]>
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> |
1da177e4 | 69 | |
b035b6de AD |
70 | /* |
71 | * Scheduler clock - returns current time in nanosec units. | |
72 | * This is default implementation. | |
73 | * Architectures and sub-architectures can override this. | |
74 | */ | |
75 | unsigned long long __attribute__((weak)) sched_clock(void) | |
76 | { | |
77 | return (unsigned long long)jiffies * (1000000000 / HZ); | |
78 | } | |
79 | ||
1da177e4 LT |
80 | /* |
81 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
82 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
83 | * and back. | |
84 | */ | |
85 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
86 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
87 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
88 | ||
89 | /* | |
90 | * 'User priority' is the nice value converted to something we | |
91 | * can work with better when scaling various scheduler parameters, | |
92 | * it's a [ 0 ... 39 ] range. | |
93 | */ | |
94 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
95 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
96 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
97 | ||
98 | /* | |
99 | * Some helpers for converting nanosecond timing to jiffy resolution | |
100 | */ | |
a4ec24b4 | 101 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (1000000000 / HZ)) |
1da177e4 LT |
102 | #define JIFFIES_TO_NS(TIME) ((TIME) * (1000000000 / HZ)) |
103 | ||
6aa645ea IM |
104 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
105 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
106 | ||
1da177e4 LT |
107 | /* |
108 | * These are the 'tuning knobs' of the scheduler: | |
109 | * | |
a4ec24b4 | 110 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
111 | * Timeslices get refilled after they expire. |
112 | */ | |
1da177e4 | 113 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 114 | |
5517d86b ED |
115 | #ifdef CONFIG_SMP |
116 | /* | |
117 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
118 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
119 | */ | |
120 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
121 | { | |
122 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
123 | } | |
124 | ||
125 | /* | |
126 | * Each time a sched group cpu_power is changed, | |
127 | * we must compute its reciprocal value | |
128 | */ | |
129 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
130 | { | |
131 | sg->__cpu_power += val; | |
132 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
133 | } | |
134 | #endif | |
135 | ||
e05606d3 IM |
136 | static inline int rt_policy(int policy) |
137 | { | |
138 | if (unlikely(policy == SCHED_FIFO) || unlikely(policy == SCHED_RR)) | |
139 | return 1; | |
140 | return 0; | |
141 | } | |
142 | ||
143 | static inline int task_has_rt_policy(struct task_struct *p) | |
144 | { | |
145 | return rt_policy(p->policy); | |
146 | } | |
147 | ||
1da177e4 | 148 | /* |
6aa645ea | 149 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 150 | */ |
6aa645ea IM |
151 | struct rt_prio_array { |
152 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
153 | struct list_head queue[MAX_RT_PRIO]; | |
154 | }; | |
155 | ||
29f59db3 SV |
156 | #ifdef CONFIG_FAIR_GROUP_SCHED |
157 | ||
68318b8e SV |
158 | #include <linux/cgroup.h> |
159 | ||
29f59db3 SV |
160 | struct cfs_rq; |
161 | ||
162 | /* task group related information */ | |
4cf86d77 | 163 | struct task_group { |
68318b8e SV |
164 | #ifdef CONFIG_FAIR_CGROUP_SCHED |
165 | struct cgroup_subsys_state css; | |
166 | #endif | |
29f59db3 SV |
167 | /* schedulable entities of this group on each cpu */ |
168 | struct sched_entity **se; | |
169 | /* runqueue "owned" by this group on each cpu */ | |
170 | struct cfs_rq **cfs_rq; | |
171 | unsigned long shares; | |
5cb350ba DG |
172 | /* spinlock to serialize modification to shares */ |
173 | spinlock_t lock; | |
29f59db3 SV |
174 | }; |
175 | ||
176 | /* Default task group's sched entity on each cpu */ | |
177 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
178 | /* Default task group's cfs_rq on each cpu */ | |
179 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
180 | ||
9b5b7751 SV |
181 | static struct sched_entity *init_sched_entity_p[NR_CPUS]; |
182 | static struct cfs_rq *init_cfs_rq_p[NR_CPUS]; | |
29f59db3 SV |
183 | |
184 | /* Default task group. | |
3a252015 | 185 | * Every task in system belong to this group at bootup. |
29f59db3 | 186 | */ |
4cf86d77 | 187 | struct task_group init_task_group = { |
3a252015 IM |
188 | .se = init_sched_entity_p, |
189 | .cfs_rq = init_cfs_rq_p, | |
190 | }; | |
9b5b7751 | 191 | |
24e377a8 | 192 | #ifdef CONFIG_FAIR_USER_SCHED |
3a252015 | 193 | # define INIT_TASK_GRP_LOAD 2*NICE_0_LOAD |
24e377a8 | 194 | #else |
3a252015 | 195 | # define INIT_TASK_GRP_LOAD NICE_0_LOAD |
24e377a8 SV |
196 | #endif |
197 | ||
4cf86d77 | 198 | static int init_task_group_load = INIT_TASK_GRP_LOAD; |
29f59db3 SV |
199 | |
200 | /* return group to which a task belongs */ | |
4cf86d77 | 201 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 202 | { |
4cf86d77 | 203 | struct task_group *tg; |
9b5b7751 | 204 | |
24e377a8 SV |
205 | #ifdef CONFIG_FAIR_USER_SCHED |
206 | tg = p->user->tg; | |
68318b8e SV |
207 | #elif defined(CONFIG_FAIR_CGROUP_SCHED) |
208 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), | |
209 | struct task_group, css); | |
24e377a8 | 210 | #else |
4cf86d77 | 211 | tg = &init_task_group; |
24e377a8 | 212 | #endif |
9b5b7751 SV |
213 | |
214 | return tg; | |
29f59db3 SV |
215 | } |
216 | ||
217 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
218 | static inline void set_task_cfs_rq(struct task_struct *p) | |
219 | { | |
4cf86d77 IM |
220 | p->se.cfs_rq = task_group(p)->cfs_rq[task_cpu(p)]; |
221 | p->se.parent = task_group(p)->se[task_cpu(p)]; | |
29f59db3 SV |
222 | } |
223 | ||
224 | #else | |
225 | ||
226 | static inline void set_task_cfs_rq(struct task_struct *p) { } | |
227 | ||
