). | |
5137 | */ | |
95cdf3b7 | 5138 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5139 | { |
95c354fe | 5140 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
5141 | int ret = 0; |
5142 | ||
95c354fe | 5143 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5144 | spin_unlock(lock); |
95c354fe NP |
5145 | if (resched && need_resched()) |
5146 | __cond_resched(); | |
5147 | else | |
5148 | cpu_relax(); | |
6df3cecb | 5149 | ret = 1; |
1da177e4 | 5150 | spin_lock(lock); |
1da177e4 | 5151 | } |
6df3cecb | 5152 | return ret; |
1da177e4 | 5153 | } |
1da177e4 LT |
5154 | EXPORT_SYMBOL(cond_resched_lock); |
5155 | ||
5156 | int __sched cond_resched_softirq(void) | |
5157 | { | |
5158 | BUG_ON(!in_softirq()); | |
5159 | ||
9414232f | 5160 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 5161 | local_bh_enable(); |
1da177e4 LT |
5162 | __cond_resched(); |
5163 | local_bh_disable(); | |
5164 | return 1; | |
5165 | } | |
5166 | return 0; | |
5167 | } | |
1da177e4 LT |
5168 | EXPORT_SYMBOL(cond_resched_softirq); |
5169 | ||
1da177e4 LT |
5170 | /** |
5171 | * yield - yield the current processor to other threads. | |
5172 | * | |
72fd4a35 | 5173 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5174 | * thread runnable and calls sys_sched_yield(). |
5175 | */ | |
5176 | void __sched yield(void) | |
5177 | { | |
5178 | set_current_state(TASK_RUNNING); | |
5179 | sys_sched_yield(); | |
5180 | } | |
1da177e4 LT |
5181 | EXPORT_SYMBOL(yield); |
5182 | ||
5183 | /* | |
41a2d6cf | 5184 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
5185 | * that process accounting knows that this is a task in IO wait state. |
5186 | * | |
5187 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
5188 | * has set its backing_dev_info: the queue against which it should throttle) | |
5189 | */ | |
5190 | void __sched io_schedule(void) | |
5191 | { | |
70b97a7f | 5192 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 5193 | |
0ff92245 | 5194 | delayacct_blkio_start(); |
1da177e4 LT |
5195 | atomic_inc(&rq->nr_iowait); |
5196 | schedule(); | |
5197 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5198 | delayacct_blkio_end(); |
1da177e4 | 5199 | } |
1da177e4 LT |
5200 | EXPORT_SYMBOL(io_schedule); |
5201 | ||
5202 | long __sched io_schedule_timeout(long timeout) | |
5203 | { | |
70b97a7f | 5204 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
5205 | long ret; |
5206 | ||
0ff92245 | 5207 | delayacct_blkio_start(); |
1da177e4 LT |
5208 | atomic_inc(&rq->nr_iowait); |
5209 | ret = schedule_timeout(timeout); | |
5210 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5211 | delayacct_blkio_end(); |
1da177e4 LT |
5212 | return ret; |
5213 | } | |
5214 | ||
5215 | /** | |
5216 | * sys_sched_get_priority_max - return maximum RT priority. | |
5217 | * @policy: scheduling class. | |
5218 | * | |
5219 | * this syscall returns the maximum rt_priority that can be used | |
5220 | * by a given scheduling class. | |
5221 | */ | |
5222 | asmlinkage long sys_sched_get_priority_max(int policy) | |
5223 | { | |
5224 | int ret = -EINVAL; | |
5225 | ||
5226 | switch (policy) { | |
5227 | case SCHED_FIFO: | |
5228 | case SCHED_RR: | |
5229 | ret = MAX_USER_RT_PRIO-1; | |
5230 | break; | |
5231 | case SCHED_NORMAL: | |
b0a9499c | 5232 | case SCHED_BATCH: |
dd41f596 | 5233 | case SCHED_IDLE: |
1da177e4 LT |
5234 | ret = 0; |
5235 | break; | |
5236 | } | |
5237 | return ret; | |
5238 | } | |
5239 | ||
5240 | /** | |
5241 | * sys_sched_get_priority_min - return minimum RT priority. | |
5242 | * @policy: scheduling class. | |
5243 | * | |
5244 | * this syscall returns the minimum rt_priority that can be used | |
5245 | * by a given scheduling class. | |
5246 | */ | |
5247 | asmlinkage long sys_sched_get_priority_min(int policy) | |
5248 | { | |
5249 | int ret = -EINVAL; | |
5250 | ||
5251 | switch (policy) { | |
5252 | case SCHED_FIFO: | |
5253 | case SCHED_RR: | |
5254 | ret = 1; | |
5255 | break; | |
5256 | case SCHED_NORMAL: | |
b0a9499c | 5257 | case SCHED_BATCH: |
dd41f596 | 5258 | case SCHED_IDLE: |
1da177e4 LT |
5259 | ret = 0; |
5260 | } | |
5261 | return ret; | |
5262 | } | |
5263 | ||
5264 | /** | |
5265 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5266 | * @pid: pid of the process. | |
5267 | * @interval: userspace pointer to the timeslice value. | |
5268 | * | |
5269 | * this syscall writes the default timeslice value of a given process | |
5270 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5271 | */ | |
5272 | asmlinkage | |
5273 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
5274 | { | |
36c8b586 | 5275 | struct task_struct *p; |
a4ec24b4 | 5276 | unsigned int time_slice; |
3a5c359a | 5277 | int retval; |
1da177e4 | 5278 | struct timespec t; |
1da177e4 LT |
5279 | |
5280 | if (pid < 0) | |
3a5c359a | 5281 | return -EINVAL; |
1da177e4 LT |
5282 | |
5283 | retval = -ESRCH; | |
5284 | read_lock(&tasklist_lock); | |
5285 | p = find_process_by_pid(pid); | |
5286 | if (!p) | |
5287 | goto out_unlock; | |
5288 | ||
5289 | retval = security_task_getscheduler(p); | |
5290 | if (retval) | |
5291 | goto out_unlock; | |
5292 | ||
77034937 IM |
5293 | /* |
5294 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
5295 | * tasks that are on an otherwise idle runqueue: | |
5296 | */ | |
5297 | time_slice = 0; | |
5298 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 5299 | time_slice = DEF_TIMESLICE; |
1868f958 | 5300 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
5301 | struct sched_entity *se = &p->se; |
5302 | unsigned long flags; | |
5303 | struct rq *rq; | |
5304 | ||
5305 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
5306 | if (rq->cfs.load.weight) |
5307 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
5308 | task_rq_unlock(rq, &flags); |
5309 | } | |
1da177e4 | 5310 | read_unlock(&tasklist_lock); |
a4ec24b4 | 5311 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5312 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5313 | return retval; |
3a5c359a | 5314 | |
1da177e4 LT |
5315 | out_unlock: |
5316 | read_unlock(&tasklist_lock); | |
5317 | return retval; | |
5318 | } | |
5319 | ||
2ed6e34f | 5320 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 | 5321 | |
82a1fcb9 | 5322 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5323 | { |
1da177e4 | 5324 | unsigned long free = 0; |
36c8b586 | 5325 | unsigned state; |
1da177e4 | 5326 | |
1da177e4 | 5327 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 5328 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5329 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5330 | #if BITS_PER_LONG == 32 |
1da177e4 | 5331 | if (state == TASK_RUNNING) |
cc4ea795 | 5332 | printk(KERN_CONT " running "); |
1da177e4 | 5333 | else |
cc4ea795 | 5334 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5335 | #else |
5336 | if (state == TASK_RUNNING) | |
cc4ea795 | 5337 | printk(KERN_CONT " running task "); |
1da177e4 | 5338 | else |
cc4ea795 | 5339 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5340 | #endif |
5341 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
5342 | { | |
10ebffde | 5343 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
5344 | while (!*n) |
5345 | n++; | |
10ebffde | 5346 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
5347 | } |
5348 | #endif | |
ba25f9dc | 5349 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 5350 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 5351 | |
5fb5e6de | 5352 | show_stack(p, NULL); |
1da177e4 LT |
5353 | } |
5354 | ||
e59e2ae2 | 5355 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5356 | { |
36c8b586 | 5357 | struct task_struct *g, *p; |
1da177e4 | 5358 | |
4bd77321 IM |
5359 | #if BITS_PER_LONG == 32 |
5360 | printk(KERN_INFO | |
5361 | " task PC stack pid father\n"); | |
1da177e4 | 5362 | #else |
4bd77321 IM |
5363 | printk(KERN_INFO |
5364 | " task PC stack pid father\n"); | |
1da177e4 LT |
5365 | #endif |
5366 | read_lock(&tasklist_lock); | |
5367 | do_each_thread(g, p) { | |
5368 | /* | |
5369 | * reset the NMI-timeout, listing all files on a slow | |
5370 | * console might take alot of time: | |
5371 | */ | |
5372 | touch_nmi_watchdog(); | |
39bc89fd | 5373 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5374 | sched_show_task(p); |
1da177e4 LT |
5375 | } while_each_thread(g, p); |
5376 | ||
04c9167f JF |
5377 | touch_all_softlockup_watchdogs(); |
5378 | ||
dd41f596 IM |
5379 | #ifdef CONFIG_SCHED_DEBUG |
5380 | sysrq_sched_debug_show(); | |
5381 | #endif | |
1da177e4 | 5382 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5383 | /* |
5384 | * Only show locks if all tasks are dumped: | |
5385 | */ | |
5386 | if (state_filter == -1) | |
5387 | debug_show_all_locks(); | |
1da177e4 LT |
5388 | } |
5389 | ||
1df21055 IM |
5390 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5391 | { | |
dd41f596 | 5392 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5393 | } |
5394 | ||
f340c0d1 IM |
5395 | /** |
5396 | * init_idle - set up an idle thread for a given CPU | |
5397 | * @idle: task in question | |
5398 | * @cpu: cpu the idle task belongs to | |
5399 | * | |
5400 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5401 | * flag, to make booting more robust. | |
5402 | */ | |
5c1e1767 | 5403 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5404 | { |
70b97a7f | 5405 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5406 | unsigned long flags; |
5407 | ||
dd41f596 IM |
5408 | __sched_fork(idle); |
5409 | idle->se.exec_start = sched_clock(); | |
5410 | ||
b29739f9 | 5411 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 5412 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 5413 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
5414 | |
5415 | spin_lock_irqsave(&rq->lock, flags); | |
5416 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
5417 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5418 | idle->oncpu = 1; | |
5419 | #endif | |
1da177e4 LT |
5420 | spin_unlock_irqrestore(&rq->lock, flags); |
5421 | ||
5422 | /* Set the preempt count _outside_ the spinlocks! */ | |
a1261f54 | 5423 | task_thread_info(idle)->preempt_count = 0; |
6478d880 | 5424 | |
dd41f596 IM |
5425 | /* |
5426 | * The idle tasks have their own, simple scheduling class: | |
5427 | */ | |
5428 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5429 | } |
5430 | ||
5431 | /* | |
5432 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5433 | * indicates which cpus entered this state. This is used | |
5434 | * in the rcu update to wait only for active cpus. For system | |
5435 | * which do not switch off the HZ timer nohz_cpu_mask should | |
5436 | * always be CPU_MASK_NONE. | |
5437 | */ | |
5438 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
5439 | ||
19978ca6 IM |
5440 | /* |
5441 | * Increase the granularity value when there are more CPUs, | |
5442 | * because with more CPUs the 'effective latency' as visible | |
5443 | * to users decreases. But the relationship is not linear, | |
5444 | * so pick a second-best guess by going with the log2 of the | |
5445 | * number of CPUs. | |
5446 | * | |
5447 | * This idea comes from the SD scheduler of Con Kolivas: | |
5448 | */ | |
5449 | static inline void sched_init_granularity(void) | |
5450 | { | |
5451 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
5452 | const unsigned long limit = 200000000; | |
5453 | ||
5454 | sysctl_sched_min_granularity *= factor; | |
5455 | if (sysctl_sched_min_granularity > limit) | |
5456 | sysctl_sched_min_granularity = limit; | |
5457 | ||
5458 | sysctl_sched_latency *= factor; | |
5459 | if (sysctl_sched_latency > limit) | |
5460 | sysctl_sched_latency = limit; | |
5461 | ||
5462 | sysctl_sched_wakeup_granularity *= factor; | |
19978ca6 IM |
5463 | } |
5464 | ||
1da177e4 LT |
5465 | #ifdef CONFIG_SMP |
5466 | /* | |
5467 | * This is how migration works: | |
5468 | * | |
70b97a7f | 5469 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5470 | * runqueue and wake up that CPU's migration thread. |
5471 | * 2) we down() the locked semaphore => thread blocks. | |
5472 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5473 | * thread off the CPU) | |
5474 | * 4) it gets the migration request and checks whether the migrated | |
5475 | * task is still in the wrong runqueue. | |
5476 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5477 | * it and puts it into the right queue. | |
5478 | * 6) migration thread up()s the semaphore. | |
5479 | * 7) we wake up and the migration is done. | |
5480 | */ | |
5481 | ||
5482 | /* | |
5483 | * Change a given task's CPU affinity. Migrate the thread to a | |
5484 | * proper CPU and schedule it away if the CPU it's executing on | |
5485 | * is removed from the allowed bitmask. | |
5486 | * | |
5487 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5488 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5489 | * call is not atomic; no spinlocks may be held. |
5490 | */ | |
cd8ba7cd | 5491 | int set_cpus_allowed_ptr(struct task_struct *p, const cpumask_t *new_mask) |
1da177e4 | 5492 | { |
70b97a7f | 5493 | struct migration_req req; |
1da177e4 | 5494 | unsigned long flags; |
70b97a7f | 5495 | struct rq *rq; |
48f24c4d | 5496 | int ret = 0; |
1da177e4 LT |
5497 | |
5498 | rq = task_rq_lock(p, &flags); | |
cd8ba7cd | 5499 | if (!cpus_intersects(*new_mask, cpu_online_map)) { |
1da177e4 LT |
5500 | ret = -EINVAL; |
5501 | goto out; | |
5502 | } | |
5503 | ||
73fe6aae | 5504 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5505 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5506 | else { |
cd8ba7cd MT |
5507 | p->cpus_allowed = *new_mask; |
5508 | p->rt.nr_cpus_allowed = cpus_weight(*new_mask); | |
73fe6aae GH |
5509 | } |
5510 | ||
1da177e4 | 5511 | /* Can the task run on the task's current CPU? If so, we're done */ |
cd8ba7cd | 5512 | if (cpu_isset(task_cpu(p), *new_mask)) |
1da177e4 LT |
5513 | goto out; |
5514 | ||
cd8ba7cd | 5515 | if (migrate_task(p, any_online_cpu(*new_mask), &req)) { |
1da177e4 LT |
5516 | /* Need help from migration thread: drop lock and wait. */ |
5517 | task_rq_unlock(rq, &flags); | |
5518 | wake_up_process(rq->migration_thread); | |
5519 | wait_for_completion(&req.done); | |
5520 | tlb_migrate_finish(p->mm); | |
5521 | return 0; | |
5522 | } | |
5523 | out: | |
5524 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5525 | |
1da177e4 LT |
5526 | return ret; |
5527 | } | |
cd8ba7cd | 5528 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5529 | |
5530 | /* | |
41a2d6cf | 5531 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5532 | * this because either it can't run here any more (set_cpus_allowed() |
5533 | * away from this CPU, or CPU going down), or because we're | |
5534 | * attempting to rebalance this task on exec (sched_exec). | |
5535 | * | |
5536 | * So we race with normal scheduler movements, but that's OK, as long | |
5537 | * as the task is no longer on this CPU. | |
efc30814 KK |
5538 | * |
5539 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5540 | */ |
efc30814 | 5541 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5542 | { |
70b97a7f | 5543 | struct rq *rq_dest, *rq_src; |
dd41f596 | 5544 | int ret = 0, on_rq; |
1da177e4 LT |
5545 | |
5546 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 5547 | return ret; |
1da177e4 LT |
5548 | |
5549 | rq_src = cpu_rq(src_cpu); | |
5550 | rq_dest = cpu_rq(dest_cpu); | |
5551 | ||
5552 | double_rq_lock(rq_src, rq_dest); | |
5553 | /* Already moved. */ | |
5554 | if (task_cpu(p) != src_cpu) | |
5555 | goto out; | |
5556 | /* Affinity changed (again). */ | |
