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 | |
1da177e4 LT |
25 | */ |
26 | ||
27 | #include <linux/mm.h> | |
28 | #include <linux/module.h> | |
29 | #include <linux/nmi.h> | |
30 | #include <linux/init.h> | |
dff06c15 | 31 | #include <linux/uaccess.h> |
1da177e4 LT |
32 | #include <linux/highmem.h> |
33 | #include <linux/smp_lock.h> | |
34 | #include <asm/mmu_context.h> | |
35 | #include <linux/interrupt.h> | |
c59ede7b | 36 | #include <linux/capability.h> |
1da177e4 LT |
37 | #include <linux/completion.h> |
38 | #include <linux/kernel_stat.h> | |
9a11b49a | 39 | #include <linux/debug_locks.h> |
1da177e4 LT |
40 | #include <linux/security.h> |
41 | #include <linux/notifier.h> | |
42 | #include <linux/profile.h> | |
7dfb7103 | 43 | #include <linux/freezer.h> |
198e2f18 | 44 | #include <linux/vmalloc.h> |
1da177e4 LT |
45 | #include <linux/blkdev.h> |
46 | #include <linux/delay.h> | |
b488893a | 47 | #include <linux/pid_namespace.h> |
1da177e4 LT |
48 | #include <linux/smp.h> |
49 | #include <linux/threads.h> | |
50 | #include <linux/timer.h> | |
51 | #include <linux/rcupdate.h> | |
52 | #include <linux/cpu.h> | |
53 | #include <linux/cpuset.h> | |
54 | #include <linux/percpu.h> | |
62d0df64 | 55 | #include <linux/cpu_acct.h> |
1da177e4 LT |
56 | #include <linux/kthread.h> |
57 | #include <linux/seq_file.h> | |
e692ab53 | 58 | #include <linux/sysctl.h> |
1da177e4 LT |
59 | #include <linux/syscalls.h> |
60 | #include <linux/times.h> | |
8f0ab514 | 61 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 62 | #include <linux/kprobes.h> |
0ff92245 | 63 | #include <linux/delayacct.h> |
5517d86b | 64 | #include <linux/reciprocal_div.h> |
dff06c15 | 65 | #include <linux/unistd.h> |
f5ff8422 | 66 | #include <linux/pagemap.h> |
1da177e4 | 67 | |
5517d86b | 68 | #include <asm/tlb.h> |
838225b4 | 69 | #include <asm/irq_regs.h> |
1da177e4 | 70 | |
b035b6de AD |
71 | /* |
72 | * Scheduler clock - returns current time in nanosec units. | |
73 | * This is default implementation. | |
74 | * Architectures and sub-architectures can override this. | |
75 | */ | |
76 | unsigned long long __attribute__((weak)) sched_clock(void) | |
77 | { | |
78 | return (unsigned long long)jiffies * (1000000000 / HZ); | |
79 | } | |
80 | ||
1da177e4 LT |
81 | /* |
82 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
83 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
84 | * and back. | |
85 | */ | |
86 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
87 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
88 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
89 | ||
90 | /* | |
91 | * 'User priority' is the nice value converted to something we | |
92 | * can work with better when scaling various scheduler parameters, | |
93 | * it's a [ 0 ... 39 ] range. | |
94 | */ | |
95 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
96 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
97 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
98 | ||
99 | /* | |
100 | * Some helpers for converting nanosecond timing to jiffy resolution | |
101 | */ | |
a4ec24b4 | 102 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (1000000000 / HZ)) |
1da177e4 LT |
103 | #define JIFFIES_TO_NS(TIME) ((TIME) * (1000000000 / HZ)) |
104 | ||
6aa645ea IM |
105 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
106 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
107 | ||
1da177e4 LT |
108 | /* |
109 | * These are the 'tuning knobs' of the scheduler: | |
110 | * | |
a4ec24b4 | 111 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
112 | * Timeslices get refilled after they expire. |
113 | */ | |
1da177e4 | 114 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 115 | |
5517d86b ED |
116 | #ifdef CONFIG_SMP |
117 | /* | |
118 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
119 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
120 | */ | |
121 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
122 | { | |
123 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
124 | } | |
125 | ||
126 | /* | |
127 | * Each time a sched group cpu_power is changed, | |
128 | * we must compute its reciprocal value | |
129 | */ | |
130 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
131 | { | |
132 | sg->__cpu_power += val; | |
133 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
134 | } | |
135 | #endif | |
136 | ||
e05606d3 IM |
137 | static inline int rt_policy(int policy) |
138 | { | |
139 | if (unlikely(policy == SCHED_FIFO) || unlikely(policy == SCHED_RR)) | |
140 | return 1; | |
141 | return 0; | |
142 | } | |
143 | ||
144 | static inline int task_has_rt_policy(struct task_struct *p) | |
145 | { | |
146 | return rt_policy(p->policy); | |
147 | } | |
148 | ||
1da177e4 | 149 | /* |
6aa645ea | 150 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 151 | */ |
6aa645ea IM |
152 | struct rt_prio_array { |
153 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
154 | struct list_head queue[MAX_RT_PRIO]; | |
155 | }; | |
156 | ||
29f59db3 SV |
157 | #ifdef CONFIG_FAIR_GROUP_SCHED |
158 | ||
68318b8e SV |
159 | #include <linux/cgroup.h> |
160 | ||
29f59db3 SV |
161 | struct cfs_rq; |
162 | ||
163 | /* task group related information */ | |
4cf86d77 | 164 | struct task_group { |
68318b8e SV |
165 | #ifdef CONFIG_FAIR_CGROUP_SCHED |
166 | struct cgroup_subsys_state css; | |
167 | #endif | |
29f59db3 SV |
168 | /* schedulable entities of this group on each cpu */ |
169 | struct sched_entity **se; | |
170 | /* runqueue "owned" by this group on each cpu */ | |
171 | struct cfs_rq **cfs_rq; | |
172 | unsigned long shares; | |
5cb350ba DG |
173 | /* spinlock to serialize modification to shares */ |
174 | spinlock_t lock; | |
29f59db3 SV |
175 | }; |
176 | ||
177 | /* Default task group's sched entity on each cpu */ | |
178 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
179 | /* Default task group's cfs_rq on each cpu */ | |
180 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
181 | ||
9b5b7751 SV |
182 | static struct sched_entity *init_sched_entity_p[NR_CPUS]; |
183 | static struct cfs_rq *init_cfs_rq_p[NR_CPUS]; | |
29f59db3 SV |
184 | |
185 | /* Default task group. | |
3a252015 | 186 | * Every task in system belong to this group at bootup. |
29f59db3 | 187 | */ |
4cf86d77 | 188 | struct task_group init_task_group = { |
3a252015 IM |
189 | .se = init_sched_entity_p, |
190 | .cfs_rq = init_cfs_rq_p, | |
191 | }; | |
9b5b7751 | 192 | |
24e377a8 | 193 | #ifdef CONFIG_FAIR_USER_SCHED |
3a252015 | 194 | # define INIT_TASK_GRP_LOAD 2*NICE_0_LOAD |
24e377a8 | 195 | #else |
3a252015 | 196 | # define INIT_TASK_GRP_LOAD NICE_0_LOAD |
24e377a8 SV |
197 | #endif |
198 | ||
4cf86d77 | 199 | static int init_task_group_load = INIT_TASK_GRP_LOAD; |
29f59db3 SV |
200 | |
201 | /* return group to which a task belongs */ | |
4cf86d77 | 202 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 203 | { |
4cf86d77 | 204 | struct task_group *tg; |
9b5b7751 | 205 | |
24e377a8 SV |
206 | #ifdef CONFIG_FAIR_USER_SCHED |
207 | tg = p->user->tg; | |
68318b8e SV |
208 | #elif defined(CONFIG_FAIR_CGROUP_SCHED) |
209 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), | |
210 | struct task_group, css); | |
24e377a8 | 211 | #else |
4cf86d77 | 212 | tg = &init_task_group; |
24e377a8 | 213 | #endif |
9b5b7751 SV |
214 | |
215 | return tg; | |
29f59db3 SV |
216 | } |
217 | ||
218 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
219 | static inline void set_task_cfs_rq(struct task_struct *p) | |
220 | { | |
4cf86d77 IM |
221 | p->se.cfs_rq = task_group(p)->cfs_rq[task_cpu(p)]; |
222 | p->se.parent = task_group(p)->se[task_cpu(p)]; | |
29f59db3 SV |
223 | } |
224 | ||
225 | #else | |
226 | ||
227 | static inline void set_task_cfs_rq(struct task_struct *p) { } | |
228 | ||
229 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
230 | ||
6aa645ea IM |
231 | /* CFS-related fields in a runqueue */ |
232 | struct cfs_rq { | |
233 | struct load_weight load; | |
234 | unsigned long nr_running; | |
235 | ||
6aa645ea | 236 | u64 exec_clock; |
e9acbff6 | 237 | u64 min_vruntime; |
6aa645ea IM |
238 | |
239 | struct rb_root tasks_timeline; | |
240 | struct rb_node *rb_leftmost; | |
241 | struct rb_node *rb_load_balance_curr; | |
6aa645ea IM |
242 | /* 'curr' points to currently running entity on this cfs_rq. |
243 | * It is set to NULL otherwise (i.e when none are currently running). | |
244 | */ | |
245 | struct sched_entity *curr; | |
ddc97297 PZ |
246 | |
247 | unsigned long nr_spread_over; | |
248 | ||
62160e3f | 249 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
250 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
251 | ||
252 | /* leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
253 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities | |
254 | * (like users, containers etc.) | |
255 | * | |
256 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
257 | * list is used during load balance. | |
258 | */ | |
259 | struct list_head leaf_cfs_rq_list; /* Better name : task_cfs_rq_list? */ | |
4cf86d77 | 260 | struct task_group *tg; /* group that "owns" this runqueue */ |
9b5b7751 | 261 | struct rcu_head rcu; |
6aa645ea IM |
262 | #endif |
263 | }; | |
1da177e4 | 264 | |
6aa645ea IM |
265 | /* Real-Time classes' related field in a runqueue: */ |
266 | struct rt_rq { | |
267 | struct rt_prio_array active; | |
268 | int rt_load_balance_idx; | |
269 | struct list_head *rt_load_balance_head, *rt_load_balance_curr; | |
270 | }; | |
271 | ||
1da177e4 LT |
272 | /* |
273 | * This is the main, per-CPU runqueue data structure. | |
274 | * | |
275 | * Locking rule: those places that want to lock multiple runqueues | |
276 | * (such as the load balancing or the thread migration code), lock | |
277 | * acquire operations must be ordered by ascending &runqueue. | |
278 | */ | |
70b97a7f | 279 | struct rq { |
d8016491 IM |
280 | /* runqueue lock: */ |
281 | spinlock_t lock; | |
1da177e4 LT |
282 | |
283 | /* | |
284 | * nr_running and cpu_load should be in the same cacheline because | |
285 | * remote CPUs use both these fields when doing load calculation. | |
286 | */ | |
287 | unsigned long nr_running; | |
6aa645ea IM |
288 | #define CPU_LOAD_IDX_MAX 5 |
289 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 290 | unsigned char idle_at_tick; |
46cb4b7c SS |
291 | #ifdef CONFIG_NO_HZ |
292 | unsigned char in_nohz_recently; | |
293 | #endif | |
d8016491 IM |
294 | /* capture load from *all* tasks on this cpu: */ |
295 | struct load_weight load; | |
6aa645ea IM |
296 | unsigned long nr_load_updates; |
297 | u64 nr_switches; | |
298 | ||
299 | struct cfs_rq cfs; | |
300 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
d8016491 IM |
301 | /* list of leaf cfs_rq on this cpu: */ |
302 | struct list_head leaf_cfs_rq_list; | |
1da177e4 | 303 | #endif |
6aa645ea | 304 | struct rt_rq rt; |
1da177e4 LT |
305 | |
306 | /* | |
307 | * This is part of a global counter where only the total sum | |
308 | * over all CPUs matters. A task can increase this counter on | |
309 | * one CPU and if it got migrated afterwards it may decrease | |
310 | * it on another CPU. Always updated under the runqueue lock: | |
311 | */ | |
312 | unsigned long nr_uninterruptible; | |
313 | ||
36c8b586 | 314 | struct task_struct *curr, *idle; |
c9819f45 | 315 | unsigned long next_balance; |
1da177e4 | 316 | struct mm_struct *prev_mm; |
6aa645ea | 317 | |
6aa645ea IM |
318 | u64 clock, prev_clock_raw; |
319 | s64 clock_max_delta; | |
320 | ||
321 | unsigned int clock_warps, clock_overflows; | |
2aa44d05 IM |
322 | u64 idle_clock; |
323 | unsigned int clock_deep_idle_events; | |
529c7726 | 324 | u64 tick_timestamp; |
6aa645ea | 325 | |
1da177e4 LT |
326 | atomic_t nr_iowait; |
327 | ||
328 | #ifdef CONFIG_SMP | |
329 | struct sched_domain *sd; | |
330 | ||
331 | /* For active balancing */ | |
332 | int active_balance; | |
333 | int push_cpu; | |
d8016491 IM |
334 | /* cpu of this runqueue: */ |
335 | int cpu; | |
1da177e4 | 336 | |
36c8b586 | 337 | struct task_struct *migration_thread; |
1da177e4 LT |
338 | struct list_head migration_queue; |
339 | #endif | |
340 | ||
341 | #ifdef CONFIG_SCHEDSTATS | |
342 | /* latency stats */ | |
343 | struct sched_info rq_sched_info; | |
344 | ||
345 | /* sys_sched_yield() stats */ | |
480b9434 KC |
346 | unsigned int yld_exp_empty; |
347 | unsigned int yld_act_empty; | |
348 | unsigned int yld_both_empty; | |
349 | unsigned int yld_count; | |
1da177e4 LT |
350 | |
351 | /* schedule() stats */ | |
480b9434 KC |
352 | unsigned int sched_switch; |
353 | unsigned int sched_count; | |
354 | unsigned int sched_goidle; | |
1da177e4 LT |
355 | |
356 | /* try_to_wake_up() stats */ | |
480b9434 KC |
357 | unsigned int ttwu_count; |
358 | unsigned int ttwu_local; | |
b8efb561 IM |
359 | |
360 | /* BKL stats */ | |
480b9434 | 361 | unsigned int bkl_count; |
1da177e4 | 362 | #endif |
fcb99371 | 363 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
364 | }; |
365 | ||
f34e3b61 | 366 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
5be9361c | 367 | static DEFINE_MUTEX(sched_hotcpu_mutex); |
1da177e4 | 368 | |
dd41f596 IM |
369 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
370 | { | |
371 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
372 | } | |
373 | ||
0a2966b4 CL |
374 | static inline int cpu_of(struct rq *rq) |
375 | { | |
376 | #ifdef CONFIG_SMP | |
377 | return rq->cpu; | |
378 | #else | |
379 | return 0; | |
380 | #endif | |
381 | } | |
382 | ||
20d315d4 | 383 | /* |
b04a0f4c IM |
384 | * Update the per-runqueue clock, as finegrained as the platform can give |
385 | * us, but without assuming monotonicity, etc.: | |
20d315d4 | 386 | */ |
b04a0f4c | 387 | static void __update_rq_clock(struct rq *rq) |
20d315d4 IM |
388 | { |
389 | u64 prev_raw = rq->prev_clock_raw; | |
390 | u64 now = sched_clock(); | |
391 | s64 delta = now - prev_raw; | |
392 | u64 clock = rq->clock; | |
393 | ||
b04a0f4c IM |
394 | #ifdef CONFIG_SCHED_DEBUG |
395 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
396 | #endif | |
20d315d4 IM |
397 | /* |
398 | * Protect against sched_clock() occasionally going backwards: | |
399 | */ | |
400 | if (unlikely(delta < 0)) { | |
401 | clock++; | |
402 | rq->clock_warps++; | |
403 | } else { | |
404 | /* | |
405 | * Catch too large forward jumps too: | |
406 | */ | |
529c7726 IM |
407 | if (unlikely(clock + delta > rq->tick_timestamp + TICK_NSEC)) { |
408 | if (clock < rq->tick_timestamp + TICK_NSEC) | |
409 | clock = rq->tick_timestamp + TICK_NSEC; | |
410 | else | |
411 | clock++; | |
20d315d4 IM |
412 | rq->clock_overflows++; |
413 | } else { | |
414 | if (unlikely(delta > rq->clock_max_delta)) | |
415 | rq->clock_max_delta = delta; | |
416 | clock += delta; | |
417 | } | |
418 | } | |
419 | ||
420 | rq->prev_clock_raw = now; | |
421 | rq->clock = clock; | |
b04a0f4c | 422 | } |
20d315d4 | 423 | |
b04a0f4c IM |
424 | static void update_rq_clock(struct rq *rq) |
425 | { | |
426 | if (likely(smp_processor_id() == cpu_of(rq))) | |
427 | __update_rq_clock(rq); | |
20d315d4 IM |
428 | } |
429 | ||
674311d5 NP |
430 | /* |
431 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 432 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
433 | * |
434 | * The domain tree of any CPU may only be accessed from within | |
435 | * preempt-disabled sections. | |
436 | */ | |
48f24c4d IM |
437 | #define for_each_domain(cpu, __sd) \ |
438 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
439 | |
440 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
441 | #define this_rq() (&__get_cpu_var(runqueues)) | |
442 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
443 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
444 | ||
bf5c91ba IM |
445 | /* |
446 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
447 | */ | |
448 | #ifdef CONFIG_SCHED_DEBUG | |
449 | # define const_debug __read_mostly | |
450 | #else | |
451 | # define const_debug static const | |
452 | #endif | |
453 | ||
454 | /* | |
455 | * Debugging: various feature bits | |
456 | */ | |
457 | enum { | |
bbdba7c0 IM |
458 | SCHED_FEAT_NEW_FAIR_SLEEPERS = 1, |
459 | SCHED_FEAT_START_DEBIT = 2, | |
06877c33 | 460 | SCHED_FEAT_TREE_AVG = 4, |
bbdba7c0 | 461 | SCHED_FEAT_APPROX_AVG = 8, |
ce6c1311 | 462 | SCHED_FEAT_WAKEUP_PREEMPT = 16, |
95938a35 | 463 | SCHED_FEAT_PREEMPT_RESTRICT = 32, |
bf5c91ba IM |
464 | }; |
465 | ||
466 | const_debug unsigned int sysctl_sched_features = | |
8401f775 IM |
467 | SCHED_FEAT_NEW_FAIR_SLEEPERS * 1 | |
468 | SCHED_FEAT_START_DEBIT * 1 | | |
469 | SCHED_FEAT_TREE_AVG * 0 | | |
470 | SCHED_FEAT_APPROX_AVG * 0 | | |
471 | SCHED_FEAT_WAKEUP_PREEMPT * 1 | | |
472 | SCHED_FEAT_PREEMPT_RESTRICT * 1; | |
bf5c91ba IM |
473 | |
474 | #define sched_feat(x) (sysctl_sched_features & SCHED_FEAT_##x) | |
475 | ||
e436d800 IM |
476 | /* |
477 | * For kernel-internal use: high-speed (but slightly incorrect) per-cpu | |
478 | * clock constructed from sched_clock(): | |
479 | */ | |
480 | unsigned long long cpu_clock(int cpu) | |
481 | { | |
e436d800 IM |
482 | unsigned long long now; |
483 | unsigned long flags; | |
b04a0f4c | 484 | struct rq *rq; |
e436d800 | 485 | |
2cd4d0ea | 486 | local_irq_save(flags); |
b04a0f4c IM |
487 | rq = cpu_rq(cpu); |
488 | update_rq_clock(rq); | |
489 | now = rq->clock; | |
2cd4d0ea | 490 | local_irq_restore(flags); |
e436d800 IM |
491 | |
492 | return now; | |
493 | } | |
a58f6f25 | 494 | EXPORT_SYMBOL_GPL(cpu_clock); |
e436d800 | 495 | |
1da177e4 | 496 | #ifndef prepare_arch_switch |
4866cde0 NP |
497 | # define prepare_arch_switch(next) do { } while (0) |
498 | #endif | |
499 | #ifndef finish_arch_switch | |
500 | # define finish_arch_switch(prev) do { } while (0) | |
501 | #endif | |
502 | ||
503 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
70b97a7f | 504 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
505 | { |
506 | return rq->curr == p; | |
507 | } | |
508 | ||
70b97a7f | 509 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
510 | { |
511 | } | |
512 | ||
70b97a7f | 513 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 514 | { |
da04c035 IM |
515 | #ifdef CONFIG_DEBUG_SPINLOCK |
516 | /* this is a valid case when another task releases the spinlock */ | |
517 | rq->lock.owner = current; | |
518 | #endif | |
8a25d5de IM |
519 | /* |
520 | * If we are tracking spinlock dependencies then we have to | |
521 | * fix up the runqueue lock - which gets 'carried over' from | |
522 | * prev into current: | |
523 | */ | |
524 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
525 | ||
4866cde0 NP |
526 | spin_unlock_irq(&rq->lock); |
527 | } | |
528 | ||
529 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 530 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
531 | { |
532 | #ifdef CONFIG_SMP | |
533 | return p->oncpu; | |
534 | #else | |
535 | return rq->curr == p; | |
536 | #endif | |
537 | } | |
538 | ||
70b97a7f | 539 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
540 | { |
541 | #ifdef CONFIG_SMP | |
542 | /* | |
543 | * We can optimise this out completely for !SMP, because the | |
544 | * SMP rebalancing from interrupt is the only thing that cares | |
545 | * here. | |
546 | */ | |
547 | next->oncpu = 1; | |
548 | #endif | |
549 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
550 | spin_unlock_irq(&rq->lock); | |
551 | #else | |
552 | spin_unlock(&rq->lock); | |
553 | #endif | |
554 | } | |
555 | ||
70b97a7f | 556 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
557 | { |
558 | #ifdef CONFIG_SMP | |
559 | /* | |
560 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
561 | * We must ensure this doesn't happen until the switch is completely | |
562 | * finished. | |
563 | */ | |
564 | smp_wmb(); | |
565 | prev->oncpu = 0; | |
566 | #endif | |
567 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
568 | local_irq_enable(); | |
1da177e4 | 569 | #endif |
4866cde0 NP |
570 | } |
571 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 572 | |
b29739f9 IM |
573 | /* |
574 | * __task_rq_lock - lock the runqueue a given task resides on. | |
575 | * Must be called interrupts disabled. | |
576 | */ | |
70b97a7f | 577 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
578 | __acquires(rq->lock) |
579 | { | |
3a5c359a AK |
580 | for (;;) { |
581 | struct rq *rq = task_rq(p); | |
582 | spin_lock(&rq->lock); | |
583 | if (likely(rq == task_rq(p))) | |
584 | return rq; | |
b29739f9 | 585 | spin_unlock(&rq->lock); |
b29739f9 | 586 | } |
b29739f9 IM |
587 | } |
588 | ||
1da177e4 LT |
589 | /* |
590 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
591 | * interrupts. Note the ordering: we can safely lookup the task_rq without | |
592 | * explicitly disabling preemption. | |
593 | */ | |
70b97a7f | 594 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
595 | __acquires(rq->lock) |
596 | { | |
70b97a7f | 597 | struct rq *rq; |
1da177e4 | 598 | |
3a5c359a AK |
599 | for (;;) { |
600 | local_irq_save(*flags); | |
601 | rq = task_rq(p); | |
602 | spin_lock(&rq->lock); | |
603 | if (likely(rq == task_rq(p))) | |
604 | return rq; | |
1da177e4 | 605 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 606 | } |
1da177e4 LT |
607 | } |
608 | ||
a9957449 | 609 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
610 | __releases(rq->lock) |
611 | { | |
612 | spin_unlock(&rq->lock); | |
613 | } | |
614 | ||
70b97a7f | 615 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
616 | __releases(rq->lock) |
617 | { | |
618 | spin_unlock_irqrestore(&rq->lock, *flags); | |
619 | } | |
620 | ||
1da177e4 | 621 | /* |
cc2a73b5 | 622 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 623 | */ |
a9957449 | 624 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
625 | __acquires(rq->lock) |
626 | { | |
70b97a7f | 627 | struct rq *rq; |
1da177e4 LT |
628 | |
629 | local_irq_disable(); | |
630 | rq = this_rq(); | |
631 | spin_lock(&rq->lock); | |
632 | ||
633 | return rq; | |
634 | } | |
635 | ||
1b9f19c2 | 636 | /* |
2aa44d05 | 637 | * We are going deep-idle (irqs are disabled): |
1b9f19c2 | 638 | */ |
2aa44d05 | 639 | void sched_clock_idle_sleep_event(void) |
1b9f19c2 | 640 | { |
2aa44d05 IM |
641 | struct rq *rq = cpu_rq(smp_processor_id()); |
642 | ||
643 | spin_lock(&rq->lock); | |
644 | __update_rq_clock(rq); | |
645 | spin_unlock(&rq->lock); | |
646 | rq->clock_deep_idle_events++; | |
647 | } | |
648 | EXPORT_SYMBOL_GPL(sched_clock_idle_sleep_event); | |
649 | ||
650 | /* | |
651 | * We just idled delta nanoseconds (called with irqs disabled): | |
652 | */ | |
653 | void sched_clock_idle_wakeup_event(u64 delta_ns) | |
654 | { | |
655 | struct rq *rq = cpu_rq(smp_processor_id()); | |
656 | u64 now = sched_clock(); | |
1b9f19c2 | 657 | |
2aa44d05 IM |
658 | rq->idle_clock += delta_ns; |
659 | /* | |
660 | * Override the previous timestamp and ignore all | |
661 | * sched_clock() deltas that occured while we idled, | |
662 | * and use the PM-provided delta_ns to advance the | |
663 | * rq clock: | |
664 | */ | |
665 | spin_lock(&rq->lock); | |
666 | rq->prev_clock_raw = now; | |
667 | rq->clock += delta_ns; | |
668 | spin_unlock(&rq->lock); | |
1b9f19c2 | 669 | } |
2aa44d05 | 670 | EXPORT_SYMBOL_GPL(sched_clock_idle_wakeup_event); |
1b9f19c2 | 671 | |
c24d20db IM |
672 | /* |
673 | * resched_task - mark a task 'to be rescheduled now'. | |
674 | * | |
675 | * On UP this means the setting of the need_resched flag, on SMP it | |
676 | * might also involve a cross-CPU call to trigger the scheduler on | |
677 | * the target CPU. | |
678 | */ | |
679 | #ifdef CONFIG_SMP | |
680 | ||
681 | #ifndef tsk_is_polling | |
682 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
683 | #endif | |
684 | ||
685 | static void resched_task(struct task_struct *p) | |
686 | { | |
687 | int cpu; | |
688 | ||
689 | assert_spin_locked(&task_rq(p)->lock); | |
690 | ||
691 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) | |
692 | return; | |
693 | ||
694 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); | |
695 | ||
696 | cpu = task_cpu(p); | |
697 | if (cpu == smp_processor_id()) | |
698 | return; | |
699 | ||
700 | /* NEED_RESCHED must be visible before we test polling */ | |
701 | smp_mb(); | |
702 | if (!tsk_is_polling(p)) | |
703 | smp_send_reschedule(cpu); | |
704 | } | |
705 | ||
706 | static void resched_cpu(int cpu) | |
707 | { | |
708 | struct rq *rq = cpu_rq(cpu); | |
709 | unsigned long flags; | |
710 | ||
711 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
712 | return; | |
713 | resched_task(cpu_curr(cpu)); | |
714 | spin_unlock_irqrestore(&rq->lock, flags); | |
715 | } | |
716 | #else | |
717 | static inline void resched_task(struct task_struct *p) | |
718 | { | |
719 | assert_spin_locked(&task_rq(p)->lock); | |
720 | set_tsk_need_resched(p); | |
721 | } | |
722 | #endif | |
723 | ||
45bf76df IM |
724 | #if BITS_PER_LONG == 32 |
725 | # define WMULT_CONST (~0UL) | |
726 | #else | |
727 | # define WMULT_CONST (1UL << 32) | |
728 | #endif | |
729 | ||
730 | #define WMULT_SHIFT 32 | |
731 | ||
194081eb IM |
732 | /* |
733 | * Shift right and round: | |
734 | */ | |
cf2ab469 | 735 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 736 | |
cb1c4fc9 | 737 | static unsigned long |
45bf76df IM |
738 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
739 | struct load_weight *lw) | |
740 | { | |
741 | u64 tmp; | |
742 | ||
743 | if (unlikely(!lw->inv_weight)) | |
194081eb | 744 | lw->inv_weight = (WMULT_CONST - lw->weight/2) / lw->weight + 1; |
45bf76df IM |
745 | |
746 | tmp = (u64)delta_exec * weight; | |
747 | /* | |
748 | * Check whether we'd overflow the 64-bit multiplication: | |
749 | */ | |
194081eb | 750 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 751 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
752 | WMULT_SHIFT/2); |
753 | else | |
cf2ab469 | 754 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 755 | |
ecf691da | 756 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
757 | } |
758 | ||
759 | static inline unsigned long | |
760 | calc_delta_fair(unsigned long delta_exec, struct load_weight *lw) | |
761 | { | |
762 | return calc_delta_mine(delta_exec, NICE_0_LOAD, lw); | |
763 | } | |
764 | ||
1091985b | 765 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
766 | { |
767 | lw->weight += inc; | |
45bf76df IM |
768 | } |
769 | ||
1091985b | 770 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
771 | { |
772 | lw->weight -= dec; | |
45bf76df IM |
773 | } |
774 | ||
2dd73a4f PW |
775 | /* |
776 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
777 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
778 | * each task makes to its run queue's load is weighted according to its | |
779 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
780 | * scaled version of the new time slice allocation that they receive on time | |
781 | * slice expiry etc. | |
782 | */ | |
783 | ||
dd41f596 IM |
784 | #define WEIGHT_IDLEPRIO 2 |
785 | #define WMULT_IDLEPRIO (1 << 31) | |
786 | ||
787 | /* | |
788 | * Nice levels are multiplicative, with a gentle 10% change for every | |
789 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
790 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
791 | * that remained on nice 0. | |
792 | * | |
793 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
794 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
795 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
796 | * If a task goes up by ~10% and another task goes down by ~10% then | |
797 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
798 | */ |
799 | static const int prio_to_weight[40] = { | |
254753dc IM |
800 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
801 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
802 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
803 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
804 | /* 0 */ 1024, 820, 655, 526, 423, | |
805 | /* 5 */ 335, 272, 215, 172, 137, | |
806 | /* 10 */ 110, 87, 70, 56, 45, | |
807 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
808 | }; |
809 | ||
5714d2de IM |
810 | /* |
811 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
812 | * | |
813 | * In cases where the weight does not change often, we can use the | |
814 | * precalculated inverse to speed up arithmetics by turning divisions | |
815 | * into multiplications: | |
816 | */ | |
dd41f596 | 817 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
818 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
819 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
820 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
821 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
822 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
823 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
824 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
825 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 826 | }; |
2dd73a4f | 827 | |
dd41f596 IM |
828 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
829 | ||
830 | /* | |
831 | * runqueue iterator, to support SMP load-balancing between different | |
832 | * scheduling classes, without having to expose their internal data | |
833 | * structures to the load-balancing proper: | |
834 | */ | |
835 | struct rq_iterator { | |
836 | void *arg; | |
837 | struct task_struct *(*start)(void *); | |
838 | struct task_struct *(*next)(void *); | |
839 | }; | |
840 | ||
841 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
842 | unsigned long max_nr_move, unsigned long max_load_move, | |
843 | struct sched_domain *sd, enum cpu_idle_type idle, | |
844 | int *all_pinned, unsigned long *load_moved, | |
a4ac01c3 | 845 | int *this_best_prio, struct rq_iterator *iterator); |
dd41f596 IM |
846 | |
847 | #include "sched_stats.h" | |
dd41f596 | 848 | #include "sched_idletask.c" |
5522d5d5 IM |
849 | #include "sched_fair.c" |
850 | #include "sched_rt.c" | |
dd41f596 IM |
851 | #ifdef CONFIG_SCHED_DEBUG |
852 | # include "sched_debug.c" | |
853 | #endif | |
854 | ||
855 | #define sched_class_highest (&rt_sched_class) | |
856 | ||
9c217245 IM |
857 | /* |
858 | * Update delta_exec, delta_fair fields for rq. | |
859 | * | |
860 | * delta_fair clock advances at a rate inversely proportional to | |
495eca49 | 861 | * total load (rq->load.weight) on the runqueue, while |
9c217245 IM |
862 | * delta_exec advances at the same rate as wall-clock (provided |
863 | * cpu is not idle). | |
864 | * | |
865 | * delta_exec / delta_fair is a measure of the (smoothened) load on this | |
866 | * runqueue over any given interval. This (smoothened) load is used | |
867 | * during load balance. | |
868 | * | |
495eca49 | 869 | * This function is called /before/ updating rq->load |
9c217245 IM |
870 | * and when switching tasks. |
871 | */ | |
29b4b623 | 872 | static inline void inc_load(struct rq *rq, const struct task_struct *p) |
9c217245 | 873 | { |
495eca49 | 874 | update_load_add(&rq->load, p->se.load.weight); |
9c217245 IM |
875 | } |
876 | ||
79b5dddf | 877 | static inline void dec_load(struct rq *rq, const struct task_struct *p) |
9c217245 | 878 | { |
495eca49 | 879 | update_load_sub(&rq->load, p->se.load.weight); |
9c217245 IM |
880 | } |
881 | ||
e5fa2237 | 882 | static void inc_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
883 | { |
884 | rq->nr_running++; | |
29b4b623 | 885 | inc_load(rq, p); |
9c217245 IM |
886 | } |
887 | ||
db53181e | 888 | static void dec_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
889 | { |
890 | rq->nr_running--; | |
79b5dddf | 891 | dec_load(rq, p); |
9c217245 IM |
892 | } |
893 | ||
45bf76df IM |
894 | static void set_load_weight(struct task_struct *p) |
895 | { | |
896 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
897 | p->se.load.weight = prio_to_weight[0] * 2; |
898 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
899 | return; | |
900 | } | |
45bf76df | 901 | |
dd41f596 IM |
902 | /* |
903 | * SCHED_IDLE tasks get minimal weight: | |
904 | */ | |
905 | if (p->policy == SCHED_IDLE) { | |
906 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
907 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
908 | return; | |
909 | } | |
71f8bd46 | 910 | |
dd41f596 IM |
911 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
912 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
913 | } |
914 | ||
8159f87e | 915 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 916 | { |
dd41f596 | 917 | sched_info_queued(p); |
fd390f6a | 918 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 919 | p->se.on_rq = 1; |
71f8bd46 IM |
920 | } |
921 | ||
69be72c1 | 922 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 923 | { |
f02231e5 | 924 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 925 | p->se.on_rq = 0; |
71f8bd46 IM |
926 | } |
927 | ||
14531189 | 928 | /* |
dd41f596 | 929 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 930 | */ |
14531189 IM |
931 | static inline int __normal_prio(struct task_struct *p) |