228 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
229 | ||
6aa645ea IM |
230 | /* CFS-related fields in a runqueue */ |
231 | struct cfs_rq { | |
232 | struct load_weight load; | |
233 | unsigned long nr_running; | |
234 | ||
6aa645ea | 235 | u64 exec_clock; |
e9acbff6 | 236 | u64 min_vruntime; |
6aa645ea IM |
237 | |
238 | struct rb_root tasks_timeline; | |
239 | struct rb_node *rb_leftmost; | |
240 | struct rb_node *rb_load_balance_curr; | |
6aa645ea IM |
241 | /* 'curr' points to currently running entity on this cfs_rq. |
242 | * It is set to NULL otherwise (i.e when none are currently running). | |
243 | */ | |
244 | struct sched_entity *curr; | |
ddc97297 PZ |
245 | |
246 | unsigned long nr_spread_over; | |
247 | ||
62160e3f | 248 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
249 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
250 | ||
251 | /* leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
252 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities | |
253 | * (like users, containers etc.) | |
254 | * | |
255 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
256 | * list is used during load balance. | |
257 | */ | |
258 | struct list_head leaf_cfs_rq_list; /* Better name : task_cfs_rq_list? */ | |
4cf86d77 | 259 | struct task_group *tg; /* group that "owns" this runqueue */ |
9b5b7751 | 260 | struct rcu_head rcu; |
6aa645ea IM |
261 | #endif |
262 | }; | |
1da177e4 | 263 | |
6aa645ea IM |
264 | /* Real-Time classes' related field in a runqueue: */ |
265 | struct rt_rq { | |
266 | struct rt_prio_array active; | |
267 | int rt_load_balance_idx; | |
268 | struct list_head *rt_load_balance_head, *rt_load_balance_curr; | |
269 | }; | |
270 | ||
1da177e4 LT |
271 | /* |
272 | * This is the main, per-CPU runqueue data structure. | |
273 | * | |
274 | * Locking rule: those places that want to lock multiple runqueues | |
275 | * (such as the load balancing or the thread migration code), lock | |
276 | * acquire operations must be ordered by ascending &runqueue. | |
277 | */ | |
70b97a7f | 278 | struct rq { |
d8016491 IM |
279 | /* runqueue lock: */ |
280 | spinlock_t lock; | |
1da177e4 LT |
281 | |
282 | /* | |
283 | * nr_running and cpu_load should be in the same cacheline because | |
284 | * remote CPUs use both these fields when doing load calculation. | |
285 | */ | |
286 | unsigned long nr_running; | |
6aa645ea IM |
287 | #define CPU_LOAD_IDX_MAX 5 |
288 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 289 | unsigned char idle_at_tick; |
46cb4b7c SS |
290 | #ifdef CONFIG_NO_HZ |
291 | unsigned char in_nohz_recently; | |
292 | #endif | |
d8016491 IM |
293 | /* capture load from *all* tasks on this cpu: */ |
294 | struct load_weight load; | |
6aa645ea IM |
295 | unsigned long nr_load_updates; |
296 | u64 nr_switches; | |
297 | ||
298 | struct cfs_rq cfs; | |
299 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
d8016491 IM |
300 | /* list of leaf cfs_rq on this cpu: */ |
301 | struct list_head leaf_cfs_rq_list; | |
1da177e4 | 302 | #endif |
6aa645ea | 303 | struct rt_rq rt; |
1da177e4 LT |
304 | |
305 | /* | |
306 | * This is part of a global counter where only the total sum | |
307 | * over all CPUs matters. A task can increase this counter on | |
308 | * one CPU and if it got migrated afterwards it may decrease | |
309 | * it on another CPU. Always updated under the runqueue lock: | |
310 | */ | |
311 | unsigned long nr_uninterruptible; | |
312 | ||
36c8b586 | 313 | struct task_struct *curr, *idle; |
c9819f45 | 314 | unsigned long next_balance; |
1da177e4 | 315 | struct mm_struct *prev_mm; |
6aa645ea | 316 | |
6aa645ea IM |
317 | u64 clock, prev_clock_raw; |
318 | s64 clock_max_delta; | |
319 | ||
320 | unsigned int clock_warps, clock_overflows; | |
2aa44d05 IM |
321 | u64 idle_clock; |
322 | unsigned int clock_deep_idle_events; | |
529c7726 | 323 | u64 tick_timestamp; |
6aa645ea | 324 | |
1da177e4 LT |
325 | atomic_t nr_iowait; |
326 | ||
327 | #ifdef CONFIG_SMP | |
328 | struct sched_domain *sd; | |
329 | ||
330 | /* For active balancing */ | |
331 | int active_balance; | |
332 | int push_cpu; | |
d8016491 IM |
333 | /* cpu of this runqueue: */ |
334 | int cpu; | |
1da177e4 | 335 | |
36c8b586 | 336 | struct task_struct *migration_thread; |
1da177e4 LT |
337 | struct list_head migration_queue; |
338 | #endif | |
339 | ||
340 | #ifdef CONFIG_SCHEDSTATS | |
341 | /* latency stats */ | |
342 | struct sched_info rq_sched_info; | |
343 | ||
344 | /* sys_sched_yield() stats */ | |
480b9434 KC |
345 | unsigned int yld_exp_empty; |
346 | unsigned int yld_act_empty; | |
347 | unsigned int yld_both_empty; | |
348 | unsigned int yld_count; | |
1da177e4 LT |
349 | |
350 | /* schedule() stats */ | |
480b9434 KC |
351 | unsigned int sched_switch; |
352 | unsigned int sched_count; | |
353 | unsigned int sched_goidle; | |
1da177e4 LT |
354 | |
355 | /* try_to_wake_up() stats */ | |
480b9434 KC |
356 | unsigned int ttwu_count; |
357 | unsigned int ttwu_local; | |
b8efb561 IM |
358 | |
359 | /* BKL stats */ | |
480b9434 | 360 | unsigned int bkl_count; |
1da177e4 | 361 | #endif |
fcb99371 | 362 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
363 | }; |
364 | ||
f34e3b61 | 365 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
5be9361c | 366 | static DEFINE_MUTEX(sched_hotcpu_mutex); |
1da177e4 | 367 | |
dd41f596 IM |
368 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
369 | { | |
370 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
371 | } | |
372 | ||
0a2966b4 CL |
373 | static inline int cpu_of(struct rq *rq) |
374 | { | |
375 | #ifdef CONFIG_SMP | |
376 | return rq->cpu; | |
377 | #else | |
378 | return 0; | |
379 | #endif | |
380 | } | |
381 | ||
20d315d4 | 382 | /* |
b04a0f4c IM |
383 | * Update the per-runqueue clock, as finegrained as the platform can give |
384 | * us, but without assuming monotonicity, etc.: | |
20d315d4 | 385 | */ |
b04a0f4c | 386 | static void __update_rq_clock(struct rq *rq) |
20d315d4 IM |
