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> |
1da177e4 | 68 | |
5517d86b | 69 | #include <asm/tlb.h> |
838225b4 | 70 | #include <asm/irq_regs.h> |
1da177e4 | 71 | |
b035b6de AD |
72 | /* |
73 | * Scheduler clock - returns current time in nanosec units. | |
74 | * This is default implementation. | |
75 | * Architectures and sub-architectures can override this. | |
76 | */ | |
77 | unsigned long long __attribute__((weak)) sched_clock(void) | |
78 | { | |
d6322faf | 79 | return (unsigned long long)jiffies * (NSEC_PER_SEC / HZ); |
b035b6de AD |
80 | } |
81 | ||
1da177e4 LT |
82 | /* |
83 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
84 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
85 | * and back. | |
86 | */ | |
87 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
88 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
89 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
90 | ||
91 | /* | |
92 | * 'User priority' is the nice value converted to something we | |
93 | * can work with better when scaling various scheduler parameters, | |
94 | * it's a [ 0 ... 39 ] range. | |
95 | */ | |
96 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
97 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
98 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
99 | ||
100 | /* | |
d7876a08 | 101 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 102 | */ |
d6322faf | 103 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 104 | |
6aa645ea IM |
105 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
106 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
107 | ||
1da177e4 LT |
108 | /* |
109 | * These are the 'tuning knobs' of the scheduler: | |
110 | * | |
a4ec24b4 | 111 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
112 | * Timeslices get refilled after they expire. |
113 | */ | |
1da177e4 | 114 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 115 | |
5517d86b ED |
116 | #ifdef CONFIG_SMP |
117 | /* | |
118 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
119 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
120 | */ | |
121 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
122 | { | |
123 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
124 | } | |
125 | ||
126 | /* | |
127 | * Each time a sched group cpu_power is changed, | |
128 | * we must compute its reciprocal value | |
129 | */ | |
130 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
131 | { | |
132 | sg->__cpu_power += val; | |
133 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
134 | } | |
135 | #endif | |
136 | ||
e05606d3 IM |
137 | static inline int rt_policy(int policy) |
138 | { | |
139 | if (unlikely(policy == SCHED_FIFO) || unlikely(policy == SCHED_RR)) | |
140 | return 1; | |
141 | return 0; | |
142 | } | |
143 | ||
144 | static inline int task_has_rt_policy(struct task_struct *p) | |
145 | { | |
146 | return rt_policy(p->policy); | |
147 | } | |
148 | ||
1da177e4 | 149 | /* |
6aa645ea | 150 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 151 | */ |
6aa645ea IM |
152 | struct rt_prio_array { |
153 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
154 | struct list_head queue[MAX_RT_PRIO]; | |
155 | }; | |
156 | ||
29f59db3 SV |
157 | #ifdef CONFIG_FAIR_GROUP_SCHED |
158 | ||
68318b8e SV |
159 | #include <linux/cgroup.h> |
160 | ||
29f59db3 SV |
161 | struct cfs_rq; |
162 | ||
163 | /* task group related information */ | |
4cf86d77 | 164 | struct task_group { |
68318b8e SV |
165 | #ifdef CONFIG_FAIR_CGROUP_SCHED |
166 | struct cgroup_subsys_state css; | |
167 | #endif | |
29f59db3 SV |
168 | /* schedulable entities of this group on each cpu */ |
169 | struct sched_entity **se; | |
170 | /* runqueue "owned" by this group on each cpu */ | |
171 | struct cfs_rq **cfs_rq; | |
6b2d7700 SV |
172 | |
173 | /* | |
174 | * shares assigned to a task group governs how much of cpu bandwidth | |
175 | * is allocated to the group. The more shares a group has, the more is | |
176 | * the cpu bandwidth allocated to it. | |
177 | * | |
178 | * For ex, lets say that there are three task groups, A, B and C which | |
179 | * have been assigned shares 1000, 2000 and 3000 respectively. Then, | |
180 | * cpu bandwidth allocated by the scheduler to task groups A, B and C | |
181 | * should be: | |
182 | * | |
183 | * Bw(A) = 1000/(1000+2000+3000) * 100 = 16.66% | |
184 | * Bw(B) = 2000/(1000+2000+3000) * 100 = 33.33% | |
03319ec8 | 185 | * Bw(C) = 3000/(1000+2000+3000) * 100 = 50% |
6b2d7700 SV |
186 | * |
187 | * The weight assigned to a task group's schedulable entities on every | |
188 | * cpu (task_group.se[a_cpu]->load.weight) is derived from the task | |
189 | * group's shares. For ex: lets say that task group A has been | |
190 | * assigned shares of 1000 and there are two CPUs in a system. Then, | |
191 | * | |
192 | * tg_A->se[0]->load.weight = tg_A->se[1]->load.weight = 1000; | |
193 | * | |
194 | * Note: It's not necessary that each of a task's group schedulable | |
03319ec8 IM |
195 | * entity have the same weight on all CPUs. If the group |
196 | * has 2 of its tasks on CPU0 and 1 task on CPU1, then a | |
197 | * better distribution of weight could be: | |
6b2d7700 SV |
198 | * |
199 | * tg_A->se[0]->load.weight = 2/3 * 2000 = 1333 | |
200 | * tg_A->se[1]->load.weight = 1/2 * 2000 = 667 | |
201 | * | |
202 | * rebalance_shares() is responsible for distributing the shares of a | |
203 | * task groups like this among the group's schedulable entities across | |
204 | * cpus. | |
205 | * | |
206 | */ | |
29f59db3 | 207 | unsigned long shares; |
6b2d7700 | 208 | |
ae8393e5 | 209 | struct rcu_head rcu; |
29f59db3 SV |
210 | }; |
211 | ||
212 | /* Default task group's sched entity on each cpu */ | |
213 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
214 | /* Default task group's cfs_rq on each cpu */ | |
215 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
216 | ||
9b5b7751 SV |
217 | static struct sched_entity *init_sched_entity_p[NR_CPUS]; |
218 | static struct cfs_rq *init_cfs_rq_p[NR_CPUS]; | |
29f59db3 | 219 | |
ec2c507f SV |
220 | /* task_group_mutex serializes add/remove of task groups and also changes to |
221 | * a task group's cpu shares. | |
222 | */ | |
223 | static DEFINE_MUTEX(task_group_mutex); | |
224 | ||
a1835615 SV |
225 | /* doms_cur_mutex serializes access to doms_cur[] array */ |
226 | static DEFINE_MUTEX(doms_cur_mutex); | |
227 | ||
6b2d7700 SV |
228 | #ifdef CONFIG_SMP |
229 | /* kernel thread that runs rebalance_shares() periodically */ | |
230 | static struct task_struct *lb_monitor_task; | |
231 | static int load_balance_monitor(void *unused); | |
232 | #endif | |
233 | ||
234 | static void set_se_shares(struct sched_entity *se, unsigned long shares); | |
235 | ||
29f59db3 | 236 | /* Default task group. |
3a252015 | 237 | * Every task in system belong to this group at bootup. |
29f59db3 | 238 | */ |
4cf86d77 | 239 | struct task_group init_task_group = { |
0eab9146 | 240 | .se = init_sched_entity_p, |
3a252015 IM |
241 | .cfs_rq = init_cfs_rq_p, |
242 | }; | |
9b5b7751 | 243 | |
24e377a8 | 244 | #ifdef CONFIG_FAIR_USER_SCHED |
0eab9146 | 245 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) |
24e377a8 | 246 | #else |
93f992cc | 247 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
24e377a8 SV |
248 | #endif |
249 | ||
0eab9146 | 250 | #define MIN_GROUP_SHARES 2 |
6b2d7700 | 251 | |
93f992cc | 252 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
29f59db3 SV |
253 | |
254 | /* return group to which a task belongs */ | |
4cf86d77 | 255 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 256 | { |
4cf86d77 | 257 | struct task_group *tg; |
9b5b7751 | 258 | |
24e377a8 SV |
259 | #ifdef CONFIG_FAIR_USER_SCHED |
260 | tg = p->user->tg; | |
68318b8e SV |
261 | #elif defined(CONFIG_FAIR_CGROUP_SCHED) |
262 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), | |
263 | struct task_group, css); | |
24e377a8 | 264 | #else |
41a2d6cf | 265 | tg = &init_task_group; |
24e377a8 | 266 | #endif |
9b5b7751 | 267 | return tg; |
29f59db3 SV |
268 | } |
269 | ||
270 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
ce96b5ac | 271 | static inline void set_task_cfs_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 272 | { |
ce96b5ac DA |
273 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
274 | p->se.parent = task_group(p)->se[cpu]; | |
29f59db3 SV |
275 | } |
276 | ||
ec2c507f SV |
277 | static inline void lock_task_group_list(void) |
278 | { | |
279 | mutex_lock(&task_group_mutex); | |
280 | } | |
281 | ||
282 | static inline void unlock_task_group_list(void) | |
283 | { | |
284 | mutex_unlock(&task_group_mutex); | |
285 | } | |
286 | ||
a1835615 SV |
287 | static inline void lock_doms_cur(void) |
288 | { | |
289 | mutex_lock(&doms_cur_mutex); | |
290 | } | |
291 | ||
292 | static inline void unlock_doms_cur(void) | |
293 | { | |
294 | mutex_unlock(&doms_cur_mutex); | |
295 | } | |
296 | ||
29f59db3 SV |
297 | #else |
298 | ||
ce96b5ac | 299 | static inline void set_task_cfs_rq(struct task_struct *p, unsigned int cpu) { } |
ec2c507f SV |
300 | static inline void lock_task_group_list(void) { } |
301 | static inline void unlock_task_group_list(void) { } | |
a1835615 SV |
302 | static inline void lock_doms_cur(void) { } |
303 | static inline void unlock_doms_cur(void) { } | |
29f59db3 SV |
304 | |
305 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
306 | ||
6aa645ea IM |
307 | /* CFS-related fields in a runqueue */ |
308 | struct cfs_rq { | |
309 | struct load_weight load; | |
310 | unsigned long nr_running; | |
311 | ||
6aa645ea | 312 | u64 exec_clock; |
e9acbff6 | 313 | u64 min_vruntime; |
6aa645ea IM |
314 | |
315 | struct rb_root tasks_timeline; | |
316 | struct rb_node *rb_leftmost; | |
317 | struct rb_node *rb_load_balance_curr; | |
6aa645ea IM |
318 | /* 'curr' points to currently running entity on this cfs_rq. |
319 | * It is set to NULL otherwise (i.e when none are currently running). | |
320 | */ | |
321 | struct sched_entity *curr; | |
ddc97297 PZ |
322 | |
323 | unsigned long nr_spread_over; | |
324 | ||
62160e3f | 325 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
326 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
327 | ||
41a2d6cf IM |
328 | /* |
329 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
330 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
331 | * (like users, containers etc.) | |
332 | * | |
333 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
334 | * list is used during load balance. | |
335 | */ | |
41a2d6cf IM |
336 | struct list_head leaf_cfs_rq_list; |
337 | struct task_group *tg; /* group that "owns" this runqueue */ | |
6aa645ea IM |
338 | #endif |
339 | }; | |
1da177e4 | 340 | |
6aa645ea IM |
341 | /* Real-Time classes' related field in a runqueue: */ |
342 | struct rt_rq { | |
343 | struct rt_prio_array active; | |
344 | int rt_load_balance_idx; | |
345 | struct list_head *rt_load_balance_head, *rt_load_balance_curr; | |
63489e45 | 346 | unsigned long rt_nr_running; |
73fe6aae | 347 | unsigned long rt_nr_migratory; |
764a9d6f SR |
348 | /* highest queued rt task prio */ |
349 | int highest_prio; | |
a22d7fc1 | 350 | int overloaded; |
6aa645ea IM |
351 | }; |
352 | ||
57d885fe GH |
353 | #ifdef CONFIG_SMP |
354 | ||
355 | /* | |
356 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
357 | * variables. Each exclusive cpuset essentially defines an island domain by |
358 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
359 | * exclusive cpuset is created, we also create and attach a new root-domain |
360 | * object. | |
361 | * | |
57d885fe GH |
362 | */ |
363 | struct root_domain { | |
364 | atomic_t refcount; | |
365 | cpumask_t span; | |
366 | cpumask_t online; | |
637f5085 | 367 | |
0eab9146 | 368 | /* |
637f5085 GH |
369 | * The "RT overload" flag: it gets set if a CPU has more than |
370 | * one runnable RT task. | |
371 | */ | |
372 | cpumask_t rto_mask; | |
0eab9146 | 373 | atomic_t rto_count; |
57d885fe GH |
374 | }; |
375 | ||
dc938520 GH |
376 | /* |
377 | * By default the system creates a single root-domain with all cpus as | |
378 | * members (mimicking the global state we have today). | |
379 | */ | |
57d885fe GH |
380 | static struct root_domain def_root_domain; |
381 | ||
382 | #endif | |
383 | ||
1da177e4 LT |
384 | /* |
385 | * This is the main, per-CPU runqueue data structure. | |
386 | * | |
387 | * Locking rule: those places that want to lock multiple runqueues | |
388 | * (such as the load balancing or the thread migration code), lock | |
389 | * acquire operations must be ordered by ascending &runqueue. | |
390 | */ | |
70b97a7f | 391 | struct rq { |
d8016491 IM |
392 | /* runqueue lock: */ |
393 | spinlock_t lock; | |
1da177e4 LT |
394 | |
395 | /* | |
396 | * nr_running and cpu_load should be in the same cacheline because | |
397 | * remote CPUs use both these fields when doing load calculation. | |
398 | */ | |
399 | unsigned long nr_running; | |
6aa645ea IM |
400 | #define CPU_LOAD_IDX_MAX 5 |
401 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 402 | unsigned char idle_at_tick; |
46cb4b7c SS |
403 | #ifdef CONFIG_NO_HZ |
404 | unsigned char in_nohz_recently; | |
405 | #endif | |
d8016491 IM |
406 | /* capture load from *all* tasks on this cpu: */ |
407 | struct load_weight load; | |
6aa645ea IM |
408 | unsigned long nr_load_updates; |
409 | u64 nr_switches; | |
410 | ||
411 | struct cfs_rq cfs; | |
412 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
d8016491 IM |
413 | /* list of leaf cfs_rq on this cpu: */ |
414 | struct list_head leaf_cfs_rq_list; | |
1da177e4 | 415 | #endif |
41a2d6cf | 416 | struct rt_rq rt; |
1da177e4 LT |
417 | |
418 | /* | |
419 | * This is part of a global counter where only the total sum | |
420 | * over all CPUs matters. A task can increase this counter on | |
421 | * one CPU and if it got migrated afterwards it may decrease | |
422 | * it on another CPU. Always updated under the runqueue lock: | |
423 | */ | |
424 | unsigned long nr_uninterruptible; | |
425 | ||
36c8b586 | 426 | struct task_struct *curr, *idle; |
c9819f45 | 427 | unsigned long next_balance; |
1da177e4 | 428 | struct mm_struct *prev_mm; |
6aa645ea | 429 | |
6aa645ea IM |
430 | u64 clock, prev_clock_raw; |
431 | s64 clock_max_delta; | |
432 | ||
433 | unsigned int clock_warps, clock_overflows; | |
2aa44d05 IM |
434 | u64 idle_clock; |
435 | unsigned int clock_deep_idle_events; | |
529c7726 | 436 | u64 tick_timestamp; |
6aa645ea | 437 | |
1da177e4 LT |
438 | atomic_t nr_iowait; |
439 | ||
440 | #ifdef CONFIG_SMP | |
0eab9146 | 441 | struct root_domain *rd; |
1da177e4 LT |
442 | struct sched_domain *sd; |
443 | ||
444 | /* For active balancing */ | |
445 | int active_balance; | |
446 | int push_cpu; | |
d8016491 IM |
447 | /* cpu of this runqueue: */ |
448 | int cpu; | |
1da177e4 | 449 | |
36c8b586 | 450 | struct task_struct *migration_thread; |
1da177e4 LT |
451 | struct list_head migration_queue; |
452 | #endif | |
453 | ||
454 | #ifdef CONFIG_SCHEDSTATS | |
455 | /* latency stats */ | |
456 | struct sched_info rq_sched_info; | |
457 | ||
458 | /* sys_sched_yield() stats */ | |
480b9434 KC |
459 | unsigned int yld_exp_empty; |
460 | unsigned int yld_act_empty; | |
461 | unsigned int yld_both_empty; | |
462 | unsigned int yld_count; | |
1da177e4 LT |
463 | |
464 | /* schedule() stats */ | |
480b9434 KC |
465 | unsigned int sched_switch; |
466 | unsigned int sched_count; | |
467 | unsigned int sched_goidle; | |
1da177e4 LT |
468 | |
469 | /* try_to_wake_up() stats */ | |
480b9434 KC |
470 | unsigned int ttwu_count; |
471 | unsigned int ttwu_local; | |
b8efb561 IM |
472 | |
473 | /* BKL stats */ | |
480b9434 | 474 | unsigned int bkl_count; |
1da177e4 | 475 | #endif |
fcb99371 | 476 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
477 | }; |
478 | ||
f34e3b61 | 479 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 480 | |
dd41f596 IM |
481 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
482 | { | |
483 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
484 | } | |
485 | ||
0a2966b4 CL |
486 | static inline int cpu_of(struct rq *rq) |
487 | { | |
488 | #ifdef CONFIG_SMP | |
489 | return rq->cpu; | |
490 | #else | |
491 | return 0; | |
492 | #endif | |
493 | } | |
494 | ||
20d315d4 | 495 | /* |
b04a0f4c IM |
496 | * Update the per-runqueue clock, as finegrained as the platform can give |
497 | * us, but without assuming monotonicity, etc.: | |
20d315d4 | 498 | */ |
b04a0f4c | 499 | static void __update_rq_clock(struct rq *rq) |
20d315d4 IM |
500 | { |
501 | u64 prev_raw = rq->prev_clock_raw; | |
502 | u64 now = sched_clock(); | |
503 | s64 delta = now - prev_raw; | |
504 | u64 clock = rq->clock; | |
505 | ||
b04a0f4c IM |
506 | #ifdef CONFIG_SCHED_DEBUG |
507 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
508 | #endif | |
20d315d4 IM |
509 | /* |
510 | * Protect against sched_clock() occasionally going backwards: | |
511 | */ | |
512 | if (unlikely(delta < 0)) { | |
513 | clock++; | |
514 | rq->clock_warps++; | |
515 | } else { | |
516 | /* | |
517 | * Catch too large forward jumps too: | |
518 | */ | |
529c7726 IM |
519 | if (unlikely(clock + delta > rq->tick_timestamp + TICK_NSEC)) { |
520 | if (clock < rq->tick_timestamp + TICK_NSEC) | |
521 | clock = rq->tick_timestamp + TICK_NSEC; | |
522 | else | |
523 | clock++; | |
20d315d4 IM |
524 | rq->clock_overflows++; |
525 | } else { | |
526 | if (unlikely(delta > rq->clock_max_delta)) | |
527 | rq->clock_max_delta = delta; | |
528 | clock += delta; | |
529 | } | |
530 | } | |
531 | ||
532 | rq->prev_clock_raw = now; | |
533 | rq->clock = clock; | |
b04a0f4c | 534 | } |
20d315d4 | 535 | |
b04a0f4c IM |
536 | static void update_rq_clock(struct rq *rq) |
537 | { | |
538 | if (likely(smp_processor_id() == cpu_of(rq))) | |
539 | __update_rq_clock(rq); | |
20d315d4 IM |
540 | } |
541 | ||
674311d5 NP |
542 | /* |
543 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 544 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
545 | * |
546 | * The domain tree of any CPU may only be accessed from within | |
547 | * preempt-disabled sections. | |
548 | */ | |
48f24c4d IM |
549 | #define for_each_domain(cpu, __sd) \ |
550 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
551 | |
552 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
553 | #define this_rq() (&__get_cpu_var(runqueues)) | |
554 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
555 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
556 | ||
bf5c91ba IM |
557 | /* |
558 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
559 | */ | |
560 | #ifdef CONFIG_SCHED_DEBUG | |
561 | # define const_debug __read_mostly | |
562 | #else | |
563 | # define const_debug static const | |
564 | #endif | |
565 | ||
566 | /* | |
567 | * Debugging: various feature bits | |
568 | */ | |
569 | enum { | |
bbdba7c0 | 570 | SCHED_FEAT_NEW_FAIR_SLEEPERS = 1, |
9612633a IM |
571 | SCHED_FEAT_WAKEUP_PREEMPT = 2, |
572 | SCHED_FEAT_START_DEBIT = 4, | |
41a2d6cf IM |
573 | SCHED_FEAT_TREE_AVG = 8, |
574 | SCHED_FEAT_APPROX_AVG = 16, | |
bf5c91ba IM |
575 | }; |
576 | ||
577 | const_debug unsigned int sysctl_sched_features = | |
8401f775 | 578 | SCHED_FEAT_NEW_FAIR_SLEEPERS * 1 | |
9612633a | 579 | SCHED_FEAT_WAKEUP_PREEMPT * 1 | |
8401f775 IM |
580 | SCHED_FEAT_START_DEBIT * 1 | |
581 | SCHED_FEAT_TREE_AVG * 0 | | |
9612633a | 582 | SCHED_FEAT_APPROX_AVG * 0; |
bf5c91ba IM |
583 | |
584 | #define sched_feat(x) (sysctl_sched_features & SCHED_FEAT_##x) | |
585 | ||
b82d9fdd PZ |
586 | /* |
587 | * Number of tasks to iterate in a single balance run. | |
588 | * Limited because this is done with IRQs disabled. | |
589 | */ | |
590 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
591 | ||
e436d800 IM |
592 | /* |
593 | * For kernel-internal use: high-speed (but slightly incorrect) per-cpu | |
594 | * clock constructed from sched_clock(): | |
595 | */ | |
596 | unsigned long long cpu_clock(int cpu) | |
597 | { | |
e436d800 IM |
598 | unsigned long long now; |
599 | unsigned long flags; | |
b04a0f4c | 600 | struct rq *rq; |
e436d800 | 601 | |
2cd4d0ea | 602 | local_irq_save(flags); |
b04a0f4c | 603 | rq = cpu_rq(cpu); |
8ced5f69 IM |
604 | /* |
605 | * Only call sched_clock() if the scheduler has already been | |
606 | * initialized (some code might call cpu_clock() very early): | |
607 | */ | |
608 | if (rq->idle) | |
609 | update_rq_clock(rq); | |
b04a0f4c | 610 | now = rq->clock; |
2cd4d0ea | 611 | local_irq_restore(flags); |
e436d800 IM |
612 | |
613 | return now; | |
614 | } | |
a58f6f25 | 615 | EXPORT_SYMBOL_GPL(cpu_clock); |
e436d800 | 616 | |
1da177e4 | 617 | #ifndef prepare_arch_switch |
4866cde0 NP |
618 | # define prepare_arch_switch(next) do { } while (0) |
619 | #endif | |
620 | #ifndef finish_arch_switch | |
621 | # define finish_arch_switch(prev) do { } while (0) | |
622 | #endif | |
623 | ||
051a1d1a DA |
624 | static inline int task_current(struct rq *rq, struct task_struct *p) |
625 | { | |
626 | return rq->curr == p; | |
627 | } | |
628 | ||
4866cde0 | 629 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 630 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 631 | { |
051a1d1a | 632 | return task_current(rq, p); |
4866cde0 NP |
633 | } |
634 | ||
70b97a7f | 635 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
636 | { |
637 | } | |
638 | ||
70b97a7f | 639 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 640 | { |
da04c035 IM |
641 | #ifdef CONFIG_DEBUG_SPINLOCK |
642 | /* this is a valid case when another task releases the spinlock */ | |
643 | rq->lock.owner = current; | |
644 | #endif | |
8a25d5de IM |
645 | /* |
646 | * If we are tracking spinlock dependencies then we have to | |
647 | * fix up the runqueue lock - which gets 'carried over' from | |
648 | * prev into current: | |
649 | */ | |
650 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
651 | ||
4866cde0 NP |
652 | spin_unlock_irq(&rq->lock); |
653 | } | |
654 | ||
655 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 656 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
657 | { |
658 | #ifdef CONFIG_SMP | |
659 | return p->oncpu; | |
660 | #else | |
051a1d1a | 661 | return task_current(rq, p); |
4866cde0 NP |
662 | #endif |
663 | } | |
664 | ||
70b97a7f | 665 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
666 | { |
667 | #ifdef CONFIG_SMP | |
668 | /* | |
669 | * We can optimise this out completely for !SMP, because the | |
670 | * SMP rebalancing from interrupt is the only thing that cares | |
671 | * here. | |
672 | */ | |
673 | next->oncpu = 1; | |
674 | #endif | |
675 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
676 | spin_unlock_irq(&rq->lock); | |
677 | #else | |
678 | spin_unlock(&rq->lock); | |
679 | #endif | |
680 | } | |
681 | ||
70b97a7f | 682 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
683 | { |
684 | #ifdef CONFIG_SMP | |
685 | /* | |
686 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
687 | * We must ensure this doesn't happen until the switch is completely | |
688 | * finished. | |
689 | */ | |
690 | smp_wmb(); | |
691 | prev->oncpu = 0; | |
692 | #endif | |
693 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
694 | local_irq_enable(); | |
1da177e4 | 695 | #endif |
4866cde0 NP |
696 | } |
697 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 698 | |
b29739f9 IM |
699 | /* |
700 | * __task_rq_lock - lock the runqueue a given task resides on. | |
701 | * Must be called interrupts disabled. | |
702 | */ | |
70b97a7f | 703 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
704 | __acquires(rq->lock) |
705 | { | |
3a5c359a AK |
706 | for (;;) { |
707 | struct rq *rq = task_rq(p); | |
708 | spin_lock(&rq->lock); | |
709 | if (likely(rq == task_rq(p))) | |
710 | return rq; | |
b29739f9 | 711 | spin_unlock(&rq->lock); |
b29739f9 | 712 | } |
b29739f9 IM |
713 | } |
714 | ||
1da177e4 LT |
715 | /* |
716 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 717 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
718 | * explicitly disabling preemption. |
719 | */ | |
70b97a7f | 720 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
721 | __acquires(rq->lock) |
722 | { | |
70b97a7f | 723 | struct rq *rq; |
1da177e4 | 724 | |
3a5c359a AK |
725 | for (;;) { |
726 | local_irq_save(*flags); | |
727 | rq = task_rq(p); | |
728 | spin_lock(&rq->lock); | |
729 | if (likely(rq == task_rq(p))) | |
730 | return rq; | |
1da177e4 | 731 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 732 | } |
1da177e4 LT |
733 | } |
734 | ||
a9957449 | 735 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
736 | __releases(rq->lock) |
737 | { | |
738 | spin_unlock(&rq->lock); | |
739 | } | |
740 | ||
70b97a7f | 741 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
742 | __releases(rq->lock) |
743 | { | |
744 | spin_unlock_irqrestore(&rq->lock, *flags); | |
745 | } | |
746 | ||
1da177e4 | 747 | /* |
cc2a73b5 | 748 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 749 | */ |
a9957449 | 750 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
751 | __acquires(rq->lock) |
752 | { | |
70b97a7f | 753 | struct rq *rq; |
1da177e4 LT |
754 | |
755 | local_irq_disable(); | |
756 | rq = this_rq(); | |
757 | spin_lock(&rq->lock); | |
758 | ||
759 | return rq; | |
760 | } | |
761 | ||
1b9f19c2 | 762 | /* |
2aa44d05 | 763 | * We are going deep-idle (irqs are disabled): |
1b9f19c2 | 764 | */ |
2aa44d05 | 765 | void sched_clock_idle_sleep_event(void) |
1b9f19c2 | 766 | { |
2aa44d05 IM |
767 | struct rq *rq = cpu_rq(smp_processor_id()); |
768 | ||
769 | spin_lock(&rq->lock); | |
770 | __update_rq_clock(rq); | |
771 | spin_unlock(&rq->lock); | |
772 | rq->clock_deep_idle_events++; | |
773 | } | |
774 | EXPORT_SYMBOL_GPL(sched_clock_idle_sleep_event); | |
775 | ||
776 | /* | |
777 | * We just idled delta nanoseconds (called with irqs disabled): | |
778 | */ | |
779 | void sched_clock_idle_wakeup_event(u64 delta_ns) | |
780 | { | |
781 | struct rq *rq = cpu_rq(smp_processor_id()); | |
782 | u64 now = sched_clock(); | |
1b9f19c2 | 783 | |
2bacec8c | 784 | touch_softlockup_watchdog(); |
2aa44d05 IM |
785 | rq->idle_clock += delta_ns; |
786 | /* | |
787 | * Override the previous timestamp and ignore all | |
788 | * sched_clock() deltas that occured while we idled, | |
789 | * and use the PM-provided delta_ns to advance the | |
790 | * rq clock: | |
791 | */ | |
792 | spin_lock(&rq->lock); | |
793 | rq->prev_clock_raw = now; | |
794 | rq->clock += delta_ns; | |
795 | spin_unlock(&rq->lock); | |
1b9f19c2 | 796 | } |
2aa44d05 | 797 | EXPORT_SYMBOL_GPL(sched_clock_idle_wakeup_event); |
1b9f19c2 | 798 | |
c24d20db IM |
799 | /* |
800 | * resched_task - mark a task 'to be rescheduled now'. | |
801 | * | |
802 | * On UP this means the setting of the need_resched flag, on SMP it | |
803 | * might also involve a cross-CPU call to trigger the scheduler on | |
804 | * the target CPU. | |
805 | */ | |
806 | #ifdef CONFIG_SMP | |
807 | ||
808 | #ifndef tsk_is_polling | |
809 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
810 | #endif | |
811 | ||
812 | static void resched_task(struct task_struct *p) | |
813 | { | |
814 | int cpu; | |
815 | ||
816 | assert_spin_locked(&task_rq(p)->lock); | |
817 | ||
818 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) | |
819 | return; | |
820 | ||
821 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); | |
822 | ||
823 | cpu = task_cpu(p); | |
824 | if (cpu == smp_processor_id()) | |
825 | return; | |
826 | ||
827 | /* NEED_RESCHED must be visible before we test polling */ | |
828 | smp_mb(); | |
829 | if (!tsk_is_polling(p)) | |
830 | smp_send_reschedule(cpu); | |
831 | } | |
832 | ||
833 | static void resched_cpu(int cpu) | |
834 | { | |
835 | struct rq *rq = cpu_rq(cpu); | |
836 | unsigned long flags; | |
837 | ||
838 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
839 | return; | |
840 | resched_task(cpu_curr(cpu)); | |
841 | spin_unlock_irqrestore(&rq->lock, flags); | |
842 | } | |
843 | #else | |
844 | static inline void resched_task(struct task_struct *p) | |
845 | { | |
846 | assert_spin_locked(&task_rq(p)->lock); | |
847 | set_tsk_need_resched(p); | |
848 | } | |
849 | #endif | |
850 | ||
45bf76df IM |
851 | #if BITS_PER_LONG == 32 |
852 | # define WMULT_CONST (~0UL) | |
853 | #else | |
854 | # define WMULT_CONST (1UL << 32) | |
855 | #endif | |
856 | ||
857 | #define WMULT_SHIFT 32 | |
858 | ||
194081eb IM |
859 | /* |
860 | * Shift right and round: | |
861 | */ | |
cf2ab469 | 862 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 863 | |
cb1c4fc9 | 864 | static unsigned long |
45bf76df IM |
865 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
866 | struct load_weight *lw) | |
867 | { | |
868 | u64 tmp; | |
869 | ||
870 | if (unlikely(!lw->inv_weight)) | |
194081eb | 871 | lw->inv_weight = (WMULT_CONST - lw->weight/2) / lw->weight + 1; |
45bf76df IM |
872 | |
873 | tmp = (u64)delta_exec * weight; | |
874 | /* | |
875 | * Check whether we'd overflow the 64-bit multiplication: | |
876 | */ | |
194081eb | 877 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 878 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
879 | WMULT_SHIFT/2); |
880 | else | |
cf2ab469 | 881 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 882 | |
ecf691da | 883 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
884 | } |
885 | ||
886 | static inline unsigned long | |
887 | calc_delta_fair(unsigned long delta_exec, struct load_weight *lw) | |
888 | { | |
889 | return calc_delta_mine(delta_exec, NICE_0_LOAD, lw); | |
890 | } | |
891 | ||
1091985b | 892 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
893 | { |
894 | lw->weight += inc; | |
45bf76df IM |
895 | } |
896 | ||
1091985b | 897 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
898 | { |
899 | lw->weight -= dec; | |
45bf76df IM |
900 | } |
901 | ||
2dd73a4f PW |
902 | /* |
903 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
904 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
905 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 906 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
907 | * scaled version of the new time slice allocation that they receive on time |
908 | * slice expiry etc. | |
909 | */ | |
910 | ||
dd41f596 IM |
911 | #define WEIGHT_IDLEPRIO 2 |
912 | #define WMULT_IDLEPRIO (1 << 31) | |
913 | ||
914 | /* | |
915 | * Nice levels are multiplicative, with a gentle 10% change for every | |
916 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
917 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
918 | * that remained on nice 0. | |
919 | * | |
920 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
921 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
922 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
923 | * If a task goes up by ~10% and another task goes down by ~10% then | |
924 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
925 | */ |
926 | static const int prio_to_weight[40] = { | |
254753dc IM |
927 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
928 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
929 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
930 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
931 | /* 0 */ 1024, 820, 655, 526, 423, | |
932 | /* 5 */ 335, 272, 215, 172, 137, | |
933 | /* 10 */ 110, 87, 70, 56, 45, | |
934 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
935 | }; |
936 | ||
5714d2de IM |
937 | /* |
938 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
939 | * | |
940 | * In cases where the weight does not change often, we can use the | |
941 | * precalculated inverse to speed up arithmetics by turning divisions | |
942 | * into multiplications: | |
943 | */ | |
dd41f596 | 944 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
945 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
946 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
947 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
948 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
949 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
950 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
951 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
952 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 953 | }; |
2dd73a4f | 954 | |
dd41f596 IM |
955 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
956 | ||
957 | /* | |
958 | * runqueue iterator, to support SMP load-balancing between different | |
959 | * scheduling classes, without having to expose their internal data | |
960 | * structures to the load-balancing proper: | |
961 | */ | |
962 | struct rq_iterator { | |
963 | void *arg; | |
964 | struct task_struct *(*start)(void *); | |
965 | struct task_struct *(*next)(void *); | |
966 | }; | |
967 | ||
e1d1484f PW |
968 | #ifdef CONFIG_SMP |
969 | static unsigned long | |
970 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
971 | unsigned long max_load_move, struct sched_domain *sd, | |
972 | enum cpu_idle_type idle, int *all_pinned, | |
973 | int *this_best_prio, struct rq_iterator *iterator); | |
974 | ||
975 | static int | |
976 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
977 | struct sched_domain *sd, enum cpu_idle_type idle, | |
978 | struct rq_iterator *iterator); | |
e1d1484f | 979 | #endif |
dd41f596 | 980 | |
d842de87 SV |
981 | #ifdef CONFIG_CGROUP_CPUACCT |
982 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
983 | #else | |
984 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
985 | #endif | |
986 | ||
58e2d4ca SV |
987 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
988 | { | |
989 | update_load_add(&rq->load, load); | |
990 | } | |
991 | ||
992 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
993 | { | |
994 | update_load_sub(&rq->load, load); | |
995 | } | |
996 | ||
e7693a36 GH |
997 | #ifdef CONFIG_SMP |
998 | static unsigned long source_load(int cpu, int type); | |
999 | static unsigned long target_load(int cpu, int type); | |
1000 | static unsigned long cpu_avg_load_per_task(int cpu); | |
1001 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1002 | #endif /* CONFIG_SMP */ | |
1003 | ||
dd41f596 | 1004 | #include "sched_stats.h" |
dd41f596 | 1005 | #include "sched_idletask.c" |
5522d5d5 IM |
1006 | #include "sched_fair.c" |
1007 | #include "sched_rt.c" | |
dd41f596 IM |
1008 | #ifdef CONFIG_SCHED_DEBUG |
1009 | # include "sched_debug.c" | |
1010 | #endif | |
1011 | ||
1012 | #define sched_class_highest (&rt_sched_class) | |
1013 | ||
e5fa2237 | 1014 | static void inc_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
1015 | { |
1016 | rq->nr_running++; | |
9c217245 IM |
1017 | } |
1018 | ||
db53181e | 1019 | static void dec_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
1020 | { |
1021 | rq->nr_running--; | |