932 | { | |
dd41f596 | 933 | return p->static_prio; |
14531189 IM |
934 | } |
935 | ||
b29739f9 IM |
936 | /* |
937 | * Calculate the expected normal priority: i.e. priority | |
938 | * without taking RT-inheritance into account. Might be | |
939 | * boosted by interactivity modifiers. Changes upon fork, | |
940 | * setprio syscalls, and whenever the interactivity | |
941 | * estimator recalculates. | |
942 | */ | |
36c8b586 | 943 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
944 | { |
945 | int prio; | |
946 | ||
e05606d3 | 947 | if (task_has_rt_policy(p)) |
b29739f9 IM |
948 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
949 | else | |
950 | prio = __normal_prio(p); | |
951 | return prio; | |
952 | } | |
953 | ||
954 | /* | |
955 | * Calculate the current priority, i.e. the priority | |
956 | * taken into account by the scheduler. This value might | |
957 | * be boosted by RT tasks, or might be boosted by | |
958 | * interactivity modifiers. Will be RT if the task got | |
959 | * RT-boosted. If not then it returns p->normal_prio. | |
960 | */ | |
36c8b586 | 961 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
962 | { |
963 | p->normal_prio = normal_prio(p); | |
964 | /* | |
965 | * If we are RT tasks or we were boosted to RT priority, | |
966 | * keep the priority unchanged. Otherwise, update priority | |
967 | * to the normal priority: | |
968 | */ | |
969 | if (!rt_prio(p->prio)) | |
970 | return p->normal_prio; | |
971 | return p->prio; | |
972 | } | |
973 | ||
1da177e4 | 974 | /* |
dd41f596 | 975 | * activate_task - move a task to the runqueue. |
1da177e4 | 976 | */ |
dd41f596 | 977 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 978 | { |
dd41f596 IM |
979 | if (p->state == TASK_UNINTERRUPTIBLE) |
980 | rq->nr_uninterruptible--; | |
1da177e4 | 981 | |
8159f87e | 982 | enqueue_task(rq, p, wakeup); |
e5fa2237 | 983 | inc_nr_running(p, rq); |
1da177e4 LT |
984 | } |
985 | ||
1da177e4 LT |
986 | /* |
987 | * deactivate_task - remove a task from the runqueue. | |
988 | */ | |
2e1cb74a | 989 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 990 | { |
dd41f596 IM |
991 | if (p->state == TASK_UNINTERRUPTIBLE) |
992 | rq->nr_uninterruptible++; | |
993 | ||
69be72c1 | 994 | dequeue_task(rq, p, sleep); |
db53181e | 995 | dec_nr_running(p, rq); |
1da177e4 LT |
996 | } |
997 | ||
1da177e4 LT |
998 | /** |
999 | * task_curr - is this task currently executing on a CPU? | |
1000 | * @p: the task in question. | |
1001 | */ | |
36c8b586 | 1002 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1003 | { |
1004 | return cpu_curr(task_cpu(p)) == p; | |
1005 | } | |
1006 | ||
2dd73a4f PW |
1007 | /* Used instead of source_load when we know the type == 0 */ |
1008 | unsigned long weighted_cpuload(const int cpu) | |
1009 | { | |
495eca49 | 1010 | return cpu_rq(cpu)->load.weight; |
dd41f596 IM |
1011 | } |
1012 | ||
1013 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1014 | { | |
1015 | #ifdef CONFIG_SMP | |
1016 | task_thread_info(p)->cpu = cpu; | |
dd41f596 | 1017 | #endif |
29f59db3 | 1018 | set_task_cfs_rq(p); |
2dd73a4f PW |
1019 | } |
1020 | ||
1da177e4 | 1021 | #ifdef CONFIG_SMP |
c65cc870 | 1022 | |
cc367732 IM |
1023 | /* |
1024 | * Is this task likely cache-hot: | |
1025 | */ | |
1026 | static inline int | |
1027 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) | |
1028 | { | |
1029 | s64 delta; | |
1030 | ||
1031 | if (p->sched_class != &fair_sched_class) | |
1032 | return 0; | |
1033 | ||
6bc1665b IM |
1034 | if (sysctl_sched_migration_cost == -1) |
1035 | return 1; | |
1036 | if (sysctl_sched_migration_cost == 0) | |
1037 | return 0; | |
1038 | ||
cc367732 IM |
1039 | delta = now - p->se.exec_start; |
1040 | ||
1041 | return delta < (s64)sysctl_sched_migration_cost; | |
1042 | } | |
1043 | ||
1044 | ||
dd41f596 | 1045 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1046 | { |
dd41f596 IM |
1047 | int old_cpu = task_cpu(p); |
1048 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1049 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1050 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1051 | u64 clock_offset; |
dd41f596 IM |
1052 | |
1053 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1054 | |
1055 | #ifdef CONFIG_SCHEDSTATS | |
1056 | if (p->se.wait_start) | |
1057 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1058 | if (p->se.sleep_start) |
1059 | p->se.sleep_start -= clock_offset; | |
1060 | if (p->se.block_start) | |
1061 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1062 | if (old_cpu != new_cpu) { |
1063 | schedstat_inc(p, se.nr_migrations); | |
1064 | if (task_hot(p, old_rq->clock, NULL)) | |
1065 | schedstat_inc(p, se.nr_forced2_migrations); | |
1066 | } | |
6cfb0d5d | 1067 | #endif |
2830cf8c SV |
1068 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1069 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1070 | |
1071 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1072 | } |
1073 | ||
70b97a7f | 1074 | struct migration_req { |
1da177e4 | 1075 | struct list_head list; |
1da177e4 | 1076 | |
36c8b586 | 1077 | struct task_struct *task; |
1da177e4 LT |
1078 | int dest_cpu; |
1079 | ||
1da177e4 | 1080 | struct completion done; |
70b97a7f | 1081 | }; |
1da177e4 LT |
1082 | |
1083 | /* | |
1084 | * The task's runqueue lock must be held. | |
1085 | * Returns true if you have to wait for migration thread. | |
1086 | */ | |
36c8b586 | 1087 | static int |
70b97a7f | 1088 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1089 | { |
70b97a7f | 1090 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1091 | |
1092 | /* | |
1093 | * If the task is not on a runqueue (and not running), then | |
1094 | * it is sufficient to simply update the task's cpu field. | |
1095 | */ | |
dd41f596 | 1096 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1097 | set_task_cpu(p, dest_cpu); |
1098 | return 0; | |
1099 | } | |
1100 | ||
1101 | init_completion(&req->done); | |
1da177e4 LT |
1102 | req->task = p; |
1103 | req->dest_cpu = dest_cpu; | |
1104 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1105 | |
1da177e4 LT |
1106 | return 1; |
1107 | } | |
1108 | ||
1109 | /* | |
1110 | * wait_task_inactive - wait for a thread to unschedule. | |
1111 | * | |
1112 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1113 | * else this function might spin for a *long* time. This function can't | |
1114 | * be called with interrupts off, or it may introduce deadlock with | |
1115 | * smp_call_function() if an IPI is sent by the same process we are | |
1116 | * waiting to become inactive. | |
1117 | */ | |
36c8b586 | 1118 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1119 | { |
1120 | unsigned long flags; | |
dd41f596 | 1121 | int running, on_rq; |
70b97a7f | 1122 | struct rq *rq; |
1da177e4 | 1123 | |
3a5c359a AK |
1124 | for (;;) { |
1125 | /* | |
1126 | * We do the initial early heuristics without holding | |
1127 | * any task-queue locks at all. We'll only try to get | |
1128 | * the runqueue lock when things look like they will | |
1129 | * work out! | |
1130 | */ | |
1131 | rq = task_rq(p); | |
fa490cfd | 1132 | |
3a5c359a AK |
1133 | /* |
1134 | * If the task is actively running on another CPU | |
1135 | * still, just relax and busy-wait without holding | |
1136 | * any locks. | |
1137 | * | |
1138 | * NOTE! Since we don't hold any locks, it's not | |
1139 | * even sure that "rq" stays as the right runqueue! | |
1140 | * But we don't care, since "task_running()" will | |
1141 | * return false if the runqueue has changed and p | |
1142 | * is actually now running somewhere else! | |
1143 | */ | |
1144 | while (task_running(rq, p)) | |
1145 | cpu_relax(); | |
fa490cfd | 1146 | |
3a5c359a AK |
1147 | /* |
1148 | * Ok, time to look more closely! We need the rq | |
1149 | * lock now, to be *sure*. If we're wrong, we'll | |
1150 | * just go back and repeat. | |
1151 | */ | |
1152 | rq = task_rq_lock(p, &flags); | |
1153 | running = task_running(rq, p); | |
1154 | on_rq = p->se.on_rq; | |
1155 | task_rq_unlock(rq, &flags); | |
fa490cfd | 1156 | |
3a5c359a AK |
1157 | /* |
1158 | * Was it really running after all now that we | |
1159 | * checked with the proper locks actually held? | |
1160 | * | |
1161 | * Oops. Go back and try again.. | |
1162 | */ | |
1163 | if (unlikely(running)) { | |
1164 | cpu_relax(); | |
1165 | continue; | |
1166 | } | |
fa490cfd | 1167 | |
3a5c359a AK |
1168 | /* |
1169 | * It's not enough that it's not actively running, | |
1170 | * it must be off the runqueue _entirely_, and not | |
1171 | * preempted! | |
1172 | * | |
1173 | * So if it wa still runnable (but just not actively | |
1174 | * running right now), it's preempted, and we should | |
1175 | * yield - it could be a while. | |
1176 | */ | |
1177 | if (unlikely(on_rq)) { | |
1178 | schedule_timeout_uninterruptible(1); | |
1179 | continue; | |
1180 | } | |
fa490cfd | 1181 | |
3a5c359a AK |
1182 | /* |
1183 | * Ahh, all good. It wasn't running, and it wasn't | |
1184 | * runnable, which means that it will never become | |
1185 | * running in the future either. We're all done! | |
1186 | */ | |
1187 | break; | |
1188 | } | |
1da177e4 LT |
1189 | } |
1190 | ||
1191 | /*** | |
1192 | * kick_process - kick a running thread to enter/exit the kernel | |
1193 | * @p: the to-be-kicked thread | |
1194 | * | |
1195 | * Cause a process which is running on another CPU to enter | |
1196 | * kernel-mode, without any delay. (to get signals handled.) | |
1197 | * | |
1198 | * NOTE: this function doesnt have to take the runqueue lock, | |
1199 | * because all it wants to ensure is that the remote task enters | |
1200 | * the kernel. If the IPI races and the task has been migrated | |
1201 | * to another CPU then no harm is done and the purpose has been | |
1202 | * achieved as well. | |
1203 | */ | |
36c8b586 | 1204 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1205 | { |
1206 | int cpu; | |
1207 | ||
1208 | preempt_disable(); | |
1209 | cpu = task_cpu(p); | |
1210 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1211 | smp_send_reschedule(cpu); | |
1212 | preempt_enable(); | |
1213 | } | |
1214 | ||
1215 | /* | |
2dd73a4f PW |
1216 | * Return a low guess at the load of a migration-source cpu weighted |
1217 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1218 | * |
1219 | * We want to under-estimate the load of migration sources, to | |
1220 | * balance conservatively. | |
1221 | */ | |
a9957449 | 1222 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 1223 | { |
70b97a7f | 1224 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1225 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1226 | |
3b0bd9bc | 1227 | if (type == 0) |
dd41f596 | 1228 | return total; |
b910472d | 1229 | |
dd41f596 | 1230 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
1231 | } |
1232 | ||
1233 | /* | |
2dd73a4f PW |
1234 | * Return a high guess at the load of a migration-target cpu weighted |
1235 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1236 | */ |
a9957449 | 1237 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 1238 | { |
70b97a7f | 1239 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1240 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1241 | |
7897986b | 1242 | if (type == 0) |
dd41f596 | 1243 | return total; |
3b0bd9bc | 1244 | |
dd41f596 | 1245 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
1246 | } |
1247 | ||
1248 | /* | |
1249 | * Return the average load per task on the cpu's run queue | |
1250 | */ | |
1251 | static inline unsigned long cpu_avg_load_per_task(int cpu) | |
1252 | { | |
70b97a7f | 1253 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1254 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f PW |
1255 | unsigned long n = rq->nr_running; |
1256 | ||
dd41f596 | 1257 | return n ? total / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1258 | } |
1259 | ||
147cbb4b NP |
1260 | /* |
1261 | * find_idlest_group finds and returns the least busy CPU group within the | |
1262 | * domain. | |
1263 | */ | |
1264 | static struct sched_group * | |
1265 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1266 | { | |
1267 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1268 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1269 | int load_idx = sd->forkexec_idx; | |
1270 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1271 | ||
1272 | do { | |
1273 | unsigned long load, avg_load; | |
1274 | int local_group; | |
1275 | int i; | |
1276 | ||
da5a5522 BD |
1277 | /* Skip over this group if it has no CPUs allowed */ |
1278 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 1279 | continue; |
da5a5522 | 1280 | |
147cbb4b | 1281 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1282 | |
1283 | /* Tally up the load of all CPUs in the group */ | |
1284 | avg_load = 0; | |
1285 | ||
1286 | for_each_cpu_mask(i, group->cpumask) { | |
1287 | /* Bias balancing toward cpus of our domain */ | |
1288 | if (local_group) | |
1289 | load = source_load(i, load_idx); | |
1290 | else | |
1291 | load = target_load(i, load_idx); | |
1292 | ||
1293 | avg_load += load; | |
1294 | } | |
1295 | ||
1296 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1297 | avg_load = sg_div_cpu_power(group, |
1298 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1299 | |
1300 | if (local_group) { | |
1301 | this_load = avg_load; | |
1302 | this = group; | |
1303 | } else if (avg_load < min_load) { | |
1304 | min_load = avg_load; | |
1305 | idlest = group; | |
1306 | } | |
3a5c359a | 1307 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
1308 | |
1309 | if (!idlest || 100*this_load < imbalance*min_load) | |
1310 | return NULL; | |
1311 | return idlest; | |
1312 | } | |
1313 | ||
1314 | /* | |
0feaece9 | 1315 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1316 | */ |
95cdf3b7 IM |
1317 | static int |
1318 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) | |
147cbb4b | 1319 | { |
da5a5522 | 1320 | cpumask_t tmp; |
147cbb4b NP |
1321 | unsigned long load, min_load = ULONG_MAX; |
1322 | int idlest = -1; | |
1323 | int i; | |
1324 | ||
da5a5522 BD |
1325 | /* Traverse only the allowed CPUs */ |
1326 | cpus_and(tmp, group->cpumask, p->cpus_allowed); | |
1327 | ||
1328 | for_each_cpu_mask(i, tmp) { | |
2dd73a4f | 1329 | load = weighted_cpuload(i); |
147cbb4b NP |
1330 | |
1331 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1332 | min_load = load; | |
1333 | idlest = i; | |
1334 | } | |
1335 | } | |
1336 | ||
1337 | return idlest; | |
1338 | } | |
1339 | ||
476d139c NP |
1340 | /* |
1341 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1342 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1343 | * SD_BALANCE_EXEC. | |
1344 | * | |
1345 | * Balance, ie. select the least loaded group. | |
1346 | * | |
1347 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1348 | * | |
1349 | * preempt must be disabled. | |
1350 | */ | |
1351 | static int sched_balance_self(int cpu, int flag) | |
1352 | { | |
1353 | struct task_struct *t = current; | |
1354 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1355 | |
c96d145e | 1356 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
1357 | /* |
1358 | * If power savings logic is enabled for a domain, stop there. | |
1359 | */ | |
5c45bf27 SS |
1360 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
1361 | break; | |
476d139c NP |
1362 | if (tmp->flags & flag) |
1363 | sd = tmp; | |
c96d145e | 1364 | } |
476d139c NP |
1365 | |
1366 | while (sd) { | |
1367 | cpumask_t span; | |
1368 | struct sched_group *group; | |
1a848870 SS |
1369 | int new_cpu, weight; |
1370 | ||
1371 | if (!(sd->flags & flag)) { | |
1372 | sd = sd->child; | |
1373 | continue; | |
1374 | } | |
476d139c NP |
1375 | |
1376 | span = sd->span; | |
1377 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1378 | if (!group) { |
1379 | sd = sd->child; | |
1380 | continue; | |
1381 | } | |
476d139c | 1382 | |
da5a5522 | 1383 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
1384 | if (new_cpu == -1 || new_cpu == cpu) { |
1385 | /* Now try balancing at a lower domain level of cpu */ | |
1386 | sd = sd->child; | |
1387 | continue; | |
1388 | } | |
476d139c | 1389 | |
1a848870 | 1390 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1391 | cpu = new_cpu; |
476d139c NP |
1392 | sd = NULL; |
1393 | weight = cpus_weight(span); | |
1394 | for_each_domain(cpu, tmp) { | |
1395 | if (weight <= cpus_weight(tmp->span)) | |
1396 | break; | |
1397 | if (tmp->flags & flag) | |
1398 | sd = tmp; | |
1399 | } | |
1400 | /* while loop will break here if sd == NULL */ | |
1401 | } | |
1402 | ||
1403 | return cpu; | |
1404 | } | |
1405 | ||
1406 | #endif /* CONFIG_SMP */ | |
1da177e4 LT |
1407 | |
1408 | /* | |
1409 | * wake_idle() will wake a task on an idle cpu if task->cpu is | |
1410 | * not idle and an idle cpu is available. The span of cpus to | |
1411 | * search starts with cpus closest then further out as needed, | |
1412 | * so we always favor a closer, idle cpu. | |
1413 | * | |
1414 | * Returns the CPU we should wake onto. | |
1415 | */ | |
1416 | #if defined(ARCH_HAS_SCHED_WAKE_IDLE) | |
36c8b586 | 1417 | static int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1418 | { |
1419 | cpumask_t tmp; | |
1420 | struct sched_domain *sd; | |
1421 | int i; | |
1422 | ||
4953198b SS |
1423 | /* |
1424 | * If it is idle, then it is the best cpu to run this task. | |
1425 | * | |
1426 | * This cpu is also the best, if it has more than one task already. | |
1427 | * Siblings must be also busy(in most cases) as they didn't already | |
1428 | * pickup the extra load from this cpu and hence we need not check | |
1429 | * sibling runqueue info. This will avoid the checks and cache miss | |
1430 | * penalities associated with that. | |
1431 | */ | |
1432 | if (idle_cpu(cpu) || cpu_rq(cpu)->nr_running > 1) | |
1da177e4 LT |
1433 | return cpu; |
1434 | ||
1435 | for_each_domain(cpu, sd) { | |
1436 | if (sd->flags & SD_WAKE_IDLE) { | |
e0f364f4 | 1437 | cpus_and(tmp, sd->span, p->cpus_allowed); |
1da177e4 | 1438 | for_each_cpu_mask(i, tmp) { |
cc367732 IM |
1439 | if (idle_cpu(i)) { |
1440 | if (i != task_cpu(p)) { | |
1441 | schedstat_inc(p, | |
1442 | se.nr_wakeups_idle); | |
1443 | } | |
1da177e4 | 1444 | return i; |
cc367732 | 1445 | } |
1da177e4 | 1446 | } |
9761eea8 | 1447 | } else { |
e0f364f4 | 1448 | break; |
9761eea8 | 1449 | } |
1da177e4 LT |
1450 | } |
1451 | return cpu; | |
1452 | } | |
1453 | #else | |
36c8b586 | 1454 | static inline int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1455 | { |
1456 | return cpu; | |
1457 | } | |
1458 | #endif | |
1459 | ||
1460 | /*** | |
1461 | * try_to_wake_up - wake up a thread | |
1462 | * @p: the to-be-woken-up thread | |
1463 | * @state: the mask of task states that can be woken | |
1464 | * @sync: do a synchronous wakeup? | |
1465 | * | |
1466 | * Put it on the run-queue if it's not already there. The "current" | |
1467 | * thread is always on the run-queue (except when the actual | |
1468 | * re-schedule is in progress), and as such you're allowed to do | |
1469 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1470 | * runnable without the overhead of this. | |
1471 | * | |
1472 | * returns failure only if the task is already active. | |
1473 | */ | |
36c8b586 | 1474 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 1475 | { |
cc367732 | 1476 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
1477 | unsigned long flags; |
1478 | long old_state; | |
70b97a7f | 1479 | struct rq *rq; |
1da177e4 | 1480 | #ifdef CONFIG_SMP |
7897986b | 1481 | struct sched_domain *sd, *this_sd = NULL; |
70b97a7f | 1482 | unsigned long load, this_load; |
1da177e4 LT |
1483 | int new_cpu; |
1484 | #endif | |
1485 | ||
1486 | rq = task_rq_lock(p, &flags); | |
1487 | old_state = p->state; | |
1488 | if (!(old_state & state)) | |
1489 | goto out; | |
1490 | ||
dd41f596 | 1491 | if (p->se.on_rq) |
1da177e4 LT |
1492 | goto out_running; |
1493 | ||
1494 | cpu = task_cpu(p); | |
cc367732 | 1495 | orig_cpu = cpu; |
1da177e4 LT |
1496 | this_cpu = smp_processor_id(); |
1497 | ||
1498 | #ifdef CONFIG_SMP | |
1499 | if (unlikely(task_running(rq, p))) | |
1500 | goto out_activate; | |
1501 | ||
7897986b NP |
1502 | new_cpu = cpu; |
1503 | ||
2d72376b | 1504 | schedstat_inc(rq, ttwu_count); |
1da177e4 LT |
1505 | if (cpu == this_cpu) { |
1506 | schedstat_inc(rq, ttwu_local); | |
7897986b NP |
1507 | goto out_set_cpu; |
1508 | } | |
1509 | ||
1510 | for_each_domain(this_cpu, sd) { | |
1511 | if (cpu_isset(cpu, sd->span)) { | |
1512 | schedstat_inc(sd, ttwu_wake_remote); | |
1513 | this_sd = sd; | |
1514 | break; | |
1da177e4 LT |
1515 | } |
1516 | } | |
1da177e4 | 1517 | |
7897986b | 1518 | if (unlikely(!cpu_isset(this_cpu, p->cpus_allowed))) |
1da177e4 LT |
1519 | goto out_set_cpu; |
1520 | ||
1da177e4 | 1521 | /* |
7897986b | 1522 | * Check for affine wakeup and passive balancing possibilities. |
1da177e4 | 1523 | */ |
7897986b NP |
1524 | if (this_sd) { |
1525 | int idx = this_sd->wake_idx; | |
1526 | unsigned int imbalance; | |
1da177e4 | 1527 | |
a3f21bce NP |
1528 | imbalance = 100 + (this_sd->imbalance_pct - 100) / 2; |
1529 | ||
7897986b NP |
1530 | load = source_load(cpu, idx); |
1531 | this_load = target_load(this_cpu, idx); | |
1da177e4 | 1532 | |
7897986b NP |
1533 | new_cpu = this_cpu; /* Wake to this CPU if we can */ |
1534 | ||
a3f21bce NP |
1535 | if (this_sd->flags & SD_WAKE_AFFINE) { |
1536 | unsigned long tl = this_load; | |
33859f7f MOS |
1537 | unsigned long tl_per_task; |
1538 | ||
71e20f18 IM |
1539 | /* |
1540 | * Attract cache-cold tasks on sync wakeups: | |
1541 | */ | |
1542 | if (sync && !task_hot(p, rq->clock, this_sd)) | |
1543 | goto out_set_cpu; | |
1544 | ||
cc367732 | 1545 | schedstat_inc(p, se.nr_wakeups_affine_attempts); |
33859f7f | 1546 | tl_per_task = cpu_avg_load_per_task(this_cpu); |
2dd73a4f | 1547 | |
1da177e4 | 1548 | /* |
a3f21bce NP |
1549 | * If sync wakeup then subtract the (maximum possible) |
1550 | * effect of the currently running task from the load | |
1551 | * of the current CPU: | |
1da177e4 | 1552 | */ |
a3f21bce | 1553 | if (sync) |
dd41f596 | 1554 | tl -= current->se.load.weight; |
a3f21bce NP |
1555 | |
1556 | if ((tl <= load && | |
2dd73a4f | 1557 | tl + target_load(cpu, idx) <= tl_per_task) || |
dd41f596 | 1558 | 100*(tl + p->se.load.weight) <= imbalance*load) { |
a3f21bce NP |
1559 | /* |
1560 | * This domain has SD_WAKE_AFFINE and | |
1561 | * p is cache cold in this domain, and | |
1562 | * there is no bad imbalance. | |
1563 | */ | |
1564 | schedstat_inc(this_sd, ttwu_move_affine); | |
cc367732 | 1565 | schedstat_inc(p, se.nr_wakeups_affine); |
a3f21bce NP |
1566 | goto out_set_cpu; |
1567 | } | |
1568 | } | |
1569 | ||
1570 | /* | |
1571 | * Start passive balancing when half the imbalance_pct | |
1572 | * limit is reached. | |
1573 | */ | |
1574 | if (this_sd->flags & SD_WAKE_BALANCE) { | |
1575 | if (imbalance*this_load <= 100*load) { | |
1576 | schedstat_inc(this_sd, ttwu_move_balance); | |
cc367732 | 1577 | schedstat_inc(p, se.nr_wakeups_passive); |
a3f21bce NP |
1578 | goto out_set_cpu; |
1579 | } | |
1da177e4 LT |
1580 | } |
1581 | } | |
1582 | ||
1583 | new_cpu = cpu; /* Could not wake to this_cpu. Wake to cpu instead */ | |
1584 | out_set_cpu: | |
1585 | new_cpu = wake_idle(new_cpu, p); | |
1586 | if (new_cpu != cpu) { | |
1587 | set_task_cpu(p, new_cpu); | |
1588 | task_rq_unlock(rq, &flags); | |
1589 | /* might preempt at this point */ | |
1590 | rq = task_rq_lock(p, &flags); | |
1591 | old_state = p->state; | |
1592 | if (!(old_state & state)) | |
1593 | goto out; | |
dd41f596 | 1594 | if (p->se.on_rq) |
1da177e4 LT |
1595 | goto out_running; |
1596 | ||
1597 | this_cpu = smp_processor_id(); | |
1598 | cpu = task_cpu(p); | |
1599 | } | |
1600 | ||
1601 | out_activate: | |
1602 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
1603 | schedstat_inc(p, se.nr_wakeups); |
1604 | if (sync) | |
1605 | schedstat_inc(p, se.nr_wakeups_sync); | |
1606 | if (orig_cpu != cpu) | |
1607 | schedstat_inc(p, se.nr_wakeups_migrate); | |
1608 | if (cpu == this_cpu) | |
1609 | schedstat_inc(p, se.nr_wakeups_local); | |
1610 | else | |
1611 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 1612 | update_rq_clock(rq); |
dd41f596 | 1613 | activate_task(rq, p, 1); |
9c63d9c0 | 1614 | check_preempt_curr(rq, p); |
1da177e4 LT |
1615 | success = 1; |
1616 | ||
1617 | out_running: | |
1618 | p->state = TASK_RUNNING; | |
1619 | out: | |
1620 | task_rq_unlock(rq, &flags); | |
1621 | ||
1622 | return success; | |
1623 | } | |
1624 | ||
36c8b586 | 1625 | int fastcall wake_up_process(struct task_struct *p) |
1da177e4 LT |
1626 | { |
1627 | return try_to_wake_up(p, TASK_STOPPED | TASK_TRACED | | |
1628 | TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE, 0); | |
1629 | } | |
1da177e4 LT |
1630 | EXPORT_SYMBOL(wake_up_process); |
1631 | ||
36c8b586 | 1632 | int fastcall wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1633 | { |
1634 | return try_to_wake_up(p, state, 0); | |
1635 | } | |
1636 | ||
1da177e4 LT |
1637 | /* |
1638 | * Perform scheduler related setup for a newly forked process p. | |
1639 | * p is forked by current. | |
dd41f596 IM |
1640 | * |
1641 | * __sched_fork() is basic setup used by init_idle() too: | |
1642 | */ | |
1643 | static void __sched_fork(struct task_struct *p) | |
1644 | { | |
dd41f596 IM |
1645 | p->se.exec_start = 0; |
1646 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1647 | p->se.prev_sum_exec_runtime = 0; |
6cfb0d5d IM |
1648 | |
1649 | #ifdef CONFIG_SCHEDSTATS | |
1650 | p->se.wait_start = 0; | |
dd41f596 IM |
1651 | p->se.sum_sleep_runtime = 0; |
1652 | p->se.sleep_start = 0; | |
dd41f596 IM |
1653 | p->se.block_start = 0; |
1654 | p->se.sleep_max = 0; | |
1655 | p->se.block_max = 0; | |
1656 | p->se.exec_max = 0; | |
eba1ed4b | 1657 | p->se.slice_max = 0; |
dd41f596 | 1658 | p->se.wait_max = 0; |
6cfb0d5d | 1659 | #endif |
476d139c | 1660 | |
dd41f596 IM |
1661 | INIT_LIST_HEAD(&p->run_list); |
1662 | p->se.on_rq = 0; | |
476d139c | 1663 | |
e107be36 AK |
1664 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1665 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1666 | #endif | |
1667 | ||
1da177e4 LT |
1668 | /* |
1669 | * We mark the process as running here, but have not actually | |
1670 | * inserted it onto the runqueue yet. This guarantees that | |
1671 | * nobody will actually run it, and a signal or other external | |
1672 | * event cannot wake it up and insert it on the runqueue either. | |
1673 | */ | |
1674 | p->state = TASK_RUNNING; | |
dd41f596 IM |
1675 | } |
1676 | ||
1677 | /* | |
1678 | * fork()/clone()-time setup: | |
1679 | */ | |
1680 | void sched_fork(struct task_struct *p, int clone_flags) | |
1681 | { | |
1682 | int cpu = get_cpu(); | |
1683 | ||
1684 | __sched_fork(p); | |
1685 | ||
1686 | #ifdef CONFIG_SMP | |
1687 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
1688 | #endif | |
02e4bac2 | 1689 | set_task_cpu(p, cpu); |
b29739f9 IM |
1690 | |
1691 | /* | |
1692 | * Make sure we do not leak PI boosting priority to the child: | |
1693 | */ | |
1694 | p->prio = current->normal_prio; | |
2ddbf952 HS |
1695 | if (!rt_prio(p->prio)) |
1696 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1697 | |
52f17b6c | 1698 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1699 | if (likely(sched_info_on())) |
52f17b6c | 1700 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1701 | #endif |
d6077cb8 | 1702 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
1703 | p->oncpu = 0; |
1704 | #endif | |
1da177e4 | 1705 | #ifdef CONFIG_PREEMPT |
4866cde0 | 1706 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1707 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1708 | #endif |
476d139c | 1709 | put_cpu(); |
1da177e4 LT |
1710 | } |
1711 | ||
1712 | /* | |
1713 | * wake_up_new_task - wake up a newly created task for the first time. | |
1714 | * | |
1715 | * This function will do some initial scheduler statistics housekeeping | |
1716 | * that must be done for every newly created context, then puts the task | |
1717 | * on the runqueue and wakes it. | |
1718 | */ | |
36c8b586 | 1719 | void fastcall wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
1720 | { |
1721 | unsigned long flags; | |
dd41f596 | 1722 | struct rq *rq; |
1da177e4 LT |
1723 | |
1724 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 1725 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 1726 | update_rq_clock(rq); |
1da177e4 LT |
1727 | |
1728 | p->prio = effective_prio(p); | |
1729 | ||
b9dca1e0 | 1730 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 1731 | activate_task(rq, p, 0); |
1da177e4 | 1732 | } else { |
1da177e4 | 1733 | /* |
dd41f596 IM |
1734 | * Let the scheduling class do new task startup |
1735 | * management (if any): | |
1da177e4 | 1736 | */ |
ee0827d8 | 1737 | p->sched_class->task_new(rq, p); |
e5fa2237 | 1738 | inc_nr_running(p, rq); |
1da177e4 | 1739 | } |
dd41f596 IM |
1740 | check_preempt_curr(rq, p); |
1741 | task_rq_unlock(rq, &flags); | |
1da177e4 LT |
1742 | } |
1743 | ||
e107be36 AK |
1744 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1745 | ||
1746 | /** | |
421cee29 RD |
1747 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
1748 | * @notifier: notifier struct to register | |
e107be36 AK |
1749 | */ |
1750 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1751 | { | |
1752 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1753 | } | |
1754 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1755 | ||
1756 | /** | |
1757 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1758 | * @notifier: notifier struct to unregister |
e107be36 AK |
1759 | * |
1760 | * This is safe to call from within a preemption notifier. | |
1761 | */ | |
1762 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1763 | { | |
1764 | hlist_del(¬ifier->link); | |
1765 | } | |
1766 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1767 | ||
1768 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1769 | { | |
1770 | struct preempt_notifier *notifier; | |
1771 | struct hlist_node *node; | |
1772 | ||
1773 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1774 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
1775 | } | |
1776 | ||
1777 | static void | |
1778 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1779 | struct task_struct *next) | |
1780 | { | |
1781 | struct preempt_notifier *notifier; | |
1782 | struct hlist_node *node; | |
1783 | ||
1784 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1785 | notifier->ops->sched_out(notifier, next); | |
1786 | } | |
1787 | ||
1788 | #else | |
1789 | ||
1790 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1791 | { | |
1792 | } | |
1793 | ||
1794 | static void | |
1795 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1796 | struct task_struct *next) | |
1797 | { | |
1798 | } | |
1799 | ||
1800 | #endif | |
1801 | ||
4866cde0 NP |
1802 | /** |
1803 | * prepare_task_switch - prepare to switch tasks | |
1804 | * @rq: the runqueue preparing to switch | |
421cee29 | 1805 | * @prev: the current task that is being switched out |
4866cde0 NP |
1806 | * @next: the task we are going to switch to. |
1807 | * | |
1808 | * This is called with the rq lock held and interrupts off. It must | |
1809 | * be paired with a subsequent finish_task_switch after the context | |
1810 | * switch. | |
1811 | * | |
1812 | * prepare_task_switch sets up locking and calls architecture specific | |
1813 | * hooks. | |
1814 | */ | |
e107be36 AK |
1815 | static inline void |
1816 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1817 | struct task_struct *next) | |
4866cde0 | 1818 | { |
e107be36 | 1819 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1820 | prepare_lock_switch(rq, next); |
1821 | prepare_arch_switch(next); | |
1822 | } | |
1823 | ||
1da177e4 LT |
1824 | /** |
1825 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1826 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1827 | * @prev: the thread we just switched away from. |
1828 | * | |
4866cde0 NP |
1829 | * finish_task_switch must be called after the context switch, paired |
1830 | * with a prepare_task_switch call before the context switch. | |
1831 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1832 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1833 | * |
1834 | * Note that we may have delayed dropping an mm in context_switch(). If | |
1835 | * so, we finish that here outside of the runqueue lock. (Doing it | |
1836 | * with the lock held can cause deadlocks; see schedule() for | |
1837 | * details.) | |
1838 | */ | |
a9957449 | 1839 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1840 | __releases(rq->lock) |
1841 | { | |
1da177e4 | 1842 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1843 | long prev_state; |
1da177e4 LT |
1844 | |
1845 | rq->prev_mm = NULL; | |
1846 | ||
1847 | /* | |