387 | { |
388 | u64 prev_raw = rq->prev_clock_raw; | |
389 | u64 now = sched_clock(); | |
390 | s64 delta = now - prev_raw; | |
391 | u64 clock = rq->clock; | |
392 | ||
b04a0f4c IM |
393 | #ifdef CONFIG_SCHED_DEBUG |
394 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
395 | #endif | |
20d315d4 IM |
396 | /* |
397 | * Protect against sched_clock() occasionally going backwards: | |
398 | */ | |
399 | if (unlikely(delta < 0)) { | |
400 | clock++; | |
401 | rq->clock_warps++; | |
402 | } else { | |
403 | /* | |
404 | * Catch too large forward jumps too: | |
405 | */ | |
529c7726 IM |
406 | if (unlikely(clock + delta > rq->tick_timestamp + TICK_NSEC)) { |
407 | if (clock < rq->tick_timestamp + TICK_NSEC) | |
408 | clock = rq->tick_timestamp + TICK_NSEC; | |
409 | else | |
410 | clock++; | |
20d315d4 IM |
411 | rq->clock_overflows++; |
412 | } else { | |
413 | if (unlikely(delta > rq->clock_max_delta)) | |
414 | rq->clock_max_delta = delta; | |
415 | clock += delta; | |
416 | } | |
417 | } | |
418 | ||
419 | rq->prev_clock_raw = now; | |
420 | rq->clock = clock; | |
b04a0f4c | 421 | } |
20d315d4 | 422 | |
b04a0f4c IM |
423 | static void update_rq_clock(struct rq *rq) |
424 | { | |
425 | if (likely(smp_processor_id() == cpu_of(rq))) | |
426 | __update_rq_clock(rq); | |
20d315d4 IM |
427 | } |
428 | ||
674311d5 NP |
429 | /* |
430 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 431 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
432 | * |
433 | * The domain tree of any CPU may only be accessed from within | |
434 | * preempt-disabled sections. | |
435 | */ | |
48f24c4d IM |
436 | #define for_each_domain(cpu, __sd) \ |
437 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
438 | |
439 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
440 | #define this_rq() (&__get_cpu_var(runqueues)) | |
441 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
442 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
443 | ||
bf5c91ba IM |
444 | /* |
445 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
446 | */ | |
447 | #ifdef CONFIG_SCHED_DEBUG | |
448 | # define const_debug __read_mostly | |
449 | #else | |
450 | # define const_debug static const | |
451 | #endif | |
452 | ||
453 | /* | |
454 | * Debugging: various feature bits | |
455 | */ | |
456 | enum { | |
bbdba7c0 IM |
457 | SCHED_FEAT_NEW_FAIR_SLEEPERS = 1, |
458 | SCHED_FEAT_START_DEBIT = 2, | |
06877c33 | 459 | SCHED_FEAT_TREE_AVG = 4, |
bbdba7c0 | 460 | SCHED_FEAT_APPROX_AVG = 8, |
ce6c1311 | 461 | SCHED_FEAT_WAKEUP_PREEMPT = 16, |
95938a35 | 462 | SCHED_FEAT_PREEMPT_RESTRICT = 32, |
bf5c91ba IM |
463 | }; |
464 | ||
465 | const_debug unsigned int sysctl_sched_features = | |
8401f775 IM |
466 | SCHED_FEAT_NEW_FAIR_SLEEPERS * 1 | |
467 | SCHED_FEAT_START_DEBIT * 1 | | |
468 | SCHED_FEAT_TREE_AVG * 0 | | |
469 | SCHED_FEAT_APPROX_AVG * 0 | | |
470 | SCHED_FEAT_WAKEUP_PREEMPT * 1 | | |
471 | SCHED_FEAT_PREEMPT_RESTRICT * 1; | |
bf5c91ba IM |
472 | |
473 | #define sched_feat(x) (sysctl_sched_features & SCHED_FEAT_##x) | |
474 | ||
e436d800 IM |
475 | /* |
476 | * For kernel-internal use: high-speed (but slightly incorrect) per-cpu | |
477 | * clock constructed from sched_clock(): | |
478 | */ | |
479 | unsigned long long cpu_clock(int cpu) | |
480 | { | |
e436d800 IM |
481 | unsigned long long now; |
482 | unsigned long flags; | |
b04a0f4c | 483 | struct rq *rq; |
e436d800 | 484 | |
2cd4d0ea | 485 | local_irq_save(flags); |
b04a0f4c IM |
486 | rq = cpu_rq(cpu); |
487 | update_rq_clock(rq); | |
488 | now = rq->clock; | |
2cd4d0ea | 489 | local_irq_restore(flags); |
e436d800 IM |
490 | |
491 | return now; | |
492 | } | |
a58f6f25 | 493 | EXPORT_SYMBOL_GPL(cpu_clock); |
e436d800 | 494 | |
1da177e4 | 495 | #ifndef prepare_arch_switch |
4866cde0 NP |
496 | # define prepare_arch_switch(next) do { } while (0) |
497 | #endif | |
498 | #ifndef finish_arch_switch | |
499 | # define finish_arch_switch(prev) do { } while (0) | |
500 | #endif | |
501 | ||
502 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
70b97a7f | 503 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
504 | { |
505 | return rq->curr == p; | |
506 | } | |
507 | ||
70b97a7f | 508 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
509 | { |
510 | } | |
511 | ||
70b97a7f | 512 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 513 | { |
da04c035 IM |
514 | #ifdef CONFIG_DEBUG_SPINLOCK |
515 | /* this is a valid case when another task releases the spinlock */ | |
516 | rq->lock.owner = current; | |
517 | #endif | |
8a25d5de IM |
518 | /* |
519 | * If we are tracking spinlock dependencies then we have to | |
520 | * fix up the runqueue lock - which gets 'carried over' from | |
521 | * prev into current: | |
522 | */ | |
523 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
524 | ||
4866cde0 NP |
525 | spin_unlock_irq(&rq->lock); |
526 | } | |
527 | ||
528 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 529 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
530 | { |
531 | #ifdef CONFIG_SMP | |
532 | return p->oncpu; | |
533 | #else | |
534 | return rq->curr == p; | |
535 | #endif | |
536 | } | |
537 | ||
70b97a7f | 538 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
539 | { |
540 | #ifdef CONFIG_SMP | |
541 | /* | |
542 | * We can optimise this out completely for !SMP, because the | |
543 | * SMP rebalancing from interrupt is the only thing that cares | |
544 | * here. | |
545 | */ | |
546 | next->oncpu = 1; | |
547 | #endif | |
548 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
549 | spin_unlock_irq(&rq->lock); | |
550 | #else | |
551 | spin_unlock(&rq->lock); | |
552 | #endif | |
553 | } | |
554 | ||
70b97a7f | 555 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
556 | { |
557 | #ifdef CONFIG_SMP | |
558 | /* | |
559 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
560 | * We must ensure this doesn't happen until the switch is completely | |
561 | * finished. | |
562 | */ | |
563 | smp_wmb(); | |