9c217245 IM |
1022 | } |
1023 | ||
45bf76df IM |
1024 | static void set_load_weight(struct task_struct *p) |
1025 | { | |
1026 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1027 | p->se.load.weight = prio_to_weight[0] * 2; |
1028 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1029 | return; | |
1030 | } | |
45bf76df | 1031 | |
dd41f596 IM |
1032 | /* |
1033 | * SCHED_IDLE tasks get minimal weight: | |
1034 | */ | |
1035 | if (p->policy == SCHED_IDLE) { | |
1036 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1037 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1038 | return; | |
1039 | } | |
71f8bd46 | 1040 | |
dd41f596 IM |
1041 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1042 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1043 | } |
1044 | ||
8159f87e | 1045 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1046 | { |
dd41f596 | 1047 | sched_info_queued(p); |
fd390f6a | 1048 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1049 | p->se.on_rq = 1; |
71f8bd46 IM |
1050 | } |
1051 | ||
69be72c1 | 1052 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1053 | { |
f02231e5 | 1054 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1055 | p->se.on_rq = 0; |
71f8bd46 IM |
1056 | } |
1057 | ||
14531189 | 1058 | /* |
dd41f596 | 1059 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1060 | */ |
14531189 IM |
1061 | static inline int __normal_prio(struct task_struct *p) |
1062 | { | |
dd41f596 | 1063 | return p->static_prio; |
14531189 IM |
1064 | } |
1065 | ||
b29739f9 IM |
1066 | /* |
1067 | * Calculate the expected normal priority: i.e. priority | |
1068 | * without taking RT-inheritance into account. Might be | |
1069 | * boosted by interactivity modifiers. Changes upon fork, | |
1070 | * setprio syscalls, and whenever the interactivity | |
1071 | * estimator recalculates. | |
1072 | */ | |
36c8b586 | 1073 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1074 | { |
1075 | int prio; | |
1076 | ||
e05606d3 | 1077 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1078 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1079 | else | |
1080 | prio = __normal_prio(p); | |
1081 | return prio; | |
1082 | } | |
1083 | ||
1084 | /* | |
1085 | * Calculate the current priority, i.e. the priority | |
1086 | * taken into account by the scheduler. This value might | |
1087 | * be boosted by RT tasks, or might be boosted by | |
1088 | * interactivity modifiers. Will be RT if the task got | |
1089 | * RT-boosted. If not then it returns p->normal_prio. | |
1090 | */ | |
36c8b586 | 1091 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1092 | { |
1093 | p->normal_prio = normal_prio(p); | |
1094 | /* | |
1095 | * If we are RT tasks or we were boosted to RT priority, | |
1096 | * keep the priority unchanged. Otherwise, update priority | |
1097 | * to the normal priority: | |
1098 | */ | |
1099 | if (!rt_prio(p->prio)) | |
1100 | return p->normal_prio; | |
1101 | return p->prio; | |
1102 | } | |
1103 | ||
1da177e4 | 1104 | /* |
dd41f596 | 1105 | * activate_task - move a task to the runqueue. |
1da177e4 | 1106 | */ |
dd41f596 | 1107 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1108 | { |
dd41f596 IM |
1109 | if (p->state == TASK_UNINTERRUPTIBLE) |
1110 | rq->nr_uninterruptible--; | |
1da177e4 | 1111 | |
8159f87e | 1112 | enqueue_task(rq, p, wakeup); |
e5fa2237 | 1113 | inc_nr_running(p, rq); |
1da177e4 LT |
1114 | } |
1115 | ||
1da177e4 LT |
1116 | /* |
1117 | * deactivate_task - remove a task from the runqueue. | |
1118 | */ | |
2e1cb74a | 1119 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1120 | { |
dd41f596 IM |
1121 | if (p->state == TASK_UNINTERRUPTIBLE) |
1122 | rq->nr_uninterruptible++; | |
1123 | ||
69be72c1 | 1124 | dequeue_task(rq, p, sleep); |
db53181e | 1125 | dec_nr_running(p, rq); |
1da177e4 LT |
1126 | } |
1127 | ||
1da177e4 LT |
1128 | /** |
1129 | * task_curr - is this task currently executing on a CPU? | |
1130 | * @p: the task in question. | |
1131 | */ | |
36c8b586 | 1132 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1133 | { |
1134 | return cpu_curr(task_cpu(p)) == p; | |
1135 | } | |
1136 | ||
2dd73a4f PW |
1137 | /* Used instead of source_load when we know the type == 0 */ |
1138 | unsigned long weighted_cpuload(const int cpu) | |
1139 | { | |
495eca49 | 1140 | return cpu_rq(cpu)->load.weight; |
dd41f596 IM |
1141 | } |
1142 | ||
1143 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1144 | { | |
ce96b5ac | 1145 | set_task_cfs_rq(p, cpu); |
dd41f596 | 1146 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1147 | /* |
1148 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1149 | * successfuly executed on another CPU. We must ensure that updates of | |
1150 | * per-task data have been completed by this moment. | |
1151 | */ | |
1152 | smp_wmb(); | |
dd41f596 | 1153 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1154 | #endif |
2dd73a4f PW |
1155 | } |
1156 | ||
cb469845 SR |
1157 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1158 | const struct sched_class *prev_class, | |
1159 | int oldprio, int running) | |
1160 | { | |
1161 | if (prev_class != p->sched_class) { | |
1162 | if (prev_class->switched_from) | |
1163 | prev_class->switched_from(rq, p, running); | |
1164 | p->sched_class->switched_to(rq, p, running); | |
1165 | } else | |
1166 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1167 | } | |
1168 | ||
1da177e4 | 1169 | #ifdef CONFIG_SMP |
c65cc870 | 1170 | |
cc367732 IM |
1171 | /* |
1172 | * Is this task likely cache-hot: | |
1173 | */ | |
e7693a36 | 1174 | static int |
cc367732 IM |
1175 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1176 | { | |
1177 | s64 delta; | |
1178 | ||
1179 | if (p->sched_class != &fair_sched_class) | |
1180 | return 0; | |
1181 | ||
6bc1665b IM |
1182 | if (sysctl_sched_migration_cost == -1) |
1183 | return 1; | |
1184 | if (sysctl_sched_migration_cost == 0) | |
1185 | return 0; | |
1186 | ||
cc367732 IM |
1187 | delta = now - p->se.exec_start; |
1188 | ||
1189 | return delta < (s64)sysctl_sched_migration_cost; | |
1190 | } | |
1191 | ||
1192 | ||
dd41f596 | 1193 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1194 | { |
dd41f596 IM |
1195 | int old_cpu = task_cpu(p); |
1196 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1197 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1198 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1199 | u64 clock_offset; |
dd41f596 IM |
1200 | |
1201 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1202 | |
1203 | #ifdef CONFIG_SCHEDSTATS | |
1204 | if (p->se.wait_start) | |
1205 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1206 | if (p->se.sleep_start) |
1207 | p->se.sleep_start -= clock_offset; | |
1208 | if (p->se.block_start) | |
1209 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1210 | if (old_cpu != new_cpu) { |
1211 | schedstat_inc(p, se.nr_migrations); | |
1212 | if (task_hot(p, old_rq->clock, NULL)) | |
1213 | schedstat_inc(p, se.nr_forced2_migrations); | |
1214 | } | |
6cfb0d5d | 1215 | #endif |
2830cf8c SV |
1216 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1217 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1218 | |
1219 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1220 | } |
1221 | ||
70b97a7f | 1222 | struct migration_req { |
1da177e4 | 1223 | struct list_head list; |
1da177e4 | 1224 | |
36c8b586 | 1225 | struct task_struct *task; |
1da177e4 LT |
1226 | int dest_cpu; |
1227 | ||
1da177e4 | 1228 | struct completion done; |
70b97a7f | 1229 | }; |
1da177e4 LT |
1230 | |
1231 | /* | |
1232 | * The task's runqueue lock must be held. | |
1233 | * Returns true if you have to wait for migration thread. | |
1234 | */ | |
36c8b586 | 1235 | static int |
70b97a7f | 1236 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1237 | { |
70b97a7f | 1238 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1239 | |
1240 | /* | |
1241 | * If the task is not on a runqueue (and not running), then | |
1242 | * it is sufficient to simply update the task's cpu field. | |
1243 | */ | |
dd41f596 | 1244 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1245 | set_task_cpu(p, dest_cpu); |
1246 | return 0; | |
1247 | } | |
1248 | ||
1249 | init_completion(&req->done); | |
1da177e4 LT |
1250 | req->task = p; |
1251 | req->dest_cpu = dest_cpu; | |
1252 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1253 | |
1da177e4 LT |
1254 | return 1; |
1255 | } | |
1256 | ||
1257 | /* | |
1258 | * wait_task_inactive - wait for a thread to unschedule. | |
1259 | * | |
1260 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1261 | * else this function might spin for a *long* time. This function can't | |
1262 | * be called with interrupts off, or it may introduce deadlock with | |
1263 | * smp_call_function() if an IPI is sent by the same process we are | |
1264 | * waiting to become inactive. | |
1265 | */ | |
36c8b586 | 1266 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1267 | { |
1268 | unsigned long flags; | |
dd41f596 | 1269 | int running, on_rq; |
70b97a7f | 1270 | struct rq *rq; |
1da177e4 | 1271 | |
3a5c359a AK |
1272 | for (;;) { |
1273 | /* | |
1274 | * We do the initial early heuristics without holding | |
1275 | * any task-queue locks at all. We'll only try to get | |
1276 | * the runqueue lock when things look like they will | |
1277 | * work out! | |
1278 | */ | |
1279 | rq = task_rq(p); | |
fa490cfd | 1280 | |
3a5c359a AK |
1281 | /* |
1282 | * If the task is actively running on another CPU | |
1283 | * still, just relax and busy-wait without holding | |
1284 | * any locks. | |
1285 | * | |
1286 | * NOTE! Since we don't hold any locks, it's not | |
1287 | * even sure that "rq" stays as the right runqueue! | |
1288 | * But we don't care, since "task_running()" will | |
1289 | * return false if the runqueue has changed and p | |
1290 | * is actually now running somewhere else! | |
1291 | */ | |
1292 | while (task_running(rq, p)) | |
1293 | cpu_relax(); | |
fa490cfd | 1294 | |
3a5c359a AK |
1295 | /* |
1296 | * Ok, time to look more closely! We need the rq | |
1297 | * lock now, to be *sure*. If we're wrong, we'll | |
1298 | * just go back and repeat. | |
1299 | */ | |
1300 | rq = task_rq_lock(p, &flags); | |
1301 | running = task_running(rq, p); | |
1302 | on_rq = p->se.on_rq; | |
1303 | task_rq_unlock(rq, &flags); | |
fa490cfd | 1304 | |
3a5c359a AK |
1305 | /* |
1306 | * Was it really running after all now that we | |
1307 | * checked with the proper locks actually held? | |
1308 | * | |
1309 | * Oops. Go back and try again.. | |
1310 | */ | |
1311 | if (unlikely(running)) { | |
1312 | cpu_relax(); | |
1313 | continue; | |
1314 | } | |
fa490cfd | 1315 | |
3a5c359a AK |
1316 | /* |
1317 | * It's not enough that it's not actively running, | |
1318 | * it must be off the runqueue _entirely_, and not | |
1319 | * preempted! | |
1320 | * | |
1321 | * So if it wa still runnable (but just not actively | |
1322 | * running right now), it's preempted, and we should | |
1323 | * yield - it could be a while. | |
1324 | */ | |
1325 | if (unlikely(on_rq)) { | |
1326 | schedule_timeout_uninterruptible(1); | |
1327 | continue; | |
1328 | } | |
fa490cfd | 1329 | |
3a5c359a AK |
1330 | /* |
1331 | * Ahh, all good. It wasn't running, and it wasn't | |
1332 | * runnable, which means that it will never become | |
1333 | * running in the future either. We're all done! | |
1334 | */ | |
1335 | break; | |
1336 | } | |
1da177e4 LT |
1337 | } |
1338 | ||
1339 | /*** | |
1340 | * kick_process - kick a running thread to enter/exit the kernel | |
1341 | * @p: the to-be-kicked thread | |
1342 | * | |
1343 | * Cause a process which is running on another CPU to enter | |
1344 | * kernel-mode, without any delay. (to get signals handled.) | |
1345 | * | |
1346 | * NOTE: this function doesnt have to take the runqueue lock, | |
1347 | * because all it wants to ensure is that the remote task enters | |
1348 | * the kernel. If the IPI races and the task has been migrated | |
1349 | * to another CPU then no harm is done and the purpose has been | |
1350 | * achieved as well. | |
1351 | */ | |
36c8b586 | 1352 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1353 | { |
1354 | int cpu; | |
1355 | ||
1356 | preempt_disable(); | |
1357 | cpu = task_cpu(p); | |
1358 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1359 | smp_send_reschedule(cpu); | |
1360 | preempt_enable(); | |
1361 | } | |
1362 | ||
1363 | /* | |
2dd73a4f PW |
1364 | * Return a low guess at the load of a migration-source cpu weighted |
1365 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1366 | * |
1367 | * We want to under-estimate the load of migration sources, to | |
1368 | * balance conservatively. | |
1369 | */ | |
a9957449 | 1370 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 1371 | { |
70b97a7f | 1372 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1373 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1374 | |
3b0bd9bc | 1375 | if (type == 0) |
dd41f596 | 1376 | return total; |
b910472d | 1377 | |
dd41f596 | 1378 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
1379 | } |
1380 | ||
1381 | /* | |
2dd73a4f PW |
1382 | * Return a high guess at the load of a migration-target cpu weighted |
1383 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1384 | */ |
a9957449 | 1385 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 1386 | { |
70b97a7f | 1387 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1388 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1389 | |
7897986b | 1390 | if (type == 0) |
dd41f596 | 1391 | return total; |
3b0bd9bc | 1392 | |
dd41f596 | 1393 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
1394 | } |
1395 | ||
1396 | /* | |
1397 | * Return the average load per task on the cpu's run queue | |
1398 | */ | |
e7693a36 | 1399 | static unsigned long cpu_avg_load_per_task(int cpu) |
2dd73a4f | 1400 | { |
70b97a7f | 1401 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1402 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f PW |
1403 | unsigned long n = rq->nr_running; |
1404 | ||
dd41f596 | 1405 | return n ? total / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1406 | } |
1407 | ||
147cbb4b NP |
1408 | /* |
1409 | * find_idlest_group finds and returns the least busy CPU group within the | |
1410 | * domain. | |
1411 | */ | |
1412 | static struct sched_group * | |
1413 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1414 | { | |
1415 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1416 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1417 | int load_idx = sd->forkexec_idx; | |
1418 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1419 | ||
1420 | do { | |
1421 | unsigned long load, avg_load; | |
1422 | int local_group; | |
1423 | int i; | |
1424 | ||
da5a5522 BD |
1425 | /* Skip over this group if it has no CPUs allowed */ |
1426 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 1427 | continue; |
da5a5522 | 1428 | |
147cbb4b | 1429 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1430 | |
1431 | /* Tally up the load of all CPUs in the group */ | |
1432 | avg_load = 0; | |
1433 | ||
1434 | for_each_cpu_mask(i, group->cpumask) { | |
1435 | /* Bias balancing toward cpus of our domain */ | |
1436 | if (local_group) | |
1437 | load = source_load(i, load_idx); | |
1438 | else | |
1439 | load = target_load(i, load_idx); | |
1440 | ||
1441 | avg_load += load; | |
1442 | } | |
1443 | ||
1444 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1445 | avg_load = sg_div_cpu_power(group, |
1446 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1447 | |
1448 | if (local_group) { | |
1449 | this_load = avg_load; | |
1450 | this = group; | |
1451 | } else if (avg_load < min_load) { | |
1452 | min_load = avg_load; | |
1453 | idlest = group; | |
1454 | } | |
3a5c359a | 1455 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
1456 | |
1457 | if (!idlest || 100*this_load < imbalance*min_load) | |
1458 | return NULL; | |
1459 | return idlest; | |
1460 | } | |
1461 | ||
1462 | /* | |
0feaece9 | 1463 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1464 | */ |
95cdf3b7 IM |
1465 | static int |
1466 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) | |
147cbb4b | 1467 | { |
da5a5522 | 1468 | cpumask_t tmp; |
147cbb4b NP |
1469 | unsigned long load, min_load = ULONG_MAX; |
1470 | int idlest = -1; | |
1471 | int i; | |
1472 | ||
da5a5522 BD |
1473 | /* Traverse only the allowed CPUs */ |
1474 | cpus_and(tmp, group->cpumask, p->cpus_allowed); | |
1475 | ||
1476 | for_each_cpu_mask(i, tmp) { | |
2dd73a4f | 1477 | load = weighted_cpuload(i); |
147cbb4b NP |
1478 | |
1479 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1480 | min_load = load; | |
1481 | idlest = i; | |
1482 | } | |
1483 | } | |
1484 | ||
1485 | return idlest; | |
1486 | } | |
1487 | ||
476d139c NP |
1488 | /* |
1489 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1490 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1491 | * SD_BALANCE_EXEC. | |
1492 | * | |
1493 | * Balance, ie. select the least loaded group. | |
1494 | * | |
1495 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1496 | * | |
1497 | * preempt must be disabled. | |
1498 | */ | |
1499 | static int sched_balance_self(int cpu, int flag) | |
1500 | { | |
1501 | struct task_struct *t = current; | |
1502 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1503 | |
c96d145e | 1504 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
1505 | /* |
1506 | * If power savings logic is enabled for a domain, stop there. | |
1507 | */ | |
5c45bf27 SS |
1508 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
1509 | break; | |
476d139c NP |
1510 | if (tmp->flags & flag) |
1511 | sd = tmp; | |
c96d145e | 1512 | } |
476d139c NP |
1513 | |
1514 | while (sd) { | |
1515 | cpumask_t span; | |
1516 | struct sched_group *group; | |
1a848870 SS |
1517 | int new_cpu, weight; |
1518 | ||
1519 | if (!(sd->flags & flag)) { | |
1520 | sd = sd->child; | |
1521 | continue; | |
1522 | } | |
476d139c NP |
1523 | |
1524 | span = sd->span; | |
1525 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1526 | if (!group) { |
1527 | sd = sd->child; | |
1528 | continue; | |
1529 | } | |
476d139c | 1530 | |
da5a5522 | 1531 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
1532 | if (new_cpu == -1 || new_cpu == cpu) { |
1533 | /* Now try balancing at a lower domain level of cpu */ | |
1534 | sd = sd->child; | |
1535 | continue; | |
1536 | } | |
476d139c | 1537 | |
1a848870 | 1538 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1539 | cpu = new_cpu; |
476d139c NP |
1540 | sd = NULL; |
1541 | weight = cpus_weight(span); | |
1542 | for_each_domain(cpu, tmp) { | |
1543 | if (weight <= cpus_weight(tmp->span)) | |
1544 | break; | |
1545 | if (tmp->flags & flag) | |
1546 | sd = tmp; | |
1547 | } | |
1548 | /* while loop will break here if sd == NULL */ | |
1549 | } | |
1550 | ||
1551 | return cpu; | |
1552 | } | |
1553 | ||
1554 | #endif /* CONFIG_SMP */ | |
1da177e4 | 1555 | |
1da177e4 LT |
1556 | /*** |
1557 | * try_to_wake_up - wake up a thread | |
1558 | * @p: the to-be-woken-up thread | |
1559 | * @state: the mask of task states that can be woken | |
1560 | * @sync: do a synchronous wakeup? | |
1561 | * | |
1562 | * Put it on the run-queue if it's not already there. The "current" | |
1563 | * thread is always on the run-queue (except when the actual | |
1564 | * re-schedule is in progress), and as such you're allowed to do | |
1565 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1566 | * runnable without the overhead of this. | |
1567 | * | |
1568 | * returns failure only if the task is already active. | |
1569 | */ | |
36c8b586 | 1570 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 1571 | { |
cc367732 | 1572 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
1573 | unsigned long flags; |
1574 | long old_state; | |
70b97a7f | 1575 | struct rq *rq; |
1da177e4 LT |
1576 | |
1577 | rq = task_rq_lock(p, &flags); | |
1578 | old_state = p->state; | |
1579 | if (!(old_state & state)) | |
1580 | goto out; | |
1581 | ||
dd41f596 | 1582 | if (p->se.on_rq) |
1da177e4 LT |
1583 | goto out_running; |
1584 | ||
1585 | cpu = task_cpu(p); | |
cc367732 | 1586 | orig_cpu = cpu; |
1da177e4 LT |
1587 | this_cpu = smp_processor_id(); |
1588 | ||
1589 | #ifdef CONFIG_SMP | |
1590 | if (unlikely(task_running(rq, p))) | |
1591 | goto out_activate; | |
1592 | ||
5d2f5a61 DA |
1593 | cpu = p->sched_class->select_task_rq(p, sync); |
1594 | if (cpu != orig_cpu) { | |
1595 | set_task_cpu(p, cpu); | |
1da177e4 LT |
1596 | task_rq_unlock(rq, &flags); |
1597 | /* might preempt at this point */ | |
1598 | rq = task_rq_lock(p, &flags); | |
1599 | old_state = p->state; | |
1600 | if (!(old_state & state)) | |
1601 | goto out; | |
dd41f596 | 1602 | if (p->se.on_rq) |
1da177e4 LT |
1603 | goto out_running; |
1604 | ||
1605 | this_cpu = smp_processor_id(); | |
1606 | cpu = task_cpu(p); | |
1607 | } | |
1608 | ||
e7693a36 GH |
1609 | #ifdef CONFIG_SCHEDSTATS |
1610 | schedstat_inc(rq, ttwu_count); | |
1611 | if (cpu == this_cpu) | |
1612 | schedstat_inc(rq, ttwu_local); | |
1613 | else { | |
1614 | struct sched_domain *sd; | |
1615 | for_each_domain(this_cpu, sd) { | |
1616 | if (cpu_isset(cpu, sd->span)) { | |
1617 | schedstat_inc(sd, ttwu_wake_remote); | |
1618 | break; | |
1619 | } | |
1620 | } | |
1621 | } | |
e7693a36 GH |
1622 | #endif |
1623 | ||
1da177e4 LT |
1624 | out_activate: |
1625 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
1626 | schedstat_inc(p, se.nr_wakeups); |
1627 | if (sync) | |
1628 | schedstat_inc(p, se.nr_wakeups_sync); | |
1629 | if (orig_cpu != cpu) | |
1630 | schedstat_inc(p, se.nr_wakeups_migrate); | |
1631 | if (cpu == this_cpu) | |
1632 | schedstat_inc(p, se.nr_wakeups_local); | |
1633 | else | |
1634 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 1635 | update_rq_clock(rq); |
dd41f596 | 1636 | activate_task(rq, p, 1); |
9c63d9c0 | 1637 | check_preempt_curr(rq, p); |
1da177e4 LT |
1638 | success = 1; |
1639 | ||
1640 | out_running: | |
1641 | p->state = TASK_RUNNING; | |
9a897c5a SR |
1642 | #ifdef CONFIG_SMP |
1643 | if (p->sched_class->task_wake_up) | |
1644 | p->sched_class->task_wake_up(rq, p); | |
1645 | #endif | |
1da177e4 LT |
1646 | out: |
1647 | task_rq_unlock(rq, &flags); | |
1648 | ||
1649 | return success; | |
1650 | } | |
1651 | ||
36c8b586 | 1652 | int fastcall wake_up_process(struct task_struct *p) |
1da177e4 LT |
1653 | { |
1654 | return try_to_wake_up(p, TASK_STOPPED | TASK_TRACED | | |
1655 | TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE, 0); | |
1656 | } | |
1da177e4 LT |
1657 | EXPORT_SYMBOL(wake_up_process); |
1658 | ||
36c8b586 | 1659 | int fastcall wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1660 | { |
1661 | return try_to_wake_up(p, state, 0); | |
1662 | } | |
1663 | ||
1da177e4 LT |
1664 | /* |
1665 | * Perform scheduler related setup for a newly forked process p. | |
1666 | * p is forked by current. | |
dd41f596 IM |
1667 | * |
1668 | * __sched_fork() is basic setup used by init_idle() too: | |
1669 | */ | |
1670 | static void __sched_fork(struct task_struct *p) | |
1671 | { | |
dd41f596 IM |
1672 | p->se.exec_start = 0; |
1673 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1674 | p->se.prev_sum_exec_runtime = 0; |
6cfb0d5d IM |
1675 | |
1676 | #ifdef CONFIG_SCHEDSTATS | |
1677 | p->se.wait_start = 0; | |
dd41f596 IM |
1678 | p->se.sum_sleep_runtime = 0; |
1679 | p->se.sleep_start = 0; | |
dd41f596 IM |
1680 | p->se.block_start = 0; |
1681 | p->se.sleep_max = 0; | |
1682 | p->se.block_max = 0; | |
1683 | p->se.exec_max = 0; | |
eba1ed4b | 1684 | p->se.slice_max = 0; |
dd41f596 | 1685 | p->se.wait_max = 0; |
6cfb0d5d | 1686 | #endif |
476d139c | 1687 | |
fa717060 | 1688 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 1689 | p->se.on_rq = 0; |
476d139c | 1690 | |
e107be36 AK |
1691 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1692 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1693 | #endif | |
1694 | ||
1da177e4 LT |
1695 | /* |
1696 | * We mark the process as running here, but have not actually | |
1697 | * inserted it onto the runqueue yet. This guarantees that | |
1698 | * nobody will actually run it, and a signal or other external | |
1699 | * event cannot wake it up and insert it on the runqueue either. | |
1700 | */ | |
1701 | p->state = TASK_RUNNING; | |
dd41f596 IM |
1702 | } |
1703 | ||
1704 | /* | |
1705 | * fork()/clone()-time setup: | |
1706 | */ | |
1707 | void sched_fork(struct task_struct *p, int clone_flags) | |
1708 | { | |
1709 | int cpu = get_cpu(); | |
1710 | ||
1711 | __sched_fork(p); | |
1712 | ||
1713 | #ifdef CONFIG_SMP | |
1714 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
1715 | #endif | |
02e4bac2 | 1716 | set_task_cpu(p, cpu); |
b29739f9 IM |
1717 | |
1718 | /* | |
1719 | * Make sure we do not leak PI boosting priority to the child: | |
1720 | */ | |
1721 | p->prio = current->normal_prio; | |
2ddbf952 HS |
1722 | if (!rt_prio(p->prio)) |
1723 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1724 | |
52f17b6c | 1725 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1726 | if (likely(sched_info_on())) |
52f17b6c | 1727 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1728 | #endif |
d6077cb8 | 1729 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
1730 | p->oncpu = 0; |
1731 | #endif | |
1da177e4 | 1732 | #ifdef CONFIG_PREEMPT |
4866cde0 | 1733 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1734 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1735 | #endif |
476d139c | 1736 | put_cpu(); |
1da177e4 LT |
1737 | } |
1738 | ||
1739 | /* | |
1740 | * wake_up_new_task - wake up a newly created task for the first time. | |
1741 | * | |
1742 | * This function will do some initial scheduler statistics housekeeping | |
1743 | * that must be done for every newly created context, then puts the task | |
1744 | * on the runqueue and wakes it. | |
1745 | */ | |
36c8b586 | 1746 | void fastcall wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
1747 | { |
1748 | unsigned long flags; | |
dd41f596 | 1749 | struct rq *rq; |
1da177e4 LT |
1750 | |
1751 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 1752 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 1753 | update_rq_clock(rq); |
1da177e4 LT |
1754 | |
1755 | p->prio = effective_prio(p); | |
1756 | ||
b9dca1e0 | 1757 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 1758 | activate_task(rq, p, 0); |
1da177e4 | 1759 | } else { |
1da177e4 | 1760 | /* |
dd41f596 IM |
1761 | * Let the scheduling class do new task startup |
1762 | * management (if any): | |
1da177e4 | 1763 | */ |
ee0827d8 | 1764 | p->sched_class->task_new(rq, p); |
e5fa2237 | 1765 | inc_nr_running(p, rq); |
1da177e4 | 1766 | } |
dd41f596 | 1767 | check_preempt_curr(rq, p); |
9a897c5a SR |
1768 | #ifdef CONFIG_SMP |
1769 | if (p->sched_class->task_wake_up) | |
1770 | p->sched_class->task_wake_up(rq, p); | |
1771 | #endif | |
dd41f596 | 1772 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
1773 | } |
1774 | ||
e107be36 AK |
1775 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1776 | ||
1777 | /** | |
421cee29 RD |
1778 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
1779 | * @notifier: notifier struct to register | |
e107be36 AK |
1780 | */ |
1781 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1782 | { | |
1783 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1784 | } | |
1785 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1786 | ||
1787 | /** | |
1788 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1789 | * @notifier: notifier struct to unregister |
e107be36 AK |
1790 | * |
1791 | * This is safe to call from within a preemption notifier. | |
1792 | */ | |
1793 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1794 | { | |
1795 | hlist_del(¬ifier->link); | |
1796 | } | |
1797 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1798 | ||
1799 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1800 | { | |
1801 | struct preempt_notifier *notifier; | |
1802 | struct hlist_node *node; | |
1803 | ||
1804 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1805 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
1806 | } | |
1807 | ||
1808 | static void | |
1809 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1810 | struct task_struct *next) | |
1811 | { | |
1812 | struct preempt_notifier *notifier; | |
1813 | struct hlist_node *node; | |
1814 | ||
1815 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1816 | notifier->ops->sched_out(notifier, next); | |
1817 | } | |
1818 | ||
1819 | #else | |
1820 | ||
1821 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1822 | { | |
1823 | } | |
1824 | ||
1825 | static void | |
1826 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1827 | struct task_struct *next) | |
1828 | { | |
1829 | } | |
1830 | ||
1831 | #endif | |
1832 | ||
4866cde0 NP |
1833 | /** |
1834 | * prepare_task_switch - prepare to switch tasks | |
1835 | * @rq: the runqueue preparing to switch | |
421cee29 | 1836 | * @prev: the current task that is being switched out |
4866cde0 NP |
1837 | * @next: the task we are going to switch to. |
1838 | * | |
1839 | * This is called with the rq lock held and interrupts off. It must | |
1840 | * be paired with a subsequent finish_task_switch after the context | |
1841 | * switch. | |
1842 | * | |
1843 | * prepare_task_switch sets up locking and calls architecture specific | |
1844 | * hooks. | |
1845 | */ | |
e107be36 AK |
1846 | static inline void |
1847 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1848 | struct task_struct *next) | |
4866cde0 | 1849 | { |
e107be36 | 1850 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1851 | prepare_lock_switch(rq, next); |
1852 | prepare_arch_switch(next); | |
1853 | } | |
1854 | ||
1da177e4 LT |
1855 | /** |
1856 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1857 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1858 | * @prev: the thread we just switched away from. |
1859 | * | |
4866cde0 NP |
1860 | * finish_task_switch must be called after the context switch, paired |
1861 | * with a prepare_task_switch call before the context switch. | |
1862 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1863 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1864 | * |
1865 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 1866 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
1867 | * with the lock held can cause deadlocks; see schedule() for |
1868 | * details.) | |
1869 | */ | |
a9957449 | 1870 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1871 | __releases(rq->lock) |
1872 | { | |
1da177e4 | 1873 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1874 | long prev_state; |
1da177e4 LT |
1875 | |
1876 | rq->prev_mm = NULL; | |
1877 | ||
1878 | /* | |
1879 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1880 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1881 | * schedule one last time. The schedule call will never return, and |
1882 | * the scheduled task must drop that reference. | |
c394cc9f | 1883 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1884 | * still held, otherwise prev could be scheduled on another cpu, die |
1885 | * there before we look at prev->state, and then the reference would | |
1886 | * be dropped twice. | |
1887 | * Manfred Spraul <manfred@colorfullife.com> | |
1888 | */ | |
55a101f8 | 1889 | prev_state = prev->state; |
4866cde0 NP |
1890 | finish_arch_switch(prev); |
1891 | finish_lock_switch(rq, prev); | |
9a897c5a SR |
1892 | #ifdef CONFIG_SMP |
1893 | if (current->sched_class->post_schedule) | |
1894 | current->sched_class->post_schedule(rq); | |
1895 | #endif | |
e8fa1362 | 1896 | |
e107be36 | 1897 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1898 | if (mm) |
1899 | mmdrop(mm); | |
c394cc9f | 1900 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1901 | /* |
1902 | * Remove function-return probe instances associated with this | |
1903 | * task and put them back on the free list. | |
9761eea8 | 1904 | */ |
c6fd91f0 | 1905 | kprobe_flush_task(prev); |
1da177e4 | 1906 | put_task_struct(prev); |
c6fd91f0 | 1907 | } |
1da177e4 LT |
1908 | } |
1909 | ||
1910 | /** | |
1911 | * schedule_tail - first thing a freshly forked thread must call. | |
1912 | * @prev: the thread we just switched away from. | |
1913 | */ | |
36c8b586 | 1914 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1915 | __releases(rq->lock) |
1916 | { | |
70b97a7f IM |
1917 | struct rq *rq = this_rq(); |
1918 | ||
4866cde0 NP |
1919 | finish_task_switch(rq, prev); |
1920 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
1921 | /* In this case, finish_task_switch does not reenable preemption */ | |
1922 | preempt_enable(); | |
1923 | #endif | |
1da177e4 | 1924 | if (current->set_child_tid) |
b488893a | 1925 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
1926 | } |
1927 | ||
1928 | /* | |
1929 | * context_switch - switch to the new MM and the new | |
1930 | * thread's register state. | |
1931 | */ | |
dd41f596 | 1932 | static inline void |
70b97a7f | 1933 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1934 | struct task_struct *next) |
1da177e4 | 1935 | { |
dd41f596 | 1936 | struct mm_struct *mm, *oldmm; |
1da177e4 | 1937 | |
e107be36 | 1938 | prepare_task_switch(rq, prev, next); |
dd41f596 IM |
1939 | mm = next->mm; |
1940 | oldmm = prev->active_mm; | |
9226d125 ZA |
1941 | /* |
1942 | * For paravirt, this is coupled with an exit in switch_to to | |
1943 | * combine the page table reload and the switch backend into | |
1944 | * one hypercall. | |
1945 | */ | |
1946 | arch_enter_lazy_cpu_mode(); | |
1947 | ||
dd41f596 | 1948 | if (unlikely(!mm)) { |
1da177e4 LT |
1949 | next->active_mm = oldmm; |
1950 | atomic_inc(&oldmm->mm_count); | |
1951 | enter_lazy_tlb(oldmm, next); | |
1952 | } else | |
1953 | switch_mm(oldmm, mm, next); | |
1954 | ||
dd41f596 | 1955 | if (unlikely(!prev->mm)) { |
1da177e4 | 1956 | prev->active_mm = NULL; |
1da177e4 LT |
1957 | rq->prev_mm = oldmm; |
1958 | } | |
3a5f5e48 IM |
1959 | /* |
1960 | * Since the runqueue lock will be released by the next | |
1961 | * task (which is an invalid locking op but in the case | |
1962 | * of the scheduler it's an obvious special-case), so we | |
1963 | * do an early lockdep release here: | |
1964 | */ | |
1965 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 1966 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 1967 | #endif |
1da177e4 LT |
1968 | |
1969 | /* Here we just switch the register state and the stack. */ | |
1970 | switch_to(prev, next, prev); | |
1971 | ||
dd41f596 IM |
1972 | barrier(); |
1973 | /* | |
1974 | * this_rq must be evaluated again because prev may have moved | |
