]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/sched.c | |
3 | * | |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 LT |
34 | #include <linux/highmem.h> |
35 | #include <linux/smp_lock.h> | |
36 | #include <asm/mmu_context.h> | |
37 | #include <linux/interrupt.h> | |
c59ede7b | 38 | #include <linux/capability.h> |
1da177e4 LT |
39 | #include <linux/completion.h> |
40 | #include <linux/kernel_stat.h> | |
9a11b49a | 41 | #include <linux/debug_locks.h> |
1da177e4 LT |
42 | #include <linux/security.h> |
43 | #include <linux/notifier.h> | |
44 | #include <linux/profile.h> | |
7dfb7103 | 45 | #include <linux/freezer.h> |
198e2f18 | 46 | #include <linux/vmalloc.h> |
1da177e4 LT |
47 | #include <linux/blkdev.h> |
48 | #include <linux/delay.h> | |
b488893a | 49 | #include <linux/pid_namespace.h> |
1da177e4 LT |
50 | #include <linux/smp.h> |
51 | #include <linux/threads.h> | |
52 | #include <linux/timer.h> | |
53 | #include <linux/rcupdate.h> | |
54 | #include <linux/cpu.h> | |
55 | #include <linux/cpuset.h> | |
56 | #include <linux/percpu.h> | |
57 | #include <linux/kthread.h> | |
58 | #include <linux/seq_file.h> | |
e692ab53 | 59 | #include <linux/sysctl.h> |
1da177e4 LT |
60 | #include <linux/syscalls.h> |
61 | #include <linux/times.h> | |
8f0ab514 | 62 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 63 | #include <linux/kprobes.h> |
0ff92245 | 64 | #include <linux/delayacct.h> |
5517d86b | 65 | #include <linux/reciprocal_div.h> |
dff06c15 | 66 | #include <linux/unistd.h> |
f5ff8422 | 67 | #include <linux/pagemap.h> |
8f4d37ec | 68 | #include <linux/hrtimer.h> |
1da177e4 | 69 | |
5517d86b | 70 | #include <asm/tlb.h> |
838225b4 | 71 | #include <asm/irq_regs.h> |
1da177e4 | 72 | |
b035b6de AD |
73 | /* |
74 | * Scheduler clock - returns current time in nanosec units. | |
75 | * This is default implementation. | |
76 | * Architectures and sub-architectures can override this. | |
77 | */ | |
78 | unsigned long long __attribute__((weak)) sched_clock(void) | |
79 | { | |
d6322faf | 80 | return (unsigned long long)jiffies * (NSEC_PER_SEC / HZ); |
b035b6de AD |
81 | } |
82 | ||
1da177e4 LT |
83 | /* |
84 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
85 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
86 | * and back. | |
87 | */ | |
88 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
89 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
90 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
91 | ||
92 | /* | |
93 | * 'User priority' is the nice value converted to something we | |
94 | * can work with better when scaling various scheduler parameters, | |
95 | * it's a [ 0 ... 39 ] range. | |
96 | */ | |
97 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
98 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
99 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
100 | ||
101 | /* | |
d7876a08 | 102 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 103 | */ |
d6322faf | 104 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 105 | |
6aa645ea IM |
106 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
107 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
108 | ||
1da177e4 LT |
109 | /* |
110 | * These are the 'tuning knobs' of the scheduler: | |
111 | * | |
a4ec24b4 | 112 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
113 | * Timeslices get refilled after they expire. |
114 | */ | |
1da177e4 | 115 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 116 | |
5517d86b ED |
117 | #ifdef CONFIG_SMP |
118 | /* | |
119 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
120 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
121 | */ | |
122 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
123 | { | |
124 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
125 | } | |
126 | ||
127 | /* | |
128 | * Each time a sched group cpu_power is changed, | |
129 | * we must compute its reciprocal value | |
130 | */ | |
131 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
132 | { | |
133 | sg->__cpu_power += val; | |
134 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
135 | } | |
136 | #endif | |
137 | ||
e05606d3 IM |
138 | static inline int rt_policy(int policy) |
139 | { | |
140 | if (unlikely(policy == SCHED_FIFO) || unlikely(policy == SCHED_RR)) | |
141 | return 1; | |
142 | return 0; | |
143 | } | |
144 | ||
145 | static inline int task_has_rt_policy(struct task_struct *p) | |
146 | { | |
147 | return rt_policy(p->policy); | |
148 | } | |
149 | ||
1da177e4 | 150 | /* |
6aa645ea | 151 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 152 | */ |
6aa645ea IM |
153 | struct rt_prio_array { |
154 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
155 | struct list_head queue[MAX_RT_PRIO]; | |
156 | }; | |
157 | ||
29f59db3 SV |
158 | #ifdef CONFIG_FAIR_GROUP_SCHED |
159 | ||
68318b8e SV |
160 | #include <linux/cgroup.h> |
161 | ||
29f59db3 SV |
162 | struct cfs_rq; |
163 | ||
164 | /* task group related information */ | |
4cf86d77 | 165 | struct task_group { |
68318b8e SV |
166 | #ifdef CONFIG_FAIR_CGROUP_SCHED |
167 | struct cgroup_subsys_state css; | |
168 | #endif | |
29f59db3 SV |
169 | /* schedulable entities of this group on each cpu */ |
170 | struct sched_entity **se; | |
171 | /* runqueue "owned" by this group on each cpu */ | |
172 | struct cfs_rq **cfs_rq; | |
6b2d7700 SV |
173 | |
174 | /* | |
175 | * shares assigned to a task group governs how much of cpu bandwidth | |
176 | * is allocated to the group. The more shares a group has, the more is | |
177 | * the cpu bandwidth allocated to it. | |
178 | * | |
179 | * For ex, lets say that there are three task groups, A, B and C which | |
180 | * have been assigned shares 1000, 2000 and 3000 respectively. Then, | |
181 | * cpu bandwidth allocated by the scheduler to task groups A, B and C | |
182 | * should be: | |
183 | * | |
184 | * Bw(A) = 1000/(1000+2000+3000) * 100 = 16.66% | |
185 | * Bw(B) = 2000/(1000+2000+3000) * 100 = 33.33% | |
03319ec8 | 186 | * Bw(C) = 3000/(1000+2000+3000) * 100 = 50% |
6b2d7700 SV |
187 | * |
188 | * The weight assigned to a task group's schedulable entities on every | |
189 | * cpu (task_group.se[a_cpu]->load.weight) is derived from the task | |
190 | * group's shares. For ex: lets say that task group A has been | |
191 | * assigned shares of 1000 and there are two CPUs in a system. Then, | |
192 | * | |
193 | * tg_A->se[0]->load.weight = tg_A->se[1]->load.weight = 1000; | |
194 | * | |
195 | * Note: It's not necessary that each of a task's group schedulable | |
03319ec8 IM |
196 | * entity have the same weight on all CPUs. If the group |
197 | * has 2 of its tasks on CPU0 and 1 task on CPU1, then a | |
198 | * better distribution of weight could be: | |
6b2d7700 SV |
199 | * |
200 | * tg_A->se[0]->load.weight = 2/3 * 2000 = 1333 | |
201 | * tg_A->se[1]->load.weight = 1/2 * 2000 = 667 | |
202 | * | |
203 | * rebalance_shares() is responsible for distributing the shares of a | |
204 | * task groups like this among the group's schedulable entities across | |
205 | * cpus. | |
206 | * | |
207 | */ | |
29f59db3 | 208 | unsigned long shares; |
6b2d7700 | 209 | |
ae8393e5 | 210 | struct rcu_head rcu; |
29f59db3 SV |
211 | }; |
212 | ||
213 | /* Default task group's sched entity on each cpu */ | |
214 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
215 | /* Default task group's cfs_rq on each cpu */ | |
216 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
217 | ||
9b5b7751 SV |
218 | static struct sched_entity *init_sched_entity_p[NR_CPUS]; |
219 | static struct cfs_rq *init_cfs_rq_p[NR_CPUS]; | |
29f59db3 | 220 | |
ec2c507f SV |
221 | /* task_group_mutex serializes add/remove of task groups and also changes to |
222 | * a task group's cpu shares. | |
223 | */ | |
224 | static DEFINE_MUTEX(task_group_mutex); | |
225 | ||
a1835615 SV |
226 | /* doms_cur_mutex serializes access to doms_cur[] array */ |
227 | static DEFINE_MUTEX(doms_cur_mutex); | |
228 | ||
6b2d7700 SV |
229 | #ifdef CONFIG_SMP |
230 | /* kernel thread that runs rebalance_shares() periodically */ | |
231 | static struct task_struct *lb_monitor_task; | |
232 | static int load_balance_monitor(void *unused); | |
233 | #endif | |
234 | ||
235 | static void set_se_shares(struct sched_entity *se, unsigned long shares); | |
236 | ||
29f59db3 | 237 | /* Default task group. |
3a252015 | 238 | * Every task in system belong to this group at bootup. |
29f59db3 | 239 | */ |
4cf86d77 | 240 | struct task_group init_task_group = { |
0eab9146 | 241 | .se = init_sched_entity_p, |
3a252015 IM |
242 | .cfs_rq = init_cfs_rq_p, |
243 | }; | |
9b5b7751 | 244 | |
24e377a8 | 245 | #ifdef CONFIG_FAIR_USER_SCHED |
0eab9146 | 246 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) |
24e377a8 | 247 | #else |
93f992cc | 248 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
24e377a8 SV |
249 | #endif |
250 | ||
0eab9146 | 251 | #define MIN_GROUP_SHARES 2 |
6b2d7700 | 252 | |
93f992cc | 253 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
29f59db3 SV |
254 | |
255 | /* return group to which a task belongs */ | |
4cf86d77 | 256 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 257 | { |
4cf86d77 | 258 | struct task_group *tg; |
9b5b7751 | 259 | |
24e377a8 SV |
260 | #ifdef CONFIG_FAIR_USER_SCHED |
261 | tg = p->user->tg; | |
68318b8e SV |
262 | #elif defined(CONFIG_FAIR_CGROUP_SCHED) |
263 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), | |
264 | struct task_group, css); | |
24e377a8 | 265 | #else |
41a2d6cf | 266 | tg = &init_task_group; |
24e377a8 | 267 | #endif |
9b5b7751 | 268 | return tg; |
29f59db3 SV |
269 | } |
270 | ||
271 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
ce96b5ac | 272 | static inline void set_task_cfs_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 273 | { |
ce96b5ac DA |
274 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
275 | p->se.parent = task_group(p)->se[cpu]; | |
29f59db3 SV |
276 | } |
277 | ||
ec2c507f SV |
278 | static inline void lock_task_group_list(void) |
279 | { | |
280 | mutex_lock(&task_group_mutex); | |
281 | } | |
282 | ||
283 | static inline void unlock_task_group_list(void) | |
284 | { | |
285 | mutex_unlock(&task_group_mutex); | |
286 | } | |
287 | ||
a1835615 SV |
288 | static inline void lock_doms_cur(void) |
289 | { | |
290 | mutex_lock(&doms_cur_mutex); | |
291 | } | |
292 | ||
293 | static inline void unlock_doms_cur(void) | |
294 | { | |
295 | mutex_unlock(&doms_cur_mutex); | |
296 | } | |
297 | ||
29f59db3 SV |
298 | #else |
299 | ||
ce96b5ac | 300 | static inline void set_task_cfs_rq(struct task_struct *p, unsigned int cpu) { } |
ec2c507f SV |
301 | static inline void lock_task_group_list(void) { } |
302 | static inline void unlock_task_group_list(void) { } | |
a1835615 SV |
303 | static inline void lock_doms_cur(void) { } |
304 | static inline void unlock_doms_cur(void) { } | |
29f59db3 SV |
305 | |
306 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
307 | ||
6aa645ea IM |
308 | /* CFS-related fields in a runqueue */ |
309 | struct cfs_rq { | |
310 | struct load_weight load; | |
311 | unsigned long nr_running; | |
312 | ||
6aa645ea | 313 | u64 exec_clock; |
e9acbff6 | 314 | u64 min_vruntime; |
6aa645ea IM |
315 | |
316 | struct rb_root tasks_timeline; | |
317 | struct rb_node *rb_leftmost; | |
318 | struct rb_node *rb_load_balance_curr; | |
6aa645ea IM |
319 | /* 'curr' points to currently running entity on this cfs_rq. |
320 | * It is set to NULL otherwise (i.e when none are currently running). | |
321 | */ | |
322 | struct sched_entity *curr; | |
ddc97297 PZ |
323 | |
324 | unsigned long nr_spread_over; | |
325 | ||
62160e3f | 326 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
327 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
328 | ||
41a2d6cf IM |
329 | /* |
330 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
331 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
332 | * (like users, containers etc.) | |
333 | * | |
334 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
335 | * list is used during load balance. | |
336 | */ | |
41a2d6cf IM |
337 | struct list_head leaf_cfs_rq_list; |
338 | struct task_group *tg; /* group that "owns" this runqueue */ | |
6aa645ea IM |
339 | #endif |
340 | }; | |
1da177e4 | 341 | |
6aa645ea IM |
342 | /* Real-Time classes' related field in a runqueue: */ |
343 | struct rt_rq { | |
344 | struct rt_prio_array active; | |
63489e45 | 345 | unsigned long rt_nr_running; |
fa85ae24 | 346 | #ifdef CONFIG_SMP |
73fe6aae | 347 | unsigned long rt_nr_migratory; |
fa85ae24 | 348 | int highest_prio; /* highest queued rt task prio */ |
a22d7fc1 | 349 | int overloaded; |
fa85ae24 PZ |
350 | #endif |
351 | u64 rt_time; | |
352 | u64 rt_throttled; | |
6aa645ea IM |
353 | }; |
354 | ||
57d885fe GH |
355 | #ifdef CONFIG_SMP |
356 | ||
357 | /* | |
358 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
359 | * variables. Each exclusive cpuset essentially defines an island domain by |
360 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
361 | * exclusive cpuset is created, we also create and attach a new root-domain |
362 | * object. | |
363 | * | |
57d885fe GH |
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 | ||
dc938520 GH |
378 | /* |
379 | * By default the system creates a single root-domain with all cpus as | |
380 | * members (mimicking the global state we have today). | |
381 | */ | |
57d885fe GH |
382 | static struct root_domain def_root_domain; |
383 | ||
384 | #endif | |
385 | ||
1da177e4 LT |
386 | /* |
387 | * This is the main, per-CPU runqueue data structure. | |
388 | * | |
389 | * Locking rule: those places that want to lock multiple runqueues | |
390 | * (such as the load balancing or the thread migration code), lock | |
391 | * acquire operations must be ordered by ascending &runqueue. | |
392 | */ | |
70b97a7f | 393 | struct rq { |
d8016491 IM |
394 | /* runqueue lock: */ |
395 | spinlock_t lock; | |
1da177e4 LT |
396 | |
397 | /* | |
398 | * nr_running and cpu_load should be in the same cacheline because | |
399 | * remote CPUs use both these fields when doing load calculation. | |
400 | */ | |
401 | unsigned long nr_running; | |
6aa645ea IM |
402 | #define CPU_LOAD_IDX_MAX 5 |
403 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 404 | unsigned char idle_at_tick; |
46cb4b7c SS |
405 | #ifdef CONFIG_NO_HZ |
406 | unsigned char in_nohz_recently; | |
407 | #endif | |
d8016491 IM |
408 | /* capture load from *all* tasks on this cpu: */ |
409 | struct load_weight load; | |
6aa645ea IM |
410 | unsigned long nr_load_updates; |
411 | u64 nr_switches; | |
412 | ||
413 | struct cfs_rq cfs; | |
414 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
d8016491 IM |
415 | /* list of leaf cfs_rq on this cpu: */ |
416 | struct list_head leaf_cfs_rq_list; | |
1da177e4 | 417 | #endif |
41a2d6cf | 418 | struct rt_rq rt; |
fa85ae24 | 419 | u64 rt_period_expire; |
1da177e4 LT |
420 | |
421 | /* | |
422 | * This is part of a global counter where only the total sum | |
423 | * over all CPUs matters. A task can increase this counter on | |
424 | * one CPU and if it got migrated afterwards it may decrease | |
425 | * it on another CPU. Always updated under the runqueue lock: | |
426 | */ | |
427 | unsigned long nr_uninterruptible; | |
428 | ||
36c8b586 | 429 | struct task_struct *curr, *idle; |
c9819f45 | 430 | unsigned long next_balance; |
1da177e4 | 431 | struct mm_struct *prev_mm; |
6aa645ea | 432 | |
6aa645ea IM |
433 | u64 clock, prev_clock_raw; |
434 | s64 clock_max_delta; | |
435 | ||
436 | unsigned int clock_warps, clock_overflows; | |
2aa44d05 IM |
437 | u64 idle_clock; |
438 | unsigned int clock_deep_idle_events; | |
529c7726 | 439 | u64 tick_timestamp; |
6aa645ea | 440 | |
1da177e4 LT |
441 | atomic_t nr_iowait; |
442 | ||
443 | #ifdef CONFIG_SMP | |
0eab9146 | 444 | struct root_domain *rd; |
1da177e4 LT |
445 | struct sched_domain *sd; |
446 | ||
447 | /* For active balancing */ | |
448 | int active_balance; | |
449 | int push_cpu; | |
d8016491 IM |
450 | /* cpu of this runqueue: */ |
451 | int cpu; | |
1da177e4 | 452 | |
36c8b586 | 453 | struct task_struct *migration_thread; |
1da177e4 LT |
454 | struct list_head migration_queue; |
455 | #endif | |
456 | ||
8f4d37ec PZ |
457 | #ifdef CONFIG_SCHED_HRTICK |
458 | unsigned long hrtick_flags; | |
459 | ktime_t hrtick_expire; | |
460 | struct hrtimer hrtick_timer; | |
461 | #endif | |
462 | ||
1da177e4 LT |
463 | #ifdef CONFIG_SCHEDSTATS |
464 | /* latency stats */ | |
465 | struct sched_info rq_sched_info; | |
466 | ||
467 | /* sys_sched_yield() stats */ | |
480b9434 KC |
468 | unsigned int yld_exp_empty; |
469 | unsigned int yld_act_empty; | |
470 | unsigned int yld_both_empty; | |
471 | unsigned int yld_count; | |
1da177e4 LT |
472 | |
473 | /* schedule() stats */ | |
480b9434 KC |
474 | unsigned int sched_switch; |
475 | unsigned int sched_count; | |
476 | unsigned int sched_goidle; | |
1da177e4 LT |
477 | |
478 | /* try_to_wake_up() stats */ | |
480b9434 KC |
479 | unsigned int ttwu_count; |
480 | unsigned int ttwu_local; | |
b8efb561 IM |
481 | |
482 | /* BKL stats */ | |
480b9434 | 483 | unsigned int bkl_count; |
1da177e4 | 484 | #endif |
fcb99371 | 485 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
486 | }; |
487 | ||
f34e3b61 | 488 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 489 | |
dd41f596 IM |
490 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
491 | { | |
492 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
493 | } | |
494 | ||
0a2966b4 CL |
495 | static inline int cpu_of(struct rq *rq) |
496 | { | |
497 | #ifdef CONFIG_SMP | |
498 | return rq->cpu; | |
499 | #else | |
500 | return 0; | |
501 | #endif | |
502 | } | |
503 | ||
20d315d4 | 504 | /* |
b04a0f4c IM |
505 | * Update the per-runqueue clock, as finegrained as the platform can give |
506 | * us, but without assuming monotonicity, etc.: | |
20d315d4 | 507 | */ |
b04a0f4c | 508 | static void __update_rq_clock(struct rq *rq) |
20d315d4 IM |
509 | { |
510 | u64 prev_raw = rq->prev_clock_raw; | |
511 | u64 now = sched_clock(); | |
512 | s64 delta = now - prev_raw; | |
513 | u64 clock = rq->clock; | |
514 | ||
b04a0f4c IM |
515 | #ifdef CONFIG_SCHED_DEBUG |
516 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
517 | #endif | |
20d315d4 IM |
518 | /* |
519 | * Protect against sched_clock() occasionally going backwards: | |
520 | */ | |
521 | if (unlikely(delta < 0)) { | |
522 | clock++; | |
523 | rq->clock_warps++; | |
524 | } else { | |
525 | /* | |
526 | * Catch too large forward jumps too: | |
527 | */ | |
529c7726 IM |
528 | if (unlikely(clock + delta > rq->tick_timestamp + TICK_NSEC)) { |
529 | if (clock < rq->tick_timestamp + TICK_NSEC) | |
530 | clock = rq->tick_timestamp + TICK_NSEC; | |
531 | else | |
532 | clock++; | |
20d315d4 IM |
533 | rq->clock_overflows++; |
534 | } else { | |
535 | if (unlikely(delta > rq->clock_max_delta)) | |
536 | rq->clock_max_delta = delta; | |
537 | clock += delta; | |
538 | } | |
539 | } | |
540 | ||
541 | rq->prev_clock_raw = now; | |
542 | rq->clock = clock; | |
b04a0f4c | 543 | } |
20d315d4 | 544 | |
b04a0f4c IM |
545 | static void update_rq_clock(struct rq *rq) |
546 | { | |
547 | if (likely(smp_processor_id() == cpu_of(rq))) | |
548 | __update_rq_clock(rq); | |
20d315d4 IM |
549 | } |
550 | ||
674311d5 NP |
551 | /* |
552 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 553 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
554 | * |
555 | * The domain tree of any CPU may only be accessed from within | |
556 | * preempt-disabled sections. | |
557 | */ | |
48f24c4d IM |
558 | #define for_each_domain(cpu, __sd) \ |
559 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
560 | |
561 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
562 | #define this_rq() (&__get_cpu_var(runqueues)) | |
563 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
564 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
565 | ||
bf5c91ba IM |
566 | /* |
567 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
568 | */ | |
569 | #ifdef CONFIG_SCHED_DEBUG | |
570 | # define const_debug __read_mostly | |
571 | #else | |
572 | # define const_debug static const | |
573 | #endif | |
574 | ||
575 | /* | |
576 | * Debugging: various feature bits | |
577 | */ | |
578 | enum { | |
bbdba7c0 | 579 | SCHED_FEAT_NEW_FAIR_SLEEPERS = 1, |
9612633a IM |
580 | SCHED_FEAT_WAKEUP_PREEMPT = 2, |
581 | SCHED_FEAT_START_DEBIT = 4, | |
41a2d6cf IM |
582 | SCHED_FEAT_TREE_AVG = 8, |
583 | SCHED_FEAT_APPROX_AVG = 16, | |
8f4d37ec PZ |
584 | SCHED_FEAT_HRTICK = 32, |
585 | SCHED_FEAT_DOUBLE_TICK = 64, | |
bf5c91ba IM |
586 | }; |
587 | ||
588 | const_debug unsigned int sysctl_sched_features = | |
8401f775 | 589 | SCHED_FEAT_NEW_FAIR_SLEEPERS * 1 | |
9612633a | 590 | SCHED_FEAT_WAKEUP_PREEMPT * 1 | |
8401f775 IM |
591 | SCHED_FEAT_START_DEBIT * 1 | |
592 | SCHED_FEAT_TREE_AVG * 0 | | |
8f4d37ec PZ |
593 | SCHED_FEAT_APPROX_AVG * 0 | |
594 | SCHED_FEAT_HRTICK * 1 | | |
595 | SCHED_FEAT_DOUBLE_TICK * 0; | |
bf5c91ba IM |
596 | |
597 | #define sched_feat(x) (sysctl_sched_features & SCHED_FEAT_##x) | |
598 | ||
b82d9fdd PZ |
599 | /* |
600 | * Number of tasks to iterate in a single balance run. | |
601 | * Limited because this is done with IRQs disabled. | |
602 | */ | |
603 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
604 | ||
fa85ae24 PZ |
605 | /* |
606 | * period over which we measure -rt task cpu usage in ms. | |
607 | * default: 1s | |
608 | */ | |
609 | const_debug unsigned int sysctl_sched_rt_period = 1000; | |
610 | ||
611 | #define SCHED_RT_FRAC_SHIFT 16 | |
612 | #define SCHED_RT_FRAC (1UL << SCHED_RT_FRAC_SHIFT) | |
613 | ||
614 | /* | |
615 | * ratio of time -rt tasks may consume. | |
616 | * default: 100% | |
617 | */ | |
618 | const_debug unsigned int sysctl_sched_rt_ratio = SCHED_RT_FRAC; | |
619 | ||
e436d800 IM |
620 | /* |
621 | * For kernel-internal use: high-speed (but slightly incorrect) per-cpu | |
622 | * clock constructed from sched_clock(): | |
623 | */ | |
624 | unsigned long long cpu_clock(int cpu) | |
625 | { | |
e436d800 IM |
626 | unsigned long long now; |
627 | unsigned long flags; | |
b04a0f4c | 628 | struct rq *rq; |
e436d800 | 629 | |
2cd4d0ea | 630 | local_irq_save(flags); |
b04a0f4c | 631 | rq = cpu_rq(cpu); |
8ced5f69 IM |
632 | /* |
633 | * Only call sched_clock() if the scheduler has already been | |
634 | * initialized (some code might call cpu_clock() very early): | |
635 | */ | |
636 | if (rq->idle) | |
637 | update_rq_clock(rq); | |
b04a0f4c | 638 | now = rq->clock; |
2cd4d0ea | 639 | local_irq_restore(flags); |
e436d800 IM |
640 | |
641 | return now; | |
642 | } | |
a58f6f25 | 643 | EXPORT_SYMBOL_GPL(cpu_clock); |
e436d800 | 644 | |
1da177e4 | 645 | #ifndef prepare_arch_switch |
4866cde0 NP |
646 | # define prepare_arch_switch(next) do { } while (0) |
647 | #endif | |
648 | #ifndef finish_arch_switch | |
649 | # define finish_arch_switch(prev) do { } while (0) | |
650 | #endif | |
651 | ||
051a1d1a DA |
652 | static inline int task_current(struct rq *rq, struct task_struct *p) |
653 | { | |
654 | return rq->curr == p; | |
655 | } | |
656 | ||
4866cde0 | 657 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 658 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 659 | { |
051a1d1a | 660 | return task_current(rq, p); |
4866cde0 NP |
661 | } |
662 | ||
70b97a7f | 663 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
664 | { |
665 | } | |
666 | ||
70b97a7f | 667 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 668 | { |
da04c035 IM |
669 | #ifdef CONFIG_DEBUG_SPINLOCK |
670 | /* this is a valid case when another task releases the spinlock */ | |
671 | rq->lock.owner = current; | |
672 | #endif | |
8a25d5de IM |
673 | /* |
674 | * If we are tracking spinlock dependencies then we have to | |
675 | * fix up the runqueue lock - which gets 'carried over' from | |
676 | * prev into current: | |
677 | */ | |
678 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
679 | ||
4866cde0 NP |
680 | spin_unlock_irq(&rq->lock); |
681 | } | |
682 | ||
683 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 684 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
685 | { |
686 | #ifdef CONFIG_SMP | |
687 | return p->oncpu; | |
688 | #else | |
051a1d1a | 689 | return task_current(rq, p); |
4866cde0 NP |
690 | #endif |
691 | } | |
692 | ||
70b97a7f | 693 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
694 | { |
695 | #ifdef CONFIG_SMP | |
696 | /* | |
697 | * We can optimise this out completely for !SMP, because the | |
698 | * SMP rebalancing from interrupt is the only thing that cares | |
699 | * here. | |
700 | */ | |
701 | next->oncpu = 1; | |
702 | #endif | |
703 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
704 | spin_unlock_irq(&rq->lock); | |
705 | #else | |
706 | spin_unlock(&rq->lock); | |
707 | #endif | |
708 | } | |
709 | ||
70b97a7f | 710 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
711 | { |
712 | #ifdef CONFIG_SMP | |
713 | /* | |
714 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
715 | * We must ensure this doesn't happen until the switch is completely | |
716 | * finished. | |
717 | */ | |
718 | smp_wmb(); | |
719 | prev->oncpu = 0; | |
720 | #endif | |
721 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
722 | local_irq_enable(); | |
1da177e4 | 723 | #endif |
4866cde0 NP |
724 | } |
725 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 726 | |
b29739f9 IM |
727 | /* |
728 | * __task_rq_lock - lock the runqueue a given task resides on. | |
729 | * Must be called interrupts disabled. | |
730 | */ | |
70b97a7f | 731 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
732 | __acquires(rq->lock) |
733 | { | |
3a5c359a AK |
734 | for (;;) { |
735 | struct rq *rq = task_rq(p); | |
736 | spin_lock(&rq->lock); | |
737 | if (likely(rq == task_rq(p))) | |
738 | return rq; | |
b29739f9 | 739 | spin_unlock(&rq->lock); |
b29739f9 | 740 | } |
b29739f9 IM |
741 | } |
742 | ||
1da177e4 LT |
743 | /* |
744 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 745 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
746 | * explicitly disabling preemption. |
747 | */ | |
70b97a7f | 748 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
749 | __acquires(rq->lock) |
750 | { | |
70b97a7f | 751 | struct rq *rq; |
1da177e4 | 752 | |
3a5c359a AK |
753 | for (;;) { |
754 | local_irq_save(*flags); | |
755 | rq = task_rq(p); | |
756 | spin_lock(&rq->lock); | |
757 | if (likely(rq == task_rq(p))) | |
758 | return rq; | |
1da177e4 | 759 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 760 | } |
1da177e4 LT |
761 | } |
762 | ||
a9957449 | 763 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
764 | __releases(rq->lock) |
765 | { | |
766 | spin_unlock(&rq->lock); | |
767 | } | |
768 | ||
70b97a7f | 769 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
770 | __releases(rq->lock) |
771 | { | |
772 | spin_unlock_irqrestore(&rq->lock, *flags); | |
773 | } | |
774 | ||
1da177e4 | 775 | /* |
cc2a73b5 | 776 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 777 | */ |
a9957449 | 778 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
779 | __acquires(rq->lock) |
780 | { | |
70b97a7f | 781 | struct rq *rq; |
1da177e4 LT |
782 | |
783 | local_irq_disable(); | |
784 | rq = this_rq(); | |
785 | spin_lock(&rq->lock); | |
786 | ||
787 | return rq; | |
788 | } | |
789 | ||
1b9f19c2 | 790 | /* |
2aa44d05 | 791 | * We are going deep-idle (irqs are disabled): |
1b9f19c2 | 792 | */ |
2aa44d05 | 793 | void sched_clock_idle_sleep_event(void) |
1b9f19c2 | 794 | { |
2aa44d05 IM |
795 | struct rq *rq = cpu_rq(smp_processor_id()); |
796 | ||
797 | spin_lock(&rq->lock); | |
798 | __update_rq_clock(rq); | |
799 | spin_unlock(&rq->lock); | |
800 | rq->clock_deep_idle_events++; | |
801 | } | |
802 | EXPORT_SYMBOL_GPL(sched_clock_idle_sleep_event); | |
803 | ||
804 | /* | |
805 | * We just idled delta nanoseconds (called with irqs disabled): | |
806 | */ | |
807 | void sched_clock_idle_wakeup_event(u64 delta_ns) | |
808 | { | |
809 | struct rq *rq = cpu_rq(smp_processor_id()); | |
810 | u64 now = sched_clock(); | |
1b9f19c2 | 811 | |
2bacec8c | 812 | touch_softlockup_watchdog(); |
2aa44d05 IM |
813 | rq->idle_clock += delta_ns; |
814 | /* | |
815 | * Override the previous timestamp and ignore all | |
816 | * sched_clock() deltas that occured while we idled, | |
817 | * and use the PM-provided delta_ns to advance the | |
818 | * rq clock: | |
819 | */ | |
820 | spin_lock(&rq->lock); | |
821 | rq->prev_clock_raw = now; | |
822 | rq->clock += delta_ns; | |
823 | spin_unlock(&rq->lock); | |
1b9f19c2 | 824 | } |
2aa44d05 | 825 | EXPORT_SYMBOL_GPL(sched_clock_idle_wakeup_event); |
1b9f19c2 | 826 | |
8f4d37ec PZ |
827 | static void __resched_task(struct task_struct *p, int tif_bit); |
828 | ||
829 | static inline void resched_task(struct task_struct *p) | |
830 | { | |
831 | __resched_task(p, TIF_NEED_RESCHED); | |
832 | } | |
833 | ||
834 | #ifdef CONFIG_SCHED_HRTICK | |
835 | /* | |
836 | * Use HR-timers to deliver accurate preemption points. | |
837 | * | |
838 | * Its all a bit involved since we cannot program an hrt while holding the | |
839 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
840 | * reschedule event. | |
841 | * | |
842 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
843 | * rq->lock. | |
844 | */ | |
845 | static inline void resched_hrt(struct task_struct *p) | |
846 | { | |
847 | __resched_task(p, TIF_HRTICK_RESCHED); | |
848 | } | |
849 | ||
850 | static inline void resched_rq(struct rq *rq) | |
851 | { | |
852 | unsigned long flags; | |
853 | ||
854 | spin_lock_irqsave(&rq->lock, flags); | |
855 | resched_task(rq->curr); | |
856 | spin_unlock_irqrestore(&rq->lock, flags); | |
857 | } | |
858 | ||
859 | enum { | |
860 | HRTICK_SET, /* re-programm hrtick_timer */ | |
861 | HRTICK_RESET, /* not a new slice */ | |
862 | }; | |
863 | ||
864 | /* | |
865 | * Use hrtick when: | |
866 | * - enabled by features | |
867 | * - hrtimer is actually high res | |
868 | */ | |
869 | static inline int hrtick_enabled(struct rq *rq) | |
870 | { | |
871 | if (!sched_feat(HRTICK)) | |
872 | return 0; | |
873 | return hrtimer_is_hres_active(&rq->hrtick_timer); | |
874 | } | |
875 | ||
876 | /* | |
877 | * Called to set the hrtick timer state. | |
878 | * | |
879 | * called with rq->lock held and irqs disabled | |
880 | */ | |
881 | static void hrtick_start(struct rq *rq, u64 delay, int reset) | |
882 | { | |
883 | assert_spin_locked(&rq->lock); | |
884 | ||
885 | /* | |
886 | * preempt at: now + delay | |
887 | */ | |
888 | rq->hrtick_expire = | |
889 | ktime_add_ns(rq->hrtick_timer.base->get_time(), delay); | |
890 | /* | |
891 | * indicate we need to program the timer | |
892 | */ | |
893 | __set_bit(HRTICK_SET, &rq->hrtick_flags); | |
894 | if (reset) | |
895 | __set_bit(HRTICK_RESET, &rq->hrtick_flags); | |
896 | ||
897 | /* | |
898 | * New slices are called from the schedule path and don't need a | |
899 | * forced reschedule. | |
900 | */ | |
901 | if (reset) | |
902 | resched_hrt(rq->curr); | |
903 | } | |
904 | ||
905 | static void hrtick_clear(struct rq *rq) | |
906 | { | |
907 | if (hrtimer_active(&rq->hrtick_timer)) | |
908 | hrtimer_cancel(&rq->hrtick_timer); | |
909 | } | |
910 | ||
911 | /* | |
912 | * Update the timer from the possible pending state. | |
913 | */ | |
914 | static void hrtick_set(struct rq *rq) | |
915 | { | |
916 | ktime_t time; | |
917 | int set, reset; | |
918 | unsigned long flags; | |
919 | ||
920 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
921 | ||
922 | spin_lock_irqsave(&rq->lock, flags); | |
923 | set = __test_and_clear_bit(HRTICK_SET, &rq->hrtick_flags); | |
924 | reset = __test_and_clear_bit(HRTICK_RESET, &rq->hrtick_flags); | |
925 | time = rq->hrtick_expire; | |
926 | clear_thread_flag(TIF_HRTICK_RESCHED); | |
927 | spin_unlock_irqrestore(&rq->lock, flags); | |
928 | ||
929 | if (set) { | |
930 | hrtimer_start(&rq->hrtick_timer, time, HRTIMER_MODE_ABS); | |
931 | if (reset && !hrtimer_active(&rq->hrtick_timer)) | |
932 | resched_rq(rq); | |
933 | } else | |
934 | hrtick_clear(rq); | |
935 | } | |
936 | ||
937 | /* | |
938 | * High-resolution timer tick. | |
939 | * Runs from hardirq context with interrupts disabled. | |
940 | */ | |
941 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
942 | { | |
943 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
944 | ||
945 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
946 | ||
947 | spin_lock(&rq->lock); | |
948 | __update_rq_clock(rq); | |
949 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); | |
950 | spin_unlock(&rq->lock); | |
951 | ||
952 | return HRTIMER_NORESTART; | |
953 | } | |
954 | ||
955 | static inline void init_rq_hrtick(struct rq *rq) | |
956 | { | |
957 | rq->hrtick_flags = 0; | |
958 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
959 | rq->hrtick_timer.function = hrtick; | |
960 | rq->hrtick_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ; | |
961 | } | |
962 | ||
963 | void hrtick_resched(void) | |
964 | { | |
965 | struct rq *rq; | |
966 | unsigned long flags; | |
967 | ||
968 | if (!test_thread_flag(TIF_HRTICK_RESCHED)) | |
969 | return; | |
970 | ||
971 | local_irq_save(flags); | |
972 | rq = cpu_rq(smp_processor_id()); | |
973 | hrtick_set(rq); | |
974 | local_irq_restore(flags); | |
975 | } | |
976 | #else | |
977 | static inline void hrtick_clear(struct rq *rq) | |
978 | { | |
979 | } | |
980 | ||
981 | static inline void hrtick_set(struct rq *rq) | |
982 | { | |
983 | } | |
984 | ||
985 | static inline void init_rq_hrtick(struct rq *rq) | |
986 | { | |
987 | } | |
988 | ||
989 | void hrtick_resched(void) | |
990 | { | |
991 | } | |
992 | #endif | |
993 | ||
c24d20db IM |
994 | /* |
995 | * resched_task - mark a task 'to be rescheduled now'. | |
996 | * | |
997 | * On UP this means the setting of the need_resched flag, on SMP it | |
998 | * might also involve a cross-CPU call to trigger the scheduler on | |
999 | * the target CPU. | |
1000 | */ | |
1001 | #ifdef CONFIG_SMP | |
1002 | ||
1003 | #ifndef tsk_is_polling | |
1004 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1005 | #endif | |
1006 | ||
8f4d37ec | 1007 | static void __resched_task(struct task_struct *p, int tif_bit) |
c24d20db IM |
1008 | { |
1009 | int cpu; | |
1010 | ||
1011 | assert_spin_locked(&task_rq(p)->lock); | |
1012 | ||
8f4d37ec | 1013 | if (unlikely(test_tsk_thread_flag(p, tif_bit))) |
c24d20db IM |
1014 | return; |
1015 | ||
8f4d37ec | 1016 | set_tsk_thread_flag(p, tif_bit); |
c24d20db IM |
1017 | |
1018 | cpu = task_cpu(p); | |
1019 | if (cpu == smp_processor_id()) | |
1020 | return; | |
1021 | ||
1022 | /* NEED_RESCHED must be visible before we test polling */ | |
1023 | smp_mb(); | |
1024 | if (!tsk_is_polling(p)) | |
1025 | smp_send_reschedule(cpu); | |
1026 | } | |
1027 | ||
1028 | static void resched_cpu(int cpu) | |
1029 | { | |
1030 | struct rq *rq = cpu_rq(cpu); | |
1031 | unsigned long flags; | |
1032 | ||
1033 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1034 | return; | |
1035 | resched_task(cpu_curr(cpu)); | |
1036 | spin_unlock_irqrestore(&rq->lock, flags); | |
1037 | } | |
1038 | #else | |
8f4d37ec | 1039 | static void __resched_task(struct task_struct *p, int tif_bit) |
c24d20db IM |
1040 | { |
1041 | assert_spin_locked(&task_rq(p)->lock); | |
8f4d37ec | 1042 | set_tsk_thread_flag(p, tif_bit); |
c24d20db IM |
1043 | } |
1044 | #endif | |
1045 | ||
45bf76df IM |
1046 | #if BITS_PER_LONG == 32 |
1047 | # define WMULT_CONST (~0UL) | |
1048 | #else | |
1049 | # define WMULT_CONST (1UL << 32) | |
1050 | #endif | |
1051 | ||
1052 | #define WMULT_SHIFT 32 | |
1053 | ||
194081eb IM |
1054 | /* |
1055 | * Shift right and round: | |
1056 | */ | |