564 | prev->oncpu = 0; | |
565 | #endif | |
566 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
567 | local_irq_enable(); | |
1da177e4 | 568 | #endif |
4866cde0 NP |
569 | } |
570 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 571 | |
b29739f9 IM |
572 | /* |
573 | * __task_rq_lock - lock the runqueue a given task resides on. | |
574 | * Must be called interrupts disabled. | |
575 | */ | |
70b97a7f | 576 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
577 | __acquires(rq->lock) |
578 | { | |
3a5c359a AK |
579 | for (;;) { |
580 | struct rq *rq = task_rq(p); | |
581 | spin_lock(&rq->lock); | |
582 | if (likely(rq == task_rq(p))) | |
583 | return rq; | |
b29739f9 | 584 | spin_unlock(&rq->lock); |
b29739f9 | 585 | } |
b29739f9 IM |
586 | } |
587 | ||
1da177e4 LT |
588 | /* |
589 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
590 | * interrupts. Note the ordering: we can safely lookup the task_rq without | |
591 | * explicitly disabling preemption. | |
592 | */ | |
70b97a7f | 593 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
594 | __acquires(rq->lock) |
595 | { | |
70b97a7f | 596 | struct rq *rq; |
1da177e4 | 597 | |
3a5c359a AK |
598 | for (;;) { |
599 | local_irq_save(*flags); | |
600 | rq = task_rq(p); | |
601 | spin_lock(&rq->lock); | |
602 | if (likely(rq == task_rq(p))) | |
603 | return rq; | |
1da177e4 | 604 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 605 | } |
1da177e4 LT |
606 | } |
607 | ||
a9957449 | 608 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
609 | __releases(rq->lock) |
610 | { | |
611 | spin_unlock(&rq->lock); | |
612 | } | |
613 | ||
70b97a7f | 614 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
615 | __releases(rq->lock) |
616 | { | |
617 | spin_unlock_irqrestore(&rq->lock, *flags); | |
618 | } | |
619 | ||
1da177e4 | 620 | /* |
cc2a73b5 | 621 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 622 | */ |
a9957449 | 623 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
624 | __acquires(rq->lock) |
625 | { | |
70b97a7f | 626 | struct rq *rq; |
1da177e4 LT |
627 | |
628 | local_irq_disable(); | |
629 | rq = this_rq(); | |
630 | spin_lock(&rq->lock); | |
631 | ||
632 | return rq; | |
633 | } | |
634 | ||
1b9f19c2 | 635 | /* |
2aa44d05 | 636 | * We are going deep-idle (irqs are disabled): |
1b9f19c2 | 637 | */ |
2aa44d05 | 638 | void sched_clock_idle_sleep_event(void) |
1b9f19c2 | 639 | { |
2aa44d05 IM |
640 | struct rq *rq = cpu_rq(smp_processor_id()); |
641 | ||
642 | spin_lock(&rq->lock); | |
643 | __update_rq_clock(rq); | |
644 | spin_unlock(&rq->lock); | |
645 | rq->clock_deep_idle_events++; | |
646 | } | |
647 | EXPORT_SYMBOL_GPL(sched_clock_idle_sleep_event); | |
648 | ||
649 | /* | |
650 | * We just idled delta nanoseconds (called with irqs disabled): | |
651 | */ | |
652 | void sched_clock_idle_wakeup_event(u64 delta_ns) | |
653 | { | |
654 | struct rq *rq = cpu_rq(smp_processor_id()); | |
655 | u64 now = sched_clock(); | |
1b9f19c2 | 656 | |
2aa44d05 IM |
657 | rq->idle_clock += delta_ns; |
658 | /* | |
659 | * Override the previous timestamp and ignore all | |
660 | * sched_clock() deltas that occured while we idled, | |
661 | * and use the PM-provided delta_ns to advance the | |
662 | * rq clock: | |
663 | */ | |
664 | spin_lock(&rq->lock); | |
665 | rq->prev_clock_raw = now; | |
666 | rq->clock += delta_ns; | |
667 | spin_unlock(&rq->lock); | |
1b9f19c2 | 668 | } |
2aa44d05 | 669 | EXPORT_SYMBOL_GPL(sched_clock_idle_wakeup_event); |
1b9f19c2 | 670 | |
c24d20db IM |
671 | /* |
672 | * resched_task - mark a task 'to be rescheduled now'. | |
673 | * | |
674 | * On UP this means the setting of the need_resched flag, on SMP it | |
675 | * might also involve a cross-CPU call to trigger the scheduler on | |
676 | * the target CPU. | |
677 | */ | |
678 | #ifdef CONFIG_SMP | |
679 | ||
680 | #ifndef tsk_is_polling | |
681 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
682 | #endif | |
683 | ||
684 | static void resched_task(struct task_struct *p) | |
685 | { | |
686 | int cpu; | |
687 | ||
688 | assert_spin_locked(&task_rq(p)->lock); | |
689 | ||
690 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) | |
691 | return; | |
692 | ||
693 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); | |
694 | ||
695 | cpu = task_cpu(p); | |
696 | if (cpu == smp_processor_id()) | |
697 | return; | |
698 | ||
699 | /* NEED_RESCHED must be visible before we test polling */ | |
700 | smp_mb(); | |
701 | if (!tsk_is_polling(p)) | |
702 | smp_send_reschedule(cpu); | |
703 | } | |
704 | ||
705 | static void resched_cpu(int cpu) | |
706 | { | |
707 | struct rq *rq = cpu_rq(cpu); | |
708 | unsigned long flags; | |
709 | ||
710 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
711 | return; | |
712 | resched_task(cpu_curr(cpu)); | |
713 | spin_unlock_irqrestore(&rq->lock, flags); | |
714 | } | |
715 | #else | |
716 | static inline void resched_task(struct task_struct *p) | |
717 | { | |
718 | assert_spin_locked(&task_rq(p)->lock); | |
719 | set_tsk_need_resched(p); | |
720 | } | |
721 | #endif | |
722 | ||
45bf76df IM |
723 | #if BITS_PER_LONG == 32 |
724 | # define WMULT_CONST (~0UL) | |
725 | #else | |
726 | # define WMULT_CONST (1UL << 32) | |
727 | #endif | |
728 | ||
729 | #define WMULT_SHIFT 32 | |
730 | ||
194081eb IM |
731 | /* |
732 | * Shift right and round: | |
733 | */ | |
cf2ab469 | 734 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 735 | |
cb1c4fc9 | 736 | static unsigned long |
45bf76df IM |
737 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
738 | struct load_weight *lw) | |
739 | { | |
740 | u64 tmp; | |
741 | ||
742 | if (unlikely(!lw->inv_weight)) | |
194081eb | 743 | lw->inv_weight = (WMULT_CONST - lw->weight/2) / lw->weight + 1; |
45bf76df IM |
744 | |
745 | tmp = (u64)delta_exec * weight; | |
746 | /* | |
747 | * Check whether we'd overflow the 64-bit multiplication: | |
748 | */ | |
194081eb | 749 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 750 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