1975 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
1976 | * frame will be invalid. | |
1977 | */ | |
1978 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
1979 | } |
1980 | ||
1981 | /* | |
1982 | * nr_running, nr_uninterruptible and nr_context_switches: | |
1983 | * | |
1984 | * externally visible scheduler statistics: current number of runnable | |
1985 | * threads, current number of uninterruptible-sleeping threads, total | |
1986 | * number of context switches performed since bootup. | |
1987 | */ | |
1988 | unsigned long nr_running(void) | |
1989 | { | |
1990 | unsigned long i, sum = 0; | |
1991 | ||
1992 | for_each_online_cpu(i) | |
1993 | sum += cpu_rq(i)->nr_running; | |
1994 | ||
1995 | return sum; | |
1996 | } | |
1997 | ||
1998 | unsigned long nr_uninterruptible(void) | |
1999 | { | |
2000 | unsigned long i, sum = 0; | |
2001 | ||
0a945022 | 2002 | for_each_possible_cpu(i) |
1da177e4 LT |
2003 | sum += cpu_rq(i)->nr_uninterruptible; |
2004 | ||
2005 | /* | |
2006 | * Since we read the counters lockless, it might be slightly | |
2007 | * inaccurate. Do not allow it to go below zero though: | |
2008 | */ | |
2009 | if (unlikely((long)sum < 0)) | |
2010 | sum = 0; | |
2011 | ||
2012 | return sum; | |
2013 | } | |
2014 | ||
2015 | unsigned long long nr_context_switches(void) | |
2016 | { | |
cc94abfc SR |
2017 | int i; |
2018 | unsigned long long sum = 0; | |
1da177e4 | 2019 | |
0a945022 | 2020 | for_each_possible_cpu(i) |
1da177e4 LT |
2021 | sum += cpu_rq(i)->nr_switches; |
2022 | ||
2023 | return sum; | |
2024 | } | |
2025 | ||
2026 | unsigned long nr_iowait(void) | |
2027 | { | |
2028 | unsigned long i, sum = 0; | |
2029 | ||
0a945022 | 2030 | for_each_possible_cpu(i) |
1da177e4 LT |
2031 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2032 | ||
2033 | return sum; | |
2034 | } | |
2035 | ||
db1b1fef JS |
2036 | unsigned long nr_active(void) |
2037 | { | |
2038 | unsigned long i, running = 0, uninterruptible = 0; | |
2039 | ||
2040 | for_each_online_cpu(i) { | |
2041 | running += cpu_rq(i)->nr_running; | |
2042 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2043 | } | |
2044 | ||
2045 | if (unlikely((long)uninterruptible < 0)) | |
2046 | uninterruptible = 0; | |
2047 | ||
2048 | return running + uninterruptible; | |
2049 | } | |
2050 | ||
48f24c4d | 2051 | /* |
dd41f596 IM |
2052 | * Update rq->cpu_load[] statistics. This function is usually called every |
2053 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2054 | */ |
dd41f596 | 2055 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2056 | { |
495eca49 | 2057 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2058 | int i, scale; |
2059 | ||
2060 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2061 | |
2062 | /* Update our load: */ | |
2063 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2064 | unsigned long old_load, new_load; | |
2065 | ||
2066 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2067 | ||
2068 | old_load = this_rq->cpu_load[i]; | |
2069 | new_load = this_load; | |
a25707f3 IM |
2070 | /* |
2071 | * Round up the averaging division if load is increasing. This | |
2072 | * prevents us from getting stuck on 9 if the load is 10, for | |
2073 | * example. | |
2074 | */ | |
2075 | if (new_load > old_load) | |
2076 | new_load += scale-1; | |
dd41f596 IM |
2077 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2078 | } | |
48f24c4d IM |
2079 | } |
2080 | ||
dd41f596 IM |
2081 | #ifdef CONFIG_SMP |
2082 | ||
1da177e4 LT |
2083 | /* |
2084 | * double_rq_lock - safely lock two runqueues | |
2085 | * | |
2086 | * Note this does not disable interrupts like task_rq_lock, | |
2087 | * you need to do so manually before calling. | |
2088 | */ | |
70b97a7f | 2089 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2090 | __acquires(rq1->lock) |
2091 | __acquires(rq2->lock) | |
2092 | { | |
054b9108 | 2093 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2094 | if (rq1 == rq2) { |
2095 | spin_lock(&rq1->lock); | |
2096 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2097 | } else { | |
c96d145e | 2098 | if (rq1 < rq2) { |
1da177e4 LT |
2099 | spin_lock(&rq1->lock); |
2100 | spin_lock(&rq2->lock); | |
2101 | } else { | |
2102 | spin_lock(&rq2->lock); | |
2103 | spin_lock(&rq1->lock); | |
2104 | } | |
2105 | } | |
6e82a3be IM |
2106 | update_rq_clock(rq1); |
2107 | update_rq_clock(rq2); | |
1da177e4 LT |
2108 | } |
2109 | ||
2110 | /* | |
2111 | * double_rq_unlock - safely unlock two runqueues | |
2112 | * | |
2113 | * Note this does not restore interrupts like task_rq_unlock, | |
2114 | * you need to do so manually after calling. | |
2115 | */ | |
70b97a7f | 2116 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2117 | __releases(rq1->lock) |
2118 | __releases(rq2->lock) | |
2119 | { | |
2120 | spin_unlock(&rq1->lock); | |
2121 | if (rq1 != rq2) | |
2122 | spin_unlock(&rq2->lock); | |
2123 | else | |
2124 | __release(rq2->lock); | |
2125 | } | |
2126 | ||
2127 | /* | |
2128 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2129 | */ | |
e8fa1362 | 2130 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2131 | __releases(this_rq->lock) |
2132 | __acquires(busiest->lock) | |
2133 | __acquires(this_rq->lock) | |
2134 | { | |
e8fa1362 SR |
2135 | int ret = 0; |
2136 | ||
054b9108 KK |
2137 | if (unlikely(!irqs_disabled())) { |
2138 | /* printk() doesn't work good under rq->lock */ | |
2139 | spin_unlock(&this_rq->lock); | |
2140 | BUG_ON(1); | |
2141 | } | |
1da177e4 | 2142 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2143 | if (busiest < this_rq) { |
1da177e4 LT |
2144 | spin_unlock(&this_rq->lock); |
2145 | spin_lock(&busiest->lock); | |
2146 | spin_lock(&this_rq->lock); | |
e8fa1362 | 2147 | ret = 1; |
1da177e4 LT |
2148 | } else |
2149 | spin_lock(&busiest->lock); | |
2150 | } | |
e8fa1362 | 2151 | return ret; |
1da177e4 LT |
2152 | } |
2153 | ||
1da177e4 LT |
2154 | /* |
2155 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2156 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2157 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2158 | * the cpu_allowed mask is restored. |
2159 | */ | |
36c8b586 | 2160 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2161 | { |
70b97a7f | 2162 | struct migration_req req; |
1da177e4 | 2163 | unsigned long flags; |
70b97a7f | 2164 | struct rq *rq; |
1da177e4 LT |
2165 | |
2166 | rq = task_rq_lock(p, &flags); | |
2167 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2168 | || unlikely(cpu_is_offline(dest_cpu))) | |
2169 | goto out; | |
2170 | ||
2171 | /* force the process onto the specified CPU */ | |
2172 | if (migrate_task(p, dest_cpu, &req)) { | |
2173 | /* Need to wait for migration thread (might exit: take ref). */ | |
2174 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2175 | |
1da177e4 LT |
2176 | get_task_struct(mt); |
2177 | task_rq_unlock(rq, &flags); | |
2178 | wake_up_process(mt); | |
2179 | put_task_struct(mt); | |
2180 | wait_for_completion(&req.done); | |
36c8b586 | 2181 | |
1da177e4 LT |
2182 | return; |
2183 | } | |
2184 | out: | |
2185 | task_rq_unlock(rq, &flags); | |
2186 | } | |
2187 | ||
2188 | /* | |
476d139c NP |
2189 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2190 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2191 | */ |
2192 | void sched_exec(void) | |
2193 | { | |
1da177e4 | 2194 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2195 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2196 | put_cpu(); |
476d139c NP |
2197 | if (new_cpu != this_cpu) |
2198 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2199 | } |
2200 | ||
2201 | /* | |
2202 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2203 | * Both runqueues must be locked. | |
2204 | */ | |
dd41f596 IM |
2205 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2206 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2207 | { |
2e1cb74a | 2208 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2209 | set_task_cpu(p, this_cpu); |
dd41f596 | 2210 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2211 | /* |
2212 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2213 | * to be always true for them. | |
2214 | */ | |
dd41f596 | 2215 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2216 | } |
2217 | ||
2218 | /* | |
2219 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2220 | */ | |
858119e1 | 2221 | static |
70b97a7f | 2222 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2223 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2224 | int *all_pinned) |
1da177e4 LT |
2225 | { |
2226 | /* | |
2227 | * We do not migrate tasks that are: | |
2228 | * 1) running (obviously), or | |
2229 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2230 | * 3) are cache-hot on their current CPU. | |
2231 | */ | |
cc367732 IM |
2232 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2233 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2234 | return 0; |
cc367732 | 2235 | } |
81026794 NP |
2236 | *all_pinned = 0; |
2237 | ||
cc367732 IM |
2238 | if (task_running(rq, p)) { |
2239 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2240 | return 0; |
cc367732 | 2241 | } |
1da177e4 | 2242 | |
da84d961 IM |
2243 | /* |
2244 | * Aggressive migration if: | |
2245 | * 1) task is cache cold, or | |
2246 | * 2) too many balance attempts have failed. | |
2247 | */ | |
2248 | ||
6bc1665b IM |
2249 | if (!task_hot(p, rq->clock, sd) || |
2250 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2251 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2252 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2253 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2254 | schedstat_inc(p, se.nr_forced_migrations); |
2255 | } | |
da84d961 IM |
2256 | #endif |
2257 | return 1; | |
2258 | } | |
2259 | ||
cc367732 IM |
2260 | if (task_hot(p, rq->clock, sd)) { |
2261 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2262 | return 0; |
cc367732 | 2263 | } |
1da177e4 LT |
2264 | return 1; |
2265 | } | |
2266 | ||
e1d1484f PW |
2267 | static unsigned long |
2268 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2269 | unsigned long max_load_move, struct sched_domain *sd, | |
2270 | enum cpu_idle_type idle, int *all_pinned, | |
2271 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2272 | { |
b82d9fdd | 2273 | int loops = 0, pulled = 0, pinned = 0, skip_for_load; |
dd41f596 IM |
2274 | struct task_struct *p; |
2275 | long rem_load_move = max_load_move; | |
1da177e4 | 2276 | |
e1d1484f | 2277 | if (max_load_move == 0) |
1da177e4 LT |
2278 | goto out; |
2279 | ||
81026794 NP |
2280 | pinned = 1; |
2281 | ||
1da177e4 | 2282 | /* |
dd41f596 | 2283 | * Start the load-balancing iterator: |
1da177e4 | 2284 | */ |
dd41f596 IM |
2285 | p = iterator->start(iterator->arg); |
2286 | next: | |
b82d9fdd | 2287 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2288 | goto out; |
50ddd969 | 2289 | /* |
b82d9fdd | 2290 | * To help distribute high priority tasks across CPUs we don't |
50ddd969 PW |
2291 | * skip a task if it will be the highest priority task (i.e. smallest |
2292 | * prio value) on its new queue regardless of its load weight | |
2293 | */ | |
dd41f596 IM |
2294 | skip_for_load = (p->se.load.weight >> 1) > rem_load_move + |
2295 | SCHED_LOAD_SCALE_FUZZ; | |
a4ac01c3 | 2296 | if ((skip_for_load && p->prio >= *this_best_prio) || |
dd41f596 | 2297 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2298 | p = iterator->next(iterator->arg); |
2299 | goto next; | |
1da177e4 LT |
2300 | } |
2301 | ||
dd41f596 | 2302 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2303 | pulled++; |
dd41f596 | 2304 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2305 | |
2dd73a4f | 2306 | /* |
b82d9fdd | 2307 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 2308 | */ |
e1d1484f | 2309 | if (rem_load_move > 0) { |
a4ac01c3 PW |
2310 | if (p->prio < *this_best_prio) |
2311 | *this_best_prio = p->prio; | |
dd41f596 IM |
2312 | p = iterator->next(iterator->arg); |
2313 | goto next; | |
1da177e4 LT |
2314 | } |
2315 | out: | |
2316 | /* | |
e1d1484f | 2317 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
2318 | * so we can safely collect pull_task() stats here rather than |
2319 | * inside pull_task(). | |
2320 | */ | |
2321 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2322 | |
2323 | if (all_pinned) | |
2324 | *all_pinned = pinned; | |
e1d1484f PW |
2325 | |
2326 | return max_load_move - rem_load_move; | |
1da177e4 LT |
2327 | } |
2328 | ||
dd41f596 | 2329 | /* |
43010659 PW |
2330 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2331 | * this_rq, as part of a balancing operation within domain "sd". | |
2332 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2333 | * |
2334 | * Called with both runqueues locked. | |
2335 | */ | |
2336 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2337 | unsigned long max_load_move, |
dd41f596 IM |
2338 | struct sched_domain *sd, enum cpu_idle_type idle, |
2339 | int *all_pinned) | |
2340 | { | |
5522d5d5 | 2341 | const struct sched_class *class = sched_class_highest; |
43010659 | 2342 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2343 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
2344 | |
2345 | do { | |
43010659 PW |
2346 | total_load_moved += |
2347 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 2348 | max_load_move - total_load_moved, |
a4ac01c3 | 2349 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 2350 | class = class->next; |
43010659 | 2351 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 2352 | |
43010659 PW |
2353 | return total_load_moved > 0; |
2354 | } | |
2355 | ||
e1d1484f PW |
2356 | static int |
2357 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2358 | struct sched_domain *sd, enum cpu_idle_type idle, | |
2359 | struct rq_iterator *iterator) | |
2360 | { | |
2361 | struct task_struct *p = iterator->start(iterator->arg); | |
2362 | int pinned = 0; | |
2363 | ||
2364 | while (p) { | |
2365 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
2366 | pull_task(busiest, p, this_rq, this_cpu); | |
2367 | /* | |
2368 | * Right now, this is only the second place pull_task() | |
2369 | * is called, so we can safely collect pull_task() | |
2370 | * stats here rather than inside pull_task(). | |
2371 | */ | |
2372 | schedstat_inc(sd, lb_gained[idle]); | |
2373 | ||
2374 | return 1; | |
2375 | } | |
2376 | p = iterator->next(iterator->arg); | |
2377 | } | |
2378 | ||
2379 | return 0; | |
2380 | } | |
2381 | ||
43010659 PW |
2382 | /* |
2383 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
2384 | * part of active balancing operations within "domain". | |
2385 | * Returns 1 if successful and 0 otherwise. | |
2386 | * | |
2387 | * Called with both runqueues locked. | |
2388 | */ | |
2389 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2390 | struct sched_domain *sd, enum cpu_idle_type idle) | |
2391 | { | |
5522d5d5 | 2392 | const struct sched_class *class; |
43010659 PW |
2393 | |
2394 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 2395 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
2396 | return 1; |
2397 | ||
2398 | return 0; | |
dd41f596 IM |
2399 | } |
2400 | ||
1da177e4 LT |
2401 | /* |
2402 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2403 | * domain. It calculates and returns the amount of weighted load which |
2404 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2405 | */ |
2406 | static struct sched_group * | |
2407 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 IM |
2408 | unsigned long *imbalance, enum cpu_idle_type idle, |
2409 | int *sd_idle, cpumask_t *cpus, int *balance) | |
1da177e4 LT |
2410 | { |
2411 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2412 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2413 | unsigned long max_pull; |
2dd73a4f PW |
2414 | unsigned long busiest_load_per_task, busiest_nr_running; |
2415 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 2416 | int load_idx, group_imb = 0; |
5c45bf27 SS |
2417 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2418 | int power_savings_balance = 1; | |
2419 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2420 | unsigned long min_nr_running = ULONG_MAX; | |
2421 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2422 | #endif | |
1da177e4 LT |
2423 | |
2424 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2425 | busiest_load_per_task = busiest_nr_running = 0; |
2426 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2427 | if (idle == CPU_NOT_IDLE) |
7897986b | 2428 | load_idx = sd->busy_idx; |
d15bcfdb | 2429 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2430 | load_idx = sd->newidle_idx; |
2431 | else | |
2432 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2433 | |
2434 | do { | |
908a7c1b | 2435 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
2436 | int local_group; |
2437 | int i; | |
908a7c1b | 2438 | int __group_imb = 0; |
783609c6 | 2439 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2440 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2441 | |
2442 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2443 | ||
783609c6 SS |
2444 | if (local_group) |
2445 | balance_cpu = first_cpu(group->cpumask); | |
2446 | ||
1da177e4 | 2447 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2448 | sum_weighted_load = sum_nr_running = avg_load = 0; |
908a7c1b KC |
2449 | max_cpu_load = 0; |
2450 | min_cpu_load = ~0UL; | |
1da177e4 LT |
2451 | |
2452 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2453 | struct rq *rq; |
2454 | ||
2455 | if (!cpu_isset(i, *cpus)) | |
2456 | continue; | |
2457 | ||
2458 | rq = cpu_rq(i); | |
2dd73a4f | 2459 | |
9439aab8 | 2460 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
2461 | *sd_idle = 0; |
2462 | ||
1da177e4 | 2463 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2464 | if (local_group) { |
2465 | if (idle_cpu(i) && !first_idle_cpu) { | |
2466 | first_idle_cpu = 1; | |
2467 | balance_cpu = i; | |
2468 | } | |
2469 | ||
a2000572 | 2470 | load = target_load(i, load_idx); |
908a7c1b | 2471 | } else { |
a2000572 | 2472 | load = source_load(i, load_idx); |
908a7c1b KC |
2473 | if (load > max_cpu_load) |
2474 | max_cpu_load = load; | |
2475 | if (min_cpu_load > load) | |
2476 | min_cpu_load = load; | |
2477 | } | |
1da177e4 LT |
2478 | |
2479 | avg_load += load; | |
2dd73a4f | 2480 | sum_nr_running += rq->nr_running; |
dd41f596 | 2481 | sum_weighted_load += weighted_cpuload(i); |
1da177e4 LT |
2482 | } |
2483 | ||
783609c6 SS |
2484 | /* |
2485 | * First idle cpu or the first cpu(busiest) in this sched group | |
2486 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
2487 | * domains. In the newly idle case, we will allow all the cpu's |
2488 | * to do the newly idle load balance. | |
783609c6 | 2489 | */ |
9439aab8 SS |
2490 | if (idle != CPU_NEWLY_IDLE && local_group && |
2491 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
2492 | *balance = 0; |
2493 | goto ret; | |
2494 | } | |
2495 | ||
1da177e4 | 2496 | total_load += avg_load; |
5517d86b | 2497 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2498 | |
2499 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2500 | avg_load = sg_div_cpu_power(group, |
2501 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2502 | |
908a7c1b KC |
2503 | if ((max_cpu_load - min_cpu_load) > SCHED_LOAD_SCALE) |
2504 | __group_imb = 1; | |
2505 | ||
5517d86b | 2506 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2507 | |
1da177e4 LT |
2508 | if (local_group) { |
2509 | this_load = avg_load; | |
2510 | this = group; | |
2dd73a4f PW |
2511 | this_nr_running = sum_nr_running; |
2512 | this_load_per_task = sum_weighted_load; | |
2513 | } else if (avg_load > max_load && | |
908a7c1b | 2514 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
2515 | max_load = avg_load; |
2516 | busiest = group; | |
2dd73a4f PW |
2517 | busiest_nr_running = sum_nr_running; |
2518 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 2519 | group_imb = __group_imb; |
1da177e4 | 2520 | } |
5c45bf27 SS |
2521 | |
2522 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2523 | /* | |
2524 | * Busy processors will not participate in power savings | |
2525 | * balance. | |
2526 | */ | |
dd41f596 IM |
2527 | if (idle == CPU_NOT_IDLE || |
2528 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
2529 | goto group_next; | |
5c45bf27 SS |
2530 | |
2531 | /* | |
2532 | * If the local group is idle or completely loaded | |
2533 | * no need to do power savings balance at this domain | |
2534 | */ | |
2535 | if (local_group && (this_nr_running >= group_capacity || | |
2536 | !this_nr_running)) | |
2537 | power_savings_balance = 0; | |
2538 | ||
dd41f596 | 2539 | /* |
5c45bf27 SS |
2540 | * If a group is already running at full capacity or idle, |
2541 | * don't include that group in power savings calculations | |
dd41f596 IM |
2542 | */ |
2543 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 2544 | || !sum_nr_running) |
dd41f596 | 2545 | goto group_next; |
5c45bf27 | 2546 | |
dd41f596 | 2547 | /* |
5c45bf27 | 2548 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
2549 | * This is the group from where we need to pick up the load |
2550 | * for saving power | |
2551 | */ | |
2552 | if ((sum_nr_running < min_nr_running) || | |
2553 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
2554 | first_cpu(group->cpumask) < |
2555 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
2556 | group_min = group; |
2557 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
2558 | min_load_per_task = sum_weighted_load / |
2559 | sum_nr_running; | |
dd41f596 | 2560 | } |
5c45bf27 | 2561 | |
dd41f596 | 2562 | /* |
5c45bf27 | 2563 | * Calculate the group which is almost near its |
dd41f596 IM |
2564 | * capacity but still has some space to pick up some load |
2565 | * from other group and save more power | |
2566 | */ | |
2567 | if (sum_nr_running <= group_capacity - 1) { | |
2568 | if (sum_nr_running > leader_nr_running || | |
2569 | (sum_nr_running == leader_nr_running && | |
2570 | first_cpu(group->cpumask) > | |
2571 | first_cpu(group_leader->cpumask))) { | |
2572 | group_leader = group; | |
2573 | leader_nr_running = sum_nr_running; | |
2574 | } | |
48f24c4d | 2575 | } |
5c45bf27 SS |
2576 | group_next: |
2577 | #endif | |
1da177e4 LT |
2578 | group = group->next; |
2579 | } while (group != sd->groups); | |
2580 | ||
2dd73a4f | 2581 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
2582 | goto out_balanced; |
2583 | ||
2584 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
2585 | ||
2586 | if (this_load >= avg_load || | |
2587 | 100*max_load <= sd->imbalance_pct*this_load) | |
2588 | goto out_balanced; | |
2589 | ||
2dd73a4f | 2590 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
2591 | if (group_imb) |
2592 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
2593 | ||
1da177e4 LT |
2594 | /* |
2595 | * We're trying to get all the cpus to the average_load, so we don't | |
2596 | * want to push ourselves above the average load, nor do we wish to | |
2597 | * reduce the max loaded cpu below the average load, as either of these | |
2598 | * actions would just result in more rebalancing later, and ping-pong | |
2599 | * tasks around. Thus we look for the minimum possible imbalance. | |
2600 | * Negative imbalances (*we* are more loaded than anyone else) will | |
2601 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 2602 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
2603 | * appear as very large values with unsigned longs. |
2604 | */ | |
2dd73a4f PW |
2605 | if (max_load <= busiest_load_per_task) |
2606 | goto out_balanced; | |
2607 | ||
2608 | /* | |
2609 | * In the presence of smp nice balancing, certain scenarios can have | |
2610 | * max load less than avg load(as we skip the groups at or below | |
2611 | * its cpu_power, while calculating max_load..) | |
2612 | */ | |
2613 | if (max_load < avg_load) { | |
2614 | *imbalance = 0; | |
2615 | goto small_imbalance; | |
2616 | } | |
0c117f1b SS |
2617 | |
2618 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 2619 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 2620 | |
1da177e4 | 2621 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
2622 | *imbalance = min(max_pull * busiest->__cpu_power, |
2623 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
2624 | / SCHED_LOAD_SCALE; |
2625 | ||
2dd73a4f PW |
2626 | /* |
2627 | * if *imbalance is less than the average load per runnable task | |
2628 | * there is no gaurantee that any tasks will be moved so we'll have | |
2629 | * a think about bumping its value to force at least one task to be | |
2630 | * moved | |
2631 | */ | |
7fd0d2dd | 2632 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 2633 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
2634 | unsigned int imbn; |
2635 | ||
2636 | small_imbalance: | |
2637 | pwr_move = pwr_now = 0; | |
2638 | imbn = 2; | |
2639 | if (this_nr_running) { | |
2640 | this_load_per_task /= this_nr_running; | |
2641 | if (busiest_load_per_task > this_load_per_task) | |
2642 | imbn = 1; | |
2643 | } else | |
2644 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 2645 | |
dd41f596 IM |
2646 | if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >= |
2647 | busiest_load_per_task * imbn) { | |
2dd73a4f | 2648 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2649 | return busiest; |
2650 | } | |
2651 | ||
2652 | /* | |
2653 | * OK, we don't have enough imbalance to justify moving tasks, | |
2654 | * however we may be able to increase total CPU power used by | |
2655 | * moving them. | |
2656 | */ | |
2657 | ||
5517d86b ED |
2658 | pwr_now += busiest->__cpu_power * |
2659 | min(busiest_load_per_task, max_load); | |
2660 | pwr_now += this->__cpu_power * | |
2661 | min(this_load_per_task, this_load); | |
1da177e4 LT |
2662 | pwr_now /= SCHED_LOAD_SCALE; |
2663 | ||
2664 | /* Amount of load we'd subtract */ | |
5517d86b ED |
2665 | tmp = sg_div_cpu_power(busiest, |
2666 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 2667 | if (max_load > tmp) |
5517d86b | 2668 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 2669 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
2670 | |
2671 | /* Amount of load we'd add */ | |
5517d86b | 2672 | if (max_load * busiest->__cpu_power < |
33859f7f | 2673 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
2674 | tmp = sg_div_cpu_power(this, |
2675 | max_load * busiest->__cpu_power); | |
1da177e4 | 2676 | else |
5517d86b ED |
2677 | tmp = sg_div_cpu_power(this, |
2678 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
2679 | pwr_move += this->__cpu_power * | |
2680 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
2681 | pwr_move /= SCHED_LOAD_SCALE; |
2682 | ||
2683 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
2684 | if (pwr_move > pwr_now) |
2685 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
2686 | } |
2687 | ||
1da177e4 LT |
2688 | return busiest; |
2689 | ||
2690 | out_balanced: | |
5c45bf27 | 2691 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 2692 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 2693 | goto ret; |
1da177e4 | 2694 | |
5c45bf27 SS |
2695 | if (this == group_leader && group_leader != group_min) { |
2696 | *imbalance = min_load_per_task; | |
2697 | return group_min; | |
2698 | } | |
5c45bf27 | 2699 | #endif |
783609c6 | 2700 | ret: |
1da177e4 LT |
2701 | *imbalance = 0; |
2702 | return NULL; | |
2703 | } | |
2704 | ||
2705 | /* | |
2706 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
2707 | */ | |
70b97a7f | 2708 | static struct rq * |
d15bcfdb | 2709 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
0a2966b4 | 2710 | unsigned long imbalance, cpumask_t *cpus) |
1da177e4 | 2711 | { |
70b97a7f | 2712 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 2713 | unsigned long max_load = 0; |
1da177e4 LT |
2714 | int i; |
2715 | ||
2716 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 2717 | unsigned long wl; |
0a2966b4 CL |
2718 | |
2719 | if (!cpu_isset(i, *cpus)) | |
2720 | continue; | |
2721 | ||
48f24c4d | 2722 | rq = cpu_rq(i); |
dd41f596 | 2723 | wl = weighted_cpuload(i); |
2dd73a4f | 2724 | |
dd41f596 | 2725 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 2726 | continue; |
1da177e4 | 2727 | |
dd41f596 IM |
2728 | if (wl > max_load) { |
2729 | max_load = wl; | |
48f24c4d | 2730 | busiest = rq; |
1da177e4 LT |
2731 | } |
2732 | } | |
2733 | ||
2734 | return busiest; | |
2735 | } | |
2736 | ||
77391d71 NP |
2737 | /* |
2738 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
2739 | * so long as it is large enough. | |
2740 | */ | |
2741 | #define MAX_PINNED_INTERVAL 512 | |
2742 | ||
1da177e4 LT |
2743 | /* |
2744 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2745 | * tasks if there is an imbalance. | |
1da177e4 | 2746 | */ |
70b97a7f | 2747 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 2748 | struct sched_domain *sd, enum cpu_idle_type idle, |
783609c6 | 2749 | int *balance) |
1da177e4 | 2750 | { |
43010659 | 2751 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 2752 | struct sched_group *group; |
1da177e4 | 2753 | unsigned long imbalance; |
70b97a7f | 2754 | struct rq *busiest; |
0a2966b4 | 2755 | cpumask_t cpus = CPU_MASK_ALL; |
fe2eea3f | 2756 | unsigned long flags; |
5969fe06 | 2757 | |
89c4710e SS |
2758 | /* |
2759 | * When power savings policy is enabled for the parent domain, idle | |
2760 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 2761 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 2762 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 2763 | */ |
d15bcfdb | 2764 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2765 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2766 | sd_idle = 1; |
1da177e4 | 2767 | |
2d72376b | 2768 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 2769 | |
0a2966b4 CL |
2770 | redo: |
2771 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
783609c6 SS |
2772 | &cpus, balance); |
2773 | ||
06066714 | 2774 | if (*balance == 0) |
783609c6 | 2775 | goto out_balanced; |
783609c6 | 2776 | |
1da177e4 LT |
2777 | if (!group) { |
2778 | schedstat_inc(sd, lb_nobusyg[idle]); | |
2779 | goto out_balanced; | |
2780 | } | |
2781 | ||
0a2966b4 | 2782 | busiest = find_busiest_queue(group, idle, imbalance, &cpus); |
1da177e4 LT |
2783 | if (!busiest) { |
2784 | schedstat_inc(sd, lb_nobusyq[idle]); | |
2785 | goto out_balanced; | |
2786 | } | |
2787 | ||
db935dbd | 2788 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
2789 | |
2790 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
2791 | ||
43010659 | 2792 | ld_moved = 0; |
1da177e4 LT |
2793 | if (busiest->nr_running > 1) { |
2794 | /* | |
2795 | * Attempt to move tasks. If find_busiest_group has found | |
2796 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 2797 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
2798 | * correctly treated as an imbalance. |
2799 | */ | |
fe2eea3f | 2800 | local_irq_save(flags); |
e17224bf | 2801 | double_rq_lock(this_rq, busiest); |
43010659 | 2802 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 2803 | imbalance, sd, idle, &all_pinned); |
e17224bf | 2804 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 2805 | local_irq_restore(flags); |
81026794 | 2806 | |
46cb4b7c SS |
2807 | /* |
2808 | * some other cpu did the load balance for us. | |
2809 | */ | |
43010659 | 2810 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
2811 | resched_cpu(this_cpu); |
2812 | ||
81026794 | 2813 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 CL |
2814 | if (unlikely(all_pinned)) { |
2815 | cpu_clear(cpu_of(busiest), cpus); | |
2816 | if (!cpus_empty(cpus)) | |
2817 | goto redo; | |
81026794 | 2818 | goto out_balanced; |
0a2966b4 | 2819 | } |
1da177e4 | 2820 | } |
81026794 | 2821 | |
43010659 | 2822 | if (!ld_moved) { |
1da177e4 LT |
2823 | schedstat_inc(sd, lb_failed[idle]); |
2824 | sd->nr_balance_failed++; | |
2825 | ||
2826 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 2827 | |
fe2eea3f | 2828 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
2829 | |
2830 | /* don't kick the migration_thread, if the curr | |
2831 | * task on busiest cpu can't be moved to this_cpu | |
2832 | */ | |
2833 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 2834 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
2835 | all_pinned = 1; |
2836 | goto out_one_pinned; | |
2837 | } | |
2838 | ||
1da177e4 LT |
2839 | if (!busiest->active_balance) { |
2840 | busiest->active_balance = 1; | |
2841 | busiest->push_cpu = this_cpu; | |
81026794 | 2842 | active_balance = 1; |
1da177e4 | 2843 | } |
fe2eea3f | 2844 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 2845 | if (active_balance) |
1da177e4 LT |
2846 | wake_up_process(busiest->migration_thread); |
2847 | ||
2848 | /* | |
2849 | * We've kicked active balancing, reset the failure | |
2850 | * counter. | |
2851 | */ | |
39507451 | 2852 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 2853 | } |
81026794 | 2854 | } else |
1da177e4 LT |
2855 | sd->nr_balance_failed = 0; |
2856 | ||
81026794 | 2857 | if (likely(!active_balance)) { |
1da177e4 LT |
2858 | /* We were unbalanced, so reset the balancing interval */ |
2859 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
2860 | } else { |
2861 | /* | |
2862 | * If we've begun active balancing, start to back off. This | |
2863 | * case may not be covered by the all_pinned logic if there | |
2864 | * is only 1 task on the busy runqueue (because we don't call | |
2865 | * move_tasks). | |
2866 | */ | |
2867 | if (sd->balance_interval < sd->max_interval) | |
2868 | sd->balance_interval *= 2; | |
1da177e4 LT |
2869 | } |
2870 | ||
43010659 | 2871 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2872 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2873 | return -1; |
43010659 | 2874 | return ld_moved; |
1da177e4 LT |
2875 | |
2876 | out_balanced: | |
1da177e4 LT |
2877 | schedstat_inc(sd, lb_balanced[idle]); |
2878 | ||
16cfb1c0 | 2879 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
2880 | |
2881 | out_one_pinned: | |
1da177e4 | 2882 | /* tune up the balancing interval */ |
77391d71 NP |
2883 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
2884 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
2885 | sd->balance_interval *= 2; |
2886 | ||
48f24c4d | 2887 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2888 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2889 | return -1; |
1da177e4 LT |
2890 | return 0; |
2891 | } | |
2892 | ||
2893 | /* | |
2894 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2895 | * tasks if there is an imbalance. | |
2896 | * | |
d15bcfdb | 2897 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
2898 | * this_rq is locked. |
2899 | */ | |