5557 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
5558 | goto out; | |
5559 | ||
dd41f596 | 5560 | on_rq = p->se.on_rq; |
6e82a3be | 5561 | if (on_rq) |
2e1cb74a | 5562 | deactivate_task(rq_src, p, 0); |
6e82a3be | 5563 | |
1da177e4 | 5564 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
5565 | if (on_rq) { |
5566 | activate_task(rq_dest, p, 0); | |
5567 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 5568 | } |
efc30814 | 5569 | ret = 1; |
1da177e4 LT |
5570 | out: |
5571 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 5572 | return ret; |
1da177e4 LT |
5573 | } |
5574 | ||
5575 | /* | |
5576 | * migration_thread - this is a highprio system thread that performs | |
5577 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5578 | * another runqueue. | |
5579 | */ | |
95cdf3b7 | 5580 | static int migration_thread(void *data) |
1da177e4 | 5581 | { |
1da177e4 | 5582 | int cpu = (long)data; |
70b97a7f | 5583 | struct rq *rq; |
1da177e4 LT |
5584 | |
5585 | rq = cpu_rq(cpu); | |
5586 | BUG_ON(rq->migration_thread != current); | |
5587 | ||
5588 | set_current_state(TASK_INTERRUPTIBLE); | |
5589 | while (!kthread_should_stop()) { | |
70b97a7f | 5590 | struct migration_req *req; |
1da177e4 | 5591 | struct list_head *head; |
1da177e4 | 5592 | |
1da177e4 LT |
5593 | spin_lock_irq(&rq->lock); |
5594 | ||
5595 | if (cpu_is_offline(cpu)) { | |
5596 | spin_unlock_irq(&rq->lock); | |
5597 | goto wait_to_die; | |
5598 | } | |
5599 | ||
5600 | if (rq->active_balance) { | |
5601 | active_load_balance(rq, cpu); | |
5602 | rq->active_balance = 0; | |
5603 | } | |
5604 | ||
5605 | head = &rq->migration_queue; | |
5606 | ||
5607 | if (list_empty(head)) { | |
5608 | spin_unlock_irq(&rq->lock); | |
5609 | schedule(); | |
5610 | set_current_state(TASK_INTERRUPTIBLE); | |
5611 | continue; | |
5612 | } | |
70b97a7f | 5613 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5614 | list_del_init(head->next); |
5615 | ||
674311d5 NP |
5616 | spin_unlock(&rq->lock); |
5617 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5618 | local_irq_enable(); | |
1da177e4 LT |
5619 | |
5620 | complete(&req->done); | |
5621 | } | |
5622 | __set_current_state(TASK_RUNNING); | |
5623 | return 0; | |
5624 | ||
5625 | wait_to_die: | |
5626 | /* Wait for kthread_stop */ | |
5627 | set_current_state(TASK_INTERRUPTIBLE); | |
5628 | while (!kthread_should_stop()) { | |
5629 | schedule(); | |
5630 | set_current_state(TASK_INTERRUPTIBLE); | |
5631 | } | |
5632 | __set_current_state(TASK_RUNNING); | |
5633 | return 0; | |
5634 | } | |
5635 | ||
5636 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5637 | |
5638 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5639 | { | |
5640 | int ret; | |
5641 | ||
5642 | local_irq_disable(); | |
5643 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5644 | local_irq_enable(); | |
5645 | return ret; | |
5646 | } | |
5647 | ||
054b9108 | 5648 | /* |
3a4fa0a2 | 5649 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
5650 | * NOTE: interrupts should be disabled by the caller |
5651 | */ | |
48f24c4d | 5652 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5653 | { |
efc30814 | 5654 | unsigned long flags; |
1da177e4 | 5655 | cpumask_t mask; |
70b97a7f IM |
5656 | struct rq *rq; |
5657 | int dest_cpu; | |
1da177e4 | 5658 | |
3a5c359a AK |
5659 | do { |
5660 | /* On same node? */ | |
5661 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
5662 | cpus_and(mask, mask, p->cpus_allowed); | |
5663 | dest_cpu = any_online_cpu(mask); | |
5664 | ||
5665 | /* On any allowed CPU? */ | |
434d53b0 | 5666 | if (dest_cpu >= nr_cpu_ids) |
3a5c359a AK |
5667 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5668 | ||
5669 | /* No more Mr. Nice Guy. */ | |
434d53b0 | 5670 | if (dest_cpu >= nr_cpu_ids) { |
f9a86fcb MT |
5671 | cpumask_t cpus_allowed; |
5672 | ||
5673 | cpuset_cpus_allowed_locked(p, &cpus_allowed); | |
470fd646 CW |
5674 | /* |
5675 | * Try to stay on the same cpuset, where the | |
5676 | * current cpuset may be a subset of all cpus. | |
5677 | * The cpuset_cpus_allowed_locked() variant of | |
41a2d6cf | 5678 | * cpuset_cpus_allowed() will not block. It must be |
470fd646 CW |
5679 | * called within calls to cpuset_lock/cpuset_unlock. |
5680 | */ | |
3a5c359a | 5681 | rq = task_rq_lock(p, &flags); |
470fd646 | 5682 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
5683 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5684 | task_rq_unlock(rq, &flags); | |
1da177e4 | 5685 | |
3a5c359a AK |
5686 | /* |
5687 | * Don't tell them about moving exiting tasks or | |
5688 | * kernel threads (both mm NULL), since they never | |
5689 | * leave kernel. | |
5690 | */ | |
41a2d6cf | 5691 | if (p->mm && printk_ratelimit()) { |
3a5c359a AK |
5692 | printk(KERN_INFO "process %d (%s) no " |
5693 | "longer affine to cpu%d\n", | |
41a2d6cf IM |
5694 | task_pid_nr(p), p->comm, dead_cpu); |
5695 | } | |
3a5c359a | 5696 | } |
f7b4cddc | 5697 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
5698 | } |
5699 | ||
5700 | /* | |
5701 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5702 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5703 | * for performance reasons the counter is not stricly tracking tasks to | |
5704 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5705 | * to keep the global sum constant after CPU-down: | |
5706 | */ | |
70b97a7f | 5707 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5708 | { |
7c16ec58 | 5709 | struct rq *rq_dest = cpu_rq(any_online_cpu(*CPU_MASK_ALL_PTR)); |
1da177e4 LT |
5710 | unsigned long flags; |
5711 | ||
5712 | local_irq_save(flags); | |
5713 | double_rq_lock(rq_src, rq_dest); | |
5714 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5715 | rq_src->nr_uninterruptible = 0; | |
5716 | double_rq_unlock(rq_src, rq_dest); | |
5717 | local_irq_restore(flags); | |
5718 | } | |
5719 | ||
5720 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5721 | static void migrate_live_tasks(int src_cpu) | |
5722 | { | |
48f24c4d | 5723 | struct task_struct *p, *t; |
1da177e4 | 5724 | |
f7b4cddc | 5725 | read_lock(&tasklist_lock); |
1da177e4 | 5726 | |
48f24c4d IM |
5727 | do_each_thread(t, p) { |
5728 | if (p == current) | |
1da177e4 LT |
5729 | continue; |
5730 | ||
48f24c4d IM |
5731 | if (task_cpu(p) == src_cpu) |
5732 | move_task_off_dead_cpu(src_cpu, p); | |
5733 | } while_each_thread(t, p); | |
1da177e4 | 5734 | |
f7b4cddc | 5735 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5736 | } |
5737 | ||
dd41f596 IM |
5738 | /* |
5739 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5740 | * It does so by boosting its priority to highest possible. |
5741 | * Used by CPU offline code. | |
1da177e4 LT |
5742 | */ |
5743 | void sched_idle_next(void) | |
5744 | { | |
48f24c4d | 5745 | int this_cpu = smp_processor_id(); |
70b97a7f | 5746 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5747 | struct task_struct *p = rq->idle; |
5748 | unsigned long flags; | |
5749 | ||
5750 | /* cpu has to be offline */ | |
48f24c4d | 5751 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5752 | |
48f24c4d IM |
5753 | /* |
5754 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5755 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5756 | */ |
5757 | spin_lock_irqsave(&rq->lock, flags); | |
5758 | ||
dd41f596 | 5759 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 5760 | |
94bc9a7b DA |
5761 | update_rq_clock(rq); |
5762 | activate_task(rq, p, 0); | |
1da177e4 LT |
5763 | |
5764 | spin_unlock_irqrestore(&rq->lock, flags); | |
5765 | } | |
5766 | ||
48f24c4d IM |
5767 | /* |
5768 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5769 | * offline. |
5770 | */ | |
5771 | void idle_task_exit(void) | |
5772 | { | |
5773 | struct mm_struct *mm = current->active_mm; | |
5774 | ||
5775 | BUG_ON(cpu_online(smp_processor_id())); | |
5776 | ||
5777 | if (mm != &init_mm) | |
5778 | switch_mm(mm, &init_mm, current); | |
5779 | mmdrop(mm); | |
5780 | } | |
5781 | ||
054b9108 | 5782 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5783 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5784 | { |
70b97a7f | 5785 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5786 | |
5787 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5788 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5789 | |
5790 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5791 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5792 | |
48f24c4d | 5793 | get_task_struct(p); |
1da177e4 LT |
5794 | |
5795 | /* | |
5796 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 5797 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
5798 | * fine. |
5799 | */ | |
f7b4cddc | 5800 | spin_unlock_irq(&rq->lock); |
48f24c4d | 5801 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 5802 | spin_lock_irq(&rq->lock); |
1da177e4 | 5803 | |
48f24c4d | 5804 | put_task_struct(p); |
1da177e4 LT |
5805 | } |
5806 | ||
5807 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5808 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5809 | { | |
70b97a7f | 5810 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5811 | struct task_struct *next; |
48f24c4d | 5812 | |
dd41f596 IM |
5813 | for ( ; ; ) { |
5814 | if (!rq->nr_running) | |
5815 | break; | |
a8e504d2 | 5816 | update_rq_clock(rq); |
ff95f3df | 5817 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
5818 | if (!next) |
5819 | break; | |
5820 | migrate_dead(dead_cpu, next); | |
e692ab53 | 5821 | |
1da177e4 LT |
5822 | } |
5823 | } | |
5824 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5825 | ||
e692ab53 NP |
5826 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5827 | ||
5828 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5829 | { |
5830 | .procname = "sched_domain", | |
c57baf1e | 5831 | .mode = 0555, |
e0361851 | 5832 | }, |
38605cae | 5833 | {0, }, |
e692ab53 NP |
5834 | }; |
5835 | ||
5836 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 5837 | { |
c57baf1e | 5838 | .ctl_name = CTL_KERN, |
e0361851 | 5839 | .procname = "kernel", |
c57baf1e | 5840 | .mode = 0555, |
e0361851 AD |
5841 | .child = sd_ctl_dir, |
5842 | }, | |
38605cae | 5843 | {0, }, |
e692ab53 NP |
5844 | }; |
5845 | ||
5846 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5847 | { | |
5848 | struct ctl_table *entry = | |
5cf9f062 | 5849 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5850 | |
e692ab53 NP |
5851 | return entry; |
5852 | } | |
5853 | ||
6382bc90 MM |
5854 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5855 | { | |
cd790076 | 5856 | struct ctl_table *entry; |
6382bc90 | 5857 | |
cd790076 MM |
5858 | /* |
5859 | * In the intermediate directories, both the child directory and | |
5860 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5861 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5862 | * static strings and all have proc handlers. |
5863 | */ | |
5864 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5865 | if (entry->child) |
5866 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5867 | if (entry->proc_handler == NULL) |
5868 | kfree(entry->procname); | |
5869 | } | |
6382bc90 MM |
5870 | |
5871 | kfree(*tablep); | |
5872 | *tablep = NULL; | |
5873 | } | |
5874 | ||
e692ab53 | 5875 | static void |
e0361851 | 5876 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5877 | const char *procname, void *data, int maxlen, |
5878 | mode_t mode, proc_handler *proc_handler) | |
5879 | { | |
e692ab53 NP |
5880 | entry->procname = procname; |
5881 | entry->data = data; | |
5882 | entry->maxlen = maxlen; | |
5883 | entry->mode = mode; | |
5884 | entry->proc_handler = proc_handler; | |
5885 | } | |
5886 | ||
5887 | static struct ctl_table * | |
5888 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5889 | { | |
ace8b3d6 | 5890 | struct ctl_table *table = sd_alloc_ctl_entry(12); |
e692ab53 | 5891 | |
ad1cdc1d MM |
5892 | if (table == NULL) |
5893 | return NULL; | |
5894 | ||
e0361851 | 5895 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5896 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5897 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5898 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5899 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5900 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5901 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5902 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5903 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5904 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5905 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5906 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5907 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5908 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5909 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5910 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5911 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5912 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5913 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5914 | &sd->cache_nice_tries, |
5915 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5916 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5917 | sizeof(int), 0644, proc_dointvec_minmax); |
6323469f | 5918 | /* &table[11] is terminator */ |
e692ab53 NP |
5919 | |
5920 | return table; | |
5921 | } | |
5922 | ||
9a4e7159 | 5923 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5924 | { |
5925 | struct ctl_table *entry, *table; | |
5926 | struct sched_domain *sd; | |
5927 | int domain_num = 0, i; | |
5928 | char buf[32]; | |
5929 | ||
5930 | for_each_domain(cpu, sd) | |
5931 | domain_num++; | |
5932 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5933 | if (table == NULL) |
5934 | return NULL; | |
e692ab53 NP |
5935 | |
5936 | i = 0; | |
5937 | for_each_domain(cpu, sd) { | |
5938 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5939 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5940 | entry->mode = 0555; |
e692ab53 NP |
5941 | entry->child = sd_alloc_ctl_domain_table(sd); |
5942 | entry++; | |
5943 | i++; | |
5944 | } | |
5945 | return table; | |
5946 | } | |
5947 | ||
5948 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5949 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
5950 | { |
5951 | int i, cpu_num = num_online_cpus(); | |
5952 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
5953 | char buf[32]; | |
5954 | ||
7378547f MM |
5955 | WARN_ON(sd_ctl_dir[0].child); |
5956 | sd_ctl_dir[0].child = entry; | |
5957 | ||
ad1cdc1d MM |
5958 | if (entry == NULL) |
5959 | return; | |
5960 | ||
97b6ea7b | 5961 | for_each_online_cpu(i) { |
e692ab53 | 5962 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5963 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5964 | entry->mode = 0555; |
e692ab53 | 5965 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5966 | entry++; |
e692ab53 | 5967 | } |
7378547f MM |
5968 | |
5969 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5970 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5971 | } | |
6382bc90 | 5972 | |
7378547f | 5973 | /* may be called multiple times per register */ |
6382bc90 MM |
5974 | static void unregister_sched_domain_sysctl(void) |
5975 | { | |
7378547f MM |
5976 | if (sd_sysctl_header) |
5977 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5978 | sd_sysctl_header = NULL; |
7378547f MM |
5979 | if (sd_ctl_dir[0].child) |
5980 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5981 | } |
e692ab53 | 5982 | #else |
6382bc90 MM |