751 | WMULT_SHIFT/2); |
752 | else | |
cf2ab469 | 753 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 754 | |
ecf691da | 755 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
756 | } |
757 | ||
758 | static inline unsigned long | |
759 | calc_delta_fair(unsigned long delta_exec, struct load_weight *lw) | |
760 | { | |
761 | return calc_delta_mine(delta_exec, NICE_0_LOAD, lw); | |
762 | } | |
763 | ||
1091985b | 764 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
765 | { |
766 | lw->weight += inc; | |
45bf76df IM |
767 | } |
768 | ||
1091985b | 769 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
770 | { |
771 | lw->weight -= dec; | |
45bf76df IM |
772 | } |
773 | ||
2dd73a4f PW |
774 | /* |
775 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
776 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
777 | * each task makes to its run queue's load is weighted according to its | |
778 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
779 | * scaled version of the new time slice allocation that they receive on time | |
780 | * slice expiry etc. | |
781 | */ | |
782 | ||
dd41f596 IM |
783 | #define WEIGHT_IDLEPRIO 2 |
784 | #define WMULT_IDLEPRIO (1 << 31) | |
785 | ||
786 | /* | |
787 | * Nice levels are multiplicative, with a gentle 10% change for every | |
788 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
789 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
790 | * that remained on nice 0. | |
791 | * | |
792 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
793 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
794 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
795 | * If a task goes up by ~10% and another task goes down by ~10% then | |
796 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
797 | */ |
798 | static const int prio_to_weight[40] = { | |
254753dc IM |
799 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
800 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
801 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
802 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
803 | /* 0 */ 1024, 820, 655, 526, 423, | |
804 | /* 5 */ 335, 272, 215, 172, 137, | |
805 | /* 10 */ 110, 87, 70, 56, 45, | |
806 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
807 | }; |
808 | ||
5714d2de IM |
809 | /* |
810 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
811 | * | |
812 | * In cases where the weight does not change often, we can use the | |
813 | * precalculated inverse to speed up arithmetics by turning divisions | |
814 | * into multiplications: | |
815 | */ | |
dd41f596 | 816 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
817 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
818 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
819 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
820 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
821 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
822 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
823 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
824 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 825 | }; |
2dd73a4f | 826 | |
dd41f596 IM |
827 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
828 | ||
829 | /* | |
830 | * runqueue iterator, to support SMP load-balancing between different | |
831 | * scheduling classes, without having to expose their internal data | |
832 | * structures to the load-balancing proper: | |
833 | */ | |
834 | struct rq_iterator { | |
835 | void *arg; | |
836 | struct task_struct *(*start)(void *); | |
837 | struct task_struct *(*next)(void *); | |
838 | }; | |
839 | ||
840 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
841 | unsigned long max_nr_move, unsigned long max_load_move, | |
842 | struct sched_domain *sd, enum cpu_idle_type idle, | |
843 | int *all_pinned, unsigned long *load_moved, | |
a4ac01c3 | 844 | int *this_best_prio, struct rq_iterator *iterator); |
dd41f596 IM |
845 | |
846 | #include "sched_stats.h" | |
dd41f596 | 847 | #include "sched_idletask.c" |
5522d5d5 IM |
848 | #include "sched_fair.c" |
849 | #include "sched_rt.c" | |
dd41f596 IM |
850 | #ifdef CONFIG_SCHED_DEBUG |
851 | # include "sched_debug.c" | |
852 | #endif | |
853 | ||
854 | #define sched_class_highest (&rt_sched_class) | |
855 | ||
9c217245 IM |
856 | /* |
857 | * Update delta_exec, delta_fair fields for rq. | |
858 | * | |
859 | * delta_fair clock advances at a rate inversely proportional to | |
495eca49 | 860 | * total load (rq->load.weight) on the runqueue, while |
9c217245 IM |
861 | * delta_exec advances at the same rate as wall-clock (provided |
862 | * cpu is not idle). | |
863 | * | |
864 | * delta_exec / delta_fair is a measure of the (smoothened) load on this | |
865 | * runqueue over any given interval. This (smoothened) load is used | |
866 | * during load balance. | |
867 | * | |
495eca49 | 868 | * This function is called /before/ updating rq->load |
9c217245 IM |
869 | * and when switching tasks. |
870 | */ | |
29b4b623 | 871 | static inline void inc_load(struct rq *rq, const struct task_struct *p) |
9c217245 | 872 | { |
495eca49 | 873 | update_load_add(&rq->load, p->se.load.weight); |
9c217245 IM |
874 | } |
875 | ||
79b5dddf | 876 | static inline void dec_load(struct rq *rq, const struct task_struct *p) |
9c217245 | 877 | { |
495eca49 | 878 | update_load_sub(&rq->load, p->se.load.weight); |
9c217245 IM |
879 | } |
880 | ||
e5fa2237 | 881 | static void inc_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
882 | { |
883 | rq->nr_running++; | |
29b4b623 | 884 | inc_load(rq, p); |
9c217245 IM |
885 | } |
886 | ||
db53181e | 887 | static void dec_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
888 | { |
889 | rq->nr_running--; | |
79b5dddf | 890 | dec_load(rq, p); |
9c217245 IM |
891 | } |
892 | ||
45bf76df IM |
893 | static void set_load_weight(struct task_struct *p) |
894 | { | |
895 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
