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% | |
185 | * Bw(C) = 3000/(1000+2000+3000) * 100 = 50% | |
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 | |
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: | |
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 | * | |
362 | * By default the system creates a single root-domain with all cpus as | |
363 | * members (mimicking the global state we have today). | |
364 | */ | |
365 | struct root_domain { | |
366 | atomic_t refcount; | |
367 | cpumask_t span; | |
368 | cpumask_t online; | |
637f5085 | 369 | |
0eab9146 | 370 | /* |
637f5085 GH |
371 | * The "RT overload" flag: it gets set if a CPU has more than |
372 | * one runnable RT task. | |
373 | */ | |
374 | cpumask_t rto_mask; | |
0eab9146 | 375 | atomic_t rto_count; |
57d885fe GH |
376 | }; |
377 | ||
378 | static struct root_domain def_root_domain; | |
379 | ||
380 | #endif | |
381 | ||
1da177e4 LT |
382 | /* |
383 | * This is the main, per-CPU runqueue data structure. | |
384 | * | |
385 | * Locking rule: those places that want to lock multiple runqueues | |
386 | * (such as the load balancing or the thread migration code), lock | |
387 | * acquire operations must be ordered by ascending &runqueue. | |
388 | */ | |
70b97a7f | 389 | struct rq { |
d8016491 IM |
390 | /* runqueue lock: */ |
391 | spinlock_t lock; | |
1da177e4 LT |
392 | |
393 | /* | |
394 | * nr_running and cpu_load should be in the same cacheline because | |
395 | * remote CPUs use both these fields when doing load calculation. | |
396 | */ | |
397 | unsigned long nr_running; | |
6aa645ea IM |
398 | #define CPU_LOAD_IDX_MAX 5 |
399 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 400 | unsigned char idle_at_tick; |
46cb4b7c SS |
401 | #ifdef CONFIG_NO_HZ |
402 | unsigned char in_nohz_recently; | |
403 | #endif | |
d8016491 IM |
404 | /* capture load from *all* tasks on this cpu: */ |
405 | struct load_weight load; | |
6aa645ea IM |
406 | unsigned long nr_load_updates; |
407 | u64 nr_switches; | |
408 | ||
409 | struct cfs_rq cfs; | |
410 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
d8016491 IM |
411 | /* list of leaf cfs_rq on this cpu: */ |
412 | struct list_head leaf_cfs_rq_list; | |
1da177e4 | 413 | #endif |
41a2d6cf | 414 | struct rt_rq rt; |
1da177e4 LT |
415 | |
416 | /* | |
417 | * This is part of a global counter where only the total sum | |
418 | * over all CPUs matters. A task can increase this counter on | |
419 | * one CPU and if it got migrated afterwards it may decrease | |
420 | * it on another CPU. Always updated under the runqueue lock: | |
421 | */ | |
422 | unsigned long nr_uninterruptible; | |
423 | ||
36c8b586 | 424 | struct task_struct *curr, *idle; |
c9819f45 | 425 | unsigned long next_balance; |
1da177e4 | 426 | struct mm_struct *prev_mm; |
6aa645ea | 427 | |
6aa645ea IM |
428 | u64 clock, prev_clock_raw; |
429 | s64 clock_max_delta; | |
430 | ||
431 | unsigned int clock_warps, clock_overflows; | |
2aa44d05 IM |
432 | u64 idle_clock; |
433 | unsigned int clock_deep_idle_events; | |
529c7726 | 434 | u64 tick_timestamp; |
6aa645ea | 435 | |
1da177e4 LT |
436 | atomic_t nr_iowait; |
437 | ||
438 | #ifdef CONFIG_SMP | |
0eab9146 | 439 | struct root_domain *rd; |
1da177e4 LT |
440 | struct sched_domain *sd; |
441 | ||
442 | /* For active balancing */ | |
443 | int active_balance; | |
444 | int push_cpu; | |
d8016491 IM |
445 | /* cpu of this runqueue: */ |
446 | int cpu; | |
1da177e4 | 447 | |
36c8b586 | 448 | struct task_struct *migration_thread; |
1da177e4 LT |
449 | struct list_head migration_queue; |
450 | #endif | |
451 | ||
452 | #ifdef CONFIG_SCHEDSTATS | |
453 | /* latency stats */ | |
454 | struct sched_info rq_sched_info; | |
455 | ||
456 | /* sys_sched_yield() stats */ | |
480b9434 KC |
457 | unsigned int yld_exp_empty; |
458 | unsigned int yld_act_empty; | |
459 | unsigned int yld_both_empty; | |
460 | unsigned int yld_count; | |
1da177e4 LT |
461 | |
462 | /* schedule() stats */ | |
480b9434 KC |
463 | unsigned int sched_switch; |
464 | unsigned int sched_count; | |
465 | unsigned int sched_goidle; | |
1da177e4 LT |
466 | |
467 | /* try_to_wake_up() stats */ | |
480b9434 KC |
468 | unsigned int ttwu_count; |
469 | unsigned int ttwu_local; | |
b8efb561 IM |
470 | |
471 | /* BKL stats */ | |
480b9434 | 472 | unsigned int bkl_count; |
1da177e4 | 473 | #endif |
fcb99371 | 474 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
475 | }; |
476 | ||
f34e3b61 | 477 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 478 | |
dd41f596 IM |
479 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
480 | { | |
481 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
482 | } | |
483 | ||
0a2966b4 CL |
484 | static inline int cpu_of(struct rq *rq) |
485 | { | |
486 | #ifdef CONFIG_SMP | |
487 | return rq->cpu; | |
488 | #else | |
489 | return 0; | |
490 | #endif | |
491 | } | |
492 | ||
20d315d4 | 493 | /* |
b04a0f4c IM |
494 | * Update the per-runqueue clock, as finegrained as the platform can give |
495 | * us, but without assuming monotonicity, etc.: | |
20d315d4 | 496 | */ |
b04a0f4c | 497 | static void __update_rq_clock(struct rq *rq) |
20d315d4 IM |
498 | { |
499 | u64 prev_raw = rq->prev_clock_raw; | |
500 | u64 now = sched_clock(); | |
501 | s64 delta = now - prev_raw; | |
502 | u64 clock = rq->clock; | |
503 | ||
b04a0f4c IM |
504 | #ifdef CONFIG_SCHED_DEBUG |
505 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
506 | #endif | |
20d315d4 IM |
507 | /* |
508 | * Protect against sched_clock() occasionally going backwards: | |
509 | */ | |
510 | if (unlikely(delta < 0)) { | |
511 | clock++; | |
512 | rq->clock_warps++; | |
513 | } else { | |
514 | /* | |
515 | * Catch too large forward jumps too: | |
516 | */ | |
529c7726 IM |
517 | if (unlikely(clock + delta > rq->tick_timestamp + TICK_NSEC)) { |
518 | if (clock < rq->tick_timestamp + TICK_NSEC) | |
519 | clock = rq->tick_timestamp + TICK_NSEC; | |
520 | else | |
521 | clock++; | |
20d315d4 IM |
522 | rq->clock_overflows++; |
523 | } else { | |
524 | if (unlikely(delta > rq->clock_max_delta)) | |
525 | rq->clock_max_delta = delta; | |
526 | clock += delta; | |
527 | } | |
528 | } | |
529 | ||
530 | rq->prev_clock_raw = now; | |
531 | rq->clock = clock; | |
b04a0f4c | 532 | } |
20d315d4 | 533 | |
b04a0f4c IM |
534 | static void update_rq_clock(struct rq *rq) |
535 | { | |
536 | if (likely(smp_processor_id() == cpu_of(rq))) | |
537 | __update_rq_clock(rq); | |
20d315d4 IM |
538 | } |
539 | ||
674311d5 NP |
540 | /* |
541 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 542 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
543 | * |
544 | * The domain tree of any CPU may only be accessed from within | |
545 | * preempt-disabled sections. | |
546 | */ | |
48f24c4d IM |
547 | #define for_each_domain(cpu, __sd) \ |
548 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
549 | |
550 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
551 | #define this_rq() (&__get_cpu_var(runqueues)) | |
552 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
553 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
554 | ||
bf5c91ba IM |
555 | /* |
556 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
557 | */ | |
558 | #ifdef CONFIG_SCHED_DEBUG | |
559 | # define const_debug __read_mostly | |
560 | #else | |
561 | # define const_debug static const | |
562 | #endif | |
563 | ||
564 | /* | |
565 | * Debugging: various feature bits | |
566 | */ | |
567 | enum { | |
bbdba7c0 | 568 | SCHED_FEAT_NEW_FAIR_SLEEPERS = 1, |
9612633a IM |
569 | SCHED_FEAT_WAKEUP_PREEMPT = 2, |
570 | SCHED_FEAT_START_DEBIT = 4, | |
41a2d6cf IM |
571 | SCHED_FEAT_TREE_AVG = 8, |
572 | SCHED_FEAT_APPROX_AVG = 16, | |
bf5c91ba IM |
573 | }; |
574 | ||
575 | const_debug unsigned int sysctl_sched_features = | |
8401f775 | 576 | SCHED_FEAT_NEW_FAIR_SLEEPERS * 1 | |
9612633a | 577 | SCHED_FEAT_WAKEUP_PREEMPT * 1 | |
8401f775 IM |
578 | SCHED_FEAT_START_DEBIT * 1 | |
579 | SCHED_FEAT_TREE_AVG * 0 | | |
9612633a | 580 | SCHED_FEAT_APPROX_AVG * 0; |
bf5c91ba IM |
581 | |
582 | #define sched_feat(x) (sysctl_sched_features & SCHED_FEAT_##x) | |
583 | ||
b82d9fdd PZ |
584 | /* |
585 | * Number of tasks to iterate in a single balance run. | |
586 | * Limited because this is done with IRQs disabled. | |
587 | */ | |
588 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
589 | ||
e436d800 IM |
590 | /* |
591 | * For kernel-internal use: high-speed (but slightly incorrect) per-cpu | |
592 | * clock constructed from sched_clock(): | |
593 | */ | |
594 | unsigned long long cpu_clock(int cpu) | |
595 | { | |
e436d800 IM |
596 | unsigned long long now; |
597 | unsigned long flags; | |
b04a0f4c | 598 | struct rq *rq; |
e436d800 | 599 | |
2cd4d0ea | 600 | local_irq_save(flags); |
b04a0f4c | 601 | rq = cpu_rq(cpu); |
8ced5f69 IM |
602 | /* |
603 | * Only call sched_clock() if the scheduler has already been | |
604 | * initialized (some code might call cpu_clock() very early): | |
605 | */ | |
606 | if (rq->idle) | |
607 | update_rq_clock(rq); | |
b04a0f4c | 608 | now = rq->clock; |
2cd4d0ea | 609 | local_irq_restore(flags); |
e436d800 IM |
610 | |
611 | return now; | |
612 | } | |
a58f6f25 | 613 | EXPORT_SYMBOL_GPL(cpu_clock); |
e436d800 | 614 | |
1da177e4 | 615 | #ifndef prepare_arch_switch |
4866cde0 NP |
616 | # define prepare_arch_switch(next) do { } while (0) |
617 | #endif | |
618 | #ifndef finish_arch_switch | |
619 | # define finish_arch_switch(prev) do { } while (0) | |
620 | #endif | |
621 | ||
051a1d1a DA |
622 | static inline int task_current(struct rq *rq, struct task_struct *p) |
623 | { | |
624 | return rq->curr == p; | |
625 | } | |
626 | ||
4866cde0 | 627 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 628 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 629 | { |
051a1d1a | 630 | return task_current(rq, p); |
4866cde0 NP |
631 | } |
632 | ||
70b97a7f | 633 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
634 | { |
635 | } | |
636 | ||
70b97a7f | 637 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 638 | { |
da04c035 IM |
639 | #ifdef CONFIG_DEBUG_SPINLOCK |
640 | /* this is a valid case when another task releases the spinlock */ | |
641 | rq->lock.owner = current; | |
642 | #endif | |
8a25d5de IM |
643 | /* |
644 | * If we are tracking spinlock dependencies then we have to | |
645 | * fix up the runqueue lock - which gets 'carried over' from | |
646 | * prev into current: | |
647 | */ | |
648 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
649 | ||
4866cde0 NP |
650 | spin_unlock_irq(&rq->lock); |
651 | } | |
652 | ||
653 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 654 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
655 | { |
656 | #ifdef CONFIG_SMP | |
657 | return p->oncpu; | |
658 | #else | |
051a1d1a | 659 | return task_current(rq, p); |
4866cde0 NP |
660 | #endif |
661 | } | |
662 | ||
70b97a7f | 663 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
664 | { |
665 | #ifdef CONFIG_SMP | |
666 | /* | |
667 | * We can optimise this out completely for !SMP, because the | |
668 | * SMP rebalancing from interrupt is the only thing that cares | |
669 | * here. | |
670 | */ | |
671 | next->oncpu = 1; | |
672 | #endif | |
673 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
674 | spin_unlock_irq(&rq->lock); | |
675 | #else | |
676 | spin_unlock(&rq->lock); | |
677 | #endif | |
678 | } | |
679 | ||
70b97a7f | 680 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
681 | { |
682 | #ifdef CONFIG_SMP | |
683 | /* | |
684 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
685 | * We must ensure this doesn't happen until the switch is completely | |
686 | * finished. | |
687 | */ | |
688 | smp_wmb(); | |
689 | prev->oncpu = 0; | |
690 | #endif | |
691 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
692 | local_irq_enable(); | |
1da177e4 | 693 | #endif |
4866cde0 NP |
694 | } |
695 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 696 | |
b29739f9 IM |
697 | /* |
698 | * __task_rq_lock - lock the runqueue a given task resides on. | |
699 | * Must be called interrupts disabled. | |
700 | */ | |
70b97a7f | 701 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
702 | __acquires(rq->lock) |
703 | { | |
3a5c359a AK |
704 | for (;;) { |
705 | struct rq *rq = task_rq(p); | |
706 | spin_lock(&rq->lock); | |
707 | if (likely(rq == task_rq(p))) | |
708 | return rq; | |
b29739f9 | 709 | spin_unlock(&rq->lock); |
b29739f9 | 710 | } |
b29739f9 IM |
711 | } |
712 | ||
1da177e4 LT |
713 | /* |
714 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 715 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
716 | * explicitly disabling preemption. |
717 | */ | |
70b97a7f | 718 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
719 | __acquires(rq->lock) |
720 | { | |
70b97a7f | 721 | struct rq *rq; |
1da177e4 | 722 | |
3a5c359a AK |
723 | for (;;) { |
724 | local_irq_save(*flags); | |
725 | rq = task_rq(p); | |
726 | spin_lock(&rq->lock); | |
727 | if (likely(rq == task_rq(p))) | |
728 | return rq; | |
1da177e4 | 729 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 730 | } |
1da177e4 LT |
731 | } |
732 | ||
a9957449 | 733 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
734 | __releases(rq->lock) |
735 | { | |
736 | spin_unlock(&rq->lock); | |
737 | } | |
738 | ||
70b97a7f | 739 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
740 | __releases(rq->lock) |
741 | { | |
742 | spin_unlock_irqrestore(&rq->lock, *flags); | |
743 | } | |
744 | ||
1da177e4 | 745 | /* |
cc2a73b5 | 746 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 747 | */ |
a9957449 | 748 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
749 | __acquires(rq->lock) |
750 | { | |
70b97a7f | 751 | struct rq *rq; |
1da177e4 LT |
752 | |
753 | local_irq_disable(); | |
754 | rq = this_rq(); | |
755 | spin_lock(&rq->lock); | |
756 | ||
757 | return rq; | |
758 | } | |
759 | ||
1b9f19c2 | 760 | /* |
2aa44d05 | 761 | * We are going deep-idle (irqs are disabled): |
1b9f19c2 | 762 | */ |
2aa44d05 | 763 | void sched_clock_idle_sleep_event(void) |
1b9f19c2 | 764 | { |
2aa44d05 IM |
765 | struct rq *rq = cpu_rq(smp_processor_id()); |
766 | ||
767 | spin_lock(&rq->lock); | |
768 | __update_rq_clock(rq); | |
769 | spin_unlock(&rq->lock); | |
770 | rq->clock_deep_idle_events++; | |
771 | } | |
772 | EXPORT_SYMBOL_GPL(sched_clock_idle_sleep_event); | |
773 | ||
774 | /* | |
775 | * We just idled delta nanoseconds (called with irqs disabled): | |
776 | */ | |
777 | void sched_clock_idle_wakeup_event(u64 delta_ns) | |
778 | { | |
779 | struct rq *rq = cpu_rq(smp_processor_id()); | |
780 | u64 now = sched_clock(); | |
1b9f19c2 | 781 | |
2bacec8c | 782 | touch_softlockup_watchdog(); |
2aa44d05 IM |
783 | rq->idle_clock += delta_ns; |
784 | /* | |
785 | * Override the previous timestamp and ignore all | |
786 | * sched_clock() deltas that occured while we idled, | |
787 | * and use the PM-provided delta_ns to advance the | |
788 | * rq clock: | |
789 | */ | |
790 | spin_lock(&rq->lock); | |
791 | rq->prev_clock_raw = now; | |
792 | rq->clock += delta_ns; | |
793 | spin_unlock(&rq->lock); | |
1b9f19c2 | 794 | } |
2aa44d05 | 795 | EXPORT_SYMBOL_GPL(sched_clock_idle_wakeup_event); |
1b9f19c2 | 796 | |
c24d20db IM |
797 | /* |
798 | * resched_task - mark a task 'to be rescheduled now'. | |
799 | * | |
800 | * On UP this means the setting of the need_resched flag, on SMP it | |
801 | * might also involve a cross-CPU call to trigger the scheduler on | |
802 | * the target CPU. | |
803 | */ | |
804 | #ifdef CONFIG_SMP | |
805 | ||
806 | #ifndef tsk_is_polling | |
807 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
808 | #endif | |
809 | ||
810 | static void resched_task(struct task_struct *p) | |
811 | { | |
812 | int cpu; | |
813 | ||
814 | assert_spin_locked(&task_rq(p)->lock); | |
815 | ||
816 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) | |
817 | return; | |
818 | ||
819 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); | |
820 | ||
821 | cpu = task_cpu(p); | |
822 | if (cpu == smp_processor_id()) | |
823 | return; | |
824 | ||
825 | /* NEED_RESCHED must be visible before we test polling */ | |
826 | smp_mb(); | |
827 | if (!tsk_is_polling(p)) | |
828 | smp_send_reschedule(cpu); | |
829 | } | |
830 | ||
831 | static void resched_cpu(int cpu) | |
832 | { | |
833 | struct rq *rq = cpu_rq(cpu); | |
834 | unsigned long flags; | |
835 | ||
836 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
837 | return; | |
838 | resched_task(cpu_curr(cpu)); | |
839 | spin_unlock_irqrestore(&rq->lock, flags); | |
840 | } | |
841 | #else | |
842 | static inline void resched_task(struct task_struct *p) | |
843 | { | |
844 | assert_spin_locked(&task_rq(p)->lock); | |
845 | set_tsk_need_resched(p); | |
846 | } | |
847 | #endif | |
848 | ||
45bf76df IM |
849 | #if BITS_PER_LONG == 32 |
850 | # define WMULT_CONST (~0UL) | |
851 | #else | |
852 | # define WMULT_CONST (1UL << 32) | |
853 | #endif | |
854 | ||
855 | #define WMULT_SHIFT 32 | |
856 | ||
194081eb IM |
857 | /* |
858 | * Shift right and round: | |
859 | */ | |
cf2ab469 | 860 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 861 | |
cb1c4fc9 | 862 | static unsigned long |
45bf76df IM |
863 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
864 | struct load_weight *lw) | |
865 | { | |
866 | u64 tmp; | |
867 | ||
868 | if (unlikely(!lw->inv_weight)) | |
194081eb | 869 | lw->inv_weight = (WMULT_CONST - lw->weight/2) / lw->weight + 1; |
45bf76df IM |
870 | |
871 | tmp = (u64)delta_exec * weight; | |
872 | /* | |
873 | * Check whether we'd overflow the 64-bit multiplication: | |
874 | */ | |
194081eb | 875 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 876 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
877 | WMULT_SHIFT/2); |
878 | else | |
cf2ab469 | 879 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 880 | |
ecf691da | 881 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
882 | } |
883 | ||
884 | static inline unsigned long | |
885 | calc_delta_fair(unsigned long delta_exec, struct load_weight *lw) | |
886 | { | |
887 | return calc_delta_mine(delta_exec, NICE_0_LOAD, lw); | |
888 | } | |
889 | ||
1091985b | 890 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
891 | { |
892 | lw->weight += inc; | |
45bf76df IM |
893 | } |
894 | ||
1091985b | 895 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
896 | { |
897 | lw->weight -= dec; | |
45bf76df IM |
898 | } |
899 | ||
2dd73a4f PW |
900 | /* |
901 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
902 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
903 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 904 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
905 | * scaled version of the new time slice allocation that they receive on time |
906 | * slice expiry etc. | |
907 | */ | |
908 | ||
dd41f596 IM |
909 | #define WEIGHT_IDLEPRIO 2 |
910 | #define WMULT_IDLEPRIO (1 << 31) | |
911 | ||
912 | /* | |
913 | * Nice levels are multiplicative, with a gentle 10% change for every | |
914 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
915 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
916 | * that remained on nice 0. | |
917 | * | |
918 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
919 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
920 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
921 | * If a task goes up by ~10% and another task goes down by ~10% then | |
922 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
923 | */ |
924 | static const int prio_to_weight[40] = { | |
254753dc IM |
925 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
926 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
927 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
928 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
929 | /* 0 */ 1024, 820, 655, 526, 423, | |
930 | /* 5 */ 335, 272, 215, 172, 137, | |
931 | /* 10 */ 110, 87, 70, 56, 45, | |
932 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
933 | }; |
934 | ||
5714d2de IM |
935 | /* |
936 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
937 | * | |
938 | * In cases where the weight does not change often, we can use the | |
939 | * precalculated inverse to speed up arithmetics by turning divisions | |
940 | * into multiplications: | |
941 | */ | |
dd41f596 | 942 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
943 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
944 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
945 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
946 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
947 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
948 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
949 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
950 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 951 | }; |
2dd73a4f | 952 | |
dd41f596 IM |
953 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
954 | ||
955 | /* | |
956 | * runqueue iterator, to support SMP load-balancing between different | |
957 | * scheduling classes, without having to expose their internal data | |
958 | * structures to the load-balancing proper: | |
959 | */ | |
960 | struct rq_iterator { | |
961 | void *arg; | |
962 | struct task_struct *(*start)(void *); | |
963 | struct task_struct *(*next)(void *); | |
964 | }; | |
965 | ||
e1d1484f PW |
966 | #ifdef CONFIG_SMP |
967 | static unsigned long | |
968 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
969 | unsigned long max_load_move, struct sched_domain *sd, | |
970 | enum cpu_idle_type idle, int *all_pinned, | |
971 | int *this_best_prio, struct rq_iterator *iterator); | |
972 | ||
973 | static int | |
974 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
975 | struct sched_domain *sd, enum cpu_idle_type idle, | |
976 | struct rq_iterator *iterator); | |
e1d1484f | 977 | #endif |
dd41f596 | 978 | |
d842de87 SV |
979 | #ifdef CONFIG_CGROUP_CPUACCT |
980 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
981 | #else | |
982 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
983 | #endif | |
984 | ||
58e2d4ca SV |
985 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
986 | { | |
987 | update_load_add(&rq->load, load); | |
988 | } | |
989 | ||
990 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
991 | { | |
992 | update_load_sub(&rq->load, load); | |
993 | } | |
994 | ||
e7693a36 GH |
995 | #ifdef CONFIG_SMP |
996 | static unsigned long source_load(int cpu, int type); | |
997 | static unsigned long target_load(int cpu, int type); | |
998 | static unsigned long cpu_avg_load_per_task(int cpu); | |
999 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1000 | #endif /* CONFIG_SMP */ | |
1001 | ||
dd41f596 | 1002 | #include "sched_stats.h" |
dd41f596 | 1003 | #include "sched_idletask.c" |
5522d5d5 IM |
1004 | #include "sched_fair.c" |
1005 | #include "sched_rt.c" | |
dd41f596 IM |
1006 | #ifdef CONFIG_SCHED_DEBUG |
1007 | # include "sched_debug.c" | |
1008 | #endif | |
1009 | ||
1010 | #define sched_class_highest (&rt_sched_class) | |
1011 | ||
e5fa2237 | 1012 | static void inc_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
1013 | { |
1014 | rq->nr_running++; | |
9c217245 IM |
1015 | } |
1016 | ||
db53181e | 1017 | static void dec_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
1018 | { |
1019 | rq->nr_running--; | |
9c217245 IM |
1020 | } |
1021 | ||
45bf76df IM |
1022 | static void set_load_weight(struct task_struct *p) |
1023 | { | |
1024 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1025 | p->se.load.weight = prio_to_weight[0] * 2; |
1026 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1027 | return; | |
1028 | } | |
45bf76df | 1029 | |
dd41f596 IM |
1030 | /* |
1031 | * SCHED_IDLE tasks get minimal weight: | |
1032 | */ | |
1033 | if (p->policy == SCHED_IDLE) { | |
1034 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1035 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1036 | return; | |
1037 | } | |
71f8bd46 | 1038 | |
dd41f596 IM |
1039 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1040 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1041 | } |
1042 | ||
8159f87e | 1043 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1044 | { |
dd41f596 | 1045 | sched_info_queued(p); |
fd390f6a | 1046 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1047 | p->se.on_rq = 1; |
71f8bd46 IM |
1048 | } |
1049 | ||
69be72c1 | 1050 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1051 | { |
f02231e5 | 1052 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1053 | p->se.on_rq = 0; |
71f8bd46 IM |
1054 | } |
1055 | ||
14531189 | 1056 | /* |
dd41f596 | 1057 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1058 | */ |
14531189 IM |
1059 | static inline int __normal_prio(struct task_struct *p) |
1060 | { | |
dd41f596 | 1061 | return p->static_prio; |
14531189 IM |
1062 | } |
1063 | ||
b29739f9 IM |
1064 | /* |
1065 | * Calculate the expected normal priority: i.e. priority | |
1066 | * without taking RT-inheritance into account. Might be | |
1067 | * boosted by interactivity modifiers. Changes upon fork, | |
1068 | * setprio syscalls, and whenever the interactivity | |
1069 | * estimator recalculates. | |
1070 | */ | |
36c8b586 | 1071 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1072 | { |
1073 | int prio; | |
1074 | ||
e05606d3 | 1075 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1076 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1077 | else | |
1078 | prio = __normal_prio(p); | |
1079 | return prio; | |
1080 | } | |
1081 | ||
1082 | /* | |
1083 | * Calculate the current priority, i.e. the priority | |
1084 | * taken into account by the scheduler. This value might | |
1085 | * be boosted by RT tasks, or might be boosted by | |
1086 | * interactivity modifiers. Will be RT if the task got | |
1087 | * RT-boosted. If not then it returns p->normal_prio. | |
1088 | */ | |
36c8b586 | 1089 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1090 | { |
1091 | p->normal_prio = normal_prio(p); | |
1092 | /* | |
1093 | * If we are RT tasks or we were boosted to RT priority, | |
1094 | * keep the priority unchanged. Otherwise, update priority | |
1095 | * to the normal priority: | |
1096 | */ | |
1097 | if (!rt_prio(p->prio)) | |
1098 | return p->normal_prio; | |
1099 | return p->prio; | |
1100 | } | |
1101 | ||
1da177e4 | 1102 | /* |
dd41f596 | 1103 | * activate_task - move a task to the runqueue. |
1da177e4 | 1104 | */ |
dd41f596 | 1105 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1106 | { |
dd41f596 IM |