1848 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1849 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1850 | * schedule one last time. The schedule call will never return, and |
1851 | * the scheduled task must drop that reference. | |
c394cc9f | 1852 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1853 | * still held, otherwise prev could be scheduled on another cpu, die |
1854 | * there before we look at prev->state, and then the reference would | |
1855 | * be dropped twice. | |
1856 | * Manfred Spraul <manfred@colorfullife.com> | |
1857 | */ | |
55a101f8 | 1858 | prev_state = prev->state; |
4866cde0 NP |
1859 | finish_arch_switch(prev); |
1860 | finish_lock_switch(rq, prev); | |
e107be36 | 1861 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1862 | if (mm) |
1863 | mmdrop(mm); | |
c394cc9f | 1864 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1865 | /* |
1866 | * Remove function-return probe instances associated with this | |
1867 | * task and put them back on the free list. | |
9761eea8 | 1868 | */ |
c6fd91f0 | 1869 | kprobe_flush_task(prev); |
1da177e4 | 1870 | put_task_struct(prev); |
c6fd91f0 | 1871 | } |
1da177e4 LT |
1872 | } |
1873 | ||
1874 | /** | |
1875 | * schedule_tail - first thing a freshly forked thread must call. | |
1876 | * @prev: the thread we just switched away from. | |
1877 | */ | |
36c8b586 | 1878 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1879 | __releases(rq->lock) |
1880 | { | |
70b97a7f IM |
1881 | struct rq *rq = this_rq(); |
1882 | ||
4866cde0 NP |
1883 | finish_task_switch(rq, prev); |
1884 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
1885 | /* In this case, finish_task_switch does not reenable preemption */ | |
1886 | preempt_enable(); | |
1887 | #endif | |
1da177e4 | 1888 | if (current->set_child_tid) |
b488893a | 1889 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
1890 | } |
1891 | ||
1892 | /* | |
1893 | * context_switch - switch to the new MM and the new | |
1894 | * thread's register state. | |
1895 | */ | |
dd41f596 | 1896 | static inline void |
70b97a7f | 1897 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1898 | struct task_struct *next) |
1da177e4 | 1899 | { |
dd41f596 | 1900 | struct mm_struct *mm, *oldmm; |
1da177e4 | 1901 | |
e107be36 | 1902 | prepare_task_switch(rq, prev, next); |
dd41f596 IM |
1903 | mm = next->mm; |
1904 | oldmm = prev->active_mm; | |
9226d125 ZA |
1905 | /* |
1906 | * For paravirt, this is coupled with an exit in switch_to to | |
1907 | * combine the page table reload and the switch backend into | |
1908 | * one hypercall. | |
1909 | */ | |
1910 | arch_enter_lazy_cpu_mode(); | |
1911 | ||
dd41f596 | 1912 | if (unlikely(!mm)) { |
1da177e4 LT |
1913 | next->active_mm = oldmm; |
1914 | atomic_inc(&oldmm->mm_count); | |
1915 | enter_lazy_tlb(oldmm, next); | |
1916 | } else | |
1917 | switch_mm(oldmm, mm, next); | |
1918 | ||
dd41f596 | 1919 | if (unlikely(!prev->mm)) { |
1da177e4 | 1920 | prev->active_mm = NULL; |
1da177e4 LT |
1921 | rq->prev_mm = oldmm; |
1922 | } | |
3a5f5e48 IM |
1923 | /* |
1924 | * Since the runqueue lock will be released by the next | |
1925 | * task (which is an invalid locking op but in the case | |
1926 | * of the scheduler it's an obvious special-case), so we | |
1927 | * do an early lockdep release here: | |
1928 | */ | |
1929 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 1930 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 1931 | #endif |
1da177e4 LT |
1932 | |
1933 | /* Here we just switch the register state and the stack. */ | |
1934 | switch_to(prev, next, prev); | |
1935 | ||
dd41f596 IM |
1936 | barrier(); |
1937 | /* | |
1938 | * this_rq must be evaluated again because prev may have moved | |
1939 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
1940 | * frame will be invalid. | |
1941 | */ | |
1942 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
1943 | } |
1944 | ||
1945 | /* | |
1946 | * nr_running, nr_uninterruptible and nr_context_switches: | |
1947 | * | |
1948 | * externally visible scheduler statistics: current number of runnable | |
1949 | * threads, current number of uninterruptible-sleeping threads, total | |
1950 | * number of context switches performed since bootup. | |
1951 | */ | |
1952 | unsigned long nr_running(void) | |
1953 | { | |
1954 | unsigned long i, sum = 0; | |
1955 | ||
1956 | for_each_online_cpu(i) | |
1957 | sum += cpu_rq(i)->nr_running; | |
1958 | ||
1959 | return sum; | |
1960 | } | |
1961 | ||
1962 | unsigned long nr_uninterruptible(void) | |
1963 | { | |
1964 | unsigned long i, sum = 0; | |
1965 | ||
0a945022 | 1966 | for_each_possible_cpu(i) |
1da177e4 LT |
1967 | sum += cpu_rq(i)->nr_uninterruptible; |
1968 | ||
1969 | /* | |
1970 | * Since we read the counters lockless, it might be slightly | |
1971 | * inaccurate. Do not allow it to go below zero though: | |
1972 | */ | |
1973 | if (unlikely((long)sum < 0)) | |
1974 | sum = 0; | |
1975 | ||
1976 | return sum; | |
1977 | } | |
1978 | ||
1979 | unsigned long long nr_context_switches(void) | |
1980 | { | |
cc94abfc SR |
1981 | int i; |
1982 | unsigned long long sum = 0; | |
1da177e4 | 1983 | |
0a945022 | 1984 | for_each_possible_cpu(i) |
1da177e4 LT |
1985 | sum += cpu_rq(i)->nr_switches; |
1986 | ||
1987 | return sum; | |
1988 | } | |
1989 | ||
1990 | unsigned long nr_iowait(void) | |
1991 | { | |
1992 | unsigned long i, sum = 0; | |
1993 | ||
0a945022 | 1994 | for_each_possible_cpu(i) |
1da177e4 LT |
1995 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
1996 | ||
1997 | return sum; | |
1998 | } | |
1999 | ||
db1b1fef JS |
2000 | unsigned long nr_active(void) |
2001 | { | |
2002 | unsigned long i, running = 0, uninterruptible = 0; | |
2003 | ||
2004 | for_each_online_cpu(i) { | |
2005 | running += cpu_rq(i)->nr_running; | |
2006 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2007 | } | |
2008 | ||
2009 | if (unlikely((long)uninterruptible < 0)) | |
2010 | uninterruptible = 0; | |
2011 | ||
2012 | return running + uninterruptible; | |
2013 | } | |
2014 | ||
48f24c4d | 2015 | /* |
dd41f596 IM |
2016 | * Update rq->cpu_load[] statistics. This function is usually called every |
2017 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2018 | */ |
dd41f596 | 2019 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2020 | { |
495eca49 | 2021 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2022 | int i, scale; |
2023 | ||
2024 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2025 | |
2026 | /* Update our load: */ | |
2027 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2028 | unsigned long old_load, new_load; | |
2029 | ||
2030 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2031 | ||
2032 | old_load = this_rq->cpu_load[i]; | |
2033 | new_load = this_load; | |
a25707f3 IM |
2034 | /* |
2035 | * Round up the averaging division if load is increasing. This | |
2036 | * prevents us from getting stuck on 9 if the load is 10, for | |
2037 | * example. | |
2038 | */ | |
2039 | if (new_load > old_load) | |
2040 | new_load += scale-1; | |
dd41f596 IM |
2041 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2042 | } | |
48f24c4d IM |
2043 | } |
2044 | ||
dd41f596 IM |
2045 | #ifdef CONFIG_SMP |
2046 | ||
1da177e4 LT |
2047 | /* |
2048 | * double_rq_lock - safely lock two runqueues | |
2049 | * | |
2050 | * Note this does not disable interrupts like task_rq_lock, | |
2051 | * you need to do so manually before calling. | |
2052 | */ | |
70b97a7f | 2053 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2054 | __acquires(rq1->lock) |
2055 | __acquires(rq2->lock) | |
2056 | { | |
054b9108 | 2057 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2058 | if (rq1 == rq2) { |
2059 | spin_lock(&rq1->lock); | |
2060 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2061 | } else { | |
c96d145e | 2062 | if (rq1 < rq2) { |
1da177e4 LT |
2063 | spin_lock(&rq1->lock); |
2064 | spin_lock(&rq2->lock); | |
2065 | } else { | |
2066 | spin_lock(&rq2->lock); | |
2067 | spin_lock(&rq1->lock); | |
2068 | } | |
2069 | } | |
6e82a3be IM |
2070 | update_rq_clock(rq1); |
2071 | update_rq_clock(rq2); | |
1da177e4 LT |
2072 | } |
2073 | ||
2074 | /* | |
2075 | * double_rq_unlock - safely unlock two runqueues | |
2076 | * | |
2077 | * Note this does not restore interrupts like task_rq_unlock, | |
2078 | * you need to do so manually after calling. | |
2079 | */ | |
70b97a7f | 2080 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2081 | __releases(rq1->lock) |
2082 | __releases(rq2->lock) | |
2083 | { | |
2084 | spin_unlock(&rq1->lock); | |
2085 | if (rq1 != rq2) | |
2086 | spin_unlock(&rq2->lock); | |
2087 | else | |
2088 | __release(rq2->lock); | |
2089 | } | |
2090 | ||
2091 | /* | |
2092 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2093 | */ | |
70b97a7f | 2094 | static void double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2095 | __releases(this_rq->lock) |
2096 | __acquires(busiest->lock) | |
2097 | __acquires(this_rq->lock) | |
2098 | { | |
054b9108 KK |
2099 | if (unlikely(!irqs_disabled())) { |
2100 | /* printk() doesn't work good under rq->lock */ | |
2101 | spin_unlock(&this_rq->lock); | |
2102 | BUG_ON(1); | |
2103 | } | |
1da177e4 | 2104 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2105 | if (busiest < this_rq) { |
1da177e4 LT |
2106 | spin_unlock(&this_rq->lock); |
2107 | spin_lock(&busiest->lock); | |
2108 | spin_lock(&this_rq->lock); | |
2109 | } else | |
2110 | spin_lock(&busiest->lock); | |
2111 | } | |
2112 | } | |
2113 | ||
1da177e4 LT |
2114 | /* |
2115 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2116 | * This is accomplished by forcing the cpu_allowed mask to only | |
2117 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then | |
2118 | * the cpu_allowed mask is restored. | |
2119 | */ | |
36c8b586 | 2120 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2121 | { |
70b97a7f | 2122 | struct migration_req req; |
1da177e4 | 2123 | unsigned long flags; |
70b97a7f | 2124 | struct rq *rq; |
1da177e4 LT |
2125 | |
2126 | rq = task_rq_lock(p, &flags); | |
2127 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2128 | || unlikely(cpu_is_offline(dest_cpu))) | |
2129 | goto out; | |
2130 | ||
2131 | /* force the process onto the specified CPU */ | |
2132 | if (migrate_task(p, dest_cpu, &req)) { | |
2133 | /* Need to wait for migration thread (might exit: take ref). */ | |
2134 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2135 | |
1da177e4 LT |
2136 | get_task_struct(mt); |
2137 | task_rq_unlock(rq, &flags); | |
2138 | wake_up_process(mt); | |
2139 | put_task_struct(mt); | |
2140 | wait_for_completion(&req.done); | |
36c8b586 | 2141 | |
1da177e4 LT |
2142 | return; |
2143 | } | |
2144 | out: | |
2145 | task_rq_unlock(rq, &flags); | |
2146 | } | |
2147 | ||
2148 | /* | |
476d139c NP |
2149 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2150 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2151 | */ |
2152 | void sched_exec(void) | |
2153 | { | |
1da177e4 | 2154 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2155 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2156 | put_cpu(); |
476d139c NP |
2157 | if (new_cpu != this_cpu) |
2158 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2159 | } |
2160 | ||
2161 | /* | |
2162 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2163 | * Both runqueues must be locked. | |
2164 | */ | |
dd41f596 IM |
2165 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2166 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2167 | { |
2e1cb74a | 2168 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2169 | set_task_cpu(p, this_cpu); |
dd41f596 | 2170 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2171 | /* |
2172 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2173 | * to be always true for them. | |
2174 | */ | |
dd41f596 | 2175 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2176 | } |
2177 | ||
2178 | /* | |
2179 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2180 | */ | |
858119e1 | 2181 | static |
70b97a7f | 2182 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2183 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2184 | int *all_pinned) |
1da177e4 LT |
2185 | { |
2186 | /* | |
2187 | * We do not migrate tasks that are: | |
2188 | * 1) running (obviously), or | |
2189 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2190 | * 3) are cache-hot on their current CPU. | |
2191 | */ | |
cc367732 IM |
2192 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2193 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2194 | return 0; |
cc367732 | 2195 | } |
81026794 NP |
2196 | *all_pinned = 0; |
2197 | ||
cc367732 IM |
2198 | if (task_running(rq, p)) { |
2199 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2200 | return 0; |
cc367732 | 2201 | } |
1da177e4 | 2202 | |
da84d961 IM |
2203 | /* |
2204 | * Aggressive migration if: | |
2205 | * 1) task is cache cold, or | |
2206 | * 2) too many balance attempts have failed. | |
2207 | */ | |
2208 | ||
6bc1665b IM |
2209 | if (!task_hot(p, rq->clock, sd) || |
2210 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2211 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2212 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2213 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2214 | schedstat_inc(p, se.nr_forced_migrations); |
2215 | } | |
da84d961 IM |
2216 | #endif |
2217 | return 1; | |
2218 | } | |
2219 | ||
cc367732 IM |
2220 | if (task_hot(p, rq->clock, sd)) { |
2221 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2222 | return 0; |
cc367732 | 2223 | } |
1da177e4 LT |
2224 | return 1; |
2225 | } | |
2226 | ||
dd41f596 | 2227 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, |
2dd73a4f | 2228 | unsigned long max_nr_move, unsigned long max_load_move, |
d15bcfdb | 2229 | struct sched_domain *sd, enum cpu_idle_type idle, |
dd41f596 | 2230 | int *all_pinned, unsigned long *load_moved, |
a4ac01c3 | 2231 | int *this_best_prio, struct rq_iterator *iterator) |
1da177e4 | 2232 | { |
dd41f596 IM |
2233 | int pulled = 0, pinned = 0, skip_for_load; |
2234 | struct task_struct *p; | |
2235 | long rem_load_move = max_load_move; | |
1da177e4 | 2236 | |
2dd73a4f | 2237 | if (max_nr_move == 0 || max_load_move == 0) |
1da177e4 LT |
2238 | goto out; |
2239 | ||
81026794 NP |
2240 | pinned = 1; |
2241 | ||
1da177e4 | 2242 | /* |
dd41f596 | 2243 | * Start the load-balancing iterator: |
1da177e4 | 2244 | */ |
dd41f596 IM |
2245 | p = iterator->start(iterator->arg); |
2246 | next: | |
2247 | if (!p) | |
1da177e4 | 2248 | goto out; |
50ddd969 PW |
2249 | /* |
2250 | * To help distribute high priority tasks accross CPUs we don't | |
2251 | * skip a task if it will be the highest priority task (i.e. smallest | |
2252 | * prio value) on its new queue regardless of its load weight | |
2253 | */ | |
dd41f596 IM |
2254 | skip_for_load = (p->se.load.weight >> 1) > rem_load_move + |
2255 | SCHED_LOAD_SCALE_FUZZ; | |
a4ac01c3 | 2256 | if ((skip_for_load && p->prio >= *this_best_prio) || |
dd41f596 | 2257 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2258 | p = iterator->next(iterator->arg); |
2259 | goto next; | |
1da177e4 LT |
2260 | } |
2261 | ||
dd41f596 | 2262 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2263 | pulled++; |
dd41f596 | 2264 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2265 | |
2dd73a4f PW |
2266 | /* |
2267 | * We only want to steal up to the prescribed number of tasks | |
2268 | * and the prescribed amount of weighted load. | |
2269 | */ | |
2270 | if (pulled < max_nr_move && rem_load_move > 0) { | |
a4ac01c3 PW |
2271 | if (p->prio < *this_best_prio) |
2272 | *this_best_prio = p->prio; | |
dd41f596 IM |
2273 | p = iterator->next(iterator->arg); |
2274 | goto next; | |
1da177e4 LT |
2275 | } |
2276 | out: | |
2277 | /* | |
2278 | * Right now, this is the only place pull_task() is called, | |
2279 | * so we can safely collect pull_task() stats here rather than | |
2280 | * inside pull_task(). | |
2281 | */ | |
2282 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2283 | |
2284 | if (all_pinned) | |
2285 | *all_pinned = pinned; | |
dd41f596 | 2286 | *load_moved = max_load_move - rem_load_move; |
1da177e4 LT |
2287 | return pulled; |
2288 | } | |
2289 | ||
dd41f596 | 2290 | /* |
43010659 PW |
2291 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2292 | * this_rq, as part of a balancing operation within domain "sd". | |
2293 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2294 | * |
2295 | * Called with both runqueues locked. | |
2296 | */ | |
2297 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2298 | unsigned long max_load_move, |
dd41f596 IM |
2299 | struct sched_domain *sd, enum cpu_idle_type idle, |
2300 | int *all_pinned) | |
2301 | { | |
5522d5d5 | 2302 | const struct sched_class *class = sched_class_highest; |
43010659 | 2303 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2304 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
2305 | |
2306 | do { | |
43010659 PW |
2307 | total_load_moved += |
2308 | class->load_balance(this_rq, this_cpu, busiest, | |
2309 | ULONG_MAX, max_load_move - total_load_moved, | |
a4ac01c3 | 2310 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 2311 | class = class->next; |
43010659 | 2312 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 2313 | |
43010659 PW |
2314 | return total_load_moved > 0; |
2315 | } | |
2316 | ||
2317 | /* | |
2318 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
2319 | * part of active balancing operations within "domain". | |
2320 | * Returns 1 if successful and 0 otherwise. | |
2321 | * | |
2322 | * Called with both runqueues locked. | |
2323 | */ | |
2324 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2325 | struct sched_domain *sd, enum cpu_idle_type idle) | |
2326 | { | |
5522d5d5 | 2327 | const struct sched_class *class; |
a4ac01c3 | 2328 | int this_best_prio = MAX_PRIO; |
43010659 PW |
2329 | |
2330 | for (class = sched_class_highest; class; class = class->next) | |
2331 | if (class->load_balance(this_rq, this_cpu, busiest, | |
a4ac01c3 PW |
2332 | 1, ULONG_MAX, sd, idle, NULL, |
2333 | &this_best_prio)) | |
43010659 PW |
2334 | return 1; |
2335 | ||
2336 | return 0; | |
dd41f596 IM |
2337 | } |
2338 | ||
1da177e4 LT |
2339 | /* |
2340 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2341 | * domain. It calculates and returns the amount of weighted load which |
2342 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2343 | */ |
2344 | static struct sched_group * | |
2345 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 IM |
2346 | unsigned long *imbalance, enum cpu_idle_type idle, |
2347 | int *sd_idle, cpumask_t *cpus, int *balance) | |
1da177e4 LT |
2348 | { |
2349 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2350 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2351 | unsigned long max_pull; |
2dd73a4f PW |
2352 | unsigned long busiest_load_per_task, busiest_nr_running; |
2353 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 2354 | int load_idx, group_imb = 0; |
5c45bf27 SS |
2355 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2356 | int power_savings_balance = 1; | |
2357 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2358 | unsigned long min_nr_running = ULONG_MAX; | |
2359 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2360 | #endif | |
1da177e4 LT |
2361 | |
2362 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2363 | busiest_load_per_task = busiest_nr_running = 0; |
2364 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2365 | if (idle == CPU_NOT_IDLE) |
7897986b | 2366 | load_idx = sd->busy_idx; |
d15bcfdb | 2367 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2368 | load_idx = sd->newidle_idx; |
2369 | else | |
2370 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2371 | |
2372 | do { | |
908a7c1b | 2373 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
2374 | int local_group; |
2375 | int i; | |
908a7c1b | 2376 | int __group_imb = 0; |
783609c6 | 2377 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2378 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2379 | |
2380 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2381 | ||
783609c6 SS |
2382 | if (local_group) |
2383 | balance_cpu = first_cpu(group->cpumask); | |
2384 | ||
1da177e4 | 2385 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2386 | sum_weighted_load = sum_nr_running = avg_load = 0; |
908a7c1b KC |
2387 | max_cpu_load = 0; |
2388 | min_cpu_load = ~0UL; | |
1da177e4 LT |
2389 | |
2390 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2391 | struct rq *rq; |
2392 | ||
2393 | if (!cpu_isset(i, *cpus)) | |
2394 | continue; | |
2395 | ||
2396 | rq = cpu_rq(i); | |
2dd73a4f | 2397 | |
9439aab8 | 2398 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
2399 | *sd_idle = 0; |
2400 | ||
1da177e4 | 2401 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2402 | if (local_group) { |
2403 | if (idle_cpu(i) && !first_idle_cpu) { | |
2404 | first_idle_cpu = 1; | |
2405 | balance_cpu = i; | |
2406 | } | |
2407 | ||
a2000572 | 2408 | load = target_load(i, load_idx); |
908a7c1b | 2409 | } else { |
a2000572 | 2410 | load = source_load(i, load_idx); |
908a7c1b KC |
2411 | if (load > max_cpu_load) |
2412 | max_cpu_load = load; | |
2413 | if (min_cpu_load > load) | |
2414 | min_cpu_load = load; | |
2415 | } | |
1da177e4 LT |
2416 | |
2417 | avg_load += load; | |
2dd73a4f | 2418 | sum_nr_running += rq->nr_running; |
dd41f596 | 2419 | sum_weighted_load += weighted_cpuload(i); |
1da177e4 LT |
2420 | } |
2421 | ||
783609c6 SS |
2422 | /* |
2423 | * First idle cpu or the first cpu(busiest) in this sched group | |
2424 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
2425 | * domains. In the newly idle case, we will allow all the cpu's |
2426 | * to do the newly idle load balance. | |
783609c6 | 2427 | */ |
9439aab8 SS |
2428 | if (idle != CPU_NEWLY_IDLE && local_group && |
2429 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
2430 | *balance = 0; |
2431 | goto ret; | |
2432 | } | |
2433 | ||
1da177e4 | 2434 | total_load += avg_load; |
5517d86b | 2435 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2436 | |
2437 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2438 | avg_load = sg_div_cpu_power(group, |
2439 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2440 | |
908a7c1b KC |
2441 | if ((max_cpu_load - min_cpu_load) > SCHED_LOAD_SCALE) |
2442 | __group_imb = 1; | |
2443 | ||
5517d86b | 2444 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2445 | |
1da177e4 LT |
2446 | if (local_group) { |
2447 | this_load = avg_load; | |
2448 | this = group; | |
2dd73a4f PW |
2449 | this_nr_running = sum_nr_running; |
2450 | this_load_per_task = sum_weighted_load; | |
2451 | } else if (avg_load > max_load && | |
908a7c1b | 2452 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
2453 | max_load = avg_load; |
2454 | busiest = group; | |
2dd73a4f PW |
2455 | busiest_nr_running = sum_nr_running; |
2456 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 2457 | group_imb = __group_imb; |
1da177e4 | 2458 | } |
5c45bf27 SS |
2459 | |
2460 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2461 | /* | |
2462 | * Busy processors will not participate in power savings | |
2463 | * balance. | |
2464 | */ | |
dd41f596 IM |
2465 | if (idle == CPU_NOT_IDLE || |
2466 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
2467 | goto group_next; | |
5c45bf27 SS |
2468 | |
2469 | /* | |
2470 | * If the local group is idle or completely loaded | |
2471 | * no need to do power savings balance at this domain | |
2472 | */ | |
2473 | if (local_group && (this_nr_running >= group_capacity || | |
2474 | !this_nr_running)) | |
2475 | power_savings_balance = 0; | |
2476 | ||
dd41f596 | 2477 | /* |
5c45bf27 SS |
2478 | * If a group is already running at full capacity or idle, |
2479 | * don't include that group in power savings calculations | |
dd41f596 IM |
2480 | */ |
2481 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 2482 | || !sum_nr_running) |
dd41f596 | 2483 | goto group_next; |
5c45bf27 | 2484 | |
dd41f596 | 2485 | /* |
5c45bf27 | 2486 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
2487 | * This is the group from where we need to pick up the load |
2488 | * for saving power | |
2489 | */ | |
2490 | if ((sum_nr_running < min_nr_running) || | |
2491 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
2492 | first_cpu(group->cpumask) < |
2493 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
2494 | group_min = group; |
2495 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
2496 | min_load_per_task = sum_weighted_load / |
2497 | sum_nr_running; | |
dd41f596 | 2498 | } |
5c45bf27 | 2499 | |
dd41f596 | 2500 | /* |
5c45bf27 | 2501 | * Calculate the group which is almost near its |
dd41f596 IM |
2502 | * capacity but still has some space to pick up some load |
2503 | * from other group and save more power | |
2504 | */ | |
2505 | if (sum_nr_running <= group_capacity - 1) { | |
2506 | if (sum_nr_running > leader_nr_running || | |
2507 | (sum_nr_running == leader_nr_running && | |
2508 | first_cpu(group->cpumask) > | |
2509 | first_cpu(group_leader->cpumask))) { | |
2510 | group_leader = group; | |
2511 | leader_nr_running = sum_nr_running; | |
2512 | } | |
48f24c4d | 2513 | } |
5c45bf27 SS |
2514 | group_next: |
2515 | #endif | |
1da177e4 LT |
2516 | group = group->next; |
2517 | } while (group != sd->groups); | |
2518 | ||
2dd73a4f | 2519 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
2520 | goto out_balanced; |
2521 | ||
2522 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
2523 | ||
2524 | if (this_load >= avg_load || | |
2525 | 100*max_load <= sd->imbalance_pct*this_load) | |
2526 | goto out_balanced; | |
2527 | ||
2dd73a4f | 2528 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
2529 | if (group_imb) |
2530 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
2531 | ||
1da177e4 LT |
2532 | /* |
2533 | * We're trying to get all the cpus to the average_load, so we don't | |
2534 | * want to push ourselves above the average load, nor do we wish to | |
2535 | * reduce the max loaded cpu below the average load, as either of these | |
2536 | * actions would just result in more rebalancing later, and ping-pong | |
2537 | * tasks around. Thus we look for the minimum possible imbalance. | |
2538 | * Negative imbalances (*we* are more loaded than anyone else) will | |
2539 | * be counted as no imbalance for these purposes -- we can't fix that | |
2540 | * by pulling tasks to us. Be careful of negative numbers as they'll | |
2541 | * appear as very large values with unsigned longs. | |
2542 | */ | |
2dd73a4f PW |
2543 | if (max_load <= busiest_load_per_task) |
2544 | goto out_balanced; | |
2545 | ||
2546 | /* | |
2547 | * In the presence of smp nice balancing, certain scenarios can have | |
2548 | * max load less than avg load(as we skip the groups at or below | |
2549 | * its cpu_power, while calculating max_load..) | |
2550 | */ | |
2551 | if (max_load < avg_load) { | |
2552 | *imbalance = 0; | |
2553 | goto small_imbalance; | |
2554 | } | |
0c117f1b SS |
2555 | |
2556 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 2557 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 2558 | |
1da177e4 | 2559 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
2560 | *imbalance = min(max_pull * busiest->__cpu_power, |
2561 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
2562 | / SCHED_LOAD_SCALE; |
2563 | ||
2dd73a4f PW |
2564 | /* |
2565 | * if *imbalance is less than the average load per runnable task | |
2566 | * there is no gaurantee that any tasks will be moved so we'll have | |
2567 | * a think about bumping its value to force at least one task to be | |
2568 | * moved | |
2569 | */ | |
7fd0d2dd | 2570 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 2571 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
2572 | unsigned int imbn; |
2573 | ||
2574 | small_imbalance: | |
2575 | pwr_move = pwr_now = 0; | |
2576 | imbn = 2; | |
2577 | if (this_nr_running) { | |
2578 | this_load_per_task /= this_nr_running; | |
2579 | if (busiest_load_per_task > this_load_per_task) | |
2580 | imbn = 1; | |
2581 | } else | |
2582 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 2583 | |
dd41f596 IM |
2584 | if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >= |
2585 | busiest_load_per_task * imbn) { | |
2dd73a4f | 2586 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2587 | return busiest; |
2588 | } | |
2589 | ||
2590 | /* | |
2591 | * OK, we don't have enough imbalance to justify moving tasks, | |
2592 | * however we may be able to increase total CPU power used by | |
2593 | * moving them. | |
2594 | */ | |
2595 | ||
5517d86b ED |
2596 | pwr_now += busiest->__cpu_power * |
2597 | min(busiest_load_per_task, max_load); | |
2598 | pwr_now += this->__cpu_power * | |
2599 | min(this_load_per_task, this_load); | |
1da177e4 LT |
2600 | pwr_now /= SCHED_LOAD_SCALE; |
2601 | ||
2602 | /* Amount of load we'd subtract */ | |
5517d86b ED |
2603 | tmp = sg_div_cpu_power(busiest, |
2604 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 2605 | if (max_load > tmp) |
5517d86b | 2606 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 2607 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
2608 | |
2609 | /* Amount of load we'd add */ | |
5517d86b | 2610 | if (max_load * busiest->__cpu_power < |
33859f7f | 2611 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
2612 | tmp = sg_div_cpu_power(this, |
2613 | max_load * busiest->__cpu_power); | |
1da177e4 | 2614 | else |
5517d86b ED |
2615 | tmp = sg_div_cpu_power(this, |
2616 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
2617 | pwr_move += this->__cpu_power * | |
2618 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
2619 | pwr_move /= SCHED_LOAD_SCALE; |
2620 | ||
2621 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
2622 | if (pwr_move > pwr_now) |
2623 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
2624 | } |
2625 | ||
1da177e4 LT |
2626 | return busiest; |
2627 | ||
2628 | out_balanced: | |
5c45bf27 | 2629 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 2630 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 2631 | goto ret; |
1da177e4 | 2632 | |
5c45bf27 SS |
2633 | if (this == group_leader && group_leader != group_min) { |
2634 | *imbalance = min_load_per_task; | |
2635 | return group_min; | |
2636 | } | |
5c45bf27 | 2637 | #endif |
783609c6 | 2638 | ret: |
1da177e4 LT |
2639 | *imbalance = 0; |
2640 | return NULL; | |
2641 | } | |
2642 | ||
2643 | /* | |
2644 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
2645 | */ | |
70b97a7f | 2646 | static struct rq * |
d15bcfdb | 2647 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
0a2966b4 | 2648 | unsigned long imbalance, cpumask_t *cpus) |
1da177e4 | 2649 | { |
70b97a7f | 2650 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 2651 | unsigned long max_load = 0; |
1da177e4 LT |
2652 | int i; |
2653 | ||
2654 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 2655 | unsigned long wl; |
0a2966b4 CL |
2656 | |
2657 | if (!cpu_isset(i, *cpus)) | |
2658 | continue; | |
2659 | ||
48f24c4d | 2660 | rq = cpu_rq(i); |
dd41f596 | 2661 | wl = weighted_cpuload(i); |
2dd73a4f | 2662 | |
dd41f596 | 2663 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 2664 | continue; |
1da177e4 | 2665 | |
dd41f596 IM |
2666 | if (wl > max_load) { |
2667 | max_load = wl; | |
48f24c4d | 2668 | busiest = rq; |
1da177e4 LT |
2669 | } |
2670 | } | |
2671 | ||
2672 | return busiest; | |
2673 | } | |
2674 | ||
77391d71 NP |
2675 | /* |
2676 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
2677 | * so long as it is large enough. | |
2678 | */ | |
2679 | #define MAX_PINNED_INTERVAL 512 | |
2680 | ||
1da177e4 LT |
2681 | /* |
2682 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2683 | * tasks if there is an imbalance. | |
1da177e4 | 2684 | */ |
70b97a7f | 2685 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 2686 | struct sched_domain *sd, enum cpu_idle_type idle, |
783609c6 | 2687 | int *balance) |
1da177e4 | 2688 | { |
43010659 | 2689 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 2690 | struct sched_group *group; |
1da177e4 | 2691 | unsigned long imbalance; |
70b97a7f | 2692 | struct rq *busiest; |
0a2966b4 | 2693 | cpumask_t cpus = CPU_MASK_ALL; |
fe2eea3f | 2694 | unsigned long flags; |
5969fe06 | 2695 | |
89c4710e SS |
2696 | /* |
2697 | * When power savings policy is enabled for the parent domain, idle | |
2698 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 2699 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 2700 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 2701 | */ |
d15bcfdb | 2702 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2703 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2704 | sd_idle = 1; |
1da177e4 | 2705 | |
2d72376b | 2706 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 2707 | |