896 | p->se.load.weight = prio_to_weight[0] * 2; |
897 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
898 | return; | |
899 | } | |
45bf76df | 900 | |
dd41f596 IM |
901 | /* |
902 | * SCHED_IDLE tasks get minimal weight: | |
903 | */ | |
904 | if (p->policy == SCHED_IDLE) { | |
905 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
906 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
907 | return; | |
908 | } | |
71f8bd46 | 909 | |
dd41f596 IM |
910 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
911 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
912 | } |
913 | ||
8159f87e | 914 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 915 | { |
dd41f596 | 916 | sched_info_queued(p); |
fd390f6a | 917 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 918 | p->se.on_rq = 1; |
71f8bd46 IM |
919 | } |
920 | ||
69be72c1 | 921 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 922 | { |
f02231e5 | 923 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 924 | p->se.on_rq = 0; |
71f8bd46 IM |
925 | } |
926 | ||
14531189 | 927 | /* |
dd41f596 | 928 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 929 | */ |
14531189 IM |
930 | static inline int __normal_prio(struct task_struct *p) |
931 | { | |
dd41f596 | 932 | return p->static_prio; |
14531189 IM |
933 | } |
934 | ||
b29739f9 IM |
935 | /* |
936 | * Calculate the expected normal priority: i.e. priority | |
937 | * without taking RT-inheritance into account. Might be | |
938 | * boosted by interactivity modifiers. Changes upon fork, | |
939 | * setprio syscalls, and whenever the interactivity | |
940 | * estimator recalculates. | |
941 | */ | |
36c8b586 | 942 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
943 | { |
944 | int prio; | |
945 | ||
e05606d3 | 946 | if (task_has_rt_policy(p)) |
b29739f9 IM |
947 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
948 | else | |
949 | prio = __normal_prio(p); | |
950 | return prio; | |
951 | } | |
952 | ||
953 | /* | |
954 | * Calculate the current priority, i.e. the priority | |
955 | * taken into account by the scheduler. This value might | |
956 | * be boosted by RT tasks, or might be boosted by | |
957 | * interactivity modifiers. Will be RT if the task got | |
958 | * RT-boosted. If not then it returns p->normal_prio. | |
959 | */ | |
36c8b586 | 960 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
961 | { |
962 | p->normal_prio = normal_prio(p); | |
963 | /* | |
964 | * If we are RT tasks or we were boosted to RT priority, | |
965 | * keep the priority unchanged. Otherwise, update priority | |
966 | * to the normal priority: | |
967 | */ | |
968 | if (!rt_prio(p->prio)) | |
969 | return p->normal_prio; | |
970 | return p->prio; | |
971 | } | |
972 | ||
1da177e4 | 973 | /* |
dd41f596 | 974 | * activate_task - move a task to the runqueue. |
1da177e4 | 975 | */ |
dd41f596 | 976 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 977 | { |
dd41f596 IM |
978 | if (p->state == TASK_UNINTERRUPTIBLE) |
979 | rq->nr_uninterruptible--; | |
1da177e4 | 980 | |
8159f87e | 981 | enqueue_task(rq, p, wakeup); |
e5fa2237 | 982 | inc_nr_running(p, rq); |
1da177e4 LT |
983 | } |
984 | ||
1da177e4 LT |
985 | /* |
986 | * deactivate_task - remove a task from the runqueue. | |
987 | */ | |
2e1cb74a | 988 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 989 | { |
dd41f596 IM |
990 | if (p->state == TASK_UNINTERRUPTIBLE) |
991 | rq->nr_uninterruptible++; | |
992 | ||
69be72c1 | 993 | dequeue_task(rq, p, sleep); |
db53181e | 994 | dec_nr_running(p, rq); |
1da177e4 LT |
995 | } |
996 | ||
1da177e4 LT |
997 | /** |
998 | * task_curr - is this task currently executing on a CPU? | |
999 | * @p: the task in question. | |
1000 | */ | |
36c8b586 | 1001 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1002 | { |
1003 | return cpu_curr(task_cpu(p)) == p; | |
1004 | } | |
1005 | ||
2dd73a4f PW |
1006 | /* Used instead of source_load when we know the type == 0 */ |
1007 | unsigned long weighted_cpuload(const int cpu) | |
1008 | { | |
495eca49 | 1009 | return cpu_rq(cpu)->load.weight; |
dd41f596 IM |
1010 | } |
1011 | ||
1012 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1013 | { | |
1014 | #ifdef CONFIG_SMP | |
1015 | task_thread_info(p)->cpu = cpu; | |
dd41f596 | 1016 | #endif |
29f59db3 | 1017 | set_task_cfs_rq(p); |
2dd73a4f PW |
1018 | } |
1019 | ||
1da177e4 | 1020 | #ifdef CONFIG_SMP |
c65cc870 | 1021 | |
cc367732 IM |
1022 | /* |
1023 | * Is this task likely cache-hot: | |
1024 | */ | |
1025 | static inline int | |
1026 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) | |
1027 | { | |
1028 | s64 delta; | |
1029 | ||
1030 | if (p->sched_class != &fair_sched_class) | |
1031 | return 0; | |
1032 | ||
6bc1665b IM |
1033 | if (sysctl_sched_migration_cost == -1) |
1034 | return 1; | |
1035 | if (sysctl_sched_migration_cost == 0) | |
1036 | return 0; | |
1037 | ||
cc367732 IM |
1038 | delta = now - p->se.exec_start; |
1039 | ||
1040 | return delta < (s64)sysctl_sched_migration_cost; | |
1041 | } | |
1042 | ||
1043 | ||
dd41f596 | 1044 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1045 | { |
dd41f596 IM |
1046 | int old_cpu = task_cpu(p); |
1047 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1048 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1049 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1050 | u64 clock_offset; |
dd41f596 IM |
1051 | |
1052 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1053 | |
1054 | #ifdef CONFIG_SCHEDSTATS | |
1055 | if (p->se.wait_start) | |
1056 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1057 | if (p->se.sleep_start) |
1058 | p->se.sleep_start -= clock_offset; | |
1059 | if (p->se.block_start) | |
1060 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1061 | if (old_cpu != new_cpu) { |
1062 | schedstat_inc(p, se.nr_migrations); | |
1063 | if (task_hot(p, old_rq->clock, NULL)) | |
1064 | schedstat_inc(p, se.nr_forced2_migrations); | |
1065 | } | |