1107 | if (p->state == TASK_UNINTERRUPTIBLE) |
1108 | rq->nr_uninterruptible--; | |
1da177e4 | 1109 | |
8159f87e | 1110 | enqueue_task(rq, p, wakeup); |
e5fa2237 | 1111 | inc_nr_running(p, rq); |
1da177e4 LT |
1112 | } |
1113 | ||
1da177e4 LT |
1114 | /* |
1115 | * deactivate_task - remove a task from the runqueue. | |
1116 | */ | |
2e1cb74a | 1117 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1118 | { |
dd41f596 IM |
1119 | if (p->state == TASK_UNINTERRUPTIBLE) |
1120 | rq->nr_uninterruptible++; | |
1121 | ||
69be72c1 | 1122 | dequeue_task(rq, p, sleep); |
db53181e | 1123 | dec_nr_running(p, rq); |
1da177e4 LT |
1124 | } |
1125 | ||
1da177e4 LT |
1126 | /** |
1127 | * task_curr - is this task currently executing on a CPU? | |
1128 | * @p: the task in question. | |
1129 | */ | |
36c8b586 | 1130 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1131 | { |
1132 | return cpu_curr(task_cpu(p)) == p; | |
1133 | } | |
1134 | ||
2dd73a4f PW |
1135 | /* Used instead of source_load when we know the type == 0 */ |
1136 | unsigned long weighted_cpuload(const int cpu) | |
1137 | { | |
495eca49 | 1138 | return cpu_rq(cpu)->load.weight; |
dd41f596 IM |
1139 | } |
1140 | ||
1141 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1142 | { | |
ce96b5ac | 1143 | set_task_cfs_rq(p, cpu); |
dd41f596 | 1144 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1145 | /* |
1146 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1147 | * successfuly executed on another CPU. We must ensure that updates of | |
1148 | * per-task data have been completed by this moment. | |
1149 | */ | |
1150 | smp_wmb(); | |
dd41f596 | 1151 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1152 | #endif |
2dd73a4f PW |
1153 | } |
1154 | ||
1da177e4 | 1155 | #ifdef CONFIG_SMP |
c65cc870 | 1156 | |
cc367732 IM |
1157 | /* |
1158 | * Is this task likely cache-hot: | |
1159 | */ | |
e7693a36 | 1160 | static int |
cc367732 IM |
1161 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1162 | { | |
1163 | s64 delta; | |
1164 | ||
1165 | if (p->sched_class != &fair_sched_class) | |
1166 | return 0; | |
1167 | ||
6bc1665b IM |
1168 | if (sysctl_sched_migration_cost == -1) |
1169 | return 1; | |
1170 | if (sysctl_sched_migration_cost == 0) | |
1171 | return 0; | |
1172 | ||
cc367732 IM |
1173 | delta = now - p->se.exec_start; |
1174 | ||
1175 | return delta < (s64)sysctl_sched_migration_cost; | |
1176 | } | |
1177 | ||
1178 | ||
dd41f596 | 1179 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1180 | { |
dd41f596 IM |
1181 | int old_cpu = task_cpu(p); |
1182 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1183 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1184 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1185 | u64 clock_offset; |
dd41f596 IM |
1186 | |
1187 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1188 | |
1189 | #ifdef CONFIG_SCHEDSTATS | |
1190 | if (p->se.wait_start) | |
1191 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1192 | if (p->se.sleep_start) |
1193 | p->se.sleep_start -= clock_offset; | |
1194 | if (p->se.block_start) | |
1195 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1196 | if (old_cpu != new_cpu) { |
1197 | schedstat_inc(p, se.nr_migrations); | |
1198 | if (task_hot(p, old_rq->clock, NULL)) | |
1199 | schedstat_inc(p, se.nr_forced2_migrations); | |
1200 | } | |
6cfb0d5d | 1201 | #endif |
2830cf8c SV |
1202 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1203 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1204 | |
1205 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1206 | } |
1207 | ||
70b97a7f | 1208 | struct migration_req { |
1da177e4 | 1209 | struct list_head list; |
1da177e4 | 1210 | |
36c8b586 | 1211 | struct task_struct *task; |
1da177e4 LT |
1212 | int dest_cpu; |
1213 | ||
1da177e4 | 1214 | struct completion done; |
70b97a7f | 1215 | }; |
1da177e4 LT |
1216 | |
1217 | /* | |
1218 | * The task's runqueue lock must be held. | |
1219 | * Returns true if you have to wait for migration thread. | |
1220 | */ | |
36c8b586 | 1221 | static int |
70b97a7f | 1222 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1223 | { |
70b97a7f | 1224 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1225 | |
1226 | /* | |
1227 | * If the task is not on a runqueue (and not running), then | |
1228 | * it is sufficient to simply update the task's cpu field. | |
1229 | */ | |
dd41f596 | 1230 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1231 | set_task_cpu(p, dest_cpu); |
1232 | return 0; | |
1233 | } | |
1234 | ||
1235 | init_completion(&req->done); | |
1da177e4 LT |
1236 | req->task = p; |
1237 | req->dest_cpu = dest_cpu; | |
1238 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1239 | |
1da177e4 LT |
1240 | return 1; |
1241 | } | |
1242 | ||
1243 | /* | |
1244 | * wait_task_inactive - wait for a thread to unschedule. | |
1245 | * | |
1246 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1247 | * else this function might spin for a *long* time. This function can't | |
1248 | * be called with interrupts off, or it may introduce deadlock with | |
1249 | * smp_call_function() if an IPI is sent by the same process we are | |
1250 | * waiting to become inactive. | |
1251 | */ | |
36c8b586 | 1252 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1253 | { |
1254 | unsigned long flags; | |
dd41f596 | 1255 | int running, on_rq; |
70b97a7f | 1256 | struct rq *rq; |
1da177e4 | 1257 | |
3a5c359a AK |
1258 | for (;;) { |
1259 | /* | |
1260 | * We do the initial early heuristics without holding | |
1261 | * any task-queue locks at all. We'll only try to get | |
1262 | * the runqueue lock when things look like they will | |
1263 | * work out! | |
1264 | */ | |
1265 | rq = task_rq(p); | |
fa490cfd | 1266 | |
3a5c359a AK |
1267 | /* |
1268 | * If the task is actively running on another CPU | |
1269 | * still, just relax and busy-wait without holding | |
1270 | * any locks. | |
1271 | * | |
1272 | * NOTE! Since we don't hold any locks, it's not | |
1273 | * even sure that "rq" stays as the right runqueue! | |
1274 | * But we don't care, since "task_running()" will | |
1275 | * return false if the runqueue has changed and p | |
1276 | * is actually now running somewhere else! | |
1277 | */ | |
1278 | while (task_running(rq, p)) | |
1279 | cpu_relax(); | |
fa490cfd | 1280 | |
3a5c359a AK |
1281 | /* |
1282 | * Ok, time to look more closely! We need the rq | |
1283 | * lock now, to be *sure*. If we're wrong, we'll | |
1284 | * just go back and repeat. | |
1285 | */ | |
1286 | rq = task_rq_lock(p, &flags); | |
1287 | running = task_running(rq, p); | |
1288 | on_rq = p->se.on_rq; | |
1289 | task_rq_unlock(rq, &flags); | |
fa490cfd | 1290 | |
3a5c359a AK |
1291 | /* |
1292 | * Was it really running after all now that we | |
1293 | * checked with the proper locks actually held? | |
1294 | * | |
1295 | * Oops. Go back and try again.. | |
1296 | */ | |
1297 | if (unlikely(running)) { | |
1298 | cpu_relax(); | |
1299 | continue; | |
1300 | } | |
fa490cfd | 1301 | |
3a5c359a AK |
1302 | /* |
1303 | * It's not enough that it's not actively running, | |
1304 | * it must be off the runqueue _entirely_, and not | |
1305 | * preempted! | |
1306 | * | |
1307 | * So if it wa still runnable (but just not actively | |
1308 | * running right now), it's preempted, and we should | |
1309 | * yield - it could be a while. | |
1310 | */ | |
1311 | if (unlikely(on_rq)) { | |
1312 | schedule_timeout_uninterruptible(1); | |
1313 | continue; | |
1314 | } | |
fa490cfd | 1315 | |
3a5c359a AK |
1316 | /* |
1317 | * Ahh, all good. It wasn't running, and it wasn't | |
1318 | * runnable, which means that it will never become | |
1319 | * running in the future either. We're all done! | |
1320 | */ | |
1321 | break; | |
1322 | } | |
1da177e4 LT |
1323 | } |
1324 | ||
1325 | /*** | |
1326 | * kick_process - kick a running thread to enter/exit the kernel | |
1327 | * @p: the to-be-kicked thread | |
1328 | * | |
1329 | * Cause a process which is running on another CPU to enter | |
1330 | * kernel-mode, without any delay. (to get signals handled.) | |
1331 | * | |
1332 | * NOTE: this function doesnt have to take the runqueue lock, | |
1333 | * because all it wants to ensure is that the remote task enters | |
1334 | * the kernel. If the IPI races and the task has been migrated | |
1335 | * to another CPU then no harm is done and the purpose has been | |
1336 | * achieved as well. | |
1337 | */ | |
36c8b586 | 1338 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1339 | { |
1340 | int cpu; | |
1341 | ||
1342 | preempt_disable(); | |
1343 | cpu = task_cpu(p); | |
1344 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1345 | smp_send_reschedule(cpu); | |
1346 | preempt_enable(); | |
1347 | } | |
1348 | ||
1349 | /* | |
2dd73a4f PW |
1350 | * Return a low guess at the load of a migration-source cpu weighted |
1351 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1352 | * |
1353 | * We want to under-estimate the load of migration sources, to | |
1354 | * balance conservatively. | |
1355 | */ | |
a9957449 | 1356 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 1357 | { |
70b97a7f | 1358 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1359 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1360 | |
3b0bd9bc | 1361 | if (type == 0) |
dd41f596 | 1362 | return total; |
b910472d | 1363 | |
dd41f596 | 1364 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
1365 | } |
1366 | ||
1367 | /* | |
2dd73a4f PW |
1368 | * Return a high guess at the load of a migration-target cpu weighted |
1369 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1370 | */ |
a9957449 | 1371 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 1372 | { |
70b97a7f | 1373 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1374 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1375 | |
7897986b | 1376 | if (type == 0) |
dd41f596 | 1377 | return total; |
3b0bd9bc | 1378 | |
dd41f596 | 1379 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
1380 | } |
1381 | ||
1382 | /* | |
1383 | * Return the average load per task on the cpu's run queue | |
1384 | */ | |
e7693a36 | 1385 | static unsigned long cpu_avg_load_per_task(int cpu) |
2dd73a4f | 1386 | { |
70b97a7f | 1387 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1388 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f PW |
1389 | unsigned long n = rq->nr_running; |
1390 | ||
dd41f596 | 1391 | return n ? total / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1392 | } |
1393 | ||
147cbb4b NP |
1394 | /* |
1395 | * find_idlest_group finds and returns the least busy CPU group within the | |
1396 | * domain. | |
1397 | */ | |
1398 | static struct sched_group * | |
1399 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1400 | { | |
1401 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1402 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1403 | int load_idx = sd->forkexec_idx; | |
1404 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1405 | ||
1406 | do { | |
1407 | unsigned long load, avg_load; | |
1408 | int local_group; | |
1409 | int i; | |
1410 | ||
da5a5522 BD |
1411 | /* Skip over this group if it has no CPUs allowed */ |
1412 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 1413 | continue; |
da5a5522 | 1414 | |
147cbb4b | 1415 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1416 | |
1417 | /* Tally up the load of all CPUs in the group */ | |
1418 | avg_load = 0; | |
1419 | ||
1420 | for_each_cpu_mask(i, group->cpumask) { | |
1421 | /* Bias balancing toward cpus of our domain */ | |
1422 | if (local_group) | |
1423 | load = source_load(i, load_idx); | |
1424 | else | |
1425 | load = target_load(i, load_idx); | |
1426 | ||
1427 | avg_load += load; | |
1428 | } | |
1429 | ||
1430 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1431 | avg_load = sg_div_cpu_power(group, |
1432 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1433 | |
1434 | if (local_group) { | |
1435 | this_load = avg_load; | |
1436 | this = group; | |
1437 | } else if (avg_load < min_load) { | |
1438 | min_load = avg_load; | |
1439 | idlest = group; | |
1440 | } | |
3a5c359a | 1441 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
1442 | |
1443 | if (!idlest || 100*this_load < imbalance*min_load) | |
1444 | return NULL; | |
1445 | return idlest; | |
1446 | } | |
1447 | ||
1448 | /* | |
0feaece9 | 1449 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1450 | */ |
95cdf3b7 IM |
1451 | static int |
1452 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) | |
147cbb4b | 1453 | { |
da5a5522 | 1454 | cpumask_t tmp; |
147cbb4b NP |
1455 | unsigned long load, min_load = ULONG_MAX; |
1456 | int idlest = -1; | |
1457 | int i; | |
1458 | ||
da5a5522 BD |
1459 | /* Traverse only the allowed CPUs */ |
1460 | cpus_and(tmp, group->cpumask, p->cpus_allowed); | |
1461 | ||
1462 | for_each_cpu_mask(i, tmp) { | |
2dd73a4f | 1463 | load = weighted_cpuload(i); |
147cbb4b NP |
1464 | |
1465 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1466 | min_load = load; | |
1467 | idlest = i; | |
1468 | } | |
1469 | } | |
1470 | ||
1471 | return idlest; | |
1472 | } | |
1473 | ||
476d139c NP |
1474 | /* |
1475 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1476 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1477 | * SD_BALANCE_EXEC. | |
1478 | * | |
1479 | * Balance, ie. select the least loaded group. | |
1480 | * | |
1481 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1482 | * | |
1483 | * preempt must be disabled. | |
1484 | */ | |
1485 | static int sched_balance_self(int cpu, int flag) | |
1486 | { | |
1487 | struct task_struct *t = current; | |
1488 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1489 | |
c96d145e | 1490 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
1491 | /* |
1492 | * If power savings logic is enabled for a domain, stop there. | |
1493 | */ | |
5c45bf27 SS |
1494 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
1495 | break; | |
476d139c NP |
1496 | if (tmp->flags & flag) |
1497 | sd = tmp; | |
c96d145e | 1498 | } |
476d139c NP |
1499 | |
1500 | while (sd) { | |
1501 | cpumask_t span; | |
1502 | struct sched_group *group; | |
1a848870 SS |
1503 | int new_cpu, weight; |
1504 | ||
1505 | if (!(sd->flags & flag)) { | |
1506 | sd = sd->child; | |
1507 | continue; | |
1508 | } | |
476d139c NP |
1509 | |
1510 | span = sd->span; | |
1511 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1512 | if (!group) { |
1513 | sd = sd->child; | |
1514 | continue; | |
1515 | } | |
476d139c | 1516 | |
da5a5522 | 1517 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
1518 | if (new_cpu == -1 || new_cpu == cpu) { |
1519 | /* Now try balancing at a lower domain level of cpu */ | |
1520 | sd = sd->child; | |
1521 | continue; | |
1522 | } | |
476d139c | 1523 | |
1a848870 | 1524 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1525 | cpu = new_cpu; |
476d139c NP |
1526 | sd = NULL; |
1527 | weight = cpus_weight(span); | |
1528 | for_each_domain(cpu, tmp) { | |
1529 | if (weight <= cpus_weight(tmp->span)) | |
1530 | break; | |
1531 | if (tmp->flags & flag) | |
1532 | sd = tmp; | |
1533 | } | |
1534 | /* while loop will break here if sd == NULL */ | |
1535 | } | |
1536 | ||
1537 | return cpu; | |
1538 | } | |
1539 | ||
1540 | #endif /* CONFIG_SMP */ | |
1da177e4 | 1541 | |
1da177e4 LT |
1542 | /*** |
1543 | * try_to_wake_up - wake up a thread | |
1544 | * @p: the to-be-woken-up thread | |
1545 | * @state: the mask of task states that can be woken | |
1546 | * @sync: do a synchronous wakeup? | |
1547 | * | |
1548 | * Put it on the run-queue if it's not already there. The "current" | |
1549 | * thread is always on the run-queue (except when the actual | |
1550 | * re-schedule is in progress), and as such you're allowed to do | |
1551 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1552 | * runnable without the overhead of this. | |
1553 | * | |
1554 | * returns failure only if the task is already active. | |
1555 | */ | |
36c8b586 | 1556 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 1557 | { |
cc367732 | 1558 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
1559 | unsigned long flags; |
1560 | long old_state; | |
70b97a7f | 1561 | struct rq *rq; |
1da177e4 LT |
1562 | |
1563 | rq = task_rq_lock(p, &flags); | |
1564 | old_state = p->state; | |
1565 | if (!(old_state & state)) | |
1566 | goto out; | |
1567 | ||
dd41f596 | 1568 | if (p->se.on_rq) |
1da177e4 LT |
1569 | goto out_running; |
1570 | ||
1571 | cpu = task_cpu(p); | |
cc367732 | 1572 | orig_cpu = cpu; |
1da177e4 LT |
1573 | this_cpu = smp_processor_id(); |
1574 | ||
1575 | #ifdef CONFIG_SMP | |
1576 | if (unlikely(task_running(rq, p))) | |
1577 | goto out_activate; | |
1578 | ||
5d2f5a61 DA |
1579 | cpu = p->sched_class->select_task_rq(p, sync); |
1580 | if (cpu != orig_cpu) { | |
1581 | set_task_cpu(p, cpu); | |
1da177e4 LT |
1582 | task_rq_unlock(rq, &flags); |
1583 | /* might preempt at this point */ | |
1584 | rq = task_rq_lock(p, &flags); | |
1585 | old_state = p->state; | |
1586 | if (!(old_state & state)) | |
1587 | goto out; | |
dd41f596 | 1588 | if (p->se.on_rq) |
1da177e4 LT |
1589 | goto out_running; |
1590 | ||
1591 | this_cpu = smp_processor_id(); | |
1592 | cpu = task_cpu(p); | |
1593 | } | |
1594 | ||
e7693a36 GH |
1595 | #ifdef CONFIG_SCHEDSTATS |
1596 | schedstat_inc(rq, ttwu_count); | |
1597 | if (cpu == this_cpu) | |
1598 | schedstat_inc(rq, ttwu_local); | |
1599 | else { | |
1600 | struct sched_domain *sd; | |
1601 | for_each_domain(this_cpu, sd) { | |
1602 | if (cpu_isset(cpu, sd->span)) { | |
1603 | schedstat_inc(sd, ttwu_wake_remote); | |
1604 | break; | |
1605 | } | |
1606 | } | |
1607 | } | |
e7693a36 GH |
1608 | #endif |
1609 | ||
1da177e4 LT |
1610 | out_activate: |
1611 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
1612 | schedstat_inc(p, se.nr_wakeups); |
1613 | if (sync) | |
1614 | schedstat_inc(p, se.nr_wakeups_sync); | |
1615 | if (orig_cpu != cpu) | |
1616 | schedstat_inc(p, se.nr_wakeups_migrate); | |
1617 | if (cpu == this_cpu) | |
1618 | schedstat_inc(p, se.nr_wakeups_local); | |
1619 | else | |
1620 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 1621 | update_rq_clock(rq); |
dd41f596 | 1622 | activate_task(rq, p, 1); |
9c63d9c0 | 1623 | check_preempt_curr(rq, p); |
1da177e4 LT |
1624 | success = 1; |
1625 | ||
1626 | out_running: | |
1627 | p->state = TASK_RUNNING; | |
9a897c5a SR |
1628 | #ifdef CONFIG_SMP |
1629 | if (p->sched_class->task_wake_up) | |
1630 | p->sched_class->task_wake_up(rq, p); | |
1631 | #endif | |
1da177e4 LT |
1632 | out: |
1633 | task_rq_unlock(rq, &flags); | |
1634 | ||
1635 | return success; | |
1636 | } | |
1637 | ||
36c8b586 | 1638 | int fastcall wake_up_process(struct task_struct *p) |
1da177e4 LT |
1639 | { |
1640 | return try_to_wake_up(p, TASK_STOPPED | TASK_TRACED | | |
1641 | TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE, 0); | |
1642 | } | |
1da177e4 LT |
1643 | EXPORT_SYMBOL(wake_up_process); |
1644 | ||
36c8b586 | 1645 | int fastcall wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1646 | { |
1647 | return try_to_wake_up(p, state, 0); | |
1648 | } | |
1649 | ||
1da177e4 LT |
1650 | /* |
1651 | * Perform scheduler related setup for a newly forked process p. | |
1652 | * p is forked by current. | |
dd41f596 IM |
1653 | * |
1654 | * __sched_fork() is basic setup used by init_idle() too: | |
1655 | */ | |
1656 | static void __sched_fork(struct task_struct *p) | |
1657 | { | |
dd41f596 IM |
1658 | p->se.exec_start = 0; |
1659 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1660 | p->se.prev_sum_exec_runtime = 0; |
6cfb0d5d IM |
1661 | |
1662 | #ifdef CONFIG_SCHEDSTATS | |
1663 | p->se.wait_start = 0; | |
dd41f596 IM |
1664 | p->se.sum_sleep_runtime = 0; |
1665 | p->se.sleep_start = 0; | |
dd41f596 IM |
1666 | p->se.block_start = 0; |
1667 | p->se.sleep_max = 0; | |
1668 | p->se.block_max = 0; | |
1669 | p->se.exec_max = 0; | |
eba1ed4b | 1670 | p->se.slice_max = 0; |
dd41f596 | 1671 | p->se.wait_max = 0; |
6cfb0d5d | 1672 | #endif |
476d139c | 1673 | |
dd41f596 IM |
1674 | INIT_LIST_HEAD(&p->run_list); |
1675 | p->se.on_rq = 0; | |
476d139c | 1676 | |
e107be36 AK |
1677 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1678 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1679 | #endif | |
1680 | ||
1da177e4 LT |
1681 | /* |
1682 | * We mark the process as running here, but have not actually | |
1683 | * inserted it onto the runqueue yet. This guarantees that | |
1684 | * nobody will actually run it, and a signal or other external | |
1685 | * event cannot wake it up and insert it on the runqueue either. | |
1686 | */ | |
1687 | p->state = TASK_RUNNING; | |
dd41f596 IM |
1688 | } |
1689 | ||
1690 | /* | |
1691 | * fork()/clone()-time setup: | |
1692 | */ | |
1693 | void sched_fork(struct task_struct *p, int clone_flags) | |
1694 | { | |
1695 | int cpu = get_cpu(); | |
1696 | ||
1697 | __sched_fork(p); | |
1698 | ||
1699 | #ifdef CONFIG_SMP | |
1700 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
1701 | #endif | |
02e4bac2 | 1702 | set_task_cpu(p, cpu); |
b29739f9 IM |
1703 | |
1704 | /* | |
1705 | * Make sure we do not leak PI boosting priority to the child: | |
1706 | */ | |
1707 | p->prio = current->normal_prio; | |
2ddbf952 HS |
1708 | if (!rt_prio(p->prio)) |
1709 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1710 | |
52f17b6c | 1711 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1712 | if (likely(sched_info_on())) |
52f17b6c | 1713 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1714 | #endif |
d6077cb8 | 1715 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
1716 | p->oncpu = 0; |
1717 | #endif | |
1da177e4 | 1718 | #ifdef CONFIG_PREEMPT |
4866cde0 | 1719 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1720 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1721 | #endif |
476d139c | 1722 | put_cpu(); |
1da177e4 LT |
1723 | } |
1724 | ||
1725 | /* | |
1726 | * wake_up_new_task - wake up a newly created task for the first time. | |
1727 | * | |
1728 | * This function will do some initial scheduler statistics housekeeping | |
1729 | * that must be done for every newly created context, then puts the task | |
1730 | * on the runqueue and wakes it. | |
1731 | */ | |
36c8b586 | 1732 | void fastcall wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
1733 | { |
1734 | unsigned long flags; | |
dd41f596 | 1735 | struct rq *rq; |
1da177e4 LT |
1736 | |
1737 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 1738 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 1739 | update_rq_clock(rq); |
1da177e4 LT |
1740 | |
1741 | p->prio = effective_prio(p); | |
1742 | ||
b9dca1e0 | 1743 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 1744 | activate_task(rq, p, 0); |
1da177e4 | 1745 | } else { |
1da177e4 | 1746 | /* |
dd41f596 IM |
1747 | * Let the scheduling class do new task startup |
1748 | * management (if any): | |
1da177e4 | 1749 | */ |
ee0827d8 | 1750 | p->sched_class->task_new(rq, p); |
e5fa2237 | 1751 | inc_nr_running(p, rq); |
1da177e4 | 1752 | } |
dd41f596 | 1753 | check_preempt_curr(rq, p); |
9a897c5a SR |
1754 | #ifdef CONFIG_SMP |
1755 | if (p->sched_class->task_wake_up) | |
1756 | p->sched_class->task_wake_up(rq, p); | |
1757 | #endif | |
dd41f596 | 1758 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
1759 | } |
1760 | ||
e107be36 AK |
1761 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1762 | ||
1763 | /** | |
421cee29 RD |
1764 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
1765 | * @notifier: notifier struct to register | |
e107be36 AK |
1766 | */ |
1767 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1768 | { | |
1769 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1770 | } | |
1771 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1772 | ||
1773 | /** | |
1774 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1775 | * @notifier: notifier struct to unregister |
e107be36 AK |
1776 | * |
1777 | * This is safe to call from within a preemption notifier. | |
1778 | */ | |
1779 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1780 | { | |
1781 | hlist_del(¬ifier->link); | |
1782 | } | |
1783 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1784 | ||
1785 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1786 | { | |
1787 | struct preempt_notifier *notifier; | |
1788 | struct hlist_node *node; | |
1789 | ||
1790 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1791 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
1792 | } | |
1793 | ||
1794 | static void | |
1795 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1796 | struct task_struct *next) | |
1797 | { | |
1798 | struct preempt_notifier *notifier; | |
1799 | struct hlist_node *node; | |
1800 | ||
1801 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1802 | notifier->ops->sched_out(notifier, next); | |
1803 | } | |
1804 | ||
1805 | #else | |
1806 | ||
1807 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1808 | { | |
1809 | } | |
1810 | ||
1811 | static void | |
1812 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1813 | struct task_struct *next) | |
1814 | { | |
1815 | } | |
1816 | ||
1817 | #endif | |
1818 | ||
4866cde0 NP |
1819 | /** |
1820 | * prepare_task_switch - prepare to switch tasks | |
1821 | * @rq: the runqueue preparing to switch | |
421cee29 | 1822 | * @prev: the current task that is being switched out |
4866cde0 NP |
1823 | * @next: the task we are going to switch to. |
1824 | * | |
1825 | * This is called with the rq lock held and interrupts off. It must | |
1826 | * be paired with a subsequent finish_task_switch after the context | |
1827 | * switch. | |
1828 | * | |
1829 | * prepare_task_switch sets up locking and calls architecture specific | |
1830 | * hooks. | |
1831 | */ | |
e107be36 AK |
1832 | static inline void |
1833 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1834 | struct task_struct *next) | |
4866cde0 | 1835 | { |
e107be36 | 1836 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1837 | prepare_lock_switch(rq, next); |
1838 | prepare_arch_switch(next); | |
1839 | } | |
1840 | ||
1da177e4 LT |
1841 | /** |
1842 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1843 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1844 | * @prev: the thread we just switched away from. |
1845 | * | |
4866cde0 NP |
1846 | * finish_task_switch must be called after the context switch, paired |
1847 | * with a prepare_task_switch call before the context switch. | |
1848 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1849 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1850 | * |