48f24c4d | 2900 | static int |
70b97a7f | 2901 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
2902 | { |
2903 | struct sched_group *group; | |
70b97a7f | 2904 | struct rq *busiest = NULL; |
1da177e4 | 2905 | unsigned long imbalance; |
43010659 | 2906 | int ld_moved = 0; |
5969fe06 | 2907 | int sd_idle = 0; |
969bb4e4 | 2908 | int all_pinned = 0; |
0a2966b4 | 2909 | cpumask_t cpus = CPU_MASK_ALL; |
5969fe06 | 2910 | |
89c4710e SS |
2911 | /* |
2912 | * When power savings policy is enabled for the parent domain, idle | |
2913 | * sibling can pick up load irrespective of busy siblings. In this case, | |
2914 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 2915 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
2916 | */ |
2917 | if (sd->flags & SD_SHARE_CPUPOWER && | |
2918 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 2919 | sd_idle = 1; |
1da177e4 | 2920 | |
2d72376b | 2921 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 2922 | redo: |
d15bcfdb | 2923 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
783609c6 | 2924 | &sd_idle, &cpus, NULL); |
1da177e4 | 2925 | if (!group) { |
d15bcfdb | 2926 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2927 | goto out_balanced; |
1da177e4 LT |
2928 | } |
2929 | ||
d15bcfdb | 2930 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, |
0a2966b4 | 2931 | &cpus); |
db935dbd | 2932 | if (!busiest) { |
d15bcfdb | 2933 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2934 | goto out_balanced; |
1da177e4 LT |
2935 | } |
2936 | ||
db935dbd NP |
2937 | BUG_ON(busiest == this_rq); |
2938 | ||
d15bcfdb | 2939 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 2940 | |
43010659 | 2941 | ld_moved = 0; |
d6d5cfaf NP |
2942 | if (busiest->nr_running > 1) { |
2943 | /* Attempt to move tasks */ | |
2944 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
2945 | /* this_rq->clock is already updated */ |
2946 | update_rq_clock(busiest); | |
43010659 | 2947 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
2948 | imbalance, sd, CPU_NEWLY_IDLE, |
2949 | &all_pinned); | |
d6d5cfaf | 2950 | spin_unlock(&busiest->lock); |
0a2966b4 | 2951 | |
969bb4e4 | 2952 | if (unlikely(all_pinned)) { |
0a2966b4 CL |
2953 | cpu_clear(cpu_of(busiest), cpus); |
2954 | if (!cpus_empty(cpus)) | |
2955 | goto redo; | |
2956 | } | |
d6d5cfaf NP |
2957 | } |
2958 | ||
43010659 | 2959 | if (!ld_moved) { |
d15bcfdb | 2960 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
2961 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
2962 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
2963 | return -1; |
2964 | } else | |
16cfb1c0 | 2965 | sd->nr_balance_failed = 0; |
1da177e4 | 2966 | |
43010659 | 2967 | return ld_moved; |
16cfb1c0 NP |
2968 | |
2969 | out_balanced: | |
d15bcfdb | 2970 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 2971 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2972 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2973 | return -1; |
16cfb1c0 | 2974 | sd->nr_balance_failed = 0; |
48f24c4d | 2975 | |
16cfb1c0 | 2976 | return 0; |
1da177e4 LT |
2977 | } |
2978 | ||
2979 | /* | |
2980 | * idle_balance is called by schedule() if this_cpu is about to become | |
2981 | * idle. Attempts to pull tasks from other CPUs. | |
2982 | */ | |
70b97a7f | 2983 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
2984 | { |
2985 | struct sched_domain *sd; | |
dd41f596 IM |
2986 | int pulled_task = -1; |
2987 | unsigned long next_balance = jiffies + HZ; | |
1da177e4 LT |
2988 | |
2989 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
2990 | unsigned long interval; |
2991 | ||
2992 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
2993 | continue; | |
2994 | ||
2995 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 2996 | /* If we've pulled tasks over stop searching: */ |
1bd77f2d | 2997 | pulled_task = load_balance_newidle(this_cpu, |
92c4ca5c CL |
2998 | this_rq, sd); |
2999 | ||
3000 | interval = msecs_to_jiffies(sd->balance_interval); | |
3001 | if (time_after(next_balance, sd->last_balance + interval)) | |
3002 | next_balance = sd->last_balance + interval; | |
3003 | if (pulled_task) | |
3004 | break; | |
1da177e4 | 3005 | } |
dd41f596 | 3006 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
3007 | /* |
3008 | * We are going idle. next_balance may be set based on | |
3009 | * a busy processor. So reset next_balance. | |
3010 | */ | |
3011 | this_rq->next_balance = next_balance; | |
dd41f596 | 3012 | } |
1da177e4 LT |
3013 | } |
3014 | ||
3015 | /* | |
3016 | * active_load_balance is run by migration threads. It pushes running tasks | |
3017 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
3018 | * running on each physical CPU where possible, and avoids physical / | |
3019 | * logical imbalances. | |
3020 | * | |
3021 | * Called with busiest_rq locked. | |
3022 | */ | |
70b97a7f | 3023 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3024 | { |
39507451 | 3025 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3026 | struct sched_domain *sd; |
3027 | struct rq *target_rq; | |
39507451 | 3028 | |
48f24c4d | 3029 | /* Is there any task to move? */ |
39507451 | 3030 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3031 | return; |
3032 | ||
3033 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3034 | |
3035 | /* | |
39507451 | 3036 | * This condition is "impossible", if it occurs |
41a2d6cf | 3037 | * we need to fix it. Originally reported by |
39507451 | 3038 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3039 | */ |
39507451 | 3040 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3041 | |
39507451 NP |
3042 | /* move a task from busiest_rq to target_rq */ |
3043 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3044 | update_rq_clock(busiest_rq); |
3045 | update_rq_clock(target_rq); | |
39507451 NP |
3046 | |
3047 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3048 | for_each_domain(target_cpu, sd) { |
39507451 | 3049 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 3050 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 3051 | break; |
c96d145e | 3052 | } |
39507451 | 3053 | |
48f24c4d | 3054 | if (likely(sd)) { |
2d72376b | 3055 | schedstat_inc(sd, alb_count); |
39507451 | 3056 | |
43010659 PW |
3057 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3058 | sd, CPU_IDLE)) | |
48f24c4d IM |
3059 | schedstat_inc(sd, alb_pushed); |
3060 | else | |
3061 | schedstat_inc(sd, alb_failed); | |
3062 | } | |
39507451 | 3063 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
3064 | } |
3065 | ||
46cb4b7c SS |
3066 | #ifdef CONFIG_NO_HZ |
3067 | static struct { | |
3068 | atomic_t load_balancer; | |
41a2d6cf | 3069 | cpumask_t cpu_mask; |
46cb4b7c SS |
3070 | } nohz ____cacheline_aligned = { |
3071 | .load_balancer = ATOMIC_INIT(-1), | |
3072 | .cpu_mask = CPU_MASK_NONE, | |
3073 | }; | |
3074 | ||
7835b98b | 3075 | /* |
46cb4b7c SS |
3076 | * This routine will try to nominate the ilb (idle load balancing) |
3077 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3078 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3079 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3080 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3081 | * arrives... | |
3082 | * | |
3083 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3084 | * for idle load balancing. ilb owner will still be part of | |
3085 | * nohz.cpu_mask.. | |
7835b98b | 3086 | * |
46cb4b7c SS |
3087 | * While stopping the tick, this cpu will become the ilb owner if there |
3088 | * is no other owner. And will be the owner till that cpu becomes busy | |
3089 | * or if all cpus in the system stop their ticks at which point | |
3090 | * there is no need for ilb owner. | |
3091 | * | |
3092 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3093 | * next busy scheduler_tick() | |
3094 | */ | |
3095 | int select_nohz_load_balancer(int stop_tick) | |
3096 | { | |
3097 | int cpu = smp_processor_id(); | |
3098 | ||
3099 | if (stop_tick) { | |
3100 | cpu_set(cpu, nohz.cpu_mask); | |
3101 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3102 | ||
3103 | /* | |
3104 | * If we are going offline and still the leader, give up! | |
3105 | */ | |
3106 | if (cpu_is_offline(cpu) && | |
3107 | atomic_read(&nohz.load_balancer) == cpu) { | |
3108 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3109 | BUG(); | |
3110 | return 0; | |
3111 | } | |
3112 | ||
3113 | /* time for ilb owner also to sleep */ | |
3114 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3115 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3116 | atomic_set(&nohz.load_balancer, -1); | |
3117 | return 0; | |
3118 | } | |
3119 | ||
3120 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3121 | /* make me the ilb owner */ | |
3122 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3123 | return 1; | |
3124 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3125 | return 1; | |
3126 | } else { | |
3127 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3128 | return 0; | |
3129 | ||
3130 | cpu_clear(cpu, nohz.cpu_mask); | |
3131 | ||
3132 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3133 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3134 | BUG(); | |
3135 | } | |
3136 | return 0; | |
3137 | } | |
3138 | #endif | |
3139 | ||
3140 | static DEFINE_SPINLOCK(balancing); | |
3141 | ||
3142 | /* | |
7835b98b CL |
3143 | * It checks each scheduling domain to see if it is due to be balanced, |
3144 | * and initiates a balancing operation if so. | |
3145 | * | |
3146 | * Balancing parameters are set up in arch_init_sched_domains. | |
3147 | */ | |
a9957449 | 3148 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3149 | { |
46cb4b7c SS |
3150 | int balance = 1; |
3151 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3152 | unsigned long interval; |
3153 | struct sched_domain *sd; | |
46cb4b7c | 3154 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3155 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3156 | int update_next_balance = 0; |
1da177e4 | 3157 | |
46cb4b7c | 3158 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3159 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3160 | continue; | |
3161 | ||
3162 | interval = sd->balance_interval; | |
d15bcfdb | 3163 | if (idle != CPU_IDLE) |
1da177e4 LT |
3164 | interval *= sd->busy_factor; |
3165 | ||
3166 | /* scale ms to jiffies */ | |
3167 | interval = msecs_to_jiffies(interval); | |
3168 | if (unlikely(!interval)) | |
3169 | interval = 1; | |
dd41f596 IM |
3170 | if (interval > HZ*NR_CPUS/10) |
3171 | interval = HZ*NR_CPUS/10; | |
3172 | ||
1da177e4 | 3173 | |
08c183f3 CL |
3174 | if (sd->flags & SD_SERIALIZE) { |
3175 | if (!spin_trylock(&balancing)) | |
3176 | goto out; | |
3177 | } | |
3178 | ||
c9819f45 | 3179 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
46cb4b7c | 3180 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
3181 | /* |
3182 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3183 | * longer idle, or one of our SMT siblings is |
3184 | * not idle. | |
3185 | */ | |
d15bcfdb | 3186 | idle = CPU_NOT_IDLE; |
1da177e4 | 3187 | } |
1bd77f2d | 3188 | sd->last_balance = jiffies; |
1da177e4 | 3189 | } |
08c183f3 CL |
3190 | if (sd->flags & SD_SERIALIZE) |
3191 | spin_unlock(&balancing); | |
3192 | out: | |
f549da84 | 3193 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3194 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3195 | update_next_balance = 1; |
3196 | } | |
783609c6 SS |
3197 | |
3198 | /* | |
3199 | * Stop the load balance at this level. There is another | |
3200 | * CPU in our sched group which is doing load balancing more | |
3201 | * actively. | |
3202 | */ | |
3203 | if (!balance) | |
3204 | break; | |
1da177e4 | 3205 | } |
f549da84 SS |
3206 | |
3207 | /* | |
3208 | * next_balance will be updated only when there is a need. | |
3209 | * When the cpu is attached to null domain for ex, it will not be | |
3210 | * updated. | |
3211 | */ | |
3212 | if (likely(update_next_balance)) | |
3213 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3214 | } |
3215 | ||
3216 | /* | |
3217 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3218 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3219 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3220 | */ | |
3221 | static void run_rebalance_domains(struct softirq_action *h) | |
3222 | { | |
dd41f596 IM |
3223 | int this_cpu = smp_processor_id(); |
3224 | struct rq *this_rq = cpu_rq(this_cpu); | |
3225 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3226 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3227 | |
dd41f596 | 3228 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3229 | |
3230 | #ifdef CONFIG_NO_HZ | |
3231 | /* | |
3232 | * If this cpu is the owner for idle load balancing, then do the | |
3233 | * balancing on behalf of the other idle cpus whose ticks are | |
3234 | * stopped. | |
3235 | */ | |
dd41f596 IM |
3236 | if (this_rq->idle_at_tick && |
3237 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3238 | cpumask_t cpus = nohz.cpu_mask; |
3239 | struct rq *rq; | |
3240 | int balance_cpu; | |
3241 | ||
dd41f596 | 3242 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3243 | for_each_cpu_mask(balance_cpu, cpus) { |
3244 | /* | |
3245 | * If this cpu gets work to do, stop the load balancing | |
3246 | * work being done for other cpus. Next load | |
3247 | * balancing owner will pick it up. | |
3248 | */ | |
3249 | if (need_resched()) | |
3250 | break; | |
3251 | ||
de0cf899 | 3252 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3253 | |
3254 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3255 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3256 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3257 | } |
3258 | } | |
3259 | #endif | |
3260 | } | |
3261 | ||
3262 | /* | |
3263 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3264 | * | |
3265 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3266 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3267 | * if the whole system is idle. | |
3268 | */ | |
dd41f596 | 3269 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3270 | { |
46cb4b7c SS |
3271 | #ifdef CONFIG_NO_HZ |
3272 | /* | |
3273 | * If we were in the nohz mode recently and busy at the current | |
3274 | * scheduler tick, then check if we need to nominate new idle | |
3275 | * load balancer. | |
3276 | */ | |
3277 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3278 | rq->in_nohz_recently = 0; | |
3279 | ||
3280 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3281 | cpu_clear(cpu, nohz.cpu_mask); | |
3282 | atomic_set(&nohz.load_balancer, -1); | |
3283 | } | |
3284 | ||
3285 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3286 | /* | |
3287 | * simple selection for now: Nominate the | |
3288 | * first cpu in the nohz list to be the next | |
3289 | * ilb owner. | |
3290 | * | |
3291 | * TBD: Traverse the sched domains and nominate | |
3292 | * the nearest cpu in the nohz.cpu_mask. | |
3293 | */ | |
3294 | int ilb = first_cpu(nohz.cpu_mask); | |
3295 | ||
3296 | if (ilb != NR_CPUS) | |
3297 | resched_cpu(ilb); | |
3298 | } | |
3299 | } | |
3300 | ||
3301 | /* | |
3302 | * If this cpu is idle and doing idle load balancing for all the | |
3303 | * cpus with ticks stopped, is it time for that to stop? | |
3304 | */ | |
3305 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3306 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3307 | resched_cpu(cpu); | |
3308 | return; | |
3309 | } | |
3310 | ||
3311 | /* | |
3312 | * If this cpu is idle and the idle load balancing is done by | |
3313 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3314 | */ | |
3315 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3316 | cpu_isset(cpu, nohz.cpu_mask)) | |
3317 | return; | |
3318 | #endif | |
3319 | if (time_after_eq(jiffies, rq->next_balance)) | |
3320 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3321 | } |
dd41f596 IM |
3322 | |
3323 | #else /* CONFIG_SMP */ | |
3324 | ||
1da177e4 LT |
3325 | /* |
3326 | * on UP we do not need to balance between CPUs: | |
3327 | */ | |
70b97a7f | 3328 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3329 | { |
3330 | } | |
dd41f596 | 3331 | |
1da177e4 LT |
3332 | #endif |
3333 | ||
1da177e4 LT |
3334 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3335 | ||
3336 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3337 | ||
3338 | /* | |
41b86e9c IM |
3339 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
3340 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 3341 | */ |
41b86e9c | 3342 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 3343 | { |
1da177e4 | 3344 | unsigned long flags; |
41b86e9c IM |
3345 | u64 ns, delta_exec; |
3346 | struct rq *rq; | |
48f24c4d | 3347 | |
41b86e9c IM |
3348 | rq = task_rq_lock(p, &flags); |
3349 | ns = p->se.sum_exec_runtime; | |
051a1d1a | 3350 | if (task_current(rq, p)) { |
a8e504d2 IM |
3351 | update_rq_clock(rq); |
3352 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
3353 | if ((s64)delta_exec > 0) |
3354 | ns += delta_exec; | |
3355 | } | |
3356 | task_rq_unlock(rq, &flags); | |
48f24c4d | 3357 | |
1da177e4 LT |
3358 | return ns; |
3359 | } | |
3360 | ||
1da177e4 LT |
3361 | /* |
3362 | * Account user cpu time to a process. | |
3363 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
3364 | * @cputime: the cpu time spent in user space since the last update |
3365 | */ | |
3366 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3367 | { | |
3368 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3369 | cputime64_t tmp; | |
3370 | ||
3371 | p->utime = cputime_add(p->utime, cputime); | |
3372 | ||
3373 | /* Add user time to cpustat. */ | |
3374 | tmp = cputime_to_cputime64(cputime); | |
3375 | if (TASK_NICE(p) > 0) | |
3376 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3377 | else | |
3378 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3379 | } | |
3380 | ||
94886b84 LV |
3381 | /* |
3382 | * Account guest cpu time to a process. | |
3383 | * @p: the process that the cpu time gets accounted to | |
3384 | * @cputime: the cpu time spent in virtual machine since the last update | |
3385 | */ | |
f7402e03 | 3386 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
3387 | { |
3388 | cputime64_t tmp; | |
3389 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3390 | ||
3391 | tmp = cputime_to_cputime64(cputime); | |
3392 | ||
3393 | p->utime = cputime_add(p->utime, cputime); | |
3394 | p->gtime = cputime_add(p->gtime, cputime); | |
3395 | ||
3396 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3397 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3398 | } | |
3399 | ||
c66f08be MN |
3400 | /* |
3401 | * Account scaled user cpu time to a process. | |
3402 | * @p: the process that the cpu time gets accounted to | |
3403 | * @cputime: the cpu time spent in user space since the last update | |
3404 | */ | |
3405 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
3406 | { | |
3407 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
3408 | } | |
3409 | ||
1da177e4 LT |
3410 | /* |
3411 | * Account system cpu time to a process. | |
3412 | * @p: the process that the cpu time gets accounted to | |
3413 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3414 | * @cputime: the cpu time spent in kernel space since the last update | |
3415 | */ | |
3416 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3417 | cputime_t cputime) | |
3418 | { | |
3419 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3420 | struct rq *rq = this_rq(); |
1da177e4 LT |
3421 | cputime64_t tmp; |
3422 | ||
9778385d CB |
3423 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) |
3424 | return account_guest_time(p, cputime); | |
94886b84 | 3425 | |
1da177e4 LT |
3426 | p->stime = cputime_add(p->stime, cputime); |
3427 | ||
3428 | /* Add system time to cpustat. */ | |
3429 | tmp = cputime_to_cputime64(cputime); | |
3430 | if (hardirq_count() - hardirq_offset) | |
3431 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3432 | else if (softirq_count()) | |
3433 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 3434 | else if (p != rq->idle) |
1da177e4 | 3435 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 3436 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
3437 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
3438 | else | |
3439 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3440 | /* Account for system time used */ | |
3441 | acct_update_integrals(p); | |
1da177e4 LT |
3442 | } |
3443 | ||
c66f08be MN |
3444 | /* |
3445 | * Account scaled system cpu time to a process. | |
3446 | * @p: the process that the cpu time gets accounted to | |
3447 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3448 | * @cputime: the cpu time spent in kernel space since the last update | |
3449 | */ | |
3450 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
3451 | { | |
3452 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
3453 | } | |
3454 | ||
1da177e4 LT |
3455 | /* |
3456 | * Account for involuntary wait time. | |
3457 | * @p: the process from which the cpu time has been stolen | |
3458 | * @steal: the cpu time spent in involuntary wait | |
3459 | */ | |
3460 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3461 | { | |
3462 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3463 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3464 | struct rq *rq = this_rq(); |
1da177e4 LT |
3465 | |
3466 | if (p == rq->idle) { | |
3467 | p->stime = cputime_add(p->stime, steal); | |
3468 | if (atomic_read(&rq->nr_iowait) > 0) | |
3469 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3470 | else | |
3471 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 3472 | } else |
1da177e4 LT |
3473 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
3474 | } | |
3475 | ||
7835b98b CL |
3476 | /* |
3477 | * This function gets called by the timer code, with HZ frequency. | |
3478 | * We call it with interrupts disabled. | |
3479 | * | |
3480 | * It also gets called by the fork code, when changing the parent's | |
3481 | * timeslices. | |
3482 | */ | |
3483 | void scheduler_tick(void) | |
3484 | { | |
7835b98b CL |
3485 | int cpu = smp_processor_id(); |
3486 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3487 | struct task_struct *curr = rq->curr; |
529c7726 | 3488 | u64 next_tick = rq->tick_timestamp + TICK_NSEC; |
dd41f596 IM |
3489 | |
3490 | spin_lock(&rq->lock); | |
546fe3c9 | 3491 | __update_rq_clock(rq); |
529c7726 IM |
3492 | /* |
3493 | * Let rq->clock advance by at least TICK_NSEC: | |
3494 | */ | |
3495 | if (unlikely(rq->clock < next_tick)) | |
3496 | rq->clock = next_tick; | |
3497 | rq->tick_timestamp = rq->clock; | |
f1a438d8 | 3498 | update_cpu_load(rq); |
dd41f596 IM |
3499 | if (curr != rq->idle) /* FIXME: needed? */ |
3500 | curr->sched_class->task_tick(rq, curr); | |
dd41f596 | 3501 | spin_unlock(&rq->lock); |
7835b98b | 3502 | |
e418e1c2 | 3503 | #ifdef CONFIG_SMP |
dd41f596 IM |
3504 | rq->idle_at_tick = idle_cpu(cpu); |
3505 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3506 | #endif |
1da177e4 LT |
3507 | } |
3508 | ||
1da177e4 LT |
3509 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3510 | ||
3511 | void fastcall add_preempt_count(int val) | |
3512 | { | |
3513 | /* | |
3514 | * Underflow? | |
3515 | */ | |
9a11b49a IM |
3516 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3517 | return; | |
1da177e4 LT |
3518 | preempt_count() += val; |
3519 | /* | |
3520 | * Spinlock count overflowing soon? | |
3521 | */ | |
33859f7f MOS |
3522 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3523 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3524 | } |
3525 | EXPORT_SYMBOL(add_preempt_count); | |
3526 | ||
3527 | void fastcall sub_preempt_count(int val) | |
3528 | { | |
3529 | /* | |
3530 | * Underflow? | |
3531 | */ | |
9a11b49a IM |
3532 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3533 | return; | |
1da177e4 LT |
3534 | /* |
3535 | * Is the spinlock portion underflowing? | |
3536 | */ | |
9a11b49a IM |
3537 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3538 | !(preempt_count() & PREEMPT_MASK))) | |
3539 | return; | |
3540 | ||
1da177e4 LT |
3541 | preempt_count() -= val; |
3542 | } | |
3543 | EXPORT_SYMBOL(sub_preempt_count); | |
3544 | ||
3545 | #endif | |
3546 | ||
3547 | /* | |
dd41f596 | 3548 | * Print scheduling while atomic bug: |
1da177e4 | 3549 | */ |
dd41f596 | 3550 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3551 | { |
838225b4 SS |
3552 | struct pt_regs *regs = get_irq_regs(); |
3553 | ||
3554 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
3555 | prev->comm, prev->pid, preempt_count()); | |
3556 | ||
dd41f596 IM |
3557 | debug_show_held_locks(prev); |
3558 | if (irqs_disabled()) | |
3559 | print_irqtrace_events(prev); | |
838225b4 SS |
3560 | |
3561 | if (regs) | |
3562 | show_regs(regs); | |
3563 | else | |
3564 | dump_stack(); | |
dd41f596 | 3565 | } |
1da177e4 | 3566 | |
dd41f596 IM |
3567 | /* |
3568 | * Various schedule()-time debugging checks and statistics: | |
3569 | */ | |
3570 | static inline void schedule_debug(struct task_struct *prev) | |
3571 | { | |
1da177e4 | 3572 | /* |
41a2d6cf | 3573 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3574 | * schedule() atomically, we ignore that path for now. |
3575 | * Otherwise, whine if we are scheduling when we should not be. | |
3576 | */ | |
dd41f596 IM |
3577 | if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state)) |
3578 | __schedule_bug(prev); | |
3579 | ||
1da177e4 LT |
3580 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3581 | ||
2d72376b | 3582 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3583 | #ifdef CONFIG_SCHEDSTATS |
3584 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3585 | schedstat_inc(this_rq(), bkl_count); |
3586 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3587 | } |
3588 | #endif | |
dd41f596 IM |
3589 | } |
3590 | ||
3591 | /* | |
3592 | * Pick up the highest-prio task: | |
3593 | */ | |
3594 | static inline struct task_struct * | |
ff95f3df | 3595 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 3596 | { |
5522d5d5 | 3597 | const struct sched_class *class; |
dd41f596 | 3598 | struct task_struct *p; |
1da177e4 LT |
3599 | |
3600 | /* | |
dd41f596 IM |
3601 | * Optimization: we know that if all tasks are in |
3602 | * the fair class we can call that function directly: | |
1da177e4 | 3603 | */ |
dd41f596 | 3604 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3605 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3606 | if (likely(p)) |
3607 | return p; | |
1da177e4 LT |
3608 | } |
3609 | ||
dd41f596 IM |
3610 | class = sched_class_highest; |
3611 | for ( ; ; ) { | |
fb8d4724 | 3612 | p = class->pick_next_task(rq); |
dd41f596 IM |
3613 | if (p) |
3614 | return p; | |
3615 | /* | |
3616 | * Will never be NULL as the idle class always | |
3617 | * returns a non-NULL p: | |
3618 | */ | |
3619 | class = class->next; | |
3620 | } | |
3621 | } | |
1da177e4 | 3622 | |
dd41f596 IM |
3623 | /* |
3624 | * schedule() is the main scheduler function. | |
3625 | */ | |
3626 | asmlinkage void __sched schedule(void) | |
3627 | { | |
3628 | struct task_struct *prev, *next; | |
3629 | long *switch_count; | |
3630 | struct rq *rq; | |
dd41f596 IM |
3631 | int cpu; |
3632 | ||
3633 | need_resched: | |
3634 | preempt_disable(); | |
3635 | cpu = smp_processor_id(); | |
3636 | rq = cpu_rq(cpu); | |
3637 | rcu_qsctr_inc(cpu); | |
3638 | prev = rq->curr; | |
3639 | switch_count = &prev->nivcsw; | |
3640 | ||
3641 | release_kernel_lock(prev); | |
3642 | need_resched_nonpreemptible: | |
3643 | ||
3644 | schedule_debug(prev); | |
1da177e4 | 3645 | |
1e819950 IM |
3646 | /* |
3647 | * Do the rq-clock update outside the rq lock: | |
3648 | */ | |
3649 | local_irq_disable(); | |
c1b3da3e | 3650 | __update_rq_clock(rq); |
1e819950 IM |
3651 | spin_lock(&rq->lock); |
3652 | clear_tsk_need_resched(prev); | |
1da177e4 | 3653 | |
1da177e4 | 3654 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
1da177e4 | 3655 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && |
dd41f596 | 3656 | unlikely(signal_pending(prev)))) { |
1da177e4 | 3657 | prev->state = TASK_RUNNING; |
dd41f596 | 3658 | } else { |
2e1cb74a | 3659 | deactivate_task(rq, prev, 1); |
1da177e4 | 3660 | } |
dd41f596 | 3661 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3662 | } |
3663 | ||
9a897c5a SR |
3664 | #ifdef CONFIG_SMP |
3665 | if (prev->sched_class->pre_schedule) | |
3666 | prev->sched_class->pre_schedule(rq, prev); | |
3667 | #endif | |
f65eda4f | 3668 | |
dd41f596 | 3669 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3670 | idle_balance(cpu, rq); |
1da177e4 | 3671 | |
31ee529c | 3672 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 3673 | next = pick_next_task(rq, prev); |
1da177e4 LT |
3674 | |
3675 | sched_info_switch(prev, next); | |
dd41f596 | 3676 | |
1da177e4 | 3677 | if (likely(prev != next)) { |
1da177e4 LT |
3678 | rq->nr_switches++; |
3679 | rq->curr = next; | |
3680 | ++*switch_count; | |
3681 | ||
dd41f596 | 3682 | context_switch(rq, prev, next); /* unlocks the rq */ |
1da177e4 LT |
3683 | } else |
3684 | spin_unlock_irq(&rq->lock); | |
3685 | ||
dd41f596 IM |
3686 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
3687 | cpu = smp_processor_id(); | |
3688 | rq = cpu_rq(cpu); | |
1da177e4 | 3689 | goto need_resched_nonpreemptible; |
dd41f596 | 3690 | } |
1da177e4 LT |
3691 | preempt_enable_no_resched(); |
3692 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3693 | goto need_resched; | |
3694 | } | |
1da177e4 LT |
3695 | EXPORT_SYMBOL(schedule); |
3696 | ||
3697 | #ifdef CONFIG_PREEMPT | |
3698 | /* | |
2ed6e34f | 3699 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3700 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3701 | * occur there and call schedule directly. |
3702 | */ | |
3703 | asmlinkage void __sched preempt_schedule(void) | |
3704 | { | |
3705 | struct thread_info *ti = current_thread_info(); | |
3706 | #ifdef CONFIG_PREEMPT_BKL | |
3707 | struct task_struct *task = current; | |
3708 | int saved_lock_depth; | |
3709 | #endif | |
3710 | /* | |
3711 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3712 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3713 | */ |
beed33a8 | 3714 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3715 | return; |
3716 | ||
3a5c359a AK |
3717 | do { |
3718 | add_preempt_count(PREEMPT_ACTIVE); | |
3719 | ||
3720 | /* | |
3721 | * We keep the big kernel semaphore locked, but we | |
3722 | * clear ->lock_depth so that schedule() doesnt | |
3723 | * auto-release the semaphore: | |
3724 | */ | |
1da177e4 | 3725 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a AK |
3726 | saved_lock_depth = task->lock_depth; |
3727 | task->lock_depth = -1; | |
1da177e4 | 3728 | #endif |
3a5c359a | 3729 | schedule(); |
1da177e4 | 3730 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a | 3731 | task->lock_depth = saved_lock_depth; |
1da177e4 | 3732 | #endif |
3a5c359a | 3733 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3734 | |
3a5c359a AK |
3735 | /* |
3736 | * Check again in case we missed a preemption opportunity | |
3737 | * between schedule and now. | |
3738 | */ | |
3739 | barrier(); | |
3740 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 3741 | } |
1da177e4 LT |
3742 | EXPORT_SYMBOL(preempt_schedule); |
3743 | ||
3744 | /* | |
2ed6e34f | 3745 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3746 | * off of irq context. |
3747 | * Note, that this is called and return with irqs disabled. This will | |
3748 | * protect us against recursive calling from irq. | |
3749 | */ | |
3750 | asmlinkage void __sched preempt_schedule_irq(void) | |
3751 | { | |
3752 | struct thread_info *ti = current_thread_info(); | |
3753 | #ifdef CONFIG_PREEMPT_BKL | |
3754 | struct task_struct *task = current; | |
3755 | int saved_lock_depth; | |
3756 | #endif | |
2ed6e34f | 3757 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3758 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3759 | ||
3a5c359a AK |
3760 | do { |
3761 | add_preempt_count(PREEMPT_ACTIVE); | |
3762 | ||
3763 | /* | |
3764 | * We keep the big kernel semaphore locked, but we | |
3765 | * clear ->lock_depth so that schedule() doesnt | |
3766 | * auto-release the semaphore: | |
3767 | */ | |
1da177e4 | 3768 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a AK |
3769 | saved_lock_depth = task->lock_depth; |
3770 | task->lock_depth = -1; | |
1da177e4 | 3771 | #endif |
3a5c359a AK |
3772 | local_irq_enable(); |
3773 | schedule(); | |
3774 | local_irq_disable(); | |
1da177e4 | 3775 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a | 3776 | task->lock_depth = saved_lock_depth; |
1da177e4 | 3777 | #endif |
3a5c359a | 3778 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3779 | |
3a5c359a AK |
3780 | /* |
3781 | * Check again in case we missed a preemption opportunity | |
3782 | * between schedule and now. | |
3783 | */ | |
3784 | barrier(); | |
3785 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
3786 | } |
3787 | ||
3788 | #endif /* CONFIG_PREEMPT */ | |
3789 | ||
95cdf3b7 IM |
3790 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
3791 | void *key) | |
1da177e4 | 3792 | { |
48f24c4d | 3793 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 3794 | } |
1da177e4 LT |
3795 | EXPORT_SYMBOL(default_wake_function); |
3796 | ||
3797 | /* | |
41a2d6cf IM |
3798 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3799 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3800 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3801 | * | |
3802 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3803 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3804 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3805 | */ | |
3806 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
3807 | int nr_exclusive, int sync, void *key) | |
3808 | { | |
2e45874c | 3809 | wait_queue_t *curr, *next; |
1da177e4 | 3810 | |
2e45874c | 3811 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3812 | unsigned flags = curr->flags; |
3813 | ||
1da177e4 | 3814 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 3815 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3816 | break; |
3817 | } | |
3818 | } | |
3819 | ||
3820 | /** | |
3821 | * __wake_up - wake up threads blocked on a waitqueue. | |
3822 | * @q: the waitqueue | |
3823 | * @mode: which threads | |
3824 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3825 | * @key: is directly passed to the wakeup function |
1da177e4 LT |
3826 | */ |
3827 | void fastcall __wake_up(wait_queue_head_t *q, unsigned int mode, | |
95cdf3b7 | 3828 | int nr_exclusive, void *key) |
1da177e4 LT |
3829 | { |
3830 | unsigned long flags; | |