6cfb0d5d | 1066 | #endif |
2830cf8c SV |
1067 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1068 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1069 | |
1070 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1071 | } |
1072 | ||
70b97a7f | 1073 | struct migration_req { |
1da177e4 | 1074 | struct list_head list; |
1da177e4 | 1075 | |
36c8b586 | 1076 | struct task_struct *task; |
1da177e4 LT |
1077 | int dest_cpu; |
1078 | ||
1da177e4 | 1079 | struct completion done; |
70b97a7f | 1080 | }; |
1da177e4 LT |
1081 | |
1082 | /* | |
1083 | * The task's runqueue lock must be held. | |
1084 | * Returns true if you have to wait for migration thread. | |
1085 | */ | |
36c8b586 | 1086 | static int |
70b97a7f | 1087 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1088 | { |
70b97a7f | 1089 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1090 | |
1091 | /* | |
1092 | * If the task is not on a runqueue (and not running), then | |
1093 | * it is sufficient to simply update the task's cpu field. | |
1094 | */ | |
dd41f596 | 1095 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1096 | set_task_cpu(p, dest_cpu); |
1097 | return 0; | |
1098 | } | |
1099 | ||
1100 | init_completion(&req->done); | |
1da177e4 LT |
1101 | req->task = p; |
1102 | req->dest_cpu = dest_cpu; | |
1103 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1104 | |
1da177e4 LT |
1105 | return 1; |
1106 | } | |
1107 | ||
1108 | /* | |
1109 | * wait_task_inactive - wait for a thread to unschedule. | |
1110 | * | |
1111 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1112 | * else this function might spin for a *long* time. This function can't | |
1113 | * be called with interrupts off, or it may introduce deadlock with | |
1114 | * smp_call_function() if an IPI is sent by the same process we are | |
1115 | * waiting to become inactive. | |
1116 | */ | |
36c8b586 | 1117 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1118 | { |
1119 | unsigned long flags; | |
dd41f596 | 1120 | int running, on_rq; |
70b97a7f | 1121 | struct rq *rq; |
1da177e4 | 1122 | |
3a5c359a AK |
1123 | for (;;) { |
1124 | /* | |
1125 | * We do the initial early heuristics without holding | |
1126 | * any task-queue locks at all. We'll only try to get | |
1127 | * the runqueue lock when things look like they will | |
1128 | * work out! | |
1129 | */ | |
1130 | rq = task_rq(p); | |
fa490cfd | 1131 | |
3a5c359a AK |
1132 | /* |
1133 | * If the task is actively running on another CPU | |
1134 | * still, just relax and busy-wait without holding | |
1135 | * any locks. | |
1136 | * | |
1137 | * NOTE! Since we don't hold any locks, it's not | |
1138 | * even sure that "rq" stays as the right runqueue! | |
1139 | * But we don't care, since "task_running()" will | |
1140 | * return false if the runqueue has changed and p | |
1141 | * is actually now running somewhere else! | |
1142 | */ | |
1143 | while (task_running(rq, p)) | |
1144 | cpu_relax(); | |
fa490cfd | 1145 | |
3a5c359a AK |
1146 | /* |
1147 | * Ok, time to look more closely! We need the rq | |
1148 | * lock now, to be *sure*. If we're wrong, we'll | |
1149 | * just go back and repeat. | |
1150 | */ | |
1151 | rq = task_rq_lock(p, &flags); | |
1152 | running = task_running(rq, p); | |
1153 | on_rq = p->se.on_rq; | |
1154 | task_rq_unlock(rq, &flags); | |
fa490cfd | 1155 | |
3a5c359a AK |
1156 | /* |
1157 | * Was it really running after all now that we | |
1158 | * checked with the proper locks actually held? | |
1159 | * | |
1160 | * Oops. Go back and try again.. | |
1161 | */ | |
1162 | if (unlikely(running)) { | |
1163 | cpu_relax(); | |
1164 | continue; | |
1165 | } | |
fa490cfd | 1166 | |
3a5c359a AK |
1167 | /* |
1168 | * It's not enough that it's not actively running, | |
1169 | * it must be off the runqueue _entirely_, and not | |
1170 | * preempted! | |
1171 | * | |
1172 | * So if it wa still runnable (but just not actively | |
1173 | * running right now), it's preempted, and we should | |
1174 | * yield - it could be a while. | |
1175 | */ | |
1176 | if (unlikely(on_rq)) { | |
1177 | schedule_timeout_uninterruptible(1); | |
1178 | continue; | |
1179 | } | |
fa490cfd | 1180 | |
3a5c359a AK |
1181 | /* |
1182 | * Ahh, all good. It wasn't running, and it wasn't | |
1183 | * runnable, which means that it will never become | |
1184 | * running in the future either. We're all done! | |
1185 | */ | |
1186 | break; | |
1187 | } | |
1da177e4 LT |
1188 | } |
1189 | ||
1190 | /*** | |
1191 | * kick_process - kick a running thread to enter/exit the kernel | |
1192 | * @p: the to-be-kicked thread | |
1193 | * | |
1194 | * Cause a process which is running on another CPU to enter | |
1195 | * kernel-mode, without any delay. (to get signals handled.) | |
1196 | * | |
1197 | * NOTE: this function doesnt have to take the runqueue lock, | |
1198 | * because all it wants to ensure is that the remote task enters | |
1199 | * the kernel. If the IPI races and the task has been migrated | |
1200 | * to another CPU then no harm is done and the purpose has been | |
1201 | * achieved as well. | |
1202 | */ | |
36c8b586 | 1203 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1204 | { |
1205 | int cpu; | |
1206 | ||
1207 | preempt_disable(); | |
1208 | cpu = task_cpu(p); | |
1209 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1210 | smp_send_reschedule(cpu); | |
1211 | preempt_enable(); | |
1212 | } | |
1213 | ||
1214 | /* | |
2dd73a4f PW |
1215 | * Return a low guess at the load of a migration-source cpu weighted |
1216 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1217 | * |
1218 | * We want to under-estimate the load of migration sources, to | |
1219 | * balance conservatively. | |
1220 | */ | |
a9957449 | 1221 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 1222 | { |
70b97a7f | 1223 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1224 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1225 | |
3b0bd9bc | 1226 | if (type == 0) |
dd41f596 | 1227 | return total; |