cf2ab469 | 1057 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1058 | |
cb1c4fc9 | 1059 | static unsigned long |
45bf76df IM |
1060 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1061 | struct load_weight *lw) | |
1062 | { | |
1063 | u64 tmp; | |
1064 | ||
1065 | if (unlikely(!lw->inv_weight)) | |
194081eb | 1066 | lw->inv_weight = (WMULT_CONST - lw->weight/2) / lw->weight + 1; |
45bf76df IM |
1067 | |
1068 | tmp = (u64)delta_exec * weight; | |
1069 | /* | |
1070 | * Check whether we'd overflow the 64-bit multiplication: | |
1071 | */ | |
194081eb | 1072 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1073 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1074 | WMULT_SHIFT/2); |
1075 | else | |
cf2ab469 | 1076 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1077 | |
ecf691da | 1078 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1079 | } |
1080 | ||
1081 | static inline unsigned long | |
1082 | calc_delta_fair(unsigned long delta_exec, struct load_weight *lw) | |
1083 | { | |
1084 | return calc_delta_mine(delta_exec, NICE_0_LOAD, lw); | |
1085 | } | |
1086 | ||
1091985b | 1087 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1088 | { |
1089 | lw->weight += inc; | |
45bf76df IM |
1090 | } |
1091 | ||
1091985b | 1092 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1093 | { |
1094 | lw->weight -= dec; | |
45bf76df IM |
1095 | } |
1096 | ||
2dd73a4f PW |
1097 | /* |
1098 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1099 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1100 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1101 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1102 | * scaled version of the new time slice allocation that they receive on time |
1103 | * slice expiry etc. | |
1104 | */ | |
1105 | ||
dd41f596 IM |
1106 | #define WEIGHT_IDLEPRIO 2 |
1107 | #define WMULT_IDLEPRIO (1 << 31) | |
1108 | ||
1109 | /* | |
1110 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1111 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1112 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1113 | * that remained on nice 0. | |
1114 | * | |
1115 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1116 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1117 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1118 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1119 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1120 | */ |
1121 | static const int prio_to_weight[40] = { | |
254753dc IM |
1122 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1123 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1124 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1125 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1126 | /* 0 */ 1024, 820, 655, 526, 423, | |
1127 | /* 5 */ 335, 272, 215, 172, 137, | |
1128 | /* 10 */ 110, 87, 70, 56, 45, | |
1129 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1130 | }; |
1131 | ||
5714d2de IM |
1132 | /* |
1133 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1134 | * | |
1135 | * In cases where the weight does not change often, we can use the | |
1136 | * precalculated inverse to speed up arithmetics by turning divisions | |
1137 | * into multiplications: | |
1138 | */ | |
dd41f596 | 1139 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1140 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1141 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1142 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1143 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1144 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1145 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1146 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1147 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1148 | }; |
2dd73a4f | 1149 | |
dd41f596 IM |
1150 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1151 | ||
1152 | /* | |
1153 | * runqueue iterator, to support SMP load-balancing between different | |
1154 | * scheduling classes, without having to expose their internal data | |
1155 | * structures to the load-balancing proper: | |
1156 | */ | |
1157 | struct rq_iterator { | |
1158 | void *arg; | |
1159 | struct task_struct *(*start)(void *); | |
1160 | struct task_struct *(*next)(void *); | |
1161 | }; | |
1162 | ||
e1d1484f PW |
1163 | #ifdef CONFIG_SMP |
1164 | static unsigned long | |
1165 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1166 | unsigned long max_load_move, struct sched_domain *sd, | |
1167 | enum cpu_idle_type idle, int *all_pinned, | |
1168 | int *this_best_prio, struct rq_iterator *iterator); | |
1169 | ||
1170 | static int | |
1171 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1172 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1173 | struct rq_iterator *iterator); | |
e1d1484f | 1174 | #endif |
dd41f596 | 1175 | |
d842de87 SV |
1176 | #ifdef CONFIG_CGROUP_CPUACCT |
1177 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
1178 | #else | |
1179 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
1180 | #endif | |
1181 | ||
58e2d4ca SV |
1182 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1183 | { | |
1184 | update_load_add(&rq->load, load); | |
1185 | } | |
1186 | ||
1187 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1188 | { | |
1189 | update_load_sub(&rq->load, load); | |
1190 | } | |
1191 | ||
e7693a36 GH |
1192 | #ifdef CONFIG_SMP |
1193 | static unsigned long source_load(int cpu, int type); | |
1194 | static unsigned long target_load(int cpu, int type); | |
1195 | static unsigned long cpu_avg_load_per_task(int cpu); | |
1196 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1197 | #endif /* CONFIG_SMP */ | |
1198 | ||
dd41f596 | 1199 | #include "sched_stats.h" |
dd41f596 | 1200 | #include "sched_idletask.c" |
5522d5d5 IM |
1201 | #include "sched_fair.c" |
1202 | #include "sched_rt.c" | |
dd41f596 IM |
1203 | #ifdef CONFIG_SCHED_DEBUG |
1204 | # include "sched_debug.c" | |
1205 | #endif | |
1206 | ||
1207 | #define sched_class_highest (&rt_sched_class) | |
1208 | ||
e5fa2237 | 1209 | static void inc_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
1210 | { |
1211 | rq->nr_running++; | |
9c217245 IM |
1212 | } |
1213 | ||
db53181e | 1214 | static void dec_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
1215 | { |
1216 | rq->nr_running--; | |
9c217245 IM |
1217 | } |
1218 | ||
45bf76df IM |
1219 | static void set_load_weight(struct task_struct *p) |
1220 | { | |
1221 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1222 | p->se.load.weight = prio_to_weight[0] * 2; |
1223 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1224 | return; | |
1225 | } | |
45bf76df | 1226 | |
dd41f596 IM |
1227 | /* |
1228 | * SCHED_IDLE tasks get minimal weight: | |
1229 | */ | |
1230 | if (p->policy == SCHED_IDLE) { | |
1231 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1232 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1233 | return; | |
1234 | } | |
71f8bd46 | 1235 | |
dd41f596 IM |
1236 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1237 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1238 | } |
1239 | ||
8159f87e | 1240 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1241 | { |
dd41f596 | 1242 | sched_info_queued(p); |
fd390f6a | 1243 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1244 | p->se.on_rq = 1; |
71f8bd46 IM |
1245 | } |
1246 | ||
69be72c1 | 1247 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1248 | { |
f02231e5 | 1249 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1250 | p->se.on_rq = 0; |
71f8bd46 IM |
1251 | } |
1252 | ||
14531189 | 1253 | /* |
dd41f596 | 1254 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1255 | */ |
14531189 IM |
1256 | static inline int __normal_prio(struct task_struct *p) |
1257 | { | |
dd41f596 | 1258 | return p->static_prio; |
14531189 IM |
1259 | } |
1260 | ||
b29739f9 IM |
1261 | /* |
1262 | * Calculate the expected normal priority: i.e. priority | |
1263 | * without taking RT-inheritance into account. Might be | |
1264 | * boosted by interactivity modifiers. Changes upon fork, | |
1265 | * setprio syscalls, and whenever the interactivity | |
1266 | * estimator recalculates. | |
1267 | */ | |
36c8b586 | 1268 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1269 | { |
1270 | int prio; | |
1271 | ||
e05606d3 | 1272 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1273 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1274 | else | |
1275 | prio = __normal_prio(p); | |
1276 | return prio; | |
1277 | } | |
1278 | ||
1279 | /* | |
1280 | * Calculate the current priority, i.e. the priority | |
1281 | * taken into account by the scheduler. This value might | |
1282 | * be boosted by RT tasks, or might be boosted by | |
1283 | * interactivity modifiers. Will be RT if the task got | |
1284 | * RT-boosted. If not then it returns p->normal_prio. | |
1285 | */ | |
36c8b586 | 1286 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1287 | { |
1288 | p->normal_prio = normal_prio(p); | |
1289 | /* | |
1290 | * If we are RT tasks or we were boosted to RT priority, | |
1291 | * keep the priority unchanged. Otherwise, update priority | |
1292 | * to the normal priority: | |
1293 | */ | |
1294 | if (!rt_prio(p->prio)) | |
1295 | return p->normal_prio; | |
1296 | return p->prio; | |
1297 | } | |
1298 | ||
1da177e4 | 1299 | /* |
dd41f596 | 1300 | * activate_task - move a task to the runqueue. |
1da177e4 | 1301 | */ |
dd41f596 | 1302 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1303 | { |
dd41f596 IM |
1304 | if (p->state == TASK_UNINTERRUPTIBLE) |
1305 | rq->nr_uninterruptible--; | |
1da177e4 | 1306 | |
8159f87e | 1307 | enqueue_task(rq, p, wakeup); |
e5fa2237 | 1308 | inc_nr_running(p, rq); |
1da177e4 LT |
1309 | } |
1310 | ||
1da177e4 LT |
1311 | /* |
1312 | * deactivate_task - remove a task from the runqueue. | |
1313 | */ | |
2e1cb74a | 1314 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1315 | { |
dd41f596 IM |
1316 | if (p->state == TASK_UNINTERRUPTIBLE) |
1317 | rq->nr_uninterruptible++; | |
1318 | ||
69be72c1 | 1319 | dequeue_task(rq, p, sleep); |
db53181e | 1320 | dec_nr_running(p, rq); |
1da177e4 LT |
1321 | } |
1322 | ||
1da177e4 LT |
1323 | /** |
1324 | * task_curr - is this task currently executing on a CPU? | |
1325 | * @p: the task in question. | |
1326 | */ | |
36c8b586 | 1327 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1328 | { |
1329 | return cpu_curr(task_cpu(p)) == p; | |
1330 | } | |
1331 | ||
2dd73a4f PW |
1332 | /* Used instead of source_load when we know the type == 0 */ |
1333 | unsigned long weighted_cpuload(const int cpu) | |
1334 | { | |
495eca49 | 1335 | return cpu_rq(cpu)->load.weight; |
dd41f596 IM |
1336 | } |
1337 | ||
1338 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1339 | { | |
ce96b5ac | 1340 | set_task_cfs_rq(p, cpu); |
dd41f596 | 1341 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1342 | /* |
1343 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1344 | * successfuly executed on another CPU. We must ensure that updates of | |
1345 | * per-task data have been completed by this moment. | |
1346 | */ | |
1347 | smp_wmb(); | |
dd41f596 | 1348 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1349 | #endif |
2dd73a4f PW |
1350 | } |
1351 | ||
cb469845 SR |
1352 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1353 | const struct sched_class *prev_class, | |
1354 | int oldprio, int running) | |
1355 | { | |
1356 | if (prev_class != p->sched_class) { | |
1357 | if (prev_class->switched_from) | |
1358 | prev_class->switched_from(rq, p, running); | |
1359 | p->sched_class->switched_to(rq, p, running); | |
1360 | } else | |
1361 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1362 | } | |
1363 | ||
1da177e4 | 1364 | #ifdef CONFIG_SMP |
c65cc870 | 1365 | |
cc367732 IM |
1366 | /* |
1367 | * Is this task likely cache-hot: | |
1368 | */ | |
e7693a36 | 1369 | static int |
cc367732 IM |
1370 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1371 | { | |
1372 | s64 delta; | |
1373 | ||
1374 | if (p->sched_class != &fair_sched_class) | |
1375 | return 0; | |
1376 | ||
6bc1665b IM |
1377 | if (sysctl_sched_migration_cost == -1) |
1378 | return 1; | |
1379 | if (sysctl_sched_migration_cost == 0) | |
1380 | return 0; | |
1381 | ||
cc367732 IM |
1382 | delta = now - p->se.exec_start; |
1383 | ||
1384 | return delta < (s64)sysctl_sched_migration_cost; | |
1385 | } | |
1386 | ||
1387 | ||
dd41f596 | 1388 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1389 | { |
dd41f596 IM |
1390 | int old_cpu = task_cpu(p); |
1391 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1392 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1393 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1394 | u64 clock_offset; |
dd41f596 IM |
1395 | |
1396 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1397 | |
1398 | #ifdef CONFIG_SCHEDSTATS | |
1399 | if (p->se.wait_start) | |
1400 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1401 | if (p->se.sleep_start) |
1402 | p->se.sleep_start -= clock_offset; | |
1403 | if (p->se.block_start) | |
1404 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1405 | if (old_cpu != new_cpu) { |
1406 | schedstat_inc(p, se.nr_migrations); | |
1407 | if (task_hot(p, old_rq->clock, NULL)) | |
1408 | schedstat_inc(p, se.nr_forced2_migrations); | |
1409 | } | |
6cfb0d5d | 1410 | #endif |
2830cf8c SV |
1411 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1412 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1413 | |
1414 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1415 | } |
1416 | ||
70b97a7f | 1417 | struct migration_req { |
1da177e4 | 1418 | struct list_head list; |
1da177e4 | 1419 | |
36c8b586 | 1420 | struct task_struct *task; |
1da177e4 LT |
1421 | int dest_cpu; |
1422 | ||
1da177e4 | 1423 | struct completion done; |
70b97a7f | 1424 | }; |
1da177e4 LT |
1425 | |
1426 | /* | |
1427 | * The task's runqueue lock must be held. | |
1428 | * Returns true if you have to wait for migration thread. | |
1429 | */ | |
36c8b586 | 1430 | static int |
70b97a7f | 1431 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1432 | { |
70b97a7f | 1433 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1434 | |
1435 | /* | |
1436 | * If the task is not on a runqueue (and not running), then | |
1437 | * it is sufficient to simply update the task's cpu field. | |
1438 | */ | |
dd41f596 | 1439 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1440 | set_task_cpu(p, dest_cpu); |
1441 | return 0; | |
1442 | } | |
1443 | ||
1444 | init_completion(&req->done); | |
1da177e4 LT |
1445 | req->task = p; |
1446 | req->dest_cpu = dest_cpu; | |
1447 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1448 | |
1da177e4 LT |
1449 | return 1; |
1450 | } | |
1451 | ||
1452 | /* | |
1453 | * wait_task_inactive - wait for a thread to unschedule. | |
1454 | * | |
1455 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1456 | * else this function might spin for a *long* time. This function can't | |
1457 | * be called with interrupts off, or it may introduce deadlock with | |
1458 | * smp_call_function() if an IPI is sent by the same process we are | |
1459 | * waiting to become inactive. | |
1460 | */ | |
36c8b586 | 1461 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1462 | { |
1463 | unsigned long flags; | |
dd41f596 | 1464 | int running, on_rq; |
70b97a7f | 1465 | struct rq *rq; |
1da177e4 | 1466 | |
3a5c359a AK |
1467 | for (;;) { |
1468 | /* | |
1469 | * We do the initial early heuristics without holding | |
1470 | * any task-queue locks at all. We'll only try to get | |
1471 | * the runqueue lock when things look like they will | |
1472 | * work out! | |
1473 | */ | |
1474 | rq = task_rq(p); | |
fa490cfd | 1475 | |
3a5c359a AK |
1476 | /* |
1477 | * If the task is actively running on another CPU | |
1478 | * still, just relax and busy-wait without holding | |
1479 | * any locks. | |
1480 | * | |
1481 | * NOTE! Since we don't hold any locks, it's not | |
1482 | * even sure that "rq" stays as the right runqueue! | |
1483 | * But we don't care, since "task_running()" will | |
1484 | * return false if the runqueue has changed and p | |
1485 | * is actually now running somewhere else! | |
1486 | */ | |
1487 | while (task_running(rq, p)) | |
1488 | cpu_relax(); | |
fa490cfd | 1489 | |
3a5c359a AK |
1490 | /* |
1491 | * Ok, time to look more closely! We need the rq | |
1492 | * lock now, to be *sure*. If we're wrong, we'll | |
1493 | * just go back and repeat. | |
1494 | */ | |
1495 | rq = task_rq_lock(p, &flags); | |
1496 | running = task_running(rq, p); | |
1497 | on_rq = p->se.on_rq; | |
1498 | task_rq_unlock(rq, &flags); | |
fa490cfd | 1499 | |
3a5c359a AK |
1500 | /* |
1501 | * Was it really running after all now that we | |
1502 | * checked with the proper locks actually held? | |
1503 | * | |
1504 | * Oops. Go back and try again.. | |
1505 | */ | |
1506 | if (unlikely(running)) { | |
1507 | cpu_relax(); | |
1508 | continue; | |
1509 | } | |
fa490cfd | 1510 | |
3a5c359a AK |
1511 | /* |
1512 | * It's not enough that it's not actively running, | |
1513 | * it must be off the runqueue _entirely_, and not | |
1514 | * preempted! | |
1515 | * | |
1516 | * So if it wa still runnable (but just not actively | |
1517 | * running right now), it's preempted, and we should | |
1518 | * yield - it could be a while. | |
1519 | */ | |
1520 | if (unlikely(on_rq)) { | |
1521 | schedule_timeout_uninterruptible(1); | |
1522 | continue; | |
1523 | } | |
fa490cfd | 1524 | |
3a5c359a AK |
1525 | /* |
1526 | * Ahh, all good. It wasn't running, and it wasn't | |
1527 | * runnable, which means that it will never become | |
1528 | * running in the future either. We're all done! | |
1529 | */ | |
1530 | break; | |
1531 | } | |
1da177e4 LT |
1532 | } |
1533 | ||
1534 | /*** | |
1535 | * kick_process - kick a running thread to enter/exit the kernel | |
1536 | * @p: the to-be-kicked thread | |
1537 | * | |
1538 | * Cause a process which is running on another CPU to enter | |
1539 | * kernel-mode, without any delay. (to get signals handled.) | |
1540 | * | |
1541 | * NOTE: this function doesnt have to take the runqueue lock, | |
1542 | * because all it wants to ensure is that the remote task enters | |
1543 | * the kernel. If the IPI races and the task has been migrated | |
1544 | * to another CPU then no harm is done and the purpose has been | |
1545 | * achieved as well. | |
1546 | */ | |
36c8b586 | 1547 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1548 | { |
1549 | int cpu; | |
1550 | ||
1551 | preempt_disable(); | |
1552 | cpu = task_cpu(p); | |
1553 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1554 | smp_send_reschedule(cpu); | |
1555 | preempt_enable(); | |
1556 | } | |
1557 | ||
1558 | /* | |
2dd73a4f PW |
1559 | * Return a low guess at the load of a migration-source cpu weighted |
1560 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1561 | * |
1562 | * We want to under-estimate the load of migration sources, to | |
1563 | * balance conservatively. | |
1564 | */ | |
a9957449 | 1565 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 1566 | { |
70b97a7f | 1567 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1568 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1569 | |
3b0bd9bc | 1570 | if (type == 0) |
dd41f596 | 1571 | return total; |
b910472d | 1572 | |
dd41f596 | 1573 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
1574 | } |
1575 | ||
1576 | /* | |
2dd73a4f PW |
1577 | * Return a high guess at the load of a migration-target cpu weighted |
1578 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1579 | */ |
a9957449 | 1580 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 1581 | { |
70b97a7f | 1582 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1583 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1584 | |
7897986b | 1585 | if (type == 0) |
dd41f596 | 1586 | return total; |
3b0bd9bc | 1587 | |
dd41f596 | 1588 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
1589 | } |
1590 | ||
1591 | /* | |
1592 | * Return the average load per task on the cpu's run queue | |
1593 | */ | |
e7693a36 | 1594 | static unsigned long cpu_avg_load_per_task(int cpu) |
2dd73a4f | 1595 | { |
70b97a7f | 1596 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1597 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f PW |
1598 | unsigned long n = rq->nr_running; |
1599 | ||
dd41f596 | 1600 | return n ? total / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1601 | } |
1602 | ||
147cbb4b NP |
1603 | /* |
1604 | * find_idlest_group finds and returns the least busy CPU group within the | |
1605 | * domain. | |
1606 | */ | |
1607 | static struct sched_group * | |
1608 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1609 | { | |
1610 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1611 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1612 | int load_idx = sd->forkexec_idx; | |
1613 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1614 | ||
1615 | do { | |
1616 | unsigned long load, avg_load; | |
1617 | int local_group; | |
1618 | int i; | |
1619 | ||
da5a5522 BD |
1620 | /* Skip over this group if it has no CPUs allowed */ |
1621 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 1622 | continue; |
da5a5522 | 1623 | |
147cbb4b | 1624 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1625 | |
1626 | /* Tally up the load of all CPUs in the group */ | |
1627 | avg_load = 0; | |
1628 | ||
1629 | for_each_cpu_mask(i, group->cpumask) { | |
1630 | /* Bias balancing toward cpus of our domain */ | |
1631 | if (local_group) | |
1632 | load = source_load(i, load_idx); | |
1633 | else | |
1634 | load = target_load(i, load_idx); | |
1635 | ||
1636 | avg_load += load; | |
1637 | } | |
1638 | ||
1639 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1640 | avg_load = sg_div_cpu_power(group, |
1641 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1642 | |
1643 | if (local_group) { | |
1644 | this_load = avg_load; | |
1645 | this = group; | |
1646 | } else if (avg_load < min_load) { | |
1647 | min_load = avg_load; | |
1648 | idlest = group; | |
1649 | } | |
3a5c359a | 1650 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
1651 | |
1652 | if (!idlest || 100*this_load < imbalance*min_load) | |
1653 | return NULL; | |
1654 | return idlest; | |
1655 | } | |
1656 | ||
1657 | /* | |
0feaece9 | 1658 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1659 | */ |
95cdf3b7 IM |
1660 | static int |
1661 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) | |
147cbb4b | 1662 | { |
da5a5522 | 1663 | cpumask_t tmp; |
147cbb4b NP |
1664 | unsigned long load, min_load = ULONG_MAX; |
1665 | int idlest = -1; | |
1666 | int i; | |
1667 | ||
da5a5522 BD |
1668 | /* Traverse only the allowed CPUs */ |
1669 | cpus_and(tmp, group->cpumask, p->cpus_allowed); | |
1670 | ||
1671 | for_each_cpu_mask(i, tmp) { | |
2dd73a4f | 1672 | load = weighted_cpuload(i); |
147cbb4b NP |
1673 | |
1674 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1675 | min_load = load; | |
1676 | idlest = i; | |
1677 | } | |
1678 | } | |
1679 | ||
1680 | return idlest; | |
1681 | } | |
1682 | ||
476d139c NP |
1683 | /* |
1684 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1685 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1686 | * SD_BALANCE_EXEC. | |
1687 | * | |
1688 | * Balance, ie. select the least loaded group. | |
1689 | * | |
1690 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1691 | * | |
1692 | * preempt must be disabled. | |
1693 | */ | |
1694 | static int sched_balance_self(int cpu, int flag) | |
1695 | { | |
1696 | struct task_struct *t = current; | |
1697 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1698 | |
c96d145e | 1699 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
1700 | /* |
1701 | * If power savings logic is enabled for a domain, stop there. | |
1702 | */ | |
5c45bf27 SS |
1703 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
1704 | break; | |
476d139c NP |
1705 | if (tmp->flags & flag) |
1706 | sd = tmp; | |
c96d145e | 1707 | } |
476d139c NP |
1708 | |
1709 | while (sd) { | |
1710 | cpumask_t span; | |
1711 | struct sched_group *group; | |
1a848870 SS |
1712 | int new_cpu, weight; |
1713 | ||
1714 | if (!(sd->flags & flag)) { | |
1715 | sd = sd->child; | |
1716 | continue; | |
1717 | } | |
476d139c NP |
1718 | |
1719 | span = sd->span; | |
1720 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1721 | if (!group) { |
1722 | sd = sd->child; | |
1723 | continue; | |
1724 | } | |
476d139c | 1725 | |
da5a5522 | 1726 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
1727 | if (new_cpu == -1 || new_cpu == cpu) { |
1728 | /* Now try balancing at a lower domain level of cpu */ | |
1729 | sd = sd->child; | |
1730 | continue; | |
1731 | } | |
476d139c | 1732 | |
1a848870 | 1733 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1734 | cpu = new_cpu; |
476d139c NP |
1735 | sd = NULL; |
1736 | weight = cpus_weight(span); | |
1737 | for_each_domain(cpu, tmp) { | |
1738 | if (weight <= cpus_weight(tmp->span)) | |
1739 | break; | |
1740 | if (tmp->flags & flag) | |
1741 | sd = tmp; | |
1742 | } | |
1743 | /* while loop will break here if sd == NULL */ | |
1744 | } | |
1745 | ||
1746 | return cpu; | |
1747 | } | |
1748 | ||
1749 | #endif /* CONFIG_SMP */ | |
1da177e4 | 1750 | |
1da177e4 LT |
1751 | /*** |
1752 | * try_to_wake_up - wake up a thread | |
1753 | * @p: the to-be-woken-up thread | |
1754 | * @state: the mask of task states that can be woken | |
1755 | * @sync: do a synchronous wakeup? | |
1756 | * | |
1757 | * Put it on the run-queue if it's not already there. The "current" | |
1758 | * thread is always on the run-queue (except when the actual | |
1759 | * re-schedule is in progress), and as such you're allowed to do | |
1760 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1761 | * runnable without the overhead of this. | |
1762 | * | |
1763 | * returns failure only if the task is already active. | |
1764 | */ | |
36c8b586 | 1765 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 1766 | { |
cc367732 | 1767 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
1768 | unsigned long flags; |
1769 | long old_state; | |
70b97a7f | 1770 | struct rq *rq; |
1da177e4 LT |
1771 | |
1772 | rq = task_rq_lock(p, &flags); | |
1773 | old_state = p->state; | |
1774 | if (!(old_state & state)) | |
1775 | goto out; | |
1776 | ||
dd41f596 | 1777 | if (p->se.on_rq) |
1da177e4 LT |
1778 | goto out_running; |
1779 | ||
1780 | cpu = task_cpu(p); | |
cc367732 | 1781 | orig_cpu = cpu; |
1da177e4 LT |
1782 | this_cpu = smp_processor_id(); |
1783 | ||
1784 | #ifdef CONFIG_SMP | |
1785 | if (unlikely(task_running(rq, p))) | |
1786 | goto out_activate; | |
1787 | ||
5d2f5a61 DA |
1788 | cpu = p->sched_class->select_task_rq(p, sync); |
1789 | if (cpu != orig_cpu) { | |
1790 | set_task_cpu(p, cpu); | |
1da177e4 LT |
1791 | task_rq_unlock(rq, &flags); |
1792 | /* might preempt at this point */ | |
1793 | rq = task_rq_lock(p, &flags); | |
1794 | old_state = p->state; | |
1795 | if (!(old_state & state)) | |
1796 | goto out; | |
dd41f596 | 1797 | if (p->se.on_rq) |
1da177e4 LT |
1798 | goto out_running; |
1799 | ||
1800 | this_cpu = smp_processor_id(); | |
1801 | cpu = task_cpu(p); | |
1802 | } | |
1803 | ||
e7693a36 GH |
1804 | #ifdef CONFIG_SCHEDSTATS |
1805 | schedstat_inc(rq, ttwu_count); | |
1806 | if (cpu == this_cpu) | |
1807 | schedstat_inc(rq, ttwu_local); | |
1808 | else { | |
1809 | struct sched_domain *sd; | |
1810 | for_each_domain(this_cpu, sd) { | |
1811 | if (cpu_isset(cpu, sd->span)) { | |
1812 | schedstat_inc(sd, ttwu_wake_remote); | |
1813 | break; | |
1814 | } | |
1815 | } | |
1816 | } | |
e7693a36 GH |
1817 | #endif |
1818 | ||
1da177e4 LT |
1819 | out_activate: |
1820 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
1821 | schedstat_inc(p, se.nr_wakeups); |
1822 | if (sync) | |
1823 | schedstat_inc(p, se.nr_wakeups_sync); | |
1824 | if (orig_cpu != cpu) | |
1825 | schedstat_inc(p, se.nr_wakeups_migrate); | |
1826 | if (cpu == this_cpu) | |
1827 | schedstat_inc(p, se.nr_wakeups_local); | |
1828 | else | |
1829 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 1830 | update_rq_clock(rq); |
dd41f596 | 1831 | activate_task(rq, p, 1); |
9c63d9c0 | 1832 | check_preempt_curr(rq, p); |
1da177e4 LT |
1833 | success = 1; |
1834 | ||
1835 | out_running: | |
1836 | p->state = TASK_RUNNING; | |
9a897c5a SR |
1837 | #ifdef CONFIG_SMP |
1838 | if (p->sched_class->task_wake_up) | |
1839 | p->sched_class->task_wake_up(rq, p); | |
1840 | #endif | |
1da177e4 LT |
1841 | out: |
1842 | task_rq_unlock(rq, &flags); | |
1843 | ||
1844 | return success; | |
1845 | } | |
1846 | ||
36c8b586 | 1847 | int fastcall wake_up_process(struct task_struct *p) |
1da177e4 LT |
1848 | { |
1849 | return try_to_wake_up(p, TASK_STOPPED | TASK_TRACED | | |
1850 | TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE, 0); | |
1851 | } | |
1da177e4 LT |
1852 | EXPORT_SYMBOL(wake_up_process); |
1853 | ||
36c8b586 | 1854 | int fastcall wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1855 | { |
1856 | return try_to_wake_up(p, state, 0); | |
1857 | } | |
1858 | ||
1da177e4 LT |
1859 | /* |
1860 | * Perform scheduler related setup for a newly forked process p. | |
1861 | * p is forked by current. | |
dd41f596 IM |
1862 | * |
1863 | * __sched_fork() is basic setup used by init_idle() too: | |
1864 | */ | |
1865 | static void __sched_fork(struct task_struct *p) | |
1866 | { | |
dd41f596 IM |
1867 | p->se.exec_start = 0; |
1868 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1869 | p->se.prev_sum_exec_runtime = 0; |
6cfb0d5d IM |
1870 | |
1871 | #ifdef CONFIG_SCHEDSTATS | |
1872 | p->se.wait_start = 0; | |
dd41f596 IM |
1873 | p->se.sum_sleep_runtime = 0; |
1874 | p->se.sleep_start = 0; | |
dd41f596 IM |
1875 | p->se.block_start = 0; |
1876 | p->se.sleep_max = 0; | |
1877 | p->se.block_max = 0; | |
1878 | p->se.exec_max = 0; | |
eba1ed4b | 1879 | p->se.slice_max = 0; |
dd41f596 | 1880 | p->se.wait_max = 0; |
6cfb0d5d | 1881 | #endif |
476d139c | 1882 | |
fa717060 | 1883 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 1884 | p->se.on_rq = 0; |
476d139c | 1885 | |
e107be36 AK |
1886 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1887 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1888 | #endif | |
1889 | ||
1da177e4 LT |
1890 | /* |
1891 | * We mark the process as running here, but have not actually | |
1892 | * inserted it onto the runqueue yet. This guarantees that | |
1893 | * nobody will actually run it, and a signal or other external | |
1894 | * event cannot wake it up and insert it on the runqueue either. | |
1895 | */ | |
1896 | p->state = TASK_RUNNING; | |
dd41f596 IM |
1897 | } |
1898 | ||
1899 | /* | |
1900 | * fork()/clone()-time setup: | |
1901 | */ | |
1902 | void sched_fork(struct task_struct *p, int clone_flags) | |
1903 | { | |
1904 | int cpu = get_cpu(); | |
1905 | ||
1906 | __sched_fork(p); | |
1907 | ||
1908 | #ifdef CONFIG_SMP | |
1909 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
1910 | #endif | |
02e4bac2 | 1911 | set_task_cpu(p, cpu); |
b29739f9 IM |
1912 | |
1913 | /* | |
1914 | * Make sure we do not leak PI boosting priority to the child: | |
1915 | */ | |
1916 | p->prio = current->normal_prio; | |
2ddbf952 HS |
1917 | if (!rt_prio(p->prio)) |
1918 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1919 | |
52f17b6c | 1920 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1921 | if (likely(sched_info_on())) |
52f17b6c | 1922 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1923 | #endif |
d6077cb8 | 1924 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
1925 | p->oncpu = 0; |
1926 | #endif | |
1da177e4 | 1927 | #ifdef CONFIG_PREEMPT |
4866cde0 | 1928 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1929 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1930 | #endif |
476d139c | 1931 | put_cpu(); |
1da177e4 LT |
1932 | } |
1933 | ||
1934 | /* | |
1935 | * wake_up_new_task - wake up a newly created task for the first time. | |
1936 | * | |
1937 | * This function will do some initial scheduler statistics housekeeping | |
1938 | * that must be done for every newly created context, then puts the task | |
1939 | * on the runqueue and wakes it. | |
1940 | */ | |
36c8b586 | 1941 | void fastcall wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
1942 | { |
1943 | unsigned long flags; | |
dd41f596 | 1944 | struct rq *rq; |
1da177e4 LT |
1945 | |
1946 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 1947 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 1948 | update_rq_clock(rq); |
1da177e4 LT |
1949 | |
1950 | p->prio = effective_prio(p); | |
1951 | ||
b9dca1e0 | 1952 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 1953 | activate_task(rq, p, 0); |
1da177e4 | 1954 | } else { |
1da177e4 | 1955 | /* |
dd41f596 IM |
1956 | * Let the scheduling class do new task startup |
1957 | * management (if any): | |
1da177e4 | 1958 | */ |
ee0827d8 | 1959 | p->sched_class->task_new(rq, p); |
e5fa2237 | 1960 | inc_nr_running(p, rq); |
1da177e4 | 1961 | } |
dd41f596 | 1962 | check_preempt_curr(rq, p); |
9a897c5a SR |
1963 | #ifdef CONFIG_SMP |
1964 | if (p->sched_class->task_wake_up) | |
1965 | p->sched_class->task_wake_up(rq, p); | |
1966 | #endif | |
dd41f596 | 1967 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
1968 | } |
1969 | ||
e107be36 AK |
1970 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1971 | ||
1972 | /** | |
421cee29 RD |
1973 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
1974 | * @notifier: notifier struct to register | |
e107be36 AK |
1975 | */ |
1976 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1977 | { | |
1978 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1979 | } | |
1980 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1981 | ||
1982 | /** | |
1983 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1984 | * @notifier: notifier struct to unregister |
e107be36 AK |
1985 | * |
1986 | * This is safe to call from within a preemption notifier. | |
1987 | */ | |
1988 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1989 | { | |
1990 | hlist_del(¬ifier->link); | |
1991 | } | |
1992 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1993 | ||
1994 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1995 | { | |
1996 | struct preempt_notifier *notifier; | |
1997 | struct hlist_node *node; | |
1998 | ||
1999 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2000 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2001 | } | |
2002 | ||
2003 | static void | |
2004 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2005 | struct task_struct *next) | |
2006 | { | |
2007 | struct preempt_notifier *notifier; | |
2008 | struct hlist_node *node; | |
2009 | ||
2010 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2011 | notifier->ops->sched_out(notifier, next); | |
2012 | } | |
2013 | ||
2014 | #else | |
2015 | ||
2016 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2017 | { | |
2018 | } | |
2019 | ||
2020 | static void | |
2021 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2022 | struct task_struct *next) | |
2023 | { | |
2024 | } | |
2025 | ||
2026 | #endif | |
2027 | ||
4866cde0 NP |
2028 | /** |
2029 | * prepare_task_switch - prepare to switch tasks | |
2030 | * @rq: the runqueue preparing to switch | |
421cee29 | 2031 | * @prev: the current task that is being switched out |
4866cde0 NP |
2032 | * @next: the task we are going to switch to. |
2033 | * | |
2034 | * This is called with the rq lock held and interrupts off. It must | |
2035 | * be paired with a subsequent finish_task_switch after the context | |
2036 | * switch. | |
2037 | * | |
2038 | * prepare_task_switch sets up locking and calls architecture specific | |