1851 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 1852 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
1853 | * with the lock held can cause deadlocks; see schedule() for |
1854 | * details.) | |
1855 | */ | |
a9957449 | 1856 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1857 | __releases(rq->lock) |
1858 | { | |
1da177e4 | 1859 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1860 | long prev_state; |
1da177e4 LT |
1861 | |
1862 | rq->prev_mm = NULL; | |
1863 | ||
1864 | /* | |
1865 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1866 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1867 | * schedule one last time. The schedule call will never return, and |
1868 | * the scheduled task must drop that reference. | |
c394cc9f | 1869 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1870 | * still held, otherwise prev could be scheduled on another cpu, die |
1871 | * there before we look at prev->state, and then the reference would | |
1872 | * be dropped twice. | |
1873 | * Manfred Spraul <manfred@colorfullife.com> | |
1874 | */ | |
55a101f8 | 1875 | prev_state = prev->state; |
4866cde0 NP |
1876 | finish_arch_switch(prev); |
1877 | finish_lock_switch(rq, prev); | |
9a897c5a SR |
1878 | #ifdef CONFIG_SMP |
1879 | if (current->sched_class->post_schedule) | |
1880 | current->sched_class->post_schedule(rq); | |
1881 | #endif | |
e8fa1362 | 1882 | |
e107be36 | 1883 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1884 | if (mm) |
1885 | mmdrop(mm); | |
c394cc9f | 1886 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1887 | /* |
1888 | * Remove function-return probe instances associated with this | |
1889 | * task and put them back on the free list. | |
9761eea8 | 1890 | */ |
c6fd91f0 | 1891 | kprobe_flush_task(prev); |
1da177e4 | 1892 | put_task_struct(prev); |
c6fd91f0 | 1893 | } |
1da177e4 LT |
1894 | } |
1895 | ||
1896 | /** | |
1897 | * schedule_tail - first thing a freshly forked thread must call. | |
1898 | * @prev: the thread we just switched away from. | |
1899 | */ | |
36c8b586 | 1900 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1901 | __releases(rq->lock) |
1902 | { | |
70b97a7f IM |
1903 | struct rq *rq = this_rq(); |
1904 | ||
4866cde0 NP |
1905 | finish_task_switch(rq, prev); |
1906 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
1907 | /* In this case, finish_task_switch does not reenable preemption */ | |
1908 | preempt_enable(); | |
1909 | #endif | |
1da177e4 | 1910 | if (current->set_child_tid) |
b488893a | 1911 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
1912 | } |
1913 | ||
1914 | /* | |
1915 | * context_switch - switch to the new MM and the new | |
1916 | * thread's register state. | |
1917 | */ | |
dd41f596 | 1918 | static inline void |
70b97a7f | 1919 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1920 | struct task_struct *next) |
1da177e4 | 1921 | { |
dd41f596 | 1922 | struct mm_struct *mm, *oldmm; |
1da177e4 | 1923 | |
e107be36 | 1924 | prepare_task_switch(rq, prev, next); |
dd41f596 IM |
1925 | mm = next->mm; |
1926 | oldmm = prev->active_mm; | |
9226d125 ZA |
1927 | /* |
1928 | * For paravirt, this is coupled with an exit in switch_to to | |
1929 | * combine the page table reload and the switch backend into | |
1930 | * one hypercall. | |
1931 | */ | |
1932 | arch_enter_lazy_cpu_mode(); | |
1933 | ||
dd41f596 | 1934 | if (unlikely(!mm)) { |
1da177e4 LT |
1935 | next->active_mm = oldmm; |
1936 | atomic_inc(&oldmm->mm_count); | |
1937 | enter_lazy_tlb(oldmm, next); | |
1938 | } else | |
1939 | switch_mm(oldmm, mm, next); | |
1940 | ||
dd41f596 | 1941 | if (unlikely(!prev->mm)) { |
1da177e4 | 1942 | prev->active_mm = NULL; |
1da177e4 LT |
1943 | rq->prev_mm = oldmm; |
1944 | } | |
3a5f5e48 IM |
1945 | /* |
1946 | * Since the runqueue lock will be released by the next | |
1947 | * task (which is an invalid locking op but in the case | |
1948 | * of the scheduler it's an obvious special-case), so we | |
1949 | * do an early lockdep release here: | |
1950 | */ | |
1951 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 1952 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 1953 | #endif |
1da177e4 LT |
1954 | |
1955 | /* Here we just switch the register state and the stack. */ | |
1956 | switch_to(prev, next, prev); | |
1957 | ||
dd41f596 IM |
1958 | barrier(); |
1959 | /* | |
1960 | * this_rq must be evaluated again because prev may have moved | |
1961 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
1962 | * frame will be invalid. | |
1963 | */ | |
1964 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
1965 | } |
1966 | ||
1967 | /* | |
1968 | * nr_running, nr_uninterruptible and nr_context_switches: | |
1969 | * | |
1970 | * externally visible scheduler statistics: current number of runnable | |
1971 | * threads, current number of uninterruptible-sleeping threads, total | |
1972 | * number of context switches performed since bootup. | |
1973 | */ | |
1974 | unsigned long nr_running(void) | |
1975 | { | |
1976 | unsigned long i, sum = 0; | |
1977 | ||
1978 | for_each_online_cpu(i) | |
1979 | sum += cpu_rq(i)->nr_running; | |
1980 | ||
1981 | return sum; | |
1982 | } | |
1983 | ||
1984 | unsigned long nr_uninterruptible(void) | |
1985 | { | |
1986 | unsigned long i, sum = 0; | |
1987 | ||
0a945022 | 1988 | for_each_possible_cpu(i) |
1da177e4 LT |
1989 | sum += cpu_rq(i)->nr_uninterruptible; |
1990 | ||
1991 | /* | |
1992 | * Since we read the counters lockless, it might be slightly | |
1993 | * inaccurate. Do not allow it to go below zero though: | |
1994 | */ | |
1995 | if (unlikely((long)sum < 0)) | |
1996 | sum = 0; | |
1997 | ||
1998 | return sum; | |
1999 | } | |
2000 | ||
2001 | unsigned long long nr_context_switches(void) | |
2002 | { | |
cc94abfc SR |
2003 | int i; |
2004 | unsigned long long sum = 0; | |
1da177e4 | 2005 | |
0a945022 | 2006 | for_each_possible_cpu(i) |
1da177e4 LT |
2007 | sum += cpu_rq(i)->nr_switches; |
2008 | ||
2009 | return sum; | |
2010 | } | |
2011 | ||
2012 | unsigned long nr_iowait(void) | |
2013 | { | |
2014 | unsigned long i, sum = 0; | |
2015 | ||
0a945022 | 2016 | for_each_possible_cpu(i) |
1da177e4 LT |
2017 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2018 | ||
2019 | return sum; | |
2020 | } | |
2021 | ||
db1b1fef JS |
2022 | unsigned long nr_active(void) |
2023 | { | |
2024 | unsigned long i, running = 0, uninterruptible = 0; | |
2025 | ||
2026 | for_each_online_cpu(i) { | |
2027 | running += cpu_rq(i)->nr_running; | |
2028 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2029 | } | |
2030 | ||
2031 | if (unlikely((long)uninterruptible < 0)) | |
2032 | uninterruptible = 0; | |
2033 | ||
2034 | return running + uninterruptible; | |
2035 | } | |
2036 | ||
48f24c4d | 2037 | /* |
dd41f596 IM |
2038 | * Update rq->cpu_load[] statistics. This function is usually called every |
2039 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2040 | */ |
dd41f596 | 2041 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2042 | { |
495eca49 | 2043 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2044 | int i, scale; |
2045 | ||
2046 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2047 | |
2048 | /* Update our load: */ | |
2049 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2050 | unsigned long old_load, new_load; | |
2051 | ||
2052 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2053 | ||
2054 | old_load = this_rq->cpu_load[i]; | |
2055 | new_load = this_load; | |
a25707f3 IM |
2056 | /* |
2057 | * Round up the averaging division if load is increasing. This | |
2058 | * prevents us from getting stuck on 9 if the load is 10, for | |
2059 | * example. | |
2060 | */ | |
2061 | if (new_load > old_load) | |
2062 | new_load += scale-1; | |
dd41f596 IM |
2063 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2064 | } | |
48f24c4d IM |
2065 | } |
2066 | ||
dd41f596 IM |
2067 | #ifdef CONFIG_SMP |
2068 | ||
1da177e4 LT |
2069 | /* |
2070 | * double_rq_lock - safely lock two runqueues | |
2071 | * | |
2072 | * Note this does not disable interrupts like task_rq_lock, | |
2073 | * you need to do so manually before calling. | |
2074 | */ | |
70b97a7f | 2075 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2076 | __acquires(rq1->lock) |
2077 | __acquires(rq2->lock) | |
2078 | { | |
054b9108 | 2079 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2080 | if (rq1 == rq2) { |
2081 | spin_lock(&rq1->lock); | |
2082 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2083 | } else { | |
c96d145e | 2084 | if (rq1 < rq2) { |
1da177e4 LT |
2085 | spin_lock(&rq1->lock); |
2086 | spin_lock(&rq2->lock); | |
2087 | } else { | |
2088 | spin_lock(&rq2->lock); | |
2089 | spin_lock(&rq1->lock); | |
2090 | } | |
2091 | } | |
6e82a3be IM |
2092 | update_rq_clock(rq1); |
2093 | update_rq_clock(rq2); | |
1da177e4 LT |
2094 | } |
2095 | ||
2096 | /* | |
2097 | * double_rq_unlock - safely unlock two runqueues | |
2098 | * | |
2099 | * Note this does not restore interrupts like task_rq_unlock, | |
2100 | * you need to do so manually after calling. | |
2101 | */ | |
70b97a7f | 2102 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2103 | __releases(rq1->lock) |
2104 | __releases(rq2->lock) | |
2105 | { | |
2106 | spin_unlock(&rq1->lock); | |
2107 | if (rq1 != rq2) | |
2108 | spin_unlock(&rq2->lock); | |
2109 | else | |
2110 | __release(rq2->lock); | |
2111 | } | |
2112 | ||
2113 | /* | |
2114 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2115 | */ | |
e8fa1362 | 2116 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2117 | __releases(this_rq->lock) |
2118 | __acquires(busiest->lock) | |
2119 | __acquires(this_rq->lock) | |
2120 | { | |
e8fa1362 SR |
2121 | int ret = 0; |
2122 | ||
054b9108 KK |
2123 | if (unlikely(!irqs_disabled())) { |
2124 | /* printk() doesn't work good under rq->lock */ | |
2125 | spin_unlock(&this_rq->lock); | |
2126 | BUG_ON(1); | |
2127 | } | |
1da177e4 | 2128 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2129 | if (busiest < this_rq) { |
1da177e4 LT |
2130 | spin_unlock(&this_rq->lock); |
2131 | spin_lock(&busiest->lock); | |
2132 | spin_lock(&this_rq->lock); | |
e8fa1362 | 2133 | ret = 1; |
1da177e4 LT |
2134 | } else |
2135 | spin_lock(&busiest->lock); | |
2136 | } | |
e8fa1362 | 2137 | return ret; |
1da177e4 LT |
2138 | } |
2139 | ||
1da177e4 LT |
2140 | /* |
2141 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2142 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2143 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2144 | * the cpu_allowed mask is restored. |
2145 | */ | |
36c8b586 | 2146 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2147 | { |
70b97a7f | 2148 | struct migration_req req; |
1da177e4 | 2149 | unsigned long flags; |
70b97a7f | 2150 | struct rq *rq; |
1da177e4 LT |
2151 | |
2152 | rq = task_rq_lock(p, &flags); | |
2153 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2154 | || unlikely(cpu_is_offline(dest_cpu))) | |
2155 | goto out; | |
2156 | ||
2157 | /* force the process onto the specified CPU */ | |
2158 | if (migrate_task(p, dest_cpu, &req)) { | |
2159 | /* Need to wait for migration thread (might exit: take ref). */ | |
2160 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2161 | |
1da177e4 LT |
2162 | get_task_struct(mt); |
2163 | task_rq_unlock(rq, &flags); | |
2164 | wake_up_process(mt); | |
2165 | put_task_struct(mt); | |
2166 | wait_for_completion(&req.done); | |
36c8b586 | 2167 | |
1da177e4 LT |
2168 | return; |
2169 | } | |
2170 | out: | |
2171 | task_rq_unlock(rq, &flags); | |
2172 | } | |
2173 | ||
2174 | /* | |
476d139c NP |
2175 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2176 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2177 | */ |
2178 | void sched_exec(void) | |
2179 | { | |
1da177e4 | 2180 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2181 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2182 | put_cpu(); |
476d139c NP |
2183 | if (new_cpu != this_cpu) |
2184 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2185 | } |
2186 | ||
2187 | /* | |
2188 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2189 | * Both runqueues must be locked. | |
2190 | */ | |
dd41f596 IM |
2191 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2192 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2193 | { |
2e1cb74a | 2194 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2195 | set_task_cpu(p, this_cpu); |
dd41f596 | 2196 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2197 | /* |
2198 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2199 | * to be always true for them. | |
2200 | */ | |
dd41f596 | 2201 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2202 | } |
2203 | ||
2204 | /* | |
2205 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2206 | */ | |
858119e1 | 2207 | static |
70b97a7f | 2208 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2209 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2210 | int *all_pinned) |
1da177e4 LT |
2211 | { |
2212 | /* | |
2213 | * We do not migrate tasks that are: | |
2214 | * 1) running (obviously), or | |
2215 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2216 | * 3) are cache-hot on their current CPU. | |
2217 | */ | |
cc367732 IM |
2218 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2219 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2220 | return 0; |
cc367732 | 2221 | } |
81026794 NP |
2222 | *all_pinned = 0; |
2223 | ||
cc367732 IM |
2224 | if (task_running(rq, p)) { |
2225 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2226 | return 0; |
cc367732 | 2227 | } |
1da177e4 | 2228 | |
da84d961 IM |
2229 | /* |
2230 | * Aggressive migration if: | |
2231 | * 1) task is cache cold, or | |
2232 | * 2) too many balance attempts have failed. | |
2233 | */ | |
2234 | ||
6bc1665b IM |
2235 | if (!task_hot(p, rq->clock, sd) || |
2236 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2237 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2238 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2239 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2240 | schedstat_inc(p, se.nr_forced_migrations); |
2241 | } | |
da84d961 IM |
2242 | #endif |
2243 | return 1; | |
2244 | } | |
2245 | ||
cc367732 IM |
2246 | if (task_hot(p, rq->clock, sd)) { |
2247 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2248 | return 0; |
cc367732 | 2249 | } |
1da177e4 LT |
2250 | return 1; |
2251 | } | |
2252 | ||
e1d1484f PW |
2253 | static unsigned long |
2254 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2255 | unsigned long max_load_move, struct sched_domain *sd, | |
2256 | enum cpu_idle_type idle, int *all_pinned, | |
2257 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2258 | { |
b82d9fdd | 2259 | int loops = 0, pulled = 0, pinned = 0, skip_for_load; |
dd41f596 IM |
2260 | struct task_struct *p; |
2261 | long rem_load_move = max_load_move; | |
1da177e4 | 2262 | |
e1d1484f | 2263 | if (max_load_move == 0) |
1da177e4 LT |
2264 | goto out; |
2265 | ||
81026794 NP |
2266 | pinned = 1; |
2267 | ||
1da177e4 | 2268 | /* |
dd41f596 | 2269 | * Start the load-balancing iterator: |
1da177e4 | 2270 | */ |
dd41f596 IM |
2271 | p = iterator->start(iterator->arg); |
2272 | next: | |
b82d9fdd | 2273 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2274 | goto out; |
50ddd969 | 2275 | /* |
b82d9fdd | 2276 | * To help distribute high priority tasks across CPUs we don't |
50ddd969 PW |
2277 | * skip a task if it will be the highest priority task (i.e. smallest |
2278 | * prio value) on its new queue regardless of its load weight | |
2279 | */ | |
dd41f596 IM |
2280 | skip_for_load = (p->se.load.weight >> 1) > rem_load_move + |
2281 | SCHED_LOAD_SCALE_FUZZ; | |
a4ac01c3 | 2282 | if ((skip_for_load && p->prio >= *this_best_prio) || |
dd41f596 | 2283 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2284 | p = iterator->next(iterator->arg); |
2285 | goto next; | |
1da177e4 LT |
2286 | } |
2287 | ||
dd41f596 | 2288 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2289 | pulled++; |
dd41f596 | 2290 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2291 | |
2dd73a4f | 2292 | /* |
b82d9fdd | 2293 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 2294 | */ |
e1d1484f | 2295 | if (rem_load_move > 0) { |
a4ac01c3 PW |
2296 | if (p->prio < *this_best_prio) |
2297 | *this_best_prio = p->prio; | |
dd41f596 IM |
2298 | p = iterator->next(iterator->arg); |
2299 | goto next; | |
1da177e4 LT |
2300 | } |
2301 | out: | |
2302 | /* | |
e1d1484f | 2303 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
2304 | * so we can safely collect pull_task() stats here rather than |
2305 | * inside pull_task(). | |
2306 | */ | |
2307 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2308 | |
2309 | if (all_pinned) | |
2310 | *all_pinned = pinned; | |
e1d1484f PW |
2311 | |
2312 | return max_load_move - rem_load_move; | |
1da177e4 LT |
2313 | } |
2314 | ||
dd41f596 | 2315 | /* |
43010659 PW |
2316 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2317 | * this_rq, as part of a balancing operation within domain "sd". | |
2318 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2319 | * |
2320 | * Called with both runqueues locked. | |
2321 | */ | |
2322 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2323 | unsigned long max_load_move, |
dd41f596 IM |
2324 | struct sched_domain *sd, enum cpu_idle_type idle, |
2325 | int *all_pinned) | |
2326 | { | |
5522d5d5 | 2327 | const struct sched_class *class = sched_class_highest; |
43010659 | 2328 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2329 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
2330 | |
2331 | do { | |
43010659 PW |
2332 | total_load_moved += |
2333 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 2334 | max_load_move - total_load_moved, |
a4ac01c3 | 2335 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 2336 | class = class->next; |
43010659 | 2337 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 2338 | |
43010659 PW |
2339 | return total_load_moved > 0; |
2340 | } | |
2341 | ||
e1d1484f PW |
2342 | static int |
2343 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2344 | struct sched_domain *sd, enum cpu_idle_type idle, | |
2345 | struct rq_iterator *iterator) | |
2346 | { | |
2347 | struct task_struct *p = iterator->start(iterator->arg); | |
2348 | int pinned = 0; | |
2349 | ||
2350 | while (p) { | |
2351 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
2352 | pull_task(busiest, p, this_rq, this_cpu); | |
2353 | /* | |
2354 | * Right now, this is only the second place pull_task() | |
2355 | * is called, so we can safely collect pull_task() | |
2356 | * stats here rather than inside pull_task(). | |
2357 | */ | |
2358 | schedstat_inc(sd, lb_gained[idle]); | |
2359 | ||
2360 | return 1; | |
2361 | } | |
2362 | p = iterator->next(iterator->arg); | |
2363 | } | |
2364 | ||
2365 | return 0; | |
2366 | } | |
2367 | ||
43010659 PW |
2368 | /* |
2369 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
2370 | * part of active balancing operations within "domain". | |
2371 | * Returns 1 if successful and 0 otherwise. | |
2372 | * | |
2373 | * Called with both runqueues locked. | |
2374 | */ | |
2375 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2376 | struct sched_domain *sd, enum cpu_idle_type idle) | |
2377 | { | |
5522d5d5 | 2378 | const struct sched_class *class; |
43010659 PW |
2379 | |
2380 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 2381 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
2382 | return 1; |
2383 | ||
2384 | return 0; | |
dd41f596 IM |
2385 | } |
2386 | ||
1da177e4 LT |
2387 | /* |
2388 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2389 | * domain. It calculates and returns the amount of weighted load which |
2390 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2391 | */ |
2392 | static struct sched_group * | |
2393 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 IM |
2394 | unsigned long *imbalance, enum cpu_idle_type idle, |
2395 | int *sd_idle, cpumask_t *cpus, int *balance) | |
1da177e4 LT |
2396 | { |
2397 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2398 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2399 | unsigned long max_pull; |
2dd73a4f PW |
2400 | unsigned long busiest_load_per_task, busiest_nr_running; |
2401 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 2402 | int load_idx, group_imb = 0; |
5c45bf27 SS |
2403 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2404 | int power_savings_balance = 1; | |
2405 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2406 | unsigned long min_nr_running = ULONG_MAX; | |
2407 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2408 | #endif | |
1da177e4 LT |
2409 | |
2410 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2411 | busiest_load_per_task = busiest_nr_running = 0; |
2412 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2413 | if (idle == CPU_NOT_IDLE) |
7897986b | 2414 | load_idx = sd->busy_idx; |
d15bcfdb | 2415 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2416 | load_idx = sd->newidle_idx; |
2417 | else | |
2418 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2419 | |
2420 | do { | |
908a7c1b | 2421 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
2422 | int local_group; |
2423 | int i; | |
908a7c1b | 2424 | int __group_imb = 0; |
783609c6 | 2425 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2426 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2427 | |
2428 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2429 | ||
783609c6 SS |
2430 | if (local_group) |
2431 | balance_cpu = first_cpu(group->cpumask); | |
2432 | ||
1da177e4 | 2433 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2434 | sum_weighted_load = sum_nr_running = avg_load = 0; |
908a7c1b KC |
2435 | max_cpu_load = 0; |
2436 | min_cpu_load = ~0UL; | |
1da177e4 LT |
2437 | |
2438 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2439 | struct rq *rq; |
2440 | ||
2441 | if (!cpu_isset(i, *cpus)) | |
2442 | continue; | |
2443 | ||
2444 | rq = cpu_rq(i); | |
2dd73a4f | 2445 | |
9439aab8 | 2446 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
2447 | *sd_idle = 0; |
2448 | ||
1da177e4 | 2449 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2450 | if (local_group) { |
2451 | if (idle_cpu(i) && !first_idle_cpu) { | |
2452 | first_idle_cpu = 1; | |
2453 | balance_cpu = i; | |
2454 | } | |
2455 | ||
a2000572 | 2456 | load = target_load(i, load_idx); |
908a7c1b | 2457 | } else { |
a2000572 | 2458 | load = source_load(i, load_idx); |
908a7c1b KC |
2459 | if (load > max_cpu_load) |
2460 | max_cpu_load = load; | |
2461 | if (min_cpu_load > load) | |
2462 | min_cpu_load = load; | |
2463 | } | |
1da177e4 LT |
2464 | |
2465 | avg_load += load; | |
2dd73a4f | 2466 | sum_nr_running += rq->nr_running; |
dd41f596 | 2467 | sum_weighted_load += weighted_cpuload(i); |
1da177e4 LT |
2468 | } |
2469 | ||
783609c6 SS |
2470 | /* |
2471 | * First idle cpu or the first cpu(busiest) in this sched group | |
2472 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
2473 | * domains. In the newly idle case, we will allow all the cpu's |
2474 | * to do the newly idle load balance. | |
783609c6 | 2475 | */ |
9439aab8 SS |
2476 | if (idle != CPU_NEWLY_IDLE && local_group && |
2477 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
2478 | *balance = 0; |
2479 | goto ret; | |
2480 | } | |
2481 | ||
1da177e4 | 2482 | total_load += avg_load; |
5517d86b | 2483 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2484 | |
2485 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2486 | avg_load = sg_div_cpu_power(group, |
2487 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2488 | |
908a7c1b KC |
2489 | if ((max_cpu_load - min_cpu_load) > SCHED_LOAD_SCALE) |
2490 | __group_imb = 1; | |
2491 | ||
5517d86b | 2492 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2493 | |
1da177e4 LT |
2494 | if (local_group) { |
2495 | this_load = avg_load; | |
2496 | this = group; | |
2dd73a4f PW |
2497 | this_nr_running = sum_nr_running; |
2498 | this_load_per_task = sum_weighted_load; | |
2499 | } else if (avg_load > max_load && | |
908a7c1b | 2500 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
2501 | max_load = avg_load; |
2502 | busiest = group; | |
2dd73a4f PW |
2503 | busiest_nr_running = sum_nr_running; |
2504 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 2505 | group_imb = __group_imb; |
1da177e4 | 2506 | } |
5c45bf27 SS |
2507 | |
2508 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2509 | /* | |
2510 | * Busy processors will not participate in power savings | |
2511 | * balance. | |
2512 | */ | |
dd41f596 IM |
2513 | if (idle == CPU_NOT_IDLE || |
2514 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
2515 | goto group_next; | |
5c45bf27 SS |
2516 | |
2517 | /* | |
2518 | * If the local group is idle or completely loaded | |
2519 | * no need to do power savings balance at this domain | |
2520 | */ | |
2521 | if (local_group && (this_nr_running >= group_capacity || | |
2522 | !this_nr_running)) | |
2523 | power_savings_balance = 0; | |
2524 | ||
dd41f596 | 2525 | /* |
5c45bf27 SS |
2526 | * If a group is already running at full capacity or idle, |
2527 | * don't include that group in power savings calculations | |
dd41f596 IM |
2528 | */ |
2529 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 2530 | || !sum_nr_running) |
dd41f596 | 2531 | goto group_next; |
5c45bf27 | 2532 | |
dd41f596 | 2533 | /* |
5c45bf27 | 2534 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
2535 | * This is the group from where we need to pick up the load |
2536 | * for saving power | |
2537 | */ | |
2538 | if ((sum_nr_running < min_nr_running) || | |
2539 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
2540 | first_cpu(group->cpumask) < |
2541 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
2542 | group_min = group; |
2543 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
2544 | min_load_per_task = sum_weighted_load / |
2545 | sum_nr_running; | |
dd41f596 | 2546 | } |
5c45bf27 | 2547 | |
dd41f596 | 2548 | /* |
5c45bf27 | 2549 | * Calculate the group which is almost near its |
dd41f596 IM |
2550 | * capacity but still has some space to pick up some load |
2551 | * from other group and save more power | |
2552 | */ | |
2553 | if (sum_nr_running <= group_capacity - 1) { | |
2554 | if (sum_nr_running > leader_nr_running || | |
2555 | (sum_nr_running == leader_nr_running && | |
2556 | first_cpu(group->cpumask) > | |
2557 | first_cpu(group_leader->cpumask))) { | |
2558 | group_leader = group; | |