3831 | ||
3832 | spin_lock_irqsave(&q->lock, flags); | |
3833 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3834 | spin_unlock_irqrestore(&q->lock, flags); | |
3835 | } | |
1da177e4 LT |
3836 | EXPORT_SYMBOL(__wake_up); |
3837 | ||
3838 | /* | |
3839 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3840 | */ | |
3841 | void fastcall __wake_up_locked(wait_queue_head_t *q, unsigned int mode) | |
3842 | { | |
3843 | __wake_up_common(q, mode, 1, 0, NULL); | |
3844 | } | |
3845 | ||
3846 | /** | |
67be2dd1 | 3847 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3848 | * @q: the waitqueue |
3849 | * @mode: which threads | |
3850 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
3851 | * | |
3852 | * The sync wakeup differs that the waker knows that it will schedule | |
3853 | * away soon, so while the target thread will be woken up, it will not | |
3854 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3855 | * with each other. This can prevent needless bouncing between CPUs. | |
3856 | * | |
3857 | * On UP it can prevent extra preemption. | |
3858 | */ | |
95cdf3b7 IM |
3859 | void fastcall |
3860 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
1da177e4 LT |
3861 | { |
3862 | unsigned long flags; | |
3863 | int sync = 1; | |
3864 | ||
3865 | if (unlikely(!q)) | |
3866 | return; | |
3867 | ||
3868 | if (unlikely(!nr_exclusive)) | |
3869 | sync = 0; | |
3870 | ||
3871 | spin_lock_irqsave(&q->lock, flags); | |
3872 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
3873 | spin_unlock_irqrestore(&q->lock, flags); | |
3874 | } | |
3875 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
3876 | ||
b15136e9 | 3877 | void complete(struct completion *x) |
1da177e4 LT |
3878 | { |
3879 | unsigned long flags; | |
3880 | ||
3881 | spin_lock_irqsave(&x->wait.lock, flags); | |
3882 | x->done++; | |
3883 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3884 | 1, 0, NULL); | |
3885 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3886 | } | |
3887 | EXPORT_SYMBOL(complete); | |
3888 | ||
b15136e9 | 3889 | void complete_all(struct completion *x) |
1da177e4 LT |
3890 | { |
3891 | unsigned long flags; | |
3892 | ||
3893 | spin_lock_irqsave(&x->wait.lock, flags); | |
3894 | x->done += UINT_MAX/2; | |
3895 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3896 | 0, 0, NULL); | |
3897 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3898 | } | |
3899 | EXPORT_SYMBOL(complete_all); | |
3900 | ||
8cbbe86d AK |
3901 | static inline long __sched |
3902 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3903 | { |
1da177e4 LT |
3904 | if (!x->done) { |
3905 | DECLARE_WAITQUEUE(wait, current); | |
3906 | ||
3907 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3908 | __add_wait_queue_tail(&x->wait, &wait); | |
3909 | do { | |
8cbbe86d AK |
3910 | if (state == TASK_INTERRUPTIBLE && |
3911 | signal_pending(current)) { | |
3912 | __remove_wait_queue(&x->wait, &wait); | |
3913 | return -ERESTARTSYS; | |
3914 | } | |
3915 | __set_current_state(state); | |
1da177e4 LT |
3916 | spin_unlock_irq(&x->wait.lock); |
3917 | timeout = schedule_timeout(timeout); | |
3918 | spin_lock_irq(&x->wait.lock); | |
3919 | if (!timeout) { | |
3920 | __remove_wait_queue(&x->wait, &wait); | |
8cbbe86d | 3921 | return timeout; |
1da177e4 LT |
3922 | } |
3923 | } while (!x->done); | |
3924 | __remove_wait_queue(&x->wait, &wait); | |
3925 | } | |
3926 | x->done--; | |
1da177e4 LT |
3927 | return timeout; |
3928 | } | |
1da177e4 | 3929 | |
8cbbe86d AK |
3930 | static long __sched |
3931 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3932 | { |
1da177e4 LT |
3933 | might_sleep(); |
3934 | ||
3935 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 3936 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 3937 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3938 | return timeout; |
3939 | } | |
1da177e4 | 3940 | |
b15136e9 | 3941 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
3942 | { |
3943 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 3944 | } |
8cbbe86d | 3945 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 3946 | |
b15136e9 | 3947 | unsigned long __sched |
8cbbe86d | 3948 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 3949 | { |
8cbbe86d | 3950 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 3951 | } |
8cbbe86d | 3952 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 3953 | |
8cbbe86d | 3954 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 3955 | { |
51e97990 AK |
3956 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
3957 | if (t == -ERESTARTSYS) | |
3958 | return t; | |
3959 | return 0; | |
0fec171c | 3960 | } |
8cbbe86d | 3961 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 3962 | |
b15136e9 | 3963 | unsigned long __sched |
8cbbe86d AK |
3964 | wait_for_completion_interruptible_timeout(struct completion *x, |
3965 | unsigned long timeout) | |
0fec171c | 3966 | { |
8cbbe86d | 3967 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 3968 | } |
8cbbe86d | 3969 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 3970 | |
8cbbe86d AK |
3971 | static long __sched |
3972 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 3973 | { |
0fec171c IM |
3974 | unsigned long flags; |
3975 | wait_queue_t wait; | |
3976 | ||
3977 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 3978 | |
8cbbe86d | 3979 | __set_current_state(state); |
1da177e4 | 3980 | |
8cbbe86d AK |
3981 | spin_lock_irqsave(&q->lock, flags); |
3982 | __add_wait_queue(q, &wait); | |
3983 | spin_unlock(&q->lock); | |
3984 | timeout = schedule_timeout(timeout); | |
3985 | spin_lock_irq(&q->lock); | |
3986 | __remove_wait_queue(q, &wait); | |
3987 | spin_unlock_irqrestore(&q->lock, flags); | |
3988 | ||
3989 | return timeout; | |
3990 | } | |
3991 | ||
3992 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
3993 | { | |
3994 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 3995 | } |
1da177e4 LT |
3996 | EXPORT_SYMBOL(interruptible_sleep_on); |
3997 | ||
0fec171c | 3998 | long __sched |
95cdf3b7 | 3999 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4000 | { |
8cbbe86d | 4001 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4002 | } |
1da177e4 LT |
4003 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4004 | ||
0fec171c | 4005 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4006 | { |
8cbbe86d | 4007 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4008 | } |
1da177e4 LT |
4009 | EXPORT_SYMBOL(sleep_on); |
4010 | ||
0fec171c | 4011 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4012 | { |
8cbbe86d | 4013 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4014 | } |
1da177e4 LT |
4015 | EXPORT_SYMBOL(sleep_on_timeout); |
4016 | ||
b29739f9 IM |
4017 | #ifdef CONFIG_RT_MUTEXES |
4018 | ||
4019 | /* | |
4020 | * rt_mutex_setprio - set the current priority of a task | |
4021 | * @p: task | |
4022 | * @prio: prio value (kernel-internal form) | |
4023 | * | |
4024 | * This function changes the 'effective' priority of a task. It does | |
4025 | * not touch ->normal_prio like __setscheduler(). | |
4026 | * | |
4027 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4028 | */ | |
36c8b586 | 4029 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4030 | { |
4031 | unsigned long flags; | |
83b699ed | 4032 | int oldprio, on_rq, running; |
70b97a7f | 4033 | struct rq *rq; |
cb469845 | 4034 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
4035 | |
4036 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4037 | ||
4038 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4039 | update_rq_clock(rq); |
b29739f9 | 4040 | |
d5f9f942 | 4041 | oldprio = p->prio; |
dd41f596 | 4042 | on_rq = p->se.on_rq; |
051a1d1a | 4043 | running = task_current(rq, p); |
83b699ed | 4044 | if (on_rq) { |
69be72c1 | 4045 | dequeue_task(rq, p, 0); |
83b699ed SV |
4046 | if (running) |
4047 | p->sched_class->put_prev_task(rq, p); | |
4048 | } | |
dd41f596 IM |
4049 | |
4050 | if (rt_prio(prio)) | |
4051 | p->sched_class = &rt_sched_class; | |
4052 | else | |
4053 | p->sched_class = &fair_sched_class; | |
4054 | ||
b29739f9 IM |
4055 | p->prio = prio; |
4056 | ||
dd41f596 | 4057 | if (on_rq) { |
83b699ed SV |
4058 | if (running) |
4059 | p->sched_class->set_curr_task(rq); | |
cb469845 | 4060 | |
8159f87e | 4061 | enqueue_task(rq, p, 0); |
cb469845 SR |
4062 | |
4063 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4064 | } |
4065 | task_rq_unlock(rq, &flags); | |
4066 | } | |
4067 | ||
4068 | #endif | |
4069 | ||
36c8b586 | 4070 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4071 | { |
dd41f596 | 4072 | int old_prio, delta, on_rq; |
1da177e4 | 4073 | unsigned long flags; |
70b97a7f | 4074 | struct rq *rq; |
1da177e4 LT |
4075 | |
4076 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4077 | return; | |
4078 | /* | |
4079 | * We have to be careful, if called from sys_setpriority(), | |
4080 | * the task might be in the middle of scheduling on another CPU. | |
4081 | */ | |
4082 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4083 | update_rq_clock(rq); |
1da177e4 LT |
4084 | /* |
4085 | * The RT priorities are set via sched_setscheduler(), but we still | |
4086 | * allow the 'normal' nice value to be set - but as expected | |
4087 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4088 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4089 | */ |
e05606d3 | 4090 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4091 | p->static_prio = NICE_TO_PRIO(nice); |
4092 | goto out_unlock; | |
4093 | } | |
dd41f596 | 4094 | on_rq = p->se.on_rq; |
58e2d4ca | 4095 | if (on_rq) |
69be72c1 | 4096 | dequeue_task(rq, p, 0); |
1da177e4 | 4097 | |
1da177e4 | 4098 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4099 | set_load_weight(p); |
b29739f9 IM |
4100 | old_prio = p->prio; |
4101 | p->prio = effective_prio(p); | |
4102 | delta = p->prio - old_prio; | |
1da177e4 | 4103 | |
dd41f596 | 4104 | if (on_rq) { |
8159f87e | 4105 | enqueue_task(rq, p, 0); |
1da177e4 | 4106 | /* |
d5f9f942 AM |
4107 | * If the task increased its priority or is running and |
4108 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4109 | */ |
d5f9f942 | 4110 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4111 | resched_task(rq->curr); |
4112 | } | |
4113 | out_unlock: | |
4114 | task_rq_unlock(rq, &flags); | |
4115 | } | |
1da177e4 LT |
4116 | EXPORT_SYMBOL(set_user_nice); |
4117 | ||
e43379f1 MM |
4118 | /* |
4119 | * can_nice - check if a task can reduce its nice value | |
4120 | * @p: task | |
4121 | * @nice: nice value | |
4122 | */ | |
36c8b586 | 4123 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4124 | { |
024f4747 MM |
4125 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4126 | int nice_rlim = 20 - nice; | |
48f24c4d | 4127 | |
e43379f1 MM |
4128 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4129 | capable(CAP_SYS_NICE)); | |
4130 | } | |
4131 | ||
1da177e4 LT |
4132 | #ifdef __ARCH_WANT_SYS_NICE |
4133 | ||
4134 | /* | |
4135 | * sys_nice - change the priority of the current process. | |
4136 | * @increment: priority increment | |
4137 | * | |
4138 | * sys_setpriority is a more generic, but much slower function that | |
4139 | * does similar things. | |
4140 | */ | |
4141 | asmlinkage long sys_nice(int increment) | |
4142 | { | |
48f24c4d | 4143 | long nice, retval; |
1da177e4 LT |
4144 | |
4145 | /* | |
4146 | * Setpriority might change our priority at the same moment. | |
4147 | * We don't have to worry. Conceptually one call occurs first | |
4148 | * and we have a single winner. | |
4149 | */ | |
e43379f1 MM |
4150 | if (increment < -40) |
4151 | increment = -40; | |
1da177e4 LT |
4152 | if (increment > 40) |
4153 | increment = 40; | |
4154 | ||
4155 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4156 | if (nice < -20) | |
4157 | nice = -20; | |
4158 | if (nice > 19) | |
4159 | nice = 19; | |
4160 | ||
e43379f1 MM |
4161 | if (increment < 0 && !can_nice(current, nice)) |
4162 | return -EPERM; | |
4163 | ||
1da177e4 LT |
4164 | retval = security_task_setnice(current, nice); |
4165 | if (retval) | |
4166 | return retval; | |
4167 | ||
4168 | set_user_nice(current, nice); | |
4169 | return 0; | |
4170 | } | |
4171 | ||
4172 | #endif | |
4173 | ||
4174 | /** | |
4175 | * task_prio - return the priority value of a given task. | |
4176 | * @p: the task in question. | |
4177 | * | |
4178 | * This is the priority value as seen by users in /proc. | |
4179 | * RT tasks are offset by -200. Normal tasks are centered | |
4180 | * around 0, value goes from -16 to +15. | |
4181 | */ | |
36c8b586 | 4182 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4183 | { |
4184 | return p->prio - MAX_RT_PRIO; | |
4185 | } | |
4186 | ||
4187 | /** | |
4188 | * task_nice - return the nice value of a given task. | |
4189 | * @p: the task in question. | |
4190 | */ | |
36c8b586 | 4191 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4192 | { |
4193 | return TASK_NICE(p); | |
4194 | } | |
1da177e4 | 4195 | EXPORT_SYMBOL_GPL(task_nice); |
1da177e4 LT |
4196 | |
4197 | /** | |
4198 | * idle_cpu - is a given cpu idle currently? | |
4199 | * @cpu: the processor in question. | |
4200 | */ | |
4201 | int idle_cpu(int cpu) | |
4202 | { | |
4203 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4204 | } | |
4205 | ||
1da177e4 LT |
4206 | /** |
4207 | * idle_task - return the idle task for a given cpu. | |
4208 | * @cpu: the processor in question. | |
4209 | */ | |
36c8b586 | 4210 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4211 | { |
4212 | return cpu_rq(cpu)->idle; | |
4213 | } | |
4214 | ||
4215 | /** | |
4216 | * find_process_by_pid - find a process with a matching PID value. | |
4217 | * @pid: the pid in question. | |
4218 | */ | |
a9957449 | 4219 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4220 | { |
228ebcbe | 4221 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4222 | } |
4223 | ||
4224 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4225 | static void |
4226 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4227 | { |
dd41f596 | 4228 | BUG_ON(p->se.on_rq); |
48f24c4d | 4229 | |
1da177e4 | 4230 | p->policy = policy; |
dd41f596 IM |
4231 | switch (p->policy) { |
4232 | case SCHED_NORMAL: | |
4233 | case SCHED_BATCH: | |
4234 | case SCHED_IDLE: | |
4235 | p->sched_class = &fair_sched_class; | |
4236 | break; | |
4237 | case SCHED_FIFO: | |
4238 | case SCHED_RR: | |
4239 | p->sched_class = &rt_sched_class; | |
4240 | break; | |
4241 | } | |
4242 | ||
1da177e4 | 4243 | p->rt_priority = prio; |
b29739f9 IM |
4244 | p->normal_prio = normal_prio(p); |
4245 | /* we are holding p->pi_lock already */ | |
4246 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4247 | set_load_weight(p); |
1da177e4 LT |
4248 | } |
4249 | ||
4250 | /** | |
72fd4a35 | 4251 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4252 | * @p: the task in question. |
4253 | * @policy: new policy. | |
4254 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4255 | * |
72fd4a35 | 4256 | * NOTE that the task may be already dead. |
1da177e4 | 4257 | */ |
95cdf3b7 IM |
4258 | int sched_setscheduler(struct task_struct *p, int policy, |
4259 | struct sched_param *param) | |
1da177e4 | 4260 | { |
83b699ed | 4261 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4262 | unsigned long flags; |
cb469845 | 4263 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 4264 | struct rq *rq; |
1da177e4 | 4265 | |
66e5393a SR |
4266 | /* may grab non-irq protected spin_locks */ |
4267 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4268 | recheck: |
4269 | /* double check policy once rq lock held */ | |
4270 | if (policy < 0) | |
4271 | policy = oldpolicy = p->policy; | |
4272 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4273 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4274 | policy != SCHED_IDLE) | |
b0a9499c | 4275 | return -EINVAL; |
1da177e4 LT |
4276 | /* |
4277 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4278 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4279 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4280 | */ |
4281 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4282 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4283 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4284 | return -EINVAL; |
e05606d3 | 4285 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4286 | return -EINVAL; |
4287 | ||
37e4ab3f OC |
4288 | /* |
4289 | * Allow unprivileged RT tasks to decrease priority: | |
4290 | */ | |
4291 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4292 | if (rt_policy(policy)) { |
8dc3e909 | 4293 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4294 | |
4295 | if (!lock_task_sighand(p, &flags)) | |
4296 | return -ESRCH; | |
4297 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4298 | unlock_task_sighand(p, &flags); | |
4299 | ||
4300 | /* can't set/change the rt policy */ | |
4301 | if (policy != p->policy && !rlim_rtprio) | |
4302 | return -EPERM; | |
4303 | ||
4304 | /* can't increase priority */ | |
4305 | if (param->sched_priority > p->rt_priority && | |
4306 | param->sched_priority > rlim_rtprio) | |
4307 | return -EPERM; | |
4308 | } | |
dd41f596 IM |
4309 | /* |
4310 | * Like positive nice levels, dont allow tasks to | |
4311 | * move out of SCHED_IDLE either: | |
4312 | */ | |
4313 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4314 | return -EPERM; | |
5fe1d75f | 4315 | |
37e4ab3f OC |
4316 | /* can't change other user's priorities */ |
4317 | if ((current->euid != p->euid) && | |
4318 | (current->euid != p->uid)) | |
4319 | return -EPERM; | |
4320 | } | |
1da177e4 LT |
4321 | |
4322 | retval = security_task_setscheduler(p, policy, param); | |
4323 | if (retval) | |
4324 | return retval; | |
b29739f9 IM |
4325 | /* |
4326 | * make sure no PI-waiters arrive (or leave) while we are | |
4327 | * changing the priority of the task: | |
4328 | */ | |
4329 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4330 | /* |
4331 | * To be able to change p->policy safely, the apropriate | |
4332 | * runqueue lock must be held. | |
4333 | */ | |
b29739f9 | 4334 | rq = __task_rq_lock(p); |
1da177e4 LT |
4335 | /* recheck policy now with rq lock held */ |
4336 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4337 | policy = oldpolicy = -1; | |
b29739f9 IM |
4338 | __task_rq_unlock(rq); |
4339 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4340 | goto recheck; |
4341 | } | |
2daa3577 | 4342 | update_rq_clock(rq); |
dd41f596 | 4343 | on_rq = p->se.on_rq; |
051a1d1a | 4344 | running = task_current(rq, p); |
83b699ed | 4345 | if (on_rq) { |
2e1cb74a | 4346 | deactivate_task(rq, p, 0); |
83b699ed SV |
4347 | if (running) |
4348 | p->sched_class->put_prev_task(rq, p); | |
4349 | } | |
f6b53205 | 4350 | |
1da177e4 | 4351 | oldprio = p->prio; |
dd41f596 | 4352 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4353 | |
dd41f596 | 4354 | if (on_rq) { |
83b699ed SV |
4355 | if (running) |
4356 | p->sched_class->set_curr_task(rq); | |
cb469845 | 4357 | |
dd41f596 | 4358 | activate_task(rq, p, 0); |
cb469845 SR |
4359 | |
4360 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 4361 | } |
b29739f9 IM |
4362 | __task_rq_unlock(rq); |
4363 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4364 | ||
95e02ca9 TG |
4365 | rt_mutex_adjust_pi(p); |
4366 | ||
1da177e4 LT |
4367 | return 0; |
4368 | } | |
4369 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4370 | ||
95cdf3b7 IM |
4371 | static int |
4372 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4373 | { |
1da177e4 LT |
4374 | struct sched_param lparam; |
4375 | struct task_struct *p; | |
36c8b586 | 4376 | int retval; |
1da177e4 LT |
4377 | |
4378 | if (!param || pid < 0) | |
4379 | return -EINVAL; | |
4380 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4381 | return -EFAULT; | |
5fe1d75f ON |
4382 | |
4383 | rcu_read_lock(); | |
4384 | retval = -ESRCH; | |
1da177e4 | 4385 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4386 | if (p != NULL) |
4387 | retval = sched_setscheduler(p, policy, &lparam); | |
4388 | rcu_read_unlock(); | |
36c8b586 | 4389 | |
1da177e4 LT |
4390 | return retval; |
4391 | } | |
4392 | ||
4393 | /** | |
4394 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4395 | * @pid: the pid in question. | |
4396 | * @policy: new policy. | |
4397 | * @param: structure containing the new RT priority. | |
4398 | */ | |
41a2d6cf IM |
4399 | asmlinkage long |
4400 | sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4401 | { |
c21761f1 JB |
4402 | /* negative values for policy are not valid */ |
4403 | if (policy < 0) | |
4404 | return -EINVAL; | |
4405 | ||
1da177e4 LT |
4406 | return do_sched_setscheduler(pid, policy, param); |
4407 | } | |
4408 | ||
4409 | /** | |
4410 | * sys_sched_setparam - set/change the RT priority of a thread | |
4411 | * @pid: the pid in question. | |
4412 | * @param: structure containing the new RT priority. | |
4413 | */ | |
4414 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4415 | { | |
4416 | return do_sched_setscheduler(pid, -1, param); | |
4417 | } | |
4418 | ||
4419 | /** | |
4420 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4421 | * @pid: the pid in question. | |
4422 | */ | |
4423 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4424 | { | |
36c8b586 | 4425 | struct task_struct *p; |
3a5c359a | 4426 | int retval; |
1da177e4 LT |
4427 | |
4428 | if (pid < 0) | |
3a5c359a | 4429 | return -EINVAL; |
1da177e4 LT |
4430 | |
4431 | retval = -ESRCH; | |
4432 | read_lock(&tasklist_lock); | |
4433 | p = find_process_by_pid(pid); | |
4434 | if (p) { | |
4435 | retval = security_task_getscheduler(p); | |
4436 | if (!retval) | |
4437 | retval = p->policy; | |
4438 | } | |
4439 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
4440 | return retval; |
4441 | } | |
4442 | ||
4443 | /** | |
4444 | * sys_sched_getscheduler - get the RT priority of a thread | |
4445 | * @pid: the pid in question. | |
4446 | * @param: structure containing the RT priority. | |
4447 | */ | |
4448 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
4449 | { | |
4450 | struct sched_param lp; | |
36c8b586 | 4451 | struct task_struct *p; |
3a5c359a | 4452 | int retval; |
1da177e4 LT |
4453 | |
4454 | if (!param || pid < 0) | |
3a5c359a | 4455 | return -EINVAL; |
1da177e4 LT |
4456 | |
4457 | read_lock(&tasklist_lock); | |
4458 | p = find_process_by_pid(pid); | |
4459 | retval = -ESRCH; | |
4460 | if (!p) | |
4461 | goto out_unlock; | |
4462 | ||
4463 | retval = security_task_getscheduler(p); | |
4464 | if (retval) | |
4465 | goto out_unlock; | |
4466 | ||
4467 | lp.sched_priority = p->rt_priority; | |
4468 | read_unlock(&tasklist_lock); | |
4469 | ||
4470 | /* | |
4471 | * This one might sleep, we cannot do it with a spinlock held ... | |
4472 | */ | |
4473 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4474 | ||
1da177e4 LT |
4475 | return retval; |
4476 | ||
4477 | out_unlock: | |
4478 | read_unlock(&tasklist_lock); | |
4479 | return retval; | |
4480 | } | |
4481 | ||
4482 | long sched_setaffinity(pid_t pid, cpumask_t new_mask) | |
4483 | { | |
1da177e4 | 4484 | cpumask_t cpus_allowed; |
36c8b586 IM |
4485 | struct task_struct *p; |
4486 | int retval; | |
1da177e4 | 4487 | |
95402b38 | 4488 | get_online_cpus(); |
1da177e4 LT |
4489 | read_lock(&tasklist_lock); |
4490 | ||
4491 | p = find_process_by_pid(pid); | |
4492 | if (!p) { | |
4493 | read_unlock(&tasklist_lock); | |
95402b38 | 4494 | put_online_cpus(); |
1da177e4 LT |
4495 | return -ESRCH; |
4496 | } | |
4497 | ||
4498 | /* | |
4499 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 4500 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
4501 | * usage count and then drop tasklist_lock. |
4502 | */ | |
4503 | get_task_struct(p); | |
4504 | read_unlock(&tasklist_lock); | |
4505 | ||
4506 | retval = -EPERM; | |
4507 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
4508 | !capable(CAP_SYS_NICE)) | |
4509 | goto out_unlock; | |
4510 | ||
e7834f8f DQ |
4511 | retval = security_task_setscheduler(p, 0, NULL); |
4512 | if (retval) | |
4513 | goto out_unlock; | |
4514 | ||
1da177e4 LT |
4515 | cpus_allowed = cpuset_cpus_allowed(p); |
4516 | cpus_and(new_mask, new_mask, cpus_allowed); | |
8707d8b8 | 4517 | again: |
1da177e4 LT |
4518 | retval = set_cpus_allowed(p, new_mask); |
4519 | ||
8707d8b8 PM |
4520 | if (!retval) { |
4521 | cpus_allowed = cpuset_cpus_allowed(p); | |
4522 | if (!cpus_subset(new_mask, cpus_allowed)) { | |
4523 | /* | |
4524 | * We must have raced with a concurrent cpuset | |
4525 | * update. Just reset the cpus_allowed to the | |
4526 | * cpuset's cpus_allowed | |
4527 | */ | |
4528 | new_mask = cpus_allowed; | |
4529 | goto again; | |
4530 | } | |
4531 | } | |
1da177e4 LT |
4532 | out_unlock: |
4533 | put_task_struct(p); | |
95402b38 | 4534 | put_online_cpus(); |
1da177e4 LT |
4535 | return retval; |
4536 | } | |
4537 | ||
4538 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
4539 | cpumask_t *new_mask) | |
4540 | { | |
4541 | if (len < sizeof(cpumask_t)) { | |
4542 | memset(new_mask, 0, sizeof(cpumask_t)); | |
4543 | } else if (len > sizeof(cpumask_t)) { | |
4544 | len = sizeof(cpumask_t); | |
4545 | } | |
4546 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
4547 | } | |
4548 | ||
4549 | /** | |
4550 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4551 | * @pid: pid of the process | |
4552 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4553 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4554 | */ | |
4555 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
4556 | unsigned long __user *user_mask_ptr) | |
4557 | { | |
4558 | cpumask_t new_mask; | |
4559 | int retval; | |
4560 | ||
4561 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
4562 | if (retval) | |
4563 | return retval; | |
4564 | ||
4565 | return sched_setaffinity(pid, new_mask); | |
4566 | } | |
4567 | ||
4568 | /* | |
4569 | * Represents all cpu's present in the system | |
4570 | * In systems capable of hotplug, this map could dynamically grow | |
4571 | * as new cpu's are detected in the system via any platform specific | |
4572 | * method, such as ACPI for e.g. | |
4573 | */ | |
4574 | ||
4cef0c61 | 4575 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
4576 | EXPORT_SYMBOL(cpu_present_map); |
4577 | ||
4578 | #ifndef CONFIG_SMP | |
4cef0c61 | 4579 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
4580 | EXPORT_SYMBOL(cpu_online_map); |
4581 | ||
4cef0c61 | 4582 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 4583 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
4584 | #endif |
4585 | ||
4586 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
4587 | { | |
36c8b586 | 4588 | struct task_struct *p; |
1da177e4 | 4589 | int retval; |
1da177e4 | 4590 | |
95402b38 | 4591 | get_online_cpus(); |
1da177e4 LT |
4592 | read_lock(&tasklist_lock); |
4593 | ||
4594 | retval = -ESRCH; | |
4595 | p = find_process_by_pid(pid); | |
4596 | if (!p) | |
4597 | goto out_unlock; | |
4598 | ||
e7834f8f DQ |
4599 | retval = security_task_getscheduler(p); |
4600 | if (retval) | |
4601 | goto out_unlock; | |
4602 | ||
2f7016d9 | 4603 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
4604 | |
4605 | out_unlock: | |
4606 | read_unlock(&tasklist_lock); | |
95402b38 | 4607 | put_online_cpus(); |
1da177e4 | 4608 | |
9531b62f | 4609 | return retval; |
1da177e4 LT |
4610 | } |
4611 | ||
4612 | /** | |
4613 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4614 | * @pid: pid of the process | |
4615 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4616 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4617 | */ | |
4618 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
4619 | unsigned long __user *user_mask_ptr) | |
4620 | { | |
4621 | int ret; | |
4622 | cpumask_t mask; | |
4623 | ||
4624 | if (len < sizeof(cpumask_t)) | |
4625 | return -EINVAL; | |
4626 | ||
4627 | ret = sched_getaffinity(pid, &mask); | |
4628 | if (ret < 0) | |
4629 | return ret; | |
4630 | ||
4631 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
4632 | return -EFAULT; | |
4633 | ||
4634 | return sizeof(cpumask_t); | |
4635 | } | |
4636 | ||
4637 | /** | |
4638 | * sys_sched_yield - yield the current processor to other threads. | |
4639 | * | |
dd41f596 IM |
4640 | * This function yields the current CPU to other tasks. If there are no |
4641 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
4642 | */ |
4643 | asmlinkage long sys_sched_yield(void) | |
4644 | { | |
70b97a7f | 4645 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4646 | |
2d72376b | 4647 | schedstat_inc(rq, yld_count); |
4530d7ab | 4648 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4649 | |
4650 | /* | |
4651 | * Since we are going to call schedule() anyway, there's | |
4652 | * no need to preempt or enable interrupts: | |
4653 | */ | |
4654 | __release(rq->lock); | |
8a25d5de | 4655 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4656 | _raw_spin_unlock(&rq->lock); |
4657 | preempt_enable_no_resched(); | |
4658 | ||
4659 | schedule(); | |
4660 | ||
4661 | return 0; | |
4662 | } | |
4663 | ||
e7b38404 | 4664 | static void __cond_resched(void) |
1da177e4 | 4665 | { |
8e0a43d8 IM |
4666 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
4667 | __might_sleep(__FILE__, __LINE__); | |
4668 | #endif | |
5bbcfd90 IM |
4669 | /* |
4670 | * The BKS might be reacquired before we have dropped | |
4671 | * PREEMPT_ACTIVE, which could trigger a second | |
4672 | * cond_resched() call. | |
4673 | */ | |
1da177e4 LT |
4674 | do { |
4675 | add_preempt_count(PREEMPT_ACTIVE); | |
4676 | schedule(); | |
4677 | sub_preempt_count(PREEMPT_ACTIVE); | |
4678 | } while (need_resched()); | |
4679 | } | |
4680 | ||
4681 | int __sched cond_resched(void) | |
4682 | { | |
9414232f IM |
4683 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
4684 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
4685 | __cond_resched(); |
4686 | return 1; | |
4687 | } | |
4688 | return 0; | |
4689 | } | |
1da177e4 LT |
4690 | EXPORT_SYMBOL(cond_resched); |
4691 | ||
4692 | /* | |
4693 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
4694 | * call schedule, and on return reacquire the lock. | |
4695 | * | |
41a2d6cf | 4696 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4697 | * operations here to prevent schedule() from being called twice (once via |
4698 | * spin_unlock(), once by hand). | |
4699 | */ | |
95cdf3b7 | 4700 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4701 | { |
6df3cecb JK |
4702 | int ret = 0; |
4703 | ||
1da177e4 LT |
4704 | if (need_lockbreak(lock)) { |
4705 | spin_unlock(lock); | |
4706 | cpu_relax(); | |
6df3cecb | 4707 | ret = 1; |
1da177e4 LT |
4708 | spin_lock(lock); |
4709 | } | |
9414232f | 4710 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
8a25d5de | 4711 | spin_release(&lock->dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4712 | _raw_spin_unlock(lock); |
4713 | preempt_enable_no_resched(); | |
4714 | __cond_resched(); | |
6df3cecb | 4715 | ret = 1; |
1da177e4 | 4716 | spin_lock(lock); |
1da177e4 | 4717 | } |
6df3cecb | 4718 | return ret; |
1da177e4 | 4719 | } |
1da177e4 LT |
4720 | EXPORT_SYMBOL(cond_resched_lock); |
4721 | ||
4722 | int __sched cond_resched_softirq(void) | |
4723 | { | |
4724 | BUG_ON(!in_softirq()); | |
4725 | ||
9414232f | 4726 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 4727 | local_bh_enable(); |
1da177e4 LT |
4728 | __cond_resched(); |
4729 | local_bh_disable(); | |
4730 | return 1; | |
4731 | } | |
4732 | return 0; | |
4733 | } | |
1da177e4 LT |
4734 | EXPORT_SYMBOL(cond_resched_softirq); |
4735 | ||
1da177e4 LT |
4736 | /** |
4737 | * yield - yield the current processor to other threads. | |
4738 | * | |
72fd4a35 | 4739 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4740 | * thread runnable and calls sys_sched_yield(). |
4741 | */ | |
4742 | void __sched yield(void) | |
4743 | { | |
4744 | set_current_state(TASK_RUNNING); | |
4745 | sys_sched_yield(); | |
4746 | } | |
1da177e4 LT |
4747 | EXPORT_SYMBOL(yield); |
4748 | ||
4749 | /* | |
41a2d6cf | 4750 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
4751 | * that process accounting knows that this is a task in IO wait state. |
4752 | * | |
4753 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
4754 | * has set its backing_dev_info: the queue against which it should throttle) | |
4755 | */ | |
4756 | void __sched io_schedule(void) | |
4757 | { | |
70b97a7f | 4758 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 4759 | |
0ff92245 | 4760 | delayacct_blkio_start(); |
1da177e4 LT |
4761 | atomic_inc(&rq->nr_iowait); |
4762 | schedule(); | |
4763 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4764 | delayacct_blkio_end(); |
1da177e4 | 4765 | } |
1da177e4 LT |
4766 | EXPORT_SYMBOL(io_schedule); |
4767 | ||
4768 | long __sched io_schedule_timeout(long timeout) | |
4769 | { | |
70b97a7f | 4770 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
4771 | long ret; |
4772 | ||
0ff92245 | 4773 | delayacct_blkio_start(); |
1da177e4 LT |
4774 | atomic_inc(&rq->nr_iowait); |
4775 | ret = schedule_timeout(timeout); | |
4776 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4777 | delayacct_blkio_end(); |
1da177e4 LT |
4778 | return ret; |
4779 | } | |
4780 | ||
4781 | /** | |
4782 | * sys_sched_get_priority_max - return maximum RT priority. | |
4783 | * @policy: scheduling class. | |
4784 | * | |
4785 | * this syscall returns the maximum rt_priority that can be used | |
4786 | * by a given scheduling class. | |
4787 | */ | |
4788 | asmlinkage long sys_sched_get_priority_max(int policy) | |
4789 | { | |
4790 | int ret = -EINVAL; | |
4791 | ||
4792 | switch (policy) { | |
4793 | case SCHED_FIFO: | |
4794 | case SCHED_RR: | |
4795 | ret = MAX_USER_RT_PRIO-1; | |
4796 | break; | |
4797 | case SCHED_NORMAL: | |
b0a9499c | 4798 | case SCHED_BATCH: |
dd41f596 | 4799 | case SCHED_IDLE: |
1da177e4 LT |
4800 | ret = 0; |
4801 | break; | |
4802 | } | |
4803 | return ret; | |
4804 | } | |
4805 | ||
4806 | /** | |
4807 | * sys_sched_get_priority_min - return minimum RT priority. | |
4808 | * @policy: scheduling class. | |
4809 | * | |
4810 | * this syscall returns the minimum rt_priority that can be used | |
4811 | * by a given scheduling class. | |
4812 | */ | |
4813 | asmlinkage long sys_sched_get_priority_min(int policy) | |
4814 | { | |
4815 | int ret = -EINVAL; | |
4816 | ||
4817 | switch (policy) { | |
4818 | case SCHED_FIFO: | |
4819 | case SCHED_RR: | |
4820 | ret = 1; | |
4821 | break; | |
4822 | case SCHED_NORMAL: | |
b0a9499c | 4823 | case SCHED_BATCH: |
dd41f596 | 4824 | case SCHED_IDLE: |
1da177e4 LT |
4825 | ret = 0; |
4826 | } | |
4827 | return ret; | |
4828 | } | |
4829 | ||
4830 | /** | |