2039 | * hooks. | |
2040 | */ | |
e107be36 AK |
2041 | static inline void |
2042 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2043 | struct task_struct *next) | |
4866cde0 | 2044 | { |
e107be36 | 2045 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2046 | prepare_lock_switch(rq, next); |
2047 | prepare_arch_switch(next); | |
2048 | } | |
2049 | ||
1da177e4 LT |
2050 | /** |
2051 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2052 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2053 | * @prev: the thread we just switched away from. |
2054 | * | |
4866cde0 NP |
2055 | * finish_task_switch must be called after the context switch, paired |
2056 | * with a prepare_task_switch call before the context switch. | |
2057 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2058 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2059 | * |
2060 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2061 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2062 | * with the lock held can cause deadlocks; see schedule() for |
2063 | * details.) | |
2064 | */ | |
a9957449 | 2065 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2066 | __releases(rq->lock) |
2067 | { | |
1da177e4 | 2068 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2069 | long prev_state; |
1da177e4 LT |
2070 | |
2071 | rq->prev_mm = NULL; | |
2072 | ||
2073 | /* | |
2074 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2075 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2076 | * schedule one last time. The schedule call will never return, and |
2077 | * the scheduled task must drop that reference. | |
c394cc9f | 2078 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2079 | * still held, otherwise prev could be scheduled on another cpu, die |
2080 | * there before we look at prev->state, and then the reference would | |
2081 | * be dropped twice. | |
2082 | * Manfred Spraul <manfred@colorfullife.com> | |
2083 | */ | |
55a101f8 | 2084 | prev_state = prev->state; |
4866cde0 NP |
2085 | finish_arch_switch(prev); |
2086 | finish_lock_switch(rq, prev); | |
9a897c5a SR |
2087 | #ifdef CONFIG_SMP |
2088 | if (current->sched_class->post_schedule) | |
2089 | current->sched_class->post_schedule(rq); | |
2090 | #endif | |
e8fa1362 | 2091 | |
e107be36 | 2092 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2093 | if (mm) |
2094 | mmdrop(mm); | |
c394cc9f | 2095 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2096 | /* |
2097 | * Remove function-return probe instances associated with this | |
2098 | * task and put them back on the free list. | |
9761eea8 | 2099 | */ |
c6fd91f0 | 2100 | kprobe_flush_task(prev); |
1da177e4 | 2101 | put_task_struct(prev); |
c6fd91f0 | 2102 | } |
1da177e4 LT |
2103 | } |
2104 | ||
2105 | /** | |
2106 | * schedule_tail - first thing a freshly forked thread must call. | |
2107 | * @prev: the thread we just switched away from. | |
2108 | */ | |
36c8b586 | 2109 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2110 | __releases(rq->lock) |
2111 | { | |
70b97a7f IM |
2112 | struct rq *rq = this_rq(); |
2113 | ||
4866cde0 NP |
2114 | finish_task_switch(rq, prev); |
2115 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2116 | /* In this case, finish_task_switch does not reenable preemption */ | |
2117 | preempt_enable(); | |
2118 | #endif | |
1da177e4 | 2119 | if (current->set_child_tid) |
b488893a | 2120 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2121 | } |
2122 | ||
2123 | /* | |
2124 | * context_switch - switch to the new MM and the new | |
2125 | * thread's register state. | |
2126 | */ | |
dd41f596 | 2127 | static inline void |
70b97a7f | 2128 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2129 | struct task_struct *next) |
1da177e4 | 2130 | { |
dd41f596 | 2131 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2132 | |
e107be36 | 2133 | prepare_task_switch(rq, prev, next); |
dd41f596 IM |
2134 | mm = next->mm; |
2135 | oldmm = prev->active_mm; | |
9226d125 ZA |
2136 | /* |
2137 | * For paravirt, this is coupled with an exit in switch_to to | |
2138 | * combine the page table reload and the switch backend into | |
2139 | * one hypercall. | |
2140 | */ | |
2141 | arch_enter_lazy_cpu_mode(); | |
2142 | ||
dd41f596 | 2143 | if (unlikely(!mm)) { |
1da177e4 LT |
2144 | next->active_mm = oldmm; |
2145 | atomic_inc(&oldmm->mm_count); | |
2146 | enter_lazy_tlb(oldmm, next); | |
2147 | } else | |
2148 | switch_mm(oldmm, mm, next); | |
2149 | ||
dd41f596 | 2150 | if (unlikely(!prev->mm)) { |
1da177e4 | 2151 | prev->active_mm = NULL; |
1da177e4 LT |
2152 | rq->prev_mm = oldmm; |
2153 | } | |
3a5f5e48 IM |
2154 | /* |
2155 | * Since the runqueue lock will be released by the next | |
2156 | * task (which is an invalid locking op but in the case | |
2157 | * of the scheduler it's an obvious special-case), so we | |
2158 | * do an early lockdep release here: | |
2159 | */ | |
2160 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2161 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2162 | #endif |
1da177e4 LT |
2163 | |
2164 | /* Here we just switch the register state and the stack. */ | |
2165 | switch_to(prev, next, prev); | |
2166 | ||
dd41f596 IM |
2167 | barrier(); |
2168 | /* | |
2169 | * this_rq must be evaluated again because prev may have moved | |
2170 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2171 | * frame will be invalid. | |
2172 | */ | |
2173 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2174 | } |
2175 | ||
2176 | /* | |
2177 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2178 | * | |
2179 | * externally visible scheduler statistics: current number of runnable | |
2180 | * threads, current number of uninterruptible-sleeping threads, total | |
2181 | * number of context switches performed since bootup. | |
2182 | */ | |
2183 | unsigned long nr_running(void) | |
2184 | { | |
2185 | unsigned long i, sum = 0; | |
2186 | ||
2187 | for_each_online_cpu(i) | |
2188 | sum += cpu_rq(i)->nr_running; | |
2189 | ||
2190 | return sum; | |
2191 | } | |
2192 | ||
2193 | unsigned long nr_uninterruptible(void) | |
2194 | { | |
2195 | unsigned long i, sum = 0; | |
2196 | ||
0a945022 | 2197 | for_each_possible_cpu(i) |
1da177e4 LT |
2198 | sum += cpu_rq(i)->nr_uninterruptible; |
2199 | ||
2200 | /* | |
2201 | * Since we read the counters lockless, it might be slightly | |
2202 | * inaccurate. Do not allow it to go below zero though: | |
2203 | */ | |
2204 | if (unlikely((long)sum < 0)) | |
2205 | sum = 0; | |
2206 | ||
2207 | return sum; | |
2208 | } | |
2209 | ||
2210 | unsigned long long nr_context_switches(void) | |
2211 | { | |
cc94abfc SR |
2212 | int i; |
2213 | unsigned long long sum = 0; | |
1da177e4 | 2214 | |
0a945022 | 2215 | for_each_possible_cpu(i) |
1da177e4 LT |
2216 | sum += cpu_rq(i)->nr_switches; |
2217 | ||
2218 | return sum; | |
2219 | } | |
2220 | ||
2221 | unsigned long nr_iowait(void) | |
2222 | { | |
2223 | unsigned long i, sum = 0; | |
2224 | ||
0a945022 | 2225 | for_each_possible_cpu(i) |
1da177e4 LT |
2226 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2227 | ||
2228 | return sum; | |
2229 | } | |
2230 | ||
db1b1fef JS |
2231 | unsigned long nr_active(void) |
2232 | { | |
2233 | unsigned long i, running = 0, uninterruptible = 0; | |
2234 | ||
2235 | for_each_online_cpu(i) { | |
2236 | running += cpu_rq(i)->nr_running; | |
2237 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2238 | } | |
2239 | ||
2240 | if (unlikely((long)uninterruptible < 0)) | |
2241 | uninterruptible = 0; | |
2242 | ||
2243 | return running + uninterruptible; | |
2244 | } | |
2245 | ||
48f24c4d | 2246 | /* |
dd41f596 IM |
2247 | * Update rq->cpu_load[] statistics. This function is usually called every |
2248 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2249 | */ |
dd41f596 | 2250 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2251 | { |
495eca49 | 2252 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2253 | int i, scale; |
2254 | ||
2255 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2256 | |
2257 | /* Update our load: */ | |
2258 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2259 | unsigned long old_load, new_load; | |
2260 | ||
2261 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2262 | ||
2263 | old_load = this_rq->cpu_load[i]; | |
2264 | new_load = this_load; | |
a25707f3 IM |
2265 | /* |
2266 | * Round up the averaging division if load is increasing. This | |
2267 | * prevents us from getting stuck on 9 if the load is 10, for | |
2268 | * example. | |
2269 | */ | |
2270 | if (new_load > old_load) | |
2271 | new_load += scale-1; | |
dd41f596 IM |
2272 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2273 | } | |
48f24c4d IM |
2274 | } |
2275 | ||
dd41f596 IM |
2276 | #ifdef CONFIG_SMP |
2277 | ||
1da177e4 LT |
2278 | /* |
2279 | * double_rq_lock - safely lock two runqueues | |
2280 | * | |
2281 | * Note this does not disable interrupts like task_rq_lock, | |
2282 | * you need to do so manually before calling. | |
2283 | */ | |
70b97a7f | 2284 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2285 | __acquires(rq1->lock) |
2286 | __acquires(rq2->lock) | |
2287 | { | |
054b9108 | 2288 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2289 | if (rq1 == rq2) { |
2290 | spin_lock(&rq1->lock); | |
2291 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2292 | } else { | |
c96d145e | 2293 | if (rq1 < rq2) { |
1da177e4 LT |
2294 | spin_lock(&rq1->lock); |
2295 | spin_lock(&rq2->lock); | |
2296 | } else { | |
2297 | spin_lock(&rq2->lock); | |
2298 | spin_lock(&rq1->lock); | |
2299 | } | |
2300 | } | |
6e82a3be IM |
2301 | update_rq_clock(rq1); |
2302 | update_rq_clock(rq2); | |
1da177e4 LT |
2303 | } |
2304 | ||
2305 | /* | |
2306 | * double_rq_unlock - safely unlock two runqueues | |
2307 | * | |
2308 | * Note this does not restore interrupts like task_rq_unlock, | |
2309 | * you need to do so manually after calling. | |
2310 | */ | |
70b97a7f | 2311 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2312 | __releases(rq1->lock) |
2313 | __releases(rq2->lock) | |
2314 | { | |
2315 | spin_unlock(&rq1->lock); | |
2316 | if (rq1 != rq2) | |
2317 | spin_unlock(&rq2->lock); | |
2318 | else | |
2319 | __release(rq2->lock); | |
2320 | } | |
2321 | ||
2322 | /* | |
2323 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2324 | */ | |
e8fa1362 | 2325 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2326 | __releases(this_rq->lock) |
2327 | __acquires(busiest->lock) | |
2328 | __acquires(this_rq->lock) | |
2329 | { | |
e8fa1362 SR |
2330 | int ret = 0; |
2331 | ||
054b9108 KK |
2332 | if (unlikely(!irqs_disabled())) { |
2333 | /* printk() doesn't work good under rq->lock */ | |
2334 | spin_unlock(&this_rq->lock); | |
2335 | BUG_ON(1); | |
2336 | } | |
1da177e4 | 2337 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2338 | if (busiest < this_rq) { |
1da177e4 LT |
2339 | spin_unlock(&this_rq->lock); |
2340 | spin_lock(&busiest->lock); | |
2341 | spin_lock(&this_rq->lock); | |
e8fa1362 | 2342 | ret = 1; |
1da177e4 LT |
2343 | } else |
2344 | spin_lock(&busiest->lock); | |
2345 | } | |
e8fa1362 | 2346 | return ret; |
1da177e4 LT |
2347 | } |
2348 | ||
1da177e4 LT |
2349 | /* |
2350 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2351 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2352 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2353 | * the cpu_allowed mask is restored. |
2354 | */ | |
36c8b586 | 2355 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2356 | { |
70b97a7f | 2357 | struct migration_req req; |
1da177e4 | 2358 | unsigned long flags; |
70b97a7f | 2359 | struct rq *rq; |
1da177e4 LT |
2360 | |
2361 | rq = task_rq_lock(p, &flags); | |
2362 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2363 | || unlikely(cpu_is_offline(dest_cpu))) | |
2364 | goto out; | |
2365 | ||
2366 | /* force the process onto the specified CPU */ | |
2367 | if (migrate_task(p, dest_cpu, &req)) { | |
2368 | /* Need to wait for migration thread (might exit: take ref). */ | |
2369 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2370 | |
1da177e4 LT |
2371 | get_task_struct(mt); |
2372 | task_rq_unlock(rq, &flags); | |
2373 | wake_up_process(mt); | |
2374 | put_task_struct(mt); | |
2375 | wait_for_completion(&req.done); | |
36c8b586 | 2376 | |
1da177e4 LT |
2377 | return; |
2378 | } | |
2379 | out: | |
2380 | task_rq_unlock(rq, &flags); | |
2381 | } | |
2382 | ||
2383 | /* | |
476d139c NP |
2384 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2385 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2386 | */ |
2387 | void sched_exec(void) | |
2388 | { | |
1da177e4 | 2389 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2390 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2391 | put_cpu(); |
476d139c NP |
2392 | if (new_cpu != this_cpu) |
2393 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2394 | } |
2395 | ||
2396 | /* | |
2397 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2398 | * Both runqueues must be locked. | |
2399 | */ | |
dd41f596 IM |
2400 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2401 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2402 | { |
2e1cb74a | 2403 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2404 | set_task_cpu(p, this_cpu); |
dd41f596 | 2405 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2406 | /* |
2407 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2408 | * to be always true for them. | |
2409 | */ | |
dd41f596 | 2410 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2411 | } |
2412 | ||
2413 | /* | |
2414 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2415 | */ | |
858119e1 | 2416 | static |
70b97a7f | 2417 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2418 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2419 | int *all_pinned) |
1da177e4 LT |
2420 | { |
2421 | /* | |
2422 | * We do not migrate tasks that are: | |
2423 | * 1) running (obviously), or | |
2424 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2425 | * 3) are cache-hot on their current CPU. | |
2426 | */ | |
cc367732 IM |
2427 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2428 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2429 | return 0; |
cc367732 | 2430 | } |
81026794 NP |
2431 | *all_pinned = 0; |
2432 | ||
cc367732 IM |
2433 | if (task_running(rq, p)) { |
2434 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2435 | return 0; |
cc367732 | 2436 | } |
1da177e4 | 2437 | |
da84d961 IM |
2438 | /* |
2439 | * Aggressive migration if: | |
2440 | * 1) task is cache cold, or | |
2441 | * 2) too many balance attempts have failed. | |
2442 | */ | |
2443 | ||
6bc1665b IM |
2444 | if (!task_hot(p, rq->clock, sd) || |
2445 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2446 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2447 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2448 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2449 | schedstat_inc(p, se.nr_forced_migrations); |
2450 | } | |
da84d961 IM |
2451 | #endif |
2452 | return 1; | |
2453 | } | |
2454 | ||
cc367732 IM |
2455 | if (task_hot(p, rq->clock, sd)) { |
2456 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2457 | return 0; |
cc367732 | 2458 | } |
1da177e4 LT |
2459 | return 1; |
2460 | } | |
2461 | ||
e1d1484f PW |
2462 | static unsigned long |
2463 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2464 | unsigned long max_load_move, struct sched_domain *sd, | |
2465 | enum cpu_idle_type idle, int *all_pinned, | |
2466 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2467 | { |
b82d9fdd | 2468 | int loops = 0, pulled = 0, pinned = 0, skip_for_load; |
dd41f596 IM |
2469 | struct task_struct *p; |
2470 | long rem_load_move = max_load_move; | |
1da177e4 | 2471 | |
e1d1484f | 2472 | if (max_load_move == 0) |
1da177e4 LT |
2473 | goto out; |
2474 | ||
81026794 NP |
2475 | pinned = 1; |
2476 | ||
1da177e4 | 2477 | /* |
dd41f596 | 2478 | * Start the load-balancing iterator: |
1da177e4 | 2479 | */ |
dd41f596 IM |
2480 | p = iterator->start(iterator->arg); |
2481 | next: | |
b82d9fdd | 2482 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2483 | goto out; |
50ddd969 | 2484 | /* |
b82d9fdd | 2485 | * To help distribute high priority tasks across CPUs we don't |
50ddd969 PW |
2486 | * skip a task if it will be the highest priority task (i.e. smallest |
2487 | * prio value) on its new queue regardless of its load weight | |
2488 | */ | |
dd41f596 IM |
2489 | skip_for_load = (p->se.load.weight >> 1) > rem_load_move + |
2490 | SCHED_LOAD_SCALE_FUZZ; | |
a4ac01c3 | 2491 | if ((skip_for_load && p->prio >= *this_best_prio) || |
dd41f596 | 2492 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2493 | p = iterator->next(iterator->arg); |
2494 | goto next; | |
1da177e4 LT |
2495 | } |
2496 | ||
dd41f596 | 2497 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2498 | pulled++; |
dd41f596 | 2499 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2500 | |
2dd73a4f | 2501 | /* |
b82d9fdd | 2502 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 2503 | */ |
e1d1484f | 2504 | if (rem_load_move > 0) { |
a4ac01c3 PW |
2505 | if (p->prio < *this_best_prio) |
2506 | *this_best_prio = p->prio; | |
dd41f596 IM |
2507 | p = iterator->next(iterator->arg); |
2508 | goto next; | |
1da177e4 LT |
2509 | } |
2510 | out: | |
2511 | /* | |
e1d1484f | 2512 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
2513 | * so we can safely collect pull_task() stats here rather than |
2514 | * inside pull_task(). | |
2515 | */ | |
2516 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2517 | |
2518 | if (all_pinned) | |
2519 | *all_pinned = pinned; | |
e1d1484f PW |
2520 | |
2521 | return max_load_move - rem_load_move; | |
1da177e4 LT |
2522 | } |
2523 | ||
dd41f596 | 2524 | /* |
43010659 PW |
2525 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2526 | * this_rq, as part of a balancing operation within domain "sd". | |
2527 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2528 | * |
2529 | * Called with both runqueues locked. | |
2530 | */ | |
2531 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2532 | unsigned long max_load_move, |
dd41f596 IM |
2533 | struct sched_domain *sd, enum cpu_idle_type idle, |
2534 | int *all_pinned) | |
2535 | { | |
5522d5d5 | 2536 | const struct sched_class *class = sched_class_highest; |
43010659 | 2537 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2538 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
2539 | |
2540 | do { | |
43010659 PW |
2541 | total_load_moved += |
2542 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 2543 | max_load_move - total_load_moved, |
a4ac01c3 | 2544 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 2545 | class = class->next; |
43010659 | 2546 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 2547 | |
43010659 PW |
2548 | return total_load_moved > 0; |
2549 | } | |
2550 | ||
e1d1484f PW |
2551 | static int |
2552 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2553 | struct sched_domain *sd, enum cpu_idle_type idle, | |
2554 | struct rq_iterator *iterator) | |
2555 | { | |
2556 | struct task_struct *p = iterator->start(iterator->arg); | |
2557 | int pinned = 0; | |
2558 | ||
2559 | while (p) { | |
2560 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
2561 | pull_task(busiest, p, this_rq, this_cpu); | |
2562 | /* | |
2563 | * Right now, this is only the second place pull_task() | |
2564 | * is called, so we can safely collect pull_task() | |
2565 | * stats here rather than inside pull_task(). | |
2566 | */ | |
2567 | schedstat_inc(sd, lb_gained[idle]); | |
2568 | ||
2569 | return 1; | |
2570 | } | |
2571 | p = iterator->next(iterator->arg); | |
2572 | } | |
2573 | ||
2574 | return 0; | |
2575 | } | |
2576 | ||
43010659 PW |
2577 | /* |
2578 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
2579 | * part of active balancing operations within "domain". | |
2580 | * Returns 1 if successful and 0 otherwise. | |
2581 | * | |
2582 | * Called with both runqueues locked. | |
2583 | */ | |
2584 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2585 | struct sched_domain *sd, enum cpu_idle_type idle) | |
2586 | { | |
5522d5d5 | 2587 | const struct sched_class *class; |
43010659 PW |
2588 | |
2589 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 2590 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
2591 | return 1; |
2592 | ||
2593 | return 0; | |
dd41f596 IM |
2594 | } |
2595 | ||
1da177e4 LT |
2596 | /* |
2597 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2598 | * domain. It calculates and returns the amount of weighted load which |
2599 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2600 | */ |
2601 | static struct sched_group * | |
2602 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 IM |
2603 | unsigned long *imbalance, enum cpu_idle_type idle, |
2604 | int *sd_idle, cpumask_t *cpus, int *balance) | |
1da177e4 LT |
2605 | { |
2606 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2607 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2608 | unsigned long max_pull; |
2dd73a4f PW |
2609 | unsigned long busiest_load_per_task, busiest_nr_running; |
2610 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 2611 | int load_idx, group_imb = 0; |
5c45bf27 SS |
2612 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2613 | int power_savings_balance = 1; | |
2614 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2615 | unsigned long min_nr_running = ULONG_MAX; | |
2616 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2617 | #endif | |
1da177e4 LT |
2618 | |
2619 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2620 | busiest_load_per_task = busiest_nr_running = 0; |
2621 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2622 | if (idle == CPU_NOT_IDLE) |
7897986b | 2623 | load_idx = sd->busy_idx; |
d15bcfdb | 2624 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2625 | load_idx = sd->newidle_idx; |
2626 | else | |
2627 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2628 | |
2629 | do { | |
908a7c1b | 2630 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
2631 | int local_group; |
2632 | int i; | |
908a7c1b | 2633 | int __group_imb = 0; |
783609c6 | 2634 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2635 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2636 | |
2637 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2638 | ||
783609c6 SS |
2639 | if (local_group) |
2640 | balance_cpu = first_cpu(group->cpumask); | |
2641 | ||
1da177e4 | 2642 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2643 | sum_weighted_load = sum_nr_running = avg_load = 0; |
908a7c1b KC |
2644 | max_cpu_load = 0; |
2645 | min_cpu_load = ~0UL; | |
1da177e4 LT |
2646 | |
2647 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2648 | struct rq *rq; |
2649 | ||
2650 | if (!cpu_isset(i, *cpus)) | |
2651 | continue; | |
2652 | ||
2653 | rq = cpu_rq(i); | |
2dd73a4f | 2654 | |
9439aab8 | 2655 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
2656 | *sd_idle = 0; |
2657 | ||
1da177e4 | 2658 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2659 | if (local_group) { |
2660 | if (idle_cpu(i) && !first_idle_cpu) { | |
2661 | first_idle_cpu = 1; | |
2662 | balance_cpu = i; | |
2663 | } | |
2664 | ||
a2000572 | 2665 | load = target_load(i, load_idx); |
908a7c1b | 2666 | } else { |
a2000572 | 2667 | load = source_load(i, load_idx); |
908a7c1b KC |
2668 | if (load > max_cpu_load) |
2669 | max_cpu_load = load; | |
2670 | if (min_cpu_load > load) | |
2671 | min_cpu_load = load; | |
2672 | } | |
1da177e4 LT |
2673 | |
2674 | avg_load += load; | |
2dd73a4f | 2675 | sum_nr_running += rq->nr_running; |
dd41f596 | 2676 | sum_weighted_load += weighted_cpuload(i); |
1da177e4 LT |
2677 | } |
2678 | ||
783609c6 SS |
2679 | /* |
2680 | * First idle cpu or the first cpu(busiest) in this sched group | |
2681 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
2682 | * domains. In the newly idle case, we will allow all the cpu's |
2683 | * to do the newly idle load balance. | |
783609c6 | 2684 | */ |
9439aab8 SS |
2685 | if (idle != CPU_NEWLY_IDLE && local_group && |
2686 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
2687 | *balance = 0; |
2688 | goto ret; | |
2689 | } | |
2690 | ||
1da177e4 | 2691 | total_load += avg_load; |
5517d86b | 2692 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2693 | |
2694 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2695 | avg_load = sg_div_cpu_power(group, |
2696 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2697 | |
908a7c1b KC |
2698 | if ((max_cpu_load - min_cpu_load) > SCHED_LOAD_SCALE) |
2699 | __group_imb = 1; | |
2700 | ||
5517d86b | 2701 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2702 | |
1da177e4 LT |
2703 | if (local_group) { |
2704 | this_load = avg_load; | |
2705 | this = group; | |
2dd73a4f PW |
2706 | this_nr_running = sum_nr_running; |
2707 | this_load_per_task = sum_weighted_load; | |
2708 | } else if (avg_load > max_load && | |
908a7c1b | 2709 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
2710 | max_load = avg_load; |
2711 | busiest = group; | |
2dd73a4f PW |
2712 | busiest_nr_running = sum_nr_running; |
2713 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 2714 | group_imb = __group_imb; |
1da177e4 | 2715 | } |
5c45bf27 SS |
2716 | |
2717 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2718 | /* | |
2719 | * Busy processors will not participate in power savings | |
2720 | * balance. | |
2721 | */ | |
dd41f596 IM |
2722 | if (idle == CPU_NOT_IDLE || |
2723 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
2724 | goto group_next; | |
5c45bf27 SS |
2725 | |
2726 | /* | |
2727 | * If the local group is idle or completely loaded | |
2728 | * no need to do power savings balance at this domain | |
2729 | */ | |
2730 | if (local_group && (this_nr_running >= group_capacity || | |
2731 | !this_nr_running)) | |
2732 | power_savings_balance = 0; | |
2733 | ||
dd41f596 | 2734 | /* |
5c45bf27 SS |
2735 | * If a group is already running at full capacity or idle, |
2736 | * don't include that group in power savings calculations | |
dd41f596 IM |
2737 | */ |
2738 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 2739 | || !sum_nr_running) |
dd41f596 | 2740 | goto group_next; |
5c45bf27 | 2741 | |
dd41f596 | 2742 | /* |
5c45bf27 | 2743 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
2744 | * This is the group from where we need to pick up the load |
2745 | * for saving power | |
2746 | */ | |
2747 | if ((sum_nr_running < min_nr_running) || | |
2748 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
2749 | first_cpu(group->cpumask) < |
2750 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
2751 | group_min = group; |
2752 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
2753 | min_load_per_task = sum_weighted_load / |
2754 | sum_nr_running; | |
dd41f596 | 2755 | } |
5c45bf27 | 2756 | |
dd41f596 | 2757 | /* |
5c45bf27 | 2758 | * Calculate the group which is almost near its |
dd41f596 IM |
2759 | * capacity but still has some space to pick up some load |
2760 | * from other group and save more power | |
2761 | */ | |
2762 | if (sum_nr_running <= group_capacity - 1) { | |
2763 | if (sum_nr_running > leader_nr_running || | |
2764 | (sum_nr_running == leader_nr_running && | |
2765 | first_cpu(group->cpumask) > | |
2766 | first_cpu(group_leader->cpumask))) { | |
2767 | group_leader = group; | |
2768 | leader_nr_running = sum_nr_running; | |
2769 | } | |
48f24c4d | 2770 | } |
5c45bf27 SS |
2771 | group_next: |
2772 | #endif | |
1da177e4 LT |
2773 | group = group->next; |
2774 | } while (group != sd->groups); | |
2775 | ||
2dd73a4f | 2776 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
2777 | goto out_balanced; |
2778 | ||
2779 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
2780 | ||
2781 | if (this_load >= avg_load || | |
2782 | 100*max_load <= sd->imbalance_pct*this_load) | |
2783 | goto out_balanced; | |
2784 | ||
2dd73a4f | 2785 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
2786 | if (group_imb) |
2787 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
2788 | ||
1da177e4 LT |
2789 | /* |
2790 | * We're trying to get all the cpus to the average_load, so we don't | |
2791 | * want to push ourselves above the average load, nor do we wish to | |
2792 | * reduce the max loaded cpu below the average load, as either of these | |
2793 | * actions would just result in more rebalancing later, and ping-pong | |
2794 | * tasks around. Thus we look for the minimum possible imbalance. | |
2795 | * Negative imbalances (*we* are more loaded than anyone else) will | |
2796 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 2797 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
2798 | * appear as very large values with unsigned longs. |
2799 | */ | |
2dd73a4f PW |
2800 | if (max_load <= busiest_load_per_task) |
2801 | goto out_balanced; | |
2802 | ||
2803 | /* | |
2804 | * In the presence of smp nice balancing, certain scenarios can have | |
2805 | * max load less than avg load(as we skip the groups at or below | |
2806 | * its cpu_power, while calculating max_load..) | |
2807 | */ | |
2808 | if (max_load < avg_load) { | |
2809 | *imbalance = 0; | |
2810 | goto small_imbalance; | |
2811 | } | |
0c117f1b SS |
2812 | |
2813 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 2814 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 2815 | |
1da177e4 | 2816 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
2817 | *imbalance = min(max_pull * busiest->__cpu_power, |
2818 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
2819 | / SCHED_LOAD_SCALE; |
2820 | ||
2dd73a4f PW |
2821 | /* |
2822 | * if *imbalance is less than the average load per runnable task | |
2823 | * there is no gaurantee that any tasks will be moved so we'll have | |
2824 | * a think about bumping its value to force at least one task to be | |
2825 | * moved | |
2826 | */ | |
7fd0d2dd | 2827 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 2828 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
2829 | unsigned int imbn; |
2830 | ||
2831 | small_imbalance: | |
2832 | pwr_move = pwr_now = 0; | |
2833 | imbn = 2; | |
2834 | if (this_nr_running) { | |
2835 | this_load_per_task /= this_nr_running; | |
2836 | if (busiest_load_per_task > this_load_per_task) | |
2837 | imbn = 1; | |
2838 | } else | |
2839 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 2840 | |
dd41f596 IM |
2841 | if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >= |
2842 | busiest_load_per_task * imbn) { | |
2dd73a4f | 2843 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2844 | return busiest; |
2845 | } | |
2846 | ||
2847 | /* | |
2848 | * OK, we don't have enough imbalance to justify moving tasks, | |
2849 | * however we may be able to increase total CPU power used by | |
2850 | * moving them. | |
2851 | */ | |
2852 | ||
5517d86b ED |
2853 | pwr_now += busiest->__cpu_power * |
2854 | min(busiest_load_per_task, max_load); | |
2855 | pwr_now += this->__cpu_power * | |
2856 | min(this_load_per_task, this_load); | |
1da177e4 LT |
2857 | pwr_now /= SCHED_LOAD_SCALE; |
2858 | ||
2859 | /* Amount of load we'd subtract */ | |
5517d86b ED |
2860 | tmp = sg_div_cpu_power(busiest, |
2861 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 2862 | if (max_load > tmp) |
5517d86b | 2863 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 2864 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
2865 | |
2866 | /* Amount of load we'd add */ | |
5517d86b | 2867 | if (max_load * busiest->__cpu_power < |
33859f7f | 2868 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
2869 | tmp = sg_div_cpu_power(this, |
2870 | max_load * busiest->__cpu_power); | |
1da177e4 | 2871 | else |
5517d86b ED |
2872 | tmp = sg_div_cpu_power(this, |
2873 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
2874 | pwr_move += this->__cpu_power * | |
2875 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
2876 | pwr_move /= SCHED_LOAD_SCALE; |
2877 | ||
2878 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
2879 | if (pwr_move > pwr_now) |
2880 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
2881 | } |
2882 | ||
1da177e4 LT |
2883 | return busiest; |
2884 | ||
2885 | out_balanced: | |
5c45bf27 | 2886 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 2887 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 2888 | goto ret; |
1da177e4 | 2889 | |
5c45bf27 SS |
2890 | if (this == group_leader && group_leader != group_min) { |
2891 | *imbalance = min_load_per_task; | |
2892 | return group_min; | |
2893 | } | |
5c45bf27 | 2894 | #endif |
783609c6 | 2895 | ret: |
1da177e4 LT |
2896 | *imbalance = 0; |
2897 | return NULL; | |
2898 | } | |
2899 | ||
2900 | /* | |
2901 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
2902 | */ | |
70b97a7f | 2903 | static struct rq * |
d15bcfdb | 2904 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
0a2966b4 | 2905 | unsigned long imbalance, cpumask_t *cpus) |
1da177e4 | 2906 | { |
70b97a7f | 2907 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 2908 | unsigned long max_load = 0; |
1da177e4 LT |
2909 | int i; |
2910 | ||
2911 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 2912 | unsigned long wl; |
0a2966b4 CL |
2913 | |
2914 | if (!cpu_isset(i, *cpus)) | |
2915 | continue; | |
2916 | ||
48f24c4d | 2917 | rq = cpu_rq(i); |
dd41f596 | 2918 | wl = weighted_cpuload(i); |
2dd73a4f | 2919 | |
dd41f596 | 2920 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 2921 | continue; |
1da177e4 | 2922 | |
dd41f596 IM |
2923 | if (wl > max_load) { |
2924 | max_load = wl; | |
48f24c4d | 2925 | busiest = rq; |
1da177e4 LT |
2926 | } |
2927 | } | |
2928 | ||
2929 | return busiest; | |
2930 | } | |
2931 | ||
77391d71 NP |
2932 | /* |
2933 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
2934 | * so long as it is large enough. | |
2935 | */ | |
2936 | #define MAX_PINNED_INTERVAL 512 | |
2937 | ||
1da177e4 LT |
2938 | /* |
2939 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2940 | * tasks if there is an imbalance. | |
1da177e4 | 2941 | */ |
70b97a7f | 2942 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 2943 | struct sched_domain *sd, enum cpu_idle_type idle, |
783609c6 | 2944 | int *balance) |
1da177e4 | 2945 | { |
43010659 | 2946 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 2947 | struct sched_group *group; |
1da177e4 | 2948 | unsigned long imbalance; |
70b97a7f | 2949 | struct rq *busiest; |
0a2966b4 | 2950 | cpumask_t cpus = CPU_MASK_ALL; |
fe2eea3f | 2951 | unsigned long flags; |
5969fe06 | 2952 | |
89c4710e SS |
2953 | /* |
2954 | * When power savings policy is enabled for the parent domain, idle | |
2955 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 2956 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 2957 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 2958 | */ |
d15bcfdb | 2959 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2960 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2961 | sd_idle = 1; |
1da177e4 | 2962 | |
2d72376b | 2963 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 2964 | |
0a2966b4 CL |
2965 | redo: |
2966 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
783609c6 SS |
2967 | &cpus, balance); |
2968 | ||
06066714 | 2969 | if (*balance == 0) |
783609c6 | 2970 | goto out_balanced; |
783609c6 | 2971 | |
1da177e4 LT |
2972 | if (!group) { |
2973 | schedstat_inc(sd, lb_nobusyg[idle]); | |
2974 | goto out_balanced; | |
2975 | } | |
2976 | ||
0a2966b4 | 2977 | busiest = find_busiest_queue(group, idle, imbalance, &cpus); |
1da177e4 LT |
2978 | if (!busiest) { |
2979 | schedstat_inc(sd, lb_nobusyq[idle]); | |
2980 | goto out_balanced; | |
2981 | } | |
2982 | ||