2559 | leader_nr_running = sum_nr_running; | |
2560 | } | |
48f24c4d | 2561 | } |
5c45bf27 SS |
2562 | group_next: |
2563 | #endif | |
1da177e4 LT |
2564 | group = group->next; |
2565 | } while (group != sd->groups); | |
2566 | ||
2dd73a4f | 2567 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
2568 | goto out_balanced; |
2569 | ||
2570 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
2571 | ||
2572 | if (this_load >= avg_load || | |
2573 | 100*max_load <= sd->imbalance_pct*this_load) | |
2574 | goto out_balanced; | |
2575 | ||
2dd73a4f | 2576 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
2577 | if (group_imb) |
2578 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
2579 | ||
1da177e4 LT |
2580 | /* |
2581 | * We're trying to get all the cpus to the average_load, so we don't | |
2582 | * want to push ourselves above the average load, nor do we wish to | |
2583 | * reduce the max loaded cpu below the average load, as either of these | |
2584 | * actions would just result in more rebalancing later, and ping-pong | |
2585 | * tasks around. Thus we look for the minimum possible imbalance. | |
2586 | * Negative imbalances (*we* are more loaded than anyone else) will | |
2587 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 2588 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
2589 | * appear as very large values with unsigned longs. |
2590 | */ | |
2dd73a4f PW |
2591 | if (max_load <= busiest_load_per_task) |
2592 | goto out_balanced; | |
2593 | ||
2594 | /* | |
2595 | * In the presence of smp nice balancing, certain scenarios can have | |
2596 | * max load less than avg load(as we skip the groups at or below | |
2597 | * its cpu_power, while calculating max_load..) | |
2598 | */ | |
2599 | if (max_load < avg_load) { | |
2600 | *imbalance = 0; | |
2601 | goto small_imbalance; | |
2602 | } | |
0c117f1b SS |
2603 | |
2604 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 2605 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 2606 | |
1da177e4 | 2607 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
2608 | *imbalance = min(max_pull * busiest->__cpu_power, |
2609 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
2610 | / SCHED_LOAD_SCALE; |
2611 | ||
2dd73a4f PW |
2612 | /* |
2613 | * if *imbalance is less than the average load per runnable task | |
2614 | * there is no gaurantee that any tasks will be moved so we'll have | |
2615 | * a think about bumping its value to force at least one task to be | |
2616 | * moved | |
2617 | */ | |
7fd0d2dd | 2618 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 2619 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
2620 | unsigned int imbn; |
2621 | ||
2622 | small_imbalance: | |
2623 | pwr_move = pwr_now = 0; | |
2624 | imbn = 2; | |
2625 | if (this_nr_running) { | |
2626 | this_load_per_task /= this_nr_running; | |
2627 | if (busiest_load_per_task > this_load_per_task) | |
2628 | imbn = 1; | |
2629 | } else | |
2630 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 2631 | |
dd41f596 IM |
2632 | if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >= |
2633 | busiest_load_per_task * imbn) { | |
2dd73a4f | 2634 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2635 | return busiest; |
2636 | } | |
2637 | ||
2638 | /* | |
2639 | * OK, we don't have enough imbalance to justify moving tasks, | |
2640 | * however we may be able to increase total CPU power used by | |
2641 | * moving them. | |
2642 | */ | |
2643 | ||
5517d86b ED |
2644 | pwr_now += busiest->__cpu_power * |
2645 | min(busiest_load_per_task, max_load); | |
2646 | pwr_now += this->__cpu_power * | |
2647 | min(this_load_per_task, this_load); | |
1da177e4 LT |
2648 | pwr_now /= SCHED_LOAD_SCALE; |
2649 | ||
2650 | /* Amount of load we'd subtract */ | |
5517d86b ED |
2651 | tmp = sg_div_cpu_power(busiest, |
2652 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 2653 | if (max_load > tmp) |
5517d86b | 2654 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 2655 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
2656 | |
2657 | /* Amount of load we'd add */ | |
5517d86b | 2658 | if (max_load * busiest->__cpu_power < |
33859f7f | 2659 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
2660 | tmp = sg_div_cpu_power(this, |
2661 | max_load * busiest->__cpu_power); | |
1da177e4 | 2662 | else |
5517d86b ED |
2663 | tmp = sg_div_cpu_power(this, |
2664 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
2665 | pwr_move += this->__cpu_power * | |
2666 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
2667 | pwr_move /= SCHED_LOAD_SCALE; |
2668 | ||
2669 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
2670 | if (pwr_move > pwr_now) |
2671 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
2672 | } |
2673 | ||
1da177e4 LT |
2674 | return busiest; |
2675 | ||
2676 | out_balanced: | |
5c45bf27 | 2677 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 2678 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 2679 | goto ret; |
1da177e4 | 2680 | |
5c45bf27 SS |
2681 | if (this == group_leader && group_leader != group_min) { |
2682 | *imbalance = min_load_per_task; | |
2683 | return group_min; | |
2684 | } | |
5c45bf27 | 2685 | #endif |
783609c6 | 2686 | ret: |
1da177e4 LT |
2687 | *imbalance = 0; |
2688 | return NULL; | |
2689 | } | |
2690 | ||
2691 | /* | |
2692 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
2693 | */ | |
70b97a7f | 2694 | static struct rq * |
d15bcfdb | 2695 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
0a2966b4 | 2696 | unsigned long imbalance, cpumask_t *cpus) |
1da177e4 | 2697 | { |
70b97a7f | 2698 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 2699 | unsigned long max_load = 0; |
1da177e4 LT |
2700 | int i; |
2701 | ||
2702 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 2703 | unsigned long wl; |
0a2966b4 CL |
2704 | |
2705 | if (!cpu_isset(i, *cpus)) | |
2706 | continue; | |
2707 | ||
48f24c4d | 2708 | rq = cpu_rq(i); |
dd41f596 | 2709 | wl = weighted_cpuload(i); |
2dd73a4f | 2710 | |
dd41f596 | 2711 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 2712 | continue; |
1da177e4 | 2713 | |
dd41f596 IM |
2714 | if (wl > max_load) { |
2715 | max_load = wl; | |
48f24c4d | 2716 | busiest = rq; |
1da177e4 LT |
2717 | } |
2718 | } | |
2719 | ||
2720 | return busiest; | |
2721 | } | |
2722 | ||
77391d71 NP |
2723 | /* |
2724 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
2725 | * so long as it is large enough. | |
2726 | */ | |
2727 | #define MAX_PINNED_INTERVAL 512 | |
2728 | ||
1da177e4 LT |
2729 | /* |
2730 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2731 | * tasks if there is an imbalance. | |
1da177e4 | 2732 | */ |
70b97a7f | 2733 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 2734 | struct sched_domain *sd, enum cpu_idle_type idle, |
783609c6 | 2735 | int *balance) |
1da177e4 | 2736 | { |
43010659 | 2737 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 2738 | struct sched_group *group; |
1da177e4 | 2739 | unsigned long imbalance; |
70b97a7f | 2740 | struct rq *busiest; |
0a2966b4 | 2741 | cpumask_t cpus = CPU_MASK_ALL; |
fe2eea3f | 2742 | unsigned long flags; |
5969fe06 | 2743 | |
89c4710e SS |
2744 | /* |
2745 | * When power savings policy is enabled for the parent domain, idle | |
2746 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 2747 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 2748 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 2749 | */ |
d15bcfdb | 2750 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2751 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2752 | sd_idle = 1; |
1da177e4 | 2753 | |
2d72376b | 2754 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 2755 | |
0a2966b4 CL |
2756 | redo: |
2757 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
783609c6 SS |
2758 | &cpus, balance); |
2759 | ||
06066714 | 2760 | if (*balance == 0) |
783609c6 | 2761 | goto out_balanced; |
783609c6 | 2762 | |
1da177e4 LT |
2763 | if (!group) { |
2764 | schedstat_inc(sd, lb_nobusyg[idle]); | |
2765 | goto out_balanced; | |
2766 | } | |
2767 | ||
0a2966b4 | 2768 | busiest = find_busiest_queue(group, idle, imbalance, &cpus); |
1da177e4 LT |
2769 | if (!busiest) { |
2770 | schedstat_inc(sd, lb_nobusyq[idle]); | |
2771 | goto out_balanced; | |
2772 | } | |
2773 | ||
db935dbd | 2774 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
2775 | |
2776 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
2777 | ||
43010659 | 2778 | ld_moved = 0; |
1da177e4 LT |
2779 | if (busiest->nr_running > 1) { |
2780 | /* | |
2781 | * Attempt to move tasks. If find_busiest_group has found | |
2782 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 2783 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
2784 | * correctly treated as an imbalance. |
2785 | */ | |
fe2eea3f | 2786 | local_irq_save(flags); |
e17224bf | 2787 | double_rq_lock(this_rq, busiest); |
43010659 | 2788 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 2789 | imbalance, sd, idle, &all_pinned); |
e17224bf | 2790 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 2791 | local_irq_restore(flags); |
81026794 | 2792 | |
46cb4b7c SS |
2793 | /* |
2794 | * some other cpu did the load balance for us. | |
2795 | */ | |
43010659 | 2796 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
2797 | resched_cpu(this_cpu); |
2798 | ||
81026794 | 2799 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 CL |
2800 | if (unlikely(all_pinned)) { |
2801 | cpu_clear(cpu_of(busiest), cpus); | |
2802 | if (!cpus_empty(cpus)) | |
2803 | goto redo; | |
81026794 | 2804 | goto out_balanced; |
0a2966b4 | 2805 | } |
1da177e4 | 2806 | } |
81026794 | 2807 | |
43010659 | 2808 | if (!ld_moved) { |
1da177e4 LT |
2809 | schedstat_inc(sd, lb_failed[idle]); |
2810 | sd->nr_balance_failed++; | |
2811 | ||
2812 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 2813 | |
fe2eea3f | 2814 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
2815 | |
2816 | /* don't kick the migration_thread, if the curr | |
2817 | * task on busiest cpu can't be moved to this_cpu | |
2818 | */ | |
2819 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 2820 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
2821 | all_pinned = 1; |
2822 | goto out_one_pinned; | |
2823 | } | |
2824 | ||
1da177e4 LT |
2825 | if (!busiest->active_balance) { |
2826 | busiest->active_balance = 1; | |
2827 | busiest->push_cpu = this_cpu; | |
81026794 | 2828 | active_balance = 1; |
1da177e4 | 2829 | } |
fe2eea3f | 2830 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 2831 | if (active_balance) |
1da177e4 LT |
2832 | wake_up_process(busiest->migration_thread); |
2833 | ||
2834 | /* | |
2835 | * We've kicked active balancing, reset the failure | |
2836 | * counter. | |
2837 | */ | |
39507451 | 2838 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 2839 | } |
81026794 | 2840 | } else |
1da177e4 LT |
2841 | sd->nr_balance_failed = 0; |
2842 | ||
81026794 | 2843 | if (likely(!active_balance)) { |
1da177e4 LT |
2844 | /* We were unbalanced, so reset the balancing interval */ |
2845 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
2846 | } else { |
2847 | /* | |
2848 | * If we've begun active balancing, start to back off. This | |
2849 | * case may not be covered by the all_pinned logic if there | |
2850 | * is only 1 task on the busy runqueue (because we don't call | |
2851 | * move_tasks). | |
2852 | */ | |
2853 | if (sd->balance_interval < sd->max_interval) | |
2854 | sd->balance_interval *= 2; | |
1da177e4 LT |
2855 | } |
2856 | ||
43010659 | 2857 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2858 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2859 | return -1; |
43010659 | 2860 | return ld_moved; |
1da177e4 LT |
2861 | |
2862 | out_balanced: | |
1da177e4 LT |
2863 | schedstat_inc(sd, lb_balanced[idle]); |
2864 | ||
16cfb1c0 | 2865 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
2866 | |
2867 | out_one_pinned: | |
1da177e4 | 2868 | /* tune up the balancing interval */ |
77391d71 NP |
2869 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
2870 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
2871 | sd->balance_interval *= 2; |
2872 | ||
48f24c4d | 2873 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2874 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2875 | return -1; |
1da177e4 LT |
2876 | return 0; |
2877 | } | |
2878 | ||
2879 | /* | |
2880 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2881 | * tasks if there is an imbalance. | |
2882 | * | |
d15bcfdb | 2883 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
2884 | * this_rq is locked. |
2885 | */ | |
48f24c4d | 2886 | static int |
70b97a7f | 2887 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
2888 | { |
2889 | struct sched_group *group; | |
70b97a7f | 2890 | struct rq *busiest = NULL; |
1da177e4 | 2891 | unsigned long imbalance; |
43010659 | 2892 | int ld_moved = 0; |
5969fe06 | 2893 | int sd_idle = 0; |
969bb4e4 | 2894 | int all_pinned = 0; |
0a2966b4 | 2895 | cpumask_t cpus = CPU_MASK_ALL; |
5969fe06 | 2896 | |
89c4710e SS |
2897 | /* |
2898 | * When power savings policy is enabled for the parent domain, idle | |
2899 | * sibling can pick up load irrespective of busy siblings. In this case, | |
2900 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 2901 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
2902 | */ |
2903 | if (sd->flags & SD_SHARE_CPUPOWER && | |
2904 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 2905 | sd_idle = 1; |
1da177e4 | 2906 | |
2d72376b | 2907 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 2908 | redo: |
d15bcfdb | 2909 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
783609c6 | 2910 | &sd_idle, &cpus, NULL); |
1da177e4 | 2911 | if (!group) { |
d15bcfdb | 2912 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2913 | goto out_balanced; |
1da177e4 LT |
2914 | } |
2915 | ||
d15bcfdb | 2916 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, |
0a2966b4 | 2917 | &cpus); |
db935dbd | 2918 | if (!busiest) { |
d15bcfdb | 2919 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2920 | goto out_balanced; |
1da177e4 LT |
2921 | } |
2922 | ||
db935dbd NP |
2923 | BUG_ON(busiest == this_rq); |
2924 | ||
d15bcfdb | 2925 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 2926 | |
43010659 | 2927 | ld_moved = 0; |
d6d5cfaf NP |
2928 | if (busiest->nr_running > 1) { |
2929 | /* Attempt to move tasks */ | |
2930 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
2931 | /* this_rq->clock is already updated */ |
2932 | update_rq_clock(busiest); | |
43010659 | 2933 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
2934 | imbalance, sd, CPU_NEWLY_IDLE, |
2935 | &all_pinned); | |
d6d5cfaf | 2936 | spin_unlock(&busiest->lock); |
0a2966b4 | 2937 | |
969bb4e4 | 2938 | if (unlikely(all_pinned)) { |
0a2966b4 CL |
2939 | cpu_clear(cpu_of(busiest), cpus); |
2940 | if (!cpus_empty(cpus)) | |
2941 | goto redo; | |
2942 | } | |
d6d5cfaf NP |
2943 | } |
2944 | ||
43010659 | 2945 | if (!ld_moved) { |
d15bcfdb | 2946 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
2947 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
2948 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
2949 | return -1; |
2950 | } else | |
16cfb1c0 | 2951 | sd->nr_balance_failed = 0; |
1da177e4 | 2952 | |
43010659 | 2953 | return ld_moved; |
16cfb1c0 NP |
2954 | |
2955 | out_balanced: | |
d15bcfdb | 2956 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 2957 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2958 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2959 | return -1; |
16cfb1c0 | 2960 | sd->nr_balance_failed = 0; |
48f24c4d | 2961 | |
16cfb1c0 | 2962 | return 0; |
1da177e4 LT |
2963 | } |
2964 | ||
2965 | /* | |
2966 | * idle_balance is called by schedule() if this_cpu is about to become | |
2967 | * idle. Attempts to pull tasks from other CPUs. | |
2968 | */ | |
70b97a7f | 2969 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
2970 | { |
2971 | struct sched_domain *sd; | |
dd41f596 IM |
2972 | int pulled_task = -1; |
2973 | unsigned long next_balance = jiffies + HZ; | |
1da177e4 LT |
2974 | |
2975 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
2976 | unsigned long interval; |
2977 | ||
2978 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
2979 | continue; | |
2980 | ||
2981 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 2982 | /* If we've pulled tasks over stop searching: */ |
1bd77f2d | 2983 | pulled_task = load_balance_newidle(this_cpu, |
92c4ca5c CL |
2984 | this_rq, sd); |
2985 | ||
2986 | interval = msecs_to_jiffies(sd->balance_interval); | |
2987 | if (time_after(next_balance, sd->last_balance + interval)) | |
2988 | next_balance = sd->last_balance + interval; | |
2989 | if (pulled_task) | |
2990 | break; | |
1da177e4 | 2991 | } |
dd41f596 | 2992 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
2993 | /* |
2994 | * We are going idle. next_balance may be set based on | |
2995 | * a busy processor. So reset next_balance. | |
2996 | */ | |
2997 | this_rq->next_balance = next_balance; | |
dd41f596 | 2998 | } |
1da177e4 LT |
2999 | } |
3000 | ||
3001 | /* | |
3002 | * active_load_balance is run by migration threads. It pushes running tasks | |
3003 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
3004 | * running on each physical CPU where possible, and avoids physical / | |
3005 | * logical imbalances. | |
3006 | * | |
3007 | * Called with busiest_rq locked. | |
3008 | */ | |
70b97a7f | 3009 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3010 | { |
39507451 | 3011 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3012 | struct sched_domain *sd; |
3013 | struct rq *target_rq; | |
39507451 | 3014 | |
48f24c4d | 3015 | /* Is there any task to move? */ |
39507451 | 3016 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3017 | return; |
3018 | ||
3019 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3020 | |
3021 | /* | |
39507451 | 3022 | * This condition is "impossible", if it occurs |
41a2d6cf | 3023 | * we need to fix it. Originally reported by |
39507451 | 3024 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3025 | */ |
39507451 | 3026 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3027 | |
39507451 NP |
3028 | /* move a task from busiest_rq to target_rq */ |
3029 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3030 | update_rq_clock(busiest_rq); |
3031 | update_rq_clock(target_rq); | |
39507451 NP |
3032 | |
3033 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3034 | for_each_domain(target_cpu, sd) { |
39507451 | 3035 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 3036 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 3037 | break; |
c96d145e | 3038 | } |
39507451 | 3039 | |
48f24c4d | 3040 | if (likely(sd)) { |
2d72376b | 3041 | schedstat_inc(sd, alb_count); |
39507451 | 3042 | |
43010659 PW |
3043 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3044 | sd, CPU_IDLE)) | |
48f24c4d IM |
3045 | schedstat_inc(sd, alb_pushed); |
3046 | else | |
3047 | schedstat_inc(sd, alb_failed); | |
3048 | } | |
39507451 | 3049 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
3050 | } |
3051 | ||
46cb4b7c SS |
3052 | #ifdef CONFIG_NO_HZ |
3053 | static struct { | |
3054 | atomic_t load_balancer; | |
41a2d6cf | 3055 | cpumask_t cpu_mask; |
46cb4b7c SS |
3056 | } nohz ____cacheline_aligned = { |
3057 | .load_balancer = ATOMIC_INIT(-1), | |
3058 | .cpu_mask = CPU_MASK_NONE, | |
3059 | }; | |
3060 | ||
7835b98b | 3061 | /* |
46cb4b7c SS |
3062 | * This routine will try to nominate the ilb (idle load balancing) |
3063 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3064 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3065 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3066 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3067 | * arrives... | |
3068 | * | |
3069 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3070 | * for idle load balancing. ilb owner will still be part of | |
3071 | * nohz.cpu_mask.. | |
7835b98b | 3072 | * |
46cb4b7c SS |
3073 | * While stopping the tick, this cpu will become the ilb owner if there |
3074 | * is no other owner. And will be the owner till that cpu becomes busy | |
3075 | * or if all cpus in the system stop their ticks at which point | |
3076 | * there is no need for ilb owner. | |
3077 | * | |
3078 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3079 | * next busy scheduler_tick() | |
3080 | */ | |
3081 | int select_nohz_load_balancer(int stop_tick) | |
3082 | { | |
3083 | int cpu = smp_processor_id(); | |
3084 | ||
3085 | if (stop_tick) { | |
3086 | cpu_set(cpu, nohz.cpu_mask); | |
3087 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3088 | ||
3089 | /* | |
3090 | * If we are going offline and still the leader, give up! | |
3091 | */ | |
3092 | if (cpu_is_offline(cpu) && | |
3093 | atomic_read(&nohz.load_balancer) == cpu) { | |
3094 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3095 | BUG(); | |
3096 | return 0; | |
3097 | } | |
3098 | ||
3099 | /* time for ilb owner also to sleep */ | |
3100 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3101 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3102 | atomic_set(&nohz.load_balancer, -1); | |
3103 | return 0; | |
3104 | } | |
3105 | ||
3106 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3107 | /* make me the ilb owner */ | |
3108 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3109 | return 1; | |
3110 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3111 | return 1; | |
3112 | } else { | |
3113 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3114 | return 0; | |
3115 | ||
3116 | cpu_clear(cpu, nohz.cpu_mask); | |
3117 | ||
3118 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3119 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3120 | BUG(); | |
3121 | } | |
3122 | return 0; | |
3123 | } | |
3124 | #endif | |
3125 | ||
3126 | static DEFINE_SPINLOCK(balancing); | |
3127 | ||
3128 | /* | |
7835b98b CL |
3129 | * It checks each scheduling domain to see if it is due to be balanced, |
3130 | * and initiates a balancing operation if so. | |
3131 | * | |
3132 | * Balancing parameters are set up in arch_init_sched_domains. | |
3133 | */ | |
a9957449 | 3134 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3135 | { |
46cb4b7c SS |
3136 | int balance = 1; |
3137 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3138 | unsigned long interval; |
3139 | struct sched_domain *sd; | |
46cb4b7c | 3140 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3141 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3142 | int update_next_balance = 0; |
1da177e4 | 3143 | |
46cb4b7c | 3144 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3145 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3146 | continue; | |
3147 | ||
3148 | interval = sd->balance_interval; | |
d15bcfdb | 3149 | if (idle != CPU_IDLE) |
1da177e4 LT |
3150 | interval *= sd->busy_factor; |
3151 | ||
3152 | /* scale ms to jiffies */ | |
3153 | interval = msecs_to_jiffies(interval); | |
3154 | if (unlikely(!interval)) | |
3155 | interval = 1; | |
dd41f596 IM |
3156 | if (interval > HZ*NR_CPUS/10) |
3157 | interval = HZ*NR_CPUS/10; | |
3158 | ||
1da177e4 | 3159 | |
08c183f3 CL |
3160 | if (sd->flags & SD_SERIALIZE) { |
3161 | if (!spin_trylock(&balancing)) | |
3162 | goto out; | |
3163 | } | |
3164 | ||
c9819f45 | 3165 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
46cb4b7c | 3166 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
3167 | /* |
3168 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3169 | * longer idle, or one of our SMT siblings is |
3170 | * not idle. | |
3171 | */ | |
d15bcfdb | 3172 | idle = CPU_NOT_IDLE; |
1da177e4 | 3173 | } |
1bd77f2d | 3174 | sd->last_balance = jiffies; |
1da177e4 | 3175 | } |
08c183f3 CL |
3176 | if (sd->flags & SD_SERIALIZE) |
3177 | spin_unlock(&balancing); | |
3178 | out: | |
f549da84 | 3179 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3180 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3181 | update_next_balance = 1; |
3182 | } | |
783609c6 SS |
3183 | |
3184 | /* | |
3185 | * Stop the load balance at this level. There is another | |
3186 | * CPU in our sched group which is doing load balancing more | |
3187 | * actively. | |
3188 | */ | |
3189 | if (!balance) | |
3190 | break; | |
1da177e4 | 3191 | } |
f549da84 SS |
3192 | |
3193 | /* | |
3194 | * next_balance will be updated only when there is a need. | |
3195 | * When the cpu is attached to null domain for ex, it will not be | |
3196 | * updated. | |
3197 | */ | |
3198 | if (likely(update_next_balance)) | |
3199 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3200 | } |
3201 | ||
3202 | /* | |
3203 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3204 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3205 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3206 | */ | |
3207 | static void run_rebalance_domains(struct softirq_action *h) | |
3208 | { | |
dd41f596 IM |
3209 | int this_cpu = smp_processor_id(); |
3210 | struct rq *this_rq = cpu_rq(this_cpu); | |
3211 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3212 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3213 | |
dd41f596 | 3214 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3215 | |
3216 | #ifdef CONFIG_NO_HZ | |
3217 | /* | |
3218 | * If this cpu is the owner for idle load balancing, then do the | |
3219 | * balancing on behalf of the other idle cpus whose ticks are | |
3220 | * stopped. | |
3221 | */ | |
dd41f596 IM |
3222 | if (this_rq->idle_at_tick && |
3223 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3224 | cpumask_t cpus = nohz.cpu_mask; |
3225 | struct rq *rq; | |
3226 | int balance_cpu; | |
3227 | ||
dd41f596 | 3228 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3229 | for_each_cpu_mask(balance_cpu, cpus) { |
3230 | /* | |
3231 | * If this cpu gets work to do, stop the load balancing | |
3232 | * work being done for other cpus. Next load | |
3233 | * balancing owner will pick it up. | |
3234 | */ | |
3235 | if (need_resched()) | |
3236 | break; | |
3237 | ||
de0cf899 | 3238 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3239 | |
3240 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3241 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3242 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3243 | } |
3244 | } | |
3245 | #endif | |
3246 | } | |
3247 | ||
3248 | /* | |
3249 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3250 | * | |
3251 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3252 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3253 | * if the whole system is idle. | |