0a2966b4 CL |
2708 | redo: |
2709 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
783609c6 SS |
2710 | &cpus, balance); |
2711 | ||
06066714 | 2712 | if (*balance == 0) |
783609c6 | 2713 | goto out_balanced; |
783609c6 | 2714 | |
1da177e4 LT |
2715 | if (!group) { |
2716 | schedstat_inc(sd, lb_nobusyg[idle]); | |
2717 | goto out_balanced; | |
2718 | } | |
2719 | ||
0a2966b4 | 2720 | busiest = find_busiest_queue(group, idle, imbalance, &cpus); |
1da177e4 LT |
2721 | if (!busiest) { |
2722 | schedstat_inc(sd, lb_nobusyq[idle]); | |
2723 | goto out_balanced; | |
2724 | } | |
2725 | ||
db935dbd | 2726 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
2727 | |
2728 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
2729 | ||
43010659 | 2730 | ld_moved = 0; |
1da177e4 LT |
2731 | if (busiest->nr_running > 1) { |
2732 | /* | |
2733 | * Attempt to move tasks. If find_busiest_group has found | |
2734 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 2735 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
2736 | * correctly treated as an imbalance. |
2737 | */ | |
fe2eea3f | 2738 | local_irq_save(flags); |
e17224bf | 2739 | double_rq_lock(this_rq, busiest); |
43010659 | 2740 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 2741 | imbalance, sd, idle, &all_pinned); |
e17224bf | 2742 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 2743 | local_irq_restore(flags); |
81026794 | 2744 | |
46cb4b7c SS |
2745 | /* |
2746 | * some other cpu did the load balance for us. | |
2747 | */ | |
43010659 | 2748 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
2749 | resched_cpu(this_cpu); |
2750 | ||
81026794 | 2751 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 CL |
2752 | if (unlikely(all_pinned)) { |
2753 | cpu_clear(cpu_of(busiest), cpus); | |
2754 | if (!cpus_empty(cpus)) | |
2755 | goto redo; | |
81026794 | 2756 | goto out_balanced; |
0a2966b4 | 2757 | } |
1da177e4 | 2758 | } |
81026794 | 2759 | |
43010659 | 2760 | if (!ld_moved) { |
1da177e4 LT |
2761 | schedstat_inc(sd, lb_failed[idle]); |
2762 | sd->nr_balance_failed++; | |
2763 | ||
2764 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 2765 | |
fe2eea3f | 2766 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
2767 | |
2768 | /* don't kick the migration_thread, if the curr | |
2769 | * task on busiest cpu can't be moved to this_cpu | |
2770 | */ | |
2771 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 2772 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
2773 | all_pinned = 1; |
2774 | goto out_one_pinned; | |
2775 | } | |
2776 | ||
1da177e4 LT |
2777 | if (!busiest->active_balance) { |
2778 | busiest->active_balance = 1; | |
2779 | busiest->push_cpu = this_cpu; | |
81026794 | 2780 | active_balance = 1; |
1da177e4 | 2781 | } |
fe2eea3f | 2782 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 2783 | if (active_balance) |
1da177e4 LT |
2784 | wake_up_process(busiest->migration_thread); |
2785 | ||
2786 | /* | |
2787 | * We've kicked active balancing, reset the failure | |
2788 | * counter. | |
2789 | */ | |
39507451 | 2790 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 2791 | } |
81026794 | 2792 | } else |
1da177e4 LT |
2793 | sd->nr_balance_failed = 0; |
2794 | ||
81026794 | 2795 | if (likely(!active_balance)) { |
1da177e4 LT |
2796 | /* We were unbalanced, so reset the balancing interval */ |
2797 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
2798 | } else { |
2799 | /* | |
2800 | * If we've begun active balancing, start to back off. This | |
2801 | * case may not be covered by the all_pinned logic if there | |
2802 | * is only 1 task on the busy runqueue (because we don't call | |
2803 | * move_tasks). | |
2804 | */ | |
2805 | if (sd->balance_interval < sd->max_interval) | |
2806 | sd->balance_interval *= 2; | |
1da177e4 LT |
2807 | } |
2808 | ||
43010659 | 2809 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2810 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2811 | return -1; |
43010659 | 2812 | return ld_moved; |
1da177e4 LT |
2813 | |
2814 | out_balanced: | |
1da177e4 LT |
2815 | schedstat_inc(sd, lb_balanced[idle]); |
2816 | ||
16cfb1c0 | 2817 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
2818 | |
2819 | out_one_pinned: | |
1da177e4 | 2820 | /* tune up the balancing interval */ |
77391d71 NP |
2821 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
2822 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
2823 | sd->balance_interval *= 2; |
2824 | ||
48f24c4d | 2825 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2826 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2827 | return -1; |
1da177e4 LT |
2828 | return 0; |
2829 | } | |
2830 | ||
2831 | /* | |
2832 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2833 | * tasks if there is an imbalance. | |
2834 | * | |
d15bcfdb | 2835 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
2836 | * this_rq is locked. |
2837 | */ | |
48f24c4d | 2838 | static int |
70b97a7f | 2839 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
2840 | { |
2841 | struct sched_group *group; | |
70b97a7f | 2842 | struct rq *busiest = NULL; |
1da177e4 | 2843 | unsigned long imbalance; |
43010659 | 2844 | int ld_moved = 0; |
5969fe06 | 2845 | int sd_idle = 0; |
969bb4e4 | 2846 | int all_pinned = 0; |
0a2966b4 | 2847 | cpumask_t cpus = CPU_MASK_ALL; |
5969fe06 | 2848 | |
89c4710e SS |
2849 | /* |
2850 | * When power savings policy is enabled for the parent domain, idle | |
2851 | * sibling can pick up load irrespective of busy siblings. In this case, | |
2852 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 2853 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
2854 | */ |
2855 | if (sd->flags & SD_SHARE_CPUPOWER && | |
2856 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 2857 | sd_idle = 1; |
1da177e4 | 2858 | |
2d72376b | 2859 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 2860 | redo: |
d15bcfdb | 2861 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
783609c6 | 2862 | &sd_idle, &cpus, NULL); |
1da177e4 | 2863 | if (!group) { |
d15bcfdb | 2864 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2865 | goto out_balanced; |
1da177e4 LT |
2866 | } |
2867 | ||
d15bcfdb | 2868 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, |
0a2966b4 | 2869 | &cpus); |
db935dbd | 2870 | if (!busiest) { |
d15bcfdb | 2871 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2872 | goto out_balanced; |
1da177e4 LT |
2873 | } |
2874 | ||
db935dbd NP |
2875 | BUG_ON(busiest == this_rq); |
2876 | ||
d15bcfdb | 2877 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 2878 | |
43010659 | 2879 | ld_moved = 0; |
d6d5cfaf NP |
2880 | if (busiest->nr_running > 1) { |
2881 | /* Attempt to move tasks */ | |
2882 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
2883 | /* this_rq->clock is already updated */ |
2884 | update_rq_clock(busiest); | |
43010659 | 2885 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
2886 | imbalance, sd, CPU_NEWLY_IDLE, |
2887 | &all_pinned); | |
d6d5cfaf | 2888 | spin_unlock(&busiest->lock); |
0a2966b4 | 2889 | |
969bb4e4 | 2890 | if (unlikely(all_pinned)) { |
0a2966b4 CL |
2891 | cpu_clear(cpu_of(busiest), cpus); |
2892 | if (!cpus_empty(cpus)) | |
2893 | goto redo; | |
2894 | } | |
d6d5cfaf NP |
2895 | } |
2896 | ||
43010659 | 2897 | if (!ld_moved) { |
d15bcfdb | 2898 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
2899 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
2900 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
2901 | return -1; |
2902 | } else | |
16cfb1c0 | 2903 | sd->nr_balance_failed = 0; |
1da177e4 | 2904 | |
43010659 | 2905 | return ld_moved; |
16cfb1c0 NP |
2906 | |
2907 | out_balanced: | |
d15bcfdb | 2908 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 2909 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2910 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2911 | return -1; |
16cfb1c0 | 2912 | sd->nr_balance_failed = 0; |
48f24c4d | 2913 | |
16cfb1c0 | 2914 | return 0; |
1da177e4 LT |
2915 | } |
2916 | ||
2917 | /* | |
2918 | * idle_balance is called by schedule() if this_cpu is about to become | |
2919 | * idle. Attempts to pull tasks from other CPUs. | |
2920 | */ | |
70b97a7f | 2921 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
2922 | { |
2923 | struct sched_domain *sd; | |
dd41f596 IM |
2924 | int pulled_task = -1; |
2925 | unsigned long next_balance = jiffies + HZ; | |
1da177e4 LT |
2926 | |
2927 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
2928 | unsigned long interval; |
2929 | ||
2930 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
2931 | continue; | |
2932 | ||
2933 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 2934 | /* If we've pulled tasks over stop searching: */ |
1bd77f2d | 2935 | pulled_task = load_balance_newidle(this_cpu, |
92c4ca5c CL |
2936 | this_rq, sd); |
2937 | ||
2938 | interval = msecs_to_jiffies(sd->balance_interval); | |
2939 | if (time_after(next_balance, sd->last_balance + interval)) | |
2940 | next_balance = sd->last_balance + interval; | |
2941 | if (pulled_task) | |
2942 | break; | |
1da177e4 | 2943 | } |
dd41f596 | 2944 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
2945 | /* |
2946 | * We are going idle. next_balance may be set based on | |
2947 | * a busy processor. So reset next_balance. | |
2948 | */ | |
2949 | this_rq->next_balance = next_balance; | |
dd41f596 | 2950 | } |
1da177e4 LT |
2951 | } |
2952 | ||
2953 | /* | |
2954 | * active_load_balance is run by migration threads. It pushes running tasks | |
2955 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
2956 | * running on each physical CPU where possible, and avoids physical / | |
2957 | * logical imbalances. | |
2958 | * | |
2959 | * Called with busiest_rq locked. | |
2960 | */ | |
70b97a7f | 2961 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 2962 | { |
39507451 | 2963 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
2964 | struct sched_domain *sd; |
2965 | struct rq *target_rq; | |
39507451 | 2966 | |
48f24c4d | 2967 | /* Is there any task to move? */ |
39507451 | 2968 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
2969 | return; |
2970 | ||
2971 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
2972 | |
2973 | /* | |
39507451 NP |
2974 | * This condition is "impossible", if it occurs |
2975 | * we need to fix it. Originally reported by | |
2976 | * Bjorn Helgaas on a 128-cpu setup. | |
1da177e4 | 2977 | */ |
39507451 | 2978 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 2979 | |
39507451 NP |
2980 | /* move a task from busiest_rq to target_rq */ |
2981 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
2982 | update_rq_clock(busiest_rq); |
2983 | update_rq_clock(target_rq); | |
39507451 NP |
2984 | |
2985 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 2986 | for_each_domain(target_cpu, sd) { |
39507451 | 2987 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 2988 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 2989 | break; |
c96d145e | 2990 | } |
39507451 | 2991 | |
48f24c4d | 2992 | if (likely(sd)) { |
2d72376b | 2993 | schedstat_inc(sd, alb_count); |
39507451 | 2994 | |
43010659 PW |
2995 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
2996 | sd, CPU_IDLE)) | |
48f24c4d IM |
2997 | schedstat_inc(sd, alb_pushed); |
2998 | else | |
2999 | schedstat_inc(sd, alb_failed); | |
3000 | } | |
39507451 | 3001 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
3002 | } |
3003 | ||
46cb4b7c SS |
3004 | #ifdef CONFIG_NO_HZ |
3005 | static struct { | |
3006 | atomic_t load_balancer; | |
3007 | cpumask_t cpu_mask; | |
3008 | } nohz ____cacheline_aligned = { | |
3009 | .load_balancer = ATOMIC_INIT(-1), | |
3010 | .cpu_mask = CPU_MASK_NONE, | |
3011 | }; | |
3012 | ||
7835b98b | 3013 | /* |
46cb4b7c SS |
3014 | * This routine will try to nominate the ilb (idle load balancing) |
3015 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3016 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3017 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3018 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3019 | * arrives... | |
3020 | * | |
3021 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3022 | * for idle load balancing. ilb owner will still be part of | |
3023 | * nohz.cpu_mask.. | |
7835b98b | 3024 | * |
46cb4b7c SS |
3025 | * While stopping the tick, this cpu will become the ilb owner if there |
3026 | * is no other owner. And will be the owner till that cpu becomes busy | |
3027 | * or if all cpus in the system stop their ticks at which point | |
3028 | * there is no need for ilb owner. | |
3029 | * | |
3030 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3031 | * next busy scheduler_tick() | |
3032 | */ | |
3033 | int select_nohz_load_balancer(int stop_tick) | |
3034 | { | |
3035 | int cpu = smp_processor_id(); | |
3036 | ||
3037 | if (stop_tick) { | |
3038 | cpu_set(cpu, nohz.cpu_mask); | |
3039 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3040 | ||
3041 | /* | |
3042 | * If we are going offline and still the leader, give up! | |
3043 | */ | |
3044 | if (cpu_is_offline(cpu) && | |
3045 | atomic_read(&nohz.load_balancer) == cpu) { | |
3046 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3047 | BUG(); | |
3048 | return 0; | |
3049 | } | |
3050 | ||
3051 | /* time for ilb owner also to sleep */ | |
3052 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3053 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3054 | atomic_set(&nohz.load_balancer, -1); | |
3055 | return 0; | |
3056 | } | |
3057 | ||
3058 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3059 | /* make me the ilb owner */ | |
3060 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3061 | return 1; | |
3062 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3063 | return 1; | |
3064 | } else { | |
3065 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3066 | return 0; | |
3067 | ||
3068 | cpu_clear(cpu, nohz.cpu_mask); | |
3069 | ||
3070 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3071 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3072 | BUG(); | |
3073 | } | |
3074 | return 0; | |
3075 | } | |
3076 | #endif | |
3077 | ||
3078 | static DEFINE_SPINLOCK(balancing); | |
3079 | ||
3080 | /* | |
7835b98b CL |
3081 | * It checks each scheduling domain to see if it is due to be balanced, |
3082 | * and initiates a balancing operation if so. | |
3083 | * | |
3084 | * Balancing parameters are set up in arch_init_sched_domains. | |
3085 | */ | |
a9957449 | 3086 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3087 | { |
46cb4b7c SS |
3088 | int balance = 1; |
3089 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3090 | unsigned long interval; |
3091 | struct sched_domain *sd; | |
46cb4b7c | 3092 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3093 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3094 | int update_next_balance = 0; |
1da177e4 | 3095 | |
46cb4b7c | 3096 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3097 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3098 | continue; | |
3099 | ||
3100 | interval = sd->balance_interval; | |
d15bcfdb | 3101 | if (idle != CPU_IDLE) |
1da177e4 LT |
3102 | interval *= sd->busy_factor; |
3103 | ||
3104 | /* scale ms to jiffies */ | |
3105 | interval = msecs_to_jiffies(interval); | |
3106 | if (unlikely(!interval)) | |
3107 | interval = 1; | |
dd41f596 IM |
3108 | if (interval > HZ*NR_CPUS/10) |
3109 | interval = HZ*NR_CPUS/10; | |
3110 | ||
1da177e4 | 3111 | |
08c183f3 CL |
3112 | if (sd->flags & SD_SERIALIZE) { |
3113 | if (!spin_trylock(&balancing)) | |
3114 | goto out; | |
3115 | } | |
3116 | ||
c9819f45 | 3117 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
46cb4b7c | 3118 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
3119 | /* |
3120 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3121 | * longer idle, or one of our SMT siblings is |
3122 | * not idle. | |
3123 | */ | |
d15bcfdb | 3124 | idle = CPU_NOT_IDLE; |
1da177e4 | 3125 | } |
1bd77f2d | 3126 | sd->last_balance = jiffies; |
1da177e4 | 3127 | } |
08c183f3 CL |
3128 | if (sd->flags & SD_SERIALIZE) |
3129 | spin_unlock(&balancing); | |
3130 | out: | |
f549da84 | 3131 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3132 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3133 | update_next_balance = 1; |
3134 | } | |
783609c6 SS |
3135 | |
3136 | /* | |
3137 | * Stop the load balance at this level. There is another | |
3138 | * CPU in our sched group which is doing load balancing more | |
3139 | * actively. | |
3140 | */ | |
3141 | if (!balance) | |
3142 | break; | |
1da177e4 | 3143 | } |
f549da84 SS |
3144 | |
3145 | /* | |
3146 | * next_balance will be updated only when there is a need. | |
3147 | * When the cpu is attached to null domain for ex, it will not be | |
3148 | * updated. | |
3149 | */ | |
3150 | if (likely(update_next_balance)) | |
3151 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3152 | } |
3153 | ||
3154 | /* | |
3155 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3156 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3157 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3158 | */ | |
3159 | static void run_rebalance_domains(struct softirq_action *h) | |
3160 | { | |
dd41f596 IM |
3161 | int this_cpu = smp_processor_id(); |
3162 | struct rq *this_rq = cpu_rq(this_cpu); | |
3163 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3164 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3165 | |
dd41f596 | 3166 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3167 | |
3168 | #ifdef CONFIG_NO_HZ | |
3169 | /* | |
3170 | * If this cpu is the owner for idle load balancing, then do the | |
3171 | * balancing on behalf of the other idle cpus whose ticks are | |
3172 | * stopped. | |
3173 | */ | |
dd41f596 IM |
3174 | if (this_rq->idle_at_tick && |
3175 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3176 | cpumask_t cpus = nohz.cpu_mask; |
3177 | struct rq *rq; | |
3178 | int balance_cpu; | |
3179 | ||
dd41f596 | 3180 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3181 | for_each_cpu_mask(balance_cpu, cpus) { |
3182 | /* | |
3183 | * If this cpu gets work to do, stop the load balancing | |
3184 | * work being done for other cpus. Next load | |
3185 | * balancing owner will pick it up. | |
3186 | */ | |
3187 | if (need_resched()) | |
3188 | break; | |
3189 | ||
de0cf899 | 3190 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3191 | |
3192 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3193 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3194 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3195 | } |
3196 | } | |
3197 | #endif | |
3198 | } | |
3199 | ||
3200 | /* | |
3201 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3202 | * | |
3203 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3204 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3205 | * if the whole system is idle. | |
3206 | */ | |
dd41f596 | 3207 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3208 | { |
46cb4b7c SS |
3209 | #ifdef CONFIG_NO_HZ |
3210 | /* | |
3211 | * If we were in the nohz mode recently and busy at the current | |
3212 | * scheduler tick, then check if we need to nominate new idle | |
3213 | * load balancer. | |
3214 | */ | |
3215 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3216 | rq->in_nohz_recently = 0; | |
3217 | ||
3218 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3219 | cpu_clear(cpu, nohz.cpu_mask); | |
3220 | atomic_set(&nohz.load_balancer, -1); | |
3221 | } | |
3222 | ||
3223 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3224 | /* | |
3225 | * simple selection for now: Nominate the | |
3226 | * first cpu in the nohz list to be the next | |
3227 | * ilb owner. | |
3228 | * | |
3229 | * TBD: Traverse the sched domains and nominate | |
3230 | * the nearest cpu in the nohz.cpu_mask. | |
3231 | */ | |
3232 | int ilb = first_cpu(nohz.cpu_mask); | |
3233 | ||
3234 | if (ilb != NR_CPUS) | |
3235 | resched_cpu(ilb); | |
3236 | } | |
3237 | } | |
3238 | ||
3239 | /* | |
3240 | * If this cpu is idle and doing idle load balancing for all the | |
3241 | * cpus with ticks stopped, is it time for that to stop? | |
3242 | */ | |
3243 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3244 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3245 | resched_cpu(cpu); | |
3246 | return; | |
3247 | } | |
3248 | ||
3249 | /* | |
3250 | * If this cpu is idle and the idle load balancing is done by | |
3251 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3252 | */ | |
3253 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3254 | cpu_isset(cpu, nohz.cpu_mask)) | |
3255 | return; | |
3256 | #endif | |
3257 | if (time_after_eq(jiffies, rq->next_balance)) | |
3258 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3259 | } |
dd41f596 IM |
3260 | |
3261 | #else /* CONFIG_SMP */ | |
3262 | ||
1da177e4 LT |
3263 | /* |
3264 | * on UP we do not need to balance between CPUs: | |
3265 | */ | |
70b97a7f | 3266 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3267 | { |
3268 | } | |
dd41f596 IM |
3269 | |
3270 | /* Avoid "used but not defined" warning on UP */ | |
3271 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3272 | unsigned long max_nr_move, unsigned long max_load_move, | |
3273 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3274 | int *all_pinned, unsigned long *load_moved, | |
a4ac01c3 | 3275 | int *this_best_prio, struct rq_iterator *iterator) |
dd41f596 IM |
3276 | { |
3277 | *load_moved = 0; | |
3278 | ||
3279 | return 0; | |
3280 | } | |
3281 | ||
1da177e4 LT |
3282 | #endif |
3283 | ||
1da177e4 LT |
3284 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3285 | ||
3286 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3287 | ||
3288 | /* | |
41b86e9c IM |
3289 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
3290 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 3291 | */ |
41b86e9c | 3292 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 3293 | { |
1da177e4 | 3294 | unsigned long flags; |
41b86e9c IM |
3295 | u64 ns, delta_exec; |
3296 | struct rq *rq; | |
48f24c4d | 3297 | |
41b86e9c IM |
3298 | rq = task_rq_lock(p, &flags); |
3299 | ns = p->se.sum_exec_runtime; | |
3300 | if (rq->curr == p) { | |
a8e504d2 IM |
3301 | update_rq_clock(rq); |
3302 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
3303 | if ((s64)delta_exec > 0) |
3304 | ns += delta_exec; | |
3305 | } | |
3306 | task_rq_unlock(rq, &flags); | |
48f24c4d | 3307 | |
1da177e4 LT |
3308 | return ns; |
3309 | } | |
3310 | ||
1da177e4 LT |
3311 | /* |
3312 | * Account user cpu time to a process. | |
3313 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
3314 | * @cputime: the cpu time spent in user space since the last update |
3315 | */ | |
3316 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3317 | { | |
3318 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3319 | cputime64_t tmp; | |
62d0df64 | 3320 | struct rq *rq = this_rq(); |
1da177e4 LT |
3321 | |
3322 | p->utime = cputime_add(p->utime, cputime); | |
3323 | ||
62d0df64 PM |
3324 | if (p != rq->idle) |
3325 | cpuacct_charge(p, cputime); | |
3326 | ||
1da177e4 LT |
3327 | /* Add user time to cpustat. */ |
3328 | tmp = cputime_to_cputime64(cputime); | |
3329 | if (TASK_NICE(p) > 0) | |
3330 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3331 | else | |
3332 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3333 | } | |
3334 | ||
94886b84 LV |
3335 | /* |
3336 | * Account guest cpu time to a process. | |
3337 | * @p: the process that the cpu time gets accounted to | |
3338 | * @cputime: the cpu time spent in virtual machine since the last update | |
3339 | */ | |
3340 | void account_guest_time(struct task_struct *p, cputime_t cputime) | |
3341 | { | |
3342 | cputime64_t tmp; | |
3343 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3344 | ||
3345 | tmp = cputime_to_cputime64(cputime); | |
3346 | ||
3347 | p->utime = cputime_add(p->utime, cputime); | |
3348 | p->gtime = cputime_add(p->gtime, cputime); | |
3349 | ||
3350 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3351 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3352 | } | |
3353 | ||
c66f08be MN |
3354 | /* |
3355 | * Account scaled user cpu time to a process. | |
3356 | * @p: the process that the cpu time gets accounted to | |
3357 | * @cputime: the cpu time spent in user space since the last update | |
3358 | */ | |
3359 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
3360 | { | |
3361 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
3362 | } | |
3363 | ||
1da177e4 LT |
3364 | /* |
3365 | * Account system cpu time to a process. | |
3366 | * @p: the process that the cpu time gets accounted to | |
3367 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3368 | * @cputime: the cpu time spent in kernel space since the last update | |
3369 | */ | |
3370 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3371 | cputime_t cputime) | |
3372 | { | |
3373 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3374 | struct rq *rq = this_rq(); |
1da177e4 LT |
3375 | cputime64_t tmp; |
3376 | ||
94886b84 LV |
3377 | if (p->flags & PF_VCPU) { |
3378 | account_guest_time(p, cputime); | |
94886b84 LV |
3379 | return; |
3380 | } | |
3381 | ||
1da177e4 LT |
3382 | p->stime = cputime_add(p->stime, cputime); |
3383 | ||
3384 | /* Add system time to cpustat. */ | |
3385 | tmp = cputime_to_cputime64(cputime); | |
3386 | if (hardirq_count() - hardirq_offset) | |
3387 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3388 | else if (softirq_count()) | |
3389 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
62d0df64 | 3390 | else if (p != rq->idle) { |
1da177e4 | 3391 | cpustat->system = cputime64_add(cpustat->system, tmp); |
62d0df64 PM |
3392 | cpuacct_charge(p, cputime); |
3393 | } else if (atomic_read(&rq->nr_iowait) > 0) | |
1da177e4 LT |
3394 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
3395 | else | |
3396 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3397 | /* Account for system time used */ | |
3398 | acct_update_integrals(p); | |
1da177e4 LT |
3399 | } |
3400 | ||
c66f08be MN |
3401 | /* |
3402 | * Account scaled system cpu time to a process. | |
3403 | * @p: the process that the cpu time gets accounted to | |
3404 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3405 | * @cputime: the cpu time spent in kernel space since the last update | |
3406 | */ | |
3407 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
3408 | { | |
3409 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
3410 | } | |
3411 | ||
1da177e4 LT |
3412 | /* |
3413 | * Account for involuntary wait time. | |
3414 | * @p: the process from which the cpu time has been stolen | |
3415 | * @steal: the cpu time spent in involuntary wait | |
3416 | */ | |
3417 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3418 | { | |
3419 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3420 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3421 | struct rq *rq = this_rq(); |
1da177e4 LT |
3422 | |
3423 | if (p == rq->idle) { | |
3424 | p->stime = cputime_add(p->stime, steal); | |
3425 | if (atomic_read(&rq->nr_iowait) > 0) | |
3426 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3427 | else | |
3428 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
62d0df64 | 3429 | } else { |
1da177e4 | 3430 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
62d0df64 PM |
3431 | cpuacct_charge(p, -tmp); |
3432 | } | |
1da177e4 LT |
3433 | } |
3434 | ||
7835b98b CL |
3435 | /* |
3436 | * This function gets called by the timer code, with HZ frequency. | |
3437 | * We call it with interrupts disabled. | |
3438 | * | |
3439 | * It also gets called by the fork code, when changing the parent's | |
3440 | * timeslices. | |
3441 | */ | |
3442 | void scheduler_tick(void) | |
3443 | { | |
7835b98b CL |
3444 | int cpu = smp_processor_id(); |
3445 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3446 | struct task_struct *curr = rq->curr; |
529c7726 | 3447 | u64 next_tick = rq->tick_timestamp + TICK_NSEC; |
dd41f596 IM |
3448 | |
3449 | spin_lock(&rq->lock); | |
546fe3c9 | 3450 | __update_rq_clock(rq); |
529c7726 IM |
3451 | /* |
3452 | * Let rq->clock advance by at least TICK_NSEC: | |
3453 | */ | |
3454 | if (unlikely(rq->clock < next_tick)) | |
3455 | rq->clock = next_tick; | |
3456 | rq->tick_timestamp = rq->clock; | |
f1a438d8 | 3457 | update_cpu_load(rq); |
dd41f596 IM |
3458 | if (curr != rq->idle) /* FIXME: needed? */ |
3459 | curr->sched_class->task_tick(rq, curr); | |
dd41f596 | 3460 | spin_unlock(&rq->lock); |
7835b98b | 3461 | |
e418e1c2 | 3462 | #ifdef CONFIG_SMP |
dd41f596 IM |
3463 | rq->idle_at_tick = idle_cpu(cpu); |
3464 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3465 | #endif |
1da177e4 LT |
3466 | } |
3467 | ||
1da177e4 LT |
3468 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3469 | ||
3470 | void fastcall add_preempt_count(int val) | |
3471 | { | |
3472 | /* | |
3473 | * Underflow? | |
3474 | */ | |
9a11b49a IM |
3475 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3476 | return; | |
1da177e4 LT |
3477 | preempt_count() += val; |
3478 | /* | |
3479 | * Spinlock count overflowing soon? | |
3480 | */ | |
33859f7f MOS |
3481 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3482 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3483 | } |
3484 | EXPORT_SYMBOL(add_preempt_count); | |
3485 | ||
3486 | void fastcall sub_preempt_count(int val) | |
3487 | { | |
3488 | /* | |
3489 | * Underflow? | |
3490 | */ | |
9a11b49a IM |
3491 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3492 | return; | |
1da177e4 LT |
3493 | /* |
3494 | * Is the spinlock portion underflowing? | |
3495 | */ | |
9a11b49a IM |
3496 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3497 | !(preempt_count() & PREEMPT_MASK))) | |
3498 | return; | |
3499 | ||
1da177e4 LT |
3500 | preempt_count() -= val; |
3501 | } | |
3502 | EXPORT_SYMBOL(sub_preempt_count); | |
3503 | ||
3504 | #endif | |
3505 | ||
3506 | /* | |
dd41f596 | 3507 | * Print scheduling while atomic bug: |
1da177e4 | 3508 | */ |
dd41f596 | 3509 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3510 | { |
838225b4 SS |
3511 | struct pt_regs *regs = get_irq_regs(); |
3512 | ||
3513 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
3514 | prev->comm, prev->pid, preempt_count()); | |
3515 | ||
dd41f596 IM |
3516 | debug_show_held_locks(prev); |
3517 | if (irqs_disabled()) | |
3518 | print_irqtrace_events(prev); | |
838225b4 SS |
3519 | |
3520 | if (regs) | |
3521 | show_regs(regs); | |
3522 | else | |
3523 | dump_stack(); | |
dd41f596 | 3524 | } |
1da177e4 | 3525 | |
dd41f596 IM |
3526 | /* |
3527 | * Various schedule()-time debugging checks and statistics: | |
3528 | */ | |
3529 | static inline void schedule_debug(struct task_struct *prev) | |
3530 | { | |
1da177e4 LT |
3531 | /* |
3532 | * Test if we are atomic. Since do_exit() needs to call into | |
3533 | * schedule() atomically, we ignore that path for now. | |
3534 | * Otherwise, whine if we are scheduling when we should not be. | |
3535 | */ | |
dd41f596 IM |
3536 | if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state)) |
3537 | __schedule_bug(prev); | |
3538 | ||
1da177e4 LT |
3539 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3540 | ||
2d72376b | 3541 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3542 | #ifdef CONFIG_SCHEDSTATS |
3543 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3544 | schedstat_inc(this_rq(), bkl_count); |
3545 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3546 | } |
3547 | #endif | |
dd41f596 IM |
3548 | } |
3549 | ||
3550 | /* | |
3551 | * Pick up the highest-prio task: | |
3552 | */ | |
3553 | static inline struct task_struct * | |
ff95f3df | 3554 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 3555 | { |
5522d5d5 | 3556 | const struct sched_class *class; |
dd41f596 | 3557 | struct task_struct *p; |
1da177e4 LT |
3558 | |
3559 | /* | |
dd41f596 IM |
3560 | * Optimization: we know that if all tasks are in |
3561 | * the fair class we can call that function directly: | |
1da177e4 | 3562 | */ |
dd41f596 | 3563 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3564 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3565 | if (likely(p)) |
3566 | return p; | |
1da177e4 LT |
3567 | } |
3568 | ||
dd41f596 IM |
3569 | class = sched_class_highest; |
3570 | for ( ; ; ) { | |
fb8d4724 | 3571 | p = class->pick_next_task(rq); |
dd41f596 IM |
3572 | if (p) |
3573 | return p; | |
3574 | /* | |
3575 | * Will never be NULL as the idle class always | |
3576 | * returns a non-NULL p: | |
3577 | */ | |
3578 | class = class->next; | |
3579 | } | |
3580 | } | |
1da177e4 | 3581 | |
dd41f596 IM |
3582 | /* |
3583 | * schedule() is the main scheduler function. | |
3584 | */ | |
3585 | asmlinkage void __sched schedule(void) | |
3586 | { | |
3587 | struct task_struct *prev, *next; | |
3588 | long *switch_count; | |