db935dbd | 2983 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
2984 | |
2985 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
2986 | ||
43010659 | 2987 | ld_moved = 0; |
1da177e4 LT |
2988 | if (busiest->nr_running > 1) { |
2989 | /* | |
2990 | * Attempt to move tasks. If find_busiest_group has found | |
2991 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 2992 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
2993 | * correctly treated as an imbalance. |
2994 | */ | |
fe2eea3f | 2995 | local_irq_save(flags); |
e17224bf | 2996 | double_rq_lock(this_rq, busiest); |
43010659 | 2997 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 2998 | imbalance, sd, idle, &all_pinned); |
e17224bf | 2999 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3000 | local_irq_restore(flags); |
81026794 | 3001 | |
46cb4b7c SS |
3002 | /* |
3003 | * some other cpu did the load balance for us. | |
3004 | */ | |
43010659 | 3005 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3006 | resched_cpu(this_cpu); |
3007 | ||
81026794 | 3008 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 CL |
3009 | if (unlikely(all_pinned)) { |
3010 | cpu_clear(cpu_of(busiest), cpus); | |
3011 | if (!cpus_empty(cpus)) | |
3012 | goto redo; | |
81026794 | 3013 | goto out_balanced; |
0a2966b4 | 3014 | } |
1da177e4 | 3015 | } |
81026794 | 3016 | |
43010659 | 3017 | if (!ld_moved) { |
1da177e4 LT |
3018 | schedstat_inc(sd, lb_failed[idle]); |
3019 | sd->nr_balance_failed++; | |
3020 | ||
3021 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3022 | |
fe2eea3f | 3023 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3024 | |
3025 | /* don't kick the migration_thread, if the curr | |
3026 | * task on busiest cpu can't be moved to this_cpu | |
3027 | */ | |
3028 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3029 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3030 | all_pinned = 1; |
3031 | goto out_one_pinned; | |
3032 | } | |
3033 | ||
1da177e4 LT |
3034 | if (!busiest->active_balance) { |
3035 | busiest->active_balance = 1; | |
3036 | busiest->push_cpu = this_cpu; | |
81026794 | 3037 | active_balance = 1; |
1da177e4 | 3038 | } |
fe2eea3f | 3039 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3040 | if (active_balance) |
1da177e4 LT |
3041 | wake_up_process(busiest->migration_thread); |
3042 | ||
3043 | /* | |
3044 | * We've kicked active balancing, reset the failure | |
3045 | * counter. | |
3046 | */ | |
39507451 | 3047 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3048 | } |
81026794 | 3049 | } else |
1da177e4 LT |
3050 | sd->nr_balance_failed = 0; |
3051 | ||
81026794 | 3052 | if (likely(!active_balance)) { |
1da177e4 LT |
3053 | /* We were unbalanced, so reset the balancing interval */ |
3054 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
3055 | } else { |
3056 | /* | |
3057 | * If we've begun active balancing, start to back off. This | |
3058 | * case may not be covered by the all_pinned logic if there | |
3059 | * is only 1 task on the busy runqueue (because we don't call | |
3060 | * move_tasks). | |
3061 | */ | |
3062 | if (sd->balance_interval < sd->max_interval) | |
3063 | sd->balance_interval *= 2; | |
1da177e4 LT |
3064 | } |
3065 | ||
43010659 | 3066 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3067 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3068 | return -1; |
43010659 | 3069 | return ld_moved; |
1da177e4 LT |
3070 | |
3071 | out_balanced: | |
1da177e4 LT |
3072 | schedstat_inc(sd, lb_balanced[idle]); |
3073 | ||
16cfb1c0 | 3074 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
3075 | |
3076 | out_one_pinned: | |
1da177e4 | 3077 | /* tune up the balancing interval */ |
77391d71 NP |
3078 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
3079 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
3080 | sd->balance_interval *= 2; |
3081 | ||
48f24c4d | 3082 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3083 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3084 | return -1; |
1da177e4 LT |
3085 | return 0; |
3086 | } | |
3087 | ||
3088 | /* | |
3089 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3090 | * tasks if there is an imbalance. | |
3091 | * | |
d15bcfdb | 3092 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
3093 | * this_rq is locked. |
3094 | */ | |
48f24c4d | 3095 | static int |
70b97a7f | 3096 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
3097 | { |
3098 | struct sched_group *group; | |
70b97a7f | 3099 | struct rq *busiest = NULL; |
1da177e4 | 3100 | unsigned long imbalance; |
43010659 | 3101 | int ld_moved = 0; |
5969fe06 | 3102 | int sd_idle = 0; |
969bb4e4 | 3103 | int all_pinned = 0; |
0a2966b4 | 3104 | cpumask_t cpus = CPU_MASK_ALL; |
5969fe06 | 3105 | |
89c4710e SS |
3106 | /* |
3107 | * When power savings policy is enabled for the parent domain, idle | |
3108 | * sibling can pick up load irrespective of busy siblings. In this case, | |
3109 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 3110 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
3111 | */ |
3112 | if (sd->flags & SD_SHARE_CPUPOWER && | |
3113 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3114 | sd_idle = 1; |
1da177e4 | 3115 | |
2d72376b | 3116 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 3117 | redo: |
d15bcfdb | 3118 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
783609c6 | 3119 | &sd_idle, &cpus, NULL); |
1da177e4 | 3120 | if (!group) { |
d15bcfdb | 3121 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3122 | goto out_balanced; |
1da177e4 LT |
3123 | } |
3124 | ||
d15bcfdb | 3125 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, |
0a2966b4 | 3126 | &cpus); |
db935dbd | 3127 | if (!busiest) { |
d15bcfdb | 3128 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3129 | goto out_balanced; |
1da177e4 LT |
3130 | } |
3131 | ||
db935dbd NP |
3132 | BUG_ON(busiest == this_rq); |
3133 | ||
d15bcfdb | 3134 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 3135 | |
43010659 | 3136 | ld_moved = 0; |
d6d5cfaf NP |
3137 | if (busiest->nr_running > 1) { |
3138 | /* Attempt to move tasks */ | |
3139 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
3140 | /* this_rq->clock is already updated */ |
3141 | update_rq_clock(busiest); | |
43010659 | 3142 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
3143 | imbalance, sd, CPU_NEWLY_IDLE, |
3144 | &all_pinned); | |
d6d5cfaf | 3145 | spin_unlock(&busiest->lock); |
0a2966b4 | 3146 | |
969bb4e4 | 3147 | if (unlikely(all_pinned)) { |
0a2966b4 CL |
3148 | cpu_clear(cpu_of(busiest), cpus); |
3149 | if (!cpus_empty(cpus)) | |
3150 | goto redo; | |
3151 | } | |
d6d5cfaf NP |
3152 | } |
3153 | ||
43010659 | 3154 | if (!ld_moved) { |
d15bcfdb | 3155 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
3156 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
3157 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
3158 | return -1; |
3159 | } else | |
16cfb1c0 | 3160 | sd->nr_balance_failed = 0; |
1da177e4 | 3161 | |
43010659 | 3162 | return ld_moved; |
16cfb1c0 NP |
3163 | |
3164 | out_balanced: | |
d15bcfdb | 3165 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 3166 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3167 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3168 | return -1; |
16cfb1c0 | 3169 | sd->nr_balance_failed = 0; |
48f24c4d | 3170 | |
16cfb1c0 | 3171 | return 0; |
1da177e4 LT |
3172 | } |
3173 | ||
3174 | /* | |
3175 | * idle_balance is called by schedule() if this_cpu is about to become | |
3176 | * idle. Attempts to pull tasks from other CPUs. | |
3177 | */ | |
70b97a7f | 3178 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
3179 | { |
3180 | struct sched_domain *sd; | |
dd41f596 IM |
3181 | int pulled_task = -1; |
3182 | unsigned long next_balance = jiffies + HZ; | |
1da177e4 LT |
3183 | |
3184 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
3185 | unsigned long interval; |
3186 | ||
3187 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
3188 | continue; | |
3189 | ||
3190 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 3191 | /* If we've pulled tasks over stop searching: */ |
1bd77f2d | 3192 | pulled_task = load_balance_newidle(this_cpu, |
92c4ca5c CL |
3193 | this_rq, sd); |
3194 | ||
3195 | interval = msecs_to_jiffies(sd->balance_interval); | |
3196 | if (time_after(next_balance, sd->last_balance + interval)) | |
3197 | next_balance = sd->last_balance + interval; | |
3198 | if (pulled_task) | |
3199 | break; | |
1da177e4 | 3200 | } |
dd41f596 | 3201 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
3202 | /* |
3203 | * We are going idle. next_balance may be set based on | |
3204 | * a busy processor. So reset next_balance. | |
3205 | */ | |
3206 | this_rq->next_balance = next_balance; | |
dd41f596 | 3207 | } |
1da177e4 LT |
3208 | } |
3209 | ||
3210 | /* | |
3211 | * active_load_balance is run by migration threads. It pushes running tasks | |
3212 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
3213 | * running on each physical CPU where possible, and avoids physical / | |
3214 | * logical imbalances. | |
3215 | * | |
3216 | * Called with busiest_rq locked. | |
3217 | */ | |
70b97a7f | 3218 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3219 | { |
39507451 | 3220 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3221 | struct sched_domain *sd; |
3222 | struct rq *target_rq; | |
39507451 | 3223 | |
48f24c4d | 3224 | /* Is there any task to move? */ |
39507451 | 3225 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3226 | return; |
3227 | ||
3228 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3229 | |
3230 | /* | |
39507451 | 3231 | * This condition is "impossible", if it occurs |
41a2d6cf | 3232 | * we need to fix it. Originally reported by |
39507451 | 3233 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3234 | */ |
39507451 | 3235 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3236 | |
39507451 NP |
3237 | /* move a task from busiest_rq to target_rq */ |
3238 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3239 | update_rq_clock(busiest_rq); |
3240 | update_rq_clock(target_rq); | |
39507451 NP |
3241 | |
3242 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3243 | for_each_domain(target_cpu, sd) { |
39507451 | 3244 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 3245 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 3246 | break; |
c96d145e | 3247 | } |
39507451 | 3248 | |
48f24c4d | 3249 | if (likely(sd)) { |
2d72376b | 3250 | schedstat_inc(sd, alb_count); |
39507451 | 3251 | |
43010659 PW |
3252 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3253 | sd, CPU_IDLE)) | |
48f24c4d IM |
3254 | schedstat_inc(sd, alb_pushed); |
3255 | else | |
3256 | schedstat_inc(sd, alb_failed); | |
3257 | } | |
39507451 | 3258 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
3259 | } |
3260 | ||
46cb4b7c SS |
3261 | #ifdef CONFIG_NO_HZ |
3262 | static struct { | |
3263 | atomic_t load_balancer; | |
41a2d6cf | 3264 | cpumask_t cpu_mask; |
46cb4b7c SS |
3265 | } nohz ____cacheline_aligned = { |
3266 | .load_balancer = ATOMIC_INIT(-1), | |
3267 | .cpu_mask = CPU_MASK_NONE, | |
3268 | }; | |
3269 | ||
7835b98b | 3270 | /* |
46cb4b7c SS |
3271 | * This routine will try to nominate the ilb (idle load balancing) |
3272 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3273 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3274 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3275 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3276 | * arrives... | |
3277 | * | |
3278 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3279 | * for idle load balancing. ilb owner will still be part of | |
3280 | * nohz.cpu_mask.. | |
7835b98b | 3281 | * |
46cb4b7c SS |
3282 | * While stopping the tick, this cpu will become the ilb owner if there |
3283 | * is no other owner. And will be the owner till that cpu becomes busy | |
3284 | * or if all cpus in the system stop their ticks at which point | |
3285 | * there is no need for ilb owner. | |
3286 | * | |
3287 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3288 | * next busy scheduler_tick() | |
3289 | */ | |
3290 | int select_nohz_load_balancer(int stop_tick) | |
3291 | { | |
3292 | int cpu = smp_processor_id(); | |
3293 | ||
3294 | if (stop_tick) { | |
3295 | cpu_set(cpu, nohz.cpu_mask); | |
3296 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3297 | ||
3298 | /* | |
3299 | * If we are going offline and still the leader, give up! | |
3300 | */ | |
3301 | if (cpu_is_offline(cpu) && | |
3302 | atomic_read(&nohz.load_balancer) == cpu) { | |
3303 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3304 | BUG(); | |
3305 | return 0; | |
3306 | } | |
3307 | ||
3308 | /* time for ilb owner also to sleep */ | |
3309 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3310 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3311 | atomic_set(&nohz.load_balancer, -1); | |
3312 | return 0; | |
3313 | } | |
3314 | ||
3315 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3316 | /* make me the ilb owner */ | |
3317 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3318 | return 1; | |
3319 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3320 | return 1; | |
3321 | } else { | |
3322 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3323 | return 0; | |
3324 | ||
3325 | cpu_clear(cpu, nohz.cpu_mask); | |
3326 | ||
3327 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3328 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3329 | BUG(); | |
3330 | } | |
3331 | return 0; | |
3332 | } | |
3333 | #endif | |
3334 | ||
3335 | static DEFINE_SPINLOCK(balancing); | |
3336 | ||
3337 | /* | |
7835b98b CL |
3338 | * It checks each scheduling domain to see if it is due to be balanced, |
3339 | * and initiates a balancing operation if so. | |
3340 | * | |
3341 | * Balancing parameters are set up in arch_init_sched_domains. | |
3342 | */ | |
a9957449 | 3343 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3344 | { |
46cb4b7c SS |
3345 | int balance = 1; |
3346 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3347 | unsigned long interval; |
3348 | struct sched_domain *sd; | |
46cb4b7c | 3349 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3350 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3351 | int update_next_balance = 0; |
1da177e4 | 3352 | |
46cb4b7c | 3353 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3354 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3355 | continue; | |
3356 | ||
3357 | interval = sd->balance_interval; | |
d15bcfdb | 3358 | if (idle != CPU_IDLE) |
1da177e4 LT |
3359 | interval *= sd->busy_factor; |
3360 | ||
3361 | /* scale ms to jiffies */ | |
3362 | interval = msecs_to_jiffies(interval); | |
3363 | if (unlikely(!interval)) | |
3364 | interval = 1; | |
dd41f596 IM |
3365 | if (interval > HZ*NR_CPUS/10) |
3366 | interval = HZ*NR_CPUS/10; | |
3367 | ||
1da177e4 | 3368 | |
08c183f3 CL |
3369 | if (sd->flags & SD_SERIALIZE) { |
3370 | if (!spin_trylock(&balancing)) | |
3371 | goto out; | |
3372 | } | |
3373 | ||
c9819f45 | 3374 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
46cb4b7c | 3375 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
3376 | /* |
3377 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3378 | * longer idle, or one of our SMT siblings is |
3379 | * not idle. | |
3380 | */ | |
d15bcfdb | 3381 | idle = CPU_NOT_IDLE; |
1da177e4 | 3382 | } |
1bd77f2d | 3383 | sd->last_balance = jiffies; |
1da177e4 | 3384 | } |
08c183f3 CL |
3385 | if (sd->flags & SD_SERIALIZE) |
3386 | spin_unlock(&balancing); | |
3387 | out: | |
f549da84 | 3388 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3389 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3390 | update_next_balance = 1; |
3391 | } | |
783609c6 SS |
3392 | |
3393 | /* | |
3394 | * Stop the load balance at this level. There is another | |
3395 | * CPU in our sched group which is doing load balancing more | |
3396 | * actively. | |
3397 | */ | |
3398 | if (!balance) | |
3399 | break; | |
1da177e4 | 3400 | } |
f549da84 SS |
3401 | |
3402 | /* | |
3403 | * next_balance will be updated only when there is a need. | |
3404 | * When the cpu is attached to null domain for ex, it will not be | |
3405 | * updated. | |
3406 | */ | |
3407 | if (likely(update_next_balance)) | |
3408 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3409 | } |
3410 | ||
3411 | /* | |
3412 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3413 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3414 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3415 | */ | |
3416 | static void run_rebalance_domains(struct softirq_action *h) | |
3417 | { | |
dd41f596 IM |
3418 | int this_cpu = smp_processor_id(); |
3419 | struct rq *this_rq = cpu_rq(this_cpu); | |
3420 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3421 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3422 | |
dd41f596 | 3423 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3424 | |
3425 | #ifdef CONFIG_NO_HZ | |
3426 | /* | |
3427 | * If this cpu is the owner for idle load balancing, then do the | |
3428 | * balancing on behalf of the other idle cpus whose ticks are | |
3429 | * stopped. | |
3430 | */ | |
dd41f596 IM |
3431 | if (this_rq->idle_at_tick && |
3432 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3433 | cpumask_t cpus = nohz.cpu_mask; |
3434 | struct rq *rq; | |
3435 | int balance_cpu; | |
3436 | ||
dd41f596 | 3437 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3438 | for_each_cpu_mask(balance_cpu, cpus) { |
3439 | /* | |
3440 | * If this cpu gets work to do, stop the load balancing | |
3441 | * work being done for other cpus. Next load | |
3442 | * balancing owner will pick it up. | |
3443 | */ | |
3444 | if (need_resched()) | |
3445 | break; | |
3446 | ||
de0cf899 | 3447 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3448 | |
3449 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3450 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3451 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3452 | } |
3453 | } | |
3454 | #endif | |
3455 | } | |
3456 | ||
3457 | /* | |
3458 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3459 | * | |
3460 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3461 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3462 | * if the whole system is idle. | |
3463 | */ | |
dd41f596 | 3464 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3465 | { |
46cb4b7c SS |
3466 | #ifdef CONFIG_NO_HZ |
3467 | /* | |
3468 | * If we were in the nohz mode recently and busy at the current | |
3469 | * scheduler tick, then check if we need to nominate new idle | |
3470 | * load balancer. | |
3471 | */ | |
3472 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3473 | rq->in_nohz_recently = 0; | |
3474 | ||
3475 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3476 | cpu_clear(cpu, nohz.cpu_mask); | |
3477 | atomic_set(&nohz.load_balancer, -1); | |
3478 | } | |
3479 | ||
3480 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3481 | /* | |
3482 | * simple selection for now: Nominate the | |
3483 | * first cpu in the nohz list to be the next | |
3484 | * ilb owner. | |
3485 | * | |
3486 | * TBD: Traverse the sched domains and nominate | |
3487 | * the nearest cpu in the nohz.cpu_mask. | |
3488 | */ | |
3489 | int ilb = first_cpu(nohz.cpu_mask); | |
3490 | ||
3491 | if (ilb != NR_CPUS) | |
3492 | resched_cpu(ilb); | |
3493 | } | |
3494 | } | |
3495 | ||
3496 | /* | |
3497 | * If this cpu is idle and doing idle load balancing for all the | |
3498 | * cpus with ticks stopped, is it time for that to stop? | |
3499 | */ | |
3500 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3501 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3502 | resched_cpu(cpu); | |
3503 | return; | |
3504 | } | |
3505 | ||
3506 | /* | |
3507 | * If this cpu is idle and the idle load balancing is done by | |
3508 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3509 | */ | |
3510 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3511 | cpu_isset(cpu, nohz.cpu_mask)) | |
3512 | return; | |
3513 | #endif | |
3514 | if (time_after_eq(jiffies, rq->next_balance)) | |
3515 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3516 | } |
dd41f596 IM |
3517 | |
3518 | #else /* CONFIG_SMP */ | |
3519 | ||
1da177e4 LT |
3520 | /* |
3521 | * on UP we do not need to balance between CPUs: | |
3522 | */ | |
70b97a7f | 3523 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3524 | { |
3525 | } | |
dd41f596 | 3526 | |
1da177e4 LT |
3527 | #endif |
3528 | ||
1da177e4 LT |
3529 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3530 | ||
3531 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3532 | ||
3533 | /* | |
41b86e9c IM |
3534 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
3535 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 3536 | */ |
41b86e9c | 3537 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 3538 | { |
1da177e4 | 3539 | unsigned long flags; |
41b86e9c IM |
3540 | u64 ns, delta_exec; |
3541 | struct rq *rq; | |
48f24c4d | 3542 | |
41b86e9c IM |
3543 | rq = task_rq_lock(p, &flags); |
3544 | ns = p->se.sum_exec_runtime; | |
051a1d1a | 3545 | if (task_current(rq, p)) { |
a8e504d2 IM |
3546 | update_rq_clock(rq); |
3547 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
3548 | if ((s64)delta_exec > 0) |
3549 | ns += delta_exec; | |
3550 | } | |
3551 | task_rq_unlock(rq, &flags); | |
48f24c4d | 3552 | |
1da177e4 LT |
3553 | return ns; |
3554 | } | |
3555 | ||
1da177e4 LT |
3556 | /* |
3557 | * Account user cpu time to a process. | |
3558 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
3559 | * @cputime: the cpu time spent in user space since the last update |
3560 | */ | |
3561 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3562 | { | |
3563 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3564 | cputime64_t tmp; | |
3565 | ||
3566 | p->utime = cputime_add(p->utime, cputime); | |
3567 | ||
3568 | /* Add user time to cpustat. */ | |
3569 | tmp = cputime_to_cputime64(cputime); | |
3570 | if (TASK_NICE(p) > 0) | |
3571 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3572 | else | |
3573 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3574 | } | |
3575 | ||
94886b84 LV |
3576 | /* |
3577 | * Account guest cpu time to a process. | |
3578 | * @p: the process that the cpu time gets accounted to | |
3579 | * @cputime: the cpu time spent in virtual machine since the last update | |
3580 | */ | |
f7402e03 | 3581 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
3582 | { |
3583 | cputime64_t tmp; | |
3584 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3585 | ||
3586 | tmp = cputime_to_cputime64(cputime); | |
3587 | ||
3588 | p->utime = cputime_add(p->utime, cputime); | |
3589 | p->gtime = cputime_add(p->gtime, cputime); | |
3590 | ||
3591 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3592 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3593 | } | |
3594 | ||
c66f08be MN |
3595 | /* |
3596 | * Account scaled user cpu time to a process. | |
3597 | * @p: the process that the cpu time gets accounted to | |
3598 | * @cputime: the cpu time spent in user space since the last update | |
3599 | */ | |
3600 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
3601 | { | |
3602 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
3603 | } | |
3604 | ||
1da177e4 LT |
3605 | /* |
3606 | * Account system cpu time to a process. | |
3607 | * @p: the process that the cpu time gets accounted to | |
3608 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3609 | * @cputime: the cpu time spent in kernel space since the last update | |
3610 | */ | |
3611 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3612 | cputime_t cputime) | |
3613 | { | |
3614 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3615 | struct rq *rq = this_rq(); |
1da177e4 LT |
3616 | cputime64_t tmp; |
3617 | ||
9778385d CB |
3618 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) |
3619 | return account_guest_time(p, cputime); | |
94886b84 | 3620 | |
1da177e4 LT |
3621 | p->stime = cputime_add(p->stime, cputime); |
3622 | ||
3623 | /* Add system time to cpustat. */ | |
3624 | tmp = cputime_to_cputime64(cputime); | |
3625 | if (hardirq_count() - hardirq_offset) | |
3626 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3627 | else if (softirq_count()) | |
3628 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 3629 | else if (p != rq->idle) |
1da177e4 | 3630 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 3631 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
3632 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
3633 | else | |
3634 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3635 | /* Account for system time used */ | |
3636 | acct_update_integrals(p); | |
1da177e4 LT |
3637 | } |
3638 | ||
c66f08be MN |
3639 | /* |
3640 | * Account scaled system cpu time to a process. | |
3641 | * @p: the process that the cpu time gets accounted to | |
3642 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3643 | * @cputime: the cpu time spent in kernel space since the last update | |
3644 | */ | |
3645 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
3646 | { | |
3647 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
3648 | } | |
3649 | ||
1da177e4 LT |
3650 | /* |
3651 | * Account for involuntary wait time. | |
3652 | * @p: the process from which the cpu time has been stolen | |
3653 | * @steal: the cpu time spent in involuntary wait | |
3654 | */ | |
3655 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3656 | { | |
3657 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3658 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3659 | struct rq *rq = this_rq(); |
1da177e4 LT |
3660 | |
3661 | if (p == rq->idle) { | |
3662 | p->stime = cputime_add(p->stime, steal); | |
3663 | if (atomic_read(&rq->nr_iowait) > 0) | |
3664 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3665 | else | |
3666 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 3667 | } else |
1da177e4 LT |
3668 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
3669 | } | |
3670 | ||
7835b98b CL |
3671 | /* |
3672 | * This function gets called by the timer code, with HZ frequency. | |
3673 | * We call it with interrupts disabled. | |
3674 | * | |
3675 | * It also gets called by the fork code, when changing the parent's | |
3676 | * timeslices. | |
3677 | */ | |
3678 | void scheduler_tick(void) | |
3679 | { | |
7835b98b CL |
3680 | int cpu = smp_processor_id(); |
3681 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3682 | struct task_struct *curr = rq->curr; |
529c7726 | 3683 | u64 next_tick = rq->tick_timestamp + TICK_NSEC; |
dd41f596 IM |
3684 | |
3685 | spin_lock(&rq->lock); | |
546fe3c9 | 3686 | __update_rq_clock(rq); |
529c7726 IM |
3687 | /* |
3688 | * Let rq->clock advance by at least TICK_NSEC: | |
3689 | */ | |
3690 | if (unlikely(rq->clock < next_tick)) | |
3691 | rq->clock = next_tick; | |
3692 | rq->tick_timestamp = rq->clock; | |
f1a438d8 | 3693 | update_cpu_load(rq); |
fa85ae24 PZ |
3694 | curr->sched_class->task_tick(rq, curr, 0); |
3695 | update_sched_rt_period(rq); | |
dd41f596 | 3696 | spin_unlock(&rq->lock); |
7835b98b | 3697 | |
e418e1c2 | 3698 | #ifdef CONFIG_SMP |
dd41f596 IM |
3699 | rq->idle_at_tick = idle_cpu(cpu); |
3700 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3701 | #endif |
1da177e4 LT |
3702 | } |
3703 | ||
1da177e4 LT |
3704 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3705 | ||
3706 | void fastcall add_preempt_count(int val) | |
3707 | { | |
3708 | /* | |
3709 | * Underflow? | |
3710 | */ | |
9a11b49a IM |
3711 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3712 | return; | |
1da177e4 LT |
3713 | preempt_count() += val; |
3714 | /* | |
3715 | * Spinlock count overflowing soon? | |
3716 | */ | |
33859f7f MOS |
3717 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3718 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3719 | } |
3720 | EXPORT_SYMBOL(add_preempt_count); | |
3721 | ||
3722 | void fastcall sub_preempt_count(int val) | |
3723 | { | |
3724 | /* | |
3725 | * Underflow? | |
3726 | */ | |
9a11b49a IM |
3727 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3728 | return; | |
1da177e4 LT |
3729 | /* |
3730 | * Is the spinlock portion underflowing? | |
3731 | */ | |
9a11b49a IM |
3732 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3733 | !(preempt_count() & PREEMPT_MASK))) | |
3734 | return; | |
3735 | ||
1da177e4 LT |
3736 | preempt_count() -= val; |
3737 | } | |
3738 | EXPORT_SYMBOL(sub_preempt_count); | |
3739 | ||
3740 | #endif | |
3741 | ||
3742 | /* | |
dd41f596 | 3743 | * Print scheduling while atomic bug: |
1da177e4 | 3744 | */ |
dd41f596 | 3745 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3746 | { |
838225b4 SS |
3747 | struct pt_regs *regs = get_irq_regs(); |
3748 | ||
3749 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
3750 | prev->comm, prev->pid, preempt_count()); | |
3751 | ||
dd41f596 IM |
3752 | debug_show_held_locks(prev); |
3753 | if (irqs_disabled()) | |
3754 | print_irqtrace_events(prev); | |
838225b4 SS |
3755 | |
3756 | if (regs) | |
3757 | show_regs(regs); | |
3758 | else | |
3759 | dump_stack(); | |
dd41f596 | 3760 | } |
1da177e4 | 3761 | |
dd41f596 IM |
3762 | /* |
3763 | * Various schedule()-time debugging checks and statistics: | |
3764 | */ | |
3765 | static inline void schedule_debug(struct task_struct *prev) | |
3766 | { | |
1da177e4 | 3767 | /* |
41a2d6cf | 3768 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3769 | * schedule() atomically, we ignore that path for now. |
3770 | * Otherwise, whine if we are scheduling when we should not be. | |
3771 | */ | |
dd41f596 IM |
3772 | if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state)) |
3773 | __schedule_bug(prev); | |
3774 | ||
1da177e4 LT |
3775 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3776 | ||
2d72376b | 3777 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3778 | #ifdef CONFIG_SCHEDSTATS |
3779 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3780 | schedstat_inc(this_rq(), bkl_count); |
3781 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3782 | } |
3783 | #endif | |
dd41f596 IM |
3784 | } |
3785 | ||
3786 | /* | |
3787 | * Pick up the highest-prio task: | |
3788 | */ | |
3789 | static inline struct task_struct * | |
ff95f3df | 3790 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 3791 | { |
5522d5d5 | 3792 | const struct sched_class *class; |
dd41f596 | 3793 | struct task_struct *p; |
1da177e4 LT |
3794 | |
3795 | /* | |
dd41f596 IM |
3796 | * Optimization: we know that if all tasks are in |
3797 | * the fair class we can call that function directly: | |
1da177e4 | 3798 | */ |
dd41f596 | 3799 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3800 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3801 | if (likely(p)) |
3802 | return p; | |
1da177e4 LT |
3803 | } |
3804 | ||
dd41f596 IM |
3805 | class = sched_class_highest; |
3806 | for ( ; ; ) { | |
fb8d4724 | 3807 | p = class->pick_next_task(rq); |
dd41f596 IM |
3808 | if (p) |
3809 | return p; | |
3810 | /* | |
3811 | * Will never be NULL as the idle class always | |
3812 | * returns a non-NULL p: | |
3813 | */ | |
3814 | class = class->next; | |
3815 | } | |
3816 | } | |
1da177e4 | 3817 | |
dd41f596 IM |
3818 | /* |
3819 | * schedule() is the main scheduler function. | |
3820 | */ | |
3821 | asmlinkage void __sched schedule(void) | |
3822 | { | |
3823 | struct task_struct *prev, *next; | |
3824 | long *switch_count; | |
3825 | struct rq *rq; | |
dd41f596 IM |
3826 | int cpu; |
3827 | ||
3828 | need_resched: | |
3829 | preempt_disable(); | |
3830 | cpu = smp_processor_id(); | |
3831 | rq = cpu_rq(cpu); | |
3832 | rcu_qsctr_inc(cpu); | |
3833 | prev = rq->curr; | |
3834 | switch_count = &prev->nivcsw; | |
3835 | ||
3836 | release_kernel_lock(prev); | |
3837 | need_resched_nonpreemptible: | |
3838 | ||
3839 | schedule_debug(prev); | |
1da177e4 | 3840 | |
8f4d37ec PZ |
3841 | hrtick_clear(rq); |
3842 | ||
1e819950 IM |
3843 | /* |
3844 | * Do the rq-clock update outside the rq lock: | |
3845 | */ | |
3846 | local_irq_disable(); | |
c1b3da3e | 3847 | __update_rq_clock(rq); |
1e819950 IM |
3848 | spin_lock(&rq->lock); |
3849 | clear_tsk_need_resched(prev); | |
1da177e4 | 3850 | |
1da177e4 | 3851 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
1da177e4 | 3852 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && |
dd41f596 | 3853 | unlikely(signal_pending(prev)))) { |
1da177e4 | 3854 | prev->state = TASK_RUNNING; |
dd41f596 | 3855 | } else { |
2e1cb74a | 3856 | deactivate_task(rq, prev, 1); |
1da177e4 | 3857 | } |
dd41f596 | 3858 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3859 | } |
3860 | ||
9a897c5a SR |
3861 | #ifdef CONFIG_SMP |
3862 | if (prev->sched_class->pre_schedule) | |
3863 | prev->sched_class->pre_schedule(rq, prev); | |
3864 | #endif | |
f65eda4f | 3865 | |
dd41f596 | 3866 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3867 | idle_balance(cpu, rq); |
1da177e4 | 3868 | |
31ee529c | 3869 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 3870 | next = pick_next_task(rq, prev); |
1da177e4 LT |
3871 | |
3872 | sched_info_switch(prev, next); | |
dd41f596 | 3873 | |
1da177e4 | 3874 | if (likely(prev != next)) { |
1da177e4 LT |
3875 | rq->nr_switches++; |
3876 | rq->curr = next; | |
3877 | ++*switch_count; | |
3878 | ||
dd41f596 | 3879 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
3880 | /* |
3881 | * the context switch might have flipped the stack from under | |
3882 | * us, hence refresh the local variables. | |
3883 | */ | |
3884 | cpu = smp_processor_id(); | |
3885 | rq = cpu_rq(cpu); | |
1da177e4 LT |
3886 | } else |
3887 | spin_unlock_irq(&rq->lock); | |
3888 | ||
8f4d37ec PZ |
3889 | hrtick_set(rq); |
3890 | ||
3891 | if (unlikely(reacquire_kernel_lock(current) < 0)) | |
1da177e4 | 3892 | goto need_resched_nonpreemptible; |
8f4d37ec | 3893 | |
1da177e4 LT |
3894 | preempt_enable_no_resched(); |
3895 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3896 | goto need_resched; | |
3897 | } | |
1da177e4 LT |
3898 | EXPORT_SYMBOL(schedule); |
3899 | ||
3900 | #ifdef CONFIG_PREEMPT | |
3901 | /* | |
2ed6e34f | 3902 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3903 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3904 | * occur there and call schedule directly. |
3905 | */ | |
3906 | asmlinkage void __sched preempt_schedule(void) | |
3907 | { | |
3908 | struct thread_info *ti = current_thread_info(); | |
3909 | #ifdef CONFIG_PREEMPT_BKL | |
3910 | struct task_struct *task = current; | |
3911 | int saved_lock_depth; | |
3912 | #endif | |
3913 | /* | |
3914 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3915 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3916 | */ |
beed33a8 | 3917 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3918 | return; |
3919 | ||
3a5c359a AK |
3920 | do { |
3921 | add_preempt_count(PREEMPT_ACTIVE); | |
3922 | ||
3923 | /* | |
3924 | * We keep the big kernel semaphore locked, but we | |
3925 | * clear ->lock_depth so that schedule() doesnt | |
3926 | * auto-release the semaphore: | |
3927 | */ | |
1da177e4 | 3928 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a AK |
3929 | saved_lock_depth = task->lock_depth; |
3930 | task->lock_depth = -1; | |
1da177e4 | 3931 | #endif |
3a5c359a | 3932 | schedule(); |
1da177e4 | 3933 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a | 3934 | task->lock_depth = saved_lock_depth; |
1da177e4 | 3935 | #endif |
3a5c359a | 3936 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3937 | |
3a5c359a AK |
3938 | /* |
3939 | * Check again in case we missed a preemption opportunity | |
3940 | * between schedule and now. | |
3941 | */ | |