5983 | static void register_sched_domain_sysctl(void) |
5984 | { | |
5985 | } | |
5986 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5987 | { |
5988 | } | |
5989 | #endif | |
5990 | ||
1da177e4 LT |
5991 | /* |
5992 | * migration_call - callback that gets triggered when a CPU is added. | |
5993 | * Here we can start up the necessary migration thread for the new CPU. | |
5994 | */ | |
48f24c4d IM |
5995 | static int __cpuinit |
5996 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5997 | { |
1da177e4 | 5998 | struct task_struct *p; |
48f24c4d | 5999 | int cpu = (long)hcpu; |
1da177e4 | 6000 | unsigned long flags; |
70b97a7f | 6001 | struct rq *rq; |
1da177e4 LT |
6002 | |
6003 | switch (action) { | |
5be9361c | 6004 | |
1da177e4 | 6005 | case CPU_UP_PREPARE: |
8bb78442 | 6006 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 6007 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
6008 | if (IS_ERR(p)) |
6009 | return NOTIFY_BAD; | |
1da177e4 LT |
6010 | kthread_bind(p, cpu); |
6011 | /* Must be high prio: stop_machine expects to yield to it. */ | |
6012 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 6013 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
6014 | task_rq_unlock(rq, &flags); |
6015 | cpu_rq(cpu)->migration_thread = p; | |
6016 | break; | |
48f24c4d | 6017 | |
1da177e4 | 6018 | case CPU_ONLINE: |
8bb78442 | 6019 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 6020 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 6021 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
6022 | |
6023 | /* Update our root-domain */ | |
6024 | rq = cpu_rq(cpu); | |
6025 | spin_lock_irqsave(&rq->lock, flags); | |
6026 | if (rq->rd) { | |
6027 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
6028 | cpu_set(cpu, rq->rd->online); | |
6029 | } | |
6030 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 6031 | break; |
48f24c4d | 6032 | |
1da177e4 LT |
6033 | #ifdef CONFIG_HOTPLUG_CPU |
6034 | case CPU_UP_CANCELED: | |
8bb78442 | 6035 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
6036 | if (!cpu_rq(cpu)->migration_thread) |
6037 | break; | |
41a2d6cf | 6038 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
6039 | kthread_bind(cpu_rq(cpu)->migration_thread, |
6040 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
6041 | kthread_stop(cpu_rq(cpu)->migration_thread); |
6042 | cpu_rq(cpu)->migration_thread = NULL; | |
6043 | break; | |
48f24c4d | 6044 | |
1da177e4 | 6045 | case CPU_DEAD: |
8bb78442 | 6046 | case CPU_DEAD_FROZEN: |
470fd646 | 6047 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
6048 | migrate_live_tasks(cpu); |
6049 | rq = cpu_rq(cpu); | |
6050 | kthread_stop(rq->migration_thread); | |
6051 | rq->migration_thread = NULL; | |
6052 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 6053 | spin_lock_irq(&rq->lock); |
a8e504d2 | 6054 | update_rq_clock(rq); |
2e1cb74a | 6055 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 6056 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
6057 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
6058 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 6059 | migrate_dead_tasks(cpu); |
d2da272a | 6060 | spin_unlock_irq(&rq->lock); |
470fd646 | 6061 | cpuset_unlock(); |
1da177e4 LT |
6062 | migrate_nr_uninterruptible(rq); |
6063 | BUG_ON(rq->nr_running != 0); | |
6064 | ||
41a2d6cf IM |
6065 | /* |
6066 | * No need to migrate the tasks: it was best-effort if | |
6067 | * they didn't take sched_hotcpu_mutex. Just wake up | |
6068 | * the requestors. | |
6069 | */ | |
1da177e4 LT |
6070 | spin_lock_irq(&rq->lock); |
6071 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
6072 | struct migration_req *req; |
6073 | ||
1da177e4 | 6074 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 6075 | struct migration_req, list); |
1da177e4 LT |
6076 | list_del_init(&req->list); |
6077 | complete(&req->done); | |
6078 | } | |
6079 | spin_unlock_irq(&rq->lock); | |
6080 | break; | |
57d885fe | 6081 | |
08f503b0 GH |
6082 | case CPU_DYING: |
6083 | case CPU_DYING_FROZEN: | |
57d885fe GH |
6084 | /* Update our root-domain */ |
6085 | rq = cpu_rq(cpu); | |
6086 | spin_lock_irqsave(&rq->lock, flags); | |
6087 | if (rq->rd) { | |
6088 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
6089 | cpu_clear(cpu, rq->rd->online); | |
6090 | } | |
6091 | spin_unlock_irqrestore(&rq->lock, flags); | |
6092 | break; | |
1da177e4 LT |
6093 | #endif |
6094 | } | |
6095 | return NOTIFY_OK; | |
6096 | } | |
6097 | ||
6098 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
6099 | * happens before everything else. | |
6100 | */ | |
26c2143b | 6101 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
6102 | .notifier_call = migration_call, |
6103 | .priority = 10 | |
6104 | }; | |
6105 | ||
e6fe6649 | 6106 | void __init migration_init(void) |
1da177e4 LT |
6107 | { |
6108 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6109 | int err; |
48f24c4d IM |
6110 | |
6111 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
6112 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6113 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6114 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6115 | register_cpu_notifier(&migration_notifier); | |
1da177e4 LT |
6116 | } |
6117 | #endif | |
6118 | ||
6119 | #ifdef CONFIG_SMP | |
476f3534 | 6120 | |
3e9830dc | 6121 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6122 | |
7c16ec58 MT |
6123 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
6124 | cpumask_t *groupmask) | |
1da177e4 | 6125 | { |
4dcf6aff | 6126 | struct sched_group *group = sd->groups; |
434d53b0 | 6127 | char str[256]; |
1da177e4 | 6128 | |
434d53b0 | 6129 | cpulist_scnprintf(str, sizeof(str), sd->span); |
7c16ec58 | 6130 | cpus_clear(*groupmask); |
4dcf6aff IM |
6131 | |
6132 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6133 | ||
6134 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
6135 | printk("does not load-balance\n"); | |
6136 | if (sd->parent) | |
6137 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
6138 | " has parent"); | |
6139 | return -1; | |
41c7ce9a NP |
6140 | } |
6141 | ||
4dcf6aff IM |
6142 | printk(KERN_CONT "span %s\n", str); |
6143 | ||
6144 | if (!cpu_isset(cpu, sd->span)) { | |
6145 | printk(KERN_ERR "ERROR: domain->span does not contain " | |
6146 | "CPU%d\n", cpu); | |
6147 | } | |
6148 | if (!cpu_isset(cpu, group->cpumask)) { | |
6149 | printk(KERN_ERR "ERROR: domain->groups does not contain" | |
6150 | " CPU%d\n", cpu); | |
6151 | } | |
1da177e4 | 6152 | |
4dcf6aff | 6153 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6154 | do { |
4dcf6aff IM |
6155 | if (!group) { |
6156 | printk("\n"); | |
6157 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6158 | break; |
6159 | } | |
6160 | ||
4dcf6aff IM |
6161 | if (!group->__cpu_power) { |
6162 | printk(KERN_CONT "\n"); | |
6163 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6164 | "set\n"); | |
6165 | break; | |
6166 | } | |
1da177e4 | 6167 | |
4dcf6aff IM |
6168 | if (!cpus_weight(group->cpumask)) { |
6169 | printk(KERN_CONT "\n"); | |
6170 | printk(KERN_ERR "ERROR: empty group\n"); | |
6171 | break; | |
6172 | } | |
1da177e4 | 6173 | |
7c16ec58 | 6174 | if (cpus_intersects(*groupmask, group->cpumask)) { |
4dcf6aff IM |
6175 | printk(KERN_CONT "\n"); |
6176 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
6177 | break; | |
6178 | } | |
1da177e4 | 6179 | |
7c16ec58 | 6180 | cpus_or(*groupmask, *groupmask, group->cpumask); |
1da177e4 | 6181 | |
434d53b0 | 6182 | cpulist_scnprintf(str, sizeof(str), group->cpumask); |
4dcf6aff | 6183 | printk(KERN_CONT " %s", str); |
1da177e4 | 6184 | |
4dcf6aff IM |
6185 | group = group->next; |
6186 | } while (group != sd->groups); | |
6187 | printk(KERN_CONT "\n"); | |
1da177e4 | 6188 | |
7c16ec58 | 6189 | if (!cpus_equal(sd->span, *groupmask)) |
4dcf6aff | 6190 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6191 | |
7c16ec58 | 6192 | if (sd->parent && !cpus_subset(*groupmask, sd->parent->span)) |
4dcf6aff IM |
6193 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6194 | "of domain->span\n"); | |
6195 | return 0; | |
6196 | } | |
1da177e4 | 6197 | |
4dcf6aff IM |
6198 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6199 | { | |
7c16ec58 | 6200 | cpumask_t *groupmask; |
4dcf6aff | 6201 | int level = 0; |
1da177e4 | 6202 | |
4dcf6aff IM |
6203 | if (!sd) { |
6204 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6205 | return; | |
6206 | } | |
1da177e4 | 6207 | |
4dcf6aff IM |
6208 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6209 | ||
7c16ec58 MT |
6210 | groupmask = kmalloc(sizeof(cpumask_t), GFP_KERNEL); |
6211 | if (!groupmask) { | |
6212 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); | |
6213 | return; | |
6214 | } | |
6215 | ||
4dcf6aff | 6216 | for (;;) { |
7c16ec58 | 6217 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6218 | break; |
1da177e4 LT |
6219 | level++; |
6220 | sd = sd->parent; | |
33859f7f | 6221 | if (!sd) |
4dcf6aff IM |
6222 | break; |
6223 | } | |
7c16ec58 | 6224 | kfree(groupmask); |
1da177e4 LT |
6225 | } |
6226 | #else | |
48f24c4d | 6227 | # define sched_domain_debug(sd, cpu) do { } while (0) |
1da177e4 LT |
6228 | #endif |
6229 | ||
1a20ff27 | 6230 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
6231 | { |
6232 | if (cpus_weight(sd->span) == 1) | |
6233 | return 1; | |
6234 | ||
6235 | /* Following flags need at least 2 groups */ | |
6236 | if (sd->flags & (SD_LOAD_BALANCE | | |
6237 | SD_BALANCE_NEWIDLE | | |
6238 | SD_BALANCE_FORK | | |
89c4710e SS |
6239 | SD_BALANCE_EXEC | |
6240 | SD_SHARE_CPUPOWER | | |
6241 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6242 | if (sd->groups != sd->groups->next) |
6243 | return 0; | |
6244 | } | |
6245 | ||
6246 | /* Following flags don't use groups */ | |
6247 | if (sd->flags & (SD_WAKE_IDLE | | |
6248 | SD_WAKE_AFFINE | | |
6249 | SD_WAKE_BALANCE)) | |
6250 | return 0; | |
6251 | ||
6252 | return 1; | |
6253 | } | |
6254 | ||
48f24c4d IM |
6255 | static int |
6256 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6257 | { |
6258 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6259 | ||
6260 | if (sd_degenerate(parent)) | |
6261 | return 1; | |
6262 | ||
6263 | if (!cpus_equal(sd->span, parent->span)) | |
6264 | return 0; | |
6265 | ||
6266 | /* Does parent contain flags not in child? */ | |
6267 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
6268 | if (cflags & SD_WAKE_AFFINE) | |
6269 | pflags &= ~SD_WAKE_BALANCE; | |
6270 | /* Flags needing groups don't count if only 1 group in parent */ | |
6271 | if (parent->groups == parent->groups->next) { | |
6272 | pflags &= ~(SD_LOAD_BALANCE | | |
6273 | SD_BALANCE_NEWIDLE | | |
6274 | SD_BALANCE_FORK | | |
89c4710e SS |
6275 | SD_BALANCE_EXEC | |
6276 | SD_SHARE_CPUPOWER | | |
6277 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
6278 | } |
6279 | if (~cflags & pflags) | |
6280 | return 0; | |
6281 | ||
6282 | return 1; | |
6283 | } | |
6284 | ||
57d885fe GH |
6285 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6286 | { | |
6287 | unsigned long flags; | |
6288 | const struct sched_class *class; | |
6289 | ||
6290 | spin_lock_irqsave(&rq->lock, flags); | |
6291 | ||
6292 | if (rq->rd) { | |
6293 | struct root_domain *old_rd = rq->rd; | |
6294 | ||
0eab9146 | 6295 | for (class = sched_class_highest; class; class = class->next) { |
57d885fe GH |
6296 | if (class->leave_domain) |
6297 | class->leave_domain(rq); | |
0eab9146 | 6298 | } |
57d885fe | 6299 | |
dc938520 GH |
6300 | cpu_clear(rq->cpu, old_rd->span); |
6301 | cpu_clear(rq->cpu, old_rd->online); | |
6302 | ||
57d885fe GH |
6303 | if (atomic_dec_and_test(&old_rd->refcount)) |
6304 | kfree(old_rd); | |
6305 | } | |
6306 | ||
6307 | atomic_inc(&rd->refcount); | |
6308 | rq->rd = rd; | |
6309 | ||
dc938520 | 6310 | cpu_set(rq->cpu, rd->span); |
1f94ef59 GH |
6311 | if (cpu_isset(rq->cpu, cpu_online_map)) |
6312 | cpu_set(rq->cpu, rd->online); | |
dc938520 | 6313 | |
0eab9146 | 6314 | for (class = sched_class_highest; class; class = class->next) { |
57d885fe GH |
6315 | if (class->join_domain) |
6316 | class->join_domain(rq); | |
0eab9146 | 6317 | } |
57d885fe GH |
6318 | |
6319 | spin_unlock_irqrestore(&rq->lock, flags); | |
6320 | } | |
6321 | ||
dc938520 | 6322 | static void init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6323 | { |
6324 | memset(rd, 0, sizeof(*rd)); | |
6325 | ||
dc938520 GH |
6326 | cpus_clear(rd->span); |
6327 | cpus_clear(rd->online); | |
57d885fe GH |
6328 | } |
6329 | ||
6330 | static void init_defrootdomain(void) | |
6331 | { | |
dc938520 | 6332 | init_rootdomain(&def_root_domain); |
57d885fe GH |
6333 | atomic_set(&def_root_domain.refcount, 1); |
6334 | } | |
6335 | ||
dc938520 | 6336 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6337 | { |
6338 | struct root_domain *rd; | |
6339 | ||
6340 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6341 | if (!rd) | |
6342 | return NULL; | |
6343 | ||
dc938520 | 6344 | init_rootdomain(rd); |
57d885fe GH |
6345 | |
6346 | return rd; | |
6347 | } | |
6348 | ||
1da177e4 | 6349 | /* |
0eab9146 | 6350 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6351 | * hold the hotplug lock. |
6352 | */ | |
0eab9146 IM |
6353 | static void |
6354 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6355 | { |
70b97a7f | 6356 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6357 | struct sched_domain *tmp; |
6358 | ||
6359 | /* Remove the sched domains which do not contribute to scheduling. */ | |
6360 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
6361 | struct sched_domain *parent = tmp->parent; | |
6362 | if (!parent) | |
6363 | break; | |
1a848870 | 6364 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6365 | tmp->parent = parent->parent; |
1a848870 SS |
6366 | if (parent->parent) |
6367 | parent->parent->child = tmp; | |
6368 | } | |
245af2c7 SS |
6369 | } |
6370 | ||
1a848870 | 6371 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6372 | sd = sd->parent; |
1a848870 SS |
6373 | if (sd) |
6374 | sd->child = NULL; | |
6375 | } | |
1da177e4 LT |
6376 | |
6377 | sched_domain_debug(sd, cpu); | |
6378 | ||
57d885fe | 6379 | rq_attach_root(rq, rd); |
674311d5 | 6380 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6381 | } |
6382 | ||
6383 | /* cpus with isolated domains */ | |
67af63a6 | 6384 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
6385 | |
6386 | /* Setup the mask of cpus configured for isolated domains */ | |
6387 | static int __init isolated_cpu_setup(char *str) | |
6388 | { | |
6389 | int ints[NR_CPUS], i; | |
6390 | ||
6391 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
6392 | cpus_clear(cpu_isolated_map); | |
6393 | for (i = 1; i <= ints[0]; i++) | |
6394 | if (ints[i] < NR_CPUS) | |
6395 | cpu_set(ints[i], cpu_isolated_map); | |
6396 | return 1; | |
6397 | } | |
6398 | ||
8927f494 | 6399 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6400 | |
6401 | /* | |
6711cab4 SS |
6402 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6403 | * to a function which identifies what group(along with sched group) a CPU | |
6404 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
6405 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
6406 | * |
6407 | * init_sched_build_groups will build a circular linked list of the groups | |
6408 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6409 | * and ->cpu_power to 0. | |
6410 | */ | |
a616058b | 6411 | static void |
7c16ec58 | 6412 | init_sched_build_groups(const cpumask_t *span, const cpumask_t *cpu_map, |