3254 | */ | |
dd41f596 | 3255 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3256 | { |
46cb4b7c SS |
3257 | #ifdef CONFIG_NO_HZ |
3258 | /* | |
3259 | * If we were in the nohz mode recently and busy at the current | |
3260 | * scheduler tick, then check if we need to nominate new idle | |
3261 | * load balancer. | |
3262 | */ | |
3263 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3264 | rq->in_nohz_recently = 0; | |
3265 | ||
3266 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3267 | cpu_clear(cpu, nohz.cpu_mask); | |
3268 | atomic_set(&nohz.load_balancer, -1); | |
3269 | } | |
3270 | ||
3271 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3272 | /* | |
3273 | * simple selection for now: Nominate the | |
3274 | * first cpu in the nohz list to be the next | |
3275 | * ilb owner. | |
3276 | * | |
3277 | * TBD: Traverse the sched domains and nominate | |
3278 | * the nearest cpu in the nohz.cpu_mask. | |
3279 | */ | |
3280 | int ilb = first_cpu(nohz.cpu_mask); | |
3281 | ||
3282 | if (ilb != NR_CPUS) | |
3283 | resched_cpu(ilb); | |
3284 | } | |
3285 | } | |
3286 | ||
3287 | /* | |
3288 | * If this cpu is idle and doing idle load balancing for all the | |
3289 | * cpus with ticks stopped, is it time for that to stop? | |
3290 | */ | |
3291 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3292 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3293 | resched_cpu(cpu); | |
3294 | return; | |
3295 | } | |
3296 | ||
3297 | /* | |
3298 | * If this cpu is idle and the idle load balancing is done by | |
3299 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3300 | */ | |
3301 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3302 | cpu_isset(cpu, nohz.cpu_mask)) | |
3303 | return; | |
3304 | #endif | |
3305 | if (time_after_eq(jiffies, rq->next_balance)) | |
3306 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3307 | } |
dd41f596 IM |
3308 | |
3309 | #else /* CONFIG_SMP */ | |
3310 | ||
1da177e4 LT |
3311 | /* |
3312 | * on UP we do not need to balance between CPUs: | |
3313 | */ | |
70b97a7f | 3314 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3315 | { |
3316 | } | |
dd41f596 | 3317 | |
1da177e4 LT |
3318 | #endif |
3319 | ||
1da177e4 LT |
3320 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3321 | ||
3322 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3323 | ||
3324 | /* | |
41b86e9c IM |
3325 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
3326 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 3327 | */ |
41b86e9c | 3328 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 3329 | { |
1da177e4 | 3330 | unsigned long flags; |
41b86e9c IM |
3331 | u64 ns, delta_exec; |
3332 | struct rq *rq; | |
48f24c4d | 3333 | |
41b86e9c IM |
3334 | rq = task_rq_lock(p, &flags); |
3335 | ns = p->se.sum_exec_runtime; | |
051a1d1a | 3336 | if (task_current(rq, p)) { |
a8e504d2 IM |
3337 | update_rq_clock(rq); |
3338 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
3339 | if ((s64)delta_exec > 0) |
3340 | ns += delta_exec; | |
3341 | } | |
3342 | task_rq_unlock(rq, &flags); | |
48f24c4d | 3343 | |
1da177e4 LT |
3344 | return ns; |
3345 | } | |
3346 | ||
1da177e4 LT |
3347 | /* |
3348 | * Account user cpu time to a process. | |
3349 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
3350 | * @cputime: the cpu time spent in user space since the last update |
3351 | */ | |
3352 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3353 | { | |
3354 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3355 | cputime64_t tmp; | |
3356 | ||
3357 | p->utime = cputime_add(p->utime, cputime); | |
3358 | ||
3359 | /* Add user time to cpustat. */ | |
3360 | tmp = cputime_to_cputime64(cputime); | |
3361 | if (TASK_NICE(p) > 0) | |
3362 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3363 | else | |
3364 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3365 | } | |
3366 | ||
94886b84 LV |
3367 | /* |
3368 | * Account guest cpu time to a process. | |
3369 | * @p: the process that the cpu time gets accounted to | |
3370 | * @cputime: the cpu time spent in virtual machine since the last update | |
3371 | */ | |
f7402e03 | 3372 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
3373 | { |
3374 | cputime64_t tmp; | |
3375 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3376 | ||
3377 | tmp = cputime_to_cputime64(cputime); | |
3378 | ||
3379 | p->utime = cputime_add(p->utime, cputime); | |
3380 | p->gtime = cputime_add(p->gtime, cputime); | |
3381 | ||
3382 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3383 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3384 | } | |
3385 | ||
c66f08be MN |
3386 | /* |
3387 | * Account scaled user cpu time to a process. | |
3388 | * @p: the process that the cpu time gets accounted to | |
3389 | * @cputime: the cpu time spent in user space since the last update | |
3390 | */ | |
3391 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
3392 | { | |
3393 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
3394 | } | |
3395 | ||
1da177e4 LT |
3396 | /* |
3397 | * Account system cpu time to a process. | |
3398 | * @p: the process that the cpu time gets accounted to | |
3399 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3400 | * @cputime: the cpu time spent in kernel space since the last update | |
3401 | */ | |
3402 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3403 | cputime_t cputime) | |
3404 | { | |
3405 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3406 | struct rq *rq = this_rq(); |
1da177e4 LT |
3407 | cputime64_t tmp; |
3408 | ||
9778385d CB |
3409 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) |
3410 | return account_guest_time(p, cputime); | |
94886b84 | 3411 | |
1da177e4 LT |
3412 | p->stime = cputime_add(p->stime, cputime); |
3413 | ||
3414 | /* Add system time to cpustat. */ | |
3415 | tmp = cputime_to_cputime64(cputime); | |
3416 | if (hardirq_count() - hardirq_offset) | |
3417 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3418 | else if (softirq_count()) | |
3419 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 3420 | else if (p != rq->idle) |
1da177e4 | 3421 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 3422 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
3423 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
3424 | else | |
3425 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3426 | /* Account for system time used */ | |
3427 | acct_update_integrals(p); | |
1da177e4 LT |
3428 | } |
3429 | ||
c66f08be MN |
3430 | /* |
3431 | * Account scaled system cpu time to a process. | |
3432 | * @p: the process that the cpu time gets accounted to | |
3433 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3434 | * @cputime: the cpu time spent in kernel space since the last update | |
3435 | */ | |
3436 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
3437 | { | |
3438 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
3439 | } | |
3440 | ||
1da177e4 LT |
3441 | /* |
3442 | * Account for involuntary wait time. | |
3443 | * @p: the process from which the cpu time has been stolen | |
3444 | * @steal: the cpu time spent in involuntary wait | |
3445 | */ | |
3446 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3447 | { | |
3448 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3449 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3450 | struct rq *rq = this_rq(); |
1da177e4 LT |
3451 | |
3452 | if (p == rq->idle) { | |
3453 | p->stime = cputime_add(p->stime, steal); | |
3454 | if (atomic_read(&rq->nr_iowait) > 0) | |
3455 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3456 | else | |
3457 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 3458 | } else |
1da177e4 LT |
3459 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
3460 | } | |
3461 | ||
7835b98b CL |
3462 | /* |
3463 | * This function gets called by the timer code, with HZ frequency. | |
3464 | * We call it with interrupts disabled. | |
3465 | * | |
3466 | * It also gets called by the fork code, when changing the parent's | |
3467 | * timeslices. | |
3468 | */ | |
3469 | void scheduler_tick(void) | |
3470 | { | |
7835b98b CL |
3471 | int cpu = smp_processor_id(); |
3472 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3473 | struct task_struct *curr = rq->curr; |
529c7726 | 3474 | u64 next_tick = rq->tick_timestamp + TICK_NSEC; |
dd41f596 IM |
3475 | |
3476 | spin_lock(&rq->lock); | |
546fe3c9 | 3477 | __update_rq_clock(rq); |
529c7726 IM |
3478 | /* |
3479 | * Let rq->clock advance by at least TICK_NSEC: | |
3480 | */ | |
3481 | if (unlikely(rq->clock < next_tick)) | |
3482 | rq->clock = next_tick; | |
3483 | rq->tick_timestamp = rq->clock; | |
f1a438d8 | 3484 | update_cpu_load(rq); |
dd41f596 IM |
3485 | if (curr != rq->idle) /* FIXME: needed? */ |
3486 | curr->sched_class->task_tick(rq, curr); | |
dd41f596 | 3487 | spin_unlock(&rq->lock); |
7835b98b | 3488 | |
e418e1c2 | 3489 | #ifdef CONFIG_SMP |
dd41f596 IM |
3490 | rq->idle_at_tick = idle_cpu(cpu); |
3491 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3492 | #endif |
1da177e4 LT |
3493 | } |
3494 | ||
1da177e4 LT |
3495 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3496 | ||
3497 | void fastcall add_preempt_count(int val) | |
3498 | { | |
3499 | /* | |
3500 | * Underflow? | |
3501 | */ | |
9a11b49a IM |
3502 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3503 | return; | |
1da177e4 LT |
3504 | preempt_count() += val; |
3505 | /* | |
3506 | * Spinlock count overflowing soon? | |
3507 | */ | |
33859f7f MOS |
3508 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3509 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3510 | } |
3511 | EXPORT_SYMBOL(add_preempt_count); | |
3512 | ||
3513 | void fastcall sub_preempt_count(int val) | |
3514 | { | |
3515 | /* | |
3516 | * Underflow? | |
3517 | */ | |
9a11b49a IM |
3518 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3519 | return; | |
1da177e4 LT |
3520 | /* |
3521 | * Is the spinlock portion underflowing? | |
3522 | */ | |
9a11b49a IM |
3523 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3524 | !(preempt_count() & PREEMPT_MASK))) | |
3525 | return; | |
3526 | ||
1da177e4 LT |
3527 | preempt_count() -= val; |
3528 | } | |
3529 | EXPORT_SYMBOL(sub_preempt_count); | |
3530 | ||
3531 | #endif | |
3532 | ||
3533 | /* | |
dd41f596 | 3534 | * Print scheduling while atomic bug: |
1da177e4 | 3535 | */ |
dd41f596 | 3536 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3537 | { |
838225b4 SS |
3538 | struct pt_regs *regs = get_irq_regs(); |
3539 | ||
3540 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
3541 | prev->comm, prev->pid, preempt_count()); | |
3542 | ||
dd41f596 IM |
3543 | debug_show_held_locks(prev); |
3544 | if (irqs_disabled()) | |
3545 | print_irqtrace_events(prev); | |
838225b4 SS |
3546 | |
3547 | if (regs) | |
3548 | show_regs(regs); | |
3549 | else | |
3550 | dump_stack(); | |
dd41f596 | 3551 | } |
1da177e4 | 3552 | |
dd41f596 IM |
3553 | /* |
3554 | * Various schedule()-time debugging checks and statistics: | |
3555 | */ | |
3556 | static inline void schedule_debug(struct task_struct *prev) | |
3557 | { | |
1da177e4 | 3558 | /* |
41a2d6cf | 3559 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3560 | * schedule() atomically, we ignore that path for now. |
3561 | * Otherwise, whine if we are scheduling when we should not be. | |
3562 | */ | |
dd41f596 IM |
3563 | if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state)) |
3564 | __schedule_bug(prev); | |
3565 | ||
1da177e4 LT |
3566 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3567 | ||
2d72376b | 3568 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3569 | #ifdef CONFIG_SCHEDSTATS |
3570 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3571 | schedstat_inc(this_rq(), bkl_count); |
3572 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3573 | } |
3574 | #endif | |
dd41f596 IM |
3575 | } |
3576 | ||
3577 | /* | |
3578 | * Pick up the highest-prio task: | |
3579 | */ | |
3580 | static inline struct task_struct * | |
ff95f3df | 3581 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 3582 | { |
5522d5d5 | 3583 | const struct sched_class *class; |
dd41f596 | 3584 | struct task_struct *p; |
1da177e4 LT |
3585 | |
3586 | /* | |
dd41f596 IM |
3587 | * Optimization: we know that if all tasks are in |
3588 | * the fair class we can call that function directly: | |
1da177e4 | 3589 | */ |
dd41f596 | 3590 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3591 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3592 | if (likely(p)) |
3593 | return p; | |
1da177e4 LT |
3594 | } |
3595 | ||
dd41f596 IM |
3596 | class = sched_class_highest; |
3597 | for ( ; ; ) { | |
fb8d4724 | 3598 | p = class->pick_next_task(rq); |
dd41f596 IM |
3599 | if (p) |
3600 | return p; | |
3601 | /* | |
3602 | * Will never be NULL as the idle class always | |
3603 | * returns a non-NULL p: | |
3604 | */ | |
3605 | class = class->next; | |
3606 | } | |
3607 | } | |
1da177e4 | 3608 | |
dd41f596 IM |
3609 | /* |
3610 | * schedule() is the main scheduler function. | |
3611 | */ | |
3612 | asmlinkage void __sched schedule(void) | |
3613 | { | |
3614 | struct task_struct *prev, *next; | |
3615 | long *switch_count; | |
3616 | struct rq *rq; | |
dd41f596 IM |
3617 | int cpu; |
3618 | ||
3619 | need_resched: | |
3620 | preempt_disable(); | |
3621 | cpu = smp_processor_id(); | |
3622 | rq = cpu_rq(cpu); | |
3623 | rcu_qsctr_inc(cpu); | |
3624 | prev = rq->curr; | |
3625 | switch_count = &prev->nivcsw; | |
3626 | ||
3627 | release_kernel_lock(prev); | |
3628 | need_resched_nonpreemptible: | |
3629 | ||
3630 | schedule_debug(prev); | |
1da177e4 | 3631 | |
1e819950 IM |
3632 | /* |
3633 | * Do the rq-clock update outside the rq lock: | |
3634 | */ | |
3635 | local_irq_disable(); | |
c1b3da3e | 3636 | __update_rq_clock(rq); |
1e819950 IM |
3637 | spin_lock(&rq->lock); |
3638 | clear_tsk_need_resched(prev); | |
1da177e4 | 3639 | |
1da177e4 | 3640 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
1da177e4 | 3641 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && |
dd41f596 | 3642 | unlikely(signal_pending(prev)))) { |
1da177e4 | 3643 | prev->state = TASK_RUNNING; |
dd41f596 | 3644 | } else { |
2e1cb74a | 3645 | deactivate_task(rq, prev, 1); |
1da177e4 | 3646 | } |
dd41f596 | 3647 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3648 | } |
3649 | ||
9a897c5a SR |
3650 | #ifdef CONFIG_SMP |
3651 | if (prev->sched_class->pre_schedule) | |
3652 | prev->sched_class->pre_schedule(rq, prev); | |
3653 | #endif | |
f65eda4f | 3654 | |
dd41f596 | 3655 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3656 | idle_balance(cpu, rq); |
1da177e4 | 3657 | |
31ee529c | 3658 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 3659 | next = pick_next_task(rq, prev); |
1da177e4 LT |
3660 | |
3661 | sched_info_switch(prev, next); | |
dd41f596 | 3662 | |
1da177e4 | 3663 | if (likely(prev != next)) { |
1da177e4 LT |
3664 | rq->nr_switches++; |
3665 | rq->curr = next; | |
3666 | ++*switch_count; | |
3667 | ||
dd41f596 | 3668 | context_switch(rq, prev, next); /* unlocks the rq */ |
1da177e4 LT |
3669 | } else |
3670 | spin_unlock_irq(&rq->lock); | |
3671 | ||
dd41f596 IM |
3672 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
3673 | cpu = smp_processor_id(); | |
3674 | rq = cpu_rq(cpu); | |
1da177e4 | 3675 | goto need_resched_nonpreemptible; |
dd41f596 | 3676 | } |
1da177e4 LT |
3677 | preempt_enable_no_resched(); |
3678 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3679 | goto need_resched; | |
3680 | } | |
1da177e4 LT |
3681 | EXPORT_SYMBOL(schedule); |
3682 | ||
3683 | #ifdef CONFIG_PREEMPT | |
3684 | /* | |
2ed6e34f | 3685 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3686 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3687 | * occur there and call schedule directly. |
3688 | */ | |
3689 | asmlinkage void __sched preempt_schedule(void) | |
3690 | { | |
3691 | struct thread_info *ti = current_thread_info(); | |
3692 | #ifdef CONFIG_PREEMPT_BKL | |
3693 | struct task_struct *task = current; | |
3694 | int saved_lock_depth; | |
3695 | #endif | |
3696 | /* | |
3697 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3698 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3699 | */ |
beed33a8 | 3700 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3701 | return; |
3702 | ||
3a5c359a AK |
3703 | do { |
3704 | add_preempt_count(PREEMPT_ACTIVE); | |
3705 | ||
3706 | /* | |
3707 | * We keep the big kernel semaphore locked, but we | |
3708 | * clear ->lock_depth so that schedule() doesnt | |
3709 | * auto-release the semaphore: | |
3710 | */ | |
1da177e4 | 3711 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a AK |
3712 | saved_lock_depth = task->lock_depth; |
3713 | task->lock_depth = -1; | |
1da177e4 | 3714 | #endif |
3a5c359a | 3715 | schedule(); |
1da177e4 | 3716 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a | 3717 | task->lock_depth = saved_lock_depth; |
1da177e4 | 3718 | #endif |
3a5c359a | 3719 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3720 | |
3a5c359a AK |
3721 | /* |
3722 | * Check again in case we missed a preemption opportunity | |
3723 | * between schedule and now. | |
3724 | */ | |
3725 | barrier(); | |
3726 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 3727 | } |
1da177e4 LT |
3728 | EXPORT_SYMBOL(preempt_schedule); |
3729 | ||
3730 | /* | |
2ed6e34f | 3731 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3732 | * off of irq context. |
3733 | * Note, that this is called and return with irqs disabled. This will | |
3734 | * protect us against recursive calling from irq. | |
3735 | */ | |
3736 | asmlinkage void __sched preempt_schedule_irq(void) | |
3737 | { | |
3738 | struct thread_info *ti = current_thread_info(); | |
3739 | #ifdef CONFIG_PREEMPT_BKL | |
3740 | struct task_struct *task = current; | |
3741 | int saved_lock_depth; | |
3742 | #endif | |
2ed6e34f | 3743 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3744 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3745 | ||
3a5c359a AK |
3746 | do { |
3747 | add_preempt_count(PREEMPT_ACTIVE); | |
3748 | ||
3749 | /* | |
3750 | * We keep the big kernel semaphore locked, but we | |
3751 | * clear ->lock_depth so that schedule() doesnt | |
3752 | * auto-release the semaphore: | |
3753 | */ | |
1da177e4 | 3754 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a AK |
3755 | saved_lock_depth = task->lock_depth; |
3756 | task->lock_depth = -1; | |
1da177e4 | 3757 | #endif |
3a5c359a AK |
3758 | local_irq_enable(); |
3759 | schedule(); | |
3760 | local_irq_disable(); | |
1da177e4 | 3761 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a | 3762 | task->lock_depth = saved_lock_depth; |
1da177e4 | 3763 | #endif |
3a5c359a | 3764 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3765 | |
3a5c359a AK |
3766 | /* |
3767 | * Check again in case we missed a preemption opportunity | |
3768 | * between schedule and now. | |
3769 | */ | |
3770 | barrier(); | |
3771 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
3772 | } |
3773 | ||
3774 | #endif /* CONFIG_PREEMPT */ | |
3775 | ||
95cdf3b7 IM |
3776 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
3777 | void *key) | |
1da177e4 | 3778 | { |
48f24c4d | 3779 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 3780 | } |
1da177e4 LT |
3781 | EXPORT_SYMBOL(default_wake_function); |
3782 | ||
3783 | /* | |
41a2d6cf IM |
3784 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3785 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3786 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3787 | * | |
3788 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3789 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3790 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3791 | */ | |
3792 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
3793 | int nr_exclusive, int sync, void *key) | |
3794 | { | |
2e45874c | 3795 | wait_queue_t *curr, *next; |
1da177e4 | 3796 | |
2e45874c | 3797 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3798 | unsigned flags = curr->flags; |
3799 | ||
1da177e4 | 3800 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 3801 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3802 | break; |
3803 | } | |
3804 | } | |
3805 | ||
3806 | /** | |
3807 | * __wake_up - wake up threads blocked on a waitqueue. | |
3808 | * @q: the waitqueue | |
3809 | * @mode: which threads | |
3810 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3811 | * @key: is directly passed to the wakeup function |
1da177e4 LT |
3812 | */ |
3813 | void fastcall __wake_up(wait_queue_head_t *q, unsigned int mode, | |
95cdf3b7 | 3814 | int nr_exclusive, void *key) |
1da177e4 LT |
3815 | { |
3816 | unsigned long flags; | |
3817 | ||
3818 | spin_lock_irqsave(&q->lock, flags); | |
3819 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3820 | spin_unlock_irqrestore(&q->lock, flags); | |
3821 | } | |
1da177e4 LT |
3822 | EXPORT_SYMBOL(__wake_up); |
3823 | ||
3824 | /* | |
3825 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3826 | */ | |
3827 | void fastcall __wake_up_locked(wait_queue_head_t *q, unsigned int mode) | |
3828 | { | |
3829 | __wake_up_common(q, mode, 1, 0, NULL); | |
3830 | } | |
3831 | ||
3832 | /** | |
67be2dd1 | 3833 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3834 | * @q: the waitqueue |
3835 | * @mode: which threads | |
3836 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
3837 | * | |
3838 | * The sync wakeup differs that the waker knows that it will schedule | |
3839 | * away soon, so while the target thread will be woken up, it will not | |
3840 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3841 | * with each other. This can prevent needless bouncing between CPUs. | |
3842 | * | |
3843 | * On UP it can prevent extra preemption. | |
3844 | */ | |
95cdf3b7 IM |
3845 | void fastcall |
3846 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
1da177e4 LT |
3847 | { |
3848 | unsigned long flags; | |
3849 | int sync = 1; | |
3850 | ||
3851 | if (unlikely(!q)) | |
3852 | return; | |
3853 | ||
3854 | if (unlikely(!nr_exclusive)) | |
3855 | sync = 0; | |
3856 | ||
3857 | spin_lock_irqsave(&q->lock, flags); | |
3858 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
3859 | spin_unlock_irqrestore(&q->lock, flags); | |
3860 | } | |
3861 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
3862 | ||
b15136e9 | 3863 | void complete(struct completion *x) |
1da177e4 LT |
3864 | { |
3865 | unsigned long flags; | |
3866 | ||
3867 | spin_lock_irqsave(&x->wait.lock, flags); | |
3868 | x->done++; | |
3869 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3870 | 1, 0, NULL); | |
3871 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3872 | } | |
3873 | EXPORT_SYMBOL(complete); | |
3874 | ||
b15136e9 | 3875 | void complete_all(struct completion *x) |
1da177e4 LT |
3876 | { |
3877 | unsigned long flags; | |
3878 | ||
3879 | spin_lock_irqsave(&x->wait.lock, flags); | |
3880 | x->done += UINT_MAX/2; | |
3881 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3882 | 0, 0, NULL); | |
3883 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3884 | } | |
3885 | EXPORT_SYMBOL(complete_all); | |
3886 | ||
8cbbe86d AK |
3887 | static inline long __sched |
3888 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3889 | { |
1da177e4 LT |
3890 | if (!x->done) { |
3891 | DECLARE_WAITQUEUE(wait, current); | |
3892 | ||
3893 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3894 | __add_wait_queue_tail(&x->wait, &wait); | |
3895 | do { | |
8cbbe86d AK |
3896 | if (state == TASK_INTERRUPTIBLE && |
3897 | signal_pending(current)) { | |
3898 | __remove_wait_queue(&x->wait, &wait); | |
3899 | return -ERESTARTSYS; | |
3900 | } | |
3901 | __set_current_state(state); | |
1da177e4 LT |
3902 | spin_unlock_irq(&x->wait.lock); |
3903 | timeout = schedule_timeout(timeout); | |
3904 | spin_lock_irq(&x->wait.lock); | |
3905 | if (!timeout) { | |
3906 | __remove_wait_queue(&x->wait, &wait); | |
8cbbe86d | 3907 | return timeout; |
1da177e4 LT |
3908 | } |
3909 | } while (!x->done); | |
3910 | __remove_wait_queue(&x->wait, &wait); | |
3911 | } | |
3912 | x->done--; | |
1da177e4 LT |
3913 | return timeout; |
3914 | } | |
1da177e4 | 3915 | |
8cbbe86d AK |
3916 | static long __sched |
3917 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3918 | { |
1da177e4 LT |
3919 | might_sleep(); |
3920 | ||
3921 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 3922 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 3923 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3924 | return timeout; |
3925 | } | |
1da177e4 | 3926 | |
b15136e9 | 3927 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
3928 | { |
3929 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 3930 | } |
8cbbe86d | 3931 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 3932 | |
b15136e9 | 3933 | unsigned long __sched |
8cbbe86d | 3934 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 3935 | { |
8cbbe86d | 3936 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 3937 | } |
8cbbe86d | 3938 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 3939 | |
8cbbe86d | 3940 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 3941 | { |
51e97990 AK |
3942 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
3943 | if (t == -ERESTARTSYS) | |
3944 | return t; | |
3945 | return 0; | |
0fec171c | 3946 | } |
8cbbe86d | 3947 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 3948 | |
b15136e9 | 3949 | unsigned long __sched |
8cbbe86d AK |
3950 | wait_for_completion_interruptible_timeout(struct completion *x, |
3951 | unsigned long timeout) | |
0fec171c | 3952 | { |
8cbbe86d | 3953 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 3954 | } |
8cbbe86d | 3955 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 3956 | |
8cbbe86d AK |
3957 | static long __sched |
3958 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 3959 | { |
0fec171c IM |
3960 | unsigned long flags; |
3961 | wait_queue_t wait; | |
3962 | ||
3963 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 3964 | |
8cbbe86d | 3965 | __set_current_state(state); |
1da177e4 | 3966 | |
8cbbe86d AK |
3967 | spin_lock_irqsave(&q->lock, flags); |
3968 | __add_wait_queue(q, &wait); | |
3969 | spin_unlock(&q->lock); | |
3970 | timeout = schedule_timeout(timeout); | |
3971 | spin_lock_irq(&q->lock); | |
3972 | __remove_wait_queue(q, &wait); | |
3973 | spin_unlock_irqrestore(&q->lock, flags); | |
3974 | ||
3975 | return timeout; | |
3976 | } | |
3977 | ||
3978 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