3942 | barrier(); | |
3943 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 3944 | } |
1da177e4 LT |
3945 | EXPORT_SYMBOL(preempt_schedule); |
3946 | ||
3947 | /* | |
2ed6e34f | 3948 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3949 | * off of irq context. |
3950 | * Note, that this is called and return with irqs disabled. This will | |
3951 | * protect us against recursive calling from irq. | |
3952 | */ | |
3953 | asmlinkage void __sched preempt_schedule_irq(void) | |
3954 | { | |
3955 | struct thread_info *ti = current_thread_info(); | |
3956 | #ifdef CONFIG_PREEMPT_BKL | |
3957 | struct task_struct *task = current; | |
3958 | int saved_lock_depth; | |
3959 | #endif | |
2ed6e34f | 3960 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3961 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3962 | ||
3a5c359a AK |
3963 | do { |
3964 | add_preempt_count(PREEMPT_ACTIVE); | |
3965 | ||
3966 | /* | |
3967 | * We keep the big kernel semaphore locked, but we | |
3968 | * clear ->lock_depth so that schedule() doesnt | |
3969 | * auto-release the semaphore: | |
3970 | */ | |
1da177e4 | 3971 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a AK |
3972 | saved_lock_depth = task->lock_depth; |
3973 | task->lock_depth = -1; | |
1da177e4 | 3974 | #endif |
3a5c359a AK |
3975 | local_irq_enable(); |
3976 | schedule(); | |
3977 | local_irq_disable(); | |
1da177e4 | 3978 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a | 3979 | task->lock_depth = saved_lock_depth; |
1da177e4 | 3980 | #endif |
3a5c359a | 3981 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3982 | |
3a5c359a AK |
3983 | /* |
3984 | * Check again in case we missed a preemption opportunity | |
3985 | * between schedule and now. | |
3986 | */ | |
3987 | barrier(); | |
3988 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
3989 | } |
3990 | ||
3991 | #endif /* CONFIG_PREEMPT */ | |
3992 | ||
95cdf3b7 IM |
3993 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
3994 | void *key) | |
1da177e4 | 3995 | { |
48f24c4d | 3996 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 3997 | } |
1da177e4 LT |
3998 | EXPORT_SYMBOL(default_wake_function); |
3999 | ||
4000 | /* | |
41a2d6cf IM |
4001 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4002 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4003 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4004 | * | |
4005 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4006 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4007 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4008 | */ | |
4009 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
4010 | int nr_exclusive, int sync, void *key) | |
4011 | { | |
2e45874c | 4012 | wait_queue_t *curr, *next; |
1da177e4 | 4013 | |
2e45874c | 4014 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4015 | unsigned flags = curr->flags; |
4016 | ||
1da177e4 | 4017 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 4018 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4019 | break; |
4020 | } | |
4021 | } | |
4022 | ||
4023 | /** | |
4024 | * __wake_up - wake up threads blocked on a waitqueue. | |
4025 | * @q: the waitqueue | |
4026 | * @mode: which threads | |
4027 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4028 | * @key: is directly passed to the wakeup function |
1da177e4 LT |
4029 | */ |
4030 | void fastcall __wake_up(wait_queue_head_t *q, unsigned int mode, | |
95cdf3b7 | 4031 | int nr_exclusive, void *key) |
1da177e4 LT |
4032 | { |
4033 | unsigned long flags; | |
4034 | ||
4035 | spin_lock_irqsave(&q->lock, flags); | |
4036 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4037 | spin_unlock_irqrestore(&q->lock, flags); | |
4038 | } | |
1da177e4 LT |
4039 | EXPORT_SYMBOL(__wake_up); |
4040 | ||
4041 | /* | |
4042 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4043 | */ | |
4044 | void fastcall __wake_up_locked(wait_queue_head_t *q, unsigned int mode) | |
4045 | { | |
4046 | __wake_up_common(q, mode, 1, 0, NULL); | |
4047 | } | |
4048 | ||
4049 | /** | |
67be2dd1 | 4050 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4051 | * @q: the waitqueue |
4052 | * @mode: which threads | |
4053 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4054 | * | |
4055 | * The sync wakeup differs that the waker knows that it will schedule | |
4056 | * away soon, so while the target thread will be woken up, it will not | |
4057 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4058 | * with each other. This can prevent needless bouncing between CPUs. | |
4059 | * | |
4060 | * On UP it can prevent extra preemption. | |
4061 | */ | |
95cdf3b7 IM |
4062 | void fastcall |
4063 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
1da177e4 LT |
4064 | { |
4065 | unsigned long flags; | |
4066 | int sync = 1; | |
4067 | ||
4068 | if (unlikely(!q)) | |
4069 | return; | |
4070 | ||
4071 | if (unlikely(!nr_exclusive)) | |
4072 | sync = 0; | |
4073 | ||
4074 | spin_lock_irqsave(&q->lock, flags); | |
4075 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
4076 | spin_unlock_irqrestore(&q->lock, flags); | |
4077 | } | |
4078 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
4079 | ||
b15136e9 | 4080 | void complete(struct completion *x) |
1da177e4 LT |
4081 | { |
4082 | unsigned long flags; | |
4083 | ||
4084 | spin_lock_irqsave(&x->wait.lock, flags); | |
4085 | x->done++; | |
4086 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
4087 | 1, 0, NULL); | |
4088 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
4089 | } | |
4090 | EXPORT_SYMBOL(complete); | |
4091 | ||
b15136e9 | 4092 | void complete_all(struct completion *x) |
1da177e4 LT |
4093 | { |
4094 | unsigned long flags; | |
4095 | ||
4096 | spin_lock_irqsave(&x->wait.lock, flags); | |
4097 | x->done += UINT_MAX/2; | |
4098 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
4099 | 0, 0, NULL); | |
4100 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
4101 | } | |
4102 | EXPORT_SYMBOL(complete_all); | |
4103 | ||
8cbbe86d AK |
4104 | static inline long __sched |
4105 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4106 | { |
1da177e4 LT |
4107 | if (!x->done) { |
4108 | DECLARE_WAITQUEUE(wait, current); | |
4109 | ||
4110 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
4111 | __add_wait_queue_tail(&x->wait, &wait); | |
4112 | do { | |
8cbbe86d AK |
4113 | if (state == TASK_INTERRUPTIBLE && |
4114 | signal_pending(current)) { | |
4115 | __remove_wait_queue(&x->wait, &wait); | |
4116 | return -ERESTARTSYS; | |
4117 | } | |
4118 | __set_current_state(state); | |
1da177e4 LT |
4119 | spin_unlock_irq(&x->wait.lock); |
4120 | timeout = schedule_timeout(timeout); | |
4121 | spin_lock_irq(&x->wait.lock); | |
4122 | if (!timeout) { | |
4123 | __remove_wait_queue(&x->wait, &wait); | |
8cbbe86d | 4124 | return timeout; |
1da177e4 LT |
4125 | } |
4126 | } while (!x->done); | |
4127 | __remove_wait_queue(&x->wait, &wait); | |
4128 | } | |
4129 | x->done--; | |
1da177e4 LT |
4130 | return timeout; |
4131 | } | |
1da177e4 | 4132 | |
8cbbe86d AK |
4133 | static long __sched |
4134 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4135 | { |
1da177e4 LT |
4136 | might_sleep(); |
4137 | ||
4138 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4139 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4140 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4141 | return timeout; |
4142 | } | |
1da177e4 | 4143 | |
b15136e9 | 4144 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4145 | { |
4146 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4147 | } |
8cbbe86d | 4148 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4149 | |
b15136e9 | 4150 | unsigned long __sched |
8cbbe86d | 4151 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4152 | { |
8cbbe86d | 4153 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4154 | } |
8cbbe86d | 4155 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4156 | |
8cbbe86d | 4157 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4158 | { |
51e97990 AK |
4159 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4160 | if (t == -ERESTARTSYS) | |
4161 | return t; | |
4162 | return 0; | |
0fec171c | 4163 | } |
8cbbe86d | 4164 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4165 | |
b15136e9 | 4166 | unsigned long __sched |
8cbbe86d AK |
4167 | wait_for_completion_interruptible_timeout(struct completion *x, |
4168 | unsigned long timeout) | |
0fec171c | 4169 | { |
8cbbe86d | 4170 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4171 | } |
8cbbe86d | 4172 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4173 | |
8cbbe86d AK |
4174 | static long __sched |
4175 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4176 | { |
0fec171c IM |
4177 | unsigned long flags; |
4178 | wait_queue_t wait; | |
4179 | ||
4180 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4181 | |
8cbbe86d | 4182 | __set_current_state(state); |
1da177e4 | 4183 | |
8cbbe86d AK |
4184 | spin_lock_irqsave(&q->lock, flags); |
4185 | __add_wait_queue(q, &wait); | |
4186 | spin_unlock(&q->lock); | |
4187 | timeout = schedule_timeout(timeout); | |
4188 | spin_lock_irq(&q->lock); | |
4189 | __remove_wait_queue(q, &wait); | |
4190 | spin_unlock_irqrestore(&q->lock, flags); | |
4191 | ||
4192 | return timeout; | |
4193 | } | |
4194 | ||
4195 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4196 | { | |
4197 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4198 | } |
1da177e4 LT |
4199 | EXPORT_SYMBOL(interruptible_sleep_on); |
4200 | ||
0fec171c | 4201 | long __sched |
95cdf3b7 | 4202 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4203 | { |
8cbbe86d | 4204 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4205 | } |
1da177e4 LT |
4206 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4207 | ||
0fec171c | 4208 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4209 | { |
8cbbe86d | 4210 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4211 | } |
1da177e4 LT |
4212 | EXPORT_SYMBOL(sleep_on); |
4213 | ||
0fec171c | 4214 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4215 | { |
8cbbe86d | 4216 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4217 | } |
1da177e4 LT |
4218 | EXPORT_SYMBOL(sleep_on_timeout); |
4219 | ||
b29739f9 IM |
4220 | #ifdef CONFIG_RT_MUTEXES |
4221 | ||
4222 | /* | |
4223 | * rt_mutex_setprio - set the current priority of a task | |
4224 | * @p: task | |
4225 | * @prio: prio value (kernel-internal form) | |
4226 | * | |
4227 | * This function changes the 'effective' priority of a task. It does | |
4228 | * not touch ->normal_prio like __setscheduler(). | |
4229 | * | |
4230 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4231 | */ | |
36c8b586 | 4232 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4233 | { |
4234 | unsigned long flags; | |
83b699ed | 4235 | int oldprio, on_rq, running; |
70b97a7f | 4236 | struct rq *rq; |
cb469845 | 4237 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
4238 | |
4239 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4240 | ||
4241 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4242 | update_rq_clock(rq); |
b29739f9 | 4243 | |
d5f9f942 | 4244 | oldprio = p->prio; |
dd41f596 | 4245 | on_rq = p->se.on_rq; |
051a1d1a | 4246 | running = task_current(rq, p); |
83b699ed | 4247 | if (on_rq) { |
69be72c1 | 4248 | dequeue_task(rq, p, 0); |
83b699ed SV |
4249 | if (running) |
4250 | p->sched_class->put_prev_task(rq, p); | |
4251 | } | |
dd41f596 IM |
4252 | |
4253 | if (rt_prio(prio)) | |
4254 | p->sched_class = &rt_sched_class; | |
4255 | else | |
4256 | p->sched_class = &fair_sched_class; | |
4257 | ||
b29739f9 IM |
4258 | p->prio = prio; |
4259 | ||
dd41f596 | 4260 | if (on_rq) { |
83b699ed SV |
4261 | if (running) |
4262 | p->sched_class->set_curr_task(rq); | |
cb469845 | 4263 | |
8159f87e | 4264 | enqueue_task(rq, p, 0); |
cb469845 SR |
4265 | |
4266 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4267 | } |
4268 | task_rq_unlock(rq, &flags); | |
4269 | } | |
4270 | ||
4271 | #endif | |
4272 | ||
36c8b586 | 4273 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4274 | { |
dd41f596 | 4275 | int old_prio, delta, on_rq; |
1da177e4 | 4276 | unsigned long flags; |
70b97a7f | 4277 | struct rq *rq; |
1da177e4 LT |
4278 | |
4279 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4280 | return; | |
4281 | /* | |
4282 | * We have to be careful, if called from sys_setpriority(), | |
4283 | * the task might be in the middle of scheduling on another CPU. | |
4284 | */ | |
4285 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4286 | update_rq_clock(rq); |
1da177e4 LT |
4287 | /* |
4288 | * The RT priorities are set via sched_setscheduler(), but we still | |
4289 | * allow the 'normal' nice value to be set - but as expected | |
4290 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4291 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4292 | */ |
e05606d3 | 4293 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4294 | p->static_prio = NICE_TO_PRIO(nice); |
4295 | goto out_unlock; | |
4296 | } | |
dd41f596 | 4297 | on_rq = p->se.on_rq; |
58e2d4ca | 4298 | if (on_rq) |
69be72c1 | 4299 | dequeue_task(rq, p, 0); |
1da177e4 | 4300 | |
1da177e4 | 4301 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4302 | set_load_weight(p); |
b29739f9 IM |
4303 | old_prio = p->prio; |
4304 | p->prio = effective_prio(p); | |
4305 | delta = p->prio - old_prio; | |
1da177e4 | 4306 | |
dd41f596 | 4307 | if (on_rq) { |
8159f87e | 4308 | enqueue_task(rq, p, 0); |
1da177e4 | 4309 | /* |
d5f9f942 AM |
4310 | * If the task increased its priority or is running and |
4311 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4312 | */ |
d5f9f942 | 4313 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4314 | resched_task(rq->curr); |
4315 | } | |
4316 | out_unlock: | |
4317 | task_rq_unlock(rq, &flags); | |
4318 | } | |
1da177e4 LT |
4319 | EXPORT_SYMBOL(set_user_nice); |
4320 | ||
e43379f1 MM |
4321 | /* |
4322 | * can_nice - check if a task can reduce its nice value | |
4323 | * @p: task | |
4324 | * @nice: nice value | |
4325 | */ | |
36c8b586 | 4326 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4327 | { |
024f4747 MM |
4328 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4329 | int nice_rlim = 20 - nice; | |
48f24c4d | 4330 | |
e43379f1 MM |
4331 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4332 | capable(CAP_SYS_NICE)); | |
4333 | } | |
4334 | ||
1da177e4 LT |
4335 | #ifdef __ARCH_WANT_SYS_NICE |
4336 | ||
4337 | /* | |
4338 | * sys_nice - change the priority of the current process. | |
4339 | * @increment: priority increment | |
4340 | * | |
4341 | * sys_setpriority is a more generic, but much slower function that | |
4342 | * does similar things. | |
4343 | */ | |
4344 | asmlinkage long sys_nice(int increment) | |
4345 | { | |
48f24c4d | 4346 | long nice, retval; |
1da177e4 LT |
4347 | |
4348 | /* | |
4349 | * Setpriority might change our priority at the same moment. | |
4350 | * We don't have to worry. Conceptually one call occurs first | |
4351 | * and we have a single winner. | |
4352 | */ | |
e43379f1 MM |
4353 | if (increment < -40) |
4354 | increment = -40; | |
1da177e4 LT |
4355 | if (increment > 40) |
4356 | increment = 40; | |
4357 | ||
4358 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4359 | if (nice < -20) | |
4360 | nice = -20; | |
4361 | if (nice > 19) | |
4362 | nice = 19; | |
4363 | ||
e43379f1 MM |
4364 | if (increment < 0 && !can_nice(current, nice)) |
4365 | return -EPERM; | |
4366 | ||
1da177e4 LT |
4367 | retval = security_task_setnice(current, nice); |
4368 | if (retval) | |
4369 | return retval; | |
4370 | ||
4371 | set_user_nice(current, nice); | |
4372 | return 0; | |
4373 | } | |
4374 | ||
4375 | #endif | |
4376 | ||
4377 | /** | |
4378 | * task_prio - return the priority value of a given task. | |
4379 | * @p: the task in question. | |
4380 | * | |
4381 | * This is the priority value as seen by users in /proc. | |
4382 | * RT tasks are offset by -200. Normal tasks are centered | |
4383 | * around 0, value goes from -16 to +15. | |
4384 | */ | |
36c8b586 | 4385 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4386 | { |
4387 | return p->prio - MAX_RT_PRIO; | |
4388 | } | |
4389 | ||
4390 | /** | |
4391 | * task_nice - return the nice value of a given task. | |
4392 | * @p: the task in question. | |
4393 | */ | |
36c8b586 | 4394 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4395 | { |
4396 | return TASK_NICE(p); | |
4397 | } | |
1da177e4 | 4398 | EXPORT_SYMBOL_GPL(task_nice); |
1da177e4 LT |
4399 | |
4400 | /** | |
4401 | * idle_cpu - is a given cpu idle currently? | |
4402 | * @cpu: the processor in question. | |
4403 | */ | |
4404 | int idle_cpu(int cpu) | |
4405 | { | |
4406 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4407 | } | |
4408 | ||
1da177e4 LT |
4409 | /** |
4410 | * idle_task - return the idle task for a given cpu. | |
4411 | * @cpu: the processor in question. | |
4412 | */ | |
36c8b586 | 4413 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4414 | { |
4415 | return cpu_rq(cpu)->idle; | |
4416 | } | |
4417 | ||
4418 | /** | |
4419 | * find_process_by_pid - find a process with a matching PID value. | |
4420 | * @pid: the pid in question. | |
4421 | */ | |
a9957449 | 4422 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4423 | { |
228ebcbe | 4424 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4425 | } |
4426 | ||
4427 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4428 | static void |
4429 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4430 | { |
dd41f596 | 4431 | BUG_ON(p->se.on_rq); |
48f24c4d | 4432 | |
1da177e4 | 4433 | p->policy = policy; |
dd41f596 IM |
4434 | switch (p->policy) { |
4435 | case SCHED_NORMAL: | |
4436 | case SCHED_BATCH: | |
4437 | case SCHED_IDLE: | |
4438 | p->sched_class = &fair_sched_class; | |
4439 | break; | |
4440 | case SCHED_FIFO: | |
4441 | case SCHED_RR: | |
4442 | p->sched_class = &rt_sched_class; | |
4443 | break; | |
4444 | } | |
4445 | ||
1da177e4 | 4446 | p->rt_priority = prio; |
b29739f9 IM |
4447 | p->normal_prio = normal_prio(p); |
4448 | /* we are holding p->pi_lock already */ | |
4449 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4450 | set_load_weight(p); |
1da177e4 LT |
4451 | } |
4452 | ||
4453 | /** | |
72fd4a35 | 4454 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4455 | * @p: the task in question. |
4456 | * @policy: new policy. | |
4457 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4458 | * |
72fd4a35 | 4459 | * NOTE that the task may be already dead. |
1da177e4 | 4460 | */ |
95cdf3b7 IM |
4461 | int sched_setscheduler(struct task_struct *p, int policy, |
4462 | struct sched_param *param) | |
1da177e4 | 4463 | { |
83b699ed | 4464 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4465 | unsigned long flags; |
cb469845 | 4466 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 4467 | struct rq *rq; |
1da177e4 | 4468 | |
66e5393a SR |
4469 | /* may grab non-irq protected spin_locks */ |
4470 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4471 | recheck: |
4472 | /* double check policy once rq lock held */ | |
4473 | if (policy < 0) | |
4474 | policy = oldpolicy = p->policy; | |
4475 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4476 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4477 | policy != SCHED_IDLE) | |
b0a9499c | 4478 | return -EINVAL; |
1da177e4 LT |
4479 | /* |
4480 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4481 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4482 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4483 | */ |
4484 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4485 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4486 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4487 | return -EINVAL; |
e05606d3 | 4488 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4489 | return -EINVAL; |
4490 | ||
37e4ab3f OC |
4491 | /* |
4492 | * Allow unprivileged RT tasks to decrease priority: | |
4493 | */ | |
4494 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4495 | if (rt_policy(policy)) { |
8dc3e909 | 4496 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4497 | |
4498 | if (!lock_task_sighand(p, &flags)) | |
4499 | return -ESRCH; | |
4500 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4501 | unlock_task_sighand(p, &flags); | |
4502 | ||
4503 | /* can't set/change the rt policy */ | |
4504 | if (policy != p->policy && !rlim_rtprio) | |
4505 | return -EPERM; | |
4506 | ||
4507 | /* can't increase priority */ | |
4508 | if (param->sched_priority > p->rt_priority && | |
4509 | param->sched_priority > rlim_rtprio) | |
4510 | return -EPERM; | |
4511 | } | |
dd41f596 IM |
4512 | /* |
4513 | * Like positive nice levels, dont allow tasks to | |
4514 | * move out of SCHED_IDLE either: | |
4515 | */ | |
4516 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4517 | return -EPERM; | |
5fe1d75f | 4518 | |
37e4ab3f OC |
4519 | /* can't change other user's priorities */ |
4520 | if ((current->euid != p->euid) && | |
4521 | (current->euid != p->uid)) | |
4522 | return -EPERM; | |
4523 | } | |
1da177e4 LT |
4524 | |
4525 | retval = security_task_setscheduler(p, policy, param); | |
4526 | if (retval) | |
4527 | return retval; | |
b29739f9 IM |
4528 | /* |
4529 | * make sure no PI-waiters arrive (or leave) while we are | |
4530 | * changing the priority of the task: | |
4531 | */ | |
4532 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4533 | /* |
4534 | * To be able to change p->policy safely, the apropriate | |
4535 | * runqueue lock must be held. | |
4536 | */ | |
b29739f9 | 4537 | rq = __task_rq_lock(p); |
1da177e4 LT |
4538 | /* recheck policy now with rq lock held */ |
4539 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4540 | policy = oldpolicy = -1; | |
b29739f9 IM |
4541 | __task_rq_unlock(rq); |
4542 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4543 | goto recheck; |
4544 | } | |
2daa3577 | 4545 | update_rq_clock(rq); |
dd41f596 | 4546 | on_rq = p->se.on_rq; |
051a1d1a | 4547 | running = task_current(rq, p); |
83b699ed | 4548 | if (on_rq) { |
2e1cb74a | 4549 | deactivate_task(rq, p, 0); |
83b699ed SV |
4550 | if (running) |
4551 | p->sched_class->put_prev_task(rq, p); | |
4552 | } | |
f6b53205 | 4553 | |
1da177e4 | 4554 | oldprio = p->prio; |
dd41f596 | 4555 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4556 | |
dd41f596 | 4557 | if (on_rq) { |
83b699ed SV |
4558 | if (running) |
4559 | p->sched_class->set_curr_task(rq); | |
cb469845 | 4560 | |
dd41f596 | 4561 | activate_task(rq, p, 0); |
cb469845 SR |
4562 | |
4563 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 4564 | } |
b29739f9 IM |
4565 | __task_rq_unlock(rq); |
4566 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4567 | ||
95e02ca9 TG |
4568 | rt_mutex_adjust_pi(p); |
4569 | ||
1da177e4 LT |
4570 | return 0; |
4571 | } | |
4572 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4573 | ||
95cdf3b7 IM |
4574 | static int |
4575 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4576 | { |
1da177e4 LT |
4577 | struct sched_param lparam; |
4578 | struct task_struct *p; | |
36c8b586 | 4579 | int retval; |
1da177e4 LT |
4580 | |
4581 | if (!param || pid < 0) | |
4582 | return -EINVAL; | |
4583 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4584 | return -EFAULT; | |
5fe1d75f ON |
4585 | |
4586 | rcu_read_lock(); | |
4587 | retval = -ESRCH; | |
1da177e4 | 4588 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4589 | if (p != NULL) |
4590 | retval = sched_setscheduler(p, policy, &lparam); | |
4591 | rcu_read_unlock(); | |
36c8b586 | 4592 | |
1da177e4 LT |
4593 | return retval; |
4594 | } | |
4595 | ||
4596 | /** | |
4597 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4598 | * @pid: the pid in question. | |
4599 | * @policy: new policy. | |
4600 | * @param: structure containing the new RT priority. | |
4601 | */ | |
41a2d6cf IM |
4602 | asmlinkage long |
4603 | sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4604 | { |
c21761f1 JB |
4605 | /* negative values for policy are not valid */ |
4606 | if (policy < 0) | |
4607 | return -EINVAL; | |
4608 | ||
1da177e4 LT |
4609 | return do_sched_setscheduler(pid, policy, param); |
4610 | } | |
4611 | ||
4612 | /** | |
4613 | * sys_sched_setparam - set/change the RT priority of a thread | |
4614 | * @pid: the pid in question. | |
4615 | * @param: structure containing the new RT priority. | |
4616 | */ | |
4617 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4618 | { | |
4619 | return do_sched_setscheduler(pid, -1, param); | |
4620 | } | |
4621 | ||
4622 | /** | |
4623 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4624 | * @pid: the pid in question. | |
4625 | */ | |
4626 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4627 | { | |
36c8b586 | 4628 | struct task_struct *p; |
3a5c359a | 4629 | int retval; |
1da177e4 LT |
4630 | |
4631 | if (pid < 0) | |
3a5c359a | 4632 | return -EINVAL; |
1da177e4 LT |
4633 | |
4634 | retval = -ESRCH; | |
4635 | read_lock(&tasklist_lock); | |
4636 | p = find_process_by_pid(pid); | |
4637 | if (p) { | |
4638 | retval = security_task_getscheduler(p); | |
4639 | if (!retval) | |
4640 | retval = p->policy; | |
4641 | } | |
4642 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
4643 | return retval; |
4644 | } | |
4645 | ||
4646 | /** | |
4647 | * sys_sched_getscheduler - get the RT priority of a thread | |
4648 | * @pid: the pid in question. | |
4649 | * @param: structure containing the RT priority. | |
4650 | */ | |
4651 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
4652 | { | |
4653 | struct sched_param lp; | |
36c8b586 | 4654 | struct task_struct *p; |
3a5c359a | 4655 | int retval; |
1da177e4 LT |
4656 | |
4657 | if (!param || pid < 0) | |
3a5c359a | 4658 | return -EINVAL; |
1da177e4 LT |
4659 | |
4660 | read_lock(&tasklist_lock); | |
4661 | p = find_process_by_pid(pid); | |
4662 | retval = -ESRCH; | |
4663 | if (!p) | |
4664 | goto out_unlock; | |
4665 | ||
4666 | retval = security_task_getscheduler(p); | |
4667 | if (retval) | |
4668 | goto out_unlock; | |
4669 | ||
4670 | lp.sched_priority = p->rt_priority; | |
4671 | read_unlock(&tasklist_lock); | |
4672 | ||
4673 | /* | |
4674 | * This one might sleep, we cannot do it with a spinlock held ... | |
4675 | */ | |
4676 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4677 | ||
1da177e4 LT |
4678 | return retval; |
4679 | ||
4680 | out_unlock: | |
4681 | read_unlock(&tasklist_lock); | |
4682 | return retval; | |
4683 | } | |
4684 | ||
4685 | long sched_setaffinity(pid_t pid, cpumask_t new_mask) | |
4686 | { | |
1da177e4 | 4687 | cpumask_t cpus_allowed; |
36c8b586 IM |
4688 | struct task_struct *p; |
4689 | int retval; | |
1da177e4 | 4690 | |
95402b38 | 4691 | get_online_cpus(); |
1da177e4 LT |
4692 | read_lock(&tasklist_lock); |
4693 | ||
4694 | p = find_process_by_pid(pid); | |
4695 | if (!p) { | |
4696 | read_unlock(&tasklist_lock); | |
95402b38 | 4697 | put_online_cpus(); |
1da177e4 LT |
4698 | return -ESRCH; |
4699 | } | |
4700 | ||
4701 | /* | |
4702 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 4703 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
4704 | * usage count and then drop tasklist_lock. |
4705 | */ | |
4706 | get_task_struct(p); | |
4707 | read_unlock(&tasklist_lock); | |
4708 | ||
4709 | retval = -EPERM; | |
4710 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
4711 | !capable(CAP_SYS_NICE)) | |
4712 | goto out_unlock; | |
4713 | ||
e7834f8f DQ |
4714 | retval = security_task_setscheduler(p, 0, NULL); |
4715 | if (retval) | |
4716 | goto out_unlock; | |
4717 | ||
1da177e4 LT |
4718 | cpus_allowed = cpuset_cpus_allowed(p); |
4719 | cpus_and(new_mask, new_mask, cpus_allowed); | |
8707d8b8 | 4720 | again: |
1da177e4 LT |
4721 | retval = set_cpus_allowed(p, new_mask); |
4722 | ||
8707d8b8 PM |
4723 | if (!retval) { |
4724 | cpus_allowed = cpuset_cpus_allowed(p); | |
4725 | if (!cpus_subset(new_mask, cpus_allowed)) { | |
4726 | /* | |
4727 | * We must have raced with a concurrent cpuset | |
4728 | * update. Just reset the cpus_allowed to the | |
4729 | * cpuset's cpus_allowed | |
4730 | */ | |
4731 | new_mask = cpus_allowed; | |
4732 | goto again; | |
4733 | } | |
4734 | } | |
1da177e4 LT |
4735 | out_unlock: |
4736 | put_task_struct(p); | |
95402b38 | 4737 | put_online_cpus(); |
1da177e4 LT |
4738 | return retval; |
4739 | } | |
4740 | ||
4741 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
4742 | cpumask_t *new_mask) | |
4743 | { | |
4744 | if (len < sizeof(cpumask_t)) { | |
4745 | memset(new_mask, 0, sizeof(cpumask_t)); | |
4746 | } else if (len > sizeof(cpumask_t)) { | |
4747 | len = sizeof(cpumask_t); | |
4748 | } | |
4749 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
4750 | } | |
4751 | ||
4752 | /** | |
4753 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4754 | * @pid: pid of the process | |
4755 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4756 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4757 | */ | |
4758 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
4759 | unsigned long __user *user_mask_ptr) | |
4760 | { | |
4761 | cpumask_t new_mask; | |
4762 | int retval; | |
4763 | ||
4764 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
4765 | if (retval) | |
4766 | return retval; | |
4767 | ||
4768 | return sched_setaffinity(pid, new_mask); | |
4769 | } | |
4770 | ||
4771 | /* | |
4772 | * Represents all cpu's present in the system | |
4773 | * In systems capable of hotplug, this map could dynamically grow | |
4774 | * as new cpu's are detected in the system via any platform specific | |
4775 | * method, such as ACPI for e.g. | |
4776 | */ | |
4777 | ||
4cef0c61 | 4778 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
4779 | EXPORT_SYMBOL(cpu_present_map); |
4780 | ||
4781 | #ifndef CONFIG_SMP | |
4cef0c61 | 4782 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
4783 | EXPORT_SYMBOL(cpu_online_map); |
4784 | ||
4cef0c61 | 4785 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 4786 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
4787 | #endif |
4788 | ||
4789 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
4790 | { | |
36c8b586 | 4791 | struct task_struct *p; |
1da177e4 | 4792 | int retval; |
1da177e4 | 4793 | |
95402b38 | 4794 | get_online_cpus(); |
1da177e4 LT |
4795 | read_lock(&tasklist_lock); |
4796 | ||
4797 | retval = -ESRCH; | |
4798 | p = find_process_by_pid(pid); | |
4799 | if (!p) | |
4800 | goto out_unlock; | |
4801 | ||
e7834f8f DQ |
4802 | retval = security_task_getscheduler(p); |
4803 | if (retval) | |
4804 | goto out_unlock; | |
4805 | ||
2f7016d9 | 4806 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
4807 | |
4808 | out_unlock: | |
4809 | read_unlock(&tasklist_lock); | |
95402b38 | 4810 | put_online_cpus(); |
1da177e4 | 4811 | |
9531b62f | 4812 | return retval; |
1da177e4 LT |
4813 | } |
4814 | ||
4815 | /** | |
4816 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4817 | * @pid: pid of the process | |
4818 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4819 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4820 | */ | |
4821 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
4822 | unsigned long __user *user_mask_ptr) | |
4823 | { | |
4824 | int ret; | |
4825 | cpumask_t mask; | |
4826 | ||
4827 | if (len < sizeof(cpumask_t)) | |
4828 | return -EINVAL; | |
4829 | ||
4830 | ret = sched_getaffinity(pid, &mask); | |
4831 | if (ret < 0) | |
4832 | return ret; | |
4833 | ||
4834 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
4835 | return -EFAULT; | |
4836 | ||
4837 | return sizeof(cpumask_t); | |
4838 | } | |
4839 | ||
4840 | /** | |
4841 | * sys_sched_yield - yield the current processor to other threads. | |
4842 | * | |
dd41f596 IM |
4843 | * This function yields the current CPU to other tasks. If there are no |
4844 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
4845 | */ |
4846 | asmlinkage long sys_sched_yield(void) | |
4847 | { | |
70b97a7f | 4848 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4849 | |
2d72376b | 4850 | schedstat_inc(rq, yld_count); |
4530d7ab | 4851 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4852 | |
4853 | /* | |
4854 | * Since we are going to call schedule() anyway, there's | |
4855 | * no need to preempt or enable interrupts: | |
4856 | */ | |
4857 | __release(rq->lock); | |
8a25d5de | 4858 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4859 | _raw_spin_unlock(&rq->lock); |
4860 | preempt_enable_no_resched(); | |
4861 | ||
4862 | schedule(); | |
4863 | ||
4864 | return 0; | |
4865 | } | |
4866 | ||
e7b38404 | 4867 | static void __cond_resched(void) |
1da177e4 | 4868 | { |
8e0a43d8 IM |
4869 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
4870 | __might_sleep(__FILE__, __LINE__); | |
4871 | #endif | |
5bbcfd90 IM |
4872 | /* |
4873 | * The BKS might be reacquired before we have dropped | |
4874 | * PREEMPT_ACTIVE, which could trigger a second | |
4875 | * cond_resched() call. | |
4876 | */ | |
1da177e4 LT |
4877 | do { |
4878 | add_preempt_count(PREEMPT_ACTIVE); | |
4879 | schedule(); | |
4880 | sub_preempt_count(PREEMPT_ACTIVE); | |
4881 | } while (need_resched()); | |
4882 | } | |
4883 | ||
02b67cc3 HX |
4884 | #if !defined(CONFIG_PREEMPT) || defined(CONFIG_PREEMPT_VOLUNTARY) |
4885 | int __sched _cond_resched(void) | |
1da177e4 | 4886 | { |
9414232f IM |
4887 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
4888 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
4889 | __cond_resched(); |
4890 | return 1; | |
4891 | } | |
4892 | return 0; | |
4893 | } | |
02b67cc3 HX |
4894 | EXPORT_SYMBOL(_cond_resched); |
4895 | #endif | |
1da177e4 LT |
4896 | |
4897 | /* | |
4898 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
4899 | * call schedule, and on return reacquire the lock. | |
4900 | * | |
41a2d6cf | 4901 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4902 | * operations here to prevent schedule() from being called twice (once via |
4903 | * spin_unlock(), once by hand). | |
4904 | */ | |
95cdf3b7 | 4905 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4906 | { |
6df3cecb JK |
4907 | int ret = 0; |
4908 | ||
1da177e4 LT |
4909 | if (need_lockbreak(lock)) { |
4910 | spin_unlock(lock); | |
4911 | cpu_relax(); | |
6df3cecb | 4912 | ret = 1; |
1da177e4 LT |
4913 | spin_lock(lock); |
4914 | } | |
9414232f | 4915 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
8a25d5de | 4916 | spin_release(&lock->dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4917 | _raw_spin_unlock(lock); |
4918 | preempt_enable_no_resched(); | |
4919 | __cond_resched(); | |
6df3cecb | 4920 | ret = 1; |
1da177e4 | 4921 | spin_lock(lock); |
1da177e4 | 4922 | } |
6df3cecb | 4923 | return ret; |
1da177e4 | 4924 | } |
1da177e4 LT |
4925 | EXPORT_SYMBOL(cond_resched_lock); |
4926 | ||
4927 | int __sched cond_resched_softirq(void) | |
4928 | { | |
4929 | BUG_ON(!in_softirq()); | |
4930 | ||
9414232f | 4931 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 4932 | local_bh_enable(); |
1da177e4 LT |
4933 | __cond_resched(); |
4934 | local_bh_disable(); | |
4935 | return 1; | |
4936 | } | |
4937 | return 0; | |
4938 | } | |
1da177e4 LT |
4939 | EXPORT_SYMBOL(cond_resched_softirq); |
4940 | ||
1da177e4 LT |
4941 | /** |
4942 | * yield - yield the current processor to other threads. | |
4943 | * | |
72fd4a35 | 4944 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4945 | * thread runnable and calls sys_sched_yield(). |
4946 | */ | |
4947 | void __sched yield(void) | |
4948 | { | |
4949 | set_current_state(TASK_RUNNING); | |
4950 | sys_sched_yield(); | |
4951 | } | |
1da177e4 LT |
4952 | EXPORT_SYMBOL(yield); |
4953 | ||
4954 | /* | |
41a2d6cf | 4955 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
