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