3979 | { | |
3980 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 3981 | } |
1da177e4 LT |
3982 | EXPORT_SYMBOL(interruptible_sleep_on); |
3983 | ||
0fec171c | 3984 | long __sched |
95cdf3b7 | 3985 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3986 | { |
8cbbe86d | 3987 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 3988 | } |
1da177e4 LT |
3989 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3990 | ||
0fec171c | 3991 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3992 | { |
8cbbe86d | 3993 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 3994 | } |
1da177e4 LT |
3995 | EXPORT_SYMBOL(sleep_on); |
3996 | ||
0fec171c | 3997 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3998 | { |
8cbbe86d | 3999 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4000 | } |
1da177e4 LT |
4001 | EXPORT_SYMBOL(sleep_on_timeout); |
4002 | ||
b29739f9 IM |
4003 | #ifdef CONFIG_RT_MUTEXES |
4004 | ||
4005 | /* | |
4006 | * rt_mutex_setprio - set the current priority of a task | |
4007 | * @p: task | |
4008 | * @prio: prio value (kernel-internal form) | |
4009 | * | |
4010 | * This function changes the 'effective' priority of a task. It does | |
4011 | * not touch ->normal_prio like __setscheduler(). | |
4012 | * | |
4013 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4014 | */ | |
36c8b586 | 4015 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4016 | { |
4017 | unsigned long flags; | |
83b699ed | 4018 | int oldprio, on_rq, running; |
70b97a7f | 4019 | struct rq *rq; |
b29739f9 IM |
4020 | |
4021 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4022 | ||
4023 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4024 | update_rq_clock(rq); |
b29739f9 | 4025 | |
d5f9f942 | 4026 | oldprio = p->prio; |
dd41f596 | 4027 | on_rq = p->se.on_rq; |
051a1d1a | 4028 | running = task_current(rq, p); |
83b699ed | 4029 | if (on_rq) { |
69be72c1 | 4030 | dequeue_task(rq, p, 0); |
83b699ed SV |
4031 | if (running) |
4032 | p->sched_class->put_prev_task(rq, p); | |
4033 | } | |
dd41f596 IM |
4034 | |
4035 | if (rt_prio(prio)) | |
4036 | p->sched_class = &rt_sched_class; | |
4037 | else | |
4038 | p->sched_class = &fair_sched_class; | |
4039 | ||
b29739f9 IM |
4040 | p->prio = prio; |
4041 | ||
dd41f596 | 4042 | if (on_rq) { |
83b699ed SV |
4043 | if (running) |
4044 | p->sched_class->set_curr_task(rq); | |
8159f87e | 4045 | enqueue_task(rq, p, 0); |
b29739f9 IM |
4046 | /* |
4047 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4048 | * our priority decreased, or if we are not currently running on |
4049 | * this runqueue and our priority is higher than the current's | |
b29739f9 | 4050 | */ |
83b699ed | 4051 | if (running) { |
d5f9f942 AM |
4052 | if (p->prio > oldprio) |
4053 | resched_task(rq->curr); | |
dd41f596 IM |
4054 | } else { |
4055 | check_preempt_curr(rq, p); | |
4056 | } | |
b29739f9 IM |
4057 | } |
4058 | task_rq_unlock(rq, &flags); | |
4059 | } | |
4060 | ||
4061 | #endif | |
4062 | ||
36c8b586 | 4063 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4064 | { |
dd41f596 | 4065 | int old_prio, delta, on_rq; |
1da177e4 | 4066 | unsigned long flags; |
70b97a7f | 4067 | struct rq *rq; |
1da177e4 LT |
4068 | |
4069 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4070 | return; | |
4071 | /* | |
4072 | * We have to be careful, if called from sys_setpriority(), | |
4073 | * the task might be in the middle of scheduling on another CPU. | |
4074 | */ | |
4075 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4076 | update_rq_clock(rq); |
1da177e4 LT |
4077 | /* |
4078 | * The RT priorities are set via sched_setscheduler(), but we still | |
4079 | * allow the 'normal' nice value to be set - but as expected | |
4080 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4081 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4082 | */ |
e05606d3 | 4083 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4084 | p->static_prio = NICE_TO_PRIO(nice); |
4085 | goto out_unlock; | |
4086 | } | |
dd41f596 | 4087 | on_rq = p->se.on_rq; |
58e2d4ca | 4088 | if (on_rq) |
69be72c1 | 4089 | dequeue_task(rq, p, 0); |
1da177e4 | 4090 | |
1da177e4 | 4091 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4092 | set_load_weight(p); |
b29739f9 IM |
4093 | old_prio = p->prio; |
4094 | p->prio = effective_prio(p); | |
4095 | delta = p->prio - old_prio; | |
1da177e4 | 4096 | |
dd41f596 | 4097 | if (on_rq) { |
8159f87e | 4098 | enqueue_task(rq, p, 0); |
1da177e4 | 4099 | /* |
d5f9f942 AM |
4100 | * If the task increased its priority or is running and |
4101 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4102 | */ |
d5f9f942 | 4103 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4104 | resched_task(rq->curr); |
4105 | } | |
4106 | out_unlock: | |
4107 | task_rq_unlock(rq, &flags); | |
4108 | } | |
1da177e4 LT |
4109 | EXPORT_SYMBOL(set_user_nice); |
4110 | ||
e43379f1 MM |
4111 | /* |
4112 | * can_nice - check if a task can reduce its nice value | |
4113 | * @p: task | |
4114 | * @nice: nice value | |
4115 | */ | |
36c8b586 | 4116 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4117 | { |
024f4747 MM |
4118 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4119 | int nice_rlim = 20 - nice; | |
48f24c4d | 4120 | |
e43379f1 MM |
4121 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4122 | capable(CAP_SYS_NICE)); | |
4123 | } | |
4124 | ||
1da177e4 LT |
4125 | #ifdef __ARCH_WANT_SYS_NICE |
4126 | ||
4127 | /* | |
4128 | * sys_nice - change the priority of the current process. | |
4129 | * @increment: priority increment | |
4130 | * | |
4131 | * sys_setpriority is a more generic, but much slower function that | |
4132 | * does similar things. | |
4133 | */ | |
4134 | asmlinkage long sys_nice(int increment) | |
4135 | { | |
48f24c4d | 4136 | long nice, retval; |
1da177e4 LT |
4137 | |
4138 | /* | |
4139 | * Setpriority might change our priority at the same moment. | |
4140 | * We don't have to worry. Conceptually one call occurs first | |
4141 | * and we have a single winner. | |
4142 | */ | |
e43379f1 MM |
4143 | if (increment < -40) |
4144 | increment = -40; | |
1da177e4 LT |
4145 | if (increment > 40) |
4146 | increment = 40; | |
4147 | ||
4148 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4149 | if (nice < -20) | |
4150 | nice = -20; | |
4151 | if (nice > 19) | |
4152 | nice = 19; | |
4153 | ||
e43379f1 MM |
4154 | if (increment < 0 && !can_nice(current, nice)) |
4155 | return -EPERM; | |
4156 | ||
1da177e4 LT |
4157 | retval = security_task_setnice(current, nice); |
4158 | if (retval) | |
4159 | return retval; | |
4160 | ||
4161 | set_user_nice(current, nice); | |
4162 | return 0; | |
4163 | } | |
4164 | ||
4165 | #endif | |
4166 | ||
4167 | /** | |
4168 | * task_prio - return the priority value of a given task. | |
4169 | * @p: the task in question. | |
4170 | * | |
4171 | * This is the priority value as seen by users in /proc. | |
4172 | * RT tasks are offset by -200. Normal tasks are centered | |
4173 | * around 0, value goes from -16 to +15. | |
4174 | */ | |
36c8b586 | 4175 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4176 | { |
4177 | return p->prio - MAX_RT_PRIO; | |
4178 | } | |
4179 | ||
4180 | /** | |
4181 | * task_nice - return the nice value of a given task. | |
4182 | * @p: the task in question. | |
4183 | */ | |
36c8b586 | 4184 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4185 | { |
4186 | return TASK_NICE(p); | |
4187 | } | |
1da177e4 | 4188 | EXPORT_SYMBOL_GPL(task_nice); |
1da177e4 LT |
4189 | |
4190 | /** | |
4191 | * idle_cpu - is a given cpu idle currently? | |
4192 | * @cpu: the processor in question. | |
4193 | */ | |
4194 | int idle_cpu(int cpu) | |
4195 | { | |
4196 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4197 | } | |
4198 | ||
1da177e4 LT |
4199 | /** |
4200 | * idle_task - return the idle task for a given cpu. | |
4201 | * @cpu: the processor in question. | |
4202 | */ | |
36c8b586 | 4203 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4204 | { |
4205 | return cpu_rq(cpu)->idle; | |
4206 | } | |
4207 | ||
4208 | /** | |
4209 | * find_process_by_pid - find a process with a matching PID value. | |
4210 | * @pid: the pid in question. | |
4211 | */ | |
a9957449 | 4212 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4213 | { |
228ebcbe | 4214 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4215 | } |
4216 | ||
4217 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4218 | static void |
4219 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4220 | { |
dd41f596 | 4221 | BUG_ON(p->se.on_rq); |
48f24c4d | 4222 | |
1da177e4 | 4223 | p->policy = policy; |
dd41f596 IM |
4224 | switch (p->policy) { |
4225 | case SCHED_NORMAL: | |
4226 | case SCHED_BATCH: | |
4227 | case SCHED_IDLE: | |
4228 | p->sched_class = &fair_sched_class; | |
4229 | break; | |
4230 | case SCHED_FIFO: | |
4231 | case SCHED_RR: | |
4232 | p->sched_class = &rt_sched_class; | |
4233 | break; | |
4234 | } | |
4235 | ||
1da177e4 | 4236 | p->rt_priority = prio; |
b29739f9 IM |
4237 | p->normal_prio = normal_prio(p); |
4238 | /* we are holding p->pi_lock already */ | |
4239 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4240 | set_load_weight(p); |
1da177e4 LT |
4241 | } |
4242 | ||
4243 | /** | |
72fd4a35 | 4244 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4245 | * @p: the task in question. |
4246 | * @policy: new policy. | |
4247 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4248 | * |
72fd4a35 | 4249 | * NOTE that the task may be already dead. |
1da177e4 | 4250 | */ |
95cdf3b7 IM |
4251 | int sched_setscheduler(struct task_struct *p, int policy, |
4252 | struct sched_param *param) | |
1da177e4 | 4253 | { |
83b699ed | 4254 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4255 | unsigned long flags; |
70b97a7f | 4256 | struct rq *rq; |
1da177e4 | 4257 | |
66e5393a SR |
4258 | /* may grab non-irq protected spin_locks */ |
4259 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4260 | recheck: |
4261 | /* double check policy once rq lock held */ | |
4262 | if (policy < 0) | |
4263 | policy = oldpolicy = p->policy; | |
4264 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4265 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4266 | policy != SCHED_IDLE) | |
b0a9499c | 4267 | return -EINVAL; |
1da177e4 LT |
4268 | /* |
4269 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4270 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4271 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4272 | */ |
4273 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4274 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4275 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4276 | return -EINVAL; |
e05606d3 | 4277 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4278 | return -EINVAL; |
4279 | ||
37e4ab3f OC |
4280 | /* |
4281 | * Allow unprivileged RT tasks to decrease priority: | |
4282 | */ | |
4283 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4284 | if (rt_policy(policy)) { |
8dc3e909 | 4285 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4286 | |
4287 | if (!lock_task_sighand(p, &flags)) | |
4288 | return -ESRCH; | |
4289 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4290 | unlock_task_sighand(p, &flags); | |
4291 | ||
4292 | /* can't set/change the rt policy */ | |
4293 | if (policy != p->policy && !rlim_rtprio) | |
4294 | return -EPERM; | |
4295 | ||
4296 | /* can't increase priority */ | |
4297 | if (param->sched_priority > p->rt_priority && | |
4298 | param->sched_priority > rlim_rtprio) | |
4299 | return -EPERM; | |
4300 | } | |
dd41f596 IM |
4301 | /* |
4302 | * Like positive nice levels, dont allow tasks to | |
4303 | * move out of SCHED_IDLE either: | |
4304 | */ | |
4305 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4306 | return -EPERM; | |
5fe1d75f | 4307 | |
37e4ab3f OC |
4308 | /* can't change other user's priorities */ |
4309 | if ((current->euid != p->euid) && | |
4310 | (current->euid != p->uid)) | |
4311 | return -EPERM; | |
4312 | } | |
1da177e4 LT |
4313 | |
4314 | retval = security_task_setscheduler(p, policy, param); | |
4315 | if (retval) | |
4316 | return retval; | |
b29739f9 IM |
4317 | /* |
4318 | * make sure no PI-waiters arrive (or leave) while we are | |
4319 | * changing the priority of the task: | |
4320 | */ | |
4321 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4322 | /* |
4323 | * To be able to change p->policy safely, the apropriate | |
4324 | * runqueue lock must be held. | |
4325 | */ | |
b29739f9 | 4326 | rq = __task_rq_lock(p); |
1da177e4 LT |
4327 | /* recheck policy now with rq lock held */ |
4328 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4329 | policy = oldpolicy = -1; | |
b29739f9 IM |
4330 | __task_rq_unlock(rq); |
4331 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4332 | goto recheck; |
4333 | } | |
2daa3577 | 4334 | update_rq_clock(rq); |
dd41f596 | 4335 | on_rq = p->se.on_rq; |
051a1d1a | 4336 | running = task_current(rq, p); |
83b699ed | 4337 | if (on_rq) { |
2e1cb74a | 4338 | deactivate_task(rq, p, 0); |
83b699ed SV |
4339 | if (running) |
4340 | p->sched_class->put_prev_task(rq, p); | |
4341 | } | |
f6b53205 | 4342 | |
1da177e4 | 4343 | oldprio = p->prio; |
dd41f596 | 4344 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4345 | |
dd41f596 | 4346 | if (on_rq) { |
83b699ed SV |
4347 | if (running) |
4348 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4349 | activate_task(rq, p, 0); |
1da177e4 LT |
4350 | /* |
4351 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4352 | * our priority decreased, or if we are not currently running on |
4353 | * this runqueue and our priority is higher than the current's | |
1da177e4 | 4354 | */ |
83b699ed | 4355 | if (running) { |
d5f9f942 AM |
4356 | if (p->prio > oldprio) |
4357 | resched_task(rq->curr); | |
dd41f596 IM |
4358 | } else { |
4359 | check_preempt_curr(rq, p); | |
4360 | } | |
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 GH |
5863 | |
5864 | if (atomic_dec_and_test(&old_rd->refcount)) | |
5865 | kfree(old_rd); | |
5866 | } | |
5867 | ||
5868 | atomic_inc(&rd->refcount); | |
5869 | rq->rd = rd; | |
5870 | ||
0eab9146 | 5871 | for (class = sched_class_highest; class; class = class->next) { |
57d885fe GH |
5872 | if (class->join_domain) |
5873 | class->join_domain(rq); | |
0eab9146 | 5874 | } |
57d885fe GH |
5875 | |
5876 | spin_unlock_irqrestore(&rq->lock, flags); | |
5877 | } | |
5878 | ||
5879 | static void init_rootdomain(struct root_domain *rd, const cpumask_t *map) | |
5880 | { | |
5881 | memset(rd, 0, sizeof(*rd)); | |
5882 | ||
5883 | rd->span = *map; | |
5884 | cpus_and(rd->online, rd->span, cpu_online_map); | |
5885 | } | |
5886 | ||
5887 | static void init_defrootdomain(void) | |
5888 | { | |
5889 | cpumask_t cpus = CPU_MASK_ALL; | |
5890 | ||
5891 | init_rootdomain(&def_root_domain, &cpus); | |
5892 | atomic_set(&def_root_domain.refcount, 1); | |
5893 | } | |
5894 | ||
5895 | static struct root_domain *alloc_rootdomain(const cpumask_t *map) | |
5896 | { | |
5897 | struct root_domain *rd; | |
5898 | ||
5899 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5900 | if (!rd) | |
5901 | return NULL; | |
5902 | ||
5903 | init_rootdomain(rd, map); | |
5904 | ||
5905 | return rd; | |
5906 | } | |
5907 | ||
1da177e4 | 5908 | /* |
0eab9146 | 5909 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5910 | * hold the hotplug lock. |
5911 | */ | |
0eab9146 IM |
5912 | static void |
5913 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5914 | { |
70b97a7f | 5915 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5916 | struct sched_domain *tmp; |
5917 | ||
5918 | /* Remove the sched domains which do not contribute to scheduling. */ | |
5919 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
5920 | struct sched_domain *parent = tmp->parent; | |
5921 | if (!parent) | |
5922 | break; | |
1a848870 | 5923 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5924 | tmp->parent = parent->parent; |
1a848870 SS |
5925 | if (parent->parent) |
5926 | parent->parent->child = tmp; | |
5927 | } | |
245af2c7 SS |
5928 | } |
5929 | ||
1a848870 | 5930 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 5931 | sd = sd->parent; |
1a848870 SS |
5932 | if (sd) |
5933 | sd->child = NULL; | |
5934 | } | |
1da177e4 LT |
5935 | |
5936 | sched_domain_debug(sd, cpu); | |
5937 | ||
57d885fe | 5938 | rq_attach_root(rq, rd); |
674311d5 | 5939 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
5940 | } |
5941 | ||
5942 | /* cpus with isolated domains */ | |
67af63a6 | 5943 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
5944 | |
5945 | /* Setup the mask of cpus configured for isolated domains */ | |
5946 | static int __init isolated_cpu_setup(char *str) | |
5947 | { | |
5948 | int ints[NR_CPUS], i; | |
5949 | ||
5950 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
5951 | cpus_clear(cpu_isolated_map); | |
5952 | for (i = 1; i <= ints[0]; i++) | |
5953 | if (ints[i] < NR_CPUS) | |
5954 | cpu_set(ints[i], cpu_isolated_map); | |
5955 | return 1; | |
5956 | } | |
5957 | ||
8927f494 | 5958 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
5959 | |
5960 | /* | |
6711cab4 SS |
5961 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
5962 | * to a function which identifies what group(along with sched group) a CPU | |
5963 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
5964 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
5965 | * |
5966 | * init_sched_build_groups will build a circular linked list of the groups | |
5967 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5968 | * and ->cpu_power to 0. | |
5969 | */ | |
a616058b | 5970 | static void |
6711cab4 SS |
5971 | init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map, |
5972 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, | |
5973 | struct sched_group **sg)) | |
1da177e4 LT |
5974 | { |
5975 | struct sched_group *first = NULL, *last = NULL; | |
5976 | cpumask_t covered = CPU_MASK_NONE; | |
5977 | int i; | |
5978 | ||
5979 | for_each_cpu_mask(i, span) { | |
6711cab4 SS |
5980 | struct sched_group *sg; |
5981 | int group = group_fn(i, cpu_map, &sg); | |
1da177e4 LT |
5982 | int j; |
5983 | ||
5984 | if (cpu_isset(i, covered)) | |
5985 | continue; | |
5986 | ||
5987 | sg->cpumask = CPU_MASK_NONE; | |
5517d86b | 5988 | sg->__cpu_power = 0; |
1da177e4 LT |
5989 | |
5990 | for_each_cpu_mask(j, span) { | |
6711cab4 | 5991 | if (group_fn(j, cpu_map, NULL) != group) |
1da177e4 LT |
5992 | continue; |
5993 | ||
5994 | cpu_set(j, covered); | |
5995 | cpu_set(j, sg->cpumask); | |
5996 | } | |
5997 | if (!first) | |
5998 | first = sg; | |
5999 | if (last) | |
6000 | last->next = sg; | |
6001 | last = sg; | |
6002 | } | |
6003 | last->next = first; | |
6004 | } | |
6005 | ||
9c1cfda2 | 6006 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6007 | |
9c1cfda2 | 6008 | #ifdef CONFIG_NUMA |
198e2f18 | 6009 | |
9c1cfda2 JH |
6010 | /** |
6011 | * find_next_best_node - find the next node to include in a sched_domain | |
6012 | * @node: node whose sched_domain we're building | |
6013 | * @used_nodes: nodes already in the sched_domain | |
6014 | * | |
41a2d6cf | 6015 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6016 | * finds the closest node not already in the @used_nodes map. |
6017 | * | |
6018 | * Should use nodemask_t. | |
6019 | */ | |
6020 | static int find_next_best_node(int node, unsigned long *used_nodes) | |
6021 | { | |
6022 | int i, n, val, min_val, best_node = 0; | |
6023 | ||
6024 | min_val = INT_MAX; | |
6025 | ||
6026 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6027 | /* Start at @node */ | |
6028 | n = (node + i) % MAX_NUMNODES; | |
6029 | ||
6030 | if (!nr_cpus_node(n)) | |
6031 | continue; | |
6032 | ||
6033 | /* Skip already used nodes */ | |
6034 | if (test_bit(n, used_nodes)) | |
6035 | continue; | |
6036 | ||
6037 | /* Simple min distance search */ | |
6038 | val = node_distance(node, n); | |
6039 | ||
6040 | if (val < min_val) { | |
6041 | min_val = val; | |
6042 | best_node = n; | |
6043 | } | |
6044 | } | |
6045 | ||
6046 | set_bit(best_node, used_nodes); | |
6047 | return best_node; | |
6048 | } | |
6049 | ||
6050 | /** | |
6051 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6052 | * @node: node whose cpumask we're constructing | |
6053 | * @size: number of nodes to include in this span | |
6054 | * | |
41a2d6cf | 6055 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6056 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6057 | * out optimally. | |
6058 | */ | |
6059 | static cpumask_t sched_domain_node_span(int node) | |
6060 | { | |
9c1cfda2 | 6061 | DECLARE_BITMAP(used_nodes, MAX_NUMNODES); |
48f24c4d IM |
6062 | cpumask_t span, nodemask; |
6063 | int i; | |
9c1cfda2 JH |
6064 | |
6065 | cpus_clear(span); | |
6066 | bitmap_zero(used_nodes, MAX_NUMNODES); | |
6067 | ||
6068 | nodemask = node_to_cpumask(node); | |
6069 | cpus_or(span, span, nodemask); | |
6070 | set_bit(node, used_nodes); | |
6071 | ||
6072 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
6073 | int next_node = find_next_best_node(node, used_nodes); | |
48f24c4d | 6074 | |
9c1cfda2 JH |
6075 | nodemask = node_to_cpumask(next_node); |
6076 | cpus_or(span, span, nodemask); | |
6077 | } | |
6078 | ||
6079 | return span; | |
6080 | } | |
6081 | #endif | |
6082 | ||
5c45bf27 | 6083 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6084 | |
9c1cfda2 | 6085 | /* |
48f24c4d | 6086 | * SMT sched-domains: |
9c1cfda2 | 6087 | */ |
1da177e4 LT |
6088 | #ifdef CONFIG_SCHED_SMT |
6089 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 6090 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 6091 | |
41a2d6cf IM |
6092 | static int |
6093 | cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1da177e4 | 6094 | { |
6711cab4 SS |
6095 | if (sg) |
6096 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
6097 | return cpu; |
6098 | } | |
6099 | #endif | |
6100 | ||
48f24c4d IM |
6101 | /* |
6102 | * multi-core sched-domains: | |
6103 | */ | |
1e9f28fa SS |
6104 | #ifdef CONFIG_SCHED_MC |
6105 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 6106 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
6107 | #endif |
6108 | ||
6109 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf IM |
6110 | static int |
6111 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1e9f28fa | 6112 | { |
6711cab4 | 6113 | int group; |
d5a7430d | 6114 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 6115 | cpus_and(mask, mask, *cpu_map); |
6711cab4 SS |
6116 | group = first_cpu(mask); |
6117 | if (sg) | |
6118 | *sg = &per_cpu(sched_group_core, group); | |
6119 | return group; | |
1e9f28fa SS |
6120 | } |
6121 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf IM |
6122 | static int |
6123 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1e9f28fa | 6124 | { |
6711cab4 SS |
6125 | if (sg) |
6126 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
6127 | return cpu; |
6128 | } | |
6129 | #endif | |
6130 | ||
1da177e4 | 6131 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 6132 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 6133 | |
41a2d6cf IM |
6134 | static int |
6135 | cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1da177e4 | 6136 | { |
6711cab4 | 6137 | int group; |
48f24c4d | 6138 | #ifdef CONFIG_SCHED_MC |
1e9f28fa | 6139 | cpumask_t mask = cpu_coregroup_map(cpu); |
a616058b | 6140 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 6141 | group = first_cpu(mask); |
1e9f28fa | 6142 | #elif defined(CONFIG_SCHED_SMT) |
d5a7430d | 6143 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 6144 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 6145 | group = first_cpu(mask); |
1da177e4 | 6146 | #else |
6711cab4 | 6147 | group = cpu; |
1da177e4 | 6148 | #endif |
6711cab4 SS |
6149 | if (sg) |
6150 | *sg = &per_cpu(sched_group_phys, group); | |
6151 | return group; | |
1da177e4 LT |
6152 | } |
6153 | ||
6154 | #ifdef CONFIG_NUMA | |
1da177e4 | 6155 | /* |
9c1cfda2 JH |
6156 | * The init_sched_build_groups can't handle what we want to do with node |
6157 | * groups, so roll our own. Now each node has its own list of groups which | |
6158 | * gets dynamically allocated. | |
1da177e4 | 6159 | */ |
9c1cfda2 | 6160 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
d1b55138 | 6161 | static struct sched_group **sched_group_nodes_bycpu[NR_CPUS]; |
1da177e4 | 6162 | |
9c1cfda2 | 6163 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 6164 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 6165 | |
6711cab4 SS |
6166 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
6167 | struct sched_group **sg) | |
9c1cfda2 | 6168 | { |
6711cab4 SS |
6169 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu)); |
6170 | int group; | |
6171 | ||
6172 | cpus_and(nodemask, nodemask, *cpu_map); | |
6173 | group = first_cpu(nodemask); | |
6174 | ||
6175 | if (sg) | |
6176 | *sg = &per_cpu(sched_group_allnodes, group); | |
6177 | return group; | |
1da177e4 | 6178 | } |
6711cab4 | 6179 | |
08069033 SS |
6180 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6181 | { | |
6182 | struct sched_group *sg = group_head; | |
6183 | int j; | |
6184 | ||
6185 | if (!sg) | |
6186 | return; | |
3a5c359a AK |
6187 | do { |
6188 | for_each_cpu_mask(j, sg->cpumask) { | |
6189 | struct sched_domain *sd; | |
08069033 | 6190 | |
3a5c359a AK |
6191 | sd = &per_cpu(phys_domains, j); |
6192 | if (j != first_cpu(sd->groups->cpumask)) { | |
6193 | /* | |
6194 | * Only add "power" once for each | |
6195 | * physical package. | |
6196 | */ | |
6197 | continue; | |
6198 | } | |
08069033 | 6199 | |
3a5c359a AK |
6200 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
6201 | } | |
6202 | sg = sg->next; | |
6203 | } while (sg != group_head); | |
08069033 | 6204 | } |
1da177e4 LT |
6205 | #endif |
6206 | ||
a616058b | 6207 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6208 | /* Free memory allocated for various sched_group structures */ |
6209 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6210 | { | |
a616058b | 6211 | int cpu, i; |
51888ca2 SV |
6212 | |
6213 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
6214 | struct sched_group **sched_group_nodes |
6215 | = sched_group_nodes_bycpu[cpu]; | |
6216 | ||
51888ca2 SV |
6217 | if (!sched_group_nodes) |
6218 | continue; | |
6219 | ||
6220 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6221 | cpumask_t nodemask = node_to_cpumask(i); | |
6222 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; | |
6223 | ||
6224 | cpus_and(nodemask, nodemask, *cpu_map); | |
6225 | if (cpus_empty(nodemask)) | |