4831 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4832 | * @pid: pid of the process. | |
4833 | * @interval: userspace pointer to the timeslice value. | |
4834 | * | |
4835 | * this syscall writes the default timeslice value of a given process | |
4836 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4837 | */ | |
4838 | asmlinkage | |
4839 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
4840 | { | |
36c8b586 | 4841 | struct task_struct *p; |
a4ec24b4 | 4842 | unsigned int time_slice; |
3a5c359a | 4843 | int retval; |
1da177e4 | 4844 | struct timespec t; |
1da177e4 LT |
4845 | |
4846 | if (pid < 0) | |
3a5c359a | 4847 | return -EINVAL; |
1da177e4 LT |
4848 | |
4849 | retval = -ESRCH; | |
4850 | read_lock(&tasklist_lock); | |
4851 | p = find_process_by_pid(pid); | |
4852 | if (!p) | |
4853 | goto out_unlock; | |
4854 | ||
4855 | retval = security_task_getscheduler(p); | |
4856 | if (retval) | |
4857 | goto out_unlock; | |
4858 | ||
77034937 IM |
4859 | /* |
4860 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
4861 | * tasks that are on an otherwise idle runqueue: | |
4862 | */ | |
4863 | time_slice = 0; | |
4864 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 4865 | time_slice = DEF_TIMESLICE; |
77034937 | 4866 | } else { |
a4ec24b4 DA |
4867 | struct sched_entity *se = &p->se; |
4868 | unsigned long flags; | |
4869 | struct rq *rq; | |
4870 | ||
4871 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
4872 | if (rq->cfs.load.weight) |
4873 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
4874 | task_rq_unlock(rq, &flags); |
4875 | } | |
1da177e4 | 4876 | read_unlock(&tasklist_lock); |
a4ec24b4 | 4877 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4878 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4879 | return retval; |
3a5c359a | 4880 | |
1da177e4 LT |
4881 | out_unlock: |
4882 | read_unlock(&tasklist_lock); | |
4883 | return retval; | |
4884 | } | |
4885 | ||
2ed6e34f | 4886 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 | 4887 | |
82a1fcb9 | 4888 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4889 | { |
1da177e4 | 4890 | unsigned long free = 0; |
36c8b586 | 4891 | unsigned state; |
1da177e4 | 4892 | |
1da177e4 | 4893 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 4894 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 4895 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4896 | #if BITS_PER_LONG == 32 |
1da177e4 | 4897 | if (state == TASK_RUNNING) |
cc4ea795 | 4898 | printk(KERN_CONT " running "); |
1da177e4 | 4899 | else |
cc4ea795 | 4900 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4901 | #else |
4902 | if (state == TASK_RUNNING) | |
cc4ea795 | 4903 | printk(KERN_CONT " running task "); |
1da177e4 | 4904 | else |
cc4ea795 | 4905 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4906 | #endif |
4907 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
4908 | { | |
10ebffde | 4909 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
4910 | while (!*n) |
4911 | n++; | |
10ebffde | 4912 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
4913 | } |
4914 | #endif | |
ba25f9dc | 4915 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 4916 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 LT |
4917 | |
4918 | if (state != TASK_RUNNING) | |
4919 | show_stack(p, NULL); | |
4920 | } | |
4921 | ||
e59e2ae2 | 4922 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4923 | { |
36c8b586 | 4924 | struct task_struct *g, *p; |
1da177e4 | 4925 | |
4bd77321 IM |
4926 | #if BITS_PER_LONG == 32 |
4927 | printk(KERN_INFO | |
4928 | " task PC stack pid father\n"); | |
1da177e4 | 4929 | #else |
4bd77321 IM |
4930 | printk(KERN_INFO |
4931 | " task PC stack pid father\n"); | |
1da177e4 LT |
4932 | #endif |
4933 | read_lock(&tasklist_lock); | |
4934 | do_each_thread(g, p) { | |
4935 | /* | |
4936 | * reset the NMI-timeout, listing all files on a slow | |
4937 | * console might take alot of time: | |
4938 | */ | |
4939 | touch_nmi_watchdog(); | |
39bc89fd | 4940 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4941 | sched_show_task(p); |
1da177e4 LT |
4942 | } while_each_thread(g, p); |
4943 | ||
04c9167f JF |
4944 | touch_all_softlockup_watchdogs(); |
4945 | ||
dd41f596 IM |
4946 | #ifdef CONFIG_SCHED_DEBUG |
4947 | sysrq_sched_debug_show(); | |
4948 | #endif | |
1da177e4 | 4949 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
4950 | /* |
4951 | * Only show locks if all tasks are dumped: | |
4952 | */ | |
4953 | if (state_filter == -1) | |
4954 | debug_show_all_locks(); | |
1da177e4 LT |
4955 | } |
4956 | ||
1df21055 IM |
4957 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4958 | { | |
dd41f596 | 4959 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4960 | } |
4961 | ||
f340c0d1 IM |
4962 | /** |
4963 | * init_idle - set up an idle thread for a given CPU | |
4964 | * @idle: task in question | |
4965 | * @cpu: cpu the idle task belongs to | |
4966 | * | |
4967 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4968 | * flag, to make booting more robust. | |
4969 | */ | |
5c1e1767 | 4970 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4971 | { |
70b97a7f | 4972 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4973 | unsigned long flags; |
4974 | ||
dd41f596 IM |
4975 | __sched_fork(idle); |
4976 | idle->se.exec_start = sched_clock(); | |
4977 | ||
b29739f9 | 4978 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 4979 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 4980 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
4981 | |
4982 | spin_lock_irqsave(&rq->lock, flags); | |
4983 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
4984 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4985 | idle->oncpu = 1; | |
4986 | #endif | |
1da177e4 LT |
4987 | spin_unlock_irqrestore(&rq->lock, flags); |
4988 | ||
4989 | /* Set the preempt count _outside_ the spinlocks! */ | |
4990 | #if defined(CONFIG_PREEMPT) && !defined(CONFIG_PREEMPT_BKL) | |
a1261f54 | 4991 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); |
1da177e4 | 4992 | #else |
a1261f54 | 4993 | task_thread_info(idle)->preempt_count = 0; |
1da177e4 | 4994 | #endif |
dd41f596 IM |
4995 | /* |
4996 | * The idle tasks have their own, simple scheduling class: | |
4997 | */ | |
4998 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
4999 | } |
5000 | ||
5001 | /* | |
5002 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5003 | * indicates which cpus entered this state. This is used | |
5004 | * in the rcu update to wait only for active cpus. For system | |
5005 | * which do not switch off the HZ timer nohz_cpu_mask should | |
5006 | * always be CPU_MASK_NONE. | |
5007 | */ | |
5008 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
5009 | ||
19978ca6 IM |
5010 | /* |
5011 | * Increase the granularity value when there are more CPUs, | |
5012 | * because with more CPUs the 'effective latency' as visible | |
5013 | * to users decreases. But the relationship is not linear, | |
5014 | * so pick a second-best guess by going with the log2 of the | |
5015 | * number of CPUs. | |
5016 | * | |
5017 | * This idea comes from the SD scheduler of Con Kolivas: | |
5018 | */ | |
5019 | static inline void sched_init_granularity(void) | |
5020 | { | |
5021 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
5022 | const unsigned long limit = 200000000; | |
5023 | ||
5024 | sysctl_sched_min_granularity *= factor; | |
5025 | if (sysctl_sched_min_granularity > limit) | |
5026 | sysctl_sched_min_granularity = limit; | |
5027 | ||
5028 | sysctl_sched_latency *= factor; | |
5029 | if (sysctl_sched_latency > limit) | |
5030 | sysctl_sched_latency = limit; | |
5031 | ||
5032 | sysctl_sched_wakeup_granularity *= factor; | |
5033 | sysctl_sched_batch_wakeup_granularity *= factor; | |
5034 | } | |
5035 | ||
1da177e4 LT |
5036 | #ifdef CONFIG_SMP |
5037 | /* | |
5038 | * This is how migration works: | |
5039 | * | |
70b97a7f | 5040 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5041 | * runqueue and wake up that CPU's migration thread. |
5042 | * 2) we down() the locked semaphore => thread blocks. | |
5043 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5044 | * thread off the CPU) | |
5045 | * 4) it gets the migration request and checks whether the migrated | |
5046 | * task is still in the wrong runqueue. | |
5047 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5048 | * it and puts it into the right queue. | |
5049 | * 6) migration thread up()s the semaphore. | |
5050 | * 7) we wake up and the migration is done. | |
5051 | */ | |
5052 | ||
5053 | /* | |
5054 | * Change a given task's CPU affinity. Migrate the thread to a | |
5055 | * proper CPU and schedule it away if the CPU it's executing on | |
5056 | * is removed from the allowed bitmask. | |
5057 | * | |
5058 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5059 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5060 | * call is not atomic; no spinlocks may be held. |
5061 | */ | |
36c8b586 | 5062 | int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask) |
1da177e4 | 5063 | { |
70b97a7f | 5064 | struct migration_req req; |
1da177e4 | 5065 | unsigned long flags; |
70b97a7f | 5066 | struct rq *rq; |
48f24c4d | 5067 | int ret = 0; |
1da177e4 LT |
5068 | |
5069 | rq = task_rq_lock(p, &flags); | |
5070 | if (!cpus_intersects(new_mask, cpu_online_map)) { | |
5071 | ret = -EINVAL; | |
5072 | goto out; | |
5073 | } | |
5074 | ||
73fe6aae GH |
5075 | if (p->sched_class->set_cpus_allowed) |
5076 | p->sched_class->set_cpus_allowed(p, &new_mask); | |
5077 | else { | |
0eab9146 | 5078 | p->cpus_allowed = new_mask; |
73fe6aae GH |
5079 | p->nr_cpus_allowed = cpus_weight(new_mask); |
5080 | } | |
5081 | ||
1da177e4 LT |
5082 | /* Can the task run on the task's current CPU? If so, we're done */ |
5083 | if (cpu_isset(task_cpu(p), new_mask)) | |
5084 | goto out; | |
5085 | ||
5086 | if (migrate_task(p, any_online_cpu(new_mask), &req)) { | |
5087 | /* Need help from migration thread: drop lock and wait. */ | |
5088 | task_rq_unlock(rq, &flags); | |
5089 | wake_up_process(rq->migration_thread); | |
5090 | wait_for_completion(&req.done); | |
5091 | tlb_migrate_finish(p->mm); | |
5092 | return 0; | |
5093 | } | |
5094 | out: | |
5095 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5096 | |
1da177e4 LT |
5097 | return ret; |
5098 | } | |
1da177e4 LT |
5099 | EXPORT_SYMBOL_GPL(set_cpus_allowed); |
5100 | ||
5101 | /* | |
41a2d6cf | 5102 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5103 | * this because either it can't run here any more (set_cpus_allowed() |
5104 | * away from this CPU, or CPU going down), or because we're | |
5105 | * attempting to rebalance this task on exec (sched_exec). | |
5106 | * | |
5107 | * So we race with normal scheduler movements, but that's OK, as long | |
5108 | * as the task is no longer on this CPU. | |
efc30814 KK |
5109 | * |
5110 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5111 | */ |
efc30814 | 5112 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5113 | { |
70b97a7f | 5114 | struct rq *rq_dest, *rq_src; |
dd41f596 | 5115 | int ret = 0, on_rq; |
1da177e4 LT |
5116 | |
5117 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 5118 | return ret; |
1da177e4 LT |
5119 | |
5120 | rq_src = cpu_rq(src_cpu); | |
5121 | rq_dest = cpu_rq(dest_cpu); | |
5122 | ||
5123 | double_rq_lock(rq_src, rq_dest); | |
5124 | /* Already moved. */ | |
5125 | if (task_cpu(p) != src_cpu) | |
5126 | goto out; | |
5127 | /* Affinity changed (again). */ | |
5128 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
5129 | goto out; | |
5130 | ||
dd41f596 | 5131 | on_rq = p->se.on_rq; |
6e82a3be | 5132 | if (on_rq) |
2e1cb74a | 5133 | deactivate_task(rq_src, p, 0); |
6e82a3be | 5134 | |
1da177e4 | 5135 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
5136 | if (on_rq) { |
5137 | activate_task(rq_dest, p, 0); | |
5138 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 5139 | } |
efc30814 | 5140 | ret = 1; |
1da177e4 LT |
5141 | out: |
5142 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 5143 | return ret; |
1da177e4 LT |
5144 | } |
5145 | ||
5146 | /* | |
5147 | * migration_thread - this is a highprio system thread that performs | |
5148 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5149 | * another runqueue. | |
5150 | */ | |
95cdf3b7 | 5151 | static int migration_thread(void *data) |
1da177e4 | 5152 | { |
1da177e4 | 5153 | int cpu = (long)data; |
70b97a7f | 5154 | struct rq *rq; |
1da177e4 LT |
5155 | |
5156 | rq = cpu_rq(cpu); | |
5157 | BUG_ON(rq->migration_thread != current); | |
5158 | ||
5159 | set_current_state(TASK_INTERRUPTIBLE); | |
5160 | while (!kthread_should_stop()) { | |
70b97a7f | 5161 | struct migration_req *req; |
1da177e4 | 5162 | struct list_head *head; |
1da177e4 | 5163 | |
1da177e4 LT |
5164 | spin_lock_irq(&rq->lock); |
5165 | ||
5166 | if (cpu_is_offline(cpu)) { | |
5167 | spin_unlock_irq(&rq->lock); | |
5168 | goto wait_to_die; | |
5169 | } | |
5170 | ||
5171 | if (rq->active_balance) { | |
5172 | active_load_balance(rq, cpu); | |
5173 | rq->active_balance = 0; | |
5174 | } | |
5175 | ||
5176 | head = &rq->migration_queue; | |
5177 | ||
5178 | if (list_empty(head)) { | |
5179 | spin_unlock_irq(&rq->lock); | |
5180 | schedule(); | |
5181 | set_current_state(TASK_INTERRUPTIBLE); | |
5182 | continue; | |
5183 | } | |
70b97a7f | 5184 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5185 | list_del_init(head->next); |
5186 | ||
674311d5 NP |
5187 | spin_unlock(&rq->lock); |
5188 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5189 | local_irq_enable(); | |
1da177e4 LT |
5190 | |
5191 | complete(&req->done); | |
5192 | } | |
5193 | __set_current_state(TASK_RUNNING); | |
5194 | return 0; | |
5195 | ||
5196 | wait_to_die: | |
5197 | /* Wait for kthread_stop */ | |
5198 | set_current_state(TASK_INTERRUPTIBLE); | |
5199 | while (!kthread_should_stop()) { | |
5200 | schedule(); | |
5201 | set_current_state(TASK_INTERRUPTIBLE); | |
5202 | } | |
5203 | __set_current_state(TASK_RUNNING); | |
5204 | return 0; | |
5205 | } | |
5206 | ||
5207 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5208 | |
5209 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5210 | { | |
5211 | int ret; | |
5212 | ||
5213 | local_irq_disable(); | |
5214 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5215 | local_irq_enable(); | |
5216 | return ret; | |
5217 | } | |
5218 | ||
054b9108 | 5219 | /* |
3a4fa0a2 | 5220 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
5221 | * NOTE: interrupts should be disabled by the caller |
5222 | */ | |
48f24c4d | 5223 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5224 | { |
efc30814 | 5225 | unsigned long flags; |
1da177e4 | 5226 | cpumask_t mask; |
70b97a7f IM |
5227 | struct rq *rq; |
5228 | int dest_cpu; | |
1da177e4 | 5229 | |
3a5c359a AK |
5230 | do { |
5231 | /* On same node? */ | |
5232 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
5233 | cpus_and(mask, mask, p->cpus_allowed); | |
5234 | dest_cpu = any_online_cpu(mask); | |
5235 | ||
5236 | /* On any allowed CPU? */ | |
5237 | if (dest_cpu == NR_CPUS) | |
5238 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
5239 | ||
5240 | /* No more Mr. Nice Guy. */ | |
5241 | if (dest_cpu == NR_CPUS) { | |
470fd646 CW |
5242 | cpumask_t cpus_allowed = cpuset_cpus_allowed_locked(p); |
5243 | /* | |
5244 | * Try to stay on the same cpuset, where the | |
5245 | * current cpuset may be a subset of all cpus. | |
5246 | * The cpuset_cpus_allowed_locked() variant of | |
41a2d6cf | 5247 | * cpuset_cpus_allowed() will not block. It must be |
470fd646 CW |
5248 | * called within calls to cpuset_lock/cpuset_unlock. |
5249 | */ | |
3a5c359a | 5250 | rq = task_rq_lock(p, &flags); |
470fd646 | 5251 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
5252 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5253 | task_rq_unlock(rq, &flags); | |
1da177e4 | 5254 | |
3a5c359a AK |
5255 | /* |
5256 | * Don't tell them about moving exiting tasks or | |
5257 | * kernel threads (both mm NULL), since they never | |
5258 | * leave kernel. | |
5259 | */ | |
41a2d6cf | 5260 | if (p->mm && printk_ratelimit()) { |
3a5c359a AK |
5261 | printk(KERN_INFO "process %d (%s) no " |
5262 | "longer affine to cpu%d\n", | |
41a2d6cf IM |
5263 | task_pid_nr(p), p->comm, dead_cpu); |
5264 | } | |
3a5c359a | 5265 | } |
f7b4cddc | 5266 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
5267 | } |
5268 | ||
5269 | /* | |
5270 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5271 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5272 | * for performance reasons the counter is not stricly tracking tasks to | |
5273 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5274 | * to keep the global sum constant after CPU-down: | |
5275 | */ | |
70b97a7f | 5276 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5277 | { |
70b97a7f | 5278 | struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL)); |
1da177e4 LT |
5279 | unsigned long flags; |
5280 | ||
5281 | local_irq_save(flags); | |
5282 | double_rq_lock(rq_src, rq_dest); | |
5283 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5284 | rq_src->nr_uninterruptible = 0; | |
5285 | double_rq_unlock(rq_src, rq_dest); | |
5286 | local_irq_restore(flags); | |
5287 | } | |
5288 | ||
5289 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5290 | static void migrate_live_tasks(int src_cpu) | |
5291 | { | |
48f24c4d | 5292 | struct task_struct *p, *t; |
1da177e4 | 5293 | |
f7b4cddc | 5294 | read_lock(&tasklist_lock); |
1da177e4 | 5295 | |
48f24c4d IM |
5296 | do_each_thread(t, p) { |
5297 | if (p == current) | |
1da177e4 LT |
5298 | continue; |
5299 | ||
48f24c4d IM |
5300 | if (task_cpu(p) == src_cpu) |
5301 | move_task_off_dead_cpu(src_cpu, p); | |
5302 | } while_each_thread(t, p); | |
1da177e4 | 5303 | |
f7b4cddc | 5304 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5305 | } |
5306 | ||
dd41f596 IM |
5307 | /* |
5308 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5309 | * It does so by boosting its priority to highest possible. |
5310 | * Used by CPU offline code. | |
1da177e4 LT |
5311 | */ |
5312 | void sched_idle_next(void) | |
5313 | { | |
48f24c4d | 5314 | int this_cpu = smp_processor_id(); |
70b97a7f | 5315 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5316 | struct task_struct *p = rq->idle; |
5317 | unsigned long flags; | |
5318 | ||
5319 | /* cpu has to be offline */ | |
48f24c4d | 5320 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5321 | |
48f24c4d IM |
5322 | /* |
5323 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5324 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5325 | */ |
5326 | spin_lock_irqsave(&rq->lock, flags); | |
5327 | ||
dd41f596 | 5328 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 5329 | |
94bc9a7b DA |
5330 | update_rq_clock(rq); |
5331 | activate_task(rq, p, 0); | |
1da177e4 LT |
5332 | |
5333 | spin_unlock_irqrestore(&rq->lock, flags); | |
5334 | } | |
5335 | ||
48f24c4d IM |
5336 | /* |
5337 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5338 | * offline. |
5339 | */ | |
5340 | void idle_task_exit(void) | |
5341 | { | |
5342 | struct mm_struct *mm = current->active_mm; | |
5343 | ||
5344 | BUG_ON(cpu_online(smp_processor_id())); | |
5345 | ||
5346 | if (mm != &init_mm) | |
5347 | switch_mm(mm, &init_mm, current); | |
5348 | mmdrop(mm); | |
5349 | } | |
5350 | ||
054b9108 | 5351 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5352 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5353 | { |
70b97a7f | 5354 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5355 | |
5356 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5357 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5358 | |
5359 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5360 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5361 | |
48f24c4d | 5362 | get_task_struct(p); |
1da177e4 LT |
5363 | |
5364 | /* | |
5365 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 5366 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
5367 | * fine. |
5368 | */ | |
f7b4cddc | 5369 | spin_unlock_irq(&rq->lock); |
48f24c4d | 5370 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 5371 | spin_lock_irq(&rq->lock); |
1da177e4 | 5372 | |
48f24c4d | 5373 | put_task_struct(p); |
1da177e4 LT |
5374 | } |
5375 | ||
5376 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5377 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5378 | { | |
70b97a7f | 5379 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5380 | struct task_struct *next; |
48f24c4d | 5381 | |
dd41f596 IM |
5382 | for ( ; ; ) { |
5383 | if (!rq->nr_running) | |
5384 | break; | |
a8e504d2 | 5385 | update_rq_clock(rq); |
ff95f3df | 5386 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
5387 | if (!next) |
5388 | break; | |
5389 | migrate_dead(dead_cpu, next); | |
e692ab53 | 5390 | |
1da177e4 LT |
5391 | } |
5392 | } | |
5393 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5394 | ||
e692ab53 NP |
5395 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5396 | ||
5397 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5398 | { |
5399 | .procname = "sched_domain", | |
c57baf1e | 5400 | .mode = 0555, |
e0361851 | 5401 | }, |
38605cae | 5402 | {0, }, |
e692ab53 NP |
5403 | }; |
5404 | ||
5405 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 5406 | { |
c57baf1e | 5407 | .ctl_name = CTL_KERN, |
e0361851 | 5408 | .procname = "kernel", |
c57baf1e | 5409 | .mode = 0555, |
e0361851 AD |
5410 | .child = sd_ctl_dir, |
5411 | }, | |
38605cae | 5412 | {0, }, |
e692ab53 NP |
5413 | }; |
5414 | ||
5415 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5416 | { | |
5417 | struct ctl_table *entry = | |
5cf9f062 | 5418 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5419 | |
e692ab53 NP |
5420 | return entry; |
5421 | } | |
5422 | ||
6382bc90 MM |
5423 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5424 | { | |
cd790076 | 5425 | struct ctl_table *entry; |
6382bc90 | 5426 | |
cd790076 MM |
5427 | /* |
5428 | * In the intermediate directories, both the child directory and | |
5429 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5430 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5431 | * static strings and all have proc handlers. |
5432 | */ | |
5433 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5434 | if (entry->child) |
5435 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5436 | if (entry->proc_handler == NULL) |
5437 | kfree(entry->procname); | |
5438 | } | |
6382bc90 MM |
5439 | |
5440 | kfree(*tablep); | |
5441 | *tablep = NULL; | |
5442 | } | |
5443 | ||
e692ab53 | 5444 | static void |
e0361851 | 5445 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5446 | const char *procname, void *data, int maxlen, |
5447 | mode_t mode, proc_handler *proc_handler) | |
5448 | { | |
e692ab53 NP |
5449 | entry->procname = procname; |
5450 | entry->data = data; | |
5451 | entry->maxlen = maxlen; | |
5452 | entry->mode = mode; | |
5453 | entry->proc_handler = proc_handler; | |
5454 | } | |
5455 | ||
5456 | static struct ctl_table * | |
5457 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5458 | { | |
ace8b3d6 | 5459 | struct ctl_table *table = sd_alloc_ctl_entry(12); |
e692ab53 | 5460 | |
ad1cdc1d MM |
5461 | if (table == NULL) |
5462 | return NULL; | |
5463 | ||
e0361851 | 5464 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5465 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5466 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5467 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5468 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5469 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5470 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5471 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5472 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5473 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5474 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5475 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5476 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5477 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5478 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5479 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5480 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5481 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5482 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5483 | &sd->cache_nice_tries, |
5484 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5485 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5486 | sizeof(int), 0644, proc_dointvec_minmax); |
6323469f | 5487 | /* &table[11] is terminator */ |
e692ab53 NP |
5488 | |
5489 | return table; | |
5490 | } | |
5491 | ||
9a4e7159 | 5492 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5493 | { |
5494 | struct ctl_table *entry, *table; | |
5495 | struct sched_domain *sd; | |
5496 | int domain_num = 0, i; | |
5497 | char buf[32]; | |
5498 | ||
5499 | for_each_domain(cpu, sd) | |
5500 | domain_num++; | |
5501 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5502 | if (table == NULL) |
5503 | return NULL; | |
e692ab53 NP |
5504 | |
5505 | i = 0; | |
5506 | for_each_domain(cpu, sd) { | |
5507 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5508 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5509 | entry->mode = 0555; |
e692ab53 NP |
5510 | entry->child = sd_alloc_ctl_domain_table(sd); |
5511 | entry++; | |
5512 | i++; | |
5513 | } | |
5514 | return table; | |
5515 | } | |
5516 | ||
5517 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5518 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
5519 | { |
5520 | int i, cpu_num = num_online_cpus(); | |
5521 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
5522 | char buf[32]; | |
5523 | ||
7378547f MM |
5524 | WARN_ON(sd_ctl_dir[0].child); |
5525 | sd_ctl_dir[0].child = entry; | |
5526 | ||
ad1cdc1d MM |
5527 | if (entry == NULL) |
5528 | return; | |
5529 | ||
97b6ea7b | 5530 | for_each_online_cpu(i) { |
e692ab53 | 5531 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5532 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5533 | entry->mode = 0555; |
e692ab53 | 5534 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5535 | entry++; |
e692ab53 | 5536 | } |
7378547f MM |
5537 | |
5538 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5539 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5540 | } | |
6382bc90 | 5541 | |
7378547f | 5542 | /* may be called multiple times per register */ |
6382bc90 MM |
5543 | static void unregister_sched_domain_sysctl(void) |
5544 | { | |
7378547f MM |
5545 | if (sd_sysctl_header) |
5546 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5547 | sd_sysctl_header = NULL; |
7378547f MM |
5548 | if (sd_ctl_dir[0].child) |
5549 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5550 | } |
e692ab53 | 5551 | #else |
6382bc90 MM |
5552 | static void register_sched_domain_sysctl(void) |
5553 | { | |
5554 | } | |
5555 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5556 | { |
5557 | } | |
5558 | #endif | |
5559 | ||
1da177e4 LT |
5560 | /* |
5561 | * migration_call - callback that gets triggered when a CPU is added. | |
5562 | * Here we can start up the necessary migration thread for the new CPU. | |
5563 | */ | |
48f24c4d IM |
5564 | static int __cpuinit |
5565 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5566 | { |
1da177e4 | 5567 | struct task_struct *p; |
48f24c4d | 5568 | int cpu = (long)hcpu; |
1da177e4 | 5569 | unsigned long flags; |
70b97a7f | 5570 | struct rq *rq; |
1da177e4 LT |
5571 | |
5572 | switch (action) { | |
5be9361c | 5573 | |
1da177e4 | 5574 | case CPU_UP_PREPARE: |
8bb78442 | 5575 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 5576 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
5577 | if (IS_ERR(p)) |
5578 | return NOTIFY_BAD; | |
1da177e4 LT |
5579 | kthread_bind(p, cpu); |
5580 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5581 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 5582 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
5583 | task_rq_unlock(rq, &flags); |
5584 | cpu_rq(cpu)->migration_thread = p; | |
5585 | break; | |
48f24c4d | 5586 | |
1da177e4 | 5587 | case CPU_ONLINE: |
8bb78442 | 5588 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 5589 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 5590 | wake_up_process(cpu_rq(cpu)->migration_thread); |
57d885fe GH |
5591 | |
5592 | /* Update our root-domain */ | |
5593 | rq = cpu_rq(cpu); | |
5594 | spin_lock_irqsave(&rq->lock, flags); | |
5595 | if (rq->rd) { | |
5596 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
5597 | cpu_set(cpu, rq->rd->online); | |
5598 | } | |
5599 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 5600 | break; |
48f24c4d | 5601 | |
1da177e4 LT |
5602 | #ifdef CONFIG_HOTPLUG_CPU |
5603 | case CPU_UP_CANCELED: | |
8bb78442 | 5604 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5605 | if (!cpu_rq(cpu)->migration_thread) |
5606 | break; | |
41a2d6cf | 5607 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
5608 | kthread_bind(cpu_rq(cpu)->migration_thread, |
5609 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
5610 | kthread_stop(cpu_rq(cpu)->migration_thread); |
5611 | cpu_rq(cpu)->migration_thread = NULL; | |
5612 | break; | |
48f24c4d | 5613 | |
1da177e4 | 5614 | case CPU_DEAD: |
8bb78442 | 5615 | case CPU_DEAD_FROZEN: |
470fd646 | 5616 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
5617 | migrate_live_tasks(cpu); |
5618 | rq = cpu_rq(cpu); | |
5619 | kthread_stop(rq->migration_thread); | |
5620 | rq->migration_thread = NULL; | |
5621 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 5622 | spin_lock_irq(&rq->lock); |
a8e504d2 | 5623 | update_rq_clock(rq); |
2e1cb74a | 5624 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 5625 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
5626 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5627 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 5628 | migrate_dead_tasks(cpu); |
d2da272a | 5629 | spin_unlock_irq(&rq->lock); |
470fd646 | 5630 | cpuset_unlock(); |
1da177e4 LT |
5631 | migrate_nr_uninterruptible(rq); |
5632 | BUG_ON(rq->nr_running != 0); | |
5633 | ||
41a2d6cf IM |
5634 | /* |
5635 | * No need to migrate the tasks: it was best-effort if | |
5636 | * they didn't take sched_hotcpu_mutex. Just wake up | |
5637 | * the requestors. | |
5638 | */ | |
1da177e4 LT |
5639 | spin_lock_irq(&rq->lock); |
5640 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
5641 | struct migration_req *req; |
5642 | ||
1da177e4 | 5643 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5644 | struct migration_req, list); |
1da177e4 LT |
5645 | list_del_init(&req->list); |
5646 | complete(&req->done); | |
5647 | } | |
5648 | spin_unlock_irq(&rq->lock); | |
5649 | break; | |
57d885fe GH |
5650 | |
5651 | case CPU_DOWN_PREPARE: | |
5652 | /* Update our root-domain */ | |
5653 | rq = cpu_rq(cpu); | |
5654 | spin_lock_irqsave(&rq->lock, flags); | |
5655 | if (rq->rd) { | |
5656 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
5657 | cpu_clear(cpu, rq->rd->online); | |
5658 | } | |
5659 | spin_unlock_irqrestore(&rq->lock, flags); | |
5660 | break; | |
1da177e4 LT |
5661 | #endif |
5662 | } | |
5663 | return NOTIFY_OK; | |
5664 | } | |
5665 | ||
5666 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
5667 | * happens before everything else. | |
5668 | */ | |
26c2143b | 5669 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5670 | .notifier_call = migration_call, |
5671 | .priority = 10 | |
5672 | }; | |
5673 | ||
e6fe6649 | 5674 | void __init migration_init(void) |
1da177e4 LT |
5675 | { |
5676 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5677 | int err; |
48f24c4d IM |
5678 | |
5679 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5680 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5681 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5682 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5683 | register_cpu_notifier(&migration_notifier); | |
1da177e4 LT |
5684 | } |
5685 | #endif | |
5686 | ||
5687 | #ifdef CONFIG_SMP | |
476f3534 CL |
5688 | |
5689 | /* Number of possible processor ids */ | |
5690 | int nr_cpu_ids __read_mostly = NR_CPUS; | |
5691 | EXPORT_SYMBOL(nr_cpu_ids); | |
5692 | ||
3e9830dc | 5693 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff IM |
5694 | |
5695 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level) | |
1da177e4 | 5696 | { |
4dcf6aff IM |
5697 | struct sched_group *group = sd->groups; |
5698 | cpumask_t groupmask; | |
5699 | char str[NR_CPUS]; | |
1da177e4 | 5700 | |
4dcf6aff IM |
5701 | cpumask_scnprintf(str, NR_CPUS, sd->span); |
5702 | cpus_clear(groupmask); | |
5703 | ||
5704 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5705 | ||
5706 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
5707 | printk("does not load-balance\n"); | |
5708 | if (sd->parent) | |
5709 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
5710 | " has parent"); | |
5711 | return -1; | |
41c7ce9a NP |
5712 | } |
5713 | ||
4dcf6aff IM |
5714 | printk(KERN_CONT "span %s\n", str); |
5715 | ||
5716 | if (!cpu_isset(cpu, sd->span)) { | |
5717 | printk(KERN_ERR "ERROR: domain->span does not contain " | |
5718 | "CPU%d\n", cpu); | |
5719 | } | |
5720 | if (!cpu_isset(cpu, group->cpumask)) { | |
5721 | printk(KERN_ERR "ERROR: domain->groups does not contain" | |
5722 | " CPU%d\n", cpu); | |
5723 | } | |
1da177e4 | 5724 | |
4dcf6aff | 5725 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5726 | do { |
4dcf6aff IM |
5727 | if (!group) { |
5728 | printk("\n"); | |
5729 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5730 | break; |
5731 | } | |
5732 | ||
4dcf6aff IM |
5733 | if (!group->__cpu_power) { |
5734 | printk(KERN_CONT "\n"); | |
5735 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5736 | "set\n"); | |
5737 | break; | |
5738 | } | |
1da177e4 | 5739 | |
4dcf6aff IM |
5740 | if (!cpus_weight(group->cpumask)) { |
5741 | printk(KERN_CONT "\n"); | |
5742 | printk(KERN_ERR "ERROR: empty group\n"); | |
5743 | break; | |
5744 | } | |
1da177e4 | 5745 | |
4dcf6aff IM |
5746 | if (cpus_intersects(groupmask, group->cpumask)) { |
5747 | printk(KERN_CONT "\n"); | |
5748 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
5749 | break; | |
5750 | } | |
1da177e4 | 5751 | |
4dcf6aff | 5752 | cpus_or(groupmask, groupmask, group->cpumask); |
1da177e4 | 5753 | |
4dcf6aff IM |
5754 | cpumask_scnprintf(str, NR_CPUS, group->cpumask); |
5755 | printk(KERN_CONT " %s", str); | |
1da177e4 | 5756 | |
4dcf6aff IM |
5757 | group = group->next; |
5758 | } while (group != sd->groups); | |
5759 | printk(KERN_CONT "\n"); | |
1da177e4 | 5760 | |
4dcf6aff IM |