6711cab4 | 6413 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, |
7c16ec58 MT |
6414 | struct sched_group **sg, |
6415 | cpumask_t *tmpmask), | |
6416 | cpumask_t *covered, cpumask_t *tmpmask) | |
1da177e4 LT |
6417 | { |
6418 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6419 | int i; |
6420 | ||
7c16ec58 MT |
6421 | cpus_clear(*covered); |
6422 | ||
6423 | for_each_cpu_mask(i, *span) { | |
6711cab4 | 6424 | struct sched_group *sg; |
7c16ec58 | 6425 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6426 | int j; |
6427 | ||
7c16ec58 | 6428 | if (cpu_isset(i, *covered)) |
1da177e4 LT |
6429 | continue; |
6430 | ||
7c16ec58 | 6431 | cpus_clear(sg->cpumask); |
5517d86b | 6432 | sg->__cpu_power = 0; |
1da177e4 | 6433 | |
7c16ec58 MT |
6434 | for_each_cpu_mask(j, *span) { |
6435 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) | |
1da177e4 LT |
6436 | continue; |
6437 | ||
7c16ec58 | 6438 | cpu_set(j, *covered); |
1da177e4 LT |
6439 | cpu_set(j, sg->cpumask); |
6440 | } | |
6441 | if (!first) | |
6442 | first = sg; | |
6443 | if (last) | |
6444 | last->next = sg; | |
6445 | last = sg; | |
6446 | } | |
6447 | last->next = first; | |
6448 | } | |
6449 | ||
9c1cfda2 | 6450 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6451 | |
9c1cfda2 | 6452 | #ifdef CONFIG_NUMA |
198e2f18 | 6453 | |
9c1cfda2 JH |
6454 | /** |
6455 | * find_next_best_node - find the next node to include in a sched_domain | |
6456 | * @node: node whose sched_domain we're building | |
6457 | * @used_nodes: nodes already in the sched_domain | |
6458 | * | |
41a2d6cf | 6459 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6460 | * finds the closest node not already in the @used_nodes map. |
6461 | * | |
6462 | * Should use nodemask_t. | |
6463 | */ | |
c5f59f08 | 6464 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6465 | { |
6466 | int i, n, val, min_val, best_node = 0; | |
6467 | ||
6468 | min_val = INT_MAX; | |
6469 | ||
6470 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6471 | /* Start at @node */ | |
6472 | n = (node + i) % MAX_NUMNODES; | |
6473 | ||
6474 | if (!nr_cpus_node(n)) | |
6475 | continue; | |
6476 | ||
6477 | /* Skip already used nodes */ | |
c5f59f08 | 6478 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6479 | continue; |
6480 | ||
6481 | /* Simple min distance search */ | |
6482 | val = node_distance(node, n); | |
6483 | ||
6484 | if (val < min_val) { | |
6485 | min_val = val; | |
6486 | best_node = n; | |
6487 | } | |
6488 | } | |
6489 | ||
c5f59f08 | 6490 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6491 | return best_node; |
6492 | } | |
6493 | ||
6494 | /** | |
6495 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6496 | * @node: node whose cpumask we're constructing | |
9c1cfda2 | 6497 | * |
41a2d6cf | 6498 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6499 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6500 | * out optimally. | |
6501 | */ | |
4bdbaad3 | 6502 | static void sched_domain_node_span(int node, cpumask_t *span) |
9c1cfda2 | 6503 | { |
c5f59f08 | 6504 | nodemask_t used_nodes; |
c5f59f08 | 6505 | node_to_cpumask_ptr(nodemask, node); |
48f24c4d | 6506 | int i; |
9c1cfda2 | 6507 | |
4bdbaad3 | 6508 | cpus_clear(*span); |
c5f59f08 | 6509 | nodes_clear(used_nodes); |
9c1cfda2 | 6510 | |
4bdbaad3 | 6511 | cpus_or(*span, *span, *nodemask); |
c5f59f08 | 6512 | node_set(node, used_nodes); |
9c1cfda2 JH |
6513 | |
6514 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6515 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6516 | |
c5f59f08 | 6517 | node_to_cpumask_ptr_next(nodemask, next_node); |
4bdbaad3 | 6518 | cpus_or(*span, *span, *nodemask); |
9c1cfda2 | 6519 | } |
9c1cfda2 JH |
6520 | } |
6521 | #endif | |
6522 | ||
5c45bf27 | 6523 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6524 | |
9c1cfda2 | 6525 | /* |
48f24c4d | 6526 | * SMT sched-domains: |
9c1cfda2 | 6527 | */ |
1da177e4 LT |
6528 | #ifdef CONFIG_SCHED_SMT |
6529 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 6530 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 6531 | |
41a2d6cf | 6532 | static int |
7c16ec58 MT |
6533 | cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6534 | cpumask_t *unused) | |
1da177e4 | 6535 | { |
6711cab4 SS |
6536 | if (sg) |
6537 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
6538 | return cpu; |
6539 | } | |
6540 | #endif | |
6541 | ||
48f24c4d IM |
6542 | /* |
6543 | * multi-core sched-domains: | |
6544 | */ | |
1e9f28fa SS |
6545 | #ifdef CONFIG_SCHED_MC |
6546 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 6547 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
6548 | #endif |
6549 | ||
6550 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 6551 | static int |
7c16ec58 MT |
6552 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6553 | cpumask_t *mask) | |
1e9f28fa | 6554 | { |
6711cab4 | 6555 | int group; |
7c16ec58 MT |
6556 | |
6557 | *mask = per_cpu(cpu_sibling_map, cpu); | |
6558 | cpus_and(*mask, *mask, *cpu_map); | |
6559 | group = first_cpu(*mask); | |
6711cab4 SS |
6560 | if (sg) |
6561 | *sg = &per_cpu(sched_group_core, group); | |
6562 | return group; | |
1e9f28fa SS |
6563 | } |
6564 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 6565 | static int |
7c16ec58 MT |
6566 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6567 | cpumask_t *unused) | |
1e9f28fa | 6568 | { |
6711cab4 SS |
6569 | if (sg) |
6570 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
6571 | return cpu; |
6572 | } | |
6573 | #endif | |
6574 | ||
1da177e4 | 6575 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 6576 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 6577 | |
41a2d6cf | 6578 | static int |
7c16ec58 MT |
6579 | cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6580 | cpumask_t *mask) | |
1da177e4 | 6581 | { |
6711cab4 | 6582 | int group; |
48f24c4d | 6583 | #ifdef CONFIG_SCHED_MC |
7c16ec58 MT |
6584 | *mask = cpu_coregroup_map(cpu); |
6585 | cpus_and(*mask, *mask, *cpu_map); | |
6586 | group = first_cpu(*mask); | |
1e9f28fa | 6587 | #elif defined(CONFIG_SCHED_SMT) |
7c16ec58 MT |
6588 | *mask = per_cpu(cpu_sibling_map, cpu); |
6589 | cpus_and(*mask, *mask, *cpu_map); | |
6590 | group = first_cpu(*mask); | |
1da177e4 | 6591 | #else |
6711cab4 | 6592 | group = cpu; |
1da177e4 | 6593 | #endif |
6711cab4 SS |
6594 | if (sg) |
6595 | *sg = &per_cpu(sched_group_phys, group); | |
6596 | return group; | |
1da177e4 LT |
6597 | } |
6598 | ||
6599 | #ifdef CONFIG_NUMA | |
1da177e4 | 6600 | /* |
9c1cfda2 JH |
6601 | * The init_sched_build_groups can't handle what we want to do with node |
6602 | * groups, so roll our own. Now each node has its own list of groups which | |
6603 | * gets dynamically allocated. | |
1da177e4 | 6604 | */ |
9c1cfda2 | 6605 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
434d53b0 | 6606 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6607 | |
9c1cfda2 | 6608 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 6609 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 6610 | |
6711cab4 | 6611 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
7c16ec58 | 6612 | struct sched_group **sg, cpumask_t *nodemask) |
9c1cfda2 | 6613 | { |
6711cab4 SS |
6614 | int group; |
6615 | ||
7c16ec58 MT |
6616 | *nodemask = node_to_cpumask(cpu_to_node(cpu)); |
6617 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
6618 | group = first_cpu(*nodemask); | |
6711cab4 SS |
6619 | |
6620 | if (sg) | |
6621 | *sg = &per_cpu(sched_group_allnodes, group); | |
6622 | return group; | |
1da177e4 | 6623 | } |
6711cab4 | 6624 | |
08069033 SS |
6625 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6626 | { | |
6627 | struct sched_group *sg = group_head; | |
6628 | int j; | |
6629 | ||
6630 | if (!sg) | |
6631 | return; | |
3a5c359a AK |
6632 | do { |
6633 | for_each_cpu_mask(j, sg->cpumask) { | |
6634 | struct sched_domain *sd; | |
08069033 | 6635 | |
3a5c359a AK |
6636 | sd = &per_cpu(phys_domains, j); |
6637 | if (j != first_cpu(sd->groups->cpumask)) { | |
6638 | /* | |
6639 | * Only add "power" once for each | |
6640 | * physical package. | |
6641 | */ | |
6642 | continue; | |
6643 | } | |
08069033 | 6644 | |
3a5c359a AK |
6645 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
6646 | } | |
6647 | sg = sg->next; | |
6648 | } while (sg != group_head); | |
08069033 | 6649 | } |
1da177e4 LT |
6650 | #endif |
6651 | ||
a616058b | 6652 | #ifdef CONFIG_NUMA |
51888ca2 | 6653 | /* Free memory allocated for various sched_group structures */ |
7c16ec58 | 6654 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
51888ca2 | 6655 | { |
a616058b | 6656 | int cpu, i; |
51888ca2 SV |
6657 | |
6658 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
6659 | struct sched_group **sched_group_nodes |
6660 | = sched_group_nodes_bycpu[cpu]; | |
6661 | ||
51888ca2 SV |
6662 | if (!sched_group_nodes) |
6663 | continue; | |
6664 | ||
6665 | for (i = 0; i < MAX_NUMNODES; i++) { | |
51888ca2 SV |
6666 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
6667 | ||
7c16ec58 MT |
6668 | *nodemask = node_to_cpumask(i); |
6669 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
6670 | if (cpus_empty(*nodemask)) | |
51888ca2 SV |
6671 | continue; |
6672 | ||
6673 | if (sg == NULL) | |
6674 | continue; | |
6675 | sg = sg->next; | |
6676 | next_sg: | |
6677 | oldsg = sg; | |
6678 | sg = sg->next; | |
6679 | kfree(oldsg); | |
6680 | if (oldsg != sched_group_nodes[i]) | |
6681 | goto next_sg; | |
6682 | } | |
6683 | kfree(sched_group_nodes); | |
6684 | sched_group_nodes_bycpu[cpu] = NULL; | |
6685 | } | |
51888ca2 | 6686 | } |
a616058b | 6687 | #else |
7c16ec58 | 6688 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
a616058b SS |
6689 | { |
6690 | } | |
6691 | #endif | |
51888ca2 | 6692 | |
89c4710e SS |
6693 | /* |
6694 | * Initialize sched groups cpu_power. | |
6695 | * | |
6696 | * cpu_power indicates the capacity of sched group, which is used while | |
6697 | * distributing the load between different sched groups in a sched domain. | |
6698 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6699 | * there are asymmetries in the topology. If there are asymmetries, group | |
6700 | * having more cpu_power will pickup more load compared to the group having | |
6701 | * less cpu_power. | |
6702 | * | |
6703 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
6704 | * the maximum number of tasks a group can handle in the presence of other idle | |
6705 | * or lightly loaded groups in the same sched domain. | |
6706 | */ | |
6707 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6708 | { | |
6709 | struct sched_domain *child; | |
6710 | struct sched_group *group; | |
6711 | ||
6712 | WARN_ON(!sd || !sd->groups); | |
6713 | ||
6714 | if (cpu != first_cpu(sd->groups->cpumask)) | |
6715 | return; | |
6716 | ||
6717 | child = sd->child; | |
6718 | ||
5517d86b ED |
6719 | sd->groups->__cpu_power = 0; |
6720 | ||
89c4710e SS |
6721 | /* |
6722 | * For perf policy, if the groups in child domain share resources | |
6723 | * (for example cores sharing some portions of the cache hierarchy | |
6724 | * or SMT), then set this domain groups cpu_power such that each group | |
6725 | * can handle only one task, when there are other idle groups in the | |
6726 | * same sched domain. | |
6727 | */ | |
6728 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
6729 | (child->flags & | |
6730 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 6731 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
6732 | return; |
6733 | } | |
6734 | ||
89c4710e SS |
6735 | /* |
6736 | * add cpu_power of each child group to this groups cpu_power | |
6737 | */ | |
6738 | group = child->groups; | |
6739 | do { | |
5517d86b | 6740 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
6741 | group = group->next; |
6742 | } while (group != child->groups); | |
6743 | } | |
6744 | ||
7c16ec58 MT |
6745 | /* |
6746 | * Initializers for schedule domains | |
6747 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6748 | */ | |
6749 | ||
6750 | #define SD_INIT(sd, type) sd_init_##type(sd) | |
6751 | #define SD_INIT_FUNC(type) \ | |
6752 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
6753 | { \ | |
6754 | memset(sd, 0, sizeof(*sd)); \ | |
6755 | *sd = SD_##type##_INIT; \ | |
6756 | } | |
6757 | ||
6758 | SD_INIT_FUNC(CPU) | |
6759 | #ifdef CONFIG_NUMA | |
6760 | SD_INIT_FUNC(ALLNODES) | |
6761 | SD_INIT_FUNC(NODE) | |
6762 | #endif | |
6763 | #ifdef CONFIG_SCHED_SMT | |
6764 | SD_INIT_FUNC(SIBLING) | |
6765 | #endif | |
6766 | #ifdef CONFIG_SCHED_MC | |
6767 | SD_INIT_FUNC(MC) | |
6768 | #endif | |
6769 | ||
6770 | /* | |
6771 | * To minimize stack usage kmalloc room for cpumasks and share the | |
6772 | * space as the usage in build_sched_domains() dictates. Used only | |
6773 | * if the amount of space is significant. | |
6774 | */ | |
6775 | struct allmasks { | |
6776 | cpumask_t tmpmask; /* make this one first */ | |
6777 | union { | |
6778 | cpumask_t nodemask; | |
6779 | cpumask_t this_sibling_map; | |
6780 | cpumask_t this_core_map; | |
6781 | }; | |
6782 | cpumask_t send_covered; | |
6783 | ||
6784 | #ifdef CONFIG_NUMA | |
6785 | cpumask_t domainspan; | |
6786 | cpumask_t covered; | |
6787 | cpumask_t notcovered; | |
6788 | #endif | |
6789 | }; | |
6790 | ||
6791 | #if NR_CPUS > 128 | |
6792 | #define SCHED_CPUMASK_ALLOC 1 | |
6793 | #define SCHED_CPUMASK_FREE(v) kfree(v) | |
6794 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks *v | |
6795 | #else | |
6796 | #define SCHED_CPUMASK_ALLOC 0 | |
6797 | #define SCHED_CPUMASK_FREE(v) | |
6798 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks _v, *v = &_v | |
6799 | #endif | |
6800 | ||
6801 | #define SCHED_CPUMASK_VAR(v, a) cpumask_t *v = (cpumask_t *) \ | |
6802 | ((unsigned long)(a) + offsetof(struct allmasks, v)) | |
6803 | ||
1da177e4 | 6804 | /* |
1a20ff27 DG |
6805 | * Build sched domains for a given set of cpus and attach the sched domains |
6806 | * to the individual cpus | |
1da177e4 | 6807 | */ |
51888ca2 | 6808 | static int build_sched_domains(const cpumask_t *cpu_map) |
1da177e4 LT |
6809 | { |
6810 | int i; | |
57d885fe | 6811 | struct root_domain *rd; |
7c16ec58 MT |
6812 | SCHED_CPUMASK_DECLARE(allmasks); |
6813 | cpumask_t *tmpmask; | |
d1b55138 JH |
6814 | #ifdef CONFIG_NUMA |
6815 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 6816 | int sd_allnodes = 0; |
d1b55138 JH |
6817 | |
6818 | /* | |
6819 | * Allocate the per-node list of sched groups | |
6820 | */ | |
5cf9f062 | 6821 | sched_group_nodes = kcalloc(MAX_NUMNODES, sizeof(struct sched_group *), |
41a2d6cf | 6822 | GFP_KERNEL); |
d1b55138 JH |
6823 | if (!sched_group_nodes) { |
6824 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 6825 | return -ENOMEM; |
d1b55138 | 6826 | } |
d1b55138 | 6827 | #endif |
1da177e4 | 6828 | |
dc938520 | 6829 | rd = alloc_rootdomain(); |
57d885fe GH |
6830 | if (!rd) { |
6831 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
7c16ec58 MT |
6832 | #ifdef CONFIG_NUMA |
6833 | kfree(sched_group_nodes); | |
6834 | #endif | |
57d885fe GH |
6835 | return -ENOMEM; |
6836 | } | |
6837 | ||
7c16ec58 MT |
6838 | #if SCHED_CPUMASK_ALLOC |
6839 | /* get space for all scratch cpumask variables */ | |
6840 | allmasks = kmalloc(sizeof(*allmasks), GFP_KERNEL); | |
6841 | if (!allmasks) { | |
6842 | printk(KERN_WARNING "Cannot alloc cpumask array\n"); | |
6843 | kfree(rd); | |
6844 | #ifdef CONFIG_NUMA | |
6845 | kfree(sched_group_nodes); | |
6846 | #endif | |
6847 | return -ENOMEM; | |
6848 | } | |
6849 | #endif | |