6226 | continue; | |
6227 | ||
6228 | if (sg == NULL) | |
6229 | continue; | |
6230 | sg = sg->next; | |
6231 | next_sg: | |
6232 | oldsg = sg; | |
6233 | sg = sg->next; | |
6234 | kfree(oldsg); | |
6235 | if (oldsg != sched_group_nodes[i]) | |
6236 | goto next_sg; | |
6237 | } | |
6238 | kfree(sched_group_nodes); | |
6239 | sched_group_nodes_bycpu[cpu] = NULL; | |
6240 | } | |
51888ca2 | 6241 | } |
a616058b SS |
6242 | #else |
6243 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6244 | { | |
6245 | } | |
6246 | #endif | |
51888ca2 | 6247 | |
89c4710e SS |
6248 | /* |
6249 | * Initialize sched groups cpu_power. | |
6250 | * | |
6251 | * cpu_power indicates the capacity of sched group, which is used while | |
6252 | * distributing the load between different sched groups in a sched domain. | |
6253 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6254 | * there are asymmetries in the topology. If there are asymmetries, group | |
6255 | * having more cpu_power will pickup more load compared to the group having | |
6256 | * less cpu_power. | |
6257 | * | |
6258 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
6259 | * the maximum number of tasks a group can handle in the presence of other idle | |
6260 | * or lightly loaded groups in the same sched domain. | |
6261 | */ | |
6262 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6263 | { | |
6264 | struct sched_domain *child; | |
6265 | struct sched_group *group; | |
6266 | ||
6267 | WARN_ON(!sd || !sd->groups); | |
6268 | ||
6269 | if (cpu != first_cpu(sd->groups->cpumask)) | |
6270 | return; | |
6271 | ||
6272 | child = sd->child; | |
6273 | ||
5517d86b ED |
6274 | sd->groups->__cpu_power = 0; |
6275 | ||
89c4710e SS |
6276 | /* |
6277 | * For perf policy, if the groups in child domain share resources | |
6278 | * (for example cores sharing some portions of the cache hierarchy | |
6279 | * or SMT), then set this domain groups cpu_power such that each group | |
6280 | * can handle only one task, when there are other idle groups in the | |
6281 | * same sched domain. | |
6282 | */ | |
6283 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
6284 | (child->flags & | |
6285 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 6286 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
6287 | return; |
6288 | } | |
6289 | ||
89c4710e SS |
6290 | /* |
6291 | * add cpu_power of each child group to this groups cpu_power | |
6292 | */ | |
6293 | group = child->groups; | |
6294 | do { | |
5517d86b | 6295 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
6296 | group = group->next; |
6297 | } while (group != child->groups); | |
6298 | } | |
6299 | ||
1da177e4 | 6300 | /* |
1a20ff27 DG |
6301 | * Build sched domains for a given set of cpus and attach the sched domains |
6302 | * to the individual cpus | |
1da177e4 | 6303 | */ |
51888ca2 | 6304 | static int build_sched_domains(const cpumask_t *cpu_map) |
1da177e4 LT |
6305 | { |
6306 | int i; | |
57d885fe | 6307 | struct root_domain *rd; |
d1b55138 JH |
6308 | #ifdef CONFIG_NUMA |
6309 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 6310 | int sd_allnodes = 0; |
d1b55138 JH |
6311 | |
6312 | /* | |
6313 | * Allocate the per-node list of sched groups | |
6314 | */ | |
5cf9f062 | 6315 | sched_group_nodes = kcalloc(MAX_NUMNODES, sizeof(struct sched_group *), |
41a2d6cf | 6316 | GFP_KERNEL); |
d1b55138 JH |
6317 | if (!sched_group_nodes) { |
6318 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 6319 | return -ENOMEM; |
d1b55138 JH |
6320 | } |
6321 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
6322 | #endif | |
1da177e4 | 6323 | |
57d885fe GH |
6324 | rd = alloc_rootdomain(cpu_map); |
6325 | if (!rd) { | |
6326 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
6327 | return -ENOMEM; | |
6328 | } | |
6329 | ||
1da177e4 | 6330 | /* |
1a20ff27 | 6331 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 6332 | */ |
1a20ff27 | 6333 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6334 | struct sched_domain *sd = NULL, *p; |
6335 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(i)); | |
6336 | ||
1a20ff27 | 6337 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6338 | |
6339 | #ifdef CONFIG_NUMA | |
dd41f596 IM |
6340 | if (cpus_weight(*cpu_map) > |
6341 | SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) { | |
9c1cfda2 JH |
6342 | sd = &per_cpu(allnodes_domains, i); |
6343 | *sd = SD_ALLNODES_INIT; | |
6344 | sd->span = *cpu_map; | |
6711cab4 | 6345 | cpu_to_allnodes_group(i, cpu_map, &sd->groups); |
9c1cfda2 | 6346 | p = sd; |
6711cab4 | 6347 | sd_allnodes = 1; |
9c1cfda2 JH |
6348 | } else |
6349 | p = NULL; | |
6350 | ||
1da177e4 | 6351 | sd = &per_cpu(node_domains, i); |
1da177e4 | 6352 | *sd = SD_NODE_INIT; |
9c1cfda2 JH |
6353 | sd->span = sched_domain_node_span(cpu_to_node(i)); |
6354 | sd->parent = p; | |
1a848870 SS |
6355 | if (p) |
6356 | p->child = sd; | |
9c1cfda2 | 6357 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
6358 | #endif |
6359 | ||
6360 | p = sd; | |
6361 | sd = &per_cpu(phys_domains, i); | |
1da177e4 LT |
6362 | *sd = SD_CPU_INIT; |
6363 | sd->span = nodemask; | |
6364 | sd->parent = p; | |
1a848870 SS |
6365 | if (p) |
6366 | p->child = sd; | |
6711cab4 | 6367 | cpu_to_phys_group(i, cpu_map, &sd->groups); |
1da177e4 | 6368 | |
1e9f28fa SS |
6369 | #ifdef CONFIG_SCHED_MC |
6370 | p = sd; | |
6371 | sd = &per_cpu(core_domains, i); | |
1e9f28fa SS |
6372 | *sd = SD_MC_INIT; |
6373 | sd->span = cpu_coregroup_map(i); | |
6374 | cpus_and(sd->span, sd->span, *cpu_map); | |
6375 | sd->parent = p; | |
1a848870 | 6376 | p->child = sd; |
6711cab4 | 6377 | cpu_to_core_group(i, cpu_map, &sd->groups); |
1e9f28fa SS |
6378 | #endif |
6379 | ||
1da177e4 LT |
6380 | #ifdef CONFIG_SCHED_SMT |
6381 | p = sd; | |
6382 | sd = &per_cpu(cpu_domains, i); | |
1da177e4 | 6383 | *sd = SD_SIBLING_INIT; |
d5a7430d | 6384 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6385 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 6386 | sd->parent = p; |
1a848870 | 6387 | p->child = sd; |
6711cab4 | 6388 | cpu_to_cpu_group(i, cpu_map, &sd->groups); |
1da177e4 LT |
6389 | #endif |
6390 | } | |
6391 | ||
6392 | #ifdef CONFIG_SCHED_SMT | |
6393 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 6394 | for_each_cpu_mask(i, *cpu_map) { |
d5a7430d | 6395 | cpumask_t this_sibling_map = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6396 | cpus_and(this_sibling_map, this_sibling_map, *cpu_map); |
1da177e4 LT |
6397 | if (i != first_cpu(this_sibling_map)) |
6398 | continue; | |
6399 | ||
dd41f596 IM |
6400 | init_sched_build_groups(this_sibling_map, cpu_map, |
6401 | &cpu_to_cpu_group); | |
1da177e4 LT |
6402 | } |
6403 | #endif | |
6404 | ||
1e9f28fa SS |
6405 | #ifdef CONFIG_SCHED_MC |
6406 | /* Set up multi-core groups */ | |
6407 | for_each_cpu_mask(i, *cpu_map) { | |
6408 | cpumask_t this_core_map = cpu_coregroup_map(i); | |
6409 | cpus_and(this_core_map, this_core_map, *cpu_map); | |
6410 | if (i != first_cpu(this_core_map)) | |
6411 | continue; | |
dd41f596 IM |
6412 | init_sched_build_groups(this_core_map, cpu_map, |
6413 | &cpu_to_core_group); | |
1e9f28fa SS |
6414 | } |
6415 | #endif | |
6416 | ||
1da177e4 LT |
6417 | /* Set up physical groups */ |
6418 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6419 | cpumask_t nodemask = node_to_cpumask(i); | |
6420 | ||
1a20ff27 | 6421 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6422 | if (cpus_empty(nodemask)) |
6423 | continue; | |
6424 | ||
6711cab4 | 6425 | init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group); |
1da177e4 LT |
6426 | } |
6427 | ||
6428 | #ifdef CONFIG_NUMA | |
6429 | /* Set up node groups */ | |
6711cab4 | 6430 | if (sd_allnodes) |
dd41f596 IM |
6431 | init_sched_build_groups(*cpu_map, cpu_map, |
6432 | &cpu_to_allnodes_group); | |
9c1cfda2 JH |
6433 | |
6434 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6435 | /* Set up node groups */ | |
6436 | struct sched_group *sg, *prev; | |
6437 | cpumask_t nodemask = node_to_cpumask(i); | |
6438 | cpumask_t domainspan; | |
6439 | cpumask_t covered = CPU_MASK_NONE; | |
6440 | int j; | |
6441 | ||
6442 | cpus_and(nodemask, nodemask, *cpu_map); | |
d1b55138 JH |
6443 | if (cpus_empty(nodemask)) { |
6444 | sched_group_nodes[i] = NULL; | |
9c1cfda2 | 6445 | continue; |
d1b55138 | 6446 | } |
9c1cfda2 JH |
6447 | |
6448 | domainspan = sched_domain_node_span(i); | |
6449 | cpus_and(domainspan, domainspan, *cpu_map); | |
6450 | ||
15f0b676 | 6451 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
6452 | if (!sg) { |
6453 | printk(KERN_WARNING "Can not alloc domain group for " | |
6454 | "node %d\n", i); | |
6455 | goto error; | |
6456 | } | |
9c1cfda2 JH |
6457 | sched_group_nodes[i] = sg; |
6458 | for_each_cpu_mask(j, nodemask) { | |
6459 | struct sched_domain *sd; | |
9761eea8 | 6460 | |
9c1cfda2 JH |
6461 | sd = &per_cpu(node_domains, j); |
6462 | sd->groups = sg; | |
9c1cfda2 | 6463 | } |
5517d86b | 6464 | sg->__cpu_power = 0; |
9c1cfda2 | 6465 | sg->cpumask = nodemask; |
51888ca2 | 6466 | sg->next = sg; |
9c1cfda2 JH |
6467 | cpus_or(covered, covered, nodemask); |
6468 | prev = sg; | |
6469 | ||
6470 | for (j = 0; j < MAX_NUMNODES; j++) { | |
6471 | cpumask_t tmp, notcovered; | |
6472 | int n = (i + j) % MAX_NUMNODES; | |
6473 | ||
6474 | cpus_complement(notcovered, covered); | |
6475 | cpus_and(tmp, notcovered, *cpu_map); | |
6476 | cpus_and(tmp, tmp, domainspan); | |
6477 | if (cpus_empty(tmp)) | |
6478 | break; | |
6479 | ||
6480 | nodemask = node_to_cpumask(n); | |
6481 | cpus_and(tmp, tmp, nodemask); | |
6482 | if (cpus_empty(tmp)) | |
6483 | continue; | |
6484 | ||
15f0b676 SV |
6485 | sg = kmalloc_node(sizeof(struct sched_group), |
6486 | GFP_KERNEL, i); | |
9c1cfda2 JH |
6487 | if (!sg) { |
6488 | printk(KERN_WARNING | |
6489 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 6490 | goto error; |
9c1cfda2 | 6491 | } |
5517d86b | 6492 | sg->__cpu_power = 0; |
9c1cfda2 | 6493 | sg->cpumask = tmp; |
51888ca2 | 6494 | sg->next = prev->next; |
9c1cfda2 JH |
6495 | cpus_or(covered, covered, tmp); |
6496 | prev->next = sg; | |
6497 | prev = sg; | |
6498 | } | |
9c1cfda2 | 6499 | } |
1da177e4 LT |
6500 | #endif |
6501 | ||
6502 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 6503 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 6504 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6505 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
6506 | ||
89c4710e | 6507 | init_sched_groups_power(i, sd); |
5c45bf27 | 6508 | } |
1da177e4 | 6509 | #endif |
1e9f28fa | 6510 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 6511 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6512 | struct sched_domain *sd = &per_cpu(core_domains, i); |
6513 | ||
89c4710e | 6514 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
6515 | } |
6516 | #endif | |
1e9f28fa | 6517 | |
5c45bf27 | 6518 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6519 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
6520 | ||
89c4710e | 6521 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6522 | } |
6523 | ||
9c1cfda2 | 6524 | #ifdef CONFIG_NUMA |
08069033 SS |
6525 | for (i = 0; i < MAX_NUMNODES; i++) |
6526 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 6527 | |
6711cab4 SS |
6528 | if (sd_allnodes) { |
6529 | struct sched_group *sg; | |
f712c0c7 | 6530 | |
6711cab4 | 6531 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg); |
f712c0c7 SS |
6532 | init_numa_sched_groups_power(sg); |
6533 | } | |
9c1cfda2 JH |
6534 | #endif |
6535 | ||
1da177e4 | 6536 | /* Attach the domains */ |
1a20ff27 | 6537 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6538 | struct sched_domain *sd; |
6539 | #ifdef CONFIG_SCHED_SMT | |
6540 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
6541 | #elif defined(CONFIG_SCHED_MC) |
6542 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
6543 | #else |
6544 | sd = &per_cpu(phys_domains, i); | |
6545 | #endif | |
57d885fe | 6546 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 6547 | } |
51888ca2 SV |
6548 | |
6549 | return 0; | |
6550 | ||
a616058b | 6551 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6552 | error: |
6553 | free_sched_groups(cpu_map); | |
6554 | return -ENOMEM; | |
a616058b | 6555 | #endif |
1da177e4 | 6556 | } |
029190c5 PJ |
6557 | |
6558 | static cpumask_t *doms_cur; /* current sched domains */ | |
6559 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
6560 | ||
6561 | /* | |
6562 | * Special case: If a kmalloc of a doms_cur partition (array of | |
6563 | * cpumask_t) fails, then fallback to a single sched domain, | |
6564 | * as determined by the single cpumask_t fallback_doms. | |
6565 | */ | |
6566 | static cpumask_t fallback_doms; | |
6567 | ||
1a20ff27 | 6568 | /* |
41a2d6cf | 6569 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6570 | * For now this just excludes isolated cpus, but could be used to |
6571 | * exclude other special cases in the future. | |
1a20ff27 | 6572 | */ |
51888ca2 | 6573 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 6574 | { |
7378547f MM |
6575 | int err; |
6576 | ||
029190c5 PJ |
6577 | ndoms_cur = 1; |
6578 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
6579 | if (!doms_cur) | |
6580 | doms_cur = &fallback_doms; | |
6581 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
7378547f | 6582 | err = build_sched_domains(doms_cur); |
6382bc90 | 6583 | register_sched_domain_sysctl(); |
7378547f MM |
6584 | |
6585 | return err; | |
1a20ff27 DG |
6586 | } |
6587 | ||
6588 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map) | |
1da177e4 | 6589 | { |
51888ca2 | 6590 | free_sched_groups(cpu_map); |
9c1cfda2 | 6591 | } |
1da177e4 | 6592 | |
1a20ff27 DG |
6593 | /* |
6594 | * Detach sched domains from a group of cpus specified in cpu_map | |
6595 | * These cpus will now be attached to the NULL domain | |
6596 | */ | |
858119e1 | 6597 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6598 | { |
6599 | int i; | |
6600 | ||
6382bc90 MM |
6601 | unregister_sched_domain_sysctl(); |
6602 | ||
1a20ff27 | 6603 | for_each_cpu_mask(i, *cpu_map) |
57d885fe | 6604 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 DG |
6605 | synchronize_sched(); |
6606 | arch_destroy_sched_domains(cpu_map); | |
6607 | } | |
6608 | ||
029190c5 PJ |
6609 | /* |
6610 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6611 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6612 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6613 | * It destroys each deleted domain and builds each new domain. | |
6614 | * | |
6615 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
41a2d6cf IM |
6616 | * The masks don't intersect (don't overlap.) We should setup one |
6617 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6618 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6619 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6620 | * it as it is. | |
6621 | * | |
41a2d6cf IM |
6622 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
6623 | * ownership of it and will kfree it when done with it. If the caller | |
029190c5 PJ |
6624 | * failed the kmalloc call, then it can pass in doms_new == NULL, |
6625 | * and partition_sched_domains() will fallback to the single partition | |
6626 | * 'fallback_doms'. | |
6627 | * | |
6628 | * Call with hotplug lock held | |
6629 | */ | |
6630 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new) | |
6631 | { | |
6632 | int i, j; | |
6633 | ||
a1835615 SV |
6634 | lock_doms_cur(); |
6635 | ||
7378547f MM |
6636 | /* always unregister in case we don't destroy any domains */ |
6637 | unregister_sched_domain_sysctl(); | |
6638 | ||
029190c5 PJ |
6639 | if (doms_new == NULL) { |
6640 | ndoms_new = 1; | |
6641 | doms_new = &fallback_doms; | |
6642 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
6643 | } | |
6644 | ||
6645 | /* Destroy deleted domains */ | |
6646 | for (i = 0; i < ndoms_cur; i++) { | |
6647 | for (j = 0; j < ndoms_new; j++) { | |
6648 | if (cpus_equal(doms_cur[i], doms_new[j])) | |
6649 | goto match1; | |
6650 | } | |
6651 | /* no match - a current sched domain not in new doms_new[] */ | |
6652 | detach_destroy_domains(doms_cur + i); | |
6653 | match1: | |
6654 | ; | |
6655 | } | |
6656 | ||
6657 | /* Build new domains */ | |
6658 | for (i = 0; i < ndoms_new; i++) { | |
6659 | for (j = 0; j < ndoms_cur; j++) { | |
6660 | if (cpus_equal(doms_new[i], doms_cur[j])) | |
6661 | goto match2; | |
6662 | } | |
6663 | /* no match - add a new doms_new */ | |
6664 | build_sched_domains(doms_new + i); | |
6665 | match2: | |
6666 | ; | |
6667 | } | |
6668 | ||
6669 | /* Remember the new sched domains */ | |
6670 | if (doms_cur != &fallback_doms) | |
6671 | kfree(doms_cur); | |
6672 | doms_cur = doms_new; | |
6673 | ndoms_cur = ndoms_new; | |
7378547f MM |
6674 | |
6675 | register_sched_domain_sysctl(); | |
a1835615 SV |
6676 | |
6677 | unlock_doms_cur(); | |
029190c5 PJ |
6678 | } |
6679 | ||
5c45bf27 | 6680 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
6707de00 | 6681 | static int arch_reinit_sched_domains(void) |
5c45bf27 SS |
6682 | { |
6683 | int err; | |
6684 | ||
95402b38 | 6685 | get_online_cpus(); |
5c45bf27 SS |
6686 | detach_destroy_domains(&cpu_online_map); |
6687 | err = arch_init_sched_domains(&cpu_online_map); | |
95402b38 | 6688 | put_online_cpus(); |
5c45bf27 SS |
6689 | |
6690 | return err; | |
6691 | } | |
6692 | ||
6693 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6694 | { | |
6695 | int ret; | |
6696 | ||
6697 | if (buf[0] != '0' && buf[0] != '1') | |
6698 | return -EINVAL; | |
6699 | ||
6700 | if (smt) | |
6701 | sched_smt_power_savings = (buf[0] == '1'); | |
6702 | else | |
6703 | sched_mc_power_savings = (buf[0] == '1'); | |
6704 | ||
6705 | ret = arch_reinit_sched_domains(); | |
6706 | ||
6707 | return ret ? ret : count; | |
6708 | } | |
6709 | ||
5c45bf27 SS |
6710 | #ifdef CONFIG_SCHED_MC |
6711 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
6712 | { | |
6713 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
6714 | } | |
48f24c4d IM |
6715 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
6716 | const char *buf, size_t count) | |
5c45bf27 SS |
6717 | { |
6718 | return sched_power_savings_store(buf, count, 0); | |
6719 | } | |
6707de00 AB |
6720 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
6721 | sched_mc_power_savings_store); | |
5c45bf27 SS |
6722 | #endif |
6723 | ||
6724 | #ifdef CONFIG_SCHED_SMT | |
6725 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
6726 | { | |
6727 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
6728 | } | |
48f24c4d IM |
6729 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
6730 | const char *buf, size_t count) | |
5c45bf27 SS |
6731 | { |
6732 | return sched_power_savings_store(buf, count, 1); | |
6733 | } | |
6707de00 AB |
6734 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
6735 | sched_smt_power_savings_store); | |
6736 | #endif | |
6737 | ||
6738 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
6739 | { | |
6740 | int err = 0; | |
6741 | ||
6742 | #ifdef CONFIG_SCHED_SMT | |
6743 | if (smt_capable()) | |
6744 | err = sysfs_create_file(&cls->kset.kobj, | |
6745 | &attr_sched_smt_power_savings.attr); | |
6746 | #endif | |
6747 | #ifdef CONFIG_SCHED_MC | |
6748 | if (!err && mc_capable()) | |
6749 | err = sysfs_create_file(&cls->kset.kobj, | |
6750 | &attr_sched_mc_power_savings.attr); | |
6751 | #endif | |
6752 | return err; | |
6753 | } | |
5c45bf27 SS |
6754 | #endif |
6755 | ||
1da177e4 | 6756 | /* |
41a2d6cf | 6757 | * Force a reinitialization of the sched domains hierarchy. The domains |
1da177e4 | 6758 | * and groups cannot be updated in place without racing with the balancing |
41c7ce9a | 6759 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
6760 | * which will prevent rebalancing while the sched domains are recalculated. |
6761 | */ | |
6762 | static int update_sched_domains(struct notifier_block *nfb, | |
6763 | unsigned long action, void *hcpu) | |
6764 | { | |
1da177e4 LT |
6765 | switch (action) { |
6766 | case CPU_UP_PREPARE: | |
8bb78442 | 6767 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 6768 | case CPU_DOWN_PREPARE: |
8bb78442 | 6769 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 6770 | detach_destroy_domains(&cpu_online_map); |
1da177e4 LT |
6771 | return NOTIFY_OK; |
6772 | ||
6773 | case CPU_UP_CANCELED: | |
8bb78442 | 6774 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 6775 | case CPU_DOWN_FAILED: |
8bb78442 | 6776 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 6777 | case CPU_ONLINE: |
8bb78442 | 6778 | case CPU_ONLINE_FROZEN: |
1da177e4 | 6779 | case CPU_DEAD: |
8bb78442 | 6780 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
6781 | /* |
6782 | * Fall through and re-initialise the domains. | |
6783 | */ | |
6784 | break; | |
6785 | default: | |
6786 | return NOTIFY_DONE; | |
6787 | } | |
6788 | ||
6789 | /* The hotplug lock is already held by cpu_up/cpu_down */ | |
1a20ff27 | 6790 | arch_init_sched_domains(&cpu_online_map); |
1da177e4 LT |
6791 | |
6792 | return NOTIFY_OK; | |
6793 | } | |
1da177e4 LT |
6794 | |
6795 | void __init sched_init_smp(void) | |
6796 | { | |
5c1e1767 NP |
6797 | cpumask_t non_isolated_cpus; |
6798 | ||
95402b38 | 6799 | get_online_cpus(); |
1a20ff27 | 6800 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 6801 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
6802 | if (cpus_empty(non_isolated_cpus)) |
6803 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
95402b38 | 6804 | put_online_cpus(); |
1da177e4 LT |
6805 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
6806 | hotcpu_notifier(update_sched_domains, 0); | |
5c1e1767 NP |
6807 | |
6808 | /* Move init over to a non-isolated CPU */ | |
6809 | if (set_cpus_allowed(current, non_isolated_cpus) < 0) | |
6810 | BUG(); | |
19978ca6 | 6811 | sched_init_granularity(); |
6b2d7700 SV |
6812 | |
6813 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6814 | if (nr_cpu_ids == 1) | |
6815 | return; | |
6816 | ||
6817 | lb_monitor_task = kthread_create(load_balance_monitor, NULL, | |
6818 | "group_balance"); | |
6819 | if (!IS_ERR(lb_monitor_task)) { | |
6820 | lb_monitor_task->flags |= PF_NOFREEZE; | |
6821 | wake_up_process(lb_monitor_task); | |
6822 | } else { | |
6823 | printk(KERN_ERR "Could not create load balance monitor thread" | |
6824 | "(error = %ld) \n", PTR_ERR(lb_monitor_task)); | |
6825 | } | |
6826 | #endif | |
1da177e4 LT |
6827 | } |
6828 | #else | |
6829 | void __init sched_init_smp(void) | |
6830 | { | |
19978ca6 | 6831 | sched_init_granularity(); |
1da177e4 LT |
6832 | } |
6833 | #endif /* CONFIG_SMP */ | |
6834 | ||
6835 | int in_sched_functions(unsigned long addr) | |
6836 | { | |
1da177e4 LT |
6837 | return in_lock_functions(addr) || |
6838 | (addr >= (unsigned long)__sched_text_start | |
6839 | && addr < (unsigned long)__sched_text_end); | |
6840 | } | |
6841 | ||
a9957449 | 6842 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
6843 | { |
6844 | cfs_rq->tasks_timeline = RB_ROOT; | |
dd41f596 IM |
6845 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6846 | cfs_rq->rq = rq; | |
6847 | #endif | |
67e9fb2a | 6848 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
6849 | } |
6850 | ||
1da177e4 LT |
6851 | void __init sched_init(void) |
6852 | { | |
476f3534 | 6853 | int highest_cpu = 0; |
dd41f596 IM |
6854 | int i, j; |
6855 | ||
57d885fe GH |
6856 | #ifdef CONFIG_SMP |
6857 | init_defrootdomain(); | |
6858 | #endif | |
6859 | ||
0a945022 | 6860 | for_each_possible_cpu(i) { |
dd41f596 | 6861 | struct rt_prio_array *array; |
70b97a7f | 6862 | struct rq *rq; |
1da177e4 LT |
6863 | |
6864 | rq = cpu_rq(i); | |
6865 | spin_lock_init(&rq->lock); | |
fcb99371 | 6866 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 6867 | rq->nr_running = 0; |
dd41f596 IM |
6868 | rq->clock = 1; |
6869 | init_cfs_rq(&rq->cfs, rq); | |
6870 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6871 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); | |
3a252015 IM |
6872 | { |
6873 | struct cfs_rq *cfs_rq = &per_cpu(init_cfs_rq, i); | |
6874 | struct sched_entity *se = | |
6875 | &per_cpu(init_sched_entity, i); | |
6876 | ||
6877 | init_cfs_rq_p[i] = cfs_rq; | |
6878 | init_cfs_rq(cfs_rq, rq); | |
4cf86d77 | 6879 | cfs_rq->tg = &init_task_group; |
3a252015 | 6880 | list_add(&cfs_rq->leaf_cfs_rq_list, |
29f59db3 SV |
6881 | &rq->leaf_cfs_rq_list); |
6882 | ||
3a252015 IM |
6883 | init_sched_entity_p[i] = se; |
6884 | se->cfs_rq = &rq->cfs; | |
6885 | se->my_q = cfs_rq; | |
4cf86d77 | 6886 | se->load.weight = init_task_group_load; |
9b5b7751 | 6887 | se->load.inv_weight = |
4cf86d77 | 6888 | div64_64(1ULL<<32, init_task_group_load); |
3a252015 IM |
6889 | se->parent = NULL; |
6890 | } | |
4cf86d77 | 6891 | init_task_group.shares = init_task_group_load; |
dd41f596 | 6892 | #endif |
1da177e4 | 6893 | |
dd41f596 IM |
6894 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6895 | rq->cpu_load[j] = 0; | |
1da177e4 | 6896 | #ifdef CONFIG_SMP |
41c7ce9a | 6897 | rq->sd = NULL; |
57d885fe GH |
6898 | rq->rd = NULL; |
6899 | rq_attach_root(rq, &def_root_domain); | |
1da177e4 | 6900 | rq->active_balance = 0; |
dd41f596 | 6901 | rq->next_balance = jiffies; |
1da177e4 | 6902 | rq->push_cpu = 0; |
0a2966b4 | 6903 | rq->cpu = i; |
1da177e4 LT |
6904 | rq->migration_thread = NULL; |
6905 | INIT_LIST_HEAD(&rq->migration_queue); | |
764a9d6f | 6906 | rq->rt.highest_prio = MAX_RT_PRIO; |
a22d7fc1 | 6907 | rq->rt.overloaded = 0; |
1da177e4 LT |
6908 | #endif |
6909 | atomic_set(&rq->nr_iowait, 0); | |
6910 | ||
dd41f596 IM |
6911 | array = &rq->rt.active; |
6912 | for (j = 0; j < MAX_RT_PRIO; j++) { | |
6913 | INIT_LIST_HEAD(array->queue + j); | |
6914 | __clear_bit(j, array->bitmap); | |
1da177e4 | 6915 | } |
476f3534 | 6916 | highest_cpu = i; |
dd41f596 IM |
6917 | /* delimiter for bitsearch: */ |
6918 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
1da177e4 LT |
6919 | } |
6920 | ||
2dd73a4f | 6921 | set_load_weight(&init_task); |
b50f60ce | 6922 | |
e107be36 AK |
6923 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6924 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6925 | #endif | |
6926 | ||
c9819f45 | 6927 | #ifdef CONFIG_SMP |
476f3534 | 6928 | nr_cpu_ids = highest_cpu + 1; |
c9819f45 CL |
6929 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); |
6930 | #endif | |
6931 | ||
b50f60ce HC |
6932 | #ifdef CONFIG_RT_MUTEXES |
6933 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
6934 | #endif | |
6935 | ||
1da177e4 LT |
6936 | /* |
6937 | * The boot idle thread does lazy MMU switching as well: | |