5761 | if (!cpus_equal(sd->span, groupmask)) |
5762 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); | |
1da177e4 | 5763 | |
4dcf6aff IM |
5764 | if (sd->parent && !cpus_subset(groupmask, sd->parent->span)) |
5765 | printk(KERN_ERR "ERROR: parent span is not a superset " | |
5766 | "of domain->span\n"); | |
5767 | return 0; | |
5768 | } | |
1da177e4 | 5769 | |
4dcf6aff IM |
5770 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5771 | { | |
5772 | int level = 0; | |
1da177e4 | 5773 | |
4dcf6aff IM |
5774 | if (!sd) { |
5775 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5776 | return; | |
5777 | } | |
1da177e4 | 5778 | |
4dcf6aff IM |
5779 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5780 | ||
5781 | for (;;) { | |
5782 | if (sched_domain_debug_one(sd, cpu, level)) | |
5783 | break; | |
1da177e4 LT |
5784 | level++; |
5785 | sd = sd->parent; | |
33859f7f | 5786 | if (!sd) |
4dcf6aff IM |
5787 | break; |
5788 | } | |
1da177e4 LT |
5789 | } |
5790 | #else | |
48f24c4d | 5791 | # define sched_domain_debug(sd, cpu) do { } while (0) |
1da177e4 LT |
5792 | #endif |
5793 | ||
1a20ff27 | 5794 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
5795 | { |
5796 | if (cpus_weight(sd->span) == 1) | |
5797 | return 1; | |
5798 | ||
5799 | /* Following flags need at least 2 groups */ | |
5800 | if (sd->flags & (SD_LOAD_BALANCE | | |
5801 | SD_BALANCE_NEWIDLE | | |
5802 | SD_BALANCE_FORK | | |
89c4710e SS |
5803 | SD_BALANCE_EXEC | |
5804 | SD_SHARE_CPUPOWER | | |
5805 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5806 | if (sd->groups != sd->groups->next) |
5807 | return 0; | |
5808 | } | |
5809 | ||
5810 | /* Following flags don't use groups */ | |
5811 | if (sd->flags & (SD_WAKE_IDLE | | |
5812 | SD_WAKE_AFFINE | | |
5813 | SD_WAKE_BALANCE)) | |
5814 | return 0; | |
5815 | ||
5816 | return 1; | |
5817 | } | |
5818 | ||
48f24c4d IM |
5819 | static int |
5820 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5821 | { |
5822 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5823 | ||
5824 | if (sd_degenerate(parent)) | |
5825 | return 1; | |
5826 | ||
5827 | if (!cpus_equal(sd->span, parent->span)) | |
5828 | return 0; | |
5829 | ||
5830 | /* Does parent contain flags not in child? */ | |
5831 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
5832 | if (cflags & SD_WAKE_AFFINE) | |
5833 | pflags &= ~SD_WAKE_BALANCE; | |
5834 | /* Flags needing groups don't count if only 1 group in parent */ | |
5835 | if (parent->groups == parent->groups->next) { | |
5836 | pflags &= ~(SD_LOAD_BALANCE | | |
5837 | SD_BALANCE_NEWIDLE | | |
5838 | SD_BALANCE_FORK | | |
89c4710e SS |
5839 | SD_BALANCE_EXEC | |
5840 | SD_SHARE_CPUPOWER | | |
5841 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
5842 | } |
5843 | if (~cflags & pflags) | |
5844 | return 0; | |
5845 | ||
5846 | return 1; | |
5847 | } | |
5848 | ||
57d885fe GH |
5849 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5850 | { | |
5851 | unsigned long flags; | |
5852 | const struct sched_class *class; | |
5853 | ||
5854 | spin_lock_irqsave(&rq->lock, flags); | |
5855 | ||
5856 | if (rq->rd) { | |
5857 | struct root_domain *old_rd = rq->rd; | |
5858 | ||
0eab9146 | 5859 | for (class = sched_class_highest; class; class = class->next) { |
57d885fe GH |
5860 | if (class->leave_domain) |
5861 | class->leave_domain(rq); | |
0eab9146 | 5862 | } |
57d885fe | 5863 | |
dc938520 GH |
5864 | cpu_clear(rq->cpu, old_rd->span); |
5865 | cpu_clear(rq->cpu, old_rd->online); | |
5866 | ||
57d885fe GH |
5867 | if (atomic_dec_and_test(&old_rd->refcount)) |
5868 | kfree(old_rd); | |
5869 | } | |
5870 | ||
5871 | atomic_inc(&rd->refcount); | |
5872 | rq->rd = rd; | |
5873 | ||
dc938520 GH |
5874 | cpu_set(rq->cpu, rd->span); |
5875 | if (cpu_isset(rq->cpu, cpu_online_map)) | |
5876 | cpu_set(rq->cpu, rd->online); | |
5877 | ||
0eab9146 | 5878 | for (class = sched_class_highest; class; class = class->next) { |
57d885fe GH |
5879 | if (class->join_domain) |
5880 | class->join_domain(rq); | |
0eab9146 | 5881 | } |
57d885fe GH |
5882 | |
5883 | spin_unlock_irqrestore(&rq->lock, flags); | |
5884 | } | |
5885 | ||
dc938520 | 5886 | static void init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5887 | { |
5888 | memset(rd, 0, sizeof(*rd)); | |
5889 | ||
dc938520 GH |
5890 | cpus_clear(rd->span); |
5891 | cpus_clear(rd->online); | |
57d885fe GH |
5892 | } |
5893 | ||
5894 | static void init_defrootdomain(void) | |
5895 | { | |
dc938520 | 5896 | init_rootdomain(&def_root_domain); |
57d885fe GH |
5897 | atomic_set(&def_root_domain.refcount, 1); |
5898 | } | |
5899 | ||
dc938520 | 5900 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5901 | { |
5902 | struct root_domain *rd; | |
5903 | ||
5904 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5905 | if (!rd) | |
5906 | return NULL; | |
5907 | ||
dc938520 | 5908 | init_rootdomain(rd); |
57d885fe GH |
5909 | |
5910 | return rd; | |
5911 | } | |
5912 | ||
1da177e4 | 5913 | /* |
0eab9146 | 5914 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5915 | * hold the hotplug lock. |
5916 | */ | |
0eab9146 IM |
5917 | static void |
5918 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5919 | { |
70b97a7f | 5920 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5921 | struct sched_domain *tmp; |
5922 | ||
5923 | /* Remove the sched domains which do not contribute to scheduling. */ | |
5924 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
5925 | struct sched_domain *parent = tmp->parent; | |
5926 | if (!parent) | |
5927 | break; | |
1a848870 | 5928 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5929 | tmp->parent = parent->parent; |
1a848870 SS |
5930 | if (parent->parent) |
5931 | parent->parent->child = tmp; | |
5932 | } | |
245af2c7 SS |
5933 | } |
5934 | ||
1a848870 | 5935 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 5936 | sd = sd->parent; |
1a848870 SS |
5937 | if (sd) |
5938 | sd->child = NULL; | |
5939 | } | |
1da177e4 LT |
5940 | |
5941 | sched_domain_debug(sd, cpu); | |
5942 | ||
57d885fe | 5943 | rq_attach_root(rq, rd); |
674311d5 | 5944 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
5945 | } |
5946 | ||
5947 | /* cpus with isolated domains */ | |
67af63a6 | 5948 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
5949 | |
5950 | /* Setup the mask of cpus configured for isolated domains */ | |
5951 | static int __init isolated_cpu_setup(char *str) | |
5952 | { | |
5953 | int ints[NR_CPUS], i; | |
5954 | ||
5955 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
5956 | cpus_clear(cpu_isolated_map); | |
5957 | for (i = 1; i <= ints[0]; i++) | |
5958 | if (ints[i] < NR_CPUS) | |
5959 | cpu_set(ints[i], cpu_isolated_map); | |
5960 | return 1; | |
5961 | } | |
5962 | ||
8927f494 | 5963 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
5964 | |
5965 | /* | |
6711cab4 SS |
5966 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
5967 | * to a function which identifies what group(along with sched group) a CPU | |
5968 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
5969 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
5970 | * |
5971 | * init_sched_build_groups will build a circular linked list of the groups | |
5972 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5973 | * and ->cpu_power to 0. | |
5974 | */ | |
a616058b | 5975 | static void |
6711cab4 SS |
5976 | init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map, |
5977 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, | |
5978 | struct sched_group **sg)) | |
1da177e4 LT |
5979 | { |
5980 | struct sched_group *first = NULL, *last = NULL; | |
5981 | cpumask_t covered = CPU_MASK_NONE; | |
5982 | int i; | |
5983 | ||
5984 | for_each_cpu_mask(i, span) { | |
6711cab4 SS |
5985 | struct sched_group *sg; |
5986 | int group = group_fn(i, cpu_map, &sg); | |
1da177e4 LT |
5987 | int j; |
5988 | ||
5989 | if (cpu_isset(i, covered)) | |
5990 | continue; | |
5991 | ||
5992 | sg->cpumask = CPU_MASK_NONE; | |
5517d86b | 5993 | sg->__cpu_power = 0; |
1da177e4 LT |
5994 | |
5995 | for_each_cpu_mask(j, span) { | |
6711cab4 | 5996 | if (group_fn(j, cpu_map, NULL) != group) |
1da177e4 LT |
5997 | continue; |
5998 | ||
5999 | cpu_set(j, covered); | |
6000 | cpu_set(j, sg->cpumask); | |
6001 | } | |
6002 | if (!first) | |
6003 | first = sg; | |
6004 | if (last) | |
6005 | last->next = sg; | |
6006 | last = sg; | |
6007 | } | |
6008 | last->next = first; | |
6009 | } | |
6010 | ||
9c1cfda2 | 6011 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6012 | |
9c1cfda2 | 6013 | #ifdef CONFIG_NUMA |
198e2f18 | 6014 | |
9c1cfda2 JH |
6015 | /** |
6016 | * find_next_best_node - find the next node to include in a sched_domain | |
6017 | * @node: node whose sched_domain we're building | |
6018 | * @used_nodes: nodes already in the sched_domain | |
6019 | * | |
41a2d6cf | 6020 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6021 | * finds the closest node not already in the @used_nodes map. |
6022 | * | |
6023 | * Should use nodemask_t. | |
6024 | */ | |
6025 | static int find_next_best_node(int node, unsigned long *used_nodes) | |
6026 | { | |
6027 | int i, n, val, min_val, best_node = 0; | |
6028 | ||
6029 | min_val = INT_MAX; | |
6030 | ||
6031 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6032 | /* Start at @node */ | |
6033 | n = (node + i) % MAX_NUMNODES; | |
6034 | ||
6035 | if (!nr_cpus_node(n)) | |
6036 | continue; | |
6037 | ||
6038 | /* Skip already used nodes */ | |
6039 | if (test_bit(n, used_nodes)) | |
6040 | continue; | |
6041 | ||
6042 | /* Simple min distance search */ | |
6043 | val = node_distance(node, n); | |
6044 | ||
6045 | if (val < min_val) { | |
6046 | min_val = val; | |
6047 | best_node = n; | |
6048 | } | |
6049 | } | |
6050 | ||
6051 | set_bit(best_node, used_nodes); | |
6052 | return best_node; | |
6053 | } | |
6054 | ||
6055 | /** | |
6056 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6057 | * @node: node whose cpumask we're constructing | |
6058 | * @size: number of nodes to include in this span | |
6059 | * | |
41a2d6cf | 6060 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6061 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6062 | * out optimally. | |
6063 | */ | |
6064 | static cpumask_t sched_domain_node_span(int node) | |
6065 | { | |
9c1cfda2 | 6066 | DECLARE_BITMAP(used_nodes, MAX_NUMNODES); |
48f24c4d IM |
6067 | cpumask_t span, nodemask; |
6068 | int i; | |
9c1cfda2 JH |
6069 | |
6070 | cpus_clear(span); | |
6071 | bitmap_zero(used_nodes, MAX_NUMNODES); | |
6072 | ||
6073 | nodemask = node_to_cpumask(node); | |
6074 | cpus_or(span, span, nodemask); | |
6075 | set_bit(node, used_nodes); | |
6076 | ||
6077 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
6078 | int next_node = find_next_best_node(node, used_nodes); | |
48f24c4d | 6079 | |
9c1cfda2 JH |
6080 | nodemask = node_to_cpumask(next_node); |
6081 | cpus_or(span, span, nodemask); | |
6082 | } | |
6083 | ||
6084 | return span; | |
6085 | } | |
6086 | #endif | |
6087 | ||
5c45bf27 | 6088 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6089 | |
9c1cfda2 | 6090 | /* |
48f24c4d | 6091 | * SMT sched-domains: |
9c1cfda2 | 6092 | */ |
1da177e4 LT |
6093 | #ifdef CONFIG_SCHED_SMT |
6094 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 6095 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 6096 | |
41a2d6cf IM |
6097 | static int |
6098 | cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1da177e4 | 6099 | { |
6711cab4 SS |
6100 | if (sg) |
6101 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
6102 | return cpu; |
6103 | } | |
6104 | #endif | |
6105 | ||
48f24c4d IM |
6106 | /* |
6107 | * multi-core sched-domains: | |
6108 | */ | |
1e9f28fa SS |
6109 | #ifdef CONFIG_SCHED_MC |
6110 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 6111 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
6112 | #endif |
6113 | ||
6114 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf IM |
6115 | static int |
6116 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1e9f28fa | 6117 | { |
6711cab4 | 6118 | int group; |
d5a7430d | 6119 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 6120 | cpus_and(mask, mask, *cpu_map); |
6711cab4 SS |
6121 | group = first_cpu(mask); |
6122 | if (sg) | |
6123 | *sg = &per_cpu(sched_group_core, group); | |
6124 | return group; | |
1e9f28fa SS |
6125 | } |
6126 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf IM |
6127 | static int |
6128 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1e9f28fa | 6129 | { |
6711cab4 SS |
6130 | if (sg) |
6131 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
6132 | return cpu; |
6133 | } | |
6134 | #endif | |
6135 | ||
1da177e4 | 6136 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 6137 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 6138 | |
41a2d6cf IM |
6139 | static int |
6140 | cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1da177e4 | 6141 | { |
6711cab4 | 6142 | int group; |
48f24c4d | 6143 | #ifdef CONFIG_SCHED_MC |
1e9f28fa | 6144 | cpumask_t mask = cpu_coregroup_map(cpu); |
a616058b | 6145 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 6146 | group = first_cpu(mask); |
1e9f28fa | 6147 | #elif defined(CONFIG_SCHED_SMT) |
d5a7430d | 6148 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 6149 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 6150 | group = first_cpu(mask); |
1da177e4 | 6151 | #else |
6711cab4 | 6152 | group = cpu; |
1da177e4 | 6153 | #endif |
6711cab4 SS |
6154 | if (sg) |
6155 | *sg = &per_cpu(sched_group_phys, group); | |
6156 | return group; | |
1da177e4 LT |
6157 | } |
6158 | ||
6159 | #ifdef CONFIG_NUMA | |
1da177e4 | 6160 | /* |
9c1cfda2 JH |
6161 | * The init_sched_build_groups can't handle what we want to do with node |
6162 | * groups, so roll our own. Now each node has its own list of groups which | |
6163 | * gets dynamically allocated. | |
1da177e4 | 6164 | */ |
9c1cfda2 | 6165 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
d1b55138 | 6166 | static struct sched_group **sched_group_nodes_bycpu[NR_CPUS]; |
1da177e4 | 6167 | |
9c1cfda2 | 6168 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 6169 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 6170 | |
6711cab4 SS |
6171 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
6172 | struct sched_group **sg) | |
9c1cfda2 | 6173 | { |
6711cab4 SS |
6174 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu)); |
6175 | int group; | |
6176 | ||
6177 | cpus_and(nodemask, nodemask, *cpu_map); | |
6178 | group = first_cpu(nodemask); | |
6179 | ||
6180 | if (sg) | |
6181 | *sg = &per_cpu(sched_group_allnodes, group); | |
6182 | return group; | |
1da177e4 | 6183 | } |
6711cab4 | 6184 | |
08069033 SS |
6185 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6186 | { | |
6187 | struct sched_group *sg = group_head; | |
6188 | int j; | |
6189 | ||
6190 | if (!sg) | |
6191 | return; | |
3a5c359a AK |
6192 | do { |
6193 | for_each_cpu_mask(j, sg->cpumask) { | |
6194 | struct sched_domain *sd; | |
08069033 | 6195 | |
3a5c359a AK |
6196 | sd = &per_cpu(phys_domains, j); |
6197 | if (j != first_cpu(sd->groups->cpumask)) { | |
6198 | /* | |
6199 | * Only add "power" once for each | |
6200 | * physical package. | |
6201 | */ | |
6202 | continue; | |
6203 | } | |
08069033 | 6204 | |
3a5c359a AK |
6205 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
6206 | } | |
6207 | sg = sg->next; | |
6208 | } while (sg != group_head); | |
08069033 | 6209 | } |
1da177e4 LT |
6210 | #endif |
6211 | ||
a616058b | 6212 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6213 | /* Free memory allocated for various sched_group structures */ |
6214 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6215 | { | |
a616058b | 6216 | int cpu, i; |
51888ca2 SV |
6217 | |
6218 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
6219 | struct sched_group **sched_group_nodes |
6220 | = sched_group_nodes_bycpu[cpu]; | |
6221 | ||
51888ca2 SV |
6222 | if (!sched_group_nodes) |
6223 | continue; | |
6224 | ||
6225 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6226 | cpumask_t nodemask = node_to_cpumask(i); | |
6227 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; | |
6228 | ||
6229 | cpus_and(nodemask, nodemask, *cpu_map); | |
6230 | if (cpus_empty(nodemask)) | |
6231 | continue; | |
6232 | ||
6233 | if (sg == NULL) | |
6234 | continue; | |
6235 | sg = sg->next; | |
6236 | next_sg: | |
6237 | oldsg = sg; | |
6238 | sg = sg->next; | |
6239 | kfree(oldsg); | |
6240 | if (oldsg != sched_group_nodes[i]) | |
6241 | goto next_sg; | |
6242 | } | |
6243 | kfree(sched_group_nodes); | |
6244 | sched_group_nodes_bycpu[cpu] = NULL; | |
6245 | } | |
51888ca2 | 6246 | } |
a616058b SS |
6247 | #else |
6248 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6249 | { | |
6250 | } | |
6251 | #endif | |
51888ca2 | 6252 | |
89c4710e SS |
6253 | /* |
6254 | * Initialize sched groups cpu_power. | |
6255 | * | |
6256 | * cpu_power indicates the capacity of sched group, which is used while | |
6257 | * distributing the load between different sched groups in a sched domain. | |
6258 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6259 | * there are asymmetries in the topology. If there are asymmetries, group | |
6260 | * having more cpu_power will pickup more load compared to the group having | |
6261 | * less cpu_power. | |
6262 | * | |
6263 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
6264 | * the maximum number of tasks a group can handle in the presence of other idle | |
6265 | * or lightly loaded groups in the same sched domain. | |
6266 | */ | |
6267 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6268 | { | |
6269 | struct sched_domain *child; | |
6270 | struct sched_group *group; | |
6271 | ||
6272 | WARN_ON(!sd || !sd->groups); | |
6273 | ||
6274 | if (cpu != first_cpu(sd->groups->cpumask)) | |
6275 | return; | |
6276 | ||
6277 | child = sd->child; | |
6278 | ||
5517d86b ED |
6279 | sd->groups->__cpu_power = 0; |
6280 | ||
89c4710e SS |
6281 | /* |
6282 | * For perf policy, if the groups in child domain share resources | |
6283 | * (for example cores sharing some portions of the cache hierarchy | |
6284 | * or SMT), then set this domain groups cpu_power such that each group | |
6285 | * can handle only one task, when there are other idle groups in the | |
6286 | * same sched domain. | |
6287 | */ | |
6288 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
6289 | (child->flags & | |
6290 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 6291 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
6292 | return; |
6293 | } | |
6294 | ||
89c4710e SS |
6295 | /* |
6296 | * add cpu_power of each child group to this groups cpu_power | |
6297 | */ | |
6298 | group = child->groups; | |
6299 | do { | |
5517d86b | 6300 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
6301 | group = group->next; |
6302 | } while (group != child->groups); | |
6303 | } | |
6304 | ||
1da177e4 | 6305 | /* |
1a20ff27 DG |
6306 | * Build sched domains for a given set of cpus and attach the sched domains |
6307 | * to the individual cpus | |
1da177e4 | 6308 | */ |
51888ca2 | 6309 | static int build_sched_domains(const cpumask_t *cpu_map) |
1da177e4 LT |
6310 | { |
6311 | int i; | |
57d885fe | 6312 | struct root_domain *rd; |
d1b55138 JH |
6313 | #ifdef CONFIG_NUMA |
6314 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 6315 | int sd_allnodes = 0; |
d1b55138 JH |
6316 | |
6317 | /* | |
6318 | * Allocate the per-node list of sched groups | |
6319 | */ | |
5cf9f062 | 6320 | sched_group_nodes = kcalloc(MAX_NUMNODES, sizeof(struct sched_group *), |
41a2d6cf | 6321 | GFP_KERNEL); |
d1b55138 JH |
6322 | if (!sched_group_nodes) { |
6323 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 6324 | return -ENOMEM; |
d1b55138 JH |
6325 | } |
6326 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
6327 | #endif | |
1da177e4 | 6328 | |
dc938520 | 6329 | rd = alloc_rootdomain(); |
57d885fe GH |
6330 | if (!rd) { |
6331 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
6332 | return -ENOMEM; | |
6333 | } | |
6334 | ||
1da177e4 | 6335 | /* |
1a20ff27 | 6336 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 6337 | */ |
1a20ff27 | 6338 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6339 | struct sched_domain *sd = NULL, *p; |
6340 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(i)); | |
6341 | ||
1a20ff27 | 6342 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6343 | |
6344 | #ifdef CONFIG_NUMA | |
dd41f596 IM |
6345 | if (cpus_weight(*cpu_map) > |
6346 | SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) { | |
9c1cfda2 JH |
6347 | sd = &per_cpu(allnodes_domains, i); |
6348 | *sd = SD_ALLNODES_INIT; | |
6349 | sd->span = *cpu_map; | |
6711cab4 | 6350 | cpu_to_allnodes_group(i, cpu_map, &sd->groups); |
9c1cfda2 | 6351 | p = sd; |
6711cab4 | 6352 | sd_allnodes = 1; |
9c1cfda2 JH |
6353 | } else |
6354 | p = NULL; | |
6355 | ||
1da177e4 | 6356 | sd = &per_cpu(node_domains, i); |
1da177e4 | 6357 | *sd = SD_NODE_INIT; |
9c1cfda2 JH |
6358 | sd->span = sched_domain_node_span(cpu_to_node(i)); |
6359 | sd->parent = p; | |
1a848870 SS |
6360 | if (p) |
6361 | p->child = sd; | |
9c1cfda2 | 6362 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
6363 | #endif |
6364 | ||
6365 | p = sd; | |
6366 | sd = &per_cpu(phys_domains, i); | |
1da177e4 LT |
6367 | *sd = SD_CPU_INIT; |
6368 | sd->span = nodemask; | |
6369 | sd->parent = p; | |
1a848870 SS |
6370 | if (p) |
6371 | p->child = sd; | |
6711cab4 | 6372 | cpu_to_phys_group(i, cpu_map, &sd->groups); |
1da177e4 | 6373 | |
1e9f28fa SS |
6374 | #ifdef CONFIG_SCHED_MC |
6375 | p = sd; | |
6376 | sd = &per_cpu(core_domains, i); | |
1e9f28fa SS |
6377 | *sd = SD_MC_INIT; |
6378 | sd->span = cpu_coregroup_map(i); | |
6379 | cpus_and(sd->span, sd->span, *cpu_map); | |
6380 | sd->parent = p; | |
1a848870 | 6381 | p->child = sd; |
6711cab4 | 6382 | cpu_to_core_group(i, cpu_map, &sd->groups); |
1e9f28fa SS |
6383 | #endif |
6384 | ||
1da177e4 LT |
6385 | #ifdef CONFIG_SCHED_SMT |
6386 | p = sd; | |
6387 | sd = &per_cpu(cpu_domains, i); | |
1da177e4 | 6388 | *sd = SD_SIBLING_INIT; |
d5a7430d | 6389 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6390 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 6391 | sd->parent = p; |
1a848870 | 6392 | p->child = sd; |
6711cab4 | 6393 | cpu_to_cpu_group(i, cpu_map, &sd->groups); |
1da177e4 LT |
6394 | #endif |
6395 | } | |
6396 | ||
6397 | #ifdef CONFIG_SCHED_SMT | |
6398 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 6399 | for_each_cpu_mask(i, *cpu_map) { |
d5a7430d | 6400 | cpumask_t this_sibling_map = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6401 | cpus_and(this_sibling_map, this_sibling_map, *cpu_map); |
1da177e4 LT |
6402 | if (i != first_cpu(this_sibling_map)) |
6403 | continue; | |
6404 | ||
dd41f596 IM |
6405 | init_sched_build_groups(this_sibling_map, cpu_map, |
6406 | &cpu_to_cpu_group); | |
1da177e4 LT |
6407 | } |
6408 | #endif | |
6409 | ||
1e9f28fa SS |
6410 | #ifdef CONFIG_SCHED_MC |
6411 | /* Set up multi-core groups */ | |
6412 | for_each_cpu_mask(i, *cpu_map) { | |
6413 | cpumask_t this_core_map = cpu_coregroup_map(i); | |
6414 | cpus_and(this_core_map, this_core_map, *cpu_map); | |
6415 | if (i != first_cpu(this_core_map)) | |
6416 | continue; | |
dd41f596 IM |
6417 | init_sched_build_groups(this_core_map, cpu_map, |
6418 | &cpu_to_core_group); | |
1e9f28fa SS |
6419 | } |
6420 | #endif | |
6421 | ||
1da177e4 LT |
6422 | /* Set up physical groups */ |
6423 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6424 | cpumask_t nodemask = node_to_cpumask(i); | |
6425 | ||
1a20ff27 | 6426 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6427 | if (cpus_empty(nodemask)) |
6428 | continue; | |
6429 | ||
6711cab4 | 6430 | init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group); |
1da177e4 LT |
6431 | } |
6432 | ||
6433 | #ifdef CONFIG_NUMA | |
6434 | /* Set up node groups */ | |
6711cab4 | 6435 | if (sd_allnodes) |
dd41f596 IM |
6436 | init_sched_build_groups(*cpu_map, cpu_map, |
6437 | &cpu_to_allnodes_group); | |
9c1cfda2 JH |
6438 | |
6439 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6440 | /* Set up node groups */ | |
6441 | struct sched_group *sg, *prev; | |
6442 | cpumask_t nodemask = node_to_cpumask(i); | |
6443 | cpumask_t domainspan; | |
6444 | cpumask_t covered = CPU_MASK_NONE; | |
6445 | int j; | |
6446 | ||
6447 | cpus_and(nodemask, nodemask, *cpu_map); | |
d1b55138 JH |
6448 | if (cpus_empty(nodemask)) { |
6449 | sched_group_nodes[i] = NULL; | |
9c1cfda2 | 6450 | continue; |
d1b55138 | 6451 | } |
9c1cfda2 JH |
6452 | |
6453 | domainspan = sched_domain_node_span(i); | |
6454 | cpus_and(domainspan, domainspan, *cpu_map); | |
6455 | ||
15f0b676 | 6456 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
6457 | if (!sg) { |
6458 | printk(KERN_WARNING "Can not alloc domain group for " | |
6459 | "node %d\n", i); | |
6460 | goto error; | |
6461 | } | |
9c1cfda2 JH |
6462 | sched_group_nodes[i] = sg; |
6463 | for_each_cpu_mask(j, nodemask) { | |
6464 | struct sched_domain *sd; | |
9761eea8 | 6465 | |
9c1cfda2 JH |
6466 | sd = &per_cpu(node_domains, j); |
6467 | sd->groups = sg; | |
9c1cfda2 | 6468 | } |
5517d86b | 6469 | sg->__cpu_power = 0; |
9c1cfda2 | 6470 | sg->cpumask = nodemask; |
51888ca2 | 6471 | sg->next = sg; |
9c1cfda2 JH |
6472 | cpus_or(covered, covered, nodemask); |
6473 | prev = sg; | |
6474 | ||
6475 | for (j = 0; j < MAX_NUMNODES; j++) { | |
6476 | cpumask_t tmp, notcovered; | |
6477 | int n = (i + j) % MAX_NUMNODES; | |
6478 | ||
6479 | cpus_complement(notcovered, covered); | |
6480 | cpus_and(tmp, notcovered, *cpu_map); | |
6481 | cpus_and(tmp, tmp, domainspan); | |
6482 | if (cpus_empty(tmp)) | |
6483 | break; | |
6484 | ||
6485 | nodemask = node_to_cpumask(n); | |
6486 | cpus_and(tmp, tmp, nodemask); | |
6487 | if (cpus_empty(tmp)) | |
6488 | continue; | |
6489 | ||
15f0b676 SV |
6490 | sg = kmalloc_node(sizeof(struct sched_group), |
6491 | GFP_KERNEL, i); | |
9c1cfda2 JH |
6492 | if (!sg) { |
6493 | printk(KERN_WARNING | |
6494 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 6495 | goto error; |
9c1cfda2 | 6496 | } |
5517d86b | 6497 | sg->__cpu_power = 0; |
9c1cfda2 | 6498 | sg->cpumask = tmp; |
51888ca2 | 6499 | sg->next = prev->next; |
9c1cfda2 JH |
6500 | cpus_or(covered, covered, tmp); |
6501 | prev->next = sg; | |
6502 | prev = sg; | |
6503 | } | |
9c1cfda2 | 6504 | } |
1da177e4 LT |
6505 | #endif |
6506 | ||
6507 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 6508 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 6509 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6510 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
6511 | ||
89c4710e | 6512 | init_sched_groups_power(i, sd); |
5c45bf27 | 6513 | } |
1da177e4 | 6514 | #endif |
1e9f28fa | 6515 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 6516 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6517 | struct sched_domain *sd = &per_cpu(core_domains, i); |
6518 | ||
89c4710e | 6519 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
6520 | } |
6521 | #endif | |
1e9f28fa | 6522 | |
5c45bf27 | 6523 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6524 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
6525 | ||
89c4710e | 6526 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6527 | } |
6528 | ||
9c1cfda2 | 6529 | #ifdef CONFIG_NUMA |
08069033 SS |
6530 | for (i = 0; i < MAX_NUMNODES; i++) |
6531 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 6532 | |
6711cab4 SS |
6533 | if (sd_allnodes) { |
6534 | struct sched_group *sg; | |
f712c0c7 | 6535 | |
6711cab4 | 6536 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg); |
f712c0c7 SS |
6537 | init_numa_sched_groups_power(sg); |
6538 | } | |
9c1cfda2 JH |
6539 | #endif |
6540 | ||
1da177e4 | 6541 | /* Attach the domains */ |
1a20ff27 | 6542 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6543 | struct sched_domain *sd; |
6544 | #ifdef CONFIG_SCHED_SMT | |
6545 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
6546 | #elif defined(CONFIG_SCHED_MC) |
6547 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
6548 | #else |
6549 | sd = &per_cpu(phys_domains, i); | |
6550 | #endif | |
57d885fe | 6551 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 6552 | } |
51888ca2 SV |
6553 | |
6554 | return 0; | |
6555 | ||
a616058b | 6556 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6557 | error: |
6558 | free_sched_groups(cpu_map); | |
6559 | return -ENOMEM; | |
a616058b | 6560 | #endif |
1da177e4 | 6561 | } |
029190c5 PJ |
6562 | |
6563 | static cpumask_t *doms_cur; /* current sched domains */ | |
6564 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
6565 | ||
6566 | /* | |
6567 | * Special case: If a kmalloc of a doms_cur partition (array of | |
6568 | * cpumask_t) fails, then fallback to a single sched domain, | |
6569 | * as determined by the single cpumask_t fallback_doms. | |
6570 | */ | |
6571 | static cpumask_t fallback_doms; | |
6572 | ||
1a20ff27 | 6573 | /* |
41a2d6cf | 6574 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6575 | * For now this just excludes isolated cpus, but could be used to |
6576 | * exclude other special cases in the future. | |
1a20ff27 | 6577 | */ |
51888ca2 | 6578 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 6579 | { |
7378547f MM |
6580 | int err; |
6581 | ||
029190c5 PJ |
6582 | ndoms_cur = 1; |
6583 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
6584 | if (!doms_cur) | |
6585 | doms_cur = &fallback_doms; | |
6586 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
7378547f | 6587 | err = build_sched_domains(doms_cur); |
6382bc90 | 6588 | register_sched_domain_sysctl(); |
7378547f MM |
6589 | |
6590 | return err; | |
1a20ff27 DG |
6591 | } |
6592 | ||
6593 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map) | |
1da177e4 | 6594 | { |
51888ca2 | 6595 | free_sched_groups(cpu_map); |
9c1cfda2 | 6596 | } |
1da177e4 | 6597 | |
1a20ff27 DG |
6598 | /* |
6599 | * Detach sched domains from a group of cpus specified in cpu_map | |
6600 | * These cpus will now be attached to the NULL domain | |
6601 | */ | |
858119e1 | 6602 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6603 | { |
6604 | int i; | |
6605 | ||
6382bc90 MM |
6606 | unregister_sched_domain_sysctl(); |
6607 | ||
1a20ff27 | 6608 | for_each_cpu_mask(i, *cpu_map) |
57d885fe | 6609 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 DG |
6610 | synchronize_sched(); |
6611 | arch_destroy_sched_domains(cpu_map); | |
6612 | } | |
6613 | ||
029190c5 PJ |
6614 | /* |
6615 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6616 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6617 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6618 | * It destroys each deleted domain and builds each new domain. | |
6619 | * | |
6620 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
41a2d6cf IM |
6621 | * The masks don't intersect (don't overlap.) We should setup one |
6622 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6623 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6624 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6625 | * it as it is. | |
6626 | * | |
41a2d6cf IM |
6627 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
6628 | * ownership of it and will kfree it when done with it. If the caller | |
029190c5 PJ |
6629 | * failed the kmalloc call, then it can pass in doms_new == NULL, |
6630 | * and partition_sched_domains() will fallback to the single partition | |
6631 | * 'fallback_doms'. | |
6632 | * | |
6633 | * Call with hotplug lock held | |
6634 | */ | |
6635 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new) | |
6636 | { | |
6637 | int i, j; | |
6638 | ||
a1835615 SV |
6639 | lock_doms_cur(); |
6640 | ||
7378547f MM |
6641 | /* always unregister in case we don't destroy any domains */ |
6642 | unregister_sched_domain_sysctl(); | |
6643 | ||
029190c5 PJ |
6644 | if (doms_new == NULL) { |
6645 | ndoms_new = 1; | |
6646 | doms_new = &fallback_doms; | |
6647 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
6648 | } | |
6649 | ||
6650 | /* Destroy deleted domains */ | |
6651 | for (i = 0; i < ndoms_cur; i++) { | |
6652 | for (j = 0; j < ndoms_new; j++) { | |
6653 | if (cpus_equal(doms_cur[i], doms_new[j])) | |
6654 | goto match1; | |
6655 | } | |
6656 | /* no match - a current sched domain not in new doms_new[] */ | |
6657 | detach_destroy_domains(doms_cur + i); | |
6658 | match1: | |
6659 | ; | |
6660 | } | |
6661 | ||
6662 | /* Build new domains */ | |
6663 | for (i = 0; i < ndoms_new; i++) { | |
6664 | for (j = 0; j < ndoms_cur; j++) { | |
6665 | if (cpus_equal(doms_new[i], doms_cur[j])) | |
6666 | goto match2; | |
6667 | } | |
6668 | /* no match - add a new doms_new */ | |
6669 | build_sched_domains(doms_new + i); | |
6670 | match2: | |
6671 | ; | |
6672 | } | |
6673 | ||
6674 | /* Remember the new sched domains */ | |
6675 | if (doms_cur != &fallback_doms) | |
6676 | kfree(doms_cur); | |
6677 | doms_cur = doms_new; | |
6678 | ndoms_cur = ndoms_new; | |
7378547f MM |
6679 | |
6680 | register_sched_domain_sysctl(); | |
a1835615 SV |
6681 | |
6682 | unlock_doms_cur(); | |
029190c5 PJ |
6683 | } |
6684 | ||
5c45bf27 | 6685 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
6707de00 | 6686 | static int arch_reinit_sched_domains(void) |
5c45bf27 SS |
6687 | { |
6688 | int err; | |
6689 | ||
95402b38 | 6690 | get_online_cpus(); |
5c45bf27 SS |
6691 | detach_destroy_domains(&cpu_online_map); |
6692 | err = arch_init_sched_domains(&cpu_online_map); | |
95402b38 | 6693 | put_online_cpus(); |
5c45bf27 SS |
6694 | |
6695 | return err; | |
6696 | } | |
6697 | ||
6698 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6699 | { | |
6700 | int ret; | |
6701 | ||
6702 | if (buf[0] != '0' && buf[0] != '1') | |
6703 | return -EINVAL; | |
6704 | ||
6705 | if (smt) | |
6706 | sched_smt_power_savings = (buf[0] == '1'); | |
6707 | else | |
6708 | sched_mc_power_savings = (buf[0] == '1'); | |
6709 | ||
6710 | ret = arch_reinit_sched_domains(); | |
6711 | ||
6712 | return ret ? ret : count; | |
6713 | } | |
6714 | ||
5c45bf27 SS |
6715 | #ifdef CONFIG_SCHED_MC |
6716 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
6717 | { | |
6718 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
6719 | } | |
48f24c4d IM |
6720 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
6721 | const char *buf, size_t count) | |
5c45bf27 SS |
6722 | { |
6723 | return sched_power_savings_store(buf, count, 0); | |
6724 | } | |
6707de00 AB |
6725 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
6726 | sched_mc_power_savings_store); | |
5c45bf27 SS |
6727 | #endif |
6728 | ||
6729 | #ifdef CONFIG_SCHED_SMT | |