6850 | tmpmask = (cpumask_t *)allmasks; | |
6851 | ||
6852 | ||
6853 | #ifdef CONFIG_NUMA | |
6854 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
6855 | #endif | |
6856 | ||
1da177e4 | 6857 | /* |
1a20ff27 | 6858 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 6859 | */ |
1a20ff27 | 6860 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 | 6861 | struct sched_domain *sd = NULL, *p; |
7c16ec58 | 6862 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
1da177e4 | 6863 | |
7c16ec58 MT |
6864 | *nodemask = node_to_cpumask(cpu_to_node(i)); |
6865 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
1da177e4 LT |
6866 | |
6867 | #ifdef CONFIG_NUMA | |
dd41f596 | 6868 | if (cpus_weight(*cpu_map) > |
7c16ec58 | 6869 | SD_NODES_PER_DOMAIN*cpus_weight(*nodemask)) { |
9c1cfda2 | 6870 | sd = &per_cpu(allnodes_domains, i); |
7c16ec58 | 6871 | SD_INIT(sd, ALLNODES); |
9c1cfda2 | 6872 | sd->span = *cpu_map; |
7c16ec58 | 6873 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 6874 | p = sd; |
6711cab4 | 6875 | sd_allnodes = 1; |
9c1cfda2 JH |
6876 | } else |
6877 | p = NULL; | |
6878 | ||
1da177e4 | 6879 | sd = &per_cpu(node_domains, i); |
7c16ec58 | 6880 | SD_INIT(sd, NODE); |
4bdbaad3 | 6881 | sched_domain_node_span(cpu_to_node(i), &sd->span); |
9c1cfda2 | 6882 | sd->parent = p; |
1a848870 SS |
6883 | if (p) |
6884 | p->child = sd; | |
9c1cfda2 | 6885 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
6886 | #endif |
6887 | ||
6888 | p = sd; | |
6889 | sd = &per_cpu(phys_domains, i); | |
7c16ec58 MT |
6890 | SD_INIT(sd, CPU); |
6891 | sd->span = *nodemask; | |
1da177e4 | 6892 | sd->parent = p; |
1a848870 SS |
6893 | if (p) |
6894 | p->child = sd; | |
7c16ec58 | 6895 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 6896 | |
1e9f28fa SS |
6897 | #ifdef CONFIG_SCHED_MC |
6898 | p = sd; | |
6899 | sd = &per_cpu(core_domains, i); | |
7c16ec58 | 6900 | SD_INIT(sd, MC); |
1e9f28fa SS |
6901 | sd->span = cpu_coregroup_map(i); |
6902 | cpus_and(sd->span, sd->span, *cpu_map); | |
6903 | sd->parent = p; | |
1a848870 | 6904 | p->child = sd; |
7c16ec58 | 6905 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
6906 | #endif |
6907 | ||
1da177e4 LT |
6908 | #ifdef CONFIG_SCHED_SMT |
6909 | p = sd; | |
6910 | sd = &per_cpu(cpu_domains, i); | |
7c16ec58 | 6911 | SD_INIT(sd, SIBLING); |
d5a7430d | 6912 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6913 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 6914 | sd->parent = p; |
1a848870 | 6915 | p->child = sd; |
7c16ec58 | 6916 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
6917 | #endif |
6918 | } | |
6919 | ||
6920 | #ifdef CONFIG_SCHED_SMT | |
6921 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 6922 | for_each_cpu_mask(i, *cpu_map) { |
7c16ec58 MT |
6923 | SCHED_CPUMASK_VAR(this_sibling_map, allmasks); |
6924 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
6925 | ||
6926 | *this_sibling_map = per_cpu(cpu_sibling_map, i); | |
6927 | cpus_and(*this_sibling_map, *this_sibling_map, *cpu_map); | |
6928 | if (i != first_cpu(*this_sibling_map)) | |
1da177e4 LT |
6929 | continue; |
6930 | ||
dd41f596 | 6931 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
6932 | &cpu_to_cpu_group, |
6933 | send_covered, tmpmask); | |
1da177e4 LT |
6934 | } |
6935 | #endif | |
6936 | ||
1e9f28fa SS |
6937 | #ifdef CONFIG_SCHED_MC |
6938 | /* Set up multi-core groups */ | |
6939 | for_each_cpu_mask(i, *cpu_map) { | |
7c16ec58 MT |
6940 | SCHED_CPUMASK_VAR(this_core_map, allmasks); |
6941 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
6942 | ||
6943 | *this_core_map = cpu_coregroup_map(i); | |
6944 | cpus_and(*this_core_map, *this_core_map, *cpu_map); | |
6945 | if (i != first_cpu(*this_core_map)) | |
1e9f28fa | 6946 | continue; |
7c16ec58 | 6947 | |
dd41f596 | 6948 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
6949 | &cpu_to_core_group, |
6950 | send_covered, tmpmask); | |
1e9f28fa SS |
6951 | } |
6952 | #endif | |
6953 | ||
1da177e4 LT |
6954 | /* Set up physical groups */ |
6955 | for (i = 0; i < MAX_NUMNODES; i++) { | |
7c16ec58 MT |
6956 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
6957 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
1da177e4 | 6958 | |
7c16ec58 MT |
6959 | *nodemask = node_to_cpumask(i); |
6960 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
6961 | if (cpus_empty(*nodemask)) | |
1da177e4 LT |
6962 | continue; |
6963 | ||
7c16ec58 MT |
6964 | init_sched_build_groups(nodemask, cpu_map, |
6965 | &cpu_to_phys_group, | |
6966 | send_covered, tmpmask); | |
1da177e4 LT |
6967 | } |
6968 | ||
6969 | #ifdef CONFIG_NUMA | |
6970 | /* Set up node groups */ | |
7c16ec58 MT |
6971 | if (sd_allnodes) { |
6972 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
6973 | ||
6974 | init_sched_build_groups(cpu_map, cpu_map, | |
6975 | &cpu_to_allnodes_group, | |
6976 | send_covered, tmpmask); | |
6977 | } | |
9c1cfda2 JH |
6978 | |
6979 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6980 | /* Set up node groups */ | |
6981 | struct sched_group *sg, *prev; | |
7c16ec58 MT |
6982 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
6983 | SCHED_CPUMASK_VAR(domainspan, allmasks); | |
6984 | SCHED_CPUMASK_VAR(covered, allmasks); | |
9c1cfda2 JH |
6985 | int j; |
6986 | ||
7c16ec58 MT |
6987 | *nodemask = node_to_cpumask(i); |
6988 | cpus_clear(*covered); | |
6989 | ||
6990 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
6991 | if (cpus_empty(*nodemask)) { | |
d1b55138 | 6992 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 6993 | continue; |
d1b55138 | 6994 | } |
9c1cfda2 | 6995 | |
4bdbaad3 | 6996 | sched_domain_node_span(i, domainspan); |
7c16ec58 | 6997 | cpus_and(*domainspan, *domainspan, *cpu_map); |
9c1cfda2 | 6998 | |
15f0b676 | 6999 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
7000 | if (!sg) { |
7001 | printk(KERN_WARNING "Can not alloc domain group for " | |
7002 | "node %d\n", i); | |
7003 | goto error; | |
7004 | } | |
9c1cfda2 | 7005 | sched_group_nodes[i] = sg; |
7c16ec58 | 7006 | for_each_cpu_mask(j, *nodemask) { |
9c1cfda2 | 7007 | struct sched_domain *sd; |
9761eea8 | 7008 | |
9c1cfda2 JH |
7009 | sd = &per_cpu(node_domains, j); |
7010 | sd->groups = sg; | |
9c1cfda2 | 7011 | } |
5517d86b | 7012 | sg->__cpu_power = 0; |
7c16ec58 | 7013 | sg->cpumask = *nodemask; |
51888ca2 | 7014 | sg->next = sg; |
7c16ec58 | 7015 | cpus_or(*covered, *covered, *nodemask); |
9c1cfda2 JH |
7016 | prev = sg; |
7017 | ||
7018 | for (j = 0; j < MAX_NUMNODES; j++) { | |
7c16ec58 | 7019 | SCHED_CPUMASK_VAR(notcovered, allmasks); |
9c1cfda2 | 7020 | int n = (i + j) % MAX_NUMNODES; |
c5f59f08 | 7021 | node_to_cpumask_ptr(pnodemask, n); |
9c1cfda2 | 7022 | |
7c16ec58 MT |
7023 | cpus_complement(*notcovered, *covered); |
7024 | cpus_and(*tmpmask, *notcovered, *cpu_map); | |
7025 | cpus_and(*tmpmask, *tmpmask, *domainspan); | |
7026 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7027 | break; |
7028 | ||
7c16ec58 MT |
7029 | cpus_and(*tmpmask, *tmpmask, *pnodemask); |
7030 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7031 | continue; |
7032 | ||
15f0b676 SV |
7033 | sg = kmalloc_node(sizeof(struct sched_group), |
7034 | GFP_KERNEL, i); | |
9c1cfda2 JH |
7035 | if (!sg) { |
7036 | printk(KERN_WARNING | |
7037 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 7038 | goto error; |
9c1cfda2 | 7039 | } |
5517d86b | 7040 | sg->__cpu_power = 0; |
7c16ec58 | 7041 | sg->cpumask = *tmpmask; |
51888ca2 | 7042 | sg->next = prev->next; |
7c16ec58 | 7043 | cpus_or(*covered, *covered, *tmpmask); |
9c1cfda2 JH |
7044 | prev->next = sg; |
7045 | prev = sg; | |
7046 | } | |
9c1cfda2 | 7047 | } |
1da177e4 LT |
7048 | #endif |
7049 | ||
7050 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7051 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 7052 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
7053 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
7054 | ||
89c4710e | 7055 | init_sched_groups_power(i, sd); |
5c45bf27 | 7056 | } |
1da177e4 | 7057 | #endif |
1e9f28fa | 7058 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 7059 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
7060 | struct sched_domain *sd = &per_cpu(core_domains, i); |
7061 | ||
89c4710e | 7062 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7063 | } |
7064 | #endif | |
1e9f28fa | 7065 | |
5c45bf27 | 7066 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
7067 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
7068 | ||
89c4710e | 7069 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7070 | } |
7071 | ||
9c1cfda2 | 7072 | #ifdef CONFIG_NUMA |
08069033 SS |
7073 | for (i = 0; i < MAX_NUMNODES; i++) |
7074 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 7075 | |
6711cab4 SS |
7076 | if (sd_allnodes) { |
7077 | struct sched_group *sg; | |
f712c0c7 | 7078 | |
7c16ec58 MT |
7079 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg, |
7080 | tmpmask); | |
f712c0c7 SS |
7081 | init_numa_sched_groups_power(sg); |
7082 | } | |
9c1cfda2 JH |
7083 | #endif |
7084 | ||
1da177e4 | 7085 | /* Attach the domains */ |
1a20ff27 | 7086 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
7087 | struct sched_domain *sd; |
7088 | #ifdef CONFIG_SCHED_SMT | |
7089 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
7090 | #elif defined(CONFIG_SCHED_MC) |
7091 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
7092 | #else |
7093 | sd = &per_cpu(phys_domains, i); | |
7094 | #endif | |
57d885fe | 7095 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 7096 | } |
51888ca2 | 7097 | |
7c16ec58 | 7098 | SCHED_CPUMASK_FREE((void *)allmasks); |
51888ca2 SV |
7099 | return 0; |
7100 | ||
a616058b | 7101 | #ifdef CONFIG_NUMA |
51888ca2 | 7102 | error: |
7c16ec58 MT |
7103 | free_sched_groups(cpu_map, tmpmask); |
7104 | SCHED_CPUMASK_FREE((void *)allmasks); | |
51888ca2 | 7105 | return -ENOMEM; |
a616058b | 7106 | #endif |
1da177e4 | 7107 | } |
029190c5 PJ |
7108 | |
7109 | static cpumask_t *doms_cur; /* current sched domains */ | |
7110 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
7111 | ||
7112 | /* | |
7113 | * Special case: If a kmalloc of a doms_cur partition (array of | |
7114 | * cpumask_t) fails, then fallback to a single sched domain, | |
7115 | * as determined by the single cpumask_t fallback_doms. | |
7116 | */ | |
7117 | static cpumask_t fallback_doms; | |
7118 | ||
22e52b07 HC |
7119 | void __attribute__((weak)) arch_update_cpu_topology(void) |
7120 | { | |
7121 | } | |
7122 | ||
1a20ff27 | 7123 | /* |
41a2d6cf | 7124 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7125 | * For now this just excludes isolated cpus, but could be used to |
7126 | * exclude other special cases in the future. | |
1a20ff27 | 7127 | */ |
51888ca2 | 7128 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7129 | { |
7378547f MM |
7130 | int err; |
7131 | ||
22e52b07 | 7132 | arch_update_cpu_topology(); |
029190c5 PJ |
7133 | ndoms_cur = 1; |
7134 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
7135 | if (!doms_cur) | |
7136 | doms_cur = &fallback_doms; | |
7137 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
7378547f | 7138 | err = build_sched_domains(doms_cur); |
6382bc90 | 7139 | register_sched_domain_sysctl(); |
7378547f MM |
7140 | |
7141 | return err; | |
1a20ff27 DG |
7142 | } |
7143 | ||
7c16ec58 MT |
7144 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map, |
7145 | cpumask_t *tmpmask) | |
1da177e4 | 7146 | { |
7c16ec58 | 7147 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7148 | } |
1da177e4 | 7149 | |
1a20ff27 DG |
7150 | /* |
7151 | * Detach sched domains from a group of cpus specified in cpu_map | |
7152 | * These cpus will now be attached to the NULL domain | |
7153 | */ | |
858119e1 | 7154 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7155 | { |
7c16ec58 | 7156 | cpumask_t tmpmask; |
1a20ff27 DG |
7157 | int i; |
7158 | ||
6382bc90 MM |
7159 | unregister_sched_domain_sysctl(); |
7160 | ||
1a20ff27 | 7161 | for_each_cpu_mask(i, *cpu_map) |
57d885fe | 7162 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7163 | synchronize_sched(); |
7c16ec58 | 7164 | arch_destroy_sched_domains(cpu_map, &tmpmask); |
1a20ff27 DG |
7165 | } |
7166 | ||
029190c5 PJ |
7167 | /* |
7168 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7169 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7170 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7171 | * It destroys each deleted domain and builds each new domain. | |
7172 | * | |
7173 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
41a2d6cf IM |
7174 | * The masks don't intersect (don't overlap.) We should setup one |
7175 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7176 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7177 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7178 | * it as it is. | |
7179 | * | |
41a2d6cf IM |
7180 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
7181 | * ownership of it and will kfree it when done with it. If the caller | |
029190c5 PJ |
7182 | * failed the kmalloc call, then it can pass in doms_new == NULL, |
7183 | * and partition_sched_domains() will fallback to the single partition | |
7184 | * 'fallback_doms'. | |
7185 | * | |
7186 | * Call with hotplug lock held | |
7187 | */ | |
7188 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new) | |
7189 | { | |
7190 | int i, j; | |
7191 | ||
a1835615 SV |
7192 | lock_doms_cur(); |
7193 | ||
7378547f MM |
7194 | /* always unregister in case we don't destroy any domains */ |
7195 | unregister_sched_domain_sysctl(); | |
7196 | ||
029190c5 PJ |
7197 | if (doms_new == NULL) { |
7198 | ndoms_new = 1; | |
7199 | doms_new = &fallback_doms; | |
7200 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
7201 | } | |
7202 | ||
7203 | /* Destroy deleted domains */ | |
7204 | for (i = 0; i < ndoms_cur; i++) { | |
7205 | for (j = 0; j < ndoms_new; j++) { | |
7206 | if (cpus_equal(doms_cur[i], doms_new[j])) | |
7207 | goto match1; | |
7208 | } | |
7209 | /* no match - a current sched domain not in new doms_new[] */ | |
7210 | detach_destroy_domains(doms_cur + i); | |
7211 | match1: | |
7212 | ; | |
7213 | } | |
7214 | ||
7215 | /* Build new domains */ | |
7216 | for (i = 0; i < ndoms_new; i++) { | |
7217 | for (j = 0; j < ndoms_cur; j++) { | |
7218 | if (cpus_equal(doms_new[i], doms_cur[j])) | |
7219 | goto match2; | |
7220 | } | |
7221 | /* no match - add a new doms_new */ | |
7222 | build_sched_domains(doms_new + i); | |
7223 | match2: | |
7224 | ; | |
7225 | } | |
7226 | ||
7227 | /* Remember the new sched domains */ | |
7228 | if (doms_cur != &fallback_doms) | |
7229 | kfree(doms_cur); | |
7230 | doms_cur = doms_new; | |
7231 | ndoms_cur = ndoms_new; | |
7378547f MM |
7232 | |
7233 | register_sched_domain_sysctl(); | |
a1835615 SV |
7234 | |
7235 | unlock_doms_cur(); | |
029190c5 PJ |
7236 | } |
7237 | ||
5c45bf27 | 7238 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
9aefd0ab | 7239 | int arch_reinit_sched_domains(void) |
5c45bf27 SS |
7240 | { |
7241 | int err; | |
7242 | ||
95402b38 | 7243 | get_online_cpus(); |
5c45bf27 SS |
7244 | detach_destroy_domains(&cpu_online_map); |
7245 | err = arch_init_sched_domains(&cpu_online_map); | |
95402b38 | 7246 | put_online_cpus(); |
5c45bf27 SS |
7247 | |
7248 | return err; | |
7249 | } | |
7250 | ||
7251 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7252 | { | |
7253 | int ret; | |
7254 | ||
7255 | if (buf[0] != '0' && buf[0] != '1') | |
7256 | return -EINVAL; | |
7257 | ||
7258 | if (smt) | |