3589 | struct rq *rq; | |
dd41f596 IM |
3590 | int cpu; |
3591 | ||
3592 | need_resched: | |
3593 | preempt_disable(); | |
3594 | cpu = smp_processor_id(); | |
3595 | rq = cpu_rq(cpu); | |
3596 | rcu_qsctr_inc(cpu); | |
3597 | prev = rq->curr; | |
3598 | switch_count = &prev->nivcsw; | |
3599 | ||
3600 | release_kernel_lock(prev); | |
3601 | need_resched_nonpreemptible: | |
3602 | ||
3603 | schedule_debug(prev); | |
1da177e4 | 3604 | |
1e819950 IM |
3605 | /* |
3606 | * Do the rq-clock update outside the rq lock: | |
3607 | */ | |
3608 | local_irq_disable(); | |
c1b3da3e | 3609 | __update_rq_clock(rq); |
1e819950 IM |
3610 | spin_lock(&rq->lock); |
3611 | clear_tsk_need_resched(prev); | |
1da177e4 | 3612 | |
1da177e4 | 3613 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
1da177e4 | 3614 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && |
dd41f596 | 3615 | unlikely(signal_pending(prev)))) { |
1da177e4 | 3616 | prev->state = TASK_RUNNING; |
dd41f596 | 3617 | } else { |
2e1cb74a | 3618 | deactivate_task(rq, prev, 1); |
1da177e4 | 3619 | } |
dd41f596 | 3620 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3621 | } |
3622 | ||
dd41f596 | 3623 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3624 | idle_balance(cpu, rq); |
1da177e4 | 3625 | |
31ee529c | 3626 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 3627 | next = pick_next_task(rq, prev); |
1da177e4 LT |
3628 | |
3629 | sched_info_switch(prev, next); | |
dd41f596 | 3630 | |
1da177e4 | 3631 | if (likely(prev != next)) { |
1da177e4 LT |
3632 | rq->nr_switches++; |
3633 | rq->curr = next; | |
3634 | ++*switch_count; | |
3635 | ||
dd41f596 | 3636 | context_switch(rq, prev, next); /* unlocks the rq */ |
1da177e4 LT |
3637 | } else |
3638 | spin_unlock_irq(&rq->lock); | |
3639 | ||
dd41f596 IM |
3640 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
3641 | cpu = smp_processor_id(); | |
3642 | rq = cpu_rq(cpu); | |
1da177e4 | 3643 | goto need_resched_nonpreemptible; |
dd41f596 | 3644 | } |
1da177e4 LT |
3645 | preempt_enable_no_resched(); |
3646 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3647 | goto need_resched; | |
3648 | } | |
1da177e4 LT |
3649 | EXPORT_SYMBOL(schedule); |
3650 | ||
3651 | #ifdef CONFIG_PREEMPT | |
3652 | /* | |
2ed6e34f | 3653 | * this is the entry point to schedule() from in-kernel preemption |
1da177e4 LT |
3654 | * off of preempt_enable. Kernel preemptions off return from interrupt |
3655 | * occur there and call schedule directly. | |
3656 | */ | |
3657 | asmlinkage void __sched preempt_schedule(void) | |
3658 | { | |
3659 | struct thread_info *ti = current_thread_info(); | |
3660 | #ifdef CONFIG_PREEMPT_BKL | |
3661 | struct task_struct *task = current; | |
3662 | int saved_lock_depth; | |
3663 | #endif | |
3664 | /* | |
3665 | * If there is a non-zero preempt_count or interrupts are disabled, | |
3666 | * we do not want to preempt the current task. Just return.. | |
3667 | */ | |
beed33a8 | 3668 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3669 | return; |
3670 | ||
3a5c359a AK |
3671 | do { |
3672 | add_preempt_count(PREEMPT_ACTIVE); | |
3673 | ||
3674 | /* | |
3675 | * We keep the big kernel semaphore locked, but we | |
3676 | * clear ->lock_depth so that schedule() doesnt | |
3677 | * auto-release the semaphore: | |
3678 | */ | |
1da177e4 | 3679 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a AK |
3680 | saved_lock_depth = task->lock_depth; |
3681 | task->lock_depth = -1; | |
1da177e4 | 3682 | #endif |
3a5c359a | 3683 | schedule(); |
1da177e4 | 3684 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a | 3685 | task->lock_depth = saved_lock_depth; |
1da177e4 | 3686 | #endif |
3a5c359a | 3687 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3688 | |
3a5c359a AK |
3689 | /* |
3690 | * Check again in case we missed a preemption opportunity | |
3691 | * between schedule and now. | |
3692 | */ | |
3693 | barrier(); | |
3694 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 3695 | } |
1da177e4 LT |
3696 | EXPORT_SYMBOL(preempt_schedule); |
3697 | ||
3698 | /* | |
2ed6e34f | 3699 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3700 | * off of irq context. |
3701 | * Note, that this is called and return with irqs disabled. This will | |
3702 | * protect us against recursive calling from irq. | |
3703 | */ | |
3704 | asmlinkage void __sched preempt_schedule_irq(void) | |
3705 | { | |
3706 | struct thread_info *ti = current_thread_info(); | |
3707 | #ifdef CONFIG_PREEMPT_BKL | |
3708 | struct task_struct *task = current; | |
3709 | int saved_lock_depth; | |
3710 | #endif | |
2ed6e34f | 3711 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3712 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3713 | ||
3a5c359a AK |
3714 | do { |
3715 | add_preempt_count(PREEMPT_ACTIVE); | |
3716 | ||
3717 | /* | |
3718 | * We keep the big kernel semaphore locked, but we | |
3719 | * clear ->lock_depth so that schedule() doesnt | |
3720 | * auto-release the semaphore: | |
3721 | */ | |
1da177e4 | 3722 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a AK |
3723 | saved_lock_depth = task->lock_depth; |
3724 | task->lock_depth = -1; | |
1da177e4 | 3725 | #endif |
3a5c359a AK |
3726 | local_irq_enable(); |
3727 | schedule(); | |
3728 | local_irq_disable(); | |
1da177e4 | 3729 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a | 3730 | task->lock_depth = saved_lock_depth; |
1da177e4 | 3731 | #endif |
3a5c359a | 3732 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3733 | |
3a5c359a AK |
3734 | /* |
3735 | * Check again in case we missed a preemption opportunity | |
3736 | * between schedule and now. | |
3737 | */ | |
3738 | barrier(); | |
3739 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
3740 | } |
3741 | ||
3742 | #endif /* CONFIG_PREEMPT */ | |
3743 | ||
95cdf3b7 IM |
3744 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
3745 | void *key) | |
1da177e4 | 3746 | { |
48f24c4d | 3747 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 3748 | } |
1da177e4 LT |
3749 | EXPORT_SYMBOL(default_wake_function); |
3750 | ||
3751 | /* | |
3752 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just | |
3753 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
3754 | * number) then we wake all the non-exclusive tasks and one exclusive task. | |
3755 | * | |
3756 | * There are circumstances in which we can try to wake a task which has already | |
3757 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns | |
3758 | * zero in this (rare) case, and we handle it by continuing to scan the queue. | |
3759 | */ | |
3760 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
3761 | int nr_exclusive, int sync, void *key) | |
3762 | { | |
2e45874c | 3763 | wait_queue_t *curr, *next; |
1da177e4 | 3764 | |
2e45874c | 3765 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3766 | unsigned flags = curr->flags; |
3767 | ||
1da177e4 | 3768 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 3769 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3770 | break; |
3771 | } | |
3772 | } | |
3773 | ||
3774 | /** | |
3775 | * __wake_up - wake up threads blocked on a waitqueue. | |
3776 | * @q: the waitqueue | |
3777 | * @mode: which threads | |
3778 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3779 | * @key: is directly passed to the wakeup function |
1da177e4 LT |
3780 | */ |
3781 | void fastcall __wake_up(wait_queue_head_t *q, unsigned int mode, | |
95cdf3b7 | 3782 | int nr_exclusive, void *key) |
1da177e4 LT |
3783 | { |
3784 | unsigned long flags; | |
3785 | ||
3786 | spin_lock_irqsave(&q->lock, flags); | |
3787 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3788 | spin_unlock_irqrestore(&q->lock, flags); | |
3789 | } | |
1da177e4 LT |
3790 | EXPORT_SYMBOL(__wake_up); |
3791 | ||
3792 | /* | |
3793 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3794 | */ | |
3795 | void fastcall __wake_up_locked(wait_queue_head_t *q, unsigned int mode) | |
3796 | { | |
3797 | __wake_up_common(q, mode, 1, 0, NULL); | |
3798 | } | |
3799 | ||
3800 | /** | |
67be2dd1 | 3801 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3802 | * @q: the waitqueue |
3803 | * @mode: which threads | |
3804 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
3805 | * | |
3806 | * The sync wakeup differs that the waker knows that it will schedule | |
3807 | * away soon, so while the target thread will be woken up, it will not | |
3808 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3809 | * with each other. This can prevent needless bouncing between CPUs. | |
3810 | * | |
3811 | * On UP it can prevent extra preemption. | |
3812 | */ | |
95cdf3b7 IM |
3813 | void fastcall |
3814 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
1da177e4 LT |
3815 | { |
3816 | unsigned long flags; | |
3817 | int sync = 1; | |
3818 | ||
3819 | if (unlikely(!q)) | |
3820 | return; | |
3821 | ||
3822 | if (unlikely(!nr_exclusive)) | |
3823 | sync = 0; | |
3824 | ||
3825 | spin_lock_irqsave(&q->lock, flags); | |
3826 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
3827 | spin_unlock_irqrestore(&q->lock, flags); | |
3828 | } | |
3829 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
3830 | ||
b15136e9 | 3831 | void complete(struct completion *x) |
1da177e4 LT |
3832 | { |
3833 | unsigned long flags; | |
3834 | ||
3835 | spin_lock_irqsave(&x->wait.lock, flags); | |
3836 | x->done++; | |
3837 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3838 | 1, 0, NULL); | |
3839 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3840 | } | |
3841 | EXPORT_SYMBOL(complete); | |
3842 | ||
b15136e9 | 3843 | void complete_all(struct completion *x) |
1da177e4 LT |
3844 | { |
3845 | unsigned long flags; | |
3846 | ||
3847 | spin_lock_irqsave(&x->wait.lock, flags); | |
3848 | x->done += UINT_MAX/2; | |
3849 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3850 | 0, 0, NULL); | |
3851 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3852 | } | |
3853 | EXPORT_SYMBOL(complete_all); | |
3854 | ||
8cbbe86d AK |
3855 | static inline long __sched |
3856 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3857 | { |
1da177e4 LT |
3858 | if (!x->done) { |
3859 | DECLARE_WAITQUEUE(wait, current); | |
3860 | ||
3861 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3862 | __add_wait_queue_tail(&x->wait, &wait); | |
3863 | do { | |
8cbbe86d AK |
3864 | if (state == TASK_INTERRUPTIBLE && |
3865 | signal_pending(current)) { | |
3866 | __remove_wait_queue(&x->wait, &wait); | |
3867 | return -ERESTARTSYS; | |
3868 | } | |
3869 | __set_current_state(state); | |
1da177e4 LT |
3870 | spin_unlock_irq(&x->wait.lock); |
3871 | timeout = schedule_timeout(timeout); | |
3872 | spin_lock_irq(&x->wait.lock); | |
3873 | if (!timeout) { | |
3874 | __remove_wait_queue(&x->wait, &wait); | |
8cbbe86d | 3875 | return timeout; |
1da177e4 LT |
3876 | } |
3877 | } while (!x->done); | |
3878 | __remove_wait_queue(&x->wait, &wait); | |
3879 | } | |
3880 | x->done--; | |
1da177e4 LT |
3881 | return timeout; |
3882 | } | |
1da177e4 | 3883 | |
8cbbe86d AK |
3884 | static long __sched |
3885 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3886 | { |
1da177e4 LT |
3887 | might_sleep(); |
3888 | ||
3889 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 3890 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 3891 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3892 | return timeout; |
3893 | } | |
1da177e4 | 3894 | |
b15136e9 | 3895 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
3896 | { |
3897 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 3898 | } |
8cbbe86d | 3899 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 3900 | |
b15136e9 | 3901 | unsigned long __sched |
8cbbe86d | 3902 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 3903 | { |
8cbbe86d | 3904 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 3905 | } |
8cbbe86d | 3906 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 3907 | |
8cbbe86d | 3908 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 3909 | { |
51e97990 AK |
3910 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
3911 | if (t == -ERESTARTSYS) | |
3912 | return t; | |
3913 | return 0; | |
0fec171c | 3914 | } |
8cbbe86d | 3915 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 3916 | |
b15136e9 | 3917 | unsigned long __sched |
8cbbe86d AK |
3918 | wait_for_completion_interruptible_timeout(struct completion *x, |
3919 | unsigned long timeout) | |
0fec171c | 3920 | { |
8cbbe86d | 3921 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 3922 | } |
8cbbe86d | 3923 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 3924 | |
8cbbe86d AK |
3925 | static long __sched |
3926 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 3927 | { |
0fec171c IM |
3928 | unsigned long flags; |
3929 | wait_queue_t wait; | |
3930 | ||
3931 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 3932 | |
8cbbe86d | 3933 | __set_current_state(state); |
1da177e4 | 3934 | |
8cbbe86d AK |
3935 | spin_lock_irqsave(&q->lock, flags); |
3936 | __add_wait_queue(q, &wait); | |
3937 | spin_unlock(&q->lock); | |
3938 | timeout = schedule_timeout(timeout); | |
3939 | spin_lock_irq(&q->lock); | |
3940 | __remove_wait_queue(q, &wait); | |
3941 | spin_unlock_irqrestore(&q->lock, flags); | |
3942 | ||
3943 | return timeout; | |
3944 | } | |
3945 | ||
3946 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
3947 | { | |
3948 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 3949 | } |
1da177e4 LT |
3950 | EXPORT_SYMBOL(interruptible_sleep_on); |
3951 | ||
0fec171c | 3952 | long __sched |
95cdf3b7 | 3953 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3954 | { |
8cbbe86d | 3955 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 3956 | } |
1da177e4 LT |
3957 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3958 | ||
0fec171c | 3959 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3960 | { |
8cbbe86d | 3961 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 3962 | } |
1da177e4 LT |
3963 | EXPORT_SYMBOL(sleep_on); |
3964 | ||
0fec171c | 3965 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3966 | { |
8cbbe86d | 3967 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 3968 | } |
1da177e4 LT |
3969 | EXPORT_SYMBOL(sleep_on_timeout); |
3970 | ||
b29739f9 IM |
3971 | #ifdef CONFIG_RT_MUTEXES |
3972 | ||
3973 | /* | |
3974 | * rt_mutex_setprio - set the current priority of a task | |
3975 | * @p: task | |
3976 | * @prio: prio value (kernel-internal form) | |
3977 | * | |
3978 | * This function changes the 'effective' priority of a task. It does | |
3979 | * not touch ->normal_prio like __setscheduler(). | |
3980 | * | |
3981 | * Used by the rt_mutex code to implement priority inheritance logic. | |
3982 | */ | |
36c8b586 | 3983 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
3984 | { |
3985 | unsigned long flags; | |
83b699ed | 3986 | int oldprio, on_rq, running; |
70b97a7f | 3987 | struct rq *rq; |
b29739f9 IM |
3988 | |
3989 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
3990 | ||
3991 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 3992 | update_rq_clock(rq); |
b29739f9 | 3993 | |
d5f9f942 | 3994 | oldprio = p->prio; |
dd41f596 | 3995 | on_rq = p->se.on_rq; |
83b699ed SV |
3996 | running = task_running(rq, p); |
3997 | if (on_rq) { | |
69be72c1 | 3998 | dequeue_task(rq, p, 0); |
83b699ed SV |
3999 | if (running) |
4000 | p->sched_class->put_prev_task(rq, p); | |
4001 | } | |
dd41f596 IM |
4002 | |
4003 | if (rt_prio(prio)) | |
4004 | p->sched_class = &rt_sched_class; | |
4005 | else | |
4006 | p->sched_class = &fair_sched_class; | |
4007 | ||
b29739f9 IM |
4008 | p->prio = prio; |
4009 | ||
dd41f596 | 4010 | if (on_rq) { |
83b699ed SV |
4011 | if (running) |
4012 | p->sched_class->set_curr_task(rq); | |
8159f87e | 4013 | enqueue_task(rq, p, 0); |
b29739f9 IM |
4014 | /* |
4015 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4016 | * our priority decreased, or if we are not currently running on |
4017 | * this runqueue and our priority is higher than the current's | |
b29739f9 | 4018 | */ |
83b699ed | 4019 | if (running) { |
d5f9f942 AM |
4020 | if (p->prio > oldprio) |
4021 | resched_task(rq->curr); | |
dd41f596 IM |
4022 | } else { |
4023 | check_preempt_curr(rq, p); | |
4024 | } | |
b29739f9 IM |
4025 | } |
4026 | task_rq_unlock(rq, &flags); | |
4027 | } | |
4028 | ||
4029 | #endif | |
4030 | ||
36c8b586 | 4031 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4032 | { |
dd41f596 | 4033 | int old_prio, delta, on_rq; |
1da177e4 | 4034 | unsigned long flags; |
70b97a7f | 4035 | struct rq *rq; |
1da177e4 LT |
4036 | |
4037 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4038 | return; | |
4039 | /* | |
4040 | * We have to be careful, if called from sys_setpriority(), | |
4041 | * the task might be in the middle of scheduling on another CPU. | |
4042 | */ | |
4043 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4044 | update_rq_clock(rq); |
1da177e4 LT |
4045 | /* |
4046 | * The RT priorities are set via sched_setscheduler(), but we still | |
4047 | * allow the 'normal' nice value to be set - but as expected | |
4048 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4049 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4050 | */ |
e05606d3 | 4051 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4052 | p->static_prio = NICE_TO_PRIO(nice); |
4053 | goto out_unlock; | |
4054 | } | |
dd41f596 IM |
4055 | on_rq = p->se.on_rq; |
4056 | if (on_rq) { | |
69be72c1 | 4057 | dequeue_task(rq, p, 0); |
79b5dddf | 4058 | dec_load(rq, p); |
2dd73a4f | 4059 | } |
1da177e4 | 4060 | |
1da177e4 | 4061 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4062 | set_load_weight(p); |
b29739f9 IM |
4063 | old_prio = p->prio; |
4064 | p->prio = effective_prio(p); | |
4065 | delta = p->prio - old_prio; | |
1da177e4 | 4066 | |
dd41f596 | 4067 | if (on_rq) { |
8159f87e | 4068 | enqueue_task(rq, p, 0); |
29b4b623 | 4069 | inc_load(rq, p); |
1da177e4 | 4070 | /* |
d5f9f942 AM |
4071 | * If the task increased its priority or is running and |
4072 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4073 | */ |
d5f9f942 | 4074 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4075 | resched_task(rq->curr); |
4076 | } | |
4077 | out_unlock: | |
4078 | task_rq_unlock(rq, &flags); | |
4079 | } | |
1da177e4 LT |
4080 | EXPORT_SYMBOL(set_user_nice); |
4081 | ||
e43379f1 MM |
4082 | /* |
4083 | * can_nice - check if a task can reduce its nice value | |
4084 | * @p: task | |
4085 | * @nice: nice value | |
4086 | */ | |
36c8b586 | 4087 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4088 | { |
024f4747 MM |
4089 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4090 | int nice_rlim = 20 - nice; | |
48f24c4d | 4091 | |
e43379f1 MM |
4092 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4093 | capable(CAP_SYS_NICE)); | |
4094 | } | |
4095 | ||
1da177e4 LT |
4096 | #ifdef __ARCH_WANT_SYS_NICE |
4097 | ||
4098 | /* | |
4099 | * sys_nice - change the priority of the current process. | |
4100 | * @increment: priority increment | |
4101 | * | |
4102 | * sys_setpriority is a more generic, but much slower function that | |
4103 | * does similar things. | |
4104 | */ | |
4105 | asmlinkage long sys_nice(int increment) | |
4106 | { | |
48f24c4d | 4107 | long nice, retval; |
1da177e4 LT |
4108 | |
4109 | /* | |
4110 | * Setpriority might change our priority at the same moment. | |
4111 | * We don't have to worry. Conceptually one call occurs first | |
4112 | * and we have a single winner. | |
4113 | */ | |
e43379f1 MM |
4114 | if (increment < -40) |
4115 | increment = -40; | |
1da177e4 LT |
4116 | if (increment > 40) |
4117 | increment = 40; | |
4118 | ||
4119 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4120 | if (nice < -20) | |
4121 | nice = -20; | |
4122 | if (nice > 19) | |
4123 | nice = 19; | |
4124 | ||
e43379f1 MM |
4125 | if (increment < 0 && !can_nice(current, nice)) |
4126 | return -EPERM; | |
4127 | ||
1da177e4 LT |
4128 | retval = security_task_setnice(current, nice); |
4129 | if (retval) | |
4130 | return retval; | |
4131 | ||
4132 | set_user_nice(current, nice); | |
4133 | return 0; | |
4134 | } | |
4135 | ||
4136 | #endif | |
4137 | ||
4138 | /** | |
4139 | * task_prio - return the priority value of a given task. | |
4140 | * @p: the task in question. | |
4141 | * | |
4142 | * This is the priority value as seen by users in /proc. | |
4143 | * RT tasks are offset by -200. Normal tasks are centered | |
4144 | * around 0, value goes from -16 to +15. | |
4145 | */ | |
36c8b586 | 4146 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4147 | { |
4148 | return p->prio - MAX_RT_PRIO; | |
4149 | } | |
4150 | ||
4151 | /** | |
4152 | * task_nice - return the nice value of a given task. | |
4153 | * @p: the task in question. | |
4154 | */ | |
36c8b586 | 4155 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4156 | { |
4157 | return TASK_NICE(p); | |
4158 | } | |
1da177e4 | 4159 | EXPORT_SYMBOL_GPL(task_nice); |
1da177e4 LT |
4160 | |
4161 | /** | |
4162 | * idle_cpu - is a given cpu idle currently? | |
4163 | * @cpu: the processor in question. | |
4164 | */ | |
4165 | int idle_cpu(int cpu) | |
4166 | { | |
4167 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4168 | } | |
4169 | ||
1da177e4 LT |
4170 | /** |
4171 | * idle_task - return the idle task for a given cpu. | |
4172 | * @cpu: the processor in question. | |
4173 | */ | |
36c8b586 | 4174 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4175 | { |
4176 | return cpu_rq(cpu)->idle; | |
4177 | } | |
4178 | ||
4179 | /** | |
4180 | * find_process_by_pid - find a process with a matching PID value. | |
4181 | * @pid: the pid in question. | |
4182 | */ | |
a9957449 | 4183 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4184 | { |
228ebcbe | 4185 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4186 | } |
4187 | ||
4188 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4189 | static void |
4190 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4191 | { |
dd41f596 | 4192 | BUG_ON(p->se.on_rq); |
48f24c4d | 4193 | |
1da177e4 | 4194 | p->policy = policy; |
dd41f596 IM |
4195 | switch (p->policy) { |
4196 | case SCHED_NORMAL: | |
4197 | case SCHED_BATCH: | |
4198 | case SCHED_IDLE: | |
4199 | p->sched_class = &fair_sched_class; | |
4200 | break; | |
4201 | case SCHED_FIFO: | |
4202 | case SCHED_RR: | |
4203 | p->sched_class = &rt_sched_class; | |
4204 | break; | |
4205 | } | |
4206 | ||
1da177e4 | 4207 | p->rt_priority = prio; |
b29739f9 IM |
4208 | p->normal_prio = normal_prio(p); |
4209 | /* we are holding p->pi_lock already */ | |
4210 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4211 | set_load_weight(p); |
1da177e4 LT |
4212 | } |
4213 | ||
4214 | /** | |
72fd4a35 | 4215 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4216 | * @p: the task in question. |
4217 | * @policy: new policy. | |
4218 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4219 | * |
72fd4a35 | 4220 | * NOTE that the task may be already dead. |
1da177e4 | 4221 | */ |
95cdf3b7 IM |
4222 | int sched_setscheduler(struct task_struct *p, int policy, |
4223 | struct sched_param *param) | |
1da177e4 | 4224 | { |
83b699ed | 4225 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4226 | unsigned long flags; |
70b97a7f | 4227 | struct rq *rq; |
1da177e4 | 4228 | |
66e5393a SR |
4229 | /* may grab non-irq protected spin_locks */ |
4230 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4231 | recheck: |
4232 | /* double check policy once rq lock held */ | |
4233 | if (policy < 0) | |
4234 | policy = oldpolicy = p->policy; | |
4235 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4236 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4237 | policy != SCHED_IDLE) | |
b0a9499c | 4238 | return -EINVAL; |
1da177e4 LT |
4239 | /* |
4240 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4241 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4242 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4243 | */ |
4244 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4245 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4246 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4247 | return -EINVAL; |
e05606d3 | 4248 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4249 | return -EINVAL; |
4250 | ||
37e4ab3f OC |
4251 | /* |
4252 | * Allow unprivileged RT tasks to decrease priority: | |
4253 | */ | |
4254 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4255 | if (rt_policy(policy)) { |
8dc3e909 | 4256 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4257 | |
4258 | if (!lock_task_sighand(p, &flags)) | |
4259 | return -ESRCH; | |
4260 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4261 | unlock_task_sighand(p, &flags); | |
4262 | ||
4263 | /* can't set/change the rt policy */ | |
4264 | if (policy != p->policy && !rlim_rtprio) | |
4265 | return -EPERM; | |
4266 | ||
4267 | /* can't increase priority */ | |
4268 | if (param->sched_priority > p->rt_priority && | |
4269 | param->sched_priority > rlim_rtprio) | |
4270 | return -EPERM; | |
4271 | } | |
dd41f596 IM |
4272 | /* |
4273 | * Like positive nice levels, dont allow tasks to | |
4274 | * move out of SCHED_IDLE either: | |
4275 | */ | |
4276 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4277 | return -EPERM; | |
5fe1d75f | 4278 | |
37e4ab3f OC |
4279 | /* can't change other user's priorities */ |
4280 | if ((current->euid != p->euid) && | |
4281 | (current->euid != p->uid)) | |
4282 | return -EPERM; | |
4283 | } | |
1da177e4 LT |
4284 | |
4285 | retval = security_task_setscheduler(p, policy, param); | |
4286 | if (retval) | |
4287 | return retval; | |
b29739f9 IM |
4288 | /* |
4289 | * make sure no PI-waiters arrive (or leave) while we are | |
4290 | * changing the priority of the task: | |
4291 | */ | |
4292 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4293 | /* |
4294 | * To be able to change p->policy safely, the apropriate | |
4295 | * runqueue lock must be held. | |
4296 | */ | |
b29739f9 | 4297 | rq = __task_rq_lock(p); |
1da177e4 LT |
4298 | /* recheck policy now with rq lock held */ |
4299 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4300 | policy = oldpolicy = -1; | |
b29739f9 IM |
4301 | __task_rq_unlock(rq); |
4302 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4303 | goto recheck; |
4304 | } | |
2daa3577 | 4305 | update_rq_clock(rq); |
dd41f596 | 4306 | on_rq = p->se.on_rq; |
83b699ed SV |
4307 | running = task_running(rq, p); |
4308 | if (on_rq) { | |
2e1cb74a | 4309 | deactivate_task(rq, p, 0); |
83b699ed SV |
4310 | if (running) |
4311 | p->sched_class->put_prev_task(rq, p); | |
4312 | } | |
f6b53205 | 4313 | |
1da177e4 | 4314 | oldprio = p->prio; |
dd41f596 | 4315 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4316 | |
dd41f596 | 4317 | if (on_rq) { |
83b699ed SV |
4318 | if (running) |
4319 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4320 | activate_task(rq, p, 0); |
1da177e4 LT |
4321 | /* |
4322 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4323 | * our priority decreased, or if we are not currently running on |
4324 | * this runqueue and our priority is higher than the current's | |
1da177e4 | 4325 | */ |
83b699ed | 4326 | if (running) { |
d5f9f942 AM |
4327 | if (p->prio > oldprio) |
4328 | resched_task(rq->curr); | |
dd41f596 IM |
4329 | } else { |
4330 | check_preempt_curr(rq, p); | |
4331 | } | |
1da177e4 | 4332 | } |
b29739f9 IM |
4333 | __task_rq_unlock(rq); |
4334 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4335 | ||
95e02ca9 TG |
4336 | rt_mutex_adjust_pi(p); |
4337 | ||
1da177e4 LT |
4338 | return 0; |
4339 | } | |
4340 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4341 | ||
95cdf3b7 IM |
4342 | static int |
4343 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4344 | { |
1da177e4 LT |
4345 | struct sched_param lparam; |
4346 | struct task_struct *p; | |
36c8b586 | 4347 | int retval; |
1da177e4 LT |
4348 | |
4349 | if (!param || pid < 0) | |
4350 | return -EINVAL; | |
4351 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4352 | return -EFAULT; | |
5fe1d75f ON |
4353 | |
4354 | rcu_read_lock(); | |
4355 | retval = -ESRCH; | |
1da177e4 | 4356 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4357 | if (p != NULL) |
4358 | retval = sched_setscheduler(p, policy, &lparam); | |
4359 | rcu_read_unlock(); | |
36c8b586 | 4360 | |
1da177e4 LT |
4361 | return retval; |
4362 | } | |
4363 | ||
4364 | /** | |
4365 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4366 | * @pid: the pid in question. | |
4367 | * @policy: new policy. | |
4368 | * @param: structure containing the new RT priority. | |
4369 | */ | |
4370 | asmlinkage long sys_sched_setscheduler(pid_t pid, int policy, | |
4371 | struct sched_param __user *param) | |
4372 | { | |
c21761f1 JB |
4373 | /* negative values for policy are not valid */ |
4374 | if (policy < 0) | |
4375 | return -EINVAL; | |
4376 | ||
1da177e4 LT |
4377 | return do_sched_setscheduler(pid, policy, param); |
4378 | } | |
4379 | ||
4380 | /** | |
4381 | * sys_sched_setparam - set/change the RT priority of a thread | |
4382 | * @pid: the pid in question. | |
4383 | * @param: structure containing the new RT priority. | |
4384 | */ | |
4385 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4386 | { | |
4387 | return do_sched_setscheduler(pid, -1, param); | |
4388 | } | |
4389 | ||
4390 | /** | |
4391 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4392 | * @pid: the pid in question. | |
4393 | */ | |
4394 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4395 | { | |
36c8b586 | 4396 | struct task_struct *p; |
3a5c359a | 4397 | int retval; |
1da177e4 LT |
4398 | |
4399 | if (pid < 0) | |
3a5c359a | 4400 | return -EINVAL; |
1da177e4 LT |
4401 | |
4402 | retval = -ESRCH; | |
4403 | read_lock(&tasklist_lock); | |
4404 | p = find_process_by_pid(pid); | |
4405 | if (p) { | |
4406 | retval = security_task_getscheduler(p); | |
4407 | if (!retval) | |
4408 | retval = p->policy; | |
4409 | } | |
4410 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
4411 | return retval; |
4412 | } | |
4413 | ||
4414 | /** | |
4415 | * sys_sched_getscheduler - get the RT priority of a thread | |
4416 | * @pid: the pid in question. | |
4417 | * @param: structure containing the RT priority. | |
4418 | */ | |
4419 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
4420 | { | |
4421 | struct sched_param lp; | |
36c8b586 | 4422 | struct task_struct *p; |
3a5c359a | 4423 | int retval; |
1da177e4 LT |
4424 | |
4425 | if (!param || pid < 0) | |
3a5c359a | 4426 | return -EINVAL; |
1da177e4 LT |
4427 | |
4428 | read_lock(&tasklist_lock); | |
4429 | p = find_process_by_pid(pid); | |
4430 | retval = -ESRCH; | |
4431 | if (!p) | |
4432 | goto out_unlock; | |
4433 | ||
4434 | retval = security_task_getscheduler(p); | |
4435 | if (retval) | |
4436 | goto out_unlock; | |
4437 | ||
4438 | lp.sched_priority = p->rt_priority; | |
4439 | read_unlock(&tasklist_lock); | |
4440 | ||
4441 | /* | |
4442 | * This one might sleep, we cannot do it with a spinlock held ... | |
4443 | */ | |
4444 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4445 | ||
1da177e4 LT |
4446 | return retval; |
4447 | ||
4448 | out_unlock: | |
4449 | read_unlock(&tasklist_lock); | |
4450 | return retval; | |
4451 | } | |
4452 | ||
4453 | long sched_setaffinity(pid_t pid, cpumask_t new_mask) | |
4454 | { | |
1da177e4 | 4455 | cpumask_t cpus_allowed; |
36c8b586 IM |
4456 | struct task_struct *p; |
4457 | int retval; | |
1da177e4 | 4458 | |
5be9361c | 4459 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4460 | read_lock(&tasklist_lock); |
4461 | ||
4462 | p = find_process_by_pid(pid); | |
4463 | if (!p) { | |
4464 | read_unlock(&tasklist_lock); | |
5be9361c | 4465 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4466 | return -ESRCH; |
4467 | } | |
4468 | ||
4469 | /* | |
4470 | * It is not safe to call set_cpus_allowed with the | |
4471 | * tasklist_lock held. We will bump the task_struct's | |
4472 | * usage count and then drop tasklist_lock. | |
4473 | */ | |
4474 | get_task_struct(p); | |
4475 | read_unlock(&tasklist_lock); | |
4476 | ||
4477 | retval = -EPERM; | |
4478 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
4479 | !capable(CAP_SYS_NICE)) | |
4480 | goto out_unlock; | |
4481 | ||
e7834f8f DQ |
4482 | retval = security_task_setscheduler(p, 0, NULL); |
4483 | if (retval) | |
4484 | goto out_unlock; | |
4485 | ||
1da177e4 LT |
4486 | cpus_allowed = cpuset_cpus_allowed(p); |
4487 | cpus_and(new_mask, new_mask, cpus_allowed); | |
8707d8b8 | 4488 | again: |
1da177e4 LT |
4489 | retval = set_cpus_allowed(p, new_mask); |
4490 | ||
8707d8b8 PM |
4491 | if (!retval) { |
4492 | cpus_allowed = cpuset_cpus_allowed(p); | |
4493 | if (!cpus_subset(new_mask, cpus_allowed)) { | |
4494 | /* | |
4495 | * We must have raced with a concurrent cpuset | |
4496 | * update. Just reset the cpus_allowed to the | |
4497 | * cpuset's cpus_allowed | |
4498 | */ | |