6730 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
6731 | { | |
6732 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
6733 | } | |
48f24c4d IM |
6734 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
6735 | const char *buf, size_t count) | |
5c45bf27 SS |
6736 | { |
6737 | return sched_power_savings_store(buf, count, 1); | |
6738 | } | |
6707de00 AB |
6739 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
6740 | sched_smt_power_savings_store); | |
6741 | #endif | |
6742 | ||
6743 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
6744 | { | |
6745 | int err = 0; | |
6746 | ||
6747 | #ifdef CONFIG_SCHED_SMT | |
6748 | if (smt_capable()) | |
6749 | err = sysfs_create_file(&cls->kset.kobj, | |
6750 | &attr_sched_smt_power_savings.attr); | |
6751 | #endif | |
6752 | #ifdef CONFIG_SCHED_MC | |
6753 | if (!err && mc_capable()) | |
6754 | err = sysfs_create_file(&cls->kset.kobj, | |
6755 | &attr_sched_mc_power_savings.attr); | |
6756 | #endif | |
6757 | return err; | |
6758 | } | |
5c45bf27 SS |
6759 | #endif |
6760 | ||
1da177e4 | 6761 | /* |
41a2d6cf | 6762 | * Force a reinitialization of the sched domains hierarchy. The domains |
1da177e4 | 6763 | * and groups cannot be updated in place without racing with the balancing |
41c7ce9a | 6764 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
6765 | * which will prevent rebalancing while the sched domains are recalculated. |
6766 | */ | |
6767 | static int update_sched_domains(struct notifier_block *nfb, | |
6768 | unsigned long action, void *hcpu) | |
6769 | { | |
1da177e4 LT |
6770 | switch (action) { |
6771 | case CPU_UP_PREPARE: | |
8bb78442 | 6772 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 6773 | case CPU_DOWN_PREPARE: |
8bb78442 | 6774 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 6775 | detach_destroy_domains(&cpu_online_map); |
1da177e4 LT |
6776 | return NOTIFY_OK; |
6777 | ||
6778 | case CPU_UP_CANCELED: | |
8bb78442 | 6779 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 6780 | case CPU_DOWN_FAILED: |
8bb78442 | 6781 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 6782 | case CPU_ONLINE: |
8bb78442 | 6783 | case CPU_ONLINE_FROZEN: |
1da177e4 | 6784 | case CPU_DEAD: |
8bb78442 | 6785 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
6786 | /* |
6787 | * Fall through and re-initialise the domains. | |
6788 | */ | |
6789 | break; | |
6790 | default: | |
6791 | return NOTIFY_DONE; | |
6792 | } | |
6793 | ||
6794 | /* The hotplug lock is already held by cpu_up/cpu_down */ | |
1a20ff27 | 6795 | arch_init_sched_domains(&cpu_online_map); |
1da177e4 LT |
6796 | |
6797 | return NOTIFY_OK; | |
6798 | } | |
1da177e4 LT |
6799 | |
6800 | void __init sched_init_smp(void) | |
6801 | { | |
5c1e1767 NP |
6802 | cpumask_t non_isolated_cpus; |
6803 | ||
95402b38 | 6804 | get_online_cpus(); |
1a20ff27 | 6805 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 6806 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
6807 | if (cpus_empty(non_isolated_cpus)) |
6808 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
95402b38 | 6809 | put_online_cpus(); |
1da177e4 LT |
6810 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
6811 | hotcpu_notifier(update_sched_domains, 0); | |
5c1e1767 NP |
6812 | |
6813 | /* Move init over to a non-isolated CPU */ | |
6814 | if (set_cpus_allowed(current, non_isolated_cpus) < 0) | |
6815 | BUG(); | |
19978ca6 | 6816 | sched_init_granularity(); |
6b2d7700 SV |
6817 | |
6818 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6819 | if (nr_cpu_ids == 1) | |
6820 | return; | |
6821 | ||
6822 | lb_monitor_task = kthread_create(load_balance_monitor, NULL, | |
6823 | "group_balance"); | |
6824 | if (!IS_ERR(lb_monitor_task)) { | |
6825 | lb_monitor_task->flags |= PF_NOFREEZE; | |
6826 | wake_up_process(lb_monitor_task); | |
6827 | } else { | |
6828 | printk(KERN_ERR "Could not create load balance monitor thread" | |
6829 | "(error = %ld) \n", PTR_ERR(lb_monitor_task)); | |
6830 | } | |
6831 | #endif | |
1da177e4 LT |
6832 | } |
6833 | #else | |
6834 | void __init sched_init_smp(void) | |
6835 | { | |
19978ca6 | 6836 | sched_init_granularity(); |
1da177e4 LT |
6837 | } |
6838 | #endif /* CONFIG_SMP */ | |
6839 | ||
6840 | int in_sched_functions(unsigned long addr) | |
6841 | { | |
1da177e4 LT |
6842 | return in_lock_functions(addr) || |
6843 | (addr >= (unsigned long)__sched_text_start | |
6844 | && addr < (unsigned long)__sched_text_end); | |
6845 | } | |
6846 | ||
a9957449 | 6847 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
6848 | { |
6849 | cfs_rq->tasks_timeline = RB_ROOT; | |
dd41f596 IM |
6850 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6851 | cfs_rq->rq = rq; | |
6852 | #endif | |
67e9fb2a | 6853 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
6854 | } |
6855 | ||
1da177e4 LT |
6856 | void __init sched_init(void) |
6857 | { | |
476f3534 | 6858 | int highest_cpu = 0; |
dd41f596 IM |
6859 | int i, j; |
6860 | ||
57d885fe GH |
6861 | #ifdef CONFIG_SMP |
6862 | init_defrootdomain(); | |
6863 | #endif | |
6864 | ||
0a945022 | 6865 | for_each_possible_cpu(i) { |
dd41f596 | 6866 | struct rt_prio_array *array; |
70b97a7f | 6867 | struct rq *rq; |
1da177e4 LT |
6868 | |
6869 | rq = cpu_rq(i); | |
6870 | spin_lock_init(&rq->lock); | |
fcb99371 | 6871 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 6872 | rq->nr_running = 0; |
dd41f596 IM |
6873 | rq->clock = 1; |
6874 | init_cfs_rq(&rq->cfs, rq); | |
6875 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6876 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); | |
3a252015 IM |
6877 | { |
6878 | struct cfs_rq *cfs_rq = &per_cpu(init_cfs_rq, i); | |
6879 | struct sched_entity *se = | |
6880 | &per_cpu(init_sched_entity, i); | |
6881 | ||
6882 | init_cfs_rq_p[i] = cfs_rq; | |
6883 | init_cfs_rq(cfs_rq, rq); | |
4cf86d77 | 6884 | cfs_rq->tg = &init_task_group; |
3a252015 | 6885 | list_add(&cfs_rq->leaf_cfs_rq_list, |
29f59db3 SV |
6886 | &rq->leaf_cfs_rq_list); |
6887 | ||
3a252015 IM |
6888 | init_sched_entity_p[i] = se; |
6889 | se->cfs_rq = &rq->cfs; | |
6890 | se->my_q = cfs_rq; | |
4cf86d77 | 6891 | se->load.weight = init_task_group_load; |
9b5b7751 | 6892 | se->load.inv_weight = |
4cf86d77 | 6893 | div64_64(1ULL<<32, init_task_group_load); |
3a252015 IM |
6894 | se->parent = NULL; |
6895 | } | |
4cf86d77 | 6896 | init_task_group.shares = init_task_group_load; |
dd41f596 | 6897 | #endif |
1da177e4 | 6898 | |
dd41f596 IM |
6899 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6900 | rq->cpu_load[j] = 0; | |
1da177e4 | 6901 | #ifdef CONFIG_SMP |
41c7ce9a | 6902 | rq->sd = NULL; |
57d885fe | 6903 | rq->rd = NULL; |
1da177e4 | 6904 | rq->active_balance = 0; |
dd41f596 | 6905 | rq->next_balance = jiffies; |
1da177e4 | 6906 | rq->push_cpu = 0; |
0a2966b4 | 6907 | rq->cpu = i; |
1da177e4 LT |
6908 | rq->migration_thread = NULL; |
6909 | INIT_LIST_HEAD(&rq->migration_queue); | |
764a9d6f | 6910 | rq->rt.highest_prio = MAX_RT_PRIO; |
a22d7fc1 | 6911 | rq->rt.overloaded = 0; |
dc938520 | 6912 | rq_attach_root(rq, &def_root_domain); |
1da177e4 LT |
6913 | #endif |
6914 | atomic_set(&rq->nr_iowait, 0); | |
6915 | ||
dd41f596 IM |
6916 | array = &rq->rt.active; |
6917 | for (j = 0; j < MAX_RT_PRIO; j++) { | |
6918 | INIT_LIST_HEAD(array->queue + j); | |
6919 | __clear_bit(j, array->bitmap); | |
1da177e4 | 6920 | } |
476f3534 | 6921 | highest_cpu = i; |
dd41f596 IM |
6922 | /* delimiter for bitsearch: */ |
6923 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
1da177e4 LT |
6924 | } |
6925 | ||
2dd73a4f | 6926 | set_load_weight(&init_task); |
b50f60ce | 6927 | |
e107be36 AK |
6928 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6929 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6930 | #endif | |
6931 | ||
c9819f45 | 6932 | #ifdef CONFIG_SMP |
476f3534 | 6933 | nr_cpu_ids = highest_cpu + 1; |
c9819f45 CL |
6934 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); |
6935 | #endif | |
6936 | ||
b50f60ce HC |
6937 | #ifdef CONFIG_RT_MUTEXES |
6938 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
6939 | #endif | |
6940 | ||
1da177e4 LT |
6941 | /* |
6942 | * The boot idle thread does lazy MMU switching as well: | |
6943 | */ | |
6944 | atomic_inc(&init_mm.mm_count); | |
6945 | enter_lazy_tlb(&init_mm, current); | |
6946 | ||
6947 | /* | |
6948 | * Make us the idle thread. Technically, schedule() should not be | |
6949 | * called from this thread, however somewhere below it might be, | |
6950 | * but because we are the idle thread, we just pick up running again | |
6951 | * when this runqueue becomes "idle". | |
6952 | */ | |
6953 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
6954 | /* |
6955 | * During early bootup we pretend to be a normal task: | |
6956 | */ | |
6957 | current->sched_class = &fair_sched_class; | |
1da177e4 LT |
6958 | } |
6959 | ||
6960 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
6961 | void __might_sleep(char *file, int line) | |
6962 | { | |
48f24c4d | 6963 | #ifdef in_atomic |
1da177e4 LT |
6964 | static unsigned long prev_jiffy; /* ratelimiting */ |
6965 | ||
6966 | if ((in_atomic() || irqs_disabled()) && | |
6967 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
6968 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6969 | return; | |
6970 | prev_jiffy = jiffies; | |
91368d73 | 6971 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
6972 | " context at %s:%d\n", file, line); |
6973 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
6974 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 6975 | debug_show_held_locks(current); |
3117df04 IM |
6976 | if (irqs_disabled()) |
6977 | print_irqtrace_events(current); | |
1da177e4 LT |
6978 | dump_stack(); |
6979 | } | |
6980 | #endif | |
6981 | } | |
6982 | EXPORT_SYMBOL(__might_sleep); | |
6983 | #endif | |
6984 | ||
6985 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
6986 | static void normalize_task(struct rq *rq, struct task_struct *p) |
6987 | { | |
6988 | int on_rq; | |
6989 | update_rq_clock(rq); | |
6990 | on_rq = p->se.on_rq; | |
6991 | if (on_rq) | |
6992 | deactivate_task(rq, p, 0); | |
6993 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
6994 | if (on_rq) { | |
6995 | activate_task(rq, p, 0); | |
6996 | resched_task(rq->curr); | |
6997 | } | |
6998 | } | |
6999 | ||
1da177e4 LT |
7000 | void normalize_rt_tasks(void) |
7001 | { | |
a0f98a1c | 7002 | struct task_struct *g, *p; |
1da177e4 | 7003 | unsigned long flags; |
70b97a7f | 7004 | struct rq *rq; |
1da177e4 LT |
7005 | |
7006 | read_lock_irq(&tasklist_lock); | |
a0f98a1c | 7007 | do_each_thread(g, p) { |
178be793 IM |
7008 | /* |
7009 | * Only normalize user tasks: | |
7010 | */ | |
7011 | if (!p->mm) | |
7012 | continue; | |
7013 | ||
6cfb0d5d | 7014 | p->se.exec_start = 0; |
6cfb0d5d | 7015 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 7016 | p->se.wait_start = 0; |
dd41f596 | 7017 | p->se.sleep_start = 0; |
dd41f596 | 7018 | p->se.block_start = 0; |
6cfb0d5d | 7019 | #endif |
dd41f596 IM |
7020 | task_rq(p)->clock = 0; |
7021 | ||
7022 | if (!rt_task(p)) { | |
7023 | /* | |
7024 | * Renice negative nice level userspace | |
7025 | * tasks back to 0: | |
7026 | */ | |
7027 | if (TASK_NICE(p) < 0 && p->mm) | |
7028 | set_user_nice(p, 0); | |
1da177e4 | 7029 | continue; |
dd41f596 | 7030 | } |
1da177e4 | 7031 | |
b29739f9 IM |
7032 | spin_lock_irqsave(&p->pi_lock, flags); |
7033 | rq = __task_rq_lock(p); | |
1da177e4 | 7034 | |
178be793 | 7035 | normalize_task(rq, p); |
3a5e4dc1 | 7036 | |
b29739f9 IM |
7037 | __task_rq_unlock(rq); |
7038 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
a0f98a1c IM |
7039 | } while_each_thread(g, p); |
7040 | ||
1da177e4 LT |
7041 | read_unlock_irq(&tasklist_lock); |
7042 | } | |
7043 | ||
7044 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
7045 | |
7046 | #ifdef CONFIG_IA64 | |
7047 | /* | |
7048 | * These functions are only useful for the IA64 MCA handling. | |
7049 | * | |
7050 | * They can only be called when the whole system has been | |
7051 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7052 | * activity can take place. Using them for anything else would | |
7053 | * be a serious bug, and as a result, they aren't even visible | |
7054 | * under any other configuration. | |
7055 | */ | |
7056 | ||
7057 | /** | |
7058 | * curr_task - return the current task for a given cpu. | |
7059 | * @cpu: the processor in question. | |
7060 | * | |
7061 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7062 | */ | |
36c8b586 | 7063 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7064 | { |
7065 | return cpu_curr(cpu); | |
7066 | } | |
7067 | ||
7068 | /** | |
7069 | * set_curr_task - set the current task for a given cpu. | |
7070 | * @cpu: the processor in question. | |
7071 | * @p: the task pointer to set. | |
7072 | * | |
7073 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7074 | * are serviced on a separate stack. It allows the architecture to switch the |
7075 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7076 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7077 | * and caller must save the original value of the current task (see | |
7078 | * curr_task() above) and restore that value before reenabling interrupts and | |
7079 | * re-starting the system. | |
7080 | * | |
7081 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7082 | */ | |
36c8b586 | 7083 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7084 | { |
7085 | cpu_curr(cpu) = p; | |
7086 | } | |
7087 | ||
7088 | #endif | |
29f59db3 SV |
7089 | |
7090 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7091 | ||
6b2d7700 SV |
7092 | #ifdef CONFIG_SMP |
7093 | /* | |
7094 | * distribute shares of all task groups among their schedulable entities, | |
7095 | * to reflect load distrbution across cpus. | |
7096 | */ | |
7097 | static int rebalance_shares(struct sched_domain *sd, int this_cpu) | |
7098 | { | |
7099 | struct cfs_rq *cfs_rq; | |
7100 | struct rq *rq = cpu_rq(this_cpu); | |
7101 | cpumask_t sdspan = sd->span; | |
7102 | int balanced = 1; | |
7103 | ||
7104 | /* Walk thr' all the task groups that we have */ | |
7105 | for_each_leaf_cfs_rq(rq, cfs_rq) { | |
7106 | int i; | |
7107 | unsigned long total_load = 0, total_shares; | |
7108 | struct task_group *tg = cfs_rq->tg; | |
7109 | ||
7110 | /* Gather total task load of this group across cpus */ | |
7111 | for_each_cpu_mask(i, sdspan) | |
7112 | total_load += tg->cfs_rq[i]->load.weight; | |
7113 | ||
0eab9146 | 7114 | /* Nothing to do if this group has no load */ |
6b2d7700 SV |
7115 | if (!total_load) |
7116 | continue; | |
7117 | ||
7118 | /* | |
7119 | * tg->shares represents the number of cpu shares the task group | |
7120 | * is eligible to hold on a single cpu. On N cpus, it is | |
7121 | * eligible to hold (N * tg->shares) number of cpu shares. | |
7122 | */ | |
7123 | total_shares = tg->shares * cpus_weight(sdspan); | |
7124 | ||
7125 | /* | |
7126 | * redistribute total_shares across cpus as per the task load | |
7127 | * distribution. | |
7128 | */ | |
7129 | for_each_cpu_mask(i, sdspan) { | |
7130 | unsigned long local_load, local_shares; | |
7131 | ||
7132 | local_load = tg->cfs_rq[i]->load.weight; | |
7133 | local_shares = (local_load * total_shares) / total_load; | |
7134 | if (!local_shares) | |
7135 | local_shares = MIN_GROUP_SHARES; | |
7136 | if (local_shares == tg->se[i]->load.weight) | |
7137 | continue; | |
7138 | ||
7139 | spin_lock_irq(&cpu_rq(i)->lock); | |
7140 | set_se_shares(tg->se[i], local_shares); | |
7141 | spin_unlock_irq(&cpu_rq(i)->lock); | |
7142 | balanced = 0; | |
7143 | } | |
7144 | } | |
7145 | ||
7146 | return balanced; | |
7147 | } | |
7148 | ||
7149 | /* | |
7150 | * How frequently should we rebalance_shares() across cpus? | |
7151 | * | |
7152 | * The more frequently we rebalance shares, the more accurate is the fairness | |
7153 | * of cpu bandwidth distribution between task groups. However higher frequency | |
7154 | * also implies increased scheduling overhead. | |
7155 | * | |
7156 | * sysctl_sched_min_bal_int_shares represents the minimum interval between | |
7157 | * consecutive calls to rebalance_shares() in the same sched domain. | |
7158 | * | |
7159 | * sysctl_sched_max_bal_int_shares represents the maximum interval between | |
7160 | * consecutive calls to rebalance_shares() in the same sched domain. | |
7161 | * | |
7162 | * These settings allows for the appropriate tradeoff between accuracy of | |
7163 | * fairness and the associated overhead. | |
7164 | * | |
7165 | */ | |
7166 | ||
7167 | /* default: 8ms, units: milliseconds */ | |
7168 | const_debug unsigned int sysctl_sched_min_bal_int_shares = 8; | |
7169 | ||
7170 | /* default: 128ms, units: milliseconds */ | |
7171 | const_debug unsigned int sysctl_sched_max_bal_int_shares = 128; | |
7172 | ||
7173 | /* kernel thread that runs rebalance_shares() periodically */ | |
7174 | static int load_balance_monitor(void *unused) | |
7175 | { | |
7176 | unsigned int timeout = sysctl_sched_min_bal_int_shares; | |
7177 | struct sched_param schedparm; | |
7178 | int ret; | |
7179 | ||
7180 | /* | |
7181 | * We don't want this thread's execution to be limited by the shares | |
7182 | * assigned to default group (init_task_group). Hence make it run | |
7183 | * as a SCHED_RR RT task at the lowest priority. | |
7184 | */ | |
7185 | schedparm.sched_priority = 1; | |
7186 | ret = sched_setscheduler(current, SCHED_RR, &schedparm); | |
7187 | if (ret) | |
7188 | printk(KERN_ERR "Couldn't set SCHED_RR policy for load balance" | |
7189 | " monitor thread (error = %d) \n", ret); | |
7190 | ||
7191 | while (!kthread_should_stop()) { | |
7192 | int i, cpu, balanced = 1; | |
7193 | ||
7194 | /* Prevent cpus going down or coming up */ | |
86ef5c9a | 7195 | get_online_cpus(); |
6b2d7700 SV |
7196 | /* lockout changes to doms_cur[] array */ |
7197 | lock_doms_cur(); | |
7198 | /* | |
7199 | * Enter a rcu read-side critical section to safely walk rq->sd | |
7200 | * chain on various cpus and to walk task group list | |
7201 | * (rq->leaf_cfs_rq_list) in rebalance_shares(). | |
7202 | */ | |
7203 | rcu_read_lock(); | |
7204 | ||
7205 | for (i = 0; i < ndoms_cur; i++) { | |
7206 | cpumask_t cpumap = doms_cur[i]; | |
7207 | struct sched_domain *sd = NULL, *sd_prev = NULL; | |
7208 | ||
7209 | cpu = first_cpu(cpumap); | |
7210 | ||
7211 | /* Find the highest domain at which to balance shares */ | |
7212 | for_each_domain(cpu, sd) { | |
7213 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
7214 | continue; | |
7215 | sd_prev = sd; | |
7216 | } | |
7217 | ||
7218 | sd = sd_prev; | |
7219 | /* sd == NULL? No load balance reqd in this domain */ | |
7220 | if (!sd) | |
7221 | continue; | |
7222 | ||
7223 | balanced &= rebalance_shares(sd, cpu); | |
7224 | } | |
7225 | ||
7226 | rcu_read_unlock(); | |
7227 | ||
7228 | unlock_doms_cur(); | |
86ef5c9a | 7229 | put_online_cpus(); |
6b2d7700 SV |
7230 | |
7231 | if (!balanced) | |
7232 | timeout = sysctl_sched_min_bal_int_shares; | |
7233 | else if (timeout < sysctl_sched_max_bal_int_shares) | |
7234 | timeout *= 2; | |
7235 | ||
7236 | msleep_interruptible(timeout); | |
7237 | } | |
7238 | ||
7239 | return 0; | |
7240 | } | |
7241 | #endif /* CONFIG_SMP */ | |
7242 | ||
29f59db3 | 7243 | /* allocate runqueue etc for a new task group */ |
4cf86d77 | 7244 | struct task_group *sched_create_group(void) |
29f59db3 | 7245 | { |
4cf86d77 | 7246 | struct task_group *tg; |
29f59db3 SV |
7247 | struct cfs_rq *cfs_rq; |
7248 | struct sched_entity *se; | |
9b5b7751 | 7249 | struct rq *rq; |
29f59db3 SV |
7250 | int i; |
7251 | ||
29f59db3 SV |
7252 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); |
7253 | if (!tg) | |
7254 | return ERR_PTR(-ENOMEM); | |
7255 | ||
9b5b7751 | 7256 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
7257 | if (!tg->cfs_rq) |
7258 | goto err; | |
9b5b7751 | 7259 | tg->se = kzalloc(sizeof(se) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
7260 | if (!tg->se) |
7261 | goto err; | |
7262 | ||
7263 | for_each_possible_cpu(i) { | |
9b5b7751 | 7264 | rq = cpu_rq(i); |
29f59db3 SV |
7265 | |
7266 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), GFP_KERNEL, | |
7267 | cpu_to_node(i)); | |
7268 | if (!cfs_rq) | |
7269 | goto err; | |
7270 | ||
7271 | se = kmalloc_node(sizeof(struct sched_entity), GFP_KERNEL, | |
7272 | cpu_to_node(i)); | |
7273 | if (!se) | |
7274 | goto err; | |
7275 | ||
7276 | memset(cfs_rq, 0, sizeof(struct cfs_rq)); | |
7277 | memset(se, 0, sizeof(struct sched_entity)); | |
7278 | ||
7279 | tg->cfs_rq[i] = cfs_rq; | |
7280 | init_cfs_rq(cfs_rq, rq); | |
7281 | cfs_rq->tg = tg; | |
29f59db3 SV |
7282 | |
7283 | tg->se[i] = se; | |
7284 | se->cfs_rq = &rq->cfs; | |
7285 | se->my_q = cfs_rq; | |
7286 | se->load.weight = NICE_0_LOAD; | |
7287 | se->load.inv_weight = div64_64(1ULL<<32, NICE_0_LOAD); | |
7288 | se->parent = NULL; | |
7289 | } | |
7290 | ||
ec2c507f SV |
7291 | tg->shares = NICE_0_LOAD; |
7292 | ||
7293 | lock_task_group_list(); | |
9b5b7751 SV |
7294 | for_each_possible_cpu(i) { |
7295 | rq = cpu_rq(i); | |
7296 | cfs_rq = tg->cfs_rq[i]; | |
7297 | list_add_rcu(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7298 | } | |
ec2c507f | 7299 | unlock_task_group_list(); |
29f59db3 | 7300 | |
9b5b7751 | 7301 | return tg; |
29f59db3 SV |
7302 | |
7303 | err: | |
7304 | for_each_possible_cpu(i) { | |
a65914b3 | 7305 | if (tg->cfs_rq) |
29f59db3 | 7306 | kfree(tg->cfs_rq[i]); |
a65914b3 | 7307 | if (tg->se) |
29f59db3 SV |
7308 | kfree(tg->se[i]); |
7309 | } | |
a65914b3 IM |
7310 | kfree(tg->cfs_rq); |
7311 | kfree(tg->se); | |
7312 | kfree(tg); | |
29f59db3 SV |
7313 | |
7314 | return ERR_PTR(-ENOMEM); | |
7315 | } | |
7316 | ||
9b5b7751 SV |
7317 | /* rcu callback to free various structures associated with a task group */ |
7318 | static void free_sched_group(struct rcu_head *rhp) | |
29f59db3 | 7319 | { |
ae8393e5 SV |
7320 | struct task_group *tg = container_of(rhp, struct task_group, rcu); |
7321 | struct cfs_rq *cfs_rq; | |
29f59db3 SV |
7322 | struct sched_entity *se; |
7323 | int i; | |
7324 | ||
29f59db3 SV |
7325 | /* now it should be safe to free those cfs_rqs */ |
7326 | for_each_possible_cpu(i) { | |
7327 | cfs_rq = tg->cfs_rq[i]; | |
7328 | kfree(cfs_rq); | |
7329 | ||
7330 | se = tg->se[i]; | |
7331 | kfree(se); | |
7332 | } | |
7333 | ||
7334 | kfree(tg->cfs_rq); | |
7335 | kfree(tg->se); | |
7336 | kfree(tg); | |
7337 | } | |
7338 | ||
9b5b7751 | 7339 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7340 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 7341 | { |
7bae49d4 | 7342 | struct cfs_rq *cfs_rq = NULL; |
9b5b7751 | 7343 | int i; |
29f59db3 | 7344 | |
ec2c507f | 7345 | lock_task_group_list(); |
9b5b7751 SV |
7346 | for_each_possible_cpu(i) { |
7347 | cfs_rq = tg->cfs_rq[i]; | |
7348 | list_del_rcu(&cfs_rq->leaf_cfs_rq_list); | |
7349 | } | |
ec2c507f | 7350 | unlock_task_group_list(); |
9b5b7751 | 7351 | |
7bae49d4 | 7352 | BUG_ON(!cfs_rq); |
9b5b7751 SV |
7353 | |
7354 | /* wait for possible concurrent references to cfs_rqs complete */ | |
ae8393e5 | 7355 | call_rcu(&tg->rcu, free_sched_group); |
29f59db3 SV |
7356 | } |
7357 | ||
9b5b7751 | 7358 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7359 | * The caller of this function should have put the task in its new group |
7360 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7361 | * reflect its new group. | |
9b5b7751 SV |
7362 | */ |
7363 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
7364 | { |
7365 | int on_rq, running; | |
7366 | unsigned long flags; | |
7367 | struct rq *rq; | |
7368 | ||
7369 | rq = task_rq_lock(tsk, &flags); | |
7370 | ||
dae51f56 | 7371 | if (tsk->sched_class != &fair_sched_class) { |
ce96b5ac | 7372 | set_task_cfs_rq(tsk, task_cpu(tsk)); |
29f59db3 | 7373 | goto done; |
dae51f56 | 7374 | } |
29f59db3 SV |
7375 | |
7376 | update_rq_clock(rq); | |
7377 | ||
051a1d1a | 7378 | running = task_current(rq, tsk); |
29f59db3 SV |
7379 | on_rq = tsk->se.on_rq; |
7380 | ||
83b699ed | 7381 | if (on_rq) { |
29f59db3 | 7382 | dequeue_task(rq, tsk, 0); |
83b699ed SV |
7383 | if (unlikely(running)) |
7384 | tsk->sched_class->put_prev_task(rq, tsk); | |
7385 | } | |
29f59db3 | 7386 | |
ce96b5ac | 7387 | set_task_cfs_rq(tsk, task_cpu(tsk)); |
29f59db3 | 7388 | |
83b699ed SV |
7389 | if (on_rq) { |
7390 | if (unlikely(running)) | |
7391 | tsk->sched_class->set_curr_task(rq); | |
7074badb | 7392 | enqueue_task(rq, tsk, 0); |
83b699ed | 7393 | } |
29f59db3 SV |
7394 | |
7395 | done: | |
7396 | task_rq_unlock(rq, &flags); | |
7397 | } | |
7398 | ||
6b2d7700 | 7399 | /* rq->lock to be locked by caller */ |
29f59db3 SV |
7400 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
7401 | { | |
7402 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
7403 | struct rq *rq = cfs_rq->rq; | |
7404 | int on_rq; | |
7405 | ||
6b2d7700 SV |
7406 | if (!shares) |
7407 | shares = MIN_GROUP_SHARES; | |
29f59db3 SV |
7408 | |
7409 | on_rq = se->on_rq; | |
6b2d7700 | 7410 | if (on_rq) { |
29f59db3 | 7411 | dequeue_entity(cfs_rq, se, 0); |
6b2d7700 SV |
7412 | dec_cpu_load(rq, se->load.weight); |
7413 | } | |
29f59db3 SV |
7414 | |
7415 | se->load.weight = shares; | |
7416 | se->load.inv_weight = div64_64((1ULL<<32), shares); | |
7417 | ||
6b2d7700 | 7418 | if (on_rq) { |
29f59db3 | 7419 | enqueue_entity(cfs_rq, se, 0); |
6b2d7700 SV |
7420 | inc_cpu_load(rq, se->load.weight); |
7421 | } | |
29f59db3 SV |
7422 | } |
7423 | ||
4cf86d77 | 7424 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
7425 | { |
7426 | int i; | |
6b2d7700 SV |
7427 | struct cfs_rq *cfs_rq; |
7428 | struct rq *rq; | |
c61935fd | 7429 | |
ec2c507f | 7430 | lock_task_group_list(); |
9b5b7751 | 7431 | if (tg->shares == shares) |
5cb350ba | 7432 | goto done; |
29f59db3 | 7433 | |
6b2d7700 SV |
7434 | if (shares < MIN_GROUP_SHARES) |
7435 | shares = MIN_GROUP_SHARES; | |
7436 | ||
7437 | /* | |
7438 | * Prevent any load balance activity (rebalance_shares, | |
7439 | * load_balance_fair) from referring to this group first, | |
7440 | * by taking it off the rq->leaf_cfs_rq_list on each cpu. | |
7441 | */ | |
7442 | for_each_possible_cpu(i) { | |
7443 | cfs_rq = tg->cfs_rq[i]; | |
7444 | list_del_rcu(&cfs_rq->leaf_cfs_rq_list); | |
7445 | } | |
7446 | ||
7447 | /* wait for any ongoing reference to this group to finish */ | |
7448 | synchronize_sched(); | |
7449 | ||
7450 | /* | |
7451 | * Now we are free to modify the group's share on each cpu | |
7452 | * w/o tripping rebalance_share or load_balance_fair. | |
7453 | */ | |
9b5b7751 | 7454 | tg->shares = shares; |
6b2d7700 SV |
7455 | for_each_possible_cpu(i) { |
7456 | spin_lock_irq(&cpu_rq(i)->lock); | |
9b5b7751 | 7457 | set_se_shares(tg->se[i], shares); |
6b2d7700 SV |
7458 | spin_unlock_irq(&cpu_rq(i)->lock); |
7459 | } | |
29f59db3 | 7460 | |
6b2d7700 SV |
7461 | /* |
7462 | * Enable load balance activity on this group, by inserting it back on | |
7463 | * each cpu's rq->leaf_cfs_rq_list. | |
7464 | */ | |
7465 | for_each_possible_cpu(i) { | |
7466 | rq = cpu_rq(i); | |
7467 | cfs_rq = tg->cfs_rq[i]; | |
7468 | list_add_rcu(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7469 | } | |
5cb350ba | 7470 | done: |
ec2c507f | 7471 | unlock_task_group_list(); |
9b5b7751 | 7472 | return 0; |
29f59db3 SV |
7473 | } |
7474 | ||
5cb350ba DG |
7475 | unsigned long sched_group_shares(struct task_group *tg) |
7476 | { | |
7477 | return tg->shares; | |
7478 | } | |
7479 | ||
3a252015 | 7480 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e SV |
7481 | |
7482 | #ifdef CONFIG_FAIR_CGROUP_SCHED | |
7483 | ||
7484 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 7485 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 7486 | { |
2b01dfe3 PM |
7487 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
7488 | struct task_group, css); | |
68318b8e SV |
7489 | } |
7490 | ||
7491 | static struct cgroup_subsys_state * | |
2b01dfe3 | 7492 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e SV |
7493 | { |
7494 | struct task_group *tg; | |
7495 | ||
2b01dfe3 | 7496 | if (!cgrp->parent) { |
68318b8e | 7497 | /* This is early initialization for the top cgroup */ |
2b01dfe3 | 7498 | init_task_group.css.cgroup = cgrp; |
68318b8e SV |
7499 | return &init_task_group.css; |
7500 | } | |
7501 | ||
7502 | /* we support only 1-level deep hierarchical scheduler atm */ | |
2b01dfe3 | 7503 | if (cgrp->parent->parent) |
68318b8e SV |
7504 | return ERR_PTR(-EINVAL); |
7505 | ||
7506 | tg = sched_create_group(); | |
7507 | if (IS_ERR(tg)) | |
7508 | return ERR_PTR(-ENOMEM); | |
7509 | ||
7510 | /* Bind the cgroup to task_group object we just created */ | |
2b01dfe3 | 7511 | tg->css.cgroup = cgrp; |
68318b8e SV |
7512 | |
7513 | return &tg->css; | |
7514 | } | |
7515 | ||
41a2d6cf IM |
7516 | static void |
7517 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 7518 | { |
2b01dfe3 | 7519 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7520 | |
7521 | sched_destroy_group(tg); | |
7522 | } | |
7523 | ||
41a2d6cf IM |
7524 | static int |
7525 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
7526 | struct task_struct *tsk) | |
68318b8e SV |
7527 | { |
7528 | /* We don't support RT-tasks being in separate groups */ | |
7529 | if (tsk->sched_class != &fair_sched_class) | |
7530 | return -EINVAL; | |
7531 | ||
7532 | return 0; | |
7533 | } | |
7534 | ||
7535 | static void | |
2b01dfe3 | 7536 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
7537 | struct cgroup *old_cont, struct task_struct *tsk) |
7538 | { | |
7539 | sched_move_task(tsk); | |
7540 | } | |
7541 | ||
2b01dfe3 PM |
7542 | static int cpu_shares_write_uint(struct cgroup *cgrp, struct cftype *cftype, |
7543 | u64 shareval) | |
68318b8e | 7544 | { |
2b01dfe3 | 7545 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
7546 | } |
7547 | ||
2b01dfe3 | 7548 | static u64 cpu_shares_read_uint(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 7549 | { |
2b01dfe3 | 7550 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7551 | |
7552 | return (u64) tg->shares; | |
7553 | } | |
7554 | ||
fe5c7cc2 PM |
7555 | static struct cftype cpu_files[] = { |
7556 | { | |
7557 | .name = "shares", | |
7558 | .read_uint = cpu_shares_read_uint, | |
7559 | .write_uint = cpu_shares_write_uint, | |
7560 | }, | |
68318b8e SV |
7561 | }; |
7562 | ||
7563 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
7564 | { | |
fe5c7cc2 | 7565 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
7566 | } |
7567 | ||
7568 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
7569 | .name = "cpu", |
7570 | .create = cpu_cgroup_create, | |
7571 | .destroy = cpu_cgroup_destroy, | |
7572 | .can_attach = cpu_cgroup_can_attach, | |
7573 | .attach = cpu_cgroup_attach, | |
7574 | .populate = cpu_cgroup_populate, | |
7575 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
7576 | .early_init = 1, |
7577 | }; | |
7578 | ||
7579 | #endif /* CONFIG_FAIR_CGROUP_SCHED */ | |
d842de87 SV |
7580 | |
7581 | #ifdef CONFIG_CGROUP_CPUACCT | |
7582 | ||
7583 | /* | |
7584 | * CPU accounting code for task groups. | |
7585 | * | |
7586 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
7587 | * (balbir@in.ibm.com). | |
7588 | */ | |
7589 | ||
7590 | /* track cpu usage of a group of tasks */ | |
7591 | struct cpuacct { | |
7592 | struct cgroup_subsys_state css; | |
7593 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
7594 | u64 *cpuusage; | |
7595 | }; | |
7596 | ||
7597 | struct cgroup_subsys cpuacct_subsys; | |
7598 | ||
7599 | /* return cpu accounting group corresponding to this container */ | |
7600 | static inline struct cpuacct *cgroup_ca(struct cgroup *cont) | |
7601 | { | |
7602 | return container_of(cgroup_subsys_state(cont, cpuacct_subsys_id), | |
7603 | struct cpuacct, css); | |
7604 | } | |
7605 | ||
7606 | /* return cpu accounting group to which this task belongs */ | |
7607 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
7608 | { | |
7609 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
7610 | struct cpuacct, css); | |
7611 | } | |
7612 | ||
7613 | /* create a new cpu accounting group */ | |
7614 | static struct cgroup_subsys_state *cpuacct_create( | |
7615 | struct cgroup_subsys *ss, struct cgroup *cont) | |
7616 | { | |
7617 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
7618 | ||
7619 | if (!ca) | |
7620 | return ERR_PTR(-ENOMEM); | |
7621 | ||
7622 | ca->cpuusage = alloc_percpu(u64); | |
7623 | if (!ca->cpuusage) { | |
7624 | kfree(ca); | |
7625 | return ERR_PTR(-ENOMEM); | |
7626 | } | |
7627 | ||
7628 | return &ca->css; | |
7629 | } | |
7630 | ||
7631 | /* destroy an existing cpu accounting group */ | |
41a2d6cf IM |
7632 | static void |
7633 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cont) | |
d842de87 SV |
7634 | { |
7635 | struct cpuacct *ca = cgroup_ca(cont); | |
7636 | ||
7637 | free_percpu(ca->cpuusage); | |
7638 | kfree(ca); | |
7639 | } | |
7640 | ||
7641 | /* return total cpu usage (in nanoseconds) of a group */ | |
7642 | static u64 cpuusage_read(struct cgroup *cont, struct cftype *cft) | |
7643 | { | |
7644 | struct cpuacct *ca = cgroup_ca(cont); | |
7645 | u64 totalcpuusage = 0; | |
7646 | int i; | |
7647 | ||
7648 | for_each_possible_cpu(i) { | |
7649 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
7650 | ||
7651 | /* | |
7652 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
7653 | * platforms. | |
7654 | */ | |
7655 | spin_lock_irq(&cpu_rq(i)->lock); | |
7656 | totalcpuusage += *cpuusage; | |
7657 | spin_unlock_irq(&cpu_rq(i)->lock); | |
7658 | } | |
7659 | ||
7660 | return totalcpuusage; | |
7661 | } | |
7662 | ||
7663 | static struct cftype files[] = { | |
7664 | { | |
7665 | .name = "usage", | |
7666 | .read_uint = cpuusage_read, | |
7667 | }, | |
7668 | }; | |
7669 | ||
7670 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
7671 | { | |
7672 | return cgroup_add_files(cont, ss, files, ARRAY_SIZE(files)); | |
7673 | } | |
7674 | ||
7675 | /* | |
7676 | * charge this task's execution time to its accounting group. | |
7677 | * | |
7678 | * called with rq->lock held. | |
7679 | */ | |
7680 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
7681 | { | |
7682 | struct cpuacct *ca; | |
7683 | ||
7684 | if (!cpuacct_subsys.active) | |
7685 | return; | |
7686 | ||
7687 | ca = task_ca(tsk); | |
7688 | if (ca) { | |
7689 | u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk)); | |
7690 | ||
7691 | *cpuusage += cputime; | |
7692 | } | |
7693 | } | |
7694 | ||
7695 | struct cgroup_subsys cpuacct_subsys = { | |
7696 | .name = "cpuacct", | |
7697 | .create = cpuacct_create, | |
7698 | .destroy = cpuacct_destroy, | |
7699 | .populate = cpuacct_populate, | |
7700 | .subsys_id = cpuacct_subsys_id, | |
7701 | }; | |
7702 | #endif /* CONFIG_CGROUP_CPUACCT */ |