7259 | sched_smt_power_savings = (buf[0] == '1'); | |
7260 | else | |
7261 | sched_mc_power_savings = (buf[0] == '1'); | |
7262 | ||
7263 | ret = arch_reinit_sched_domains(); | |
7264 | ||
7265 | return ret ? ret : count; | |
7266 | } | |
7267 | ||
5c45bf27 SS |
7268 | #ifdef CONFIG_SCHED_MC |
7269 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
7270 | { | |
7271 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7272 | } | |
48f24c4d IM |
7273 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
7274 | const char *buf, size_t count) | |
5c45bf27 SS |
7275 | { |
7276 | return sched_power_savings_store(buf, count, 0); | |
7277 | } | |
6707de00 AB |
7278 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
7279 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7280 | #endif |
7281 | ||
7282 | #ifdef CONFIG_SCHED_SMT | |
7283 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
7284 | { | |
7285 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7286 | } | |
48f24c4d IM |
7287 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
7288 | const char *buf, size_t count) | |
5c45bf27 SS |
7289 | { |
7290 | return sched_power_savings_store(buf, count, 1); | |
7291 | } | |
6707de00 AB |
7292 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
7293 | sched_smt_power_savings_store); | |
7294 | #endif | |
7295 | ||
7296 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
7297 | { | |
7298 | int err = 0; | |
7299 | ||
7300 | #ifdef CONFIG_SCHED_SMT | |
7301 | if (smt_capable()) | |
7302 | err = sysfs_create_file(&cls->kset.kobj, | |
7303 | &attr_sched_smt_power_savings.attr); | |
7304 | #endif | |
7305 | #ifdef CONFIG_SCHED_MC | |
7306 | if (!err && mc_capable()) | |
7307 | err = sysfs_create_file(&cls->kset.kobj, | |
7308 | &attr_sched_mc_power_savings.attr); | |
7309 | #endif | |
7310 | return err; | |
7311 | } | |
5c45bf27 SS |
7312 | #endif |
7313 | ||
1da177e4 | 7314 | /* |
41a2d6cf | 7315 | * Force a reinitialization of the sched domains hierarchy. The domains |
1da177e4 | 7316 | * and groups cannot be updated in place without racing with the balancing |
41c7ce9a | 7317 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
7318 | * which will prevent rebalancing while the sched domains are recalculated. |
7319 | */ | |
7320 | static int update_sched_domains(struct notifier_block *nfb, | |
7321 | unsigned long action, void *hcpu) | |
7322 | { | |
1da177e4 LT |
7323 | switch (action) { |
7324 | case CPU_UP_PREPARE: | |
8bb78442 | 7325 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 7326 | case CPU_DOWN_PREPARE: |
8bb78442 | 7327 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 7328 | detach_destroy_domains(&cpu_online_map); |
1da177e4 LT |
7329 | return NOTIFY_OK; |
7330 | ||
7331 | case CPU_UP_CANCELED: | |
8bb78442 | 7332 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 7333 | case CPU_DOWN_FAILED: |
8bb78442 | 7334 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7335 | case CPU_ONLINE: |
8bb78442 | 7336 | case CPU_ONLINE_FROZEN: |
1da177e4 | 7337 | case CPU_DEAD: |
8bb78442 | 7338 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
7339 | /* |
7340 | * Fall through and re-initialise the domains. | |
7341 | */ | |
7342 | break; | |
7343 | default: | |
7344 | return NOTIFY_DONE; | |
7345 | } | |
7346 | ||
7347 | /* The hotplug lock is already held by cpu_up/cpu_down */ | |
1a20ff27 | 7348 | arch_init_sched_domains(&cpu_online_map); |
1da177e4 LT |
7349 | |
7350 | return NOTIFY_OK; | |
7351 | } | |
1da177e4 LT |
7352 | |
7353 | void __init sched_init_smp(void) | |
7354 | { | |
5c1e1767 NP |
7355 | cpumask_t non_isolated_cpus; |
7356 | ||
434d53b0 MT |
7357 | #if defined(CONFIG_NUMA) |
7358 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7359 | GFP_KERNEL); | |
7360 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7361 | #endif | |
95402b38 | 7362 | get_online_cpus(); |
1a20ff27 | 7363 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 7364 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
7365 | if (cpus_empty(non_isolated_cpus)) |
7366 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
95402b38 | 7367 | put_online_cpus(); |
1da177e4 LT |
7368 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7369 | hotcpu_notifier(update_sched_domains, 0); | |
5c1e1767 NP |
7370 | |
7371 | /* Move init over to a non-isolated CPU */ | |
7c16ec58 | 7372 | if (set_cpus_allowed_ptr(current, &non_isolated_cpus) < 0) |
5c1e1767 | 7373 | BUG(); |
19978ca6 | 7374 | sched_init_granularity(); |
1da177e4 LT |
7375 | } |
7376 | #else | |
7377 | void __init sched_init_smp(void) | |
7378 | { | |
434d53b0 MT |
7379 | #if defined(CONFIG_NUMA) |
7380 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7381 | GFP_KERNEL); | |
7382 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7383 | #endif | |
19978ca6 | 7384 | sched_init_granularity(); |
1da177e4 LT |
7385 | } |
7386 | #endif /* CONFIG_SMP */ | |
7387 | ||
7388 | int in_sched_functions(unsigned long addr) | |
7389 | { | |
1da177e4 LT |
7390 | return in_lock_functions(addr) || |
7391 | (addr >= (unsigned long)__sched_text_start | |
7392 | && addr < (unsigned long)__sched_text_end); | |
7393 | } | |
7394 | ||
a9957449 | 7395 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7396 | { |
7397 | cfs_rq->tasks_timeline = RB_ROOT; | |
dd41f596 IM |
7398 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7399 | cfs_rq->rq = rq; | |
7400 | #endif | |
67e9fb2a | 7401 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7402 | } |
7403 | ||
fa85ae24 PZ |
7404 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7405 | { | |
7406 | struct rt_prio_array *array; | |
7407 | int i; | |
7408 | ||
7409 | array = &rt_rq->active; | |
7410 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7411 | INIT_LIST_HEAD(array->queue + i); | |
7412 | __clear_bit(i, array->bitmap); | |
7413 | } | |
7414 | /* delimiter for bitsearch: */ | |
7415 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7416 | ||
052f1dc7 | 7417 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
48d5e258 PZ |
7418 | rt_rq->highest_prio = MAX_RT_PRIO; |
7419 | #endif | |
fa85ae24 PZ |
7420 | #ifdef CONFIG_SMP |
7421 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 PZ |
7422 | rt_rq->overloaded = 0; |
7423 | #endif | |
7424 | ||
7425 | rt_rq->rt_time = 0; | |
7426 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
7427 | rt_rq->rt_runtime = 0; |
7428 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 7429 | |
052f1dc7 | 7430 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7431 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7432 | rt_rq->rq = rq; |
7433 | #endif | |
fa85ae24 PZ |
7434 | } |
7435 | ||
6f505b16 PZ |
7436 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7437 | static void init_tg_cfs_entry(struct rq *rq, struct task_group *tg, | |
7438 | struct cfs_rq *cfs_rq, struct sched_entity *se, | |
7439 | int cpu, int add) | |
7440 | { | |
7441 | tg->cfs_rq[cpu] = cfs_rq; | |
7442 | init_cfs_rq(cfs_rq, rq); | |
7443 | cfs_rq->tg = tg; | |
7444 | if (add) | |
7445 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7446 | ||
7447 | tg->se[cpu] = se; | |
7448 | se->cfs_rq = &rq->cfs; | |
7449 | se->my_q = cfs_rq; | |
7450 | se->load.weight = tg->shares; | |
7451 | se->load.inv_weight = div64_64(1ULL<<32, se->load.weight); | |
7452 | se->parent = NULL; | |
7453 | } | |
052f1dc7 | 7454 | #endif |
6f505b16 | 7455 | |
052f1dc7 | 7456 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
7457 | static void init_tg_rt_entry(struct rq *rq, struct task_group *tg, |
7458 | struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, | |
7459 | int cpu, int add) | |
7460 | { | |
7461 | tg->rt_rq[cpu] = rt_rq; | |
7462 | init_rt_rq(rt_rq, rq); | |
7463 | rt_rq->tg = tg; | |
7464 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 7465 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7466 | if (add) |
7467 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
7468 | ||
7469 | tg->rt_se[cpu] = rt_se; | |
7470 | rt_se->rt_rq = &rq->rt; | |
7471 | rt_se->my_q = rt_rq; | |
7472 | rt_se->parent = NULL; | |
7473 | INIT_LIST_HEAD(&rt_se->run_list); | |
7474 | } | |
7475 | #endif | |
7476 | ||
1da177e4 LT |
7477 | void __init sched_init(void) |
7478 | { | |
dd41f596 | 7479 | int i, j; |
434d53b0 MT |
7480 | unsigned long alloc_size = 0, ptr; |
7481 | ||
7482 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7483 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7484 | #endif | |
7485 | #ifdef CONFIG_RT_GROUP_SCHED | |
7486 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7487 | #endif | |
7488 | /* | |
7489 | * As sched_init() is called before page_alloc is setup, | |
7490 | * we use alloc_bootmem(). | |
7491 | */ | |
7492 | if (alloc_size) { | |
7493 | ptr = (unsigned long)alloc_bootmem_low(alloc_size); | |
7494 | ||
7495 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7496 | init_task_group.se = (struct sched_entity **)ptr; | |
7497 | ptr += nr_cpu_ids * sizeof(void **); | |
7498 | ||
7499 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7500 | ptr += nr_cpu_ids * sizeof(void **); | |
7501 | #endif | |
7502 | #ifdef CONFIG_RT_GROUP_SCHED | |
7503 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7504 | ptr += nr_cpu_ids * sizeof(void **); | |
7505 | ||
7506 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
7507 | #endif | |
7508 | } | |
dd41f596 | 7509 | |
57d885fe GH |
7510 | #ifdef CONFIG_SMP |
7511 | init_defrootdomain(); | |
7512 | #endif | |
7513 | ||
d0b27fa7 PZ |
7514 | init_rt_bandwidth(&def_rt_bandwidth, |
7515 | global_rt_period(), global_rt_runtime()); | |
7516 | ||
7517 | #ifdef CONFIG_RT_GROUP_SCHED | |
7518 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
7519 | global_rt_period(), global_rt_runtime()); | |
7520 | #endif | |
7521 | ||
052f1dc7 | 7522 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 PZ |
7523 | list_add(&init_task_group.list, &task_groups); |
7524 | #endif | |
7525 | ||
0a945022 | 7526 | for_each_possible_cpu(i) { |
70b97a7f | 7527 | struct rq *rq; |
1da177e4 LT |
7528 | |
7529 | rq = cpu_rq(i); | |
7530 | spin_lock_init(&rq->lock); | |
fcb99371 | 7531 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 7532 | rq->nr_running = 0; |
dd41f596 | 7533 | rq->clock = 1; |
15934a37 | 7534 | update_last_tick_seen(rq); |
dd41f596 | 7535 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7536 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7537 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 7538 | init_task_group.shares = init_task_group_load; |
6f505b16 PZ |
7539 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
7540 | init_tg_cfs_entry(rq, &init_task_group, | |
7541 | &per_cpu(init_cfs_rq, i), | |
7542 | &per_cpu(init_sched_entity, i), i, 1); | |
7543 | ||
052f1dc7 PZ |
7544 | #endif |
7545 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 PZ |
7546 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
7547 | init_tg_rt_entry(rq, &init_task_group, | |
7548 | &per_cpu(init_rt_rq, i), | |
7549 | &per_cpu(init_sched_rt_entity, i), i, 1); | |
ac086bc2 PZ |
7550 | #else |
7551 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
dd41f596 | 7552 | #endif |
1da177e4 | 7553 | |
dd41f596 IM |
7554 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7555 | rq->cpu_load[j] = 0; | |
1da177e4 | 7556 | #ifdef CONFIG_SMP |
41c7ce9a | 7557 | rq->sd = NULL; |
57d885fe | 7558 | rq->rd = NULL; |
1da177e4 | 7559 | rq->active_balance = 0; |
dd41f596 | 7560 | rq->next_balance = jiffies; |
1da177e4 | 7561 | rq->push_cpu = 0; |
0a2966b4 | 7562 | rq->cpu = i; |
1da177e4 LT |
7563 | rq->migration_thread = NULL; |
7564 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 7565 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 7566 | #endif |
8f4d37ec | 7567 | init_rq_hrtick(rq); |
1da177e4 | 7568 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7569 | } |
7570 | ||
2dd73a4f | 7571 | set_load_weight(&init_task); |
b50f60ce | 7572 | |
e107be36 AK |
7573 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7574 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7575 | #endif | |
7576 | ||
c9819f45 CL |
7577 | #ifdef CONFIG_SMP |
7578 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); | |
7579 | #endif | |
7580 | ||
b50f60ce HC |
7581 | #ifdef CONFIG_RT_MUTEXES |
7582 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
7583 | #endif | |
7584 | ||
1da177e4 LT |
7585 | /* |
7586 | * The boot idle thread does lazy MMU switching as well: | |
7587 | */ | |
7588 | atomic_inc(&init_mm.mm_count); | |
7589 | enter_lazy_tlb(&init_mm, current); | |
7590 | ||
7591 | /* | |
7592 | * Make us the idle thread. Technically, schedule() should not be | |
7593 | * called from this thread, however somewhere below it might be, | |
7594 | * but because we are the idle thread, we just pick up running again | |
7595 | * when this runqueue becomes "idle". | |
7596 | */ | |
7597 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
7598 | /* |
7599 | * During early bootup we pretend to be a normal task: | |
7600 | */ | |
7601 | current->sched_class = &fair_sched_class; | |
6892b75e IM |
7602 | |
7603 | scheduler_running = 1; | |
1da177e4 LT |
7604 | } |
7605 | ||
7606 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
7607 | void __might_sleep(char *file, int line) | |
7608 | { | |
48f24c4d | 7609 | #ifdef in_atomic |
1da177e4 LT |
7610 | static unsigned long prev_jiffy; /* ratelimiting */ |
7611 | ||
7612 | if ((in_atomic() || irqs_disabled()) && | |
7613 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
7614 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7615 | return; | |
7616 | prev_jiffy = jiffies; | |
91368d73 | 7617 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
7618 | " context at %s:%d\n", file, line); |
7619 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
7620 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 7621 | debug_show_held_locks(current); |
3117df04 IM |
7622 | if (irqs_disabled()) |
7623 | print_irqtrace_events(current); | |
1da177e4 LT |
7624 | dump_stack(); |
7625 | } | |
7626 | #endif | |
7627 | } | |
7628 | EXPORT_SYMBOL(__might_sleep); | |
7629 | #endif | |
7630 | ||
7631 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7632 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7633 | { | |
7634 | int on_rq; | |
7635 | update_rq_clock(rq); | |
7636 | on_rq = p->se.on_rq; | |
7637 | if (on_rq) | |
7638 | deactivate_task(rq, p, 0); | |
7639 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
7640 | if (on_rq) { | |
7641 | activate_task(rq, p, 0); | |
7642 | resched_task(rq->curr); | |
7643 | } | |
7644 | } | |
7645 | ||
1da177e4 LT |
7646 | void normalize_rt_tasks(void) |
7647 | { | |
a0f98a1c | 7648 | struct task_struct *g, *p; |
1da177e4 | 7649 | unsigned long flags; |
70b97a7f | 7650 | struct rq *rq; |
1da177e4 | 7651 | |
4cf5d77a | 7652 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7653 | do_each_thread(g, p) { |
178be793 IM |
7654 | /* |
7655 | * Only normalize user tasks: | |
7656 | */ | |
7657 | if (!p->mm) | |
7658 | continue; | |
7659 | ||
6cfb0d5d | 7660 | p->se.exec_start = 0; |
6cfb0d5d | 7661 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 7662 | p->se.wait_start = 0; |
dd41f596 | 7663 | p->se.sleep_start = 0; |
dd41f596 | 7664 | p->se.block_start = 0; |
6cfb0d5d | 7665 | #endif |
dd41f596 IM |
7666 | task_rq(p)->clock = 0; |
7667 | ||
7668 | if (!rt_task(p)) { | |
7669 | /* | |
7670 | * Renice negative nice level userspace | |
7671 | * tasks back to 0: | |
7672 | */ | |
7673 | if (TASK_NICE(p) < 0 && p->mm) | |