4956 | * that process accounting knows that this is a task in IO wait state. |
4957 | * | |
4958 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
4959 | * has set its backing_dev_info: the queue against which it should throttle) | |
4960 | */ | |
4961 | void __sched io_schedule(void) | |
4962 | { | |
70b97a7f | 4963 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 4964 | |
0ff92245 | 4965 | delayacct_blkio_start(); |
1da177e4 LT |
4966 | atomic_inc(&rq->nr_iowait); |
4967 | schedule(); | |
4968 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4969 | delayacct_blkio_end(); |
1da177e4 | 4970 | } |
1da177e4 LT |
4971 | EXPORT_SYMBOL(io_schedule); |
4972 | ||
4973 | long __sched io_schedule_timeout(long timeout) | |
4974 | { | |
70b97a7f | 4975 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
4976 | long ret; |
4977 | ||
0ff92245 | 4978 | delayacct_blkio_start(); |
1da177e4 LT |
4979 | atomic_inc(&rq->nr_iowait); |
4980 | ret = schedule_timeout(timeout); | |
4981 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4982 | delayacct_blkio_end(); |
1da177e4 LT |
4983 | return ret; |
4984 | } | |
4985 | ||
4986 | /** | |
4987 | * sys_sched_get_priority_max - return maximum RT priority. | |
4988 | * @policy: scheduling class. | |
4989 | * | |
4990 | * this syscall returns the maximum rt_priority that can be used | |
4991 | * by a given scheduling class. | |
4992 | */ | |
4993 | asmlinkage long sys_sched_get_priority_max(int policy) | |
4994 | { | |
4995 | int ret = -EINVAL; | |
4996 | ||
4997 | switch (policy) { | |
4998 | case SCHED_FIFO: | |
4999 | case SCHED_RR: | |
5000 | ret = MAX_USER_RT_PRIO-1; | |
5001 | break; | |
5002 | case SCHED_NORMAL: | |
b0a9499c | 5003 | case SCHED_BATCH: |
dd41f596 | 5004 | case SCHED_IDLE: |
1da177e4 LT |
5005 | ret = 0; |
5006 | break; | |
5007 | } | |
5008 | return ret; | |
5009 | } | |
5010 | ||
5011 | /** | |
5012 | * sys_sched_get_priority_min - return minimum RT priority. | |
5013 | * @policy: scheduling class. | |
5014 | * | |
5015 | * this syscall returns the minimum rt_priority that can be used | |
5016 | * by a given scheduling class. | |
5017 | */ | |
5018 | asmlinkage long sys_sched_get_priority_min(int policy) | |
5019 | { | |
5020 | int ret = -EINVAL; | |
5021 | ||
5022 | switch (policy) { | |
5023 | case SCHED_FIFO: | |
5024 | case SCHED_RR: | |
5025 | ret = 1; | |
5026 | break; | |
5027 | case SCHED_NORMAL: | |
b0a9499c | 5028 | case SCHED_BATCH: |
dd41f596 | 5029 | case SCHED_IDLE: |
1da177e4 LT |
5030 | ret = 0; |
5031 | } | |
5032 | return ret; | |
5033 | } | |
5034 | ||
5035 | /** | |
5036 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5037 | * @pid: pid of the process. | |
5038 | * @interval: userspace pointer to the timeslice value. | |
5039 | * | |
5040 | * this syscall writes the default timeslice value of a given process | |
5041 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5042 | */ | |
5043 | asmlinkage | |
5044 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
5045 | { | |
36c8b586 | 5046 | struct task_struct *p; |
a4ec24b4 | 5047 | unsigned int time_slice; |
3a5c359a | 5048 | int retval; |
1da177e4 | 5049 | struct timespec t; |
1da177e4 LT |
5050 | |
5051 | if (pid < 0) | |
3a5c359a | 5052 | return -EINVAL; |
1da177e4 LT |
5053 | |
5054 | retval = -ESRCH; | |
5055 | read_lock(&tasklist_lock); | |
5056 | p = find_process_by_pid(pid); | |
5057 | if (!p) | |
5058 | goto out_unlock; | |
5059 | ||
5060 | retval = security_task_getscheduler(p); | |
5061 | if (retval) | |
5062 | goto out_unlock; | |
5063 | ||
77034937 IM |
5064 | /* |
5065 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
5066 | * tasks that are on an otherwise idle runqueue: | |
5067 | */ | |
5068 | time_slice = 0; | |
5069 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 5070 | time_slice = DEF_TIMESLICE; |
77034937 | 5071 | } else { |
a4ec24b4 DA |
5072 | struct sched_entity *se = &p->se; |
5073 | unsigned long flags; | |
5074 | struct rq *rq; | |
5075 | ||
5076 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
5077 | if (rq->cfs.load.weight) |
5078 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
5079 | task_rq_unlock(rq, &flags); |
5080 | } | |
1da177e4 | 5081 | read_unlock(&tasklist_lock); |
a4ec24b4 | 5082 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5083 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5084 | return retval; |
3a5c359a | 5085 | |
1da177e4 LT |
5086 | out_unlock: |
5087 | read_unlock(&tasklist_lock); | |
5088 | return retval; | |
5089 | } | |
5090 | ||
2ed6e34f | 5091 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 | 5092 | |
82a1fcb9 | 5093 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5094 | { |
1da177e4 | 5095 | unsigned long free = 0; |
36c8b586 | 5096 | unsigned state; |
1da177e4 | 5097 | |
1da177e4 | 5098 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 5099 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5100 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5101 | #if BITS_PER_LONG == 32 |
1da177e4 | 5102 | if (state == TASK_RUNNING) |
cc4ea795 | 5103 | printk(KERN_CONT " running "); |
1da177e4 | 5104 | else |
cc4ea795 | 5105 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5106 | #else |
5107 | if (state == TASK_RUNNING) | |
cc4ea795 | 5108 | printk(KERN_CONT " running task "); |
1da177e4 | 5109 | else |
cc4ea795 | 5110 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5111 | #endif |
5112 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
5113 | { | |
10ebffde | 5114 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
5115 | while (!*n) |
5116 | n++; | |
10ebffde | 5117 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
5118 | } |
5119 | #endif | |
ba25f9dc | 5120 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 5121 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 LT |
5122 | |
5123 | if (state != TASK_RUNNING) | |
5124 | show_stack(p, NULL); | |
5125 | } | |
5126 | ||
e59e2ae2 | 5127 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5128 | { |
36c8b586 | 5129 | struct task_struct *g, *p; |
1da177e4 | 5130 | |
4bd77321 IM |
5131 | #if BITS_PER_LONG == 32 |
5132 | printk(KERN_INFO | |
5133 | " task PC stack pid father\n"); | |
1da177e4 | 5134 | #else |
4bd77321 IM |
5135 | printk(KERN_INFO |
5136 | " task PC stack pid father\n"); | |
1da177e4 LT |
5137 | #endif |
5138 | read_lock(&tasklist_lock); | |
5139 | do_each_thread(g, p) { | |
5140 | /* | |
5141 | * reset the NMI-timeout, listing all files on a slow | |
5142 | * console might take alot of time: | |
5143 | */ | |
5144 | touch_nmi_watchdog(); | |
39bc89fd | 5145 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5146 | sched_show_task(p); |
1da177e4 LT |
5147 | } while_each_thread(g, p); |
5148 | ||
04c9167f JF |
5149 | touch_all_softlockup_watchdogs(); |
5150 | ||
dd41f596 IM |
5151 | #ifdef CONFIG_SCHED_DEBUG |
5152 | sysrq_sched_debug_show(); | |
5153 | #endif | |
1da177e4 | 5154 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5155 | /* |
5156 | * Only show locks if all tasks are dumped: | |
5157 | */ | |
5158 | if (state_filter == -1) | |
5159 | debug_show_all_locks(); | |
1da177e4 LT |
5160 | } |
5161 | ||
1df21055 IM |
5162 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5163 | { | |
dd41f596 | 5164 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5165 | } |
5166 | ||
f340c0d1 IM |
5167 | /** |
5168 | * init_idle - set up an idle thread for a given CPU | |
5169 | * @idle: task in question | |
5170 | * @cpu: cpu the idle task belongs to | |
5171 | * | |
5172 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5173 | * flag, to make booting more robust. | |
5174 | */ | |
5c1e1767 | 5175 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5176 | { |
70b97a7f | 5177 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5178 | unsigned long flags; |
5179 | ||
dd41f596 IM |
5180 | __sched_fork(idle); |
5181 | idle->se.exec_start = sched_clock(); | |
5182 | ||
b29739f9 | 5183 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 5184 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 5185 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
5186 | |
5187 | spin_lock_irqsave(&rq->lock, flags); | |
5188 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
5189 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5190 | idle->oncpu = 1; | |
5191 | #endif | |
1da177e4 LT |
5192 | spin_unlock_irqrestore(&rq->lock, flags); |
5193 | ||
5194 | /* Set the preempt count _outside_ the spinlocks! */ | |
5195 | #if defined(CONFIG_PREEMPT) && !defined(CONFIG_PREEMPT_BKL) | |
a1261f54 | 5196 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); |
1da177e4 | 5197 | #else |
a1261f54 | 5198 | task_thread_info(idle)->preempt_count = 0; |
1da177e4 | 5199 | #endif |
dd41f596 IM |
5200 | /* |
5201 | * The idle tasks have their own, simple scheduling class: | |
5202 | */ | |
5203 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5204 | } |
5205 | ||
5206 | /* | |
5207 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5208 | * indicates which cpus entered this state. This is used | |
5209 | * in the rcu update to wait only for active cpus. For system | |
5210 | * which do not switch off the HZ timer nohz_cpu_mask should | |
5211 | * always be CPU_MASK_NONE. | |
5212 | */ | |
5213 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
5214 | ||
19978ca6 IM |
5215 | /* |
5216 | * Increase the granularity value when there are more CPUs, | |
5217 | * because with more CPUs the 'effective latency' as visible | |
5218 | * to users decreases. But the relationship is not linear, | |
5219 | * so pick a second-best guess by going with the log2 of the | |
5220 | * number of CPUs. | |
5221 | * | |
5222 | * This idea comes from the SD scheduler of Con Kolivas: | |
5223 | */ | |
5224 | static inline void sched_init_granularity(void) | |
5225 | { | |
5226 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
5227 | const unsigned long limit = 200000000; | |
5228 | ||
5229 | sysctl_sched_min_granularity *= factor; | |
5230 | if (sysctl_sched_min_granularity > limit) | |
5231 | sysctl_sched_min_granularity = limit; | |
5232 | ||
5233 | sysctl_sched_latency *= factor; | |
5234 | if (sysctl_sched_latency > limit) | |
5235 | sysctl_sched_latency = limit; | |
5236 | ||
5237 | sysctl_sched_wakeup_granularity *= factor; | |
5238 | sysctl_sched_batch_wakeup_granularity *= factor; | |
5239 | } | |
5240 | ||
1da177e4 LT |
5241 | #ifdef CONFIG_SMP |
5242 | /* | |
5243 | * This is how migration works: | |
5244 | * | |
70b97a7f | 5245 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5246 | * runqueue and wake up that CPU's migration thread. |
5247 | * 2) we down() the locked semaphore => thread blocks. | |
5248 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5249 | * thread off the CPU) | |
5250 | * 4) it gets the migration request and checks whether the migrated | |
5251 | * task is still in the wrong runqueue. | |
5252 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5253 | * it and puts it into the right queue. | |
5254 | * 6) migration thread up()s the semaphore. | |
5255 | * 7) we wake up and the migration is done. | |
5256 | */ | |
5257 | ||
5258 | /* | |
5259 | * Change a given task's CPU affinity. Migrate the thread to a | |
5260 | * proper CPU and schedule it away if the CPU it's executing on | |
5261 | * is removed from the allowed bitmask. | |
5262 | * | |
5263 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5264 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5265 | * call is not atomic; no spinlocks may be held. |
5266 | */ | |
36c8b586 | 5267 | int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask) |
1da177e4 | 5268 | { |
70b97a7f | 5269 | struct migration_req req; |
1da177e4 | 5270 | unsigned long flags; |
70b97a7f | 5271 | struct rq *rq; |
48f24c4d | 5272 | int ret = 0; |
1da177e4 LT |
5273 | |
5274 | rq = task_rq_lock(p, &flags); | |
5275 | if (!cpus_intersects(new_mask, cpu_online_map)) { | |
5276 | ret = -EINVAL; | |
5277 | goto out; | |
5278 | } | |
5279 | ||
73fe6aae GH |
5280 | if (p->sched_class->set_cpus_allowed) |
5281 | p->sched_class->set_cpus_allowed(p, &new_mask); | |
5282 | else { | |
0eab9146 | 5283 | p->cpus_allowed = new_mask; |
73fe6aae GH |
5284 | p->nr_cpus_allowed = cpus_weight(new_mask); |
5285 | } | |
5286 | ||
1da177e4 LT |
5287 | /* Can the task run on the task's current CPU? If so, we're done */ |
5288 | if (cpu_isset(task_cpu(p), new_mask)) | |
5289 | goto out; | |
5290 | ||
5291 | if (migrate_task(p, any_online_cpu(new_mask), &req)) { | |
5292 | /* Need help from migration thread: drop lock and wait. */ | |
5293 | task_rq_unlock(rq, &flags); | |
5294 | wake_up_process(rq->migration_thread); | |
5295 | wait_for_completion(&req.done); | |
5296 | tlb_migrate_finish(p->mm); | |
5297 | return 0; | |
5298 | } | |
5299 | out: | |
5300 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5301 | |
1da177e4 LT |
5302 | return ret; |
5303 | } | |
1da177e4 LT |
5304 | EXPORT_SYMBOL_GPL(set_cpus_allowed); |
5305 | ||
5306 | /* | |
41a2d6cf | 5307 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5308 | * this because either it can't run here any more (set_cpus_allowed() |
5309 | * away from this CPU, or CPU going down), or because we're | |
5310 | * attempting to rebalance this task on exec (sched_exec). | |
5311 | * | |
5312 | * So we race with normal scheduler movements, but that's OK, as long | |
5313 | * as the task is no longer on this CPU. | |
efc30814 KK |
5314 | * |
5315 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5316 | */ |
efc30814 | 5317 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5318 | { |
70b97a7f | 5319 | struct rq *rq_dest, *rq_src; |
dd41f596 | 5320 | int ret = 0, on_rq; |
1da177e4 LT |
5321 | |
5322 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 5323 | return ret; |
1da177e4 LT |
5324 | |
5325 | rq_src = cpu_rq(src_cpu); | |
5326 | rq_dest = cpu_rq(dest_cpu); | |
5327 | ||
5328 | double_rq_lock(rq_src, rq_dest); | |
5329 | /* Already moved. */ | |
5330 | if (task_cpu(p) != src_cpu) | |
5331 | goto out; | |
5332 | /* Affinity changed (again). */ | |
5333 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
5334 | goto out; | |
5335 | ||
dd41f596 | 5336 | on_rq = p->se.on_rq; |
6e82a3be | 5337 | if (on_rq) |
2e1cb74a | 5338 | deactivate_task(rq_src, p, 0); |
6e82a3be | 5339 | |
1da177e4 | 5340 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
5341 | if (on_rq) { |
5342 | activate_task(rq_dest, p, 0); | |
5343 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 5344 | } |
efc30814 | 5345 | ret = 1; |
1da177e4 LT |
5346 | out: |
5347 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 5348 | return ret; |
1da177e4 LT |
5349 | } |
5350 | ||
5351 | /* | |
5352 | * migration_thread - this is a highprio system thread that performs | |
5353 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5354 | * another runqueue. | |
5355 | */ | |
95cdf3b7 | 5356 | static int migration_thread(void *data) |
1da177e4 | 5357 | { |
1da177e4 | 5358 | int cpu = (long)data; |
70b97a7f | 5359 | struct rq *rq; |
1da177e4 LT |
5360 | |
5361 | rq = cpu_rq(cpu); | |
5362 | BUG_ON(rq->migration_thread != current); | |
5363 | ||
5364 | set_current_state(TASK_INTERRUPTIBLE); | |
5365 | while (!kthread_should_stop()) { | |
70b97a7f | 5366 | struct migration_req *req; |
1da177e4 | 5367 | struct list_head *head; |
1da177e4 | 5368 | |
1da177e4 LT |
5369 | spin_lock_irq(&rq->lock); |
5370 | ||
5371 | if (cpu_is_offline(cpu)) { | |
5372 | spin_unlock_irq(&rq->lock); | |
5373 | goto wait_to_die; | |
5374 | } | |
5375 | ||
5376 | if (rq->active_balance) { | |
5377 | active_load_balance(rq, cpu); | |
5378 | rq->active_balance = 0; | |
5379 | } | |
5380 | ||
5381 | head = &rq->migration_queue; | |
5382 | ||
5383 | if (list_empty(head)) { | |
5384 | spin_unlock_irq(&rq->lock); | |
5385 | schedule(); | |
5386 | set_current_state(TASK_INTERRUPTIBLE); | |
5387 | continue; | |
5388 | } | |
70b97a7f | 5389 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5390 | list_del_init(head->next); |
5391 | ||
674311d5 NP |
5392 | spin_unlock(&rq->lock); |
5393 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5394 | local_irq_enable(); | |
1da177e4 LT |
5395 | |
5396 | complete(&req->done); | |
5397 | } | |
5398 | __set_current_state(TASK_RUNNING); | |
5399 | return 0; | |
5400 | ||
5401 | wait_to_die: | |
5402 | /* Wait for kthread_stop */ | |
5403 | set_current_state(TASK_INTERRUPTIBLE); | |
5404 | while (!kthread_should_stop()) { | |
5405 | schedule(); | |
5406 | set_current_state(TASK_INTERRUPTIBLE); | |
5407 | } | |
5408 | __set_current_state(TASK_RUNNING); | |
5409 | return 0; | |
5410 | } | |
5411 | ||
5412 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5413 | |
5414 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5415 | { | |
5416 | int ret; | |
5417 | ||
5418 | local_irq_disable(); | |
5419 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5420 | local_irq_enable(); | |
5421 | return ret; | |
5422 | } | |
5423 | ||
054b9108 | 5424 | /* |
3a4fa0a2 | 5425 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
5426 | * NOTE: interrupts should be disabled by the caller |
5427 | */ | |
48f24c4d | 5428 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5429 | { |
efc30814 | 5430 | unsigned long flags; |
1da177e4 | 5431 | cpumask_t mask; |
70b97a7f IM |
5432 | struct rq *rq; |
5433 | int dest_cpu; | |
1da177e4 | 5434 | |
3a5c359a AK |
5435 | do { |
5436 | /* On same node? */ | |
5437 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
5438 | cpus_and(mask, mask, p->cpus_allowed); | |
5439 | dest_cpu = any_online_cpu(mask); | |
5440 | ||
5441 | /* On any allowed CPU? */ | |
5442 | if (dest_cpu == NR_CPUS) | |
5443 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
5444 | ||
5445 | /* No more Mr. Nice Guy. */ | |
5446 | if (dest_cpu == NR_CPUS) { | |
470fd646 CW |
5447 | cpumask_t cpus_allowed = cpuset_cpus_allowed_locked(p); |
5448 | /* | |
5449 | * Try to stay on the same cpuset, where the | |
5450 | * current cpuset may be a subset of all cpus. | |
5451 | * The cpuset_cpus_allowed_locked() variant of | |
41a2d6cf | 5452 | * cpuset_cpus_allowed() will not block. It must be |
470fd646 CW |
5453 | * called within calls to cpuset_lock/cpuset_unlock. |
5454 | */ | |
3a5c359a | 5455 | rq = task_rq_lock(p, &flags); |
470fd646 | 5456 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
5457 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5458 | task_rq_unlock(rq, &flags); | |
1da177e4 | 5459 | |
3a5c359a AK |
5460 | /* |
5461 | * Don't tell them about moving exiting tasks or | |
5462 | * kernel threads (both mm NULL), since they never | |
5463 | * leave kernel. | |
5464 | */ | |
41a2d6cf | 5465 | if (p->mm && printk_ratelimit()) { |
3a5c359a AK |
5466 | printk(KERN_INFO "process %d (%s) no " |
5467 | "longer affine to cpu%d\n", | |
41a2d6cf IM |
5468 | task_pid_nr(p), p->comm, dead_cpu); |
5469 | } | |
3a5c359a | 5470 | } |
f7b4cddc | 5471 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
5472 | } |
5473 | ||
5474 | /* | |
5475 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5476 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5477 | * for performance reasons the counter is not stricly tracking tasks to | |
5478 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5479 | * to keep the global sum constant after CPU-down: | |
5480 | */ | |
70b97a7f | 5481 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5482 | { |
70b97a7f | 5483 | struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL)); |
1da177e4 LT |
5484 | unsigned long flags; |
5485 | ||
5486 | local_irq_save(flags); | |
5487 | double_rq_lock(rq_src, rq_dest); | |
5488 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5489 | rq_src->nr_uninterruptible = 0; | |
5490 | double_rq_unlock(rq_src, rq_dest); | |
5491 | local_irq_restore(flags); | |
5492 | } | |
5493 | ||
5494 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5495 | static void migrate_live_tasks(int src_cpu) | |
5496 | { | |
48f24c4d | 5497 | struct task_struct *p, *t; |
1da177e4 | 5498 | |
f7b4cddc | 5499 | read_lock(&tasklist_lock); |
1da177e4 | 5500 | |
48f24c4d IM |
5501 | do_each_thread(t, p) { |
5502 | if (p == current) | |
1da177e4 LT |
5503 | continue; |
5504 | ||
48f24c4d IM |
5505 | if (task_cpu(p) == src_cpu) |
5506 | move_task_off_dead_cpu(src_cpu, p); | |
5507 | } while_each_thread(t, p); | |
1da177e4 | 5508 | |
f7b4cddc | 5509 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5510 | } |
5511 | ||
dd41f596 IM |
5512 | /* |
5513 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5514 | * It does so by boosting its priority to highest possible. |
5515 | * Used by CPU offline code. | |
1da177e4 LT |
5516 | */ |
5517 | void sched_idle_next(void) | |
5518 | { | |
48f24c4d | 5519 | int this_cpu = smp_processor_id(); |
70b97a7f | 5520 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5521 | struct task_struct *p = rq->idle; |
5522 | unsigned long flags; | |
5523 | ||
5524 | /* cpu has to be offline */ | |
48f24c4d | 5525 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5526 | |
48f24c4d IM |
5527 | /* |
5528 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5529 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5530 | */ |
5531 | spin_lock_irqsave(&rq->lock, flags); | |
5532 | ||
dd41f596 | 5533 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 5534 | |
94bc9a7b DA |
5535 | update_rq_clock(rq); |
5536 | activate_task(rq, p, 0); | |
1da177e4 LT |
5537 | |
5538 | spin_unlock_irqrestore(&rq->lock, flags); | |
5539 | } | |
5540 | ||
48f24c4d IM |
5541 | /* |
5542 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5543 | * offline. |
5544 | */ | |
5545 | void idle_task_exit(void) | |
5546 | { | |
5547 | struct mm_struct *mm = current->active_mm; | |
5548 | ||
5549 | BUG_ON(cpu_online(smp_processor_id())); | |
5550 | ||
5551 | if (mm != &init_mm) | |
5552 | switch_mm(mm, &init_mm, current); | |
5553 | mmdrop(mm); | |
5554 | } | |
5555 | ||
054b9108 | 5556 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5557 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5558 | { |
70b97a7f | 5559 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5560 | |
5561 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5562 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5563 | |
5564 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5565 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5566 | |
48f24c4d | 5567 | get_task_struct(p); |
1da177e4 LT |
5568 | |
5569 | /* | |
5570 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 5571 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
5572 | * fine. |
5573 | */ | |
f7b4cddc | 5574 | spin_unlock_irq(&rq->lock); |
48f24c4d | 5575 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 5576 | spin_lock_irq(&rq->lock); |
1da177e4 | 5577 | |
48f24c4d | 5578 | put_task_struct(p); |
1da177e4 LT |
5579 | } |
5580 | ||
5581 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5582 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5583 | { | |
70b97a7f | 5584 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5585 | struct task_struct *next; |
48f24c4d | 5586 | |
dd41f596 IM |
5587 | for ( ; ; ) { |
5588 | if (!rq->nr_running) | |
5589 | break; | |
a8e504d2 | 5590 | update_rq_clock(rq); |
ff95f3df | 5591 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
5592 | if (!next) |
5593 | break; | |
5594 | migrate_dead(dead_cpu, next); | |
e692ab53 | 5595 | |
1da177e4 LT |
5596 | } |
5597 | } | |
5598 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5599 | ||
e692ab53 NP |
5600 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5601 | ||
5602 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5603 | { |
5604 | .procname = "sched_domain", | |
c57baf1e | 5605 | .mode = 0555, |
e0361851 | 5606 | }, |
38605cae | 5607 | {0, }, |
e692ab53 NP |
5608 | }; |
5609 | ||
5610 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 5611 | { |
c57baf1e | 5612 | .ctl_name = CTL_KERN, |
e0361851 | 5613 | .procname = "kernel", |
c57baf1e | 5614 | .mode = 0555, |
e0361851 AD |
5615 | .child = sd_ctl_dir, |
5616 | }, | |
38605cae | 5617 | {0, }, |
e692ab53 NP |
5618 | }; |
5619 | ||
5620 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5621 | { | |
5622 | struct ctl_table *entry = | |
5cf9f062 | 5623 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5624 | |
e692ab53 NP |
5625 | return entry; |
5626 | } | |
5627 | ||
6382bc90 MM |
5628 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5629 | { | |
cd790076 | 5630 | struct ctl_table *entry; |
6382bc90 | 5631 | |
cd790076 MM |
5632 | /* |
5633 | * In the intermediate directories, both the child directory and | |
5634 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5635 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5636 | * static strings and all have proc handlers. |
5637 | */ | |
5638 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5639 | if (entry->child) |
5640 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5641 | if (entry->proc_handler == NULL) |
5642 | kfree(entry->procname); | |
5643 | } | |
6382bc90 MM |
5644 | |
5645 | kfree(*tablep); | |
5646 | *tablep = NULL; | |
5647 | } | |
5648 | ||
e692ab53 | 5649 | static void |
e0361851 | 5650 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5651 | const char *procname, void *data, int maxlen, |
5652 | mode_t mode, proc_handler *proc_handler) | |
5653 | { | |
e692ab53 NP |
5654 | entry->procname = procname; |
5655 | entry->data = data; | |
5656 | entry->maxlen = maxlen; | |
5657 | entry->mode = mode; | |
5658 | entry->proc_handler = proc_handler; | |
5659 | } | |
5660 | ||
5661 | static struct ctl_table * | |
5662 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5663 | { | |
ace8b3d6 | 5664 | struct ctl_table *table = sd_alloc_ctl_entry(12); |
e692ab53 | 5665 | |
ad1cdc1d MM |
5666 | if (table == NULL) |
5667 | return NULL; | |
5668 | ||
e0361851 | 5669 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5670 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5671 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5672 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5673 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5674 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5675 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5676 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5677 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5678 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5679 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5680 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5681 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5682 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5683 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5684 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5685 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5686 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5687 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5688 | &sd->cache_nice_tries, |
5689 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5690 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5691 | sizeof(int), 0644, proc_dointvec_minmax); |
6323469f | 5692 | /* &table[11] is terminator */ |
e692ab53 NP |
5693 | |
5694 | return table; | |
5695 | } | |
5696 | ||
9a4e7159 | 5697 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5698 | { |
5699 | struct ctl_table *entry, *table; | |
5700 | struct sched_domain *sd; | |
5701 | int domain_num = 0, i; | |
5702 | char buf[32]; | |
5703 | ||
5704 | for_each_domain(cpu, sd) | |
5705 | domain_num++; | |
5706 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5707 | if (table == NULL) |
5708 | return NULL; | |
e692ab53 NP |
5709 | |
5710 | i = 0; | |
5711 | for_each_domain(cpu, sd) { | |
5712 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5713 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5714 | entry->mode = 0555; |
e692ab53 NP |
5715 | entry->child = sd_alloc_ctl_domain_table(sd); |
5716 | entry++; | |
5717 | i++; | |
5718 | } | |
5719 | return table; | |
5720 | } | |
5721 | ||
5722 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5723 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
5724 | { |
5725 | int i, cpu_num = num_online_cpus(); | |
5726 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
5727 | char buf[32]; | |
5728 | ||
7378547f MM |
5729 | WARN_ON(sd_ctl_dir[0].child); |
5730 | sd_ctl_dir[0].child = entry; | |
5731 | ||
ad1cdc1d MM |
5732 | if (entry == NULL) |
5733 | return; | |
5734 | ||
97b6ea7b | 5735 | for_each_online_cpu(i) { |
e692ab53 | 5736 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5737 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5738 | entry->mode = 0555; |
e692ab53 | 5739 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5740 | entry++; |
e692ab53 | 5741 | } |
7378547f MM |
5742 | |
5743 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5744 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5745 | } | |
6382bc90 | 5746 | |
7378547f | 5747 | /* may be called multiple times per register */ |
6382bc90 MM |
5748 | static void unregister_sched_domain_sysctl(void) |
5749 | { | |
7378547f MM |
5750 | if (sd_sysctl_header) |
5751 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5752 | sd_sysctl_header = NULL; |
7378547f MM |
5753 | if (sd_ctl_dir[0].child) |
5754 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5755 | } |
e692ab53 | 5756 | #else |
6382bc90 MM |
5757 | static void register_sched_domain_sysctl(void) |
5758 | { | |
5759 | } | |
5760 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5761 | { |
5762 | } | |
5763 | #endif | |
5764 | ||
1da177e4 LT |
5765 | /* |
5766 | * migration_call - callback that gets triggered when a CPU is added. | |
5767 | * Here we can start up the necessary migration thread for the new CPU. | |
5768 | */ | |
48f24c4d IM |
5769 | static int __cpuinit |
5770 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5771 | { |
1da177e4 | 5772 | struct task_struct *p; |
48f24c4d | 5773 | int cpu = (long)hcpu; |
1da177e4 | 5774 | unsigned long flags; |
70b97a7f | 5775 | struct rq *rq; |
1da177e4 LT |
5776 | |
5777 | switch (action) { | |
5be9361c | 5778 | |
1da177e4 | 5779 | case CPU_UP_PREPARE: |
8bb78442 | 5780 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 5781 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
5782 | if (IS_ERR(p)) |
5783 | return NOTIFY_BAD; | |
1da177e4 LT |
5784 | kthread_bind(p, cpu); |
5785 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5786 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 5787 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
5788 | task_rq_unlock(rq, &flags); |
5789 | cpu_rq(cpu)->migration_thread = p; | |
5790 | break; | |
48f24c4d | 5791 | |
1da177e4 | 5792 | case CPU_ONLINE: |
8bb78442 | 5793 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 5794 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 5795 | wake_up_process(cpu_rq(cpu)->migration_thread); |
57d885fe GH |
5796 | |
5797 | /* Update our root-domain */ | |
5798 | rq = cpu_rq(cpu); | |
5799 | spin_lock_irqsave(&rq->lock, flags); | |
5800 | if (rq->rd) { | |
5801 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
5802 | cpu_set(cpu, rq->rd->online); | |
5803 | } | |
5804 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 5805 | break; |
48f24c4d | 5806 | |
1da177e4 LT |
5807 | #ifdef CONFIG_HOTPLUG_CPU |
5808 | case CPU_UP_CANCELED: | |
8bb78442 | 5809 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5810 | if (!cpu_rq(cpu)->migration_thread) |
5811 | break; | |
41a2d6cf | 5812 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
5813 | kthread_bind(cpu_rq(cpu)->migration_thread, |
5814 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
5815 | kthread_stop(cpu_rq(cpu)->migration_thread); |
5816 | cpu_rq(cpu)->migration_thread = NULL; | |
5817 | break; | |
48f24c4d | 5818 | |
1da177e4 | 5819 | case CPU_DEAD: |
8bb78442 | 5820 | case CPU_DEAD_FROZEN: |
470fd646 | 5821 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
5822 | migrate_live_tasks(cpu); |
5823 | rq = cpu_rq(cpu); | |
5824 | kthread_stop(rq->migration_thread); | |
5825 | rq->migration_thread = NULL; | |
5826 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 5827 | spin_lock_irq(&rq->lock); |
a8e504d2 | 5828 | update_rq_clock(rq); |
2e1cb74a | 5829 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 5830 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
5831 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5832 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 5833 | migrate_dead_tasks(cpu); |
d2da272a | 5834 | spin_unlock_irq(&rq->lock); |
470fd646 | 5835 | cpuset_unlock(); |
1da177e4 LT |
5836 | migrate_nr_uninterruptible(rq); |
5837 | BUG_ON(rq->nr_running != 0); | |
5838 | ||
41a2d6cf IM |
5839 | /* |
5840 | * No need to migrate the tasks: it was best-effort if | |
5841 | * they didn't take sched_hotcpu_mutex. Just wake up | |
5842 | * the requestors. | |
5843 | */ | |
1da177e4 LT |
5844 | spin_lock_irq(&rq->lock); |
5845 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
5846 | struct migration_req *req; |
5847 | ||
1da177e4 | 5848 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5849 | struct migration_req, list); |
1da177e4 LT |
5850 | list_del_init(&req->list); |
5851 | complete(&req->done); | |
5852 | } | |
5853 | spin_unlock_irq(&rq->lock); | |
5854 | break; | |
57d885fe GH |
5855 | |
5856 | case CPU_DOWN_PREPARE: | |
5857 | /* Update our root-domain */ | |
5858 | rq = cpu_rq(cpu); | |
5859 | spin_lock_irqsave(&rq->lock, flags); | |
5860 | if (rq->rd) { | |
5861 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
5862 | cpu_clear(cpu, rq->rd->online); | |
5863 | } | |
5864 | spin_unlock_irqrestore(&rq->lock, flags); | |
5865 | break; | |
1da177e4 LT |
5866 | #endif |
5867 | } | |
5868 | return NOTIFY_OK; | |
5869 | } | |
5870 | ||
5871 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
5872 | * happens before everything else. | |
5873 | */ | |
26c2143b | 5874 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5875 | .notifier_call = migration_call, |
5876 | .priority = 10 | |
5877 | }; | |
5878 | ||
e6fe6649 | 5879 | void __init migration_init(void) |
1da177e4 LT |
5880 | { |
5881 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5882 | int err; |
48f24c4d IM |
5883 | |
5884 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5885 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5886 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5887 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5888 | register_cpu_notifier(&migration_notifier); | |
1da177e4 LT |
5889 | } |
5890 | #endif | |
5891 | ||
5892 | #ifdef CONFIG_SMP | |
476f3534 CL |
5893 | |
5894 | /* Number of possible processor ids */ | |
5895 | int nr_cpu_ids __read_mostly = NR_CPUS; | |
5896 | EXPORT_SYMBOL(nr_cpu_ids); | |
5897 | ||
3e9830dc | 5898 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff IM |
5899 | |
5900 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level) | |
1da177e4 | 5901 | { |
4dcf6aff IM |
5902 | struct sched_group *group = sd->groups; |
5903 | cpumask_t groupmask; | |
5904 | char str[NR_CPUS]; | |
1da177e4 | 5905 | |
4dcf6aff IM |
5906 | cpumask_scnprintf(str, NR_CPUS, sd->span); |
5907 | cpus_clear(groupmask); | |
5908 | ||
5909 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5910 | ||
5911 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
5912 | printk("does not load-balance\n"); | |
5913 | if (sd->parent) | |
5914 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
5915 | " has parent"); | |
5916 | return -1; | |
41c7ce9a NP |
5917 | } |
5918 | ||
4dcf6aff IM |