4499 | new_mask = cpus_allowed; | |
4500 | goto again; | |
4501 | } | |
4502 | } | |
1da177e4 LT |
4503 | out_unlock: |
4504 | put_task_struct(p); | |
5be9361c | 4505 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4506 | return retval; |
4507 | } | |
4508 | ||
4509 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
4510 | cpumask_t *new_mask) | |
4511 | { | |
4512 | if (len < sizeof(cpumask_t)) { | |
4513 | memset(new_mask, 0, sizeof(cpumask_t)); | |
4514 | } else if (len > sizeof(cpumask_t)) { | |
4515 | len = sizeof(cpumask_t); | |
4516 | } | |
4517 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
4518 | } | |
4519 | ||
4520 | /** | |
4521 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4522 | * @pid: pid of the process | |
4523 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4524 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4525 | */ | |
4526 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
4527 | unsigned long __user *user_mask_ptr) | |
4528 | { | |
4529 | cpumask_t new_mask; | |
4530 | int retval; | |
4531 | ||
4532 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
4533 | if (retval) | |
4534 | return retval; | |
4535 | ||
4536 | return sched_setaffinity(pid, new_mask); | |
4537 | } | |
4538 | ||
4539 | /* | |
4540 | * Represents all cpu's present in the system | |
4541 | * In systems capable of hotplug, this map could dynamically grow | |
4542 | * as new cpu's are detected in the system via any platform specific | |
4543 | * method, such as ACPI for e.g. | |
4544 | */ | |
4545 | ||
4cef0c61 | 4546 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
4547 | EXPORT_SYMBOL(cpu_present_map); |
4548 | ||
4549 | #ifndef CONFIG_SMP | |
4cef0c61 | 4550 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
4551 | EXPORT_SYMBOL(cpu_online_map); |
4552 | ||
4cef0c61 | 4553 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 4554 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
4555 | #endif |
4556 | ||
4557 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
4558 | { | |
36c8b586 | 4559 | struct task_struct *p; |
1da177e4 | 4560 | int retval; |
1da177e4 | 4561 | |
5be9361c | 4562 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4563 | read_lock(&tasklist_lock); |
4564 | ||
4565 | retval = -ESRCH; | |
4566 | p = find_process_by_pid(pid); | |
4567 | if (!p) | |
4568 | goto out_unlock; | |
4569 | ||
e7834f8f DQ |
4570 | retval = security_task_getscheduler(p); |
4571 | if (retval) | |
4572 | goto out_unlock; | |
4573 | ||
2f7016d9 | 4574 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
4575 | |
4576 | out_unlock: | |
4577 | read_unlock(&tasklist_lock); | |
5be9361c | 4578 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 | 4579 | |
9531b62f | 4580 | return retval; |
1da177e4 LT |
4581 | } |
4582 | ||
4583 | /** | |
4584 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4585 | * @pid: pid of the process | |
4586 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4587 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4588 | */ | |
4589 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
4590 | unsigned long __user *user_mask_ptr) | |
4591 | { | |
4592 | int ret; | |
4593 | cpumask_t mask; | |
4594 | ||
4595 | if (len < sizeof(cpumask_t)) | |
4596 | return -EINVAL; | |
4597 | ||
4598 | ret = sched_getaffinity(pid, &mask); | |
4599 | if (ret < 0) | |
4600 | return ret; | |
4601 | ||
4602 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
4603 | return -EFAULT; | |
4604 | ||
4605 | return sizeof(cpumask_t); | |
4606 | } | |
4607 | ||
4608 | /** | |
4609 | * sys_sched_yield - yield the current processor to other threads. | |
4610 | * | |
dd41f596 IM |
4611 | * This function yields the current CPU to other tasks. If there are no |
4612 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
4613 | */ |
4614 | asmlinkage long sys_sched_yield(void) | |
4615 | { | |
70b97a7f | 4616 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4617 | |
2d72376b | 4618 | schedstat_inc(rq, yld_count); |
4530d7ab | 4619 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4620 | |
4621 | /* | |
4622 | * Since we are going to call schedule() anyway, there's | |
4623 | * no need to preempt or enable interrupts: | |
4624 | */ | |
4625 | __release(rq->lock); | |
8a25d5de | 4626 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4627 | _raw_spin_unlock(&rq->lock); |
4628 | preempt_enable_no_resched(); | |
4629 | ||
4630 | schedule(); | |
4631 | ||
4632 | return 0; | |
4633 | } | |
4634 | ||
e7b38404 | 4635 | static void __cond_resched(void) |
1da177e4 | 4636 | { |
8e0a43d8 IM |
4637 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
4638 | __might_sleep(__FILE__, __LINE__); | |
4639 | #endif | |
5bbcfd90 IM |
4640 | /* |
4641 | * The BKS might be reacquired before we have dropped | |
4642 | * PREEMPT_ACTIVE, which could trigger a second | |
4643 | * cond_resched() call. | |
4644 | */ | |
1da177e4 LT |
4645 | do { |
4646 | add_preempt_count(PREEMPT_ACTIVE); | |
4647 | schedule(); | |
4648 | sub_preempt_count(PREEMPT_ACTIVE); | |
4649 | } while (need_resched()); | |
4650 | } | |
4651 | ||
4652 | int __sched cond_resched(void) | |
4653 | { | |
9414232f IM |
4654 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
4655 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
4656 | __cond_resched(); |
4657 | return 1; | |
4658 | } | |
4659 | return 0; | |
4660 | } | |
1da177e4 LT |
4661 | EXPORT_SYMBOL(cond_resched); |
4662 | ||
4663 | /* | |
4664 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
4665 | * call schedule, and on return reacquire the lock. | |
4666 | * | |
4667 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level | |
4668 | * operations here to prevent schedule() from being called twice (once via | |
4669 | * spin_unlock(), once by hand). | |
4670 | */ | |
95cdf3b7 | 4671 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4672 | { |
6df3cecb JK |
4673 | int ret = 0; |
4674 | ||
1da177e4 LT |
4675 | if (need_lockbreak(lock)) { |
4676 | spin_unlock(lock); | |
4677 | cpu_relax(); | |
6df3cecb | 4678 | ret = 1; |
1da177e4 LT |
4679 | spin_lock(lock); |
4680 | } | |
9414232f | 4681 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
8a25d5de | 4682 | spin_release(&lock->dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4683 | _raw_spin_unlock(lock); |
4684 | preempt_enable_no_resched(); | |
4685 | __cond_resched(); | |
6df3cecb | 4686 | ret = 1; |
1da177e4 | 4687 | spin_lock(lock); |
1da177e4 | 4688 | } |
6df3cecb | 4689 | return ret; |
1da177e4 | 4690 | } |
1da177e4 LT |
4691 | EXPORT_SYMBOL(cond_resched_lock); |
4692 | ||
4693 | int __sched cond_resched_softirq(void) | |
4694 | { | |
4695 | BUG_ON(!in_softirq()); | |
4696 | ||
9414232f | 4697 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 4698 | local_bh_enable(); |
1da177e4 LT |
4699 | __cond_resched(); |
4700 | local_bh_disable(); | |
4701 | return 1; | |
4702 | } | |
4703 | return 0; | |
4704 | } | |
1da177e4 LT |
4705 | EXPORT_SYMBOL(cond_resched_softirq); |
4706 | ||
1da177e4 LT |
4707 | /** |
4708 | * yield - yield the current processor to other threads. | |
4709 | * | |
72fd4a35 | 4710 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4711 | * thread runnable and calls sys_sched_yield(). |
4712 | */ | |
4713 | void __sched yield(void) | |
4714 | { | |
4715 | set_current_state(TASK_RUNNING); | |
4716 | sys_sched_yield(); | |
4717 | } | |
1da177e4 LT |
4718 | EXPORT_SYMBOL(yield); |
4719 | ||
4720 | /* | |
4721 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so | |
4722 | * that process accounting knows that this is a task in IO wait state. | |
4723 | * | |
4724 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
4725 | * has set its backing_dev_info: the queue against which it should throttle) | |
4726 | */ | |
4727 | void __sched io_schedule(void) | |
4728 | { | |
70b97a7f | 4729 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 4730 | |
0ff92245 | 4731 | delayacct_blkio_start(); |
1da177e4 LT |
4732 | atomic_inc(&rq->nr_iowait); |
4733 | schedule(); | |
4734 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4735 | delayacct_blkio_end(); |
1da177e4 | 4736 | } |
1da177e4 LT |
4737 | EXPORT_SYMBOL(io_schedule); |
4738 | ||
4739 | long __sched io_schedule_timeout(long timeout) | |
4740 | { | |
70b97a7f | 4741 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
4742 | long ret; |
4743 | ||
0ff92245 | 4744 | delayacct_blkio_start(); |
1da177e4 LT |
4745 | atomic_inc(&rq->nr_iowait); |
4746 | ret = schedule_timeout(timeout); | |
4747 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4748 | delayacct_blkio_end(); |
1da177e4 LT |
4749 | return ret; |
4750 | } | |
4751 | ||
4752 | /** | |
4753 | * sys_sched_get_priority_max - return maximum RT priority. | |
4754 | * @policy: scheduling class. | |
4755 | * | |
4756 | * this syscall returns the maximum rt_priority that can be used | |
4757 | * by a given scheduling class. | |
4758 | */ | |
4759 | asmlinkage long sys_sched_get_priority_max(int policy) | |
4760 | { | |
4761 | int ret = -EINVAL; | |
4762 | ||
4763 | switch (policy) { | |
4764 | case SCHED_FIFO: | |
4765 | case SCHED_RR: | |
4766 | ret = MAX_USER_RT_PRIO-1; | |
4767 | break; | |
4768 | case SCHED_NORMAL: | |
b0a9499c | 4769 | case SCHED_BATCH: |
dd41f596 | 4770 | case SCHED_IDLE: |
1da177e4 LT |
4771 | ret = 0; |
4772 | break; | |
4773 | } | |
4774 | return ret; | |
4775 | } | |
4776 | ||
4777 | /** | |
4778 | * sys_sched_get_priority_min - return minimum RT priority. | |
4779 | * @policy: scheduling class. | |
4780 | * | |
4781 | * this syscall returns the minimum rt_priority that can be used | |
4782 | * by a given scheduling class. | |
4783 | */ | |
4784 | asmlinkage long sys_sched_get_priority_min(int policy) | |
4785 | { | |
4786 | int ret = -EINVAL; | |
4787 | ||
4788 | switch (policy) { | |
4789 | case SCHED_FIFO: | |
4790 | case SCHED_RR: | |
4791 | ret = 1; | |
4792 | break; | |
4793 | case SCHED_NORMAL: | |
b0a9499c | 4794 | case SCHED_BATCH: |
dd41f596 | 4795 | case SCHED_IDLE: |
1da177e4 LT |
4796 | ret = 0; |
4797 | } | |
4798 | return ret; | |
4799 | } | |
4800 | ||
4801 | /** | |
4802 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4803 | * @pid: pid of the process. | |
4804 | * @interval: userspace pointer to the timeslice value. | |
4805 | * | |
4806 | * this syscall writes the default timeslice value of a given process | |
4807 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4808 | */ | |
4809 | asmlinkage | |
4810 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
4811 | { | |
36c8b586 | 4812 | struct task_struct *p; |
a4ec24b4 | 4813 | unsigned int time_slice; |
3a5c359a | 4814 | int retval; |
1da177e4 | 4815 | struct timespec t; |
1da177e4 LT |
4816 | |
4817 | if (pid < 0) | |
3a5c359a | 4818 | return -EINVAL; |
1da177e4 LT |
4819 | |
4820 | retval = -ESRCH; | |
4821 | read_lock(&tasklist_lock); | |
4822 | p = find_process_by_pid(pid); | |
4823 | if (!p) | |
4824 | goto out_unlock; | |
4825 | ||
4826 | retval = security_task_getscheduler(p); | |
4827 | if (retval) | |
4828 | goto out_unlock; | |
4829 | ||
a4ec24b4 DA |
4830 | if (p->policy == SCHED_FIFO) |
4831 | time_slice = 0; | |
4832 | else if (p->policy == SCHED_RR) | |
4833 | time_slice = DEF_TIMESLICE; | |
4834 | else { | |
4835 | struct sched_entity *se = &p->se; | |
4836 | unsigned long flags; | |
4837 | struct rq *rq; | |
4838 | ||
4839 | rq = task_rq_lock(p, &flags); | |
4840 | time_slice = NS_TO_JIFFIES(sched_slice(cfs_rq_of(se), se)); | |
4841 | task_rq_unlock(rq, &flags); | |
4842 | } | |
1da177e4 | 4843 | read_unlock(&tasklist_lock); |
a4ec24b4 | 4844 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4845 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4846 | return retval; |
3a5c359a | 4847 | |
1da177e4 LT |
4848 | out_unlock: |
4849 | read_unlock(&tasklist_lock); | |
4850 | return retval; | |
4851 | } | |
4852 | ||
2ed6e34f | 4853 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 IM |
4854 | |
4855 | static void show_task(struct task_struct *p) | |
1da177e4 | 4856 | { |
1da177e4 | 4857 | unsigned long free = 0; |
36c8b586 | 4858 | unsigned state; |
1da177e4 | 4859 | |
1da177e4 | 4860 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 4861 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 4862 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4863 | #if BITS_PER_LONG == 32 |
1da177e4 | 4864 | if (state == TASK_RUNNING) |
cc4ea795 | 4865 | printk(KERN_CONT " running "); |
1da177e4 | 4866 | else |
cc4ea795 | 4867 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4868 | #else |
4869 | if (state == TASK_RUNNING) | |
cc4ea795 | 4870 | printk(KERN_CONT " running task "); |
1da177e4 | 4871 | else |
cc4ea795 | 4872 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4873 | #endif |
4874 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
4875 | { | |
10ebffde | 4876 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
4877 | while (!*n) |
4878 | n++; | |
10ebffde | 4879 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
4880 | } |
4881 | #endif | |
ba25f9dc PE |
4882 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
4883 | task_pid_nr(p), task_pid_nr(p->parent)); | |
1da177e4 LT |
4884 | |
4885 | if (state != TASK_RUNNING) | |
4886 | show_stack(p, NULL); | |
4887 | } | |
4888 | ||
e59e2ae2 | 4889 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4890 | { |
36c8b586 | 4891 | struct task_struct *g, *p; |
1da177e4 | 4892 | |
4bd77321 IM |
4893 | #if BITS_PER_LONG == 32 |
4894 | printk(KERN_INFO | |
4895 | " task PC stack pid father\n"); | |
1da177e4 | 4896 | #else |
4bd77321 IM |
4897 | printk(KERN_INFO |
4898 | " task PC stack pid father\n"); | |
1da177e4 LT |
4899 | #endif |
4900 | read_lock(&tasklist_lock); | |
4901 | do_each_thread(g, p) { | |
4902 | /* | |
4903 | * reset the NMI-timeout, listing all files on a slow | |
4904 | * console might take alot of time: | |
4905 | */ | |
4906 | touch_nmi_watchdog(); | |
39bc89fd | 4907 | if (!state_filter || (p->state & state_filter)) |
e59e2ae2 | 4908 | show_task(p); |
1da177e4 LT |
4909 | } while_each_thread(g, p); |
4910 | ||
04c9167f JF |
4911 | touch_all_softlockup_watchdogs(); |
4912 | ||
dd41f596 IM |
4913 | #ifdef CONFIG_SCHED_DEBUG |
4914 | sysrq_sched_debug_show(); | |
4915 | #endif | |
1da177e4 | 4916 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
4917 | /* |
4918 | * Only show locks if all tasks are dumped: | |
4919 | */ | |
4920 | if (state_filter == -1) | |
4921 | debug_show_all_locks(); | |
1da177e4 LT |
4922 | } |
4923 | ||
1df21055 IM |
4924 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4925 | { | |
dd41f596 | 4926 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4927 | } |
4928 | ||
f340c0d1 IM |
4929 | /** |
4930 | * init_idle - set up an idle thread for a given CPU | |
4931 | * @idle: task in question | |
4932 | * @cpu: cpu the idle task belongs to | |
4933 | * | |
4934 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4935 | * flag, to make booting more robust. | |
4936 | */ | |
5c1e1767 | 4937 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4938 | { |
70b97a7f | 4939 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4940 | unsigned long flags; |
4941 | ||
dd41f596 IM |
4942 | __sched_fork(idle); |
4943 | idle->se.exec_start = sched_clock(); | |
4944 | ||
b29739f9 | 4945 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 4946 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 4947 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
4948 | |
4949 | spin_lock_irqsave(&rq->lock, flags); | |
4950 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
4951 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4952 | idle->oncpu = 1; | |
4953 | #endif | |
1da177e4 LT |
4954 | spin_unlock_irqrestore(&rq->lock, flags); |
4955 | ||
4956 | /* Set the preempt count _outside_ the spinlocks! */ | |
4957 | #if defined(CONFIG_PREEMPT) && !defined(CONFIG_PREEMPT_BKL) | |
a1261f54 | 4958 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); |
1da177e4 | 4959 | #else |
a1261f54 | 4960 | task_thread_info(idle)->preempt_count = 0; |
1da177e4 | 4961 | #endif |
dd41f596 IM |
4962 | /* |
4963 | * The idle tasks have their own, simple scheduling class: | |
4964 | */ | |
4965 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
4966 | } |
4967 | ||
4968 | /* | |
4969 | * In a system that switches off the HZ timer nohz_cpu_mask | |
4970 | * indicates which cpus entered this state. This is used | |
4971 | * in the rcu update to wait only for active cpus. For system | |
4972 | * which do not switch off the HZ timer nohz_cpu_mask should | |
4973 | * always be CPU_MASK_NONE. | |
4974 | */ | |
4975 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
4976 | ||
4977 | #ifdef CONFIG_SMP | |
4978 | /* | |
4979 | * This is how migration works: | |
4980 | * | |
70b97a7f | 4981 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
4982 | * runqueue and wake up that CPU's migration thread. |
4983 | * 2) we down() the locked semaphore => thread blocks. | |
4984 | * 3) migration thread wakes up (implicitly it forces the migrated | |
4985 | * thread off the CPU) | |
4986 | * 4) it gets the migration request and checks whether the migrated | |
4987 | * task is still in the wrong runqueue. | |
4988 | * 5) if it's in the wrong runqueue then the migration thread removes | |
4989 | * it and puts it into the right queue. | |
4990 | * 6) migration thread up()s the semaphore. | |
4991 | * 7) we wake up and the migration is done. | |
4992 | */ | |
4993 | ||
4994 | /* | |
4995 | * Change a given task's CPU affinity. Migrate the thread to a | |
4996 | * proper CPU and schedule it away if the CPU it's executing on | |
4997 | * is removed from the allowed bitmask. | |
4998 | * | |
4999 | * NOTE: the caller must have a valid reference to the task, the | |
5000 | * task must not exit() & deallocate itself prematurely. The | |
5001 | * call is not atomic; no spinlocks may be held. | |
5002 | */ | |
36c8b586 | 5003 | int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask) |
1da177e4 | 5004 | { |
70b97a7f | 5005 | struct migration_req req; |
1da177e4 | 5006 | unsigned long flags; |
70b97a7f | 5007 | struct rq *rq; |
48f24c4d | 5008 | int ret = 0; |
1da177e4 LT |
5009 | |
5010 | rq = task_rq_lock(p, &flags); | |
5011 | if (!cpus_intersects(new_mask, cpu_online_map)) { | |
5012 | ret = -EINVAL; | |
5013 | goto out; | |
5014 | } | |
5015 | ||
5016 | p->cpus_allowed = new_mask; | |
5017 | /* Can the task run on the task's current CPU? If so, we're done */ | |
5018 | if (cpu_isset(task_cpu(p), new_mask)) | |
5019 | goto out; | |
5020 | ||
5021 | if (migrate_task(p, any_online_cpu(new_mask), &req)) { | |
5022 | /* Need help from migration thread: drop lock and wait. */ | |
5023 | task_rq_unlock(rq, &flags); | |
5024 | wake_up_process(rq->migration_thread); | |
5025 | wait_for_completion(&req.done); | |
5026 | tlb_migrate_finish(p->mm); | |
5027 | return 0; | |
5028 | } | |
5029 | out: | |
5030 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5031 | |
1da177e4 LT |
5032 | return ret; |
5033 | } | |
1da177e4 LT |
5034 | EXPORT_SYMBOL_GPL(set_cpus_allowed); |
5035 | ||
5036 | /* | |
5037 | * Move (not current) task off this cpu, onto dest cpu. We're doing | |
5038 | * this because either it can't run here any more (set_cpus_allowed() | |
5039 | * away from this CPU, or CPU going down), or because we're | |
5040 | * attempting to rebalance this task on exec (sched_exec). | |
5041 | * | |
5042 | * So we race with normal scheduler movements, but that's OK, as long | |
5043 | * as the task is no longer on this CPU. | |
efc30814 KK |
5044 | * |
5045 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5046 | */ |
efc30814 | 5047 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5048 | { |
70b97a7f | 5049 | struct rq *rq_dest, *rq_src; |
dd41f596 | 5050 | int ret = 0, on_rq; |
1da177e4 LT |
5051 | |
5052 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 5053 | return ret; |
1da177e4 LT |
5054 | |
5055 | rq_src = cpu_rq(src_cpu); | |
5056 | rq_dest = cpu_rq(dest_cpu); | |
5057 | ||
5058 | double_rq_lock(rq_src, rq_dest); | |
5059 | /* Already moved. */ | |
5060 | if (task_cpu(p) != src_cpu) | |
5061 | goto out; | |
5062 | /* Affinity changed (again). */ | |
5063 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
5064 | goto out; | |
5065 | ||
dd41f596 | 5066 | on_rq = p->se.on_rq; |
6e82a3be | 5067 | if (on_rq) |
2e1cb74a | 5068 | deactivate_task(rq_src, p, 0); |
6e82a3be | 5069 | |
1da177e4 | 5070 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
5071 | if (on_rq) { |
5072 | activate_task(rq_dest, p, 0); | |
5073 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 5074 | } |
efc30814 | 5075 | ret = 1; |
1da177e4 LT |
5076 | out: |
5077 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 5078 | return ret; |
1da177e4 LT |
5079 | } |
5080 | ||
5081 | /* | |
5082 | * migration_thread - this is a highprio system thread that performs | |
5083 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5084 | * another runqueue. | |
5085 | */ | |
95cdf3b7 | 5086 | static int migration_thread(void *data) |
1da177e4 | 5087 | { |
1da177e4 | 5088 | int cpu = (long)data; |
70b97a7f | 5089 | struct rq *rq; |
1da177e4 LT |
5090 | |
5091 | rq = cpu_rq(cpu); | |
5092 | BUG_ON(rq->migration_thread != current); | |
5093 | ||
5094 | set_current_state(TASK_INTERRUPTIBLE); | |
5095 | while (!kthread_should_stop()) { | |
70b97a7f | 5096 | struct migration_req *req; |
1da177e4 | 5097 | struct list_head *head; |
1da177e4 | 5098 | |
1da177e4 LT |
5099 | spin_lock_irq(&rq->lock); |
5100 | ||
5101 | if (cpu_is_offline(cpu)) { | |
5102 | spin_unlock_irq(&rq->lock); | |
5103 | goto wait_to_die; | |
5104 | } | |
5105 | ||
5106 | if (rq->active_balance) { | |
5107 | active_load_balance(rq, cpu); | |
5108 | rq->active_balance = 0; | |
5109 | } | |
5110 | ||
5111 | head = &rq->migration_queue; | |
5112 | ||
5113 | if (list_empty(head)) { | |
5114 | spin_unlock_irq(&rq->lock); | |
5115 | schedule(); | |
5116 | set_current_state(TASK_INTERRUPTIBLE); | |
5117 | continue; | |
5118 | } | |
70b97a7f | 5119 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5120 | list_del_init(head->next); |
5121 | ||
674311d5 NP |
5122 | spin_unlock(&rq->lock); |
5123 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5124 | local_irq_enable(); | |
1da177e4 LT |
5125 | |
5126 | complete(&req->done); | |
5127 | } | |
5128 | __set_current_state(TASK_RUNNING); | |
5129 | return 0; | |
5130 | ||
5131 | wait_to_die: | |
5132 | /* Wait for kthread_stop */ | |
5133 | set_current_state(TASK_INTERRUPTIBLE); | |
5134 | while (!kthread_should_stop()) { | |
5135 | schedule(); | |
5136 | set_current_state(TASK_INTERRUPTIBLE); | |
5137 | } | |
5138 | __set_current_state(TASK_RUNNING); | |
5139 | return 0; | |
5140 | } | |
5141 | ||
5142 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5143 | |
5144 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5145 | { | |
5146 | int ret; | |
5147 | ||
5148 | local_irq_disable(); | |
5149 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5150 | local_irq_enable(); | |
5151 | return ret; | |
5152 | } | |
5153 | ||
054b9108 | 5154 | /* |
3a4fa0a2 | 5155 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
5156 | * NOTE: interrupts should be disabled by the caller |
5157 | */ | |
48f24c4d | 5158 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5159 | { |
efc30814 | 5160 | unsigned long flags; |
1da177e4 | 5161 | cpumask_t mask; |
70b97a7f IM |
5162 | struct rq *rq; |
5163 | int dest_cpu; | |
1da177e4 | 5164 | |
3a5c359a AK |
5165 | do { |
5166 | /* On same node? */ | |
5167 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
5168 | cpus_and(mask, mask, p->cpus_allowed); | |
5169 | dest_cpu = any_online_cpu(mask); | |
5170 | ||
5171 | /* On any allowed CPU? */ | |
5172 | if (dest_cpu == NR_CPUS) | |
5173 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
5174 | ||
5175 | /* No more Mr. Nice Guy. */ | |
5176 | if (dest_cpu == NR_CPUS) { | |
470fd646 CW |
5177 | cpumask_t cpus_allowed = cpuset_cpus_allowed_locked(p); |
5178 | /* | |
5179 | * Try to stay on the same cpuset, where the | |
5180 | * current cpuset may be a subset of all cpus. | |
5181 | * The cpuset_cpus_allowed_locked() variant of | |
5182 | * cpuset_cpus_allowed() will not block. It must be | |
5183 | * called within calls to cpuset_lock/cpuset_unlock. | |
5184 | */ | |
3a5c359a | 5185 | rq = task_rq_lock(p, &flags); |
470fd646 | 5186 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
5187 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5188 | task_rq_unlock(rq, &flags); | |
1da177e4 | 5189 | |
3a5c359a AK |
5190 | /* |
5191 | * Don't tell them about moving exiting tasks or | |
5192 | * kernel threads (both mm NULL), since they never | |
5193 | * leave kernel. | |
5194 | */ | |
5195 | if (p->mm && printk_ratelimit()) | |
5196 | printk(KERN_INFO "process %d (%s) no " | |
5197 | "longer affine to cpu%d\n", | |
ba25f9dc | 5198 | task_pid_nr(p), p->comm, dead_cpu); |
3a5c359a | 5199 | } |
f7b4cddc | 5200 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
5201 | } |
5202 | ||
5203 | /* | |
5204 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5205 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5206 | * for performance reasons the counter is not stricly tracking tasks to | |
5207 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5208 | * to keep the global sum constant after CPU-down: | |
5209 | */ | |
70b97a7f | 5210 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5211 | { |
70b97a7f | 5212 | struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL)); |
1da177e4 LT |
5213 | unsigned long flags; |
5214 | ||
5215 | local_irq_save(flags); | |
5216 | double_rq_lock(rq_src, rq_dest); | |
5217 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5218 | rq_src->nr_uninterruptible = 0; | |
5219 | double_rq_unlock(rq_src, rq_dest); | |
5220 | local_irq_restore(flags); | |
5221 | } | |
5222 | ||
5223 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5224 | static void migrate_live_tasks(int src_cpu) | |
5225 | { | |
48f24c4d | 5226 | struct task_struct *p, *t; |
1da177e4 | 5227 | |
f7b4cddc | 5228 | read_lock(&tasklist_lock); |
1da177e4 | 5229 | |
48f24c4d IM |
5230 | do_each_thread(t, p) { |
5231 | if (p == current) | |
1da177e4 LT |
5232 | continue; |
5233 | ||
48f24c4d IM |
5234 | if (task_cpu(p) == src_cpu) |
5235 | move_task_off_dead_cpu(src_cpu, p); | |
5236 | } while_each_thread(t, p); | |
1da177e4 | 5237 | |
f7b4cddc | 5238 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5239 | } |
5240 | ||
a9957449 AD |
5241 | /* |
5242 | * activate_idle_task - move idle task to the _front_ of runqueue. | |
5243 | */ | |
5244 | static void activate_idle_task(struct task_struct *p, struct rq *rq) | |
5245 | { | |
5246 | update_rq_clock(rq); | |
5247 | ||
5248 | if (p->state == TASK_UNINTERRUPTIBLE) | |
5249 | rq->nr_uninterruptible--; | |
5250 | ||
5251 | enqueue_task(rq, p, 0); | |
5252 | inc_nr_running(p, rq); | |
5253 | } | |
5254 | ||
dd41f596 IM |
5255 | /* |
5256 | * Schedules idle task to be the next runnable task on current CPU. | |
1da177e4 | 5257 | * It does so by boosting its priority to highest possible and adding it to |
48f24c4d | 5258 | * the _front_ of the runqueue. Used by CPU offline code. |
1da177e4 LT |
5259 | */ |
5260 | void sched_idle_next(void) | |
5261 | { | |
48f24c4d | 5262 | int this_cpu = smp_processor_id(); |
70b97a7f | 5263 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5264 | struct task_struct *p = rq->idle; |
5265 | unsigned long flags; | |
5266 | ||
5267 | /* cpu has to be offline */ | |
48f24c4d | 5268 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5269 | |
48f24c4d IM |
5270 | /* |
5271 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5272 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5273 | */ |
5274 | spin_lock_irqsave(&rq->lock, flags); | |
5275 | ||
dd41f596 | 5276 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d IM |
5277 | |
5278 | /* Add idle task to the _front_ of its priority queue: */ | |
dd41f596 | 5279 | activate_idle_task(p, rq); |
1da177e4 LT |
5280 | |
5281 | spin_unlock_irqrestore(&rq->lock, flags); | |
5282 | } | |
5283 | ||
48f24c4d IM |
5284 | /* |
5285 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5286 | * offline. |
5287 | */ | |
5288 | void idle_task_exit(void) | |
5289 | { | |
5290 | struct mm_struct *mm = current->active_mm; | |
5291 | ||
5292 | BUG_ON(cpu_online(smp_processor_id())); | |
5293 | ||
5294 | if (mm != &init_mm) | |
5295 | switch_mm(mm, &init_mm, current); | |
5296 | mmdrop(mm); | |
5297 | } | |
5298 | ||
054b9108 | 5299 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5300 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5301 | { |
70b97a7f | 5302 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5303 | |
5304 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5305 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5306 | |
5307 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5308 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5309 | |
48f24c4d | 5310 | get_task_struct(p); |
1da177e4 LT |
5311 | |
5312 | /* | |
5313 | * Drop lock around migration; if someone else moves it, | |
5314 | * that's OK. No task can be added to this CPU, so iteration is | |
5315 | * fine. | |
5316 | */ | |
f7b4cddc | 5317 | spin_unlock_irq(&rq->lock); |
48f24c4d | 5318 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 5319 | spin_lock_irq(&rq->lock); |
1da177e4 | 5320 | |
48f24c4d | 5321 | put_task_struct(p); |
1da177e4 LT |
5322 | } |
5323 | ||
5324 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5325 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5326 | { | |
70b97a7f | 5327 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5328 | struct task_struct *next; |
48f24c4d | 5329 | |
dd41f596 IM |
5330 | for ( ; ; ) { |
5331 | if (!rq->nr_running) | |
5332 | break; | |
a8e504d2 | 5333 | update_rq_clock(rq); |
ff95f3df | 5334 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
5335 | if (!next) |
5336 | break; | |
5337 | migrate_dead(dead_cpu, next); | |
e692ab53 | 5338 | |
1da177e4 LT |
5339 | } |
5340 | } | |
5341 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5342 | ||
e692ab53 NP |
5343 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5344 | ||
5345 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5346 | { |
5347 | .procname = "sched_domain", | |
c57baf1e | 5348 | .mode = 0555, |
e0361851 | 5349 | }, |
e692ab53 NP |
5350 | {0,}, |
5351 | }; | |
5352 | ||
5353 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 5354 | { |
c57baf1e | 5355 | .ctl_name = CTL_KERN, |
e0361851 | 5356 | .procname = "kernel", |
c57baf1e | 5357 | .mode = 0555, |
e0361851 AD |
5358 | .child = sd_ctl_dir, |
5359 | }, | |
e692ab53 NP |
5360 | {0,}, |
5361 | }; | |
5362 | ||
5363 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5364 | { | |
5365 | struct ctl_table *entry = | |
5cf9f062 | 5366 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5367 | |
e692ab53 NP |
5368 | return entry; |
5369 | } | |
5370 | ||
6382bc90 MM |
5371 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5372 | { | |
cd790076 | 5373 | struct ctl_table *entry; |
6382bc90 | 5374 | |
cd790076 MM |
5375 | /* |
5376 | * In the intermediate directories, both the child directory and | |
5377 | * procname are dynamically allocated and could fail but the mode | |
5378 | * will always be set. In the lowest directory the names are | |
5379 | * static strings and all have proc handlers. | |
5380 | */ | |
5381 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5382 | if (entry->child) |
5383 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5384 | if (entry->proc_handler == NULL) |
5385 | kfree(entry->procname); | |
5386 | } | |
6382bc90 MM |
5387 | |
5388 | kfree(*tablep); | |
5389 | *tablep = NULL; | |
5390 | } | |
5391 | ||
e692ab53 | 5392 | static void |
e0361851 | 5393 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5394 | const char *procname, void *data, int maxlen, |
5395 | mode_t mode, proc_handler *proc_handler) | |
5396 | { | |
e692ab53 NP |
5397 | entry->procname = procname; |
5398 | entry->data = data; | |
5399 | entry->maxlen = maxlen; | |
5400 | entry->mode = mode; | |
5401 | entry->proc_handler = proc_handler; | |
5402 | } | |
5403 | ||
5404 | static struct ctl_table * | |
5405 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5406 | { | |
ace8b3d6 | 5407 | struct ctl_table *table = sd_alloc_ctl_entry(12); |
e692ab53 | 5408 | |
ad1cdc1d MM |
5409 | if (table == NULL) |
5410 | return NULL; | |
5411 | ||