7674 | set_user_nice(p, 0); | |
1da177e4 | 7675 | continue; |
dd41f596 | 7676 | } |
1da177e4 | 7677 | |
4cf5d77a | 7678 | spin_lock(&p->pi_lock); |
b29739f9 | 7679 | rq = __task_rq_lock(p); |
1da177e4 | 7680 | |
178be793 | 7681 | normalize_task(rq, p); |
3a5e4dc1 | 7682 | |
b29739f9 | 7683 | __task_rq_unlock(rq); |
4cf5d77a | 7684 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7685 | } while_each_thread(g, p); |
7686 | ||
4cf5d77a | 7687 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7688 | } |
7689 | ||
7690 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
7691 | |
7692 | #ifdef CONFIG_IA64 | |
7693 | /* | |
7694 | * These functions are only useful for the IA64 MCA handling. | |
7695 | * | |
7696 | * They can only be called when the whole system has been | |
7697 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7698 | * activity can take place. Using them for anything else would | |
7699 | * be a serious bug, and as a result, they aren't even visible | |
7700 | * under any other configuration. | |
7701 | */ | |
7702 | ||
7703 | /** | |
7704 | * curr_task - return the current task for a given cpu. | |
7705 | * @cpu: the processor in question. | |
7706 | * | |
7707 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7708 | */ | |
36c8b586 | 7709 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7710 | { |
7711 | return cpu_curr(cpu); | |
7712 | } | |
7713 | ||
7714 | /** | |
7715 | * set_curr_task - set the current task for a given cpu. | |
7716 | * @cpu: the processor in question. | |
7717 | * @p: the task pointer to set. | |
7718 | * | |
7719 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7720 | * are serviced on a separate stack. It allows the architecture to switch the |
7721 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7722 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7723 | * and caller must save the original value of the current task (see | |
7724 | * curr_task() above) and restore that value before reenabling interrupts and | |
7725 | * re-starting the system. | |
7726 | * | |
7727 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7728 | */ | |
36c8b586 | 7729 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7730 | { |
7731 | cpu_curr(cpu) = p; | |
7732 | } | |
7733 | ||
7734 | #endif | |
29f59db3 | 7735 | |
bccbe08a PZ |
7736 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7737 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
7738 | { |
7739 | int i; | |
7740 | ||
7741 | for_each_possible_cpu(i) { | |
7742 | if (tg->cfs_rq) | |
7743 | kfree(tg->cfs_rq[i]); | |
7744 | if (tg->se) | |
7745 | kfree(tg->se[i]); | |
6f505b16 PZ |
7746 | } |
7747 | ||
7748 | kfree(tg->cfs_rq); | |
7749 | kfree(tg->se); | |
6f505b16 PZ |
7750 | } |
7751 | ||
bccbe08a | 7752 | static int alloc_fair_sched_group(struct task_group *tg) |
29f59db3 | 7753 | { |
29f59db3 SV |
7754 | struct cfs_rq *cfs_rq; |
7755 | struct sched_entity *se; | |
9b5b7751 | 7756 | struct rq *rq; |
29f59db3 SV |
7757 | int i; |
7758 | ||
434d53b0 | 7759 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
7760 | if (!tg->cfs_rq) |
7761 | goto err; | |
434d53b0 | 7762 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
7763 | if (!tg->se) |
7764 | goto err; | |
052f1dc7 PZ |
7765 | |
7766 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
7767 | |
7768 | for_each_possible_cpu(i) { | |
9b5b7751 | 7769 | rq = cpu_rq(i); |
29f59db3 | 7770 | |
6f505b16 PZ |
7771 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), |
7772 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
7773 | if (!cfs_rq) |
7774 | goto err; | |
7775 | ||
6f505b16 PZ |
7776 | se = kmalloc_node(sizeof(struct sched_entity), |
7777 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
7778 | if (!se) |
7779 | goto err; | |
7780 | ||
052f1dc7 | 7781 | init_tg_cfs_entry(rq, tg, cfs_rq, se, i, 0); |
bccbe08a PZ |
7782 | } |
7783 | ||
7784 | return 1; | |
7785 | ||
7786 | err: | |
7787 | return 0; | |
7788 | } | |
7789 | ||
7790 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
7791 | { | |
7792 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
7793 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
7794 | } | |
7795 | ||
7796 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
7797 | { | |
7798 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
7799 | } | |
7800 | #else | |
7801 | static inline void free_fair_sched_group(struct task_group *tg) | |
7802 | { | |
7803 | } | |
7804 | ||
7805 | static inline int alloc_fair_sched_group(struct task_group *tg) | |
7806 | { | |
7807 | return 1; | |
7808 | } | |
7809 | ||
7810 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
7811 | { | |
7812 | } | |
7813 | ||
7814 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
7815 | { | |
7816 | } | |
052f1dc7 PZ |
7817 | #endif |
7818 | ||
7819 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
7820 | static void free_rt_sched_group(struct task_group *tg) |
7821 | { | |
7822 | int i; | |
7823 | ||
d0b27fa7 PZ |
7824 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
7825 | ||
bccbe08a PZ |
7826 | for_each_possible_cpu(i) { |
7827 | if (tg->rt_rq) | |
7828 | kfree(tg->rt_rq[i]); | |
7829 | if (tg->rt_se) | |
7830 | kfree(tg->rt_se[i]); | |
7831 | } | |
7832 | ||
7833 | kfree(tg->rt_rq); | |
7834 | kfree(tg->rt_se); | |
7835 | } | |
7836 | ||
7837 | static int alloc_rt_sched_group(struct task_group *tg) | |
7838 | { | |
7839 | struct rt_rq *rt_rq; | |
7840 | struct sched_rt_entity *rt_se; | |
7841 | struct rq *rq; | |
7842 | int i; | |
7843 | ||
434d53b0 | 7844 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
7845 | if (!tg->rt_rq) |
7846 | goto err; | |
434d53b0 | 7847 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
7848 | if (!tg->rt_se) |
7849 | goto err; | |
7850 | ||
d0b27fa7 PZ |
7851 | init_rt_bandwidth(&tg->rt_bandwidth, |
7852 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
7853 | |
7854 | for_each_possible_cpu(i) { | |
7855 | rq = cpu_rq(i); | |
7856 | ||
6f505b16 PZ |
7857 | rt_rq = kmalloc_node(sizeof(struct rt_rq), |
7858 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
7859 | if (!rt_rq) | |
7860 | goto err; | |
29f59db3 | 7861 | |
6f505b16 PZ |
7862 | rt_se = kmalloc_node(sizeof(struct sched_rt_entity), |
7863 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
7864 | if (!rt_se) | |
7865 | goto err; | |
29f59db3 | 7866 | |
6f505b16 | 7867 | init_tg_rt_entry(rq, tg, rt_rq, rt_se, i, 0); |
29f59db3 SV |
7868 | } |
7869 | ||
bccbe08a PZ |
7870 | return 1; |
7871 | ||
7872 | err: | |
7873 | return 0; | |
7874 | } | |
7875 | ||
7876 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
7877 | { | |
7878 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
7879 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
7880 | } | |
7881 | ||
7882 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
7883 | { | |
7884 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
7885 | } | |
7886 | #else | |
7887 | static inline void free_rt_sched_group(struct task_group *tg) | |
7888 | { | |
7889 | } | |
7890 | ||
7891 | static inline int alloc_rt_sched_group(struct task_group *tg) | |
7892 | { | |
7893 | return 1; | |
7894 | } | |
7895 | ||
7896 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
7897 | { | |
7898 | } | |
7899 | ||
7900 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
7901 | { | |
7902 | } | |
7903 | #endif | |
7904 | ||
d0b27fa7 | 7905 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
7906 | static void free_sched_group(struct task_group *tg) |
7907 | { | |
7908 | free_fair_sched_group(tg); | |
7909 | free_rt_sched_group(tg); | |
7910 | kfree(tg); | |
7911 | } | |
7912 | ||
7913 | /* allocate runqueue etc for a new task group */ | |
7914 | struct task_group *sched_create_group(void) | |
7915 | { | |
7916 | struct task_group *tg; | |
7917 | unsigned long flags; | |
7918 | int i; | |
7919 | ||
7920 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7921 | if (!tg) | |
7922 | return ERR_PTR(-ENOMEM); | |
7923 | ||
7924 | if (!alloc_fair_sched_group(tg)) | |
7925 | goto err; | |
7926 | ||
7927 | if (!alloc_rt_sched_group(tg)) | |
7928 | goto err; | |
7929 | ||
8ed36996 | 7930 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 7931 | for_each_possible_cpu(i) { |
bccbe08a PZ |
7932 | register_fair_sched_group(tg, i); |
7933 | register_rt_sched_group(tg, i); | |
9b5b7751 | 7934 | } |
6f505b16 | 7935 | list_add_rcu(&tg->list, &task_groups); |
8ed36996 | 7936 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 7937 | |
9b5b7751 | 7938 | return tg; |
29f59db3 SV |
7939 | |
7940 | err: | |
6f505b16 | 7941 | free_sched_group(tg); |
29f59db3 SV |
7942 | return ERR_PTR(-ENOMEM); |
7943 | } | |
7944 | ||
9b5b7751 | 7945 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7946 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7947 | { |
29f59db3 | 7948 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7949 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7950 | } |
7951 | ||
9b5b7751 | 7952 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7953 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 7954 | { |
8ed36996 | 7955 | unsigned long flags; |
9b5b7751 | 7956 | int i; |
29f59db3 | 7957 | |
8ed36996 | 7958 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 7959 | for_each_possible_cpu(i) { |
bccbe08a PZ |
7960 | unregister_fair_sched_group(tg, i); |
7961 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 7962 | } |
6f505b16 | 7963 | list_del_rcu(&tg->list); |
8ed36996 | 7964 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 7965 | |
9b5b7751 | 7966 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 7967 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
7968 | } |
7969 | ||
9b5b7751 | 7970 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7971 | * The caller of this function should have put the task in its new group |
7972 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7973 | * reflect its new group. | |
9b5b7751 SV |
7974 | */ |
7975 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
7976 | { |
7977 | int on_rq, running; | |
7978 | unsigned long flags; | |
7979 | struct rq *rq; | |
7980 | ||
7981 | rq = task_rq_lock(tsk, &flags); | |
7982 | ||
29f59db3 SV |
7983 | update_rq_clock(rq); |
7984 | ||
051a1d1a | 7985 | running = task_current(rq, tsk); |
29f59db3 SV |
7986 | on_rq = tsk->se.on_rq; |
7987 | ||
0e1f3483 | 7988 | if (on_rq) |
29f59db3 | 7989 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
7990 | if (unlikely(running)) |
7991 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 7992 | |
6f505b16 | 7993 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 7994 | |
810b3817 PZ |
7995 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7996 | if (tsk->sched_class->moved_group) | |
7997 | tsk->sched_class->moved_group(tsk); | |
7998 | #endif | |
7999 | ||
0e1f3483 HS |
8000 | if (unlikely(running)) |
8001 | tsk->sched_class->set_curr_task(rq); | |
8002 | if (on_rq) | |
7074badb | 8003 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8004 | |
29f59db3 SV |
8005 | task_rq_unlock(rq, &flags); |
8006 | } | |
d0b27fa7 | 8007 | #endif |
29f59db3 | 8008 | |
052f1dc7 | 8009 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
8010 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8011 | { | |
8012 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8013 | struct rq *rq = cfs_rq->rq; | |
8014 | int on_rq; | |
8015 | ||
62fb1851 | 8016 | spin_lock_irq(&rq->lock); |
29f59db3 SV |
8017 | |
8018 | on_rq = se->on_rq; | |
62fb1851 | 8019 | if (on_rq) |
29f59db3 SV |
8020 | dequeue_entity(cfs_rq, se, 0); |
8021 | ||
8022 | se->load.weight = shares; | |
8023 | se->load.inv_weight = div64_64((1ULL<<32), shares); | |
8024 | ||
62fb1851 | 8025 | if (on_rq) |
29f59db3 | 8026 | enqueue_entity(cfs_rq, se, 0); |
62fb1851 PZ |
8027 | |
8028 | spin_unlock_irq(&rq->lock); | |
29f59db3 SV |
8029 | } |
8030 | ||
8ed36996 PZ |
8031 | static DEFINE_MUTEX(shares_mutex); |
8032 | ||
4cf86d77 | 8033 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8034 | { |
8035 | int i; | |
8ed36996 | 8036 | unsigned long flags; |
c61935fd | 8037 | |
62fb1851 PZ |
8038 | /* |
8039 | * A weight of 0 or 1 can cause arithmetics problems. | |
8040 | * (The default weight is 1024 - so there's no practical | |
8041 | * limitation from this.) | |
8042 | */ | |
8043 | if (shares < 2) | |
8044 | shares = 2; | |
8045 | ||
8ed36996 | 8046 | mutex_lock(&shares_mutex); |
9b5b7751 | 8047 | if (tg->shares == shares) |
5cb350ba | 8048 | goto done; |
29f59db3 | 8049 | |
8ed36996 | 8050 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8051 | for_each_possible_cpu(i) |
8052 | unregister_fair_sched_group(tg, i); | |
8ed36996 | 8053 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8054 | |
8055 | /* wait for any ongoing reference to this group to finish */ | |
8056 | synchronize_sched(); | |
8057 | ||
8058 | /* | |
8059 | * Now we are free to modify the group's share on each cpu | |
8060 | * w/o tripping rebalance_share or load_balance_fair. | |
8061 | */ | |
9b5b7751 | 8062 | tg->shares = shares; |
62fb1851 | 8063 | for_each_possible_cpu(i) |
9b5b7751 | 8064 | set_se_shares(tg->se[i], shares); |
29f59db3 | 8065 | |
6b2d7700 SV |
8066 | /* |
8067 | * Enable load balance activity on this group, by inserting it back on | |
8068 | * each cpu's rq->leaf_cfs_rq_list. | |
8069 | */ | |
8ed36996 | 8070 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8071 | for_each_possible_cpu(i) |
8072 | register_fair_sched_group(tg, i); | |
8ed36996 | 8073 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8074 | done: |
8ed36996 | 8075 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8076 | return 0; |
29f59db3 SV |
8077 | } |
8078 | ||
5cb350ba DG |
8079 | unsigned long sched_group_shares(struct task_group *tg) |
8080 | { | |
8081 | return tg->shares; | |
8082 | } | |
052f1dc7 | 8083 | #endif |
5cb350ba | 8084 | |
052f1dc7 | 8085 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8086 | /* |
9f0c1e56 | 8087 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8088 | */ |
9f0c1e56 PZ |
8089 | static DEFINE_MUTEX(rt_constraints_mutex); |
8090 | ||
8091 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8092 | { | |
8093 | if (runtime == RUNTIME_INF) | |
8094 | return 1ULL << 16; | |
8095 | ||
2692a240 | 8096 | return div64_64(runtime << 16, period); |
9f0c1e56 PZ |
8097 | } |
8098 | ||
8099 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) | |
6f505b16 PZ |
8100 | { |
8101 | struct task_group *tgi; | |
8102 | unsigned long total = 0; | |
9f0c1e56 | 8103 | unsigned long global_ratio = |
d0b27fa7 | 8104 | to_ratio(global_rt_period(), global_rt_runtime()); |
6f505b16 PZ |
8105 | |
8106 | rcu_read_lock(); | |
9f0c1e56 PZ |
8107 | list_for_each_entry_rcu(tgi, &task_groups, list) { |
8108 | if (tgi == tg) | |
8109 | continue; | |
6f505b16 | 8110 | |
d0b27fa7 PZ |
8111 | total += to_ratio(ktime_to_ns(tgi->rt_bandwidth.rt_period), |
8112 | tgi->rt_bandwidth.rt_runtime); | |
9f0c1e56 PZ |
8113 | } |
8114 | rcu_read_unlock(); | |
6f505b16 | 8115 | |
9f0c1e56 | 8116 | return total + to_ratio(period, runtime) < global_ratio; |
6f505b16 PZ |
8117 | } |
8118 | ||
521f1a24 DG |
8119 | /* Must be called with tasklist_lock held */ |
8120 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
8121 | { | |
8122 | struct task_struct *g, *p; | |
8123 | do_each_thread(g, p) { | |
8124 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8125 | return 1; | |
8126 | } while_each_thread(g, p); | |
8127 | return 0; | |