5919 | printk(KERN_CONT "span %s\n", str); |
5920 | ||
5921 | if (!cpu_isset(cpu, sd->span)) { | |
5922 | printk(KERN_ERR "ERROR: domain->span does not contain " | |
5923 | "CPU%d\n", cpu); | |
5924 | } | |
5925 | if (!cpu_isset(cpu, group->cpumask)) { | |
5926 | printk(KERN_ERR "ERROR: domain->groups does not contain" | |
5927 | " CPU%d\n", cpu); | |
5928 | } | |
1da177e4 | 5929 | |
4dcf6aff | 5930 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5931 | do { |
4dcf6aff IM |
5932 | if (!group) { |
5933 | printk("\n"); | |
5934 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5935 | break; |
5936 | } | |
5937 | ||
4dcf6aff IM |
5938 | if (!group->__cpu_power) { |
5939 | printk(KERN_CONT "\n"); | |
5940 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5941 | "set\n"); | |
5942 | break; | |
5943 | } | |
1da177e4 | 5944 | |
4dcf6aff IM |
5945 | if (!cpus_weight(group->cpumask)) { |
5946 | printk(KERN_CONT "\n"); | |
5947 | printk(KERN_ERR "ERROR: empty group\n"); | |
5948 | break; | |
5949 | } | |
1da177e4 | 5950 | |
4dcf6aff IM |
5951 | if (cpus_intersects(groupmask, group->cpumask)) { |
5952 | printk(KERN_CONT "\n"); | |
5953 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
5954 | break; | |
5955 | } | |
1da177e4 | 5956 | |
4dcf6aff | 5957 | cpus_or(groupmask, groupmask, group->cpumask); |
1da177e4 | 5958 | |
4dcf6aff IM |
5959 | cpumask_scnprintf(str, NR_CPUS, group->cpumask); |
5960 | printk(KERN_CONT " %s", str); | |
1da177e4 | 5961 | |
4dcf6aff IM |
5962 | group = group->next; |
5963 | } while (group != sd->groups); | |
5964 | printk(KERN_CONT "\n"); | |
1da177e4 | 5965 | |
4dcf6aff IM |
5966 | if (!cpus_equal(sd->span, groupmask)) |
5967 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); | |
1da177e4 | 5968 | |
4dcf6aff IM |
5969 | if (sd->parent && !cpus_subset(groupmask, sd->parent->span)) |
5970 | printk(KERN_ERR "ERROR: parent span is not a superset " | |
5971 | "of domain->span\n"); | |
5972 | return 0; | |
5973 | } | |
1da177e4 | 5974 | |
4dcf6aff IM |
5975 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5976 | { | |
5977 | int level = 0; | |
1da177e4 | 5978 | |
4dcf6aff IM |
5979 | if (!sd) { |
5980 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5981 | return; | |
5982 | } | |
1da177e4 | 5983 | |
4dcf6aff IM |
5984 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5985 | ||
5986 | for (;;) { | |
5987 | if (sched_domain_debug_one(sd, cpu, level)) | |
5988 | break; | |
1da177e4 LT |
5989 | level++; |
5990 | sd = sd->parent; | |
33859f7f | 5991 | if (!sd) |
4dcf6aff IM |
5992 | break; |
5993 | } | |
1da177e4 LT |
5994 | } |
5995 | #else | |
48f24c4d | 5996 | # define sched_domain_debug(sd, cpu) do { } while (0) |
1da177e4 LT |
5997 | #endif |
5998 | ||
1a20ff27 | 5999 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
6000 | { |
6001 | if (cpus_weight(sd->span) == 1) | |
6002 | return 1; | |
6003 | ||
6004 | /* Following flags need at least 2 groups */ | |
6005 | if (sd->flags & (SD_LOAD_BALANCE | | |
6006 | SD_BALANCE_NEWIDLE | | |
6007 | SD_BALANCE_FORK | | |
89c4710e SS |
6008 | SD_BALANCE_EXEC | |
6009 | SD_SHARE_CPUPOWER | | |
6010 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6011 | if (sd->groups != sd->groups->next) |
6012 | return 0; | |
6013 | } | |
6014 | ||
6015 | /* Following flags don't use groups */ | |
6016 | if (sd->flags & (SD_WAKE_IDLE | | |
6017 | SD_WAKE_AFFINE | | |
6018 | SD_WAKE_BALANCE)) | |
6019 | return 0; | |
6020 | ||
6021 | return 1; | |
6022 | } | |
6023 | ||
48f24c4d IM |
6024 | static int |
6025 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6026 | { |
6027 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6028 | ||
6029 | if (sd_degenerate(parent)) | |
6030 | return 1; | |
6031 | ||
6032 | if (!cpus_equal(sd->span, parent->span)) | |
6033 | return 0; | |
6034 | ||
6035 | /* Does parent contain flags not in child? */ | |
6036 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
6037 | if (cflags & SD_WAKE_AFFINE) | |
6038 | pflags &= ~SD_WAKE_BALANCE; | |
6039 | /* Flags needing groups don't count if only 1 group in parent */ | |
6040 | if (parent->groups == parent->groups->next) { | |
6041 | pflags &= ~(SD_LOAD_BALANCE | | |
6042 | SD_BALANCE_NEWIDLE | | |
6043 | SD_BALANCE_FORK | | |
89c4710e SS |
6044 | SD_BALANCE_EXEC | |
6045 | SD_SHARE_CPUPOWER | | |
6046 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
6047 | } |
6048 | if (~cflags & pflags) | |
6049 | return 0; | |
6050 | ||
6051 | return 1; | |
6052 | } | |
6053 | ||
57d885fe GH |
6054 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6055 | { | |
6056 | unsigned long flags; | |
6057 | const struct sched_class *class; | |
6058 | ||
6059 | spin_lock_irqsave(&rq->lock, flags); | |
6060 | ||
6061 | if (rq->rd) { | |
6062 | struct root_domain *old_rd = rq->rd; | |
6063 | ||
0eab9146 | 6064 | for (class = sched_class_highest; class; class = class->next) { |
57d885fe GH |
6065 | if (class->leave_domain) |
6066 | class->leave_domain(rq); | |
0eab9146 | 6067 | } |
57d885fe | 6068 | |
dc938520 GH |
6069 | cpu_clear(rq->cpu, old_rd->span); |
6070 | cpu_clear(rq->cpu, old_rd->online); | |
6071 | ||
57d885fe GH |
6072 | if (atomic_dec_and_test(&old_rd->refcount)) |
6073 | kfree(old_rd); | |
6074 | } | |
6075 | ||
6076 | atomic_inc(&rd->refcount); | |
6077 | rq->rd = rd; | |
6078 | ||
dc938520 GH |
6079 | cpu_set(rq->cpu, rd->span); |
6080 | if (cpu_isset(rq->cpu, cpu_online_map)) | |
6081 | cpu_set(rq->cpu, rd->online); | |
6082 | ||
0eab9146 | 6083 | for (class = sched_class_highest; class; class = class->next) { |
57d885fe GH |
6084 | if (class->join_domain) |
6085 | class->join_domain(rq); | |
0eab9146 | 6086 | } |
57d885fe GH |
6087 | |
6088 | spin_unlock_irqrestore(&rq->lock, flags); | |
6089 | } | |
6090 | ||
dc938520 | 6091 | static void init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6092 | { |
6093 | memset(rd, 0, sizeof(*rd)); | |
6094 | ||
dc938520 GH |
6095 | cpus_clear(rd->span); |
6096 | cpus_clear(rd->online); | |
57d885fe GH |
6097 | } |
6098 | ||
6099 | static void init_defrootdomain(void) | |
6100 | { | |
dc938520 | 6101 | init_rootdomain(&def_root_domain); |
57d885fe GH |
6102 | atomic_set(&def_root_domain.refcount, 1); |
6103 | } | |
6104 | ||
dc938520 | 6105 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6106 | { |
6107 | struct root_domain *rd; | |
6108 | ||
6109 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6110 | if (!rd) | |
6111 | return NULL; | |
6112 | ||
dc938520 | 6113 | init_rootdomain(rd); |
57d885fe GH |
6114 | |
6115 | return rd; | |
6116 | } | |
6117 | ||
1da177e4 | 6118 | /* |
0eab9146 | 6119 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6120 | * hold the hotplug lock. |
6121 | */ | |
0eab9146 IM |
6122 | static void |
6123 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6124 | { |
70b97a7f | 6125 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6126 | struct sched_domain *tmp; |
6127 | ||
6128 | /* Remove the sched domains which do not contribute to scheduling. */ | |
6129 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
6130 | struct sched_domain *parent = tmp->parent; | |
6131 | if (!parent) | |
6132 | break; | |
1a848870 | 6133 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6134 | tmp->parent = parent->parent; |
1a848870 SS |
6135 | if (parent->parent) |
6136 | parent->parent->child = tmp; | |
6137 | } | |
245af2c7 SS |
6138 | } |
6139 | ||
1a848870 | 6140 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6141 | sd = sd->parent; |
1a848870 SS |
6142 | if (sd) |
6143 | sd->child = NULL; | |
6144 | } | |
1da177e4 LT |
6145 | |
6146 | sched_domain_debug(sd, cpu); | |
6147 | ||
57d885fe | 6148 | rq_attach_root(rq, rd); |
674311d5 | 6149 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6150 | } |
6151 | ||
6152 | /* cpus with isolated domains */ | |
67af63a6 | 6153 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
6154 | |
6155 | /* Setup the mask of cpus configured for isolated domains */ | |
6156 | static int __init isolated_cpu_setup(char *str) | |
6157 | { | |
6158 | int ints[NR_CPUS], i; | |
6159 | ||
6160 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
6161 | cpus_clear(cpu_isolated_map); | |
6162 | for (i = 1; i <= ints[0]; i++) | |
6163 | if (ints[i] < NR_CPUS) | |
6164 | cpu_set(ints[i], cpu_isolated_map); | |
6165 | return 1; | |
6166 | } | |
6167 | ||
8927f494 | 6168 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6169 | |
6170 | /* | |
6711cab4 SS |
6171 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6172 | * to a function which identifies what group(along with sched group) a CPU | |
6173 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
6174 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
6175 | * |
6176 | * init_sched_build_groups will build a circular linked list of the groups | |
6177 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6178 | * and ->cpu_power to 0. | |
6179 | */ | |
a616058b | 6180 | static void |
6711cab4 SS |
6181 | init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map, |
6182 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, | |
6183 | struct sched_group **sg)) | |
1da177e4 LT |
6184 | { |
6185 | struct sched_group *first = NULL, *last = NULL; | |
6186 | cpumask_t covered = CPU_MASK_NONE; | |
6187 | int i; | |
6188 | ||
6189 | for_each_cpu_mask(i, span) { | |
6711cab4 SS |
6190 | struct sched_group *sg; |
6191 | int group = group_fn(i, cpu_map, &sg); | |
1da177e4 LT |
6192 | int j; |
6193 | ||
6194 | if (cpu_isset(i, covered)) | |
6195 | continue; | |
6196 | ||
6197 | sg->cpumask = CPU_MASK_NONE; | |
5517d86b | 6198 | sg->__cpu_power = 0; |
1da177e4 LT |
6199 | |
6200 | for_each_cpu_mask(j, span) { | |
6711cab4 | 6201 | if (group_fn(j, cpu_map, NULL) != group) |
1da177e4 LT |
6202 | continue; |
6203 | ||
6204 | cpu_set(j, covered); | |
6205 | cpu_set(j, sg->cpumask); | |
6206 | } | |
6207 | if (!first) | |
6208 | first = sg; | |
6209 | if (last) | |
6210 | last->next = sg; | |
6211 | last = sg; | |
6212 | } | |
6213 | last->next = first; | |
6214 | } | |
6215 | ||
9c1cfda2 | 6216 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6217 | |
9c1cfda2 | 6218 | #ifdef CONFIG_NUMA |
198e2f18 | 6219 | |
9c1cfda2 JH |
6220 | /** |
6221 | * find_next_best_node - find the next node to include in a sched_domain | |
6222 | * @node: node whose sched_domain we're building | |
6223 | * @used_nodes: nodes already in the sched_domain | |
6224 | * | |
41a2d6cf | 6225 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6226 | * finds the closest node not already in the @used_nodes map. |
6227 | * | |
6228 | * Should use nodemask_t. | |
6229 | */ | |
6230 | static int find_next_best_node(int node, unsigned long *used_nodes) | |
6231 | { | |
6232 | int i, n, val, min_val, best_node = 0; | |
6233 | ||
6234 | min_val = INT_MAX; | |
6235 | ||
6236 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6237 | /* Start at @node */ | |
6238 | n = (node + i) % MAX_NUMNODES; | |
6239 | ||
6240 | if (!nr_cpus_node(n)) | |
6241 | continue; | |
6242 | ||
6243 | /* Skip already used nodes */ | |
6244 | if (test_bit(n, used_nodes)) | |
6245 | continue; | |
6246 | ||
6247 | /* Simple min distance search */ | |
6248 | val = node_distance(node, n); | |
6249 | ||
6250 | if (val < min_val) { | |
6251 | min_val = val; | |
6252 | best_node = n; | |
6253 | } | |
6254 | } | |
6255 | ||
6256 | set_bit(best_node, used_nodes); | |
6257 | return best_node; | |
6258 | } | |
6259 | ||
6260 | /** | |
6261 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6262 | * @node: node whose cpumask we're constructing | |
6263 | * @size: number of nodes to include in this span | |
6264 | * | |
41a2d6cf | 6265 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6266 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6267 | * out optimally. | |
6268 | */ | |
6269 | static cpumask_t sched_domain_node_span(int node) | |
6270 | { | |
9c1cfda2 | 6271 | DECLARE_BITMAP(used_nodes, MAX_NUMNODES); |
48f24c4d IM |
6272 | cpumask_t span, nodemask; |
6273 | int i; | |
9c1cfda2 JH |
6274 | |
6275 | cpus_clear(span); | |
6276 | bitmap_zero(used_nodes, MAX_NUMNODES); | |
6277 | ||
6278 | nodemask = node_to_cpumask(node); | |
6279 | cpus_or(span, span, nodemask); | |
6280 | set_bit(node, used_nodes); | |
6281 | ||
6282 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
6283 | int next_node = find_next_best_node(node, used_nodes); | |
48f24c4d | 6284 | |
9c1cfda2 JH |
6285 | nodemask = node_to_cpumask(next_node); |
6286 | cpus_or(span, span, nodemask); | |
6287 | } | |
6288 | ||
6289 | return span; | |
6290 | } | |
6291 | #endif | |
6292 | ||
5c45bf27 | 6293 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6294 | |
9c1cfda2 | 6295 | /* |
48f24c4d | 6296 | * SMT sched-domains: |
9c1cfda2 | 6297 | */ |
1da177e4 LT |
6298 | #ifdef CONFIG_SCHED_SMT |
6299 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 6300 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 6301 | |
41a2d6cf IM |
6302 | static int |
6303 | cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1da177e4 | 6304 | { |
6711cab4 SS |
6305 | if (sg) |
6306 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
6307 | return cpu; |
6308 | } | |
6309 | #endif | |
6310 | ||
48f24c4d IM |
6311 | /* |
6312 | * multi-core sched-domains: | |
6313 | */ | |
1e9f28fa SS |
6314 | #ifdef CONFIG_SCHED_MC |
6315 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 6316 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
6317 | #endif |
6318 | ||
6319 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf IM |
6320 | static int |
6321 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1e9f28fa | 6322 | { |
6711cab4 | 6323 | int group; |
d5a7430d | 6324 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 6325 | cpus_and(mask, mask, *cpu_map); |
6711cab4 SS |
6326 | group = first_cpu(mask); |
6327 | if (sg) | |
6328 | *sg = &per_cpu(sched_group_core, group); | |
6329 | return group; | |
1e9f28fa SS |
6330 | } |
6331 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf IM |
6332 | static int |
6333 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1e9f28fa | 6334 | { |
6711cab4 SS |
6335 | if (sg) |
6336 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
6337 | return cpu; |
6338 | } | |
6339 | #endif | |
6340 | ||
1da177e4 | 6341 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 6342 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 6343 | |
41a2d6cf IM |
6344 | static int |
6345 | cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1da177e4 | 6346 | { |
6711cab4 | 6347 | int group; |
48f24c4d | 6348 | #ifdef CONFIG_SCHED_MC |
1e9f28fa | 6349 | cpumask_t mask = cpu_coregroup_map(cpu); |
a616058b | 6350 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 6351 | group = first_cpu(mask); |
1e9f28fa | 6352 | #elif defined(CONFIG_SCHED_SMT) |
d5a7430d | 6353 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 6354 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 6355 | group = first_cpu(mask); |
1da177e4 | 6356 | #else |
6711cab4 | 6357 | group = cpu; |
1da177e4 | 6358 | #endif |
6711cab4 SS |
6359 | if (sg) |
6360 | *sg = &per_cpu(sched_group_phys, group); | |
6361 | return group; | |
1da177e4 LT |
6362 | } |
6363 | ||
6364 | #ifdef CONFIG_NUMA | |
1da177e4 | 6365 | /* |
9c1cfda2 JH |
6366 | * The init_sched_build_groups can't handle what we want to do with node |
6367 | * groups, so roll our own. Now each node has its own list of groups which | |
6368 | * gets dynamically allocated. | |
1da177e4 | 6369 | */ |
9c1cfda2 | 6370 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
d1b55138 | 6371 | static struct sched_group **sched_group_nodes_bycpu[NR_CPUS]; |
1da177e4 | 6372 | |
9c1cfda2 | 6373 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 6374 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 6375 | |
6711cab4 SS |
6376 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
6377 | struct sched_group **sg) | |
9c1cfda2 | 6378 | { |
6711cab4 SS |
6379 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu)); |
6380 | int group; | |
6381 | ||
6382 | cpus_and(nodemask, nodemask, *cpu_map); | |
6383 | group = first_cpu(nodemask); | |
6384 | ||
6385 | if (sg) | |
6386 | *sg = &per_cpu(sched_group_allnodes, group); | |
6387 | return group; | |
1da177e4 | 6388 | } |
6711cab4 | 6389 | |
08069033 SS |
6390 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6391 | { | |
6392 | struct sched_group *sg = group_head; | |
6393 | int j; | |
6394 | ||
6395 | if (!sg) | |
6396 | return; | |
3a5c359a AK |
6397 | do { |
6398 | for_each_cpu_mask(j, sg->cpumask) { | |
6399 | struct sched_domain *sd; | |
08069033 | 6400 | |
3a5c359a AK |
6401 | sd = &per_cpu(phys_domains, j); |
6402 | if (j != first_cpu(sd->groups->cpumask)) { | |
6403 | /* | |
6404 | * Only add "power" once for each | |
6405 | * physical package. | |
6406 | */ | |
6407 | continue; | |
6408 | } | |
08069033 | 6409 | |
3a5c359a AK |
6410 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
6411 | } | |
6412 | sg = sg->next; | |
6413 | } while (sg != group_head); | |
08069033 | 6414 | } |
1da177e4 LT |
6415 | #endif |
6416 | ||
a616058b | 6417 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6418 | /* Free memory allocated for various sched_group structures */ |
6419 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6420 | { | |
a616058b | 6421 | int cpu, i; |
51888ca2 SV |
6422 | |
6423 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
6424 | struct sched_group **sched_group_nodes |
6425 | = sched_group_nodes_bycpu[cpu]; | |
6426 | ||
51888ca2 SV |
6427 | if (!sched_group_nodes) |
6428 | continue; | |
6429 | ||
6430 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6431 | cpumask_t nodemask = node_to_cpumask(i); | |
6432 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; | |
6433 | ||
6434 | cpus_and(nodemask, nodemask, *cpu_map); | |
6435 | if (cpus_empty(nodemask)) | |
6436 | continue; | |
6437 | ||
6438 | if (sg == NULL) | |
6439 | continue; | |
6440 | sg = sg->next; | |
6441 | next_sg: | |
6442 | oldsg = sg; | |
6443 | sg = sg->next; | |
6444 | kfree(oldsg); | |
6445 | if (oldsg != sched_group_nodes[i]) | |
6446 | goto next_sg; | |
6447 | } | |
6448 | kfree(sched_group_nodes); | |
6449 | sched_group_nodes_bycpu[cpu] = NULL; | |
6450 | } | |
51888ca2 | 6451 | } |
a616058b SS |
6452 | #else |
6453 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6454 | { | |
6455 | } | |
6456 | #endif | |
51888ca2 | 6457 | |
89c4710e SS |
6458 | /* |
6459 | * Initialize sched groups cpu_power. | |
6460 | * | |
6461 | * cpu_power indicates the capacity of sched group, which is used while | |
6462 | * distributing the load between different sched groups in a sched domain. | |
6463 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6464 | * there are asymmetries in the topology. If there are asymmetries, group | |
6465 | * having more cpu_power will pickup more load compared to the group having | |
6466 | * less cpu_power. | |
6467 | * | |
6468 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
6469 | * the maximum number of tasks a group can handle in the presence of other idle | |
6470 | * or lightly loaded groups in the same sched domain. | |
6471 | */ | |
6472 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6473 | { | |
6474 | struct sched_domain *child; | |
6475 | struct sched_group *group; | |
6476 | ||
6477 | WARN_ON(!sd || !sd->groups); | |
6478 | ||
6479 | if (cpu != first_cpu(sd->groups->cpumask)) | |
6480 | return; | |
6481 | ||
6482 | child = sd->child; | |
6483 | ||
5517d86b ED |
6484 | sd->groups->__cpu_power = 0; |
6485 | ||
89c4710e SS |
6486 | /* |
6487 | * For perf policy, if the groups in child domain share resources | |
6488 | * (for example cores sharing some portions of the cache hierarchy | |
6489 | * or SMT), then set this domain groups cpu_power such that each group | |
6490 | * can handle only one task, when there are other idle groups in the | |
6491 | * same sched domain. | |
6492 | */ | |
6493 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
6494 | (child->flags & | |
6495 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 6496 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
6497 | return; |
6498 | } | |
6499 | ||
89c4710e SS |
6500 | /* |
6501 | * add cpu_power of each child group to this groups cpu_power | |
6502 | */ | |
6503 | group = child->groups; | |
6504 | do { | |
5517d86b | 6505 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
6506 | group = group->next; |
6507 | } while (group != child->groups); | |
6508 | } | |
6509 | ||
1da177e4 | 6510 | /* |
1a20ff27 DG |
6511 | * Build sched domains for a given set of cpus and attach the sched domains |
6512 | * to the individual cpus | |
1da177e4 | 6513 | */ |
51888ca2 | 6514 | static int build_sched_domains(const cpumask_t *cpu_map) |
1da177e4 LT |
6515 | { |
6516 | int i; | |
57d885fe | 6517 | struct root_domain *rd; |
d1b55138 JH |
6518 | #ifdef CONFIG_NUMA |
6519 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 6520 | int sd_allnodes = 0; |
d1b55138 JH |
6521 | |
6522 | /* | |
6523 | * Allocate the per-node list of sched groups | |
6524 | */ | |
5cf9f062 | 6525 | sched_group_nodes = kcalloc(MAX_NUMNODES, sizeof(struct sched_group *), |
41a2d6cf | 6526 | GFP_KERNEL); |
d1b55138 JH |
6527 | if (!sched_group_nodes) { |
6528 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 6529 | return -ENOMEM; |
d1b55138 JH |
6530 | } |
6531 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
6532 | #endif | |
1da177e4 | 6533 | |
dc938520 | 6534 | rd = alloc_rootdomain(); |
57d885fe GH |
6535 | if (!rd) { |
6536 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
6537 | return -ENOMEM; | |
6538 | } | |
6539 | ||
1da177e4 | 6540 | /* |
1a20ff27 | 6541 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 6542 | */ |
1a20ff27 | 6543 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6544 | struct sched_domain *sd = NULL, *p; |
6545 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(i)); | |
6546 | ||
1a20ff27 | 6547 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6548 | |
6549 | #ifdef CONFIG_NUMA | |
dd41f596 IM |
6550 | if (cpus_weight(*cpu_map) > |
6551 | SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) { | |
9c1cfda2 JH |
6552 | sd = &per_cpu(allnodes_domains, i); |
6553 | *sd = SD_ALLNODES_INIT; | |
6554 | sd->span = *cpu_map; | |
6711cab4 | 6555 | cpu_to_allnodes_group(i, cpu_map, &sd->groups); |
9c1cfda2 | 6556 | p = sd; |
6711cab4 | 6557 | sd_allnodes = 1; |
9c1cfda2 JH |
6558 | } else |
6559 | p = NULL; | |
6560 | ||
1da177e4 | 6561 | sd = &per_cpu(node_domains, i); |
1da177e4 | 6562 | *sd = SD_NODE_INIT; |
9c1cfda2 JH |
6563 | sd->span = sched_domain_node_span(cpu_to_node(i)); |
6564 | sd->parent = p; | |
1a848870 SS |
6565 | if (p) |
6566 | p->child = sd; | |
9c1cfda2 | 6567 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
6568 | #endif |
6569 | ||
6570 | p = sd; | |
6571 | sd = &per_cpu(phys_domains, i); | |
1da177e4 LT |
6572 | *sd = SD_CPU_INIT; |
6573 | sd->span = nodemask; | |
6574 | sd->parent = p; | |
1a848870 SS |
6575 | if (p) |
6576 | p->child = sd; | |
6711cab4 | 6577 | cpu_to_phys_group(i, cpu_map, &sd->groups); |
1da177e4 | 6578 | |
1e9f28fa SS |
6579 | #ifdef CONFIG_SCHED_MC |
6580 | p = sd; | |
6581 | sd = &per_cpu(core_domains, i); | |
1e9f28fa SS |
6582 | *sd = SD_MC_INIT; |
6583 | sd->span = cpu_coregroup_map(i); | |
6584 | cpus_and(sd->span, sd->span, *cpu_map); | |
6585 | sd->parent = p; | |
1a848870 | 6586 | p->child = sd; |
6711cab4 | 6587 | cpu_to_core_group(i, cpu_map, &sd->groups); |
1e9f28fa SS |
6588 | #endif |
6589 | ||
1da177e4 LT |
6590 | #ifdef CONFIG_SCHED_SMT |
6591 | p = sd; | |
6592 | sd = &per_cpu(cpu_domains, i); | |
1da177e4 | 6593 | *sd = SD_SIBLING_INIT; |
d5a7430d | 6594 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6595 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 6596 | sd->parent = p; |
1a848870 | 6597 | p->child = sd; |
6711cab4 | 6598 | cpu_to_cpu_group(i, cpu_map, &sd->groups); |
1da177e4 LT |
6599 | #endif |
6600 | } | |
6601 | ||
6602 | #ifdef CONFIG_SCHED_SMT | |
6603 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 6604 | for_each_cpu_mask(i, *cpu_map) { |
d5a7430d | 6605 | cpumask_t this_sibling_map = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6606 | cpus_and(this_sibling_map, this_sibling_map, *cpu_map); |
1da177e4 LT |
6607 | if (i != first_cpu(this_sibling_map)) |
6608 | continue; | |
6609 | ||
dd41f596 IM |
6610 | init_sched_build_groups(this_sibling_map, cpu_map, |
6611 | &cpu_to_cpu_group); | |
1da177e4 LT |
6612 | } |
6613 | #endif | |
6614 | ||
1e9f28fa SS |
6615 | #ifdef CONFIG_SCHED_MC |
6616 | /* Set up multi-core groups */ | |
6617 | for_each_cpu_mask(i, *cpu_map) { | |
6618 | cpumask_t this_core_map = cpu_coregroup_map(i); | |
6619 | cpus_and(this_core_map, this_core_map, *cpu_map); | |
6620 | if (i != first_cpu(this_core_map)) | |
6621 | continue; | |
dd41f596 IM |
6622 | init_sched_build_groups(this_core_map, cpu_map, |
6623 | &cpu_to_core_group); | |
1e9f28fa SS |
6624 | } |
6625 | #endif | |
6626 | ||
1da177e4 LT |
6627 | /* Set up physical groups */ |
6628 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6629 | cpumask_t nodemask = node_to_cpumask(i); | |
6630 | ||
1a20ff27 | 6631 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6632 | if (cpus_empty(nodemask)) |
6633 | continue; | |
6634 | ||
6711cab4 | 6635 | init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group); |
1da177e4 LT |
6636 | } |
6637 | ||
6638 | #ifdef CONFIG_NUMA | |
6639 | /* Set up node groups */ | |
6711cab4 | 6640 | if (sd_allnodes) |
dd41f596 IM |
6641 | init_sched_build_groups(*cpu_map, cpu_map, |
6642 | &cpu_to_allnodes_group); | |
9c1cfda2 JH |
6643 | |
6644 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6645 | /* Set up node groups */ | |
6646 | struct sched_group *sg, *prev; | |
6647 | cpumask_t nodemask = node_to_cpumask(i); | |
6648 | cpumask_t domainspan; | |
6649 | cpumask_t covered = CPU_MASK_NONE; | |
6650 | int j; | |
6651 | ||
6652 | cpus_and(nodemask, nodemask, *cpu_map); | |
d1b55138 JH |
6653 | if (cpus_empty(nodemask)) { |
6654 | sched_group_nodes[i] = NULL; | |
9c1cfda2 | 6655 | continue; |
d1b55138 | 6656 | } |
9c1cfda2 JH |
6657 | |
6658 | domainspan = sched_domain_node_span(i); | |
6659 | cpus_and(domainspan, domainspan, *cpu_map); | |
6660 | ||
15f0b676 | 6661 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
6662 | if (!sg) { |
6663 | printk(KERN_WARNING "Can not alloc domain group for " | |
6664 | "node %d\n", i); | |
6665 | goto error; | |
6666 | } | |
9c1cfda2 JH |
6667 | sched_group_nodes[i] = sg; |
6668 | for_each_cpu_mask(j, nodemask) { | |
6669 | struct sched_domain *sd; | |
9761eea8 | 6670 | |
9c1cfda2 JH |
6671 | sd = &per_cpu(node_domains, j); |
6672 | sd->groups = sg; | |
9c1cfda2 | 6673 | } |
5517d86b | 6674 | sg->__cpu_power = 0; |
9c1cfda2 | 6675 | sg->cpumask = nodemask; |
51888ca2 | 6676 | sg->next = sg; |
9c1cfda2 JH |
6677 | cpus_or(covered, covered, nodemask); |
6678 | prev = sg; | |
6679 | ||
6680 | for (j = 0; j < MAX_NUMNODES; j++) { | |
6681 | cpumask_t tmp, notcovered; | |
6682 | int n = (i + j) % MAX_NUMNODES; | |
6683 | ||
6684 | cpus_complement(notcovered, covered); | |
6685 | cpus_and(tmp, notcovered, *cpu_map); | |
6686 | cpus_and(tmp, tmp, domainspan); | |
6687 | if (cpus_empty(tmp)) | |
6688 | break; | |
6689 | ||
6690 | nodemask = node_to_cpumask(n); | |
6691 | cpus_and(tmp, tmp, nodemask); | |
6692 | if (cpus_empty(tmp)) | |
6693 | continue; | |
6694 | ||
15f0b676 SV |
6695 | sg = kmalloc_node(sizeof(struct sched_group), |
6696 | GFP_KERNEL, i); | |
9c1cfda2 JH |
6697 | if (!sg) { |
6698 | printk(KERN_WARNING | |
6699 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 6700 | goto error; |
9c1cfda2 | 6701 | } |
5517d86b | 6702 | sg->__cpu_power = 0; |
9c1cfda2 | 6703 | sg->cpumask = tmp; |
51888ca2 | 6704 | sg->next = prev->next; |
9c1cfda2 JH |
6705 | cpus_or(covered, covered, tmp); |
6706 | prev->next = sg; | |
6707 | prev = sg; | |
6708 | } | |
9c1cfda2 | 6709 | } |
1da177e4 LT |
6710 | #endif |
6711 | ||
6712 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 6713 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 6714 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6715 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
6716 | ||
89c4710e | 6717 | init_sched_groups_power(i, sd); |
5c45bf27 | 6718 | } |
1da177e4 | 6719 | #endif |
1e9f28fa | 6720 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 6721 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6722 | struct sched_domain *sd = &per_cpu(core_domains, i); |
6723 | ||
89c4710e | 6724 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
6725 | } |
6726 | #endif | |
1e9f28fa | 6727 | |
5c45bf27 | 6728 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6729 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
6730 | ||
89c4710e | 6731 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6732 | } |
6733 | ||
9c1cfda2 | 6734 | #ifdef CONFIG_NUMA |
08069033 SS |
6735 | for (i = 0; i < MAX_NUMNODES; i++) |
6736 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 6737 | |
6711cab4 SS |
6738 | if (sd_allnodes) { |
6739 | struct sched_group *sg; | |
f712c0c7 | 6740 | |
6711cab4 | 6741 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg); |
f712c0c7 SS |
6742 | init_numa_sched_groups_power(sg); |
6743 | } | |
9c1cfda2 JH |
6744 | #endif |
6745 | ||
1da177e4 | 6746 | /* Attach the domains */ |
1a20ff27 | 6747 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6748 | struct sched_domain *sd; |
6749 | #ifdef CONFIG_SCHED_SMT | |
6750 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
6751 | #elif defined(CONFIG_SCHED_MC) |
6752 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
6753 | #else |
6754 | sd = &per_cpu(phys_domains, i); | |
6755 | #endif | |
57d885fe | 6756 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 6757 | } |
51888ca2 SV |
6758 | |
6759 | return 0; | |
6760 | ||
a616058b | 6761 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6762 | error: |
6763 | free_sched_groups(cpu_map); | |
6764 | return -ENOMEM; | |
a616058b | 6765 | #endif |
1da177e4 | 6766 | } |
029190c5 PJ |
6767 | |
6768 | static cpumask_t *doms_cur; /* current sched domains */ | |
6769 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
6770 | ||
6771 | /* | |
6772 | * Special case: If a kmalloc of a doms_cur partition (array of | |
6773 | * cpumask_t) fails, then fallback to a single sched domain, | |
6774 | * as determined by the single cpumask_t fallback_doms. | |
6775 | */ | |
6776 | static cpumask_t fallback_doms; | |
6777 | ||
1a20ff27 | 6778 | /* |
41a2d6cf | 6779 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6780 | * For now this just excludes isolated cpus, but could be used to |
6781 | * exclude other special cases in the future. | |
1a20ff27 | 6782 | */ |
51888ca2 | 6783 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 6784 | { |
7378547f MM |
6785 | int err; |
6786 | ||
029190c5 PJ |
6787 | ndoms_cur = 1; |
6788 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
6789 | if (!doms_cur) | |
6790 | doms_cur = &fallback_doms; | |
6791 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
7378547f | 6792 | err = build_sched_domains(doms_cur); |
6382bc90 | 6793 | register_sched_domain_sysctl(); |
7378547f MM |
6794 | |
6795 | return err; | |
1a20ff27 DG |
6796 | } |
6797 | ||
6798 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map) | |
1da177e4 | 6799 | { |
51888ca2 | 6800 | free_sched_groups(cpu_map); |
9c1cfda2 | 6801 | } |
1da177e4 | 6802 | |
1a20ff27 DG |
6803 | /* |
6804 | * Detach sched domains from a group of cpus specified in cpu_map | |
6805 | * These cpus will now be attached to the NULL domain | |
6806 | */ | |
858119e1 | 6807 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6808 | { |
6809 | int i; | |
6810 | ||
6382bc90 MM |
6811 | unregister_sched_domain_sysctl(); |
6812 | ||
1a20ff27 | 6813 | for_each_cpu_mask(i, *cpu_map) |
57d885fe | 6814 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 DG |
6815 | synchronize_sched(); |
6816 | arch_destroy_sched_domains(cpu_map); | |
6817 | } | |
6818 | ||
029190c5 PJ |
6819 | /* |
6820 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6821 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6822 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6823 | * It destroys each deleted domain and builds each new domain. | |
6824 | * | |
6825 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
41a2d6cf IM |
6826 | * The masks don't intersect (don't overlap.) We should setup one |
6827 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6828 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6829 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6830 | * it as it is. | |
6831 | * | |
41a2d6cf IM |
6832 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
6833 | * ownership of it and will kfree it when done with it. If the caller | |
029190c5 PJ |
6834 | * failed the kmalloc call, then it can pass in doms_new == NULL, |
6835 | * and partition_sched_domains() will fallback to the single partition | |
6836 | * 'fallback_doms'. | |
6837 | * | |
6838 | * Call with hotplug lock held | |
6839 | */ | |
6840 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new) | |
6841 | { | |
6842 | int i, j; | |
6843 | ||
a1835615 SV |
6844 | lock_doms_cur(); |
6845 | ||
7378547f MM |
6846 | /* always unregister in case we don't destroy any domains */ |
6847 | unregister_sched_domain_sysctl(); | |
6848 | ||
029190c5 PJ |
6849 | if (doms_new == NULL) { |
6850 | ndoms_new = 1; | |
6851 | doms_new = &fallback_doms; | |
6852 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
6853 | } | |
6854 | ||
6855 | /* Destroy deleted domains */ | |
6856 | for (i = 0; i < ndoms_cur; i++) { | |
6857 | for (j = 0; j < ndoms_new; j++) { | |
6858 | if (cpus_equal(doms_cur[i], doms_new[j])) | |
6859 | goto match1; | |
6860 | } | |
6861 | /* no match - a current sched domain not in new doms_new[] */ | |
6862 | detach_destroy_domains(doms_cur + i); | |
6863 | match1: | |
6864 | ; | |
6865 | } | |
6866 | ||
6867 | /* Build new domains */ | |
6868 | for (i = 0; i < ndoms_new; i++) { | |
6869 | for (j = 0; j < ndoms_cur; j++) { | |
6870 | if (cpus_equal(doms_new[i], doms_cur[j])) | |
6871 | goto match2; | |
6872 | } | |
6873 | /* no match - add a new doms_new */ | |
6874 | build_sched_domains(doms_new + i); | |
6875 | match2: | |
6876 | ; | |
6877 | } | |
6878 | ||
6879 | /* Remember the new sched domains */ | |
6880 | if (doms_cur != &fallback_doms) | |
6881 | kfree(doms_cur); | |
6882 | doms_cur = doms_new; | |
6883 | ndoms_cur = ndoms_new; | |
7378547f MM |
6884 | |
6885 | register_sched_domain_sysctl(); | |
a1835615 SV |
6886 | |
6887 | unlock_doms_cur(); | |
029190c5 PJ |
6888 | } |
6889 | ||
5c45bf27 | 6890 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
6707de00 | 6891 | static int arch_reinit_sched_domains(void) |
5c45bf27 SS |
6892 | { |
6893 | int err; | |
6894 | ||
95402b38 | 6895 | get_online_cpus(); |
5c45bf27 SS |
6896 | detach_destroy_domains(&cpu_online_map); |
6897 | err = arch_init_sched_domains(&cpu_online_map); | |
95402b38 | 6898 | put_online_cpus(); |
5c45bf27 SS |
6899 | |
6900 | return err; | |
6901 | } | |
6902 | ||
6903 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6904 | { | |
6905 | int ret; | |
6906 | ||
6907 | if (buf[0] != '0' && buf[0] != '1') | |
6908 | return -EINVAL; | |
6909 | ||
6910 | if (smt) | |
6911 | sched_smt_power_savings = (buf[0] == '1'); | |
6912 | else | |
6913 | sched_mc_power_savings = (buf[0] == '1'); | |
6914 | ||
6915 | ret = arch_reinit_sched_domains(); | |
6916 | ||
6917 | return ret ? ret : count; | |
6918 | } | |
6919 | ||
5c45bf27 SS |