e0361851 | 5412 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5413 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5414 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5415 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5416 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5417 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5418 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5419 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5420 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5421 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5422 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5423 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5424 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5425 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5426 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5427 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5428 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5429 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5430 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5431 | &sd->cache_nice_tries, |
5432 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5433 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5434 | sizeof(int), 0644, proc_dointvec_minmax); |
6323469f | 5435 | /* &table[11] is terminator */ |
e692ab53 NP |
5436 | |
5437 | return table; | |
5438 | } | |
5439 | ||
8401f775 | 5440 | static ctl_table * sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5441 | { |
5442 | struct ctl_table *entry, *table; | |
5443 | struct sched_domain *sd; | |
5444 | int domain_num = 0, i; | |
5445 | char buf[32]; | |
5446 | ||
5447 | for_each_domain(cpu, sd) | |
5448 | domain_num++; | |
5449 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5450 | if (table == NULL) |
5451 | return NULL; | |
e692ab53 NP |
5452 | |
5453 | i = 0; | |
5454 | for_each_domain(cpu, sd) { | |
5455 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5456 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5457 | entry->mode = 0555; |
e692ab53 NP |
5458 | entry->child = sd_alloc_ctl_domain_table(sd); |
5459 | entry++; | |
5460 | i++; | |
5461 | } | |
5462 | return table; | |
5463 | } | |
5464 | ||
5465 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5466 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
5467 | { |
5468 | int i, cpu_num = num_online_cpus(); | |
5469 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
5470 | char buf[32]; | |
5471 | ||
7378547f MM |
5472 | WARN_ON(sd_ctl_dir[0].child); |
5473 | sd_ctl_dir[0].child = entry; | |
5474 | ||
ad1cdc1d MM |
5475 | if (entry == NULL) |
5476 | return; | |
5477 | ||
97b6ea7b | 5478 | for_each_online_cpu(i) { |
e692ab53 | 5479 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5480 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5481 | entry->mode = 0555; |
e692ab53 | 5482 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5483 | entry++; |
e692ab53 | 5484 | } |
7378547f MM |
5485 | |
5486 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5487 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5488 | } | |
6382bc90 | 5489 | |
7378547f | 5490 | /* may be called multiple times per register */ |
6382bc90 MM |
5491 | static void unregister_sched_domain_sysctl(void) |
5492 | { | |
7378547f MM |
5493 | if (sd_sysctl_header) |
5494 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5495 | sd_sysctl_header = NULL; |
7378547f MM |
5496 | if (sd_ctl_dir[0].child) |
5497 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5498 | } |
e692ab53 | 5499 | #else |
6382bc90 MM |
5500 | static void register_sched_domain_sysctl(void) |
5501 | { | |
5502 | } | |
5503 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5504 | { |
5505 | } | |
5506 | #endif | |
5507 | ||
1da177e4 LT |
5508 | /* |
5509 | * migration_call - callback that gets triggered when a CPU is added. | |
5510 | * Here we can start up the necessary migration thread for the new CPU. | |
5511 | */ | |
48f24c4d IM |
5512 | static int __cpuinit |
5513 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5514 | { |
1da177e4 | 5515 | struct task_struct *p; |
48f24c4d | 5516 | int cpu = (long)hcpu; |
1da177e4 | 5517 | unsigned long flags; |
70b97a7f | 5518 | struct rq *rq; |
1da177e4 LT |
5519 | |
5520 | switch (action) { | |
5be9361c GS |
5521 | case CPU_LOCK_ACQUIRE: |
5522 | mutex_lock(&sched_hotcpu_mutex); | |
5523 | break; | |
5524 | ||
1da177e4 | 5525 | case CPU_UP_PREPARE: |
8bb78442 | 5526 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 5527 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
5528 | if (IS_ERR(p)) |
5529 | return NOTIFY_BAD; | |
1da177e4 LT |
5530 | kthread_bind(p, cpu); |
5531 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5532 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 5533 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
5534 | task_rq_unlock(rq, &flags); |
5535 | cpu_rq(cpu)->migration_thread = p; | |
5536 | break; | |
48f24c4d | 5537 | |
1da177e4 | 5538 | case CPU_ONLINE: |
8bb78442 | 5539 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 5540 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 LT |
5541 | wake_up_process(cpu_rq(cpu)->migration_thread); |
5542 | break; | |
48f24c4d | 5543 | |
1da177e4 LT |
5544 | #ifdef CONFIG_HOTPLUG_CPU |
5545 | case CPU_UP_CANCELED: | |
8bb78442 | 5546 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5547 | if (!cpu_rq(cpu)->migration_thread) |
5548 | break; | |
1da177e4 | 5549 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
5550 | kthread_bind(cpu_rq(cpu)->migration_thread, |
5551 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
5552 | kthread_stop(cpu_rq(cpu)->migration_thread); |
5553 | cpu_rq(cpu)->migration_thread = NULL; | |
5554 | break; | |
48f24c4d | 5555 | |
1da177e4 | 5556 | case CPU_DEAD: |
8bb78442 | 5557 | case CPU_DEAD_FROZEN: |
470fd646 | 5558 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
5559 | migrate_live_tasks(cpu); |
5560 | rq = cpu_rq(cpu); | |
5561 | kthread_stop(rq->migration_thread); | |
5562 | rq->migration_thread = NULL; | |
5563 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 5564 | spin_lock_irq(&rq->lock); |
a8e504d2 | 5565 | update_rq_clock(rq); |
2e1cb74a | 5566 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 5567 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
5568 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5569 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 5570 | migrate_dead_tasks(cpu); |
d2da272a | 5571 | spin_unlock_irq(&rq->lock); |
470fd646 | 5572 | cpuset_unlock(); |
1da177e4 LT |
5573 | migrate_nr_uninterruptible(rq); |
5574 | BUG_ON(rq->nr_running != 0); | |
5575 | ||
5576 | /* No need to migrate the tasks: it was best-effort if | |
5be9361c | 5577 | * they didn't take sched_hotcpu_mutex. Just wake up |
1da177e4 LT |
5578 | * the requestors. */ |
5579 | spin_lock_irq(&rq->lock); | |
5580 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
5581 | struct migration_req *req; |
5582 | ||
1da177e4 | 5583 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5584 | struct migration_req, list); |
1da177e4 LT |
5585 | list_del_init(&req->list); |
5586 | complete(&req->done); | |
5587 | } | |
5588 | spin_unlock_irq(&rq->lock); | |
5589 | break; | |
5590 | #endif | |
5be9361c GS |
5591 | case CPU_LOCK_RELEASE: |
5592 | mutex_unlock(&sched_hotcpu_mutex); | |
5593 | break; | |
1da177e4 LT |
5594 | } |
5595 | return NOTIFY_OK; | |
5596 | } | |
5597 | ||
5598 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
5599 | * happens before everything else. | |
5600 | */ | |
26c2143b | 5601 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5602 | .notifier_call = migration_call, |
5603 | .priority = 10 | |
5604 | }; | |
5605 | ||
5606 | int __init migration_init(void) | |
5607 | { | |
5608 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5609 | int err; |
48f24c4d IM |
5610 | |
5611 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5612 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5613 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5614 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5615 | register_cpu_notifier(&migration_notifier); | |
48f24c4d | 5616 | |
1da177e4 LT |
5617 | return 0; |
5618 | } | |
5619 | #endif | |
5620 | ||
5621 | #ifdef CONFIG_SMP | |
476f3534 CL |
5622 | |
5623 | /* Number of possible processor ids */ | |
5624 | int nr_cpu_ids __read_mostly = NR_CPUS; | |
5625 | EXPORT_SYMBOL(nr_cpu_ids); | |
5626 | ||
3e9830dc | 5627 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff IM |
5628 | |
5629 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level) | |
1da177e4 | 5630 | { |
4dcf6aff IM |
5631 | struct sched_group *group = sd->groups; |
5632 | cpumask_t groupmask; | |
5633 | char str[NR_CPUS]; | |
1da177e4 | 5634 | |
4dcf6aff IM |
5635 | cpumask_scnprintf(str, NR_CPUS, sd->span); |
5636 | cpus_clear(groupmask); | |
5637 | ||
5638 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5639 | ||
5640 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
5641 | printk("does not load-balance\n"); | |
5642 | if (sd->parent) | |
5643 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
5644 | " has parent"); | |
5645 | return -1; | |
41c7ce9a NP |
5646 | } |
5647 | ||
4dcf6aff IM |
5648 | printk(KERN_CONT "span %s\n", str); |
5649 | ||
5650 | if (!cpu_isset(cpu, sd->span)) { | |
5651 | printk(KERN_ERR "ERROR: domain->span does not contain " | |
5652 | "CPU%d\n", cpu); | |
5653 | } | |
5654 | if (!cpu_isset(cpu, group->cpumask)) { | |
5655 | printk(KERN_ERR "ERROR: domain->groups does not contain" | |
5656 | " CPU%d\n", cpu); | |
5657 | } | |
1da177e4 | 5658 | |
4dcf6aff | 5659 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5660 | do { |
4dcf6aff IM |
5661 | if (!group) { |
5662 | printk("\n"); | |
5663 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5664 | break; |
5665 | } | |
5666 | ||
4dcf6aff IM |
5667 | if (!group->__cpu_power) { |
5668 | printk(KERN_CONT "\n"); | |
5669 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5670 | "set\n"); | |
5671 | break; | |
5672 | } | |
1da177e4 | 5673 | |
4dcf6aff IM |
5674 | if (!cpus_weight(group->cpumask)) { |
5675 | printk(KERN_CONT "\n"); | |
5676 | printk(KERN_ERR "ERROR: empty group\n"); | |
5677 | break; | |
5678 | } | |
1da177e4 | 5679 | |
4dcf6aff IM |
5680 | if (cpus_intersects(groupmask, group->cpumask)) { |
5681 | printk(KERN_CONT "\n"); | |
5682 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
5683 | break; | |
5684 | } | |
1da177e4 | 5685 | |
4dcf6aff | 5686 | cpus_or(groupmask, groupmask, group->cpumask); |
1da177e4 | 5687 | |
4dcf6aff IM |
5688 | cpumask_scnprintf(str, NR_CPUS, group->cpumask); |
5689 | printk(KERN_CONT " %s", str); | |
1da177e4 | 5690 | |
4dcf6aff IM |
5691 | group = group->next; |
5692 | } while (group != sd->groups); | |
5693 | printk(KERN_CONT "\n"); | |
1da177e4 | 5694 | |
4dcf6aff IM |
5695 | if (!cpus_equal(sd->span, groupmask)) |
5696 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); | |
1da177e4 | 5697 | |
4dcf6aff IM |
5698 | if (sd->parent && !cpus_subset(groupmask, sd->parent->span)) |
5699 | printk(KERN_ERR "ERROR: parent span is not a superset " | |
5700 | "of domain->span\n"); | |
5701 | return 0; | |
5702 | } | |
1da177e4 | 5703 | |
4dcf6aff IM |
5704 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5705 | { | |
5706 | int level = 0; | |
1da177e4 | 5707 | |
4dcf6aff IM |
5708 | if (!sd) { |
5709 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5710 | return; | |
5711 | } | |
1da177e4 | 5712 | |
4dcf6aff IM |
5713 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5714 | ||
5715 | for (;;) { | |
5716 | if (sched_domain_debug_one(sd, cpu, level)) | |
5717 | break; | |
1da177e4 LT |
5718 | level++; |
5719 | sd = sd->parent; | |
33859f7f | 5720 | if (!sd) |
4dcf6aff IM |
5721 | break; |
5722 | } | |
1da177e4 LT |
5723 | } |
5724 | #else | |
48f24c4d | 5725 | # define sched_domain_debug(sd, cpu) do { } while (0) |
1da177e4 LT |
5726 | #endif |
5727 | ||
1a20ff27 | 5728 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
5729 | { |
5730 | if (cpus_weight(sd->span) == 1) | |
5731 | return 1; | |
5732 | ||
5733 | /* Following flags need at least 2 groups */ | |
5734 | if (sd->flags & (SD_LOAD_BALANCE | | |
5735 | SD_BALANCE_NEWIDLE | | |
5736 | SD_BALANCE_FORK | | |
89c4710e SS |
5737 | SD_BALANCE_EXEC | |
5738 | SD_SHARE_CPUPOWER | | |
5739 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5740 | if (sd->groups != sd->groups->next) |
5741 | return 0; | |
5742 | } | |
5743 | ||
5744 | /* Following flags don't use groups */ | |
5745 | if (sd->flags & (SD_WAKE_IDLE | | |
5746 | SD_WAKE_AFFINE | | |
5747 | SD_WAKE_BALANCE)) | |
5748 | return 0; | |
5749 | ||
5750 | return 1; | |
5751 | } | |
5752 | ||
48f24c4d IM |
5753 | static int |
5754 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5755 | { |
5756 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5757 | ||
5758 | if (sd_degenerate(parent)) | |
5759 | return 1; | |
5760 | ||
5761 | if (!cpus_equal(sd->span, parent->span)) | |
5762 | return 0; | |
5763 | ||
5764 | /* Does parent contain flags not in child? */ | |
5765 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
5766 | if (cflags & SD_WAKE_AFFINE) | |
5767 | pflags &= ~SD_WAKE_BALANCE; | |
5768 | /* Flags needing groups don't count if only 1 group in parent */ | |
5769 | if (parent->groups == parent->groups->next) { | |
5770 | pflags &= ~(SD_LOAD_BALANCE | | |
5771 | SD_BALANCE_NEWIDLE | | |
5772 | SD_BALANCE_FORK | | |
89c4710e SS |
5773 | SD_BALANCE_EXEC | |
5774 | SD_SHARE_CPUPOWER | | |
5775 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
5776 | } |
5777 | if (~cflags & pflags) | |
5778 | return 0; | |
5779 | ||
5780 | return 1; | |
5781 | } | |
5782 | ||
1da177e4 LT |
5783 | /* |
5784 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must | |
5785 | * hold the hotplug lock. | |
5786 | */ | |
9c1cfda2 | 5787 | static void cpu_attach_domain(struct sched_domain *sd, int cpu) |
1da177e4 | 5788 | { |
70b97a7f | 5789 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5790 | struct sched_domain *tmp; |
5791 | ||
5792 | /* Remove the sched domains which do not contribute to scheduling. */ | |
5793 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
5794 | struct sched_domain *parent = tmp->parent; | |
5795 | if (!parent) | |
5796 | break; | |
1a848870 | 5797 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5798 | tmp->parent = parent->parent; |
1a848870 SS |
5799 | if (parent->parent) |
5800 | parent->parent->child = tmp; | |
5801 | } | |
245af2c7 SS |
5802 | } |
5803 | ||
1a848870 | 5804 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 5805 | sd = sd->parent; |
1a848870 SS |
5806 | if (sd) |
5807 | sd->child = NULL; | |
5808 | } | |
1da177e4 LT |
5809 | |
5810 | sched_domain_debug(sd, cpu); | |
5811 | ||
674311d5 | 5812 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
5813 | } |
5814 | ||
5815 | /* cpus with isolated domains */ | |
67af63a6 | 5816 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
5817 | |
5818 | /* Setup the mask of cpus configured for isolated domains */ | |
5819 | static int __init isolated_cpu_setup(char *str) | |
5820 | { | |
5821 | int ints[NR_CPUS], i; | |
5822 | ||
5823 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
5824 | cpus_clear(cpu_isolated_map); | |
5825 | for (i = 1; i <= ints[0]; i++) | |
5826 | if (ints[i] < NR_CPUS) | |
5827 | cpu_set(ints[i], cpu_isolated_map); | |
5828 | return 1; | |
5829 | } | |
5830 | ||
8927f494 | 5831 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
5832 | |
5833 | /* | |
6711cab4 SS |
5834 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
5835 | * to a function which identifies what group(along with sched group) a CPU | |
5836 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
5837 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
5838 | * |
5839 | * init_sched_build_groups will build a circular linked list of the groups | |
5840 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5841 | * and ->cpu_power to 0. | |
5842 | */ | |
a616058b | 5843 | static void |
6711cab4 SS |
5844 | init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map, |
5845 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, | |
5846 | struct sched_group **sg)) | |
1da177e4 LT |
5847 | { |
5848 | struct sched_group *first = NULL, *last = NULL; | |
5849 | cpumask_t covered = CPU_MASK_NONE; | |
5850 | int i; | |
5851 | ||
5852 | for_each_cpu_mask(i, span) { | |
6711cab4 SS |
5853 | struct sched_group *sg; |
5854 | int group = group_fn(i, cpu_map, &sg); | |
1da177e4 LT |
5855 | int j; |
5856 | ||
5857 | if (cpu_isset(i, covered)) | |
5858 | continue; | |
5859 | ||
5860 | sg->cpumask = CPU_MASK_NONE; | |
5517d86b | 5861 | sg->__cpu_power = 0; |
1da177e4 LT |
5862 | |
5863 | for_each_cpu_mask(j, span) { | |
6711cab4 | 5864 | if (group_fn(j, cpu_map, NULL) != group) |
1da177e4 LT |
5865 | continue; |
5866 | ||
5867 | cpu_set(j, covered); | |
5868 | cpu_set(j, sg->cpumask); | |
5869 | } | |
5870 | if (!first) | |
5871 | first = sg; | |
5872 | if (last) | |
5873 | last->next = sg; | |
5874 | last = sg; | |
5875 | } | |
5876 | last->next = first; | |
5877 | } | |
5878 | ||
9c1cfda2 | 5879 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 5880 | |
9c1cfda2 | 5881 | #ifdef CONFIG_NUMA |
198e2f18 | 5882 | |
9c1cfda2 JH |
5883 | /** |
5884 | * find_next_best_node - find the next node to include in a sched_domain | |
5885 | * @node: node whose sched_domain we're building | |
5886 | * @used_nodes: nodes already in the sched_domain | |
5887 | * | |
5888 | * Find the next node to include in a given scheduling domain. Simply | |
5889 | * finds the closest node not already in the @used_nodes map. | |
5890 | * | |
5891 | * Should use nodemask_t. | |
5892 | */ | |
5893 | static int find_next_best_node(int node, unsigned long *used_nodes) | |
5894 | { | |
5895 | int i, n, val, min_val, best_node = 0; | |
5896 | ||
5897 | min_val = INT_MAX; | |
5898 | ||
5899 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5900 | /* Start at @node */ | |
5901 | n = (node + i) % MAX_NUMNODES; | |
5902 | ||
5903 | if (!nr_cpus_node(n)) | |
5904 | continue; | |
5905 | ||
5906 | /* Skip already used nodes */ | |
5907 | if (test_bit(n, used_nodes)) | |
5908 | continue; | |
5909 | ||
5910 | /* Simple min distance search */ | |
5911 | val = node_distance(node, n); | |
5912 | ||
5913 | if (val < min_val) { | |
5914 | min_val = val; | |
5915 | best_node = n; | |
5916 | } | |
5917 | } | |
5918 | ||
5919 | set_bit(best_node, used_nodes); | |
5920 | return best_node; | |
5921 | } | |
5922 | ||
5923 | /** | |
5924 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
5925 | * @node: node whose cpumask we're constructing | |
5926 | * @size: number of nodes to include in this span | |
5927 | * | |
5928 | * Given a node, construct a good cpumask for its sched_domain to span. It | |
5929 | * should be one that prevents unnecessary balancing, but also spreads tasks | |
5930 | * out optimally. | |
5931 | */ | |
5932 | static cpumask_t sched_domain_node_span(int node) | |
5933 | { | |
9c1cfda2 | 5934 | DECLARE_BITMAP(used_nodes, MAX_NUMNODES); |
48f24c4d IM |
5935 | cpumask_t span, nodemask; |
5936 | int i; | |
9c1cfda2 JH |
5937 | |
5938 | cpus_clear(span); | |
5939 | bitmap_zero(used_nodes, MAX_NUMNODES); | |
5940 | ||
5941 | nodemask = node_to_cpumask(node); | |
5942 | cpus_or(span, span, nodemask); | |
5943 | set_bit(node, used_nodes); | |
5944 | ||
5945 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
5946 | int next_node = find_next_best_node(node, used_nodes); | |
48f24c4d | 5947 | |
9c1cfda2 JH |
5948 | nodemask = node_to_cpumask(next_node); |
5949 | cpus_or(span, span, nodemask); | |
5950 | } | |
5951 | ||
5952 | return span; | |
5953 | } | |
5954 | #endif | |
5955 | ||
5c45bf27 | 5956 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 5957 | |
9c1cfda2 | 5958 | /* |
48f24c4d | 5959 | * SMT sched-domains: |
9c1cfda2 | 5960 | */ |
1da177e4 LT |
5961 | #ifdef CONFIG_SCHED_SMT |
5962 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 5963 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 5964 | |
6711cab4 SS |
5965 | static int cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, |
5966 | struct sched_group **sg) | |
1da177e4 | 5967 | { |
6711cab4 SS |
5968 | if (sg) |
5969 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
5970 | return cpu; |
5971 | } | |
5972 | #endif | |
5973 | ||
48f24c4d IM |
5974 | /* |
5975 | * multi-core sched-domains: | |
5976 | */ | |
1e9f28fa SS |
5977 | #ifdef CONFIG_SCHED_MC |
5978 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 5979 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
5980 | #endif |
5981 | ||
5982 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
6711cab4 SS |
5983 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5984 | struct sched_group **sg) | |
1e9f28fa | 5985 | { |
6711cab4 | 5986 | int group; |
d5a7430d | 5987 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 5988 | cpus_and(mask, mask, *cpu_map); |
6711cab4 SS |
5989 | group = first_cpu(mask); |
5990 | if (sg) | |
5991 | *sg = &per_cpu(sched_group_core, group); | |
5992 | return group; | |
1e9f28fa SS |
5993 | } |
5994 | #elif defined(CONFIG_SCHED_MC) | |
6711cab4 SS |
5995 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5996 | struct sched_group **sg) | |
1e9f28fa | 5997 | { |
6711cab4 SS |
5998 | if (sg) |
5999 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
6000 | return cpu; |
6001 | } | |
6002 | #endif | |
6003 | ||
1da177e4 | 6004 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 6005 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 6006 | |
6711cab4 SS |
6007 | static int cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, |
6008 | struct sched_group **sg) | |
1da177e4 | 6009 | { |
6711cab4 | 6010 | int group; |
48f24c4d | 6011 | #ifdef CONFIG_SCHED_MC |
1e9f28fa | 6012 | cpumask_t mask = cpu_coregroup_map(cpu); |
a616058b | 6013 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 6014 | group = first_cpu(mask); |
1e9f28fa | 6015 | #elif defined(CONFIG_SCHED_SMT) |
d5a7430d | 6016 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 6017 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 6018 | group = first_cpu(mask); |
1da177e4 | 6019 | #else |
6711cab4 | 6020 | group = cpu; |
1da177e4 | 6021 | #endif |
6711cab4 SS |
6022 | if (sg) |
6023 | *sg = &per_cpu(sched_group_phys, group); | |
6024 | return group; | |
1da177e4 LT |
6025 | } |
6026 | ||
6027 | #ifdef CONFIG_NUMA | |
1da177e4 | 6028 | /* |
9c1cfda2 JH |
6029 | * The init_sched_build_groups can't handle what we want to do with node |
6030 | * groups, so roll our own. Now each node has its own list of groups which | |
6031 | * gets dynamically allocated. | |
1da177e4 | 6032 | */ |
9c1cfda2 | 6033 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
d1b55138 | 6034 | static struct sched_group **sched_group_nodes_bycpu[NR_CPUS]; |
1da177e4 | 6035 | |
9c1cfda2 | 6036 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 6037 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 6038 | |
6711cab4 SS |
6039 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
6040 | struct sched_group **sg) | |
9c1cfda2 | 6041 | { |
6711cab4 SS |
6042 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu)); |
6043 | int group; | |
6044 | ||
6045 | cpus_and(nodemask, nodemask, *cpu_map); | |
6046 | group = first_cpu(nodemask); | |
6047 | ||
6048 | if (sg) | |
6049 | *sg = &per_cpu(sched_group_allnodes, group); | |
6050 | return group; | |
1da177e4 | 6051 | } |
6711cab4 | 6052 | |
08069033 SS |
6053 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6054 | { | |
6055 | struct sched_group *sg = group_head; | |
6056 | int j; | |
6057 | ||
6058 | if (!sg) | |
6059 | return; | |
3a5c359a AK |
6060 | do { |
6061 | for_each_cpu_mask(j, sg->cpumask) { | |
6062 | struct sched_domain *sd; | |
08069033 | 6063 | |
3a5c359a AK |
6064 | sd = &per_cpu(phys_domains, j); |
6065 | if (j != first_cpu(sd->groups->cpumask)) { | |
6066 | /* | |
6067 | * Only add "power" once for each | |
6068 | * physical package. | |
6069 | */ | |
6070 | continue; | |
6071 | } | |
08069033 | 6072 | |
3a5c359a AK |
6073 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
6074 | } | |
6075 | sg = sg->next; | |
6076 | } while (sg != group_head); | |
08069033 | 6077 | } |
1da177e4 LT |
6078 | #endif |
6079 | ||
a616058b | 6080 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6081 | /* Free memory allocated for various sched_group structures */ |
6082 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6083 | { | |
a616058b | 6084 | int cpu, i; |
51888ca2 SV |
6085 | |
6086 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
6087 | struct sched_group **sched_group_nodes |
6088 | = sched_group_nodes_bycpu[cpu]; | |
6089 | ||
51888ca2 SV |
6090 | if (!sched_group_nodes) |
6091 | continue; | |
6092 | ||
6093 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6094 | cpumask_t nodemask = node_to_cpumask(i); | |
6095 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; | |
6096 | ||
6097 | cpus_and(nodemask, nodemask, *cpu_map); | |
6098 | if (cpus_empty(nodemask)) | |
6099 | continue; | |
6100 | ||
6101 | if (sg == NULL) | |
6102 | continue; | |
6103 | sg = sg->next; | |
6104 | next_sg: | |
6105 | oldsg = sg; | |
6106 | sg = sg->next; | |
6107 | kfree(oldsg); | |
6108 | if (oldsg != sched_group_nodes[i]) | |
6109 | goto next_sg; | |
6110 | } | |
6111 | kfree(sched_group_nodes); | |
6112 | sched_group_nodes_bycpu[cpu] = NULL; | |
6113 | } | |
51888ca2 | 6114 | } |
a616058b SS |
6115 | #else |
6116 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6117 | { | |
6118 | } | |
6119 | #endif | |
51888ca2 | 6120 | |
89c4710e SS |
6121 | /* |
6122 | * Initialize sched groups cpu_power. | |
6123 | * | |
6124 | * cpu_power indicates the capacity of sched group, which is used while | |
6125 | * distributing the load between different sched groups in a sched domain. | |
6126 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6127 | * there are asymmetries in the topology. If there are asymmetries, group | |
6128 | * having more cpu_power will pickup more load compared to the group having | |
6129 | * less cpu_power. | |
6130 | * | |
6131 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
6132 | * the maximum number of tasks a group can handle in the presence of other idle | |
6133 | * or lightly loaded groups in the same sched domain. | |
6134 | */ | |
6135 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6136 | { | |
6137 | struct sched_domain *child; | |
6138 | struct sched_group *group; | |
6139 | ||
6140 | WARN_ON(!sd || !sd->groups); | |
6141 | ||
6142 | if (cpu != first_cpu(sd->groups->cpumask)) | |
6143 | return; | |
6144 | ||
6145 | child = sd->child; | |
6146 | ||
5517d86b ED |
6147 | sd->groups->__cpu_power = 0; |
6148 | ||
89c4710e SS |
6149 | /* |
6150 | * For perf policy, if the groups in child domain share resources | |
6151 | * (for example cores sharing some portions of the cache hierarchy | |
6152 | * or SMT), then set this domain groups cpu_power such that each group | |
6153 | * can handle only one task, when there are other idle groups in the | |
6154 | * same sched domain. | |
6155 | */ | |
6156 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
6157 | (child->flags & | |
6158 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 6159 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
6160 | return; |
6161 | } | |
6162 | ||
89c4710e SS |
6163 | /* |
6164 | * add cpu_power of each child group to this groups cpu_power | |
6165 | */ | |
6166 | group = child->groups; | |
6167 | do { | |
5517d86b | 6168 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
6169 | group = group->next; |
6170 | } while (group != child->groups); | |
6171 | } | |
6172 | ||
1da177e4 | 6173 | /* |
1a20ff27 DG |
6174 | * Build sched domains for a given set of cpus and attach the sched domains |
6175 | * to the individual cpus | |
1da177e4 | 6176 | */ |
51888ca2 | 6177 | static int build_sched_domains(const cpumask_t *cpu_map) |
1da177e4 LT |
6178 | { |
6179 | int i; | |
d1b55138 JH |
6180 | #ifdef CONFIG_NUMA |
6181 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 6182 | int sd_allnodes = 0; |
d1b55138 JH |
6183 | |
6184 | /* | |
6185 | * Allocate the per-node list of sched groups | |
6186 | */ | |
5cf9f062 | 6187 | sched_group_nodes = kcalloc(MAX_NUMNODES, sizeof(struct sched_group *), |
d3a5aa98 | 6188 | GFP_KERNEL); |
d1b55138 JH |
6189 | if (!sched_group_nodes) { |
6190 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 6191 | return -ENOMEM; |
d1b55138 JH |
6192 | } |
6193 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
6194 | #endif | |
1da177e4 LT |
6195 | |
6196 | /* | |
1a20ff27 | 6197 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 6198 | */ |
1a20ff27 | 6199 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6200 | struct sched_domain *sd = NULL, *p; |
6201 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(i)); | |
6202 | ||
1a20ff27 | 6203 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6204 | |
6205 | #ifdef CONFIG_NUMA | |
dd41f596 IM |
6206 | if (cpus_weight(*cpu_map) > |
6207 | SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) { | |
9c1cfda2 JH |
6208 | sd = &per_cpu(allnodes_domains, i); |
6209 | *sd = SD_ALLNODES_INIT; | |
6210 | sd->span = *cpu_map; | |
6711cab4 | 6211 | cpu_to_allnodes_group(i, cpu_map, &sd->groups); |
9c1cfda2 | 6212 | p = sd; |
6711cab4 | 6213 | sd_allnodes = 1; |
9c1cfda2 JH |
6214 | } else |
6215 | p = NULL; | |
6216 | ||
1da177e4 | 6217 | sd = &per_cpu(node_domains, i); |
1da177e4 | 6218 | *sd = SD_NODE_INIT; |
9c1cfda2 JH |
6219 | sd->span = sched_domain_node_span(cpu_to_node(i)); |
6220 | sd->parent = p; | |
1a848870 SS |
6221 | if (p) |
6222 | p->child = sd; | |
9c1cfda2 | 6223 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
6224 | #endif |
6225 | ||
6226 | p = sd; | |
6227 | sd = &per_cpu(phys_domains, i); | |
1da177e4 LT |
6228 | *sd = SD_CPU_INIT; |
6229 | sd->span = nodemask; | |
6230 | sd->parent = p; | |
1a848870 SS |
6231 | if (p) |
6232 | p->child = sd; | |
6711cab4 | 6233 | cpu_to_phys_group(i, cpu_map, &sd->groups); |
1da177e4 | 6234 | |
1e9f28fa SS |
6235 | #ifdef CONFIG_SCHED_MC |
6236 | p = sd; | |
6237 | sd = &per_cpu(core_domains, i); | |
1e9f28fa SS |
6238 | *sd = SD_MC_INIT; |
6239 | sd->span = cpu_coregroup_map(i); | |
6240 | cpus_and(sd->span, sd->span, *cpu_map); | |
6241 | sd->parent = p; | |
1a848870 | 6242 | p->child = sd; |
6711cab4 | 6243 | cpu_to_core_group(i, cpu_map, &sd->groups); |
1e9f28fa SS |
6244 | #endif |
6245 | ||
1da177e4 LT |
6246 | #ifdef CONFIG_SCHED_SMT |
6247 | p = sd; | |
6248 | sd = &per_cpu(cpu_domains, i); | |
1da177e4 | 6249 | *sd = SD_SIBLING_INIT; |
d5a7430d | 6250 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6251 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 6252 | sd->parent = p; |
1a848870 | 6253 | p->child = sd; |
6711cab4 | 6254 | cpu_to_cpu_group(i, cpu_map, &sd->groups); |
1da177e4 LT |
6255 | #endif |
6256 | } | |
6257 | ||
6258 | #ifdef CONFIG_SCHED_SMT | |
6259 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 6260 | for_each_cpu_mask(i, *cpu_map) { |
d5a7430d | 6261 | cpumask_t this_sibling_map = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6262 | cpus_and(this_sibling_map, this_sibling_map, *cpu_map); |
1da177e4 LT |
6263 | if (i != first_cpu(this_sibling_map)) |
6264 | continue; | |
6265 | ||
dd41f596 IM |
6266 | init_sched_build_groups(this_sibling_map, cpu_map, |
6267 | &cpu_to_cpu_group); | |
1da177e4 LT |
6268 | } |
6269 | #endif | |
6270 | ||
1e9f28fa SS |
6271 | #ifdef CONFIG_SCHED_MC |
6272 | /* Set up multi-core groups */ | |
6273 | for_each_cpu_mask(i, *cpu_map) { | |
6274 | cpumask_t this_core_map = cpu_coregroup_map(i); | |
6275 | cpus_and(this_core_map, this_core_map, *cpu_map); | |
6276 | if (i != first_cpu(this_core_map)) | |
6277 | continue; | |
dd41f596 IM |
6278 | init_sched_build_groups(this_core_map, cpu_map, |
6279 | &cpu_to_core_group); | |
1e9f28fa SS |
6280 | } |
6281 | #endif | |
6282 | ||
1da177e4 LT |
6283 | /* Set up physical groups */ |
6284 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6285 | cpumask_t nodemask = node_to_cpumask(i); | |
6286 | ||
1a20ff27 | 6287 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6288 | if (cpus_empty(nodemask)) |
6289 | continue; | |
6290 | ||
6711cab4 | 6291 | init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group); |
1da177e4 LT |
6292 | } |
6293 | ||
6294 | #ifdef CONFIG_NUMA | |
6295 | /* Set up node groups */ | |
6711cab4 | 6296 | if (sd_allnodes) |
dd41f596 IM |
6297 | init_sched_build_groups(*cpu_map, cpu_map, |
6298 | &cpu_to_allnodes_group); | |
9c1cfda2 JH |
6299 | |
6300 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6301 | /* Set up node groups */ | |
6302 | struct sched_group *sg, *prev; | |
6303 | cpumask_t nodemask = node_to_cpumask(i); | |
6304 | cpumask_t domainspan; | |
6305 | cpumask_t covered = CPU_MASK_NONE; | |
6306 | int j; | |
6307 | ||
6308 | cpus_and(nodemask, nodemask, *cpu_map); | |
d1b55138 JH |
6309 | if (cpus_empty(nodemask)) { |