8128 | } | |
8129 | ||
d0b27fa7 PZ |
8130 | static int tg_set_bandwidth(struct task_group *tg, |
8131 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8132 | { |
ac086bc2 | 8133 | int i, err = 0; |
9f0c1e56 | 8134 | |
9f0c1e56 | 8135 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8136 | read_lock(&tasklist_lock); |
ac086bc2 | 8137 | if (rt_runtime == 0 && tg_has_rt_tasks(tg)) { |
521f1a24 DG |
8138 | err = -EBUSY; |
8139 | goto unlock; | |
8140 | } | |
9f0c1e56 PZ |
8141 | if (!__rt_schedulable(tg, rt_period, rt_runtime)) { |
8142 | err = -EINVAL; | |
8143 | goto unlock; | |
8144 | } | |
ac086bc2 PZ |
8145 | |
8146 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
8147 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8148 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8149 | |
8150 | for_each_possible_cpu(i) { | |
8151 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8152 | ||
8153 | spin_lock(&rt_rq->rt_runtime_lock); | |
8154 | rt_rq->rt_runtime = rt_runtime; | |
8155 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8156 | } | |
8157 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 8158 | unlock: |
521f1a24 | 8159 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8160 | mutex_unlock(&rt_constraints_mutex); |
8161 | ||
8162 | return err; | |
6f505b16 PZ |
8163 | } |
8164 | ||
d0b27fa7 PZ |
8165 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8166 | { | |
8167 | u64 rt_runtime, rt_period; | |
8168 | ||
8169 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8170 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8171 | if (rt_runtime_us < 0) | |
8172 | rt_runtime = RUNTIME_INF; | |
8173 | ||
8174 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8175 | } | |
8176 | ||
9f0c1e56 PZ |
8177 | long sched_group_rt_runtime(struct task_group *tg) |
8178 | { | |
8179 | u64 rt_runtime_us; | |
8180 | ||
d0b27fa7 | 8181 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8182 | return -1; |
8183 | ||
d0b27fa7 | 8184 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8185 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8186 | return rt_runtime_us; | |
8187 | } | |
d0b27fa7 PZ |
8188 | |
8189 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8190 | { | |
8191 | u64 rt_runtime, rt_period; | |
8192 | ||
8193 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8194 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8195 | ||
8196 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8197 | } | |
8198 | ||
8199 | long sched_group_rt_period(struct task_group *tg) | |
8200 | { | |
8201 | u64 rt_period_us; | |
8202 | ||
8203 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8204 | do_div(rt_period_us, NSEC_PER_USEC); | |
8205 | return rt_period_us; | |
8206 | } | |
8207 | ||
8208 | static int sched_rt_global_constraints(void) | |
8209 | { | |
8210 | int ret = 0; | |
8211 | ||
8212 | mutex_lock(&rt_constraints_mutex); | |
8213 | if (!__rt_schedulable(NULL, 1, 0)) | |
8214 | ret = -EINVAL; | |
8215 | mutex_unlock(&rt_constraints_mutex); | |
8216 | ||
8217 | return ret; | |
8218 | } | |
8219 | #else | |
8220 | static int sched_rt_global_constraints(void) | |
8221 | { | |
ac086bc2 PZ |
8222 | unsigned long flags; |
8223 | int i; | |
8224 | ||
8225 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); | |
8226 | for_each_possible_cpu(i) { | |
8227 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8228 | ||
8229 | spin_lock(&rt_rq->rt_runtime_lock); | |
8230 | rt_rq->rt_runtime = global_rt_runtime(); | |
8231 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8232 | } | |
8233 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
8234 | ||
d0b27fa7 PZ |
8235 | return 0; |
8236 | } | |
052f1dc7 | 8237 | #endif |
d0b27fa7 PZ |
8238 | |
8239 | int sched_rt_handler(struct ctl_table *table, int write, | |
8240 | struct file *filp, void __user *buffer, size_t *lenp, | |
8241 | loff_t *ppos) | |
8242 | { | |
8243 | int ret; | |
8244 | int old_period, old_runtime; | |
8245 | static DEFINE_MUTEX(mutex); | |
8246 | ||
8247 | mutex_lock(&mutex); | |
8248 | old_period = sysctl_sched_rt_period; | |
8249 | old_runtime = sysctl_sched_rt_runtime; | |
8250 | ||
8251 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
8252 | ||
8253 | if (!ret && write) { | |
8254 | ret = sched_rt_global_constraints(); | |
8255 | if (ret) { | |
8256 | sysctl_sched_rt_period = old_period; | |
8257 | sysctl_sched_rt_runtime = old_runtime; | |
8258 | } else { | |
8259 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8260 | def_rt_bandwidth.rt_period = | |
8261 | ns_to_ktime(global_rt_period()); | |
8262 | } | |
8263 | } | |
8264 | mutex_unlock(&mutex); | |
8265 | ||
8266 | return ret; | |
8267 | } | |
68318b8e | 8268 | |
052f1dc7 | 8269 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8270 | |
8271 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8272 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8273 | { |
2b01dfe3 PM |
8274 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8275 | struct task_group, css); | |
68318b8e SV |
8276 | } |
8277 | ||
8278 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8279 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e SV |
8280 | { |
8281 | struct task_group *tg; | |
8282 | ||
2b01dfe3 | 8283 | if (!cgrp->parent) { |
68318b8e | 8284 | /* This is early initialization for the top cgroup */ |
2b01dfe3 | 8285 | init_task_group.css.cgroup = cgrp; |
68318b8e SV |
8286 | return &init_task_group.css; |
8287 | } | |
8288 | ||
8289 | /* we support only 1-level deep hierarchical scheduler atm */ | |
2b01dfe3 | 8290 | if (cgrp->parent->parent) |
68318b8e SV |
8291 | return ERR_PTR(-EINVAL); |
8292 | ||
8293 | tg = sched_create_group(); | |
8294 | if (IS_ERR(tg)) | |
8295 | return ERR_PTR(-ENOMEM); | |
8296 | ||
8297 | /* Bind the cgroup to task_group object we just created */ | |
2b01dfe3 | 8298 | tg->css.cgroup = cgrp; |
68318b8e SV |
8299 | |
8300 | return &tg->css; | |
8301 | } | |
8302 | ||
41a2d6cf IM |
8303 | static void |
8304 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8305 | { |
2b01dfe3 | 8306 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8307 | |
8308 | sched_destroy_group(tg); | |
8309 | } | |
8310 | ||
41a2d6cf IM |
8311 | static int |
8312 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8313 | struct task_struct *tsk) | |
68318b8e | 8314 | { |
b68aa230 PZ |
8315 | #ifdef CONFIG_RT_GROUP_SCHED |
8316 | /* Don't accept realtime tasks when there is no way for them to run */ | |
d0b27fa7 | 8317 | if (rt_task(tsk) && cgroup_tg(cgrp)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
8318 | return -EINVAL; |
8319 | #else | |
68318b8e SV |
8320 | /* We don't support RT-tasks being in separate groups */ |
8321 | if (tsk->sched_class != &fair_sched_class) | |
8322 | return -EINVAL; | |
b68aa230 | 8323 | #endif |
68318b8e SV |
8324 | |
8325 | return 0; | |
8326 | } | |
8327 | ||
8328 | static void | |
2b01dfe3 | 8329 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
8330 | struct cgroup *old_cont, struct task_struct *tsk) |
8331 | { | |
8332 | sched_move_task(tsk); | |
8333 | } | |
8334 | ||
052f1dc7 | 8335 | #ifdef CONFIG_FAIR_GROUP_SCHED |
2b01dfe3 PM |
8336 | static int cpu_shares_write_uint(struct cgroup *cgrp, struct cftype *cftype, |
8337 | u64 shareval) | |
68318b8e | 8338 | { |
2b01dfe3 | 8339 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8340 | } |
8341 | ||
2b01dfe3 | 8342 | static u64 cpu_shares_read_uint(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8343 | { |
2b01dfe3 | 8344 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8345 | |
8346 | return (u64) tg->shares; | |
8347 | } | |
052f1dc7 | 8348 | #endif |
68318b8e | 8349 | |
052f1dc7 | 8350 | #ifdef CONFIG_RT_GROUP_SCHED |
ac086bc2 | 8351 | static ssize_t cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
9f0c1e56 PZ |
8352 | struct file *file, |
8353 | const char __user *userbuf, | |
8354 | size_t nbytes, loff_t *unused_ppos) | |
6f505b16 | 8355 | { |
9f0c1e56 PZ |
8356 | char buffer[64]; |
8357 | int retval = 0; | |
8358 | s64 val; | |
8359 | char *end; | |
8360 | ||
8361 | if (!nbytes) | |
8362 | return -EINVAL; | |
8363 | if (nbytes >= sizeof(buffer)) | |
8364 | return -E2BIG; | |
8365 | if (copy_from_user(buffer, userbuf, nbytes)) | |
8366 | return -EFAULT; | |
8367 | ||
8368 | buffer[nbytes] = 0; /* nul-terminate */ | |
8369 | ||
8370 | /* strip newline if necessary */ | |
8371 | if (nbytes && (buffer[nbytes-1] == '\n')) | |
8372 | buffer[nbytes-1] = 0; | |
8373 | val = simple_strtoll(buffer, &end, 0); | |
8374 | if (*end) | |
8375 | return -EINVAL; | |
8376 | ||
8377 | /* Pass to subsystem */ | |
8378 | retval = sched_group_set_rt_runtime(cgroup_tg(cgrp), val); | |
8379 | if (!retval) | |
8380 | retval = nbytes; | |
8381 | return retval; | |
6f505b16 PZ |
8382 | } |
8383 | ||
9f0c1e56 PZ |
8384 | static ssize_t cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft, |
8385 | struct file *file, | |
8386 | char __user *buf, size_t nbytes, | |
8387 | loff_t *ppos) | |
6f505b16 | 8388 | { |
9f0c1e56 PZ |
8389 | char tmp[64]; |
8390 | long val = sched_group_rt_runtime(cgroup_tg(cgrp)); | |
8391 | int len = sprintf(tmp, "%ld\n", val); | |
6f505b16 | 8392 | |
9f0c1e56 | 8393 | return simple_read_from_buffer(buf, nbytes, ppos, tmp, len); |
6f505b16 | 8394 | } |
d0b27fa7 PZ |
8395 | |
8396 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8397 | u64 rt_period_us) | |
8398 | { | |
8399 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8400 | } | |
8401 | ||
8402 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8403 | { | |
8404 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8405 | } | |
052f1dc7 | 8406 | #endif |
6f505b16 | 8407 | |
fe5c7cc2 | 8408 | static struct cftype cpu_files[] = { |
052f1dc7 | 8409 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8410 | { |
8411 | .name = "shares", | |
8412 | .read_uint = cpu_shares_read_uint, | |
8413 | .write_uint = cpu_shares_write_uint, | |
8414 | }, | |
052f1dc7 PZ |
8415 | #endif |
8416 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8417 | { |
9f0c1e56 PZ |
8418 | .name = "rt_runtime_us", |
8419 | .read = cpu_rt_runtime_read, | |
8420 | .write = cpu_rt_runtime_write, | |
6f505b16 | 8421 | }, |
d0b27fa7 PZ |
8422 | { |
8423 | .name = "rt_period_us", | |
8424 | .read_uint = cpu_rt_period_read_uint, | |
8425 | .write_uint = cpu_rt_period_write_uint, | |
8426 | }, | |
052f1dc7 | 8427 | #endif |
68318b8e SV |
8428 | }; |
8429 | ||
8430 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8431 | { | |
fe5c7cc2 | 8432 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8433 | } |
8434 | ||
8435 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8436 | .name = "cpu", |
8437 | .create = cpu_cgroup_create, | |
8438 | .destroy = cpu_cgroup_destroy, | |
8439 | .can_attach = cpu_cgroup_can_attach, | |
8440 | .attach = cpu_cgroup_attach, | |
8441 | .populate = cpu_cgroup_populate, | |
8442 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8443 | .early_init = 1, |
8444 | }; | |
8445 | ||
052f1dc7 | 8446 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8447 | |
8448 | #ifdef CONFIG_CGROUP_CPUACCT | |
8449 | ||
8450 | /* | |
8451 | * CPU accounting code for task groups. | |
8452 | * | |
8453 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8454 | * (balbir@in.ibm.com). | |
8455 | */ | |
8456 | ||
8457 | /* track cpu usage of a group of tasks */ | |
8458 | struct cpuacct { | |
8459 | struct cgroup_subsys_state css; | |
8460 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
8461 | u64 *cpuusage; | |
8462 | }; | |
8463 | ||
8464 | struct cgroup_subsys cpuacct_subsys; | |
8465 | ||
8466 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8467 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8468 | { |
32cd756a | 8469 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8470 | struct cpuacct, css); |
8471 | } | |
8472 | ||
8473 | /* return cpu accounting group to which this task belongs */ | |
8474 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8475 | { | |
8476 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8477 | struct cpuacct, css); | |
8478 | } | |
8479 | ||
8480 | /* create a new cpu accounting group */ | |
8481 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8482 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
8483 | { |
8484 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
8485 | ||
8486 | if (!ca) | |
8487 | return ERR_PTR(-ENOMEM); | |
8488 | ||
8489 | ca->cpuusage = alloc_percpu(u64); | |
8490 | if (!ca->cpuusage) { | |
8491 | kfree(ca); | |
8492 | return ERR_PTR(-ENOMEM); | |
8493 | } | |
8494 | ||
8495 | return &ca->css; | |
8496 | } | |
8497 | ||
8498 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 8499 | static void |
32cd756a | 8500 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8501 | { |
32cd756a | 8502 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8503 | |
8504 | free_percpu(ca->cpuusage); | |
8505 | kfree(ca); | |
8506 | } | |
8507 | ||
8508 | /* return total cpu usage (in nanoseconds) of a group */ | |
32cd756a | 8509 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 8510 | { |
32cd756a | 8511 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8512 | u64 totalcpuusage = 0; |
8513 | int i; | |
8514 | ||
8515 | for_each_possible_cpu(i) { | |
8516 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
8517 | ||
8518 | /* | |
8519 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
8520 | * platforms. | |
8521 | */ | |
8522 | spin_lock_irq(&cpu_rq(i)->lock); | |
8523 | totalcpuusage += *cpuusage; | |
8524 | spin_unlock_irq(&cpu_rq(i)->lock); | |
8525 | } | |
8526 | ||
8527 | return totalcpuusage; | |
8528 | } | |
8529 | ||
0297b803 DG |
8530 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
8531 | u64 reset) | |
8532 | { | |
8533 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8534 | int err = 0; | |
8535 | int i; | |
8536 | ||
8537 | if (reset) { | |
8538 | err = -EINVAL; | |
8539 | goto out; | |
8540 | } | |
8541 | ||
8542 | for_each_possible_cpu(i) { | |
8543 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
8544 | ||
8545 | spin_lock_irq(&cpu_rq(i)->lock); | |
8546 | *cpuusage = 0; | |
8547 | spin_unlock_irq(&cpu_rq(i)->lock); | |
8548 | } | |
8549 | out: | |
8550 | return err; | |
8551 | } | |
8552 | ||
d842de87 SV |
8553 | static struct cftype files[] = { |
8554 | { | |
8555 | .name = "usage", | |
8556 | .read_uint = cpuusage_read, | |
0297b803 | 8557 | .write_uint = cpuusage_write, |
d842de87 SV |
8558 | }, |
8559 | }; | |
8560 | ||
32cd756a | 8561 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8562 | { |
32cd756a | 8563 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
8564 | } |
8565 | ||
8566 | /* | |
8567 | * charge this task's execution time to its accounting group. | |
8568 | * | |
8569 | * called with rq->lock held. | |
8570 | */ | |
8571 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
8572 | { | |
8573 | struct cpuacct *ca; | |
8574 | ||
8575 | if (!cpuacct_subsys.active) | |
8576 | return; | |
8577 | ||
8578 | ca = task_ca(tsk); | |
8579 | if (ca) { | |
8580 | u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk)); | |
8581 | ||
8582 | *cpuusage += cputime; | |
8583 | } | |
8584 | } | |
8585 | ||
8586 | struct cgroup_subsys cpuacct_subsys = { | |
8587 | .name = "cpuacct", | |
8588 | .create = cpuacct_create, | |
8589 | .destroy = cpuacct_destroy, | |
8590 | .populate = cpuacct_populate, | |
8591 | .subsys_id = cpuacct_subsys_id, | |
8592 | }; | |
8593 | #endif /* CONFIG_CGROUP_CPUACCT */ |