6938 | */ | |
6939 | atomic_inc(&init_mm.mm_count); | |
6940 | enter_lazy_tlb(&init_mm, current); | |
6941 | ||
6942 | /* | |
6943 | * Make us the idle thread. Technically, schedule() should not be | |
6944 | * called from this thread, however somewhere below it might be, | |
6945 | * but because we are the idle thread, we just pick up running again | |
6946 | * when this runqueue becomes "idle". | |
6947 | */ | |
6948 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
6949 | /* |
6950 | * During early bootup we pretend to be a normal task: | |
6951 | */ | |
6952 | current->sched_class = &fair_sched_class; | |
1da177e4 LT |
6953 | } |
6954 | ||
6955 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
6956 | void __might_sleep(char *file, int line) | |
6957 | { | |
48f24c4d | 6958 | #ifdef in_atomic |
1da177e4 LT |
6959 | static unsigned long prev_jiffy; /* ratelimiting */ |
6960 | ||
6961 | if ((in_atomic() || irqs_disabled()) && | |
6962 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
6963 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6964 | return; | |
6965 | prev_jiffy = jiffies; | |
91368d73 | 6966 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
6967 | " context at %s:%d\n", file, line); |
6968 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
6969 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 6970 | debug_show_held_locks(current); |
3117df04 IM |
6971 | if (irqs_disabled()) |
6972 | print_irqtrace_events(current); | |
1da177e4 LT |
6973 | dump_stack(); |
6974 | } | |
6975 | #endif | |
6976 | } | |
6977 | EXPORT_SYMBOL(__might_sleep); | |
6978 | #endif | |
6979 | ||
6980 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
6981 | static void normalize_task(struct rq *rq, struct task_struct *p) |
6982 | { | |
6983 | int on_rq; | |
6984 | update_rq_clock(rq); | |
6985 | on_rq = p->se.on_rq; | |
6986 | if (on_rq) | |
6987 | deactivate_task(rq, p, 0); | |
6988 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
6989 | if (on_rq) { | |
6990 | activate_task(rq, p, 0); | |
6991 | resched_task(rq->curr); | |
6992 | } | |
6993 | } | |
6994 | ||
1da177e4 LT |
6995 | void normalize_rt_tasks(void) |
6996 | { | |
a0f98a1c | 6997 | struct task_struct *g, *p; |
1da177e4 | 6998 | unsigned long flags; |
70b97a7f | 6999 | struct rq *rq; |
1da177e4 LT |
7000 | |
7001 | read_lock_irq(&tasklist_lock); | |
a0f98a1c | 7002 | do_each_thread(g, p) { |
178be793 IM |
7003 | /* |
7004 | * Only normalize user tasks: | |
7005 | */ | |
7006 | if (!p->mm) | |
7007 | continue; | |
7008 | ||
6cfb0d5d | 7009 | p->se.exec_start = 0; |
6cfb0d5d | 7010 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 7011 | p->se.wait_start = 0; |
dd41f596 | 7012 | p->se.sleep_start = 0; |
dd41f596 | 7013 | p->se.block_start = 0; |
6cfb0d5d | 7014 | #endif |
dd41f596 IM |
7015 | task_rq(p)->clock = 0; |
7016 | ||
7017 | if (!rt_task(p)) { | |
7018 | /* | |
7019 | * Renice negative nice level userspace | |
7020 | * tasks back to 0: | |
7021 | */ | |
7022 | if (TASK_NICE(p) < 0 && p->mm) | |
7023 | set_user_nice(p, 0); | |
1da177e4 | 7024 | continue; |
dd41f596 | 7025 | } |
1da177e4 | 7026 | |
b29739f9 IM |
7027 | spin_lock_irqsave(&p->pi_lock, flags); |
7028 | rq = __task_rq_lock(p); | |
1da177e4 | 7029 | |
178be793 | 7030 | normalize_task(rq, p); |
3a5e4dc1 | 7031 | |
b29739f9 IM |
7032 | __task_rq_unlock(rq); |
7033 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
a0f98a1c IM |
7034 | } while_each_thread(g, p); |
7035 | ||
1da177e4 LT |
7036 | read_unlock_irq(&tasklist_lock); |
7037 | } | |
7038 | ||
7039 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
7040 | |
7041 | #ifdef CONFIG_IA64 | |
7042 | /* | |
7043 | * These functions are only useful for the IA64 MCA handling. | |
7044 | * | |
7045 | * They can only be called when the whole system has been | |
7046 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7047 | * activity can take place. Using them for anything else would | |
7048 | * be a serious bug, and as a result, they aren't even visible | |
7049 | * under any other configuration. | |
7050 | */ | |
7051 | ||
7052 | /** | |
7053 | * curr_task - return the current task for a given cpu. | |
7054 | * @cpu: the processor in question. | |
7055 | * | |
7056 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7057 | */ | |
36c8b586 | 7058 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7059 | { |
7060 | return cpu_curr(cpu); | |
7061 | } | |
7062 | ||
7063 | /** | |
7064 | * set_curr_task - set the current task for a given cpu. | |
7065 | * @cpu: the processor in question. | |
7066 | * @p: the task pointer to set. | |
7067 | * | |
7068 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7069 | * are serviced on a separate stack. It allows the architecture to switch the |
7070 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7071 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7072 | * and caller must save the original value of the current task (see | |
7073 | * curr_task() above) and restore that value before reenabling interrupts and | |
7074 | * re-starting the system. | |
7075 | * | |
7076 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7077 | */ | |
36c8b586 | 7078 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7079 | { |
7080 | cpu_curr(cpu) = p; | |
7081 | } | |
7082 | ||
7083 | #endif | |
29f59db3 SV |
7084 | |
7085 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7086 | ||
6b2d7700 SV |
7087 | #ifdef CONFIG_SMP |
7088 | /* | |
7089 | * distribute shares of all task groups among their schedulable entities, | |
7090 | * to reflect load distrbution across cpus. | |
7091 | */ | |
7092 | static int rebalance_shares(struct sched_domain *sd, int this_cpu) | |
7093 | { | |
7094 | struct cfs_rq *cfs_rq; | |
7095 | struct rq *rq = cpu_rq(this_cpu); | |
7096 | cpumask_t sdspan = sd->span; | |
7097 | int balanced = 1; | |
7098 | ||
7099 | /* Walk thr' all the task groups that we have */ | |
7100 | for_each_leaf_cfs_rq(rq, cfs_rq) { | |
7101 | int i; | |
7102 | unsigned long total_load = 0, total_shares; | |
7103 | struct task_group *tg = cfs_rq->tg; | |
7104 | ||
7105 | /* Gather total task load of this group across cpus */ | |
7106 | for_each_cpu_mask(i, sdspan) | |
7107 | total_load += tg->cfs_rq[i]->load.weight; | |
7108 | ||
0eab9146 | 7109 | /* Nothing to do if this group has no load */ |
6b2d7700 SV |
7110 | if (!total_load) |
7111 | continue; | |
7112 | ||
7113 | /* | |
7114 | * tg->shares represents the number of cpu shares the task group | |
7115 | * is eligible to hold on a single cpu. On N cpus, it is | |
7116 | * eligible to hold (N * tg->shares) number of cpu shares. | |
7117 | */ | |
7118 | total_shares = tg->shares * cpus_weight(sdspan); | |
7119 | ||
7120 | /* | |
7121 | * redistribute total_shares across cpus as per the task load | |
7122 | * distribution. | |
7123 | */ | |
7124 | for_each_cpu_mask(i, sdspan) { | |
7125 | unsigned long local_load, local_shares; | |
7126 | ||
7127 | local_load = tg->cfs_rq[i]->load.weight; | |
7128 | local_shares = (local_load * total_shares) / total_load; | |
7129 | if (!local_shares) | |
7130 | local_shares = MIN_GROUP_SHARES; | |
7131 | if (local_shares == tg->se[i]->load.weight) | |
7132 | continue; | |
7133 | ||
7134 | spin_lock_irq(&cpu_rq(i)->lock); | |
7135 | set_se_shares(tg->se[i], local_shares); | |
7136 | spin_unlock_irq(&cpu_rq(i)->lock); | |
7137 | balanced = 0; | |
7138 | } | |
7139 | } | |
7140 | ||
7141 | return balanced; | |
7142 | } | |
7143 | ||
7144 | /* | |
7145 | * How frequently should we rebalance_shares() across cpus? | |
7146 | * | |
7147 | * The more frequently we rebalance shares, the more accurate is the fairness | |
7148 | * of cpu bandwidth distribution between task groups. However higher frequency | |
7149 | * also implies increased scheduling overhead. | |
7150 | * | |
7151 | * sysctl_sched_min_bal_int_shares represents the minimum interval between | |
7152 | * consecutive calls to rebalance_shares() in the same sched domain. | |
7153 | * | |
7154 | * sysctl_sched_max_bal_int_shares represents the maximum interval between | |
7155 | * consecutive calls to rebalance_shares() in the same sched domain. | |
7156 | * | |
7157 | * These settings allows for the appropriate tradeoff between accuracy of | |
7158 | * fairness and the associated overhead. | |
7159 | * | |
7160 | */ | |
7161 | ||
7162 | /* default: 8ms, units: milliseconds */ | |
7163 | const_debug unsigned int sysctl_sched_min_bal_int_shares = 8; | |
7164 | ||
7165 | /* default: 128ms, units: milliseconds */ | |
7166 | const_debug unsigned int sysctl_sched_max_bal_int_shares = 128; | |
7167 | ||
7168 | /* kernel thread that runs rebalance_shares() periodically */ | |
7169 | static int load_balance_monitor(void *unused) | |
7170 | { | |
7171 | unsigned int timeout = sysctl_sched_min_bal_int_shares; | |
7172 | struct sched_param schedparm; | |
7173 | int ret; | |
7174 | ||
7175 | /* | |
7176 | * We don't want this thread's execution to be limited by the shares | |
7177 | * assigned to default group (init_task_group). Hence make it run | |
7178 | * as a SCHED_RR RT task at the lowest priority. | |
7179 | */ | |
7180 | schedparm.sched_priority = 1; | |
7181 | ret = sched_setscheduler(current, SCHED_RR, &schedparm); | |
7182 | if (ret) | |
7183 | printk(KERN_ERR "Couldn't set SCHED_RR policy for load balance" | |
7184 | " monitor thread (error = %d) \n", ret); | |
7185 | ||
7186 | while (!kthread_should_stop()) { | |
7187 | int i, cpu, balanced = 1; | |
7188 | ||
7189 | /* Prevent cpus going down or coming up */ | |
86ef5c9a | 7190 | get_online_cpus(); |
6b2d7700 SV |
7191 | /* lockout changes to doms_cur[] array */ |
7192 | lock_doms_cur(); | |
7193 | /* | |
7194 | * Enter a rcu read-side critical section to safely walk rq->sd | |
7195 | * chain on various cpus and to walk task group list | |
7196 | * (rq->leaf_cfs_rq_list) in rebalance_shares(). | |
7197 | */ | |
7198 | rcu_read_lock(); | |
7199 | ||
7200 | for (i = 0; i < ndoms_cur; i++) { | |
7201 | cpumask_t cpumap = doms_cur[i]; | |
7202 | struct sched_domain *sd = NULL, *sd_prev = NULL; | |
7203 | ||
7204 | cpu = first_cpu(cpumap); | |
7205 | ||
7206 | /* Find the highest domain at which to balance shares */ | |
7207 | for_each_domain(cpu, sd) { | |
7208 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
7209 | continue; | |
7210 | sd_prev = sd; | |
7211 | } | |
7212 | ||
7213 | sd = sd_prev; | |
7214 | /* sd == NULL? No load balance reqd in this domain */ | |
7215 | if (!sd) | |
7216 | continue; | |
7217 | ||
7218 | balanced &= rebalance_shares(sd, cpu); | |
7219 | } | |
7220 | ||
7221 | rcu_read_unlock(); | |
7222 | ||
7223 | unlock_doms_cur(); | |
86ef5c9a | 7224 | put_online_cpus(); |
6b2d7700 SV |
7225 | |
7226 | if (!balanced) | |
7227 | timeout = sysctl_sched_min_bal_int_shares; | |
7228 | else if (timeout < sysctl_sched_max_bal_int_shares) | |
7229 | timeout *= 2; | |
7230 | ||
7231 | msleep_interruptible(timeout); | |
7232 | } | |
7233 | ||
7234 | return 0; | |
7235 | } | |
7236 | #endif /* CONFIG_SMP */ | |
7237 | ||
29f59db3 | 7238 | /* allocate runqueue etc for a new task group */ |
4cf86d77 | 7239 | struct task_group *sched_create_group(void) |
29f59db3 | 7240 | { |
4cf86d77 | 7241 | struct task_group *tg; |
29f59db3 SV |
7242 | struct cfs_rq *cfs_rq; |
7243 | struct sched_entity *se; | |
9b5b7751 | 7244 | struct rq *rq; |
29f59db3 SV |
7245 | int i; |
7246 | ||
29f59db3 SV |
7247 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); |
7248 | if (!tg) | |
7249 | return ERR_PTR(-ENOMEM); | |
7250 | ||
9b5b7751 | 7251 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
7252 | if (!tg->cfs_rq) |
7253 | goto err; | |
9b5b7751 | 7254 | tg->se = kzalloc(sizeof(se) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
7255 | if (!tg->se) |
7256 | goto err; | |
7257 | ||
7258 | for_each_possible_cpu(i) { | |
9b5b7751 | 7259 | rq = cpu_rq(i); |
29f59db3 SV |
7260 | |
7261 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), GFP_KERNEL, | |
7262 | cpu_to_node(i)); | |
7263 | if (!cfs_rq) | |
7264 | goto err; | |
7265 | ||
7266 | se = kmalloc_node(sizeof(struct sched_entity), GFP_KERNEL, | |
7267 | cpu_to_node(i)); | |
7268 | if (!se) | |
7269 | goto err; | |
7270 | ||
7271 | memset(cfs_rq, 0, sizeof(struct cfs_rq)); | |
7272 | memset(se, 0, sizeof(struct sched_entity)); | |
7273 | ||
7274 | tg->cfs_rq[i] = cfs_rq; | |
7275 | init_cfs_rq(cfs_rq, rq); | |
7276 | cfs_rq->tg = tg; | |
29f59db3 SV |
7277 | |
7278 | tg->se[i] = se; | |
7279 | se->cfs_rq = &rq->cfs; | |
7280 | se->my_q = cfs_rq; | |
7281 | se->load.weight = NICE_0_LOAD; | |
7282 | se->load.inv_weight = div64_64(1ULL<<32, NICE_0_LOAD); | |
7283 | se->parent = NULL; | |
7284 | } | |
7285 | ||
ec2c507f SV |
7286 | tg->shares = NICE_0_LOAD; |
7287 | ||
7288 | lock_task_group_list(); | |
9b5b7751 SV |
7289 | for_each_possible_cpu(i) { |
7290 | rq = cpu_rq(i); | |
7291 | cfs_rq = tg->cfs_rq[i]; | |
7292 | list_add_rcu(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7293 | } | |
ec2c507f | 7294 | unlock_task_group_list(); |
29f59db3 | 7295 | |
9b5b7751 | 7296 | return tg; |
29f59db3 SV |
7297 | |
7298 | err: | |
7299 | for_each_possible_cpu(i) { | |
a65914b3 | 7300 | if (tg->cfs_rq) |
29f59db3 | 7301 | kfree(tg->cfs_rq[i]); |
a65914b3 | 7302 | if (tg->se) |
29f59db3 SV |
7303 | kfree(tg->se[i]); |
7304 | } | |
a65914b3 IM |
7305 | kfree(tg->cfs_rq); |
7306 | kfree(tg->se); | |
7307 | kfree(tg); | |
29f59db3 SV |
7308 | |
7309 | return ERR_PTR(-ENOMEM); | |
7310 | } | |
7311 | ||
9b5b7751 SV |
7312 | /* rcu callback to free various structures associated with a task group */ |
7313 | static void free_sched_group(struct rcu_head *rhp) | |
29f59db3 | 7314 | { |
ae8393e5 SV |
7315 | struct task_group *tg = container_of(rhp, struct task_group, rcu); |
7316 | struct cfs_rq *cfs_rq; | |
29f59db3 SV |
7317 | struct sched_entity *se; |
7318 | int i; | |
7319 | ||
29f59db3 SV |
7320 | /* now it should be safe to free those cfs_rqs */ |
7321 | for_each_possible_cpu(i) { | |
7322 | cfs_rq = tg->cfs_rq[i]; | |
7323 | kfree(cfs_rq); | |
7324 | ||
7325 | se = tg->se[i]; | |
7326 | kfree(se); | |
7327 | } | |
7328 | ||
7329 | kfree(tg->cfs_rq); | |
7330 | kfree(tg->se); | |
7331 | kfree(tg); | |
7332 | } | |
7333 | ||
9b5b7751 | 7334 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7335 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 7336 | { |
7bae49d4 | 7337 | struct cfs_rq *cfs_rq = NULL; |
9b5b7751 | 7338 | int i; |
29f59db3 | 7339 | |
ec2c507f | 7340 | lock_task_group_list(); |
9b5b7751 SV |
7341 | for_each_possible_cpu(i) { |
7342 | cfs_rq = tg->cfs_rq[i]; | |
7343 | list_del_rcu(&cfs_rq->leaf_cfs_rq_list); | |
7344 | } | |
ec2c507f | 7345 | unlock_task_group_list(); |
9b5b7751 | 7346 | |
7bae49d4 | 7347 | BUG_ON(!cfs_rq); |
9b5b7751 SV |
7348 | |
7349 | /* wait for possible concurrent references to cfs_rqs complete */ | |
ae8393e5 | 7350 | call_rcu(&tg->rcu, free_sched_group); |
29f59db3 SV |
7351 | } |
7352 | ||
9b5b7751 | 7353 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7354 | * The caller of this function should have put the task in its new group |
7355 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7356 | * reflect its new group. | |
9b5b7751 SV |
7357 | */ |
7358 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
7359 | { |
7360 | int on_rq, running; | |
7361 | unsigned long flags; | |
7362 | struct rq *rq; | |
7363 | ||
7364 | rq = task_rq_lock(tsk, &flags); | |
7365 | ||
dae51f56 | 7366 | if (tsk->sched_class != &fair_sched_class) { |
ce96b5ac | 7367 | set_task_cfs_rq(tsk, task_cpu(tsk)); |
29f59db3 | 7368 | goto done; |
dae51f56 | 7369 | } |
29f59db3 SV |
7370 | |
7371 | update_rq_clock(rq); | |
7372 | ||
051a1d1a | 7373 | running = task_current(rq, tsk); |
29f59db3 SV |
7374 | on_rq = tsk->se.on_rq; |
7375 | ||
83b699ed | 7376 | if (on_rq) { |
29f59db3 | 7377 | dequeue_task(rq, tsk, 0); |
83b699ed SV |
7378 | if (unlikely(running)) |
7379 | tsk->sched_class->put_prev_task(rq, tsk); | |
7380 | } | |
29f59db3 | 7381 | |
ce96b5ac | 7382 | set_task_cfs_rq(tsk, task_cpu(tsk)); |
29f59db3 | 7383 | |
83b699ed SV |
7384 | if (on_rq) { |
7385 | if (unlikely(running)) | |
7386 | tsk->sched_class->set_curr_task(rq); | |
7074badb | 7387 | enqueue_task(rq, tsk, 0); |
83b699ed | 7388 | } |
29f59db3 SV |
7389 | |
7390 | done: | |
7391 | task_rq_unlock(rq, &flags); | |
7392 | } | |
7393 | ||
6b2d7700 | 7394 | /* rq->lock to be locked by caller */ |
29f59db3 SV |
7395 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
7396 | { | |
7397 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
7398 | struct rq *rq = cfs_rq->rq; | |
7399 | int on_rq; | |
7400 | ||
6b2d7700 SV |
7401 | if (!shares) |
7402 | shares = MIN_GROUP_SHARES; | |
29f59db3 SV |
7403 | |
7404 | on_rq = se->on_rq; | |
6b2d7700 | 7405 | if (on_rq) { |
29f59db3 | 7406 | dequeue_entity(cfs_rq, se, 0); |
6b2d7700 SV |
7407 | dec_cpu_load(rq, se->load.weight); |
7408 | } | |
29f59db3 SV |
7409 | |
7410 | se->load.weight = shares; | |
7411 | se->load.inv_weight = div64_64((1ULL<<32), shares); | |
7412 | ||
6b2d7700 | 7413 | if (on_rq) { |
29f59db3 | 7414 | enqueue_entity(cfs_rq, se, 0); |
6b2d7700 SV |
7415 | inc_cpu_load(rq, se->load.weight); |
7416 | } | |
29f59db3 SV |
7417 | } |
7418 | ||
4cf86d77 | 7419 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
7420 | { |
7421 | int i; | |
6b2d7700 SV |
7422 | struct cfs_rq *cfs_rq; |
7423 | struct rq *rq; | |
c61935fd | 7424 | |
ec2c507f | 7425 | lock_task_group_list(); |
9b5b7751 | 7426 | if (tg->shares == shares) |
5cb350ba | 7427 | goto done; |
29f59db3 | 7428 | |
6b2d7700 SV |
7429 | if (shares < MIN_GROUP_SHARES) |
7430 | shares = MIN_GROUP_SHARES; | |
7431 | ||
7432 | /* | |
7433 | * Prevent any load balance activity (rebalance_shares, | |
7434 | * load_balance_fair) from referring to this group first, | |
7435 | * by taking it off the rq->leaf_cfs_rq_list on each cpu. | |
7436 | */ | |
7437 | for_each_possible_cpu(i) { | |
7438 | cfs_rq = tg->cfs_rq[i]; | |
7439 | list_del_rcu(&cfs_rq->leaf_cfs_rq_list); | |
7440 | } | |
7441 | ||
7442 | /* wait for any ongoing reference to this group to finish */ | |
7443 | synchronize_sched(); | |
7444 | ||
7445 | /* | |
7446 | * Now we are free to modify the group's share on each cpu | |
7447 | * w/o tripping rebalance_share or load_balance_fair. | |
7448 | */ | |
9b5b7751 | 7449 | tg->shares = shares; |
6b2d7700 SV |
7450 | for_each_possible_cpu(i) { |
7451 | spin_lock_irq(&cpu_rq(i)->lock); | |
9b5b7751 | 7452 | set_se_shares(tg->se[i], shares); |
6b2d7700 SV |
7453 | spin_unlock_irq(&cpu_rq(i)->lock); |
7454 | } | |
29f59db3 | 7455 | |
6b2d7700 SV |
7456 | /* |
7457 | * Enable load balance activity on this group, by inserting it back on | |
7458 | * each cpu's rq->leaf_cfs_rq_list. | |
7459 | */ | |
7460 | for_each_possible_cpu(i) { | |
7461 | rq = cpu_rq(i); | |
7462 | cfs_rq = tg->cfs_rq[i]; | |
7463 | list_add_rcu(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7464 | } | |
5cb350ba | 7465 | done: |
ec2c507f | 7466 | unlock_task_group_list(); |
9b5b7751 | 7467 | return 0; |
29f59db3 SV |
7468 | } |
7469 | ||
5cb350ba DG |
7470 | unsigned long sched_group_shares(struct task_group *tg) |
7471 | { | |
7472 | return tg->shares; | |
7473 | } | |
7474 | ||
3a252015 | 7475 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e SV |
7476 | |
7477 | #ifdef CONFIG_FAIR_CGROUP_SCHED | |
7478 | ||
7479 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 7480 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 7481 | { |
2b01dfe3 PM |
7482 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
7483 | struct task_group, css); | |
68318b8e SV |
7484 | } |
7485 | ||
7486 | static struct cgroup_subsys_state * | |
2b01dfe3 | 7487 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e SV |
7488 | { |
7489 | struct task_group *tg; | |
7490 | ||
2b01dfe3 | 7491 | if (!cgrp->parent) { |
68318b8e | 7492 | /* This is early initialization for the top cgroup */ |
2b01dfe3 | 7493 | init_task_group.css.cgroup = cgrp; |
68318b8e SV |
7494 | return &init_task_group.css; |
7495 | } | |
7496 | ||
7497 | /* we support only 1-level deep hierarchical scheduler atm */ | |
2b01dfe3 | 7498 | if (cgrp->parent->parent) |
68318b8e SV |
7499 | return ERR_PTR(-EINVAL); |
7500 | ||
7501 | tg = sched_create_group(); | |
7502 | if (IS_ERR(tg)) | |
7503 | return ERR_PTR(-ENOMEM); | |
7504 | ||
7505 | /* Bind the cgroup to task_group object we just created */ | |
2b01dfe3 | 7506 | tg->css.cgroup = cgrp; |
68318b8e SV |
7507 | |
7508 | return &tg->css; | |
7509 | } | |
7510 | ||
41a2d6cf IM |
7511 | static void |
7512 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 7513 | { |
2b01dfe3 | 7514 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7515 | |
7516 | sched_destroy_group(tg); | |
7517 | } | |
7518 | ||
41a2d6cf IM |
7519 | static int |
7520 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
7521 | struct task_struct *tsk) | |
68318b8e SV |
7522 | { |
7523 | /* We don't support RT-tasks being in separate groups */ | |
7524 | if (tsk->sched_class != &fair_sched_class) | |
7525 | return -EINVAL; | |
7526 | ||
7527 | return 0; | |
7528 | } | |
7529 | ||
7530 | static void | |
2b01dfe3 | 7531 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
7532 | struct cgroup *old_cont, struct task_struct *tsk) |
7533 | { | |
7534 | sched_move_task(tsk); | |
7535 | } | |
7536 | ||
2b01dfe3 PM |
7537 | static int cpu_shares_write_uint(struct cgroup *cgrp, struct cftype *cftype, |
7538 | u64 shareval) | |
68318b8e | 7539 | { |
2b01dfe3 | 7540 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
7541 | } |
7542 | ||
2b01dfe3 | 7543 | static u64 cpu_shares_read_uint(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 7544 | { |
2b01dfe3 | 7545 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7546 | |
7547 | return (u64) tg->shares; | |
7548 | } | |
7549 | ||
fe5c7cc2 PM |
7550 | static struct cftype cpu_files[] = { |
7551 | { | |
7552 | .name = "shares", | |
7553 | .read_uint = cpu_shares_read_uint, | |
7554 | .write_uint = cpu_shares_write_uint, | |
7555 | }, | |
68318b8e SV |
7556 | }; |
7557 | ||
7558 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
7559 | { | |
fe5c7cc2 | 7560 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
7561 | } |
7562 | ||
7563 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
7564 | .name = "cpu", |
7565 | .create = cpu_cgroup_create, | |
7566 | .destroy = cpu_cgroup_destroy, | |
7567 | .can_attach = cpu_cgroup_can_attach, | |
7568 | .attach = cpu_cgroup_attach, | |
7569 | .populate = cpu_cgroup_populate, | |
7570 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
7571 | .early_init = 1, |
7572 | }; | |
7573 | ||
7574 | #endif /* CONFIG_FAIR_CGROUP_SCHED */ | |
d842de87 SV |
7575 | |
7576 | #ifdef CONFIG_CGROUP_CPUACCT | |
7577 | ||
7578 | /* | |
7579 | * CPU accounting code for task groups. | |
7580 | * | |
7581 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
7582 | * (balbir@in.ibm.com). | |
7583 | */ | |
7584 | ||
7585 | /* track cpu usage of a group of tasks */ | |
7586 | struct cpuacct { | |
7587 | struct cgroup_subsys_state css; | |
7588 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
7589 | u64 *cpuusage; | |
7590 | }; | |
7591 | ||
7592 | struct cgroup_subsys cpuacct_subsys; | |
7593 | ||
7594 | /* return cpu accounting group corresponding to this container */ | |
7595 | static inline struct cpuacct *cgroup_ca(struct cgroup *cont) | |
7596 | { | |
7597 | return container_of(cgroup_subsys_state(cont, cpuacct_subsys_id), | |
7598 | struct cpuacct, css); | |
7599 | } | |
7600 | ||
7601 | /* return cpu accounting group to which this task belongs */ | |
7602 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
7603 | { | |
7604 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
7605 | struct cpuacct, css); | |
7606 | } | |
7607 | ||
7608 | /* create a new cpu accounting group */ | |
7609 | static struct cgroup_subsys_state *cpuacct_create( | |
7610 | struct cgroup_subsys *ss, struct cgroup *cont) | |
7611 | { | |
7612 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
7613 | ||
7614 | if (!ca) | |
7615 | return ERR_PTR(-ENOMEM); | |
7616 | ||
7617 | ca->cpuusage = alloc_percpu(u64); | |
7618 | if (!ca->cpuusage) { | |
7619 | kfree(ca); | |
7620 | return ERR_PTR(-ENOMEM); | |
7621 | } | |
7622 | ||
7623 | return &ca->css; | |
7624 | } | |
7625 | ||
7626 | /* destroy an existing cpu accounting group */ | |
41a2d6cf IM |
7627 | static void |
7628 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cont) | |
d842de87 SV |
7629 | { |
7630 | struct cpuacct *ca = cgroup_ca(cont); | |
7631 | ||
7632 | free_percpu(ca->cpuusage); | |
7633 | kfree(ca); | |
7634 | } | |
7635 | ||
7636 | /* return total cpu usage (in nanoseconds) of a group */ | |
7637 | static u64 cpuusage_read(struct cgroup *cont, struct cftype *cft) | |
7638 | { | |
7639 | struct cpuacct *ca = cgroup_ca(cont); | |
7640 | u64 totalcpuusage = 0; | |
7641 | int i; | |
7642 | ||
7643 | for_each_possible_cpu(i) { | |
7644 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
7645 | ||
7646 | /* | |
7647 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
7648 | * platforms. | |
7649 | */ | |
7650 | spin_lock_irq(&cpu_rq(i)->lock); | |
7651 | totalcpuusage += *cpuusage; | |
7652 | spin_unlock_irq(&cpu_rq(i)->lock); | |
7653 | } | |
7654 | ||
7655 | return totalcpuusage; | |
7656 | } | |
7657 | ||
7658 | static struct cftype files[] = { | |
7659 | { | |
7660 | .name = "usage", | |
7661 | .read_uint = cpuusage_read, | |
7662 | }, | |
7663 | }; | |
7664 | ||
7665 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
7666 | { | |
7667 | return cgroup_add_files(cont, ss, files, ARRAY_SIZE(files)); | |
7668 | } | |
7669 | ||
7670 | /* | |
7671 | * charge this task's execution time to its accounting group. | |
7672 | * | |
7673 | * called with rq->lock held. | |
7674 | */ | |
7675 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
7676 | { | |
7677 | struct cpuacct *ca; | |
7678 | ||
7679 | if (!cpuacct_subsys.active) | |
7680 | return; | |
7681 | ||
7682 | ca = task_ca(tsk); | |
7683 | if (ca) { | |
7684 | u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk)); | |
7685 | ||
7686 | *cpuusage += cputime; | |
7687 | } | |
7688 | } | |
7689 | ||
7690 | struct cgroup_subsys cpuacct_subsys = { | |
7691 | .name = "cpuacct", | |
7692 | .create = cpuacct_create, | |
7693 | .destroy = cpuacct_destroy, | |
7694 | .populate = cpuacct_populate, | |
7695 | .subsys_id = cpuacct_subsys_id, | |
7696 | }; | |
7697 | #endif /* CONFIG_CGROUP_CPUACCT */ |