6310 | sched_group_nodes[i] = NULL; | |
9c1cfda2 | 6311 | continue; |
d1b55138 | 6312 | } |
9c1cfda2 JH |
6313 | |
6314 | domainspan = sched_domain_node_span(i); | |
6315 | cpus_and(domainspan, domainspan, *cpu_map); | |
6316 | ||
15f0b676 | 6317 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
6318 | if (!sg) { |
6319 | printk(KERN_WARNING "Can not alloc domain group for " | |
6320 | "node %d\n", i); | |
6321 | goto error; | |
6322 | } | |
9c1cfda2 JH |
6323 | sched_group_nodes[i] = sg; |
6324 | for_each_cpu_mask(j, nodemask) { | |
6325 | struct sched_domain *sd; | |
9761eea8 | 6326 | |
9c1cfda2 JH |
6327 | sd = &per_cpu(node_domains, j); |
6328 | sd->groups = sg; | |
9c1cfda2 | 6329 | } |
5517d86b | 6330 | sg->__cpu_power = 0; |
9c1cfda2 | 6331 | sg->cpumask = nodemask; |
51888ca2 | 6332 | sg->next = sg; |
9c1cfda2 JH |
6333 | cpus_or(covered, covered, nodemask); |
6334 | prev = sg; | |
6335 | ||
6336 | for (j = 0; j < MAX_NUMNODES; j++) { | |
6337 | cpumask_t tmp, notcovered; | |
6338 | int n = (i + j) % MAX_NUMNODES; | |
6339 | ||
6340 | cpus_complement(notcovered, covered); | |
6341 | cpus_and(tmp, notcovered, *cpu_map); | |
6342 | cpus_and(tmp, tmp, domainspan); | |
6343 | if (cpus_empty(tmp)) | |
6344 | break; | |
6345 | ||
6346 | nodemask = node_to_cpumask(n); | |
6347 | cpus_and(tmp, tmp, nodemask); | |
6348 | if (cpus_empty(tmp)) | |
6349 | continue; | |
6350 | ||
15f0b676 SV |
6351 | sg = kmalloc_node(sizeof(struct sched_group), |
6352 | GFP_KERNEL, i); | |
9c1cfda2 JH |
6353 | if (!sg) { |
6354 | printk(KERN_WARNING | |
6355 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 6356 | goto error; |
9c1cfda2 | 6357 | } |
5517d86b | 6358 | sg->__cpu_power = 0; |
9c1cfda2 | 6359 | sg->cpumask = tmp; |
51888ca2 | 6360 | sg->next = prev->next; |
9c1cfda2 JH |
6361 | cpus_or(covered, covered, tmp); |
6362 | prev->next = sg; | |
6363 | prev = sg; | |
6364 | } | |
9c1cfda2 | 6365 | } |
1da177e4 LT |
6366 | #endif |
6367 | ||
6368 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 6369 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 6370 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6371 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
6372 | ||
89c4710e | 6373 | init_sched_groups_power(i, sd); |
5c45bf27 | 6374 | } |
1da177e4 | 6375 | #endif |
1e9f28fa | 6376 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 6377 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6378 | struct sched_domain *sd = &per_cpu(core_domains, i); |
6379 | ||
89c4710e | 6380 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
6381 | } |
6382 | #endif | |
1e9f28fa | 6383 | |
5c45bf27 | 6384 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6385 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
6386 | ||
89c4710e | 6387 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6388 | } |
6389 | ||
9c1cfda2 | 6390 | #ifdef CONFIG_NUMA |
08069033 SS |
6391 | for (i = 0; i < MAX_NUMNODES; i++) |
6392 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 6393 | |
6711cab4 SS |
6394 | if (sd_allnodes) { |
6395 | struct sched_group *sg; | |
f712c0c7 | 6396 | |
6711cab4 | 6397 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg); |
f712c0c7 SS |
6398 | init_numa_sched_groups_power(sg); |
6399 | } | |
9c1cfda2 JH |
6400 | #endif |
6401 | ||
1da177e4 | 6402 | /* Attach the domains */ |
1a20ff27 | 6403 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6404 | struct sched_domain *sd; |
6405 | #ifdef CONFIG_SCHED_SMT | |
6406 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
6407 | #elif defined(CONFIG_SCHED_MC) |
6408 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
6409 | #else |
6410 | sd = &per_cpu(phys_domains, i); | |
6411 | #endif | |
6412 | cpu_attach_domain(sd, i); | |
6413 | } | |
51888ca2 SV |
6414 | |
6415 | return 0; | |
6416 | ||
a616058b | 6417 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6418 | error: |
6419 | free_sched_groups(cpu_map); | |
6420 | return -ENOMEM; | |
a616058b | 6421 | #endif |
1da177e4 | 6422 | } |
029190c5 PJ |
6423 | |
6424 | static cpumask_t *doms_cur; /* current sched domains */ | |
6425 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
6426 | ||
6427 | /* | |
6428 | * Special case: If a kmalloc of a doms_cur partition (array of | |
6429 | * cpumask_t) fails, then fallback to a single sched domain, | |
6430 | * as determined by the single cpumask_t fallback_doms. | |
6431 | */ | |
6432 | static cpumask_t fallback_doms; | |
6433 | ||
1a20ff27 DG |
6434 | /* |
6435 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. | |
029190c5 PJ |
6436 | * For now this just excludes isolated cpus, but could be used to |
6437 | * exclude other special cases in the future. | |
1a20ff27 | 6438 | */ |
51888ca2 | 6439 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 6440 | { |
7378547f MM |
6441 | int err; |
6442 | ||
029190c5 PJ |
6443 | ndoms_cur = 1; |
6444 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
6445 | if (!doms_cur) | |
6446 | doms_cur = &fallback_doms; | |
6447 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
7378547f | 6448 | err = build_sched_domains(doms_cur); |
6382bc90 | 6449 | register_sched_domain_sysctl(); |
7378547f MM |
6450 | |
6451 | return err; | |
1a20ff27 DG |
6452 | } |
6453 | ||
6454 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map) | |
1da177e4 | 6455 | { |
51888ca2 | 6456 | free_sched_groups(cpu_map); |
9c1cfda2 | 6457 | } |
1da177e4 | 6458 | |
1a20ff27 DG |
6459 | /* |
6460 | * Detach sched domains from a group of cpus specified in cpu_map | |
6461 | * These cpus will now be attached to the NULL domain | |
6462 | */ | |
858119e1 | 6463 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6464 | { |
6465 | int i; | |
6466 | ||
6382bc90 MM |
6467 | unregister_sched_domain_sysctl(); |
6468 | ||
1a20ff27 DG |
6469 | for_each_cpu_mask(i, *cpu_map) |
6470 | cpu_attach_domain(NULL, i); | |
6471 | synchronize_sched(); | |
6472 | arch_destroy_sched_domains(cpu_map); | |
6473 | } | |
6474 | ||
029190c5 PJ |
6475 | /* |
6476 | * Partition sched domains as specified by the 'ndoms_new' | |
6477 | * cpumasks in the array doms_new[] of cpumasks. This compares | |
6478 | * doms_new[] to the current sched domain partitioning, doms_cur[]. | |
6479 | * It destroys each deleted domain and builds each new domain. | |
6480 | * | |
6481 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
6482 | * The masks don't intersect (don't overlap.) We should setup one | |
6483 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6484 | * not be load balanced. If the same cpumask appears both in the | |
6485 | * current 'doms_cur' domains and in the new 'doms_new', we can leave | |
6486 | * it as it is. | |
6487 | * | |
6488 | * The passed in 'doms_new' should be kmalloc'd. This routine takes | |
6489 | * ownership of it and will kfree it when done with it. If the caller | |
6490 | * failed the kmalloc call, then it can pass in doms_new == NULL, | |
6491 | * and partition_sched_domains() will fallback to the single partition | |
6492 | * 'fallback_doms'. | |
6493 | * | |
6494 | * Call with hotplug lock held | |
6495 | */ | |
6496 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new) | |
6497 | { | |
6498 | int i, j; | |
6499 | ||
7378547f MM |
6500 | /* always unregister in case we don't destroy any domains */ |
6501 | unregister_sched_domain_sysctl(); | |
6502 | ||
029190c5 PJ |
6503 | if (doms_new == NULL) { |
6504 | ndoms_new = 1; | |
6505 | doms_new = &fallback_doms; | |
6506 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
6507 | } | |
6508 | ||
6509 | /* Destroy deleted domains */ | |
6510 | for (i = 0; i < ndoms_cur; i++) { | |
6511 | for (j = 0; j < ndoms_new; j++) { | |
6512 | if (cpus_equal(doms_cur[i], doms_new[j])) | |
6513 | goto match1; | |
6514 | } | |
6515 | /* no match - a current sched domain not in new doms_new[] */ | |
6516 | detach_destroy_domains(doms_cur + i); | |
6517 | match1: | |
6518 | ; | |
6519 | } | |
6520 | ||
6521 | /* Build new domains */ | |
6522 | for (i = 0; i < ndoms_new; i++) { | |
6523 | for (j = 0; j < ndoms_cur; j++) { | |
6524 | if (cpus_equal(doms_new[i], doms_cur[j])) | |
6525 | goto match2; | |
6526 | } | |
6527 | /* no match - add a new doms_new */ | |
6528 | build_sched_domains(doms_new + i); | |
6529 | match2: | |
6530 | ; | |
6531 | } | |
6532 | ||
6533 | /* Remember the new sched domains */ | |
6534 | if (doms_cur != &fallback_doms) | |
6535 | kfree(doms_cur); | |
6536 | doms_cur = doms_new; | |
6537 | ndoms_cur = ndoms_new; | |
7378547f MM |
6538 | |
6539 | register_sched_domain_sysctl(); | |
029190c5 PJ |
6540 | } |
6541 | ||
5c45bf27 | 6542 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
6707de00 | 6543 | static int arch_reinit_sched_domains(void) |
5c45bf27 SS |
6544 | { |
6545 | int err; | |
6546 | ||
5be9361c | 6547 | mutex_lock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6548 | detach_destroy_domains(&cpu_online_map); |
6549 | err = arch_init_sched_domains(&cpu_online_map); | |
5be9361c | 6550 | mutex_unlock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6551 | |
6552 | return err; | |
6553 | } | |
6554 | ||
6555 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6556 | { | |
6557 | int ret; | |
6558 | ||
6559 | if (buf[0] != '0' && buf[0] != '1') | |
6560 | return -EINVAL; | |
6561 | ||
6562 | if (smt) | |
6563 | sched_smt_power_savings = (buf[0] == '1'); | |
6564 | else | |
6565 | sched_mc_power_savings = (buf[0] == '1'); | |
6566 | ||
6567 | ret = arch_reinit_sched_domains(); | |
6568 | ||
6569 | return ret ? ret : count; | |
6570 | } | |
6571 | ||
5c45bf27 SS |
6572 | #ifdef CONFIG_SCHED_MC |
6573 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
6574 | { | |
6575 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
6576 | } | |
48f24c4d IM |
6577 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
6578 | const char *buf, size_t count) | |
5c45bf27 SS |
6579 | { |
6580 | return sched_power_savings_store(buf, count, 0); | |
6581 | } | |
6707de00 AB |
6582 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
6583 | sched_mc_power_savings_store); | |
5c45bf27 SS |
6584 | #endif |
6585 | ||
6586 | #ifdef CONFIG_SCHED_SMT | |
6587 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
6588 | { | |
6589 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
6590 | } | |
48f24c4d IM |
6591 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
6592 | const char *buf, size_t count) | |
5c45bf27 SS |
6593 | { |
6594 | return sched_power_savings_store(buf, count, 1); | |
6595 | } | |
6707de00 AB |
6596 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
6597 | sched_smt_power_savings_store); | |
6598 | #endif | |
6599 | ||
6600 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
6601 | { | |
6602 | int err = 0; | |
6603 | ||
6604 | #ifdef CONFIG_SCHED_SMT | |
6605 | if (smt_capable()) | |
6606 | err = sysfs_create_file(&cls->kset.kobj, | |
6607 | &attr_sched_smt_power_savings.attr); | |
6608 | #endif | |
6609 | #ifdef CONFIG_SCHED_MC | |
6610 | if (!err && mc_capable()) | |
6611 | err = sysfs_create_file(&cls->kset.kobj, | |
6612 | &attr_sched_mc_power_savings.attr); | |
6613 | #endif | |
6614 | return err; | |
6615 | } | |
5c45bf27 SS |
6616 | #endif |
6617 | ||
1da177e4 LT |
6618 | /* |
6619 | * Force a reinitialization of the sched domains hierarchy. The domains | |
6620 | * and groups cannot be updated in place without racing with the balancing | |
41c7ce9a | 6621 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
6622 | * which will prevent rebalancing while the sched domains are recalculated. |
6623 | */ | |
6624 | static int update_sched_domains(struct notifier_block *nfb, | |
6625 | unsigned long action, void *hcpu) | |
6626 | { | |
1da177e4 LT |
6627 | switch (action) { |
6628 | case CPU_UP_PREPARE: | |
8bb78442 | 6629 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 6630 | case CPU_DOWN_PREPARE: |
8bb78442 | 6631 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 6632 | detach_destroy_domains(&cpu_online_map); |
1da177e4 LT |
6633 | return NOTIFY_OK; |
6634 | ||
6635 | case CPU_UP_CANCELED: | |
8bb78442 | 6636 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 6637 | case CPU_DOWN_FAILED: |
8bb78442 | 6638 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 6639 | case CPU_ONLINE: |
8bb78442 | 6640 | case CPU_ONLINE_FROZEN: |
1da177e4 | 6641 | case CPU_DEAD: |
8bb78442 | 6642 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
6643 | /* |
6644 | * Fall through and re-initialise the domains. | |
6645 | */ | |
6646 | break; | |
6647 | default: | |
6648 | return NOTIFY_DONE; | |
6649 | } | |
6650 | ||
6651 | /* The hotplug lock is already held by cpu_up/cpu_down */ | |
1a20ff27 | 6652 | arch_init_sched_domains(&cpu_online_map); |
1da177e4 LT |
6653 | |
6654 | return NOTIFY_OK; | |
6655 | } | |
1da177e4 LT |
6656 | |
6657 | void __init sched_init_smp(void) | |
6658 | { | |
5c1e1767 NP |
6659 | cpumask_t non_isolated_cpus; |
6660 | ||
5be9361c | 6661 | mutex_lock(&sched_hotcpu_mutex); |
1a20ff27 | 6662 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 6663 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
6664 | if (cpus_empty(non_isolated_cpus)) |
6665 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
5be9361c | 6666 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
6667 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
6668 | hotcpu_notifier(update_sched_domains, 0); | |
5c1e1767 NP |
6669 | |
6670 | /* Move init over to a non-isolated CPU */ | |
6671 | if (set_cpus_allowed(current, non_isolated_cpus) < 0) | |
6672 | BUG(); | |
1da177e4 LT |
6673 | } |
6674 | #else | |
6675 | void __init sched_init_smp(void) | |
6676 | { | |
6677 | } | |
6678 | #endif /* CONFIG_SMP */ | |
6679 | ||
6680 | int in_sched_functions(unsigned long addr) | |
6681 | { | |
6682 | /* Linker adds these: start and end of __sched functions */ | |
6683 | extern char __sched_text_start[], __sched_text_end[]; | |
48f24c4d | 6684 | |
1da177e4 LT |
6685 | return in_lock_functions(addr) || |
6686 | (addr >= (unsigned long)__sched_text_start | |
6687 | && addr < (unsigned long)__sched_text_end); | |
6688 | } | |
6689 | ||
a9957449 | 6690 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
6691 | { |
6692 | cfs_rq->tasks_timeline = RB_ROOT; | |
dd41f596 IM |
6693 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6694 | cfs_rq->rq = rq; | |
6695 | #endif | |
67e9fb2a | 6696 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
6697 | } |
6698 | ||
1da177e4 LT |
6699 | void __init sched_init(void) |
6700 | { | |
476f3534 | 6701 | int highest_cpu = 0; |
dd41f596 IM |
6702 | int i, j; |
6703 | ||
0a945022 | 6704 | for_each_possible_cpu(i) { |
dd41f596 | 6705 | struct rt_prio_array *array; |
70b97a7f | 6706 | struct rq *rq; |
1da177e4 LT |
6707 | |
6708 | rq = cpu_rq(i); | |
6709 | spin_lock_init(&rq->lock); | |
fcb99371 | 6710 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 6711 | rq->nr_running = 0; |
dd41f596 IM |
6712 | rq->clock = 1; |
6713 | init_cfs_rq(&rq->cfs, rq); | |
6714 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6715 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); | |
3a252015 IM |
6716 | { |
6717 | struct cfs_rq *cfs_rq = &per_cpu(init_cfs_rq, i); | |
6718 | struct sched_entity *se = | |
6719 | &per_cpu(init_sched_entity, i); | |
6720 | ||
6721 | init_cfs_rq_p[i] = cfs_rq; | |
6722 | init_cfs_rq(cfs_rq, rq); | |
4cf86d77 | 6723 | cfs_rq->tg = &init_task_group; |
3a252015 | 6724 | list_add(&cfs_rq->leaf_cfs_rq_list, |
29f59db3 SV |
6725 | &rq->leaf_cfs_rq_list); |
6726 | ||
3a252015 IM |
6727 | init_sched_entity_p[i] = se; |
6728 | se->cfs_rq = &rq->cfs; | |
6729 | se->my_q = cfs_rq; | |
4cf86d77 | 6730 | se->load.weight = init_task_group_load; |
9b5b7751 | 6731 | se->load.inv_weight = |
4cf86d77 | 6732 | div64_64(1ULL<<32, init_task_group_load); |
3a252015 IM |
6733 | se->parent = NULL; |
6734 | } | |
4cf86d77 | 6735 | init_task_group.shares = init_task_group_load; |
5cb350ba | 6736 | spin_lock_init(&init_task_group.lock); |
dd41f596 | 6737 | #endif |
1da177e4 | 6738 | |
dd41f596 IM |
6739 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6740 | rq->cpu_load[j] = 0; | |
1da177e4 | 6741 | #ifdef CONFIG_SMP |
41c7ce9a | 6742 | rq->sd = NULL; |
1da177e4 | 6743 | rq->active_balance = 0; |
dd41f596 | 6744 | rq->next_balance = jiffies; |
1da177e4 | 6745 | rq->push_cpu = 0; |
0a2966b4 | 6746 | rq->cpu = i; |
1da177e4 LT |
6747 | rq->migration_thread = NULL; |
6748 | INIT_LIST_HEAD(&rq->migration_queue); | |
6749 | #endif | |
6750 | atomic_set(&rq->nr_iowait, 0); | |
6751 | ||
dd41f596 IM |
6752 | array = &rq->rt.active; |
6753 | for (j = 0; j < MAX_RT_PRIO; j++) { | |
6754 | INIT_LIST_HEAD(array->queue + j); | |
6755 | __clear_bit(j, array->bitmap); | |
1da177e4 | 6756 | } |
476f3534 | 6757 | highest_cpu = i; |
dd41f596 IM |
6758 | /* delimiter for bitsearch: */ |
6759 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
1da177e4 LT |
6760 | } |
6761 | ||
2dd73a4f | 6762 | set_load_weight(&init_task); |
b50f60ce | 6763 | |
e107be36 AK |
6764 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6765 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6766 | #endif | |
6767 | ||
c9819f45 | 6768 | #ifdef CONFIG_SMP |
476f3534 | 6769 | nr_cpu_ids = highest_cpu + 1; |
c9819f45 CL |
6770 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); |
6771 | #endif | |
6772 | ||
b50f60ce HC |
6773 | #ifdef CONFIG_RT_MUTEXES |
6774 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
6775 | #endif | |
6776 | ||
1da177e4 LT |
6777 | /* |
6778 | * The boot idle thread does lazy MMU switching as well: | |
6779 | */ | |
6780 | atomic_inc(&init_mm.mm_count); | |
6781 | enter_lazy_tlb(&init_mm, current); | |
6782 | ||
6783 | /* | |
6784 | * Make us the idle thread. Technically, schedule() should not be | |
6785 | * called from this thread, however somewhere below it might be, | |
6786 | * but because we are the idle thread, we just pick up running again | |
6787 | * when this runqueue becomes "idle". | |
6788 | */ | |
6789 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
6790 | /* |
6791 | * During early bootup we pretend to be a normal task: | |
6792 | */ | |
6793 | current->sched_class = &fair_sched_class; | |
1da177e4 LT |
6794 | } |
6795 | ||
6796 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
6797 | void __might_sleep(char *file, int line) | |
6798 | { | |
48f24c4d | 6799 | #ifdef in_atomic |
1da177e4 LT |
6800 | static unsigned long prev_jiffy; /* ratelimiting */ |
6801 | ||
6802 | if ((in_atomic() || irqs_disabled()) && | |
6803 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
6804 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6805 | return; | |
6806 | prev_jiffy = jiffies; | |
91368d73 | 6807 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
6808 | " context at %s:%d\n", file, line); |
6809 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
6810 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 6811 | debug_show_held_locks(current); |
3117df04 IM |
6812 | if (irqs_disabled()) |
6813 | print_irqtrace_events(current); | |
1da177e4 LT |
6814 | dump_stack(); |
6815 | } | |
6816 | #endif | |
6817 | } | |
6818 | EXPORT_SYMBOL(__might_sleep); | |
6819 | #endif | |
6820 | ||
6821 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
6822 | static void normalize_task(struct rq *rq, struct task_struct *p) |
6823 | { | |
6824 | int on_rq; | |
6825 | update_rq_clock(rq); | |
6826 | on_rq = p->se.on_rq; | |
6827 | if (on_rq) | |
6828 | deactivate_task(rq, p, 0); | |
6829 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
6830 | if (on_rq) { | |
6831 | activate_task(rq, p, 0); | |
6832 | resched_task(rq->curr); | |
6833 | } | |
6834 | } | |
6835 | ||
1da177e4 LT |
6836 | void normalize_rt_tasks(void) |
6837 | { | |
a0f98a1c | 6838 | struct task_struct *g, *p; |
1da177e4 | 6839 | unsigned long flags; |
70b97a7f | 6840 | struct rq *rq; |
1da177e4 LT |
6841 | |
6842 | read_lock_irq(&tasklist_lock); | |
a0f98a1c | 6843 | do_each_thread(g, p) { |
178be793 IM |
6844 | /* |
6845 | * Only normalize user tasks: | |
6846 | */ | |
6847 | if (!p->mm) | |
6848 | continue; | |
6849 | ||
6cfb0d5d | 6850 | p->se.exec_start = 0; |
6cfb0d5d | 6851 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 6852 | p->se.wait_start = 0; |
dd41f596 | 6853 | p->se.sleep_start = 0; |
dd41f596 | 6854 | p->se.block_start = 0; |
6cfb0d5d | 6855 | #endif |
dd41f596 IM |
6856 | task_rq(p)->clock = 0; |
6857 | ||
6858 | if (!rt_task(p)) { | |
6859 | /* | |
6860 | * Renice negative nice level userspace | |
6861 | * tasks back to 0: | |
6862 | */ | |
6863 | if (TASK_NICE(p) < 0 && p->mm) | |
6864 | set_user_nice(p, 0); | |
1da177e4 | 6865 | continue; |
dd41f596 | 6866 | } |
1da177e4 | 6867 | |
b29739f9 IM |
6868 | spin_lock_irqsave(&p->pi_lock, flags); |
6869 | rq = __task_rq_lock(p); | |
1da177e4 | 6870 | |
178be793 | 6871 | normalize_task(rq, p); |
3a5e4dc1 | 6872 | |
b29739f9 IM |
6873 | __task_rq_unlock(rq); |
6874 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
a0f98a1c IM |
6875 | } while_each_thread(g, p); |
6876 | ||
1da177e4 LT |
6877 | read_unlock_irq(&tasklist_lock); |
6878 | } | |
6879 | ||
6880 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
6881 | |
6882 | #ifdef CONFIG_IA64 | |
6883 | /* | |
6884 | * These functions are only useful for the IA64 MCA handling. | |
6885 | * | |
6886 | * They can only be called when the whole system has been | |
6887 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6888 | * activity can take place. Using them for anything else would | |
6889 | * be a serious bug, and as a result, they aren't even visible | |
6890 | * under any other configuration. | |
6891 | */ | |
6892 | ||
6893 | /** | |
6894 | * curr_task - return the current task for a given cpu. | |
6895 | * @cpu: the processor in question. | |
6896 | * | |
6897 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6898 | */ | |
36c8b586 | 6899 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
6900 | { |
6901 | return cpu_curr(cpu); | |
6902 | } | |
6903 | ||
6904 | /** | |
6905 | * set_curr_task - set the current task for a given cpu. | |
6906 | * @cpu: the processor in question. | |
6907 | * @p: the task pointer to set. | |
6908 | * | |
6909 | * Description: This function must only be used when non-maskable interrupts | |
6910 | * are serviced on a separate stack. It allows the architecture to switch the | |
6911 | * notion of the current task on a cpu in a non-blocking manner. This function | |
6912 | * must be called with all CPU's synchronized, and interrupts disabled, the | |
6913 | * and caller must save the original value of the current task (see | |
6914 | * curr_task() above) and restore that value before reenabling interrupts and | |
6915 | * re-starting the system. | |
6916 | * | |
6917 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6918 | */ | |
36c8b586 | 6919 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
6920 | { |
6921 | cpu_curr(cpu) = p; | |
6922 | } | |
6923 | ||
6924 | #endif | |
29f59db3 SV |
6925 | |
6926 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6927 | ||
29f59db3 | 6928 | /* allocate runqueue etc for a new task group */ |
4cf86d77 | 6929 | struct task_group *sched_create_group(void) |
29f59db3 | 6930 | { |
4cf86d77 | 6931 | struct task_group *tg; |
29f59db3 SV |
6932 | struct cfs_rq *cfs_rq; |
6933 | struct sched_entity *se; | |
9b5b7751 | 6934 | struct rq *rq; |
29f59db3 SV |
6935 | int i; |
6936 | ||
29f59db3 SV |
6937 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); |
6938 | if (!tg) | |
6939 | return ERR_PTR(-ENOMEM); | |
6940 | ||
9b5b7751 | 6941 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
6942 | if (!tg->cfs_rq) |
6943 | goto err; | |
9b5b7751 | 6944 | tg->se = kzalloc(sizeof(se) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
6945 | if (!tg->se) |
6946 | goto err; | |
6947 | ||
6948 | for_each_possible_cpu(i) { | |
9b5b7751 | 6949 | rq = cpu_rq(i); |
29f59db3 SV |
6950 | |
6951 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), GFP_KERNEL, | |
6952 | cpu_to_node(i)); | |
6953 | if (!cfs_rq) | |
6954 | goto err; | |
6955 | ||
6956 | se = kmalloc_node(sizeof(struct sched_entity), GFP_KERNEL, | |
6957 | cpu_to_node(i)); | |
6958 | if (!se) | |
6959 | goto err; | |
6960 | ||
6961 | memset(cfs_rq, 0, sizeof(struct cfs_rq)); | |
6962 | memset(se, 0, sizeof(struct sched_entity)); | |
6963 | ||
6964 | tg->cfs_rq[i] = cfs_rq; | |
6965 | init_cfs_rq(cfs_rq, rq); | |
6966 | cfs_rq->tg = tg; | |
29f59db3 SV |
6967 | |
6968 | tg->se[i] = se; | |
6969 | se->cfs_rq = &rq->cfs; | |
6970 | se->my_q = cfs_rq; | |
6971 | se->load.weight = NICE_0_LOAD; | |
6972 | se->load.inv_weight = div64_64(1ULL<<32, NICE_0_LOAD); | |
6973 | se->parent = NULL; | |
6974 | } | |
6975 | ||
9b5b7751 SV |
6976 | for_each_possible_cpu(i) { |
6977 | rq = cpu_rq(i); | |
6978 | cfs_rq = tg->cfs_rq[i]; | |
6979 | list_add_rcu(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
6980 | } | |
29f59db3 | 6981 | |
9b5b7751 | 6982 | tg->shares = NICE_0_LOAD; |
5cb350ba | 6983 | spin_lock_init(&tg->lock); |
29f59db3 | 6984 | |
9b5b7751 | 6985 | return tg; |
29f59db3 SV |
6986 | |
6987 | err: | |
6988 | for_each_possible_cpu(i) { | |
a65914b3 | 6989 | if (tg->cfs_rq) |
29f59db3 | 6990 | kfree(tg->cfs_rq[i]); |
a65914b3 | 6991 | if (tg->se) |
29f59db3 SV |
6992 | kfree(tg->se[i]); |
6993 | } | |
a65914b3 IM |
6994 | kfree(tg->cfs_rq); |
6995 | kfree(tg->se); | |
6996 | kfree(tg); | |
29f59db3 SV |
6997 | |
6998 | return ERR_PTR(-ENOMEM); | |
6999 | } | |
7000 | ||
9b5b7751 SV |
7001 | /* rcu callback to free various structures associated with a task group */ |
7002 | static void free_sched_group(struct rcu_head *rhp) | |
29f59db3 | 7003 | { |
9b5b7751 | 7004 | struct cfs_rq *cfs_rq = container_of(rhp, struct cfs_rq, rcu); |
4cf86d77 | 7005 | struct task_group *tg = cfs_rq->tg; |
29f59db3 SV |
7006 | struct sched_entity *se; |
7007 | int i; | |
7008 | ||
29f59db3 SV |
7009 | /* now it should be safe to free those cfs_rqs */ |
7010 | for_each_possible_cpu(i) { | |
7011 | cfs_rq = tg->cfs_rq[i]; | |
7012 | kfree(cfs_rq); | |
7013 | ||
7014 | se = tg->se[i]; | |
7015 | kfree(se); | |
7016 | } | |
7017 | ||
7018 | kfree(tg->cfs_rq); | |
7019 | kfree(tg->se); | |
7020 | kfree(tg); | |
7021 | } | |
7022 | ||
9b5b7751 | 7023 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7024 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 7025 | { |
9b5b7751 SV |
7026 | struct cfs_rq *cfs_rq; |
7027 | int i; | |
29f59db3 | 7028 | |
9b5b7751 SV |
7029 | for_each_possible_cpu(i) { |
7030 | cfs_rq = tg->cfs_rq[i]; | |
7031 | list_del_rcu(&cfs_rq->leaf_cfs_rq_list); | |
7032 | } | |
7033 | ||
7034 | cfs_rq = tg->cfs_rq[0]; | |
7035 | ||
7036 | /* wait for possible concurrent references to cfs_rqs complete */ | |
7037 | call_rcu(&cfs_rq->rcu, free_sched_group); | |
29f59db3 SV |
7038 | } |
7039 | ||
9b5b7751 | 7040 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7041 | * The caller of this function should have put the task in its new group |
7042 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7043 | * reflect its new group. | |
9b5b7751 SV |
7044 | */ |
7045 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
7046 | { |
7047 | int on_rq, running; | |
7048 | unsigned long flags; | |
7049 | struct rq *rq; | |
7050 | ||
7051 | rq = task_rq_lock(tsk, &flags); | |
7052 | ||
7053 | if (tsk->sched_class != &fair_sched_class) | |
7054 | goto done; | |
7055 | ||
7056 | update_rq_clock(rq); | |
7057 | ||
7058 | running = task_running(rq, tsk); | |
7059 | on_rq = tsk->se.on_rq; | |
7060 | ||
83b699ed | 7061 | if (on_rq) { |
29f59db3 | 7062 | dequeue_task(rq, tsk, 0); |
83b699ed SV |
7063 | if (unlikely(running)) |
7064 | tsk->sched_class->put_prev_task(rq, tsk); | |
7065 | } | |
29f59db3 SV |
7066 | |
7067 | set_task_cfs_rq(tsk); | |
7068 | ||
83b699ed SV |
7069 | if (on_rq) { |
7070 | if (unlikely(running)) | |
7071 | tsk->sched_class->set_curr_task(rq); | |
7074badb | 7072 | enqueue_task(rq, tsk, 0); |
83b699ed | 7073 | } |
29f59db3 SV |
7074 | |
7075 | done: | |
7076 | task_rq_unlock(rq, &flags); | |
7077 | } | |
7078 | ||
7079 | static void set_se_shares(struct sched_entity *se, unsigned long shares) | |
7080 | { | |
7081 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
7082 | struct rq *rq = cfs_rq->rq; | |
7083 | int on_rq; | |
7084 | ||
7085 | spin_lock_irq(&rq->lock); | |
7086 | ||
7087 | on_rq = se->on_rq; | |
7088 | if (on_rq) | |
7089 | dequeue_entity(cfs_rq, se, 0); | |
7090 | ||
7091 | se->load.weight = shares; | |
7092 | se->load.inv_weight = div64_64((1ULL<<32), shares); | |
7093 | ||
7094 | if (on_rq) | |
7095 | enqueue_entity(cfs_rq, se, 0); | |
7096 | ||
7097 | spin_unlock_irq(&rq->lock); | |
7098 | } | |
7099 | ||
4cf86d77 | 7100 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
7101 | { |
7102 | int i; | |
29f59db3 | 7103 | |
5cb350ba | 7104 | spin_lock(&tg->lock); |
9b5b7751 | 7105 | if (tg->shares == shares) |
5cb350ba | 7106 | goto done; |
29f59db3 | 7107 | |
9b5b7751 | 7108 | tg->shares = shares; |
29f59db3 | 7109 | for_each_possible_cpu(i) |
9b5b7751 | 7110 | set_se_shares(tg->se[i], shares); |
29f59db3 | 7111 | |
5cb350ba DG |
7112 | done: |
7113 | spin_unlock(&tg->lock); | |
9b5b7751 | 7114 | return 0; |
29f59db3 SV |
7115 | } |
7116 | ||
5cb350ba DG |
7117 | unsigned long sched_group_shares(struct task_group *tg) |
7118 | { | |
7119 | return tg->shares; | |
7120 | } | |
7121 | ||
3a252015 | 7122 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e SV |
7123 | |
7124 | #ifdef CONFIG_FAIR_CGROUP_SCHED | |
7125 | ||
7126 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 7127 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 7128 | { |
2b01dfe3 PM |
7129 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
7130 | struct task_group, css); | |
68318b8e SV |
7131 | } |
7132 | ||
7133 | static struct cgroup_subsys_state * | |
2b01dfe3 | 7134 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e SV |
7135 | { |
7136 | struct task_group *tg; | |
7137 | ||
2b01dfe3 | 7138 | if (!cgrp->parent) { |
68318b8e | 7139 | /* This is early initialization for the top cgroup */ |
2b01dfe3 | 7140 | init_task_group.css.cgroup = cgrp; |
68318b8e SV |
7141 | return &init_task_group.css; |
7142 | } | |
7143 | ||
7144 | /* we support only 1-level deep hierarchical scheduler atm */ | |
2b01dfe3 | 7145 | if (cgrp->parent->parent) |
68318b8e SV |
7146 | return ERR_PTR(-EINVAL); |
7147 | ||
7148 | tg = sched_create_group(); | |
7149 | if (IS_ERR(tg)) | |
7150 | return ERR_PTR(-ENOMEM); | |
7151 | ||
7152 | /* Bind the cgroup to task_group object we just created */ | |
2b01dfe3 | 7153 | tg->css.cgroup = cgrp; |
68318b8e SV |
7154 | |
7155 | return &tg->css; | |
7156 | } | |
7157 | ||
7158 | static void cpu_cgroup_destroy(struct cgroup_subsys *ss, | |
2b01dfe3 | 7159 | struct cgroup *cgrp) |
68318b8e | 7160 | { |
2b01dfe3 | 7161 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7162 | |
7163 | sched_destroy_group(tg); | |
7164 | } | |
7165 | ||
7166 | static int cpu_cgroup_can_attach(struct cgroup_subsys *ss, | |
2b01dfe3 | 7167 | struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e SV |
7168 | { |
7169 | /* We don't support RT-tasks being in separate groups */ | |
7170 | if (tsk->sched_class != &fair_sched_class) | |
7171 | return -EINVAL; | |
7172 | ||
7173 | return 0; | |
7174 | } | |
7175 | ||
7176 | static void | |
2b01dfe3 | 7177 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
7178 | struct cgroup *old_cont, struct task_struct *tsk) |
7179 | { | |
7180 | sched_move_task(tsk); | |
7181 | } | |
7182 | ||
2b01dfe3 PM |
7183 | static int cpu_shares_write_uint(struct cgroup *cgrp, struct cftype *cftype, |
7184 | u64 shareval) | |
68318b8e | 7185 | { |
2b01dfe3 | 7186 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
7187 | } |
7188 | ||
2b01dfe3 | 7189 | static u64 cpu_shares_read_uint(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 7190 | { |
2b01dfe3 | 7191 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7192 | |
7193 | return (u64) tg->shares; | |
7194 | } | |
7195 | ||
7196 | static struct cftype cpu_shares = { | |
7197 | .name = "shares", | |
7198 | .read_uint = cpu_shares_read_uint, | |
2b01dfe3 | 7199 | .write_uint = cpu_shares_write_uint, |
68318b8e SV |
7200 | }; |
7201 | ||
7202 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
7203 | { | |
7204 | return cgroup_add_file(cont, ss, &cpu_shares); | |
7205 | } | |
7206 | ||
7207 | struct cgroup_subsys cpu_cgroup_subsys = { | |
7208 | .name = "cpu", | |
7209 | .create = cpu_cgroup_create, | |
7210 | .destroy = cpu_cgroup_destroy, | |
7211 | .can_attach = cpu_cgroup_can_attach, | |
7212 | .attach = cpu_cgroup_attach, | |
7213 | .populate = cpu_cgroup_populate, | |
7214 | .subsys_id = cpu_cgroup_subsys_id, | |
7215 | .early_init = 1, | |
7216 | }; | |
7217 | ||
7218 | #endif /* CONFIG_FAIR_CGROUP_SCHED */ |