6920 | #ifdef CONFIG_SCHED_MC |
6921 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
6922 | { | |
6923 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
6924 | } | |
48f24c4d IM |
6925 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
6926 | const char *buf, size_t count) | |
5c45bf27 SS |
6927 | { |
6928 | return sched_power_savings_store(buf, count, 0); | |
6929 | } | |
6707de00 AB |
6930 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
6931 | sched_mc_power_savings_store); | |
5c45bf27 SS |
6932 | #endif |
6933 | ||
6934 | #ifdef CONFIG_SCHED_SMT | |
6935 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
6936 | { | |
6937 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
6938 | } | |
48f24c4d IM |
6939 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
6940 | const char *buf, size_t count) | |
5c45bf27 SS |
6941 | { |
6942 | return sched_power_savings_store(buf, count, 1); | |
6943 | } | |
6707de00 AB |
6944 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
6945 | sched_smt_power_savings_store); | |
6946 | #endif | |
6947 | ||
6948 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
6949 | { | |
6950 | int err = 0; | |
6951 | ||
6952 | #ifdef CONFIG_SCHED_SMT | |
6953 | if (smt_capable()) | |
6954 | err = sysfs_create_file(&cls->kset.kobj, | |
6955 | &attr_sched_smt_power_savings.attr); | |
6956 | #endif | |
6957 | #ifdef CONFIG_SCHED_MC | |
6958 | if (!err && mc_capable()) | |
6959 | err = sysfs_create_file(&cls->kset.kobj, | |
6960 | &attr_sched_mc_power_savings.attr); | |
6961 | #endif | |
6962 | return err; | |
6963 | } | |
5c45bf27 SS |
6964 | #endif |
6965 | ||
1da177e4 | 6966 | /* |
41a2d6cf | 6967 | * Force a reinitialization of the sched domains hierarchy. The domains |
1da177e4 | 6968 | * and groups cannot be updated in place without racing with the balancing |
41c7ce9a | 6969 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
6970 | * which will prevent rebalancing while the sched domains are recalculated. |
6971 | */ | |
6972 | static int update_sched_domains(struct notifier_block *nfb, | |
6973 | unsigned long action, void *hcpu) | |
6974 | { | |
1da177e4 LT |
6975 | switch (action) { |
6976 | case CPU_UP_PREPARE: | |
8bb78442 | 6977 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 6978 | case CPU_DOWN_PREPARE: |
8bb78442 | 6979 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 6980 | detach_destroy_domains(&cpu_online_map); |
1da177e4 LT |
6981 | return NOTIFY_OK; |
6982 | ||
6983 | case CPU_UP_CANCELED: | |
8bb78442 | 6984 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 6985 | case CPU_DOWN_FAILED: |
8bb78442 | 6986 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 6987 | case CPU_ONLINE: |
8bb78442 | 6988 | case CPU_ONLINE_FROZEN: |
1da177e4 | 6989 | case CPU_DEAD: |
8bb78442 | 6990 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
6991 | /* |
6992 | * Fall through and re-initialise the domains. | |
6993 | */ | |
6994 | break; | |
6995 | default: | |
6996 | return NOTIFY_DONE; | |
6997 | } | |
6998 | ||
6999 | /* The hotplug lock is already held by cpu_up/cpu_down */ | |
1a20ff27 | 7000 | arch_init_sched_domains(&cpu_online_map); |
1da177e4 LT |
7001 | |
7002 | return NOTIFY_OK; | |
7003 | } | |
1da177e4 LT |
7004 | |
7005 | void __init sched_init_smp(void) | |
7006 | { | |
5c1e1767 NP |
7007 | cpumask_t non_isolated_cpus; |
7008 | ||
95402b38 | 7009 | get_online_cpus(); |
1a20ff27 | 7010 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 7011 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
7012 | if (cpus_empty(non_isolated_cpus)) |
7013 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
95402b38 | 7014 | put_online_cpus(); |
1da177e4 LT |
7015 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7016 | hotcpu_notifier(update_sched_domains, 0); | |
5c1e1767 NP |
7017 | |
7018 | /* Move init over to a non-isolated CPU */ | |
7019 | if (set_cpus_allowed(current, non_isolated_cpus) < 0) | |
7020 | BUG(); | |
19978ca6 | 7021 | sched_init_granularity(); |
6b2d7700 SV |
7022 | |
7023 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7024 | if (nr_cpu_ids == 1) | |
7025 | return; | |
7026 | ||
7027 | lb_monitor_task = kthread_create(load_balance_monitor, NULL, | |
7028 | "group_balance"); | |
7029 | if (!IS_ERR(lb_monitor_task)) { | |
7030 | lb_monitor_task->flags |= PF_NOFREEZE; | |
7031 | wake_up_process(lb_monitor_task); | |
7032 | } else { | |
7033 | printk(KERN_ERR "Could not create load balance monitor thread" | |
7034 | "(error = %ld) \n", PTR_ERR(lb_monitor_task)); | |
7035 | } | |
7036 | #endif | |
1da177e4 LT |
7037 | } |
7038 | #else | |
7039 | void __init sched_init_smp(void) | |
7040 | { | |
19978ca6 | 7041 | sched_init_granularity(); |
1da177e4 LT |
7042 | } |
7043 | #endif /* CONFIG_SMP */ | |
7044 | ||
7045 | int in_sched_functions(unsigned long addr) | |
7046 | { | |
1da177e4 LT |
7047 | return in_lock_functions(addr) || |
7048 | (addr >= (unsigned long)__sched_text_start | |
7049 | && addr < (unsigned long)__sched_text_end); | |
7050 | } | |
7051 | ||
a9957449 | 7052 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7053 | { |
7054 | cfs_rq->tasks_timeline = RB_ROOT; | |
dd41f596 IM |
7055 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7056 | cfs_rq->rq = rq; | |
7057 | #endif | |
67e9fb2a | 7058 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7059 | } |
7060 | ||
fa85ae24 PZ |
7061 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7062 | { | |
7063 | struct rt_prio_array *array; | |
7064 | int i; | |
7065 | ||
7066 | array = &rt_rq->active; | |
7067 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7068 | INIT_LIST_HEAD(array->queue + i); | |
7069 | __clear_bit(i, array->bitmap); | |
7070 | } | |
7071 | /* delimiter for bitsearch: */ | |
7072 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7073 | ||
7074 | #ifdef CONFIG_SMP | |
7075 | rt_rq->rt_nr_migratory = 0; | |
7076 | rt_rq->highest_prio = MAX_RT_PRIO; | |
7077 | rt_rq->overloaded = 0; | |
7078 | #endif | |
7079 | ||
7080 | rt_rq->rt_time = 0; | |
7081 | rt_rq->rt_throttled = 0; | |
7082 | } | |
7083 | ||
1da177e4 LT |
7084 | void __init sched_init(void) |
7085 | { | |
476f3534 | 7086 | int highest_cpu = 0; |
dd41f596 IM |
7087 | int i, j; |
7088 | ||
57d885fe GH |
7089 | #ifdef CONFIG_SMP |
7090 | init_defrootdomain(); | |
7091 | #endif | |
7092 | ||
0a945022 | 7093 | for_each_possible_cpu(i) { |
70b97a7f | 7094 | struct rq *rq; |
1da177e4 LT |
7095 | |
7096 | rq = cpu_rq(i); | |
7097 | spin_lock_init(&rq->lock); | |
fcb99371 | 7098 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 7099 | rq->nr_running = 0; |
dd41f596 IM |
7100 | rq->clock = 1; |
7101 | init_cfs_rq(&rq->cfs, rq); | |
7102 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7103 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); | |
3a252015 IM |
7104 | { |
7105 | struct cfs_rq *cfs_rq = &per_cpu(init_cfs_rq, i); | |
7106 | struct sched_entity *se = | |
7107 | &per_cpu(init_sched_entity, i); | |
7108 | ||
7109 | init_cfs_rq_p[i] = cfs_rq; | |
7110 | init_cfs_rq(cfs_rq, rq); | |
4cf86d77 | 7111 | cfs_rq->tg = &init_task_group; |
3a252015 | 7112 | list_add(&cfs_rq->leaf_cfs_rq_list, |
29f59db3 SV |
7113 | &rq->leaf_cfs_rq_list); |
7114 | ||
3a252015 IM |
7115 | init_sched_entity_p[i] = se; |
7116 | se->cfs_rq = &rq->cfs; | |
7117 | se->my_q = cfs_rq; | |
4cf86d77 | 7118 | se->load.weight = init_task_group_load; |
9b5b7751 | 7119 | se->load.inv_weight = |
4cf86d77 | 7120 | div64_64(1ULL<<32, init_task_group_load); |
3a252015 IM |
7121 | se->parent = NULL; |
7122 | } | |
4cf86d77 | 7123 | init_task_group.shares = init_task_group_load; |
dd41f596 | 7124 | #endif |
fa85ae24 PZ |
7125 | init_rt_rq(&rq->rt, rq); |
7126 | rq->rt_period_expire = 0; | |
1da177e4 | 7127 | |
dd41f596 IM |
7128 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7129 | rq->cpu_load[j] = 0; | |
1da177e4 | 7130 | #ifdef CONFIG_SMP |
41c7ce9a | 7131 | rq->sd = NULL; |
57d885fe | 7132 | rq->rd = NULL; |
1da177e4 | 7133 | rq->active_balance = 0; |
dd41f596 | 7134 | rq->next_balance = jiffies; |
1da177e4 | 7135 | rq->push_cpu = 0; |
0a2966b4 | 7136 | rq->cpu = i; |
1da177e4 LT |
7137 | rq->migration_thread = NULL; |
7138 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 7139 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 7140 | #endif |
8f4d37ec | 7141 | init_rq_hrtick(rq); |
1da177e4 | 7142 | atomic_set(&rq->nr_iowait, 0); |
476f3534 | 7143 | highest_cpu = i; |
1da177e4 LT |
7144 | } |
7145 | ||
2dd73a4f | 7146 | set_load_weight(&init_task); |
b50f60ce | 7147 | |
e107be36 AK |
7148 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7149 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7150 | #endif | |
7151 | ||
c9819f45 | 7152 | #ifdef CONFIG_SMP |
476f3534 | 7153 | nr_cpu_ids = highest_cpu + 1; |
c9819f45 CL |
7154 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); |
7155 | #endif | |
7156 | ||
b50f60ce HC |
7157 | #ifdef CONFIG_RT_MUTEXES |
7158 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
7159 | #endif | |
7160 | ||
1da177e4 LT |
7161 | /* |
7162 | * The boot idle thread does lazy MMU switching as well: | |
7163 | */ | |
7164 | atomic_inc(&init_mm.mm_count); | |
7165 | enter_lazy_tlb(&init_mm, current); | |
7166 | ||
7167 | /* | |
7168 | * Make us the idle thread. Technically, schedule() should not be | |
7169 | * called from this thread, however somewhere below it might be, | |
7170 | * but because we are the idle thread, we just pick up running again | |
7171 | * when this runqueue becomes "idle". | |
7172 | */ | |
7173 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
7174 | /* |
7175 | * During early bootup we pretend to be a normal task: | |
7176 | */ | |
7177 | current->sched_class = &fair_sched_class; | |
1da177e4 LT |
7178 | } |
7179 | ||
7180 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
7181 | void __might_sleep(char *file, int line) | |
7182 | { | |
48f24c4d | 7183 | #ifdef in_atomic |
1da177e4 LT |
7184 | static unsigned long prev_jiffy; /* ratelimiting */ |
7185 | ||
7186 | if ((in_atomic() || irqs_disabled()) && | |
7187 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
7188 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7189 | return; | |
7190 | prev_jiffy = jiffies; | |
91368d73 | 7191 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
7192 | " context at %s:%d\n", file, line); |
7193 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
7194 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 7195 | debug_show_held_locks(current); |
3117df04 IM |
7196 | if (irqs_disabled()) |
7197 | print_irqtrace_events(current); | |
1da177e4 LT |
7198 | dump_stack(); |
7199 | } | |
7200 | #endif | |
7201 | } | |
7202 | EXPORT_SYMBOL(__might_sleep); | |
7203 | #endif | |
7204 | ||
7205 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7206 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7207 | { | |
7208 | int on_rq; | |
7209 | update_rq_clock(rq); | |
7210 | on_rq = p->se.on_rq; | |
7211 | if (on_rq) | |
7212 | deactivate_task(rq, p, 0); | |
7213 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
7214 | if (on_rq) { | |
7215 | activate_task(rq, p, 0); | |
7216 | resched_task(rq->curr); | |
7217 | } | |
7218 | } | |
7219 | ||
1da177e4 LT |
7220 | void normalize_rt_tasks(void) |
7221 | { | |
a0f98a1c | 7222 | struct task_struct *g, *p; |
1da177e4 | 7223 | unsigned long flags; |
70b97a7f | 7224 | struct rq *rq; |
1da177e4 LT |
7225 | |
7226 | read_lock_irq(&tasklist_lock); | |
a0f98a1c | 7227 | do_each_thread(g, p) { |
178be793 IM |
7228 | /* |
7229 | * Only normalize user tasks: | |
7230 | */ | |
7231 | if (!p->mm) | |
7232 | continue; | |
7233 | ||
6cfb0d5d | 7234 | p->se.exec_start = 0; |
6cfb0d5d | 7235 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 7236 | p->se.wait_start = 0; |
dd41f596 | 7237 | p->se.sleep_start = 0; |
dd41f596 | 7238 | p->se.block_start = 0; |
6cfb0d5d | 7239 | #endif |
dd41f596 IM |
7240 | task_rq(p)->clock = 0; |
7241 | ||
7242 | if (!rt_task(p)) { | |
7243 | /* | |
7244 | * Renice negative nice level userspace | |
7245 | * tasks back to 0: | |
7246 | */ | |
7247 | if (TASK_NICE(p) < 0 && p->mm) | |
7248 | set_user_nice(p, 0); | |
1da177e4 | 7249 | continue; |
dd41f596 | 7250 | } |
1da177e4 | 7251 | |
b29739f9 IM |
7252 | spin_lock_irqsave(&p->pi_lock, flags); |
7253 | rq = __task_rq_lock(p); | |
1da177e4 | 7254 | |
178be793 | 7255 | normalize_task(rq, p); |
3a5e4dc1 | 7256 | |
b29739f9 IM |
7257 | __task_rq_unlock(rq); |
7258 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
a0f98a1c IM |
7259 | } while_each_thread(g, p); |
7260 | ||
1da177e4 LT |
7261 | read_unlock_irq(&tasklist_lock); |
7262 | } | |
7263 | ||
7264 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
7265 | |
7266 | #ifdef CONFIG_IA64 | |
7267 | /* | |
7268 | * These functions are only useful for the IA64 MCA handling. | |
7269 | * | |
7270 | * They can only be called when the whole system has been | |
7271 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7272 | * activity can take place. Using them for anything else would | |
7273 | * be a serious bug, and as a result, they aren't even visible | |
7274 | * under any other configuration. | |
7275 | */ | |
7276 | ||
7277 | /** | |
7278 | * curr_task - return the current task for a given cpu. | |
7279 | * @cpu: the processor in question. | |
7280 | * | |
7281 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7282 | */ | |
36c8b586 | 7283 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7284 | { |
7285 | return cpu_curr(cpu); | |
7286 | } | |
7287 | ||
7288 | /** | |
7289 | * set_curr_task - set the current task for a given cpu. | |
7290 | * @cpu: the processor in question. | |
7291 | * @p: the task pointer to set. | |
7292 | * | |
7293 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7294 | * are serviced on a separate stack. It allows the architecture to switch the |
7295 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7296 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7297 | * and caller must save the original value of the current task (see | |
7298 | * curr_task() above) and restore that value before reenabling interrupts and | |
7299 | * re-starting the system. | |
7300 | * | |
7301 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7302 | */ | |
36c8b586 | 7303 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7304 | { |
7305 | cpu_curr(cpu) = p; | |
7306 | } | |
7307 | ||
7308 | #endif | |
29f59db3 SV |
7309 | |
7310 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7311 | ||
6b2d7700 SV |
7312 | #ifdef CONFIG_SMP |
7313 | /* | |
7314 | * distribute shares of all task groups among their schedulable entities, | |
fa85ae24 | 7315 | * to reflect load distribution across cpus. |
6b2d7700 SV |
7316 | */ |
7317 | static int rebalance_shares(struct sched_domain *sd, int this_cpu) | |
7318 | { | |
7319 | struct cfs_rq *cfs_rq; | |
7320 | struct rq *rq = cpu_rq(this_cpu); | |
7321 | cpumask_t sdspan = sd->span; | |
7322 | int balanced = 1; | |
7323 | ||
7324 | /* Walk thr' all the task groups that we have */ | |
7325 | for_each_leaf_cfs_rq(rq, cfs_rq) { | |
7326 | int i; | |
7327 | unsigned long total_load = 0, total_shares; | |
7328 | struct task_group *tg = cfs_rq->tg; | |
7329 | ||
7330 | /* Gather total task load of this group across cpus */ | |
7331 | for_each_cpu_mask(i, sdspan) | |
7332 | total_load += tg->cfs_rq[i]->load.weight; | |
7333 | ||
0eab9146 | 7334 | /* Nothing to do if this group has no load */ |
6b2d7700 SV |
7335 | if (!total_load) |
7336 | continue; | |
7337 | ||
7338 | /* | |
7339 | * tg->shares represents the number of cpu shares the task group | |
7340 | * is eligible to hold on a single cpu. On N cpus, it is | |
7341 | * eligible to hold (N * tg->shares) number of cpu shares. | |
7342 | */ | |
7343 | total_shares = tg->shares * cpus_weight(sdspan); | |
7344 | ||
7345 | /* | |
7346 | * redistribute total_shares across cpus as per the task load | |
7347 | * distribution. | |
7348 | */ | |
7349 | for_each_cpu_mask(i, sdspan) { | |
7350 | unsigned long local_load, local_shares; | |
7351 | ||
7352 | local_load = tg->cfs_rq[i]->load.weight; | |
7353 | local_shares = (local_load * total_shares) / total_load; | |
7354 | if (!local_shares) | |
7355 | local_shares = MIN_GROUP_SHARES; | |
7356 | if (local_shares == tg->se[i]->load.weight) | |
7357 | continue; | |
7358 | ||
7359 | spin_lock_irq(&cpu_rq(i)->lock); | |
7360 | set_se_shares(tg->se[i], local_shares); | |
7361 | spin_unlock_irq(&cpu_rq(i)->lock); | |
7362 | balanced = 0; | |
7363 | } | |
7364 | } | |
7365 | ||
7366 | return balanced; | |
7367 | } | |
7368 | ||
7369 | /* | |
7370 | * How frequently should we rebalance_shares() across cpus? | |
7371 | * | |
7372 | * The more frequently we rebalance shares, the more accurate is the fairness | |
7373 | * of cpu bandwidth distribution between task groups. However higher frequency | |
7374 | * also implies increased scheduling overhead. | |
7375 | * | |
7376 | * sysctl_sched_min_bal_int_shares represents the minimum interval between | |
7377 | * consecutive calls to rebalance_shares() in the same sched domain. | |
7378 | * | |
7379 | * sysctl_sched_max_bal_int_shares represents the maximum interval between | |
7380 | * consecutive calls to rebalance_shares() in the same sched domain. | |
7381 | * | |
fa85ae24 | 7382 | * These settings allows for the appropriate trade-off between accuracy of |
6b2d7700 SV |
7383 | * fairness and the associated overhead. |
7384 | * | |
7385 | */ | |
7386 | ||
7387 | /* default: 8ms, units: milliseconds */ | |
7388 | const_debug unsigned int sysctl_sched_min_bal_int_shares = 8; | |
7389 | ||
7390 | /* default: 128ms, units: milliseconds */ | |
7391 | const_debug unsigned int sysctl_sched_max_bal_int_shares = 128; | |
7392 | ||
7393 | /* kernel thread that runs rebalance_shares() periodically */ | |
7394 | static int load_balance_monitor(void *unused) | |
7395 | { | |
7396 | unsigned int timeout = sysctl_sched_min_bal_int_shares; | |
7397 | struct sched_param schedparm; | |
7398 | int ret; | |
7399 | ||
7400 | /* | |
7401 | * We don't want this thread's execution to be limited by the shares | |
7402 | * assigned to default group (init_task_group). Hence make it run | |
7403 | * as a SCHED_RR RT task at the lowest priority. | |
7404 | */ | |
7405 | schedparm.sched_priority = 1; | |
7406 | ret = sched_setscheduler(current, SCHED_RR, &schedparm); | |
7407 | if (ret) | |
7408 | printk(KERN_ERR "Couldn't set SCHED_RR policy for load balance" | |
7409 | " monitor thread (error = %d) \n", ret); | |
7410 | ||
7411 | while (!kthread_should_stop()) { | |
7412 | int i, cpu, balanced = 1; | |
7413 | ||
7414 | /* Prevent cpus going down or coming up */ | |
86ef5c9a | 7415 | get_online_cpus(); |
6b2d7700 SV |
7416 | /* lockout changes to doms_cur[] array */ |
7417 | lock_doms_cur(); | |
7418 | /* | |
7419 | * Enter a rcu read-side critical section to safely walk rq->sd | |
7420 | * chain on various cpus and to walk task group list | |
7421 | * (rq->leaf_cfs_rq_list) in rebalance_shares(). | |
7422 | */ | |
7423 | rcu_read_lock(); | |
7424 | ||
7425 | for (i = 0; i < ndoms_cur; i++) { | |
7426 | cpumask_t cpumap = doms_cur[i]; | |
7427 | struct sched_domain *sd = NULL, *sd_prev = NULL; | |
7428 | ||
7429 | cpu = first_cpu(cpumap); | |
7430 | ||
7431 | /* Find the highest domain at which to balance shares */ | |
7432 | for_each_domain(cpu, sd) { | |
7433 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
7434 | continue; | |
7435 | sd_prev = sd; | |
7436 | } | |
7437 | ||
7438 | sd = sd_prev; | |
7439 | /* sd == NULL? No load balance reqd in this domain */ | |
7440 | if (!sd) | |
7441 | continue; | |
7442 | ||
7443 | balanced &= rebalance_shares(sd, cpu); | |
7444 | } | |
7445 | ||
7446 | rcu_read_unlock(); | |
7447 | ||
7448 | unlock_doms_cur(); | |
86ef5c9a | 7449 | put_online_cpus(); |
6b2d7700 SV |
7450 | |
7451 | if (!balanced) | |
7452 | timeout = sysctl_sched_min_bal_int_shares; | |
7453 | else if (timeout < sysctl_sched_max_bal_int_shares) | |
7454 | timeout *= 2; | |
7455 | ||
7456 | msleep_interruptible(timeout); | |
7457 | } | |
7458 | ||
7459 | return 0; | |
7460 | } | |
7461 | #endif /* CONFIG_SMP */ | |
7462 | ||
29f59db3 | 7463 | /* allocate runqueue etc for a new task group */ |
4cf86d77 | 7464 | struct task_group *sched_create_group(void) |
29f59db3 | 7465 | { |
4cf86d77 | 7466 | struct task_group *tg; |
29f59db3 SV |
7467 | struct cfs_rq *cfs_rq; |
7468 | struct sched_entity *se; | |
9b5b7751 | 7469 | struct rq *rq; |
29f59db3 SV |
7470 | int i; |
7471 | ||
29f59db3 SV |
7472 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); |
7473 | if (!tg) | |
7474 | return ERR_PTR(-ENOMEM); | |
7475 | ||
9b5b7751 | 7476 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
7477 | if (!tg->cfs_rq) |
7478 | goto err; | |
9b5b7751 | 7479 | tg->se = kzalloc(sizeof(se) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
7480 | if (!tg->se) |
7481 | goto err; | |
7482 | ||
7483 | for_each_possible_cpu(i) { | |
9b5b7751 | 7484 | rq = cpu_rq(i); |
29f59db3 SV |
7485 | |
7486 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), GFP_KERNEL, | |
7487 | cpu_to_node(i)); | |
7488 | if (!cfs_rq) | |
7489 | goto err; | |
7490 | ||
7491 | se = kmalloc_node(sizeof(struct sched_entity), GFP_KERNEL, | |
7492 | cpu_to_node(i)); | |
7493 | if (!se) | |
7494 | goto err; | |
7495 | ||
7496 | memset(cfs_rq, 0, sizeof(struct cfs_rq)); | |
7497 | memset(se, 0, sizeof(struct sched_entity)); | |
7498 | ||
7499 | tg->cfs_rq[i] = cfs_rq; | |
7500 | init_cfs_rq(cfs_rq, rq); | |
7501 | cfs_rq->tg = tg; | |
29f59db3 SV |
7502 | |
7503 | tg->se[i] = se; | |
7504 | se->cfs_rq = &rq->cfs; | |
7505 | se->my_q = cfs_rq; | |
7506 | se->load.weight = NICE_0_LOAD; | |
7507 | se->load.inv_weight = div64_64(1ULL<<32, NICE_0_LOAD); | |
7508 | se->parent = NULL; | |
7509 | } | |
7510 | ||
ec2c507f SV |
7511 | tg->shares = NICE_0_LOAD; |
7512 | ||
7513 | lock_task_group_list(); | |
9b5b7751 SV |
7514 | for_each_possible_cpu(i) { |
7515 | rq = cpu_rq(i); | |
7516 | cfs_rq = tg->cfs_rq[i]; | |
7517 | list_add_rcu(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7518 | } | |
ec2c507f | 7519 | unlock_task_group_list(); |
29f59db3 | 7520 | |
9b5b7751 | 7521 | return tg; |
29f59db3 SV |
7522 | |
7523 | err: | |
7524 | for_each_possible_cpu(i) { | |
a65914b3 | 7525 | if (tg->cfs_rq) |
29f59db3 | 7526 | kfree(tg->cfs_rq[i]); |
a65914b3 | 7527 | if (tg->se) |
29f59db3 SV |
7528 | kfree(tg->se[i]); |
7529 | } | |
a65914b3 IM |
7530 | kfree(tg->cfs_rq); |
7531 | kfree(tg->se); | |
7532 | kfree(tg); | |
29f59db3 SV |
7533 | |
7534 | return ERR_PTR(-ENOMEM); | |
7535 | } | |
7536 | ||
9b5b7751 SV |
7537 | /* rcu callback to free various structures associated with a task group */ |
7538 | static void free_sched_group(struct rcu_head *rhp) | |
29f59db3 | 7539 | { |
ae8393e5 SV |
7540 | struct task_group *tg = container_of(rhp, struct task_group, rcu); |
7541 | struct cfs_rq *cfs_rq; | |
29f59db3 SV |
7542 | struct sched_entity *se; |
7543 | int i; | |
7544 | ||
29f59db3 SV |
7545 | /* now it should be safe to free those cfs_rqs */ |
7546 | for_each_possible_cpu(i) { | |
7547 | cfs_rq = tg->cfs_rq[i]; | |
7548 | kfree(cfs_rq); | |
7549 | ||
7550 | se = tg->se[i]; | |
7551 | kfree(se); | |
7552 | } | |
7553 | ||
7554 | kfree(tg->cfs_rq); | |
7555 | kfree(tg->se); | |
7556 | kfree(tg); | |
7557 | } | |
7558 | ||
9b5b7751 | 7559 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7560 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 7561 | { |
7bae49d4 | 7562 | struct cfs_rq *cfs_rq = NULL; |
9b5b7751 | 7563 | int i; |
29f59db3 | 7564 | |
ec2c507f | 7565 | lock_task_group_list(); |
9b5b7751 SV |
7566 | for_each_possible_cpu(i) { |
7567 | cfs_rq = tg->cfs_rq[i]; | |
7568 | list_del_rcu(&cfs_rq->leaf_cfs_rq_list); | |
7569 | } | |
ec2c507f | 7570 | unlock_task_group_list(); |
9b5b7751 | 7571 | |
7bae49d4 | 7572 | BUG_ON(!cfs_rq); |
9b5b7751 SV |
7573 | |
7574 | /* wait for possible concurrent references to cfs_rqs complete */ | |
ae8393e5 | 7575 | call_rcu(&tg->rcu, free_sched_group); |
29f59db3 SV |
7576 | } |
7577 | ||
9b5b7751 | 7578 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7579 | * The caller of this function should have put the task in its new group |
7580 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7581 | * reflect its new group. | |
9b5b7751 SV |
7582 | */ |
7583 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
7584 | { |
7585 | int on_rq, running; | |
7586 | unsigned long flags; | |
7587 | struct rq *rq; | |
7588 | ||
7589 | rq = task_rq_lock(tsk, &flags); | |
7590 | ||
dae51f56 | 7591 | if (tsk->sched_class != &fair_sched_class) { |
ce96b5ac | 7592 | set_task_cfs_rq(tsk, task_cpu(tsk)); |
29f59db3 | 7593 | goto done; |
dae51f56 | 7594 | } |
29f59db3 SV |
7595 | |
7596 | update_rq_clock(rq); | |
7597 | ||
051a1d1a | 7598 | running = task_current(rq, tsk); |
29f59db3 SV |
7599 | on_rq = tsk->se.on_rq; |
7600 | ||
83b699ed | 7601 | if (on_rq) { |
29f59db3 | 7602 | dequeue_task(rq, tsk, 0); |
83b699ed SV |
7603 | if (unlikely(running)) |
7604 | tsk->sched_class->put_prev_task(rq, tsk); | |
7605 | } | |
29f59db3 | 7606 | |
ce96b5ac | 7607 | set_task_cfs_rq(tsk, task_cpu(tsk)); |
29f59db3 | 7608 | |
83b699ed SV |
7609 | if (on_rq) { |
7610 | if (unlikely(running)) | |
7611 | tsk->sched_class->set_curr_task(rq); | |
7074badb | 7612 | enqueue_task(rq, tsk, 0); |
83b699ed | 7613 | } |
29f59db3 SV |
7614 | |
7615 | done: | |
7616 | task_rq_unlock(rq, &flags); | |
7617 | } | |
7618 | ||
6b2d7700 | 7619 | /* rq->lock to be locked by caller */ |
29f59db3 SV |
7620 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
7621 | { | |
7622 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
7623 | struct rq *rq = cfs_rq->rq; | |
7624 | int on_rq; | |
7625 | ||
6b2d7700 SV |
7626 | if (!shares) |
7627 | shares = MIN_GROUP_SHARES; | |
29f59db3 SV |
7628 | |
7629 | on_rq = se->on_rq; | |
6b2d7700 | 7630 | if (on_rq) { |
29f59db3 | 7631 | dequeue_entity(cfs_rq, se, 0); |
6b2d7700 SV |
7632 | dec_cpu_load(rq, se->load.weight); |
7633 | } | |
29f59db3 SV |
7634 | |
7635 | se->load.weight = shares; | |
7636 | se->load.inv_weight = div64_64((1ULL<<32), shares); | |
7637 | ||
6b2d7700 | 7638 | if (on_rq) { |
29f59db3 | 7639 | enqueue_entity(cfs_rq, se, 0); |
6b2d7700 SV |
7640 | inc_cpu_load(rq, se->load.weight); |
7641 | } | |
29f59db3 SV |
7642 | } |
7643 | ||
4cf86d77 | 7644 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
7645 | { |
7646 | int i; | |
6b2d7700 SV |
7647 | struct cfs_rq *cfs_rq; |
7648 | struct rq *rq; | |
c61935fd | 7649 | |
ec2c507f | 7650 | lock_task_group_list(); |
9b5b7751 | 7651 | if (tg->shares == shares) |
5cb350ba | 7652 | goto done; |
29f59db3 | 7653 | |
6b2d7700 SV |
7654 | if (shares < MIN_GROUP_SHARES) |
7655 | shares = MIN_GROUP_SHARES; | |
7656 | ||
7657 | /* | |
7658 | * Prevent any load balance activity (rebalance_shares, | |
7659 | * load_balance_fair) from referring to this group first, | |
7660 | * by taking it off the rq->leaf_cfs_rq_list on each cpu. | |
7661 | */ | |
7662 | for_each_possible_cpu(i) { | |
7663 | cfs_rq = tg->cfs_rq[i]; | |
7664 | list_del_rcu(&cfs_rq->leaf_cfs_rq_list); | |
7665 | } | |
7666 | ||
7667 | /* wait for any ongoing reference to this group to finish */ | |
7668 | synchronize_sched(); | |
7669 | ||
7670 | /* | |
7671 | * Now we are free to modify the group's share on each cpu | |
7672 | * w/o tripping rebalance_share or load_balance_fair. | |
7673 | */ | |
9b5b7751 | 7674 | tg->shares = shares; |
6b2d7700 SV |
7675 | for_each_possible_cpu(i) { |
7676 | spin_lock_irq(&cpu_rq(i)->lock); | |
9b5b7751 | 7677 | set_se_shares(tg->se[i], shares); |
6b2d7700 SV |
7678 | spin_unlock_irq(&cpu_rq(i)->lock); |
7679 | } | |
29f59db3 | 7680 | |
6b2d7700 SV |
7681 | /* |
7682 | * Enable load balance activity on this group, by inserting it back on | |
7683 | * each cpu's rq->leaf_cfs_rq_list. | |
7684 | */ | |
7685 | for_each_possible_cpu(i) { | |
7686 | rq = cpu_rq(i); | |
7687 | cfs_rq = tg->cfs_rq[i]; | |
7688 | list_add_rcu(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7689 | } | |
5cb350ba | 7690 | done: |
ec2c507f | 7691 | unlock_task_group_list(); |
9b5b7751 | 7692 | return 0; |
29f59db3 SV |
7693 | } |
7694 | ||
5cb350ba DG |
7695 | unsigned long sched_group_shares(struct task_group *tg) |
7696 | { | |
7697 | return tg->shares; | |
7698 | } | |
7699 | ||
3a252015 | 7700 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e SV |
7701 | |
7702 | #ifdef CONFIG_FAIR_CGROUP_SCHED | |
7703 | ||
7704 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 7705 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 7706 | { |
2b01dfe3 PM |
7707 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
7708 | struct task_group, css); | |
68318b8e SV |
7709 | } |
7710 | ||
7711 | static struct cgroup_subsys_state * | |
2b01dfe3 | 7712 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e SV |
7713 | { |
7714 | struct task_group *tg; | |
7715 | ||
2b01dfe3 | 7716 | if (!cgrp->parent) { |
68318b8e | 7717 | /* This is early initialization for the top cgroup */ |
2b01dfe3 | 7718 | init_task_group.css.cgroup = cgrp; |
68318b8e SV |
7719 | return &init_task_group.css; |
7720 | } | |
7721 | ||
7722 | /* we support only 1-level deep hierarchical scheduler atm */ | |
2b01dfe3 | 7723 | if (cgrp->parent->parent) |
68318b8e SV |
7724 | return ERR_PTR(-EINVAL); |
7725 | ||
7726 | tg = sched_create_group(); | |
7727 | if (IS_ERR(tg)) | |
7728 | return ERR_PTR(-ENOMEM); | |
7729 | ||
7730 | /* Bind the cgroup to task_group object we just created */ | |
2b01dfe3 | 7731 | tg->css.cgroup = cgrp; |
68318b8e SV |
7732 | |
7733 | return &tg->css; | |
7734 | } | |
7735 | ||
41a2d6cf IM |
7736 | static void |
7737 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 7738 | { |
2b01dfe3 | 7739 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7740 | |
7741 | sched_destroy_group(tg); | |
7742 | } | |
7743 | ||
41a2d6cf IM |
7744 | static int |
7745 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
7746 | struct task_struct *tsk) | |
68318b8e SV |
7747 | { |
7748 | /* We don't support RT-tasks being in separate groups */ | |
7749 | if (tsk->sched_class != &fair_sched_class) | |
7750 | return -EINVAL; | |
7751 | ||
7752 | return 0; | |
7753 | } | |
7754 | ||
7755 | static void | |
2b01dfe3 | 7756 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
7757 | struct cgroup *old_cont, struct task_struct *tsk) |
7758 | { | |
7759 | sched_move_task(tsk); | |
7760 | } | |
7761 | ||
2b01dfe3 PM |
7762 | static int cpu_shares_write_uint(struct cgroup *cgrp, struct cftype *cftype, |
7763 | u64 shareval) | |
68318b8e | 7764 | { |
2b01dfe3 | 7765 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
7766 | } |
7767 | ||
2b01dfe3 | 7768 | static u64 cpu_shares_read_uint(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 7769 | { |
2b01dfe3 | 7770 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7771 | |
7772 | return (u64) tg->shares; | |
7773 | } | |
7774 | ||
fe5c7cc2 PM |
7775 | static struct cftype cpu_files[] = { |
7776 | { | |
7777 | .name = "shares", | |
7778 | .read_uint = cpu_shares_read_uint, | |
7779 | .write_uint = cpu_shares_write_uint, | |
7780 | }, | |
68318b8e SV |
7781 | }; |
7782 | ||
7783 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
7784 | { | |
fe5c7cc2 | 7785 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
7786 | } |
7787 | ||
7788 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
7789 | .name = "cpu", |
7790 | .create = cpu_cgroup_create, | |
7791 | .destroy = cpu_cgroup_destroy, | |
7792 | .can_attach = cpu_cgroup_can_attach, | |
7793 | .attach = cpu_cgroup_attach, | |
7794 | .populate = cpu_cgroup_populate, | |
7795 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
7796 | .early_init = 1, |
7797 | }; | |
7798 | ||
7799 | #endif /* CONFIG_FAIR_CGROUP_SCHED */ | |
d842de87 SV |
7800 | |
7801 | #ifdef CONFIG_CGROUP_CPUACCT | |
7802 | ||
7803 | /* | |
7804 | * CPU accounting code for task groups. | |
7805 | * | |
7806 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
7807 | * (balbir@in.ibm.com). | |
7808 | */ | |
7809 | ||
7810 | /* track cpu usage of a group of tasks */ | |
7811 | struct cpuacct { | |
7812 | struct cgroup_subsys_state css; | |
7813 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
7814 | u64 *cpuusage; | |
7815 | }; | |
7816 | ||
7817 | struct cgroup_subsys cpuacct_subsys; | |
7818 | ||
7819 | /* return cpu accounting group corresponding to this container */ | |
7820 | static inline struct cpuacct *cgroup_ca(struct cgroup *cont) | |
7821 | { | |
7822 | return container_of(cgroup_subsys_state(cont, cpuacct_subsys_id), | |
7823 | struct cpuacct, css); | |
7824 | } | |
7825 | ||
7826 | /* return cpu accounting group to which this task belongs */ | |
7827 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
7828 | { | |
7829 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
7830 | struct cpuacct, css); | |
7831 | } | |
7832 | ||
7833 | /* create a new cpu accounting group */ | |
7834 | static struct cgroup_subsys_state *cpuacct_create( | |
7835 | struct cgroup_subsys *ss, struct cgroup *cont) | |
7836 | { | |
7837 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
7838 | ||
7839 | if (!ca) | |
7840 | return ERR_PTR(-ENOMEM); | |
7841 | ||
7842 | ca->cpuusage = alloc_percpu(u64); | |
7843 | if (!ca->cpuusage) { | |
7844 | kfree(ca); | |
7845 | return ERR_PTR(-ENOMEM); | |
7846 | } | |
7847 | ||
7848 | return &ca->css; | |
7849 | } | |
7850 | ||
7851 | /* destroy an existing cpu accounting group */ | |
41a2d6cf IM |
7852 | static void |
7853 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cont) | |
d842de87 SV |
7854 | { |
7855 | struct cpuacct *ca = cgroup_ca(cont); | |
7856 | ||
7857 | free_percpu(ca->cpuusage); | |
7858 | kfree(ca); | |
7859 | } | |
7860 | ||
7861 | /* return total cpu usage (in nanoseconds) of a group */ | |
7862 | static u64 cpuusage_read(struct cgroup *cont, struct cftype *cft) | |
7863 | { | |
7864 | struct cpuacct *ca = cgroup_ca(cont); | |
7865 | u64 totalcpuusage = 0; | |
7866 | int i; | |
7867 | ||
7868 | for_each_possible_cpu(i) { | |
7869 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
7870 | ||
7871 | /* | |
7872 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
7873 | * platforms. | |
7874 | */ | |
7875 | spin_lock_irq(&cpu_rq(i)->lock); | |
7876 | totalcpuusage += *cpuusage; | |
7877 | spin_unlock_irq(&cpu_rq(i)->lock); | |
7878 | } | |
7879 | ||
7880 | return totalcpuusage; | |
7881 | } | |
7882 | ||
7883 | static struct cftype files[] = { | |
7884 | { | |
7885 | .name = "usage", | |
7886 | .read_uint = cpuusage_read, | |
7887 | }, | |
7888 | }; | |
7889 | ||
7890 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
7891 | { | |
7892 | return cgroup_add_files(cont, ss, files, ARRAY_SIZE(files)); | |
7893 | } | |
7894 | ||
7895 | /* | |
7896 | * charge this task's execution time to its accounting group. | |
7897 | * | |
7898 | * called with rq->lock held. | |
7899 | */ | |
7900 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
7901 | { | |
7902 | struct cpuacct *ca; | |
7903 | ||
7904 | if (!cpuacct_subsys.active) | |
7905 | return; | |
7906 | ||
7907 | ca = task_ca(tsk); | |
7908 | if (ca) { | |
7909 | u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk)); | |
7910 | ||
7911 | *cpuusage += cputime; | |
7912 | } | |
7913 | } | |
7914 | ||
7915 | struct cgroup_subsys cpuacct_subsys = { | |
7916 | .name = "cpuacct", | |
7917 | .create = cpuacct_create, | |
7918 | .destroy = cpuacct_destroy, | |
7919 | .populate = cpuacct_populate, | |
7920 | .subsys_id = cpuacct_subsys_id, | |
7921 | }; | |
7922 | #endif /* CONFIG_CGROUP_CPUACCT */ |