b910472d | 1228 | |
dd41f596 | 1229 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
1230 | } |
1231 | ||
1232 | /* | |
2dd73a4f PW |
1233 | * Return a high guess at the load of a migration-target cpu weighted |
1234 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1235 | */ |
a9957449 | 1236 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 1237 | { |
70b97a7f | 1238 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1239 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1240 | |
7897986b | 1241 | if (type == 0) |
dd41f596 | 1242 | return total; |
3b0bd9bc | 1243 | |
dd41f596 | 1244 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
1245 | } |
1246 | ||
1247 | /* | |
1248 | * Return the average load per task on the cpu's run queue | |
1249 | */ | |
1250 | static inline unsigned long cpu_avg_load_per_task(int cpu) | |
1251 | { | |
70b97a7f | 1252 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1253 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f PW |
1254 | unsigned long n = rq->nr_running; |
1255 | ||
dd41f596 | 1256 | return n ? total / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1257 | } |
1258 | ||
147cbb4b NP |
1259 | /* |
1260 | * find_idlest_group finds and returns the least busy CPU group within the | |
1261 | * domain. | |
1262 | */ | |
1263 | static struct sched_group * | |
1264 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1265 | { | |
1266 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1267 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1268 | int load_idx = sd->forkexec_idx; | |
1269 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1270 | ||
1271 | do { | |
1272 | unsigned long load, avg_load; | |
1273 | int local_group; | |
1274 | int i; | |
1275 | ||
da5a5522 BD |
1276 | /* Skip over this group if it has no CPUs allowed */ |
1277 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 1278 | continue; |
da5a5522 | 1279 | |
147cbb4b | 1280 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1281 | |
1282 | /* Tally up the load of all CPUs in the group */ | |
1283 | avg_load = 0; | |
1284 | ||
1285 | for_each_cpu_mask(i, group->cpumask) { | |
1286 | /* Bias balancing toward cpus of our domain */ | |
1287 | if (local_group) | |
1288 | load = source_load(i, load_idx); | |
1289 | else | |
1290 | load = target_load(i, load_idx); | |
1291 | ||
1292 | avg_load += load; | |
1293 | } | |
1294 | ||
1295 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1296 | avg_load = sg_div_cpu_power(group, |
1297 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1298 | |
1299 | if (local_group) { | |
1300 | this_load = avg_load; | |
1301 | this = group; | |
1302 | } else if (avg_load < min_load) { | |
1303 | min_load = avg_load; | |
1304 | idlest = group; | |
1305 | } | |
3a5c359a | 1306 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
1307 | |
1308 | if (!idlest || 100*this_load < imbalance*min_load) | |
1309 | return NULL; | |
1310 | return idlest; | |
1311 | } | |
1312 | ||
1313 | /* | |
0feaece9 | 1314 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1315 | */ |
95cdf3b7 IM |
1316 | static int |
1317 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) | |
147cbb4b | 1318 | { |
da5a5522 | 1319 | cpumask_t tmp; |
147cbb4b NP |
1320 | unsigned long load, min_load = ULONG_MAX; |
1321 | int idlest = -1; | |
1322 | int i; | |
1323 | ||
da5a5522 BD |
1324 | /* Traverse only the allowed CPUs */ |
1325 | cpus_and(tmp, group->cpumask, p->cpus_allowed); | |
1326 | ||
1327 | for_each_cpu_mask(i, tmp) { | |
2dd73a4f | 1328 | load = weighted_cpuload(i); |
147cbb4b NP |
1329 | |
1330 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1331 | min_load = load; | |
1332 | idlest = i; | |
1333 | } | |
1334 | } | |
1335 | ||
1336 | return idlest; | |
1337 | } | |
1338 | ||
476d139c NP |
1339 | /* |
1340 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1341 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1342 | * SD_BALANCE_EXEC. | |
1343 | * | |
1344 | * Balance, ie. select the least loaded group. | |
1345 | * | |
1346 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1347 | * | |
1348 | * preempt must be disabled. | |
1349 | */ | |
1350 | static int sched_balance_self(int cpu, int flag) | |
1351 | { | |
1352 | struct task_struct *t = current; | |
1353 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1354 | |
c96d145e | 1355 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
1356 | /* |
1357 | * If power savings logic is enabled for a domain, stop there. | |
1358 | */ | |
5c45bf27 SS |
1359 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
1360 | break; | |
476d139c NP |
1361 | if (tmp->flags & flag) |
1362 | sd = tmp; | |
c96d145e | 1363 | } |
476d139c NP |
1364 | |
1365 | while (sd) { | |
1366 | cpumask_t span; | |
1367 | struct sched_group *group; | |
1a848870 SS |
1368 | int new_cpu, weight; |
1369 | ||
1370 | if (!(sd->flags & flag)) { | |
1371 | sd = sd->child; | |
1372 | continue; | |
1373 | } | |
476d139c NP |
1374 | |
1375 | span = sd->span; | |
1376 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1377 | if (!group) { |
1378 | sd = sd->child; | |
1379 | continue; | |
1380 | } | |
476d139c | 1381 | |
da5a5522 | 1382 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
1383 | if (new_cpu == -1 || new_cpu == cpu) { |
1384 | /* Now try balancing at a lower domain level of cpu */ | |
1385 | sd = sd->child; | |
1386 | continue; | |
1387 | } | |
476d139c | 1388 | |
1a848870 | 1389 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1390 | cpu = new_cpu; |
476d139c NP |
1391 | sd = NULL; |
1392 | weight = cpus_weight(span); | |
1393 | for_each_domain(cpu, tmp) { | |
1394 | if (weight <= cpus_weight(tmp->span)) | |
1395 | break; | |
1396 | if (tmp->flags & flag) | |
1397 | sd = tmp; | |
1398 | } | |
1399 | /* while loop will break here if sd == NULL */ | |
1400 | } | |
1401 | ||
1402 | return cpu; | |
1403 | } | |
1404 | ||
1405 | #endif /* CONFIG_SMP */ | |
1da177e4 LT |
1406 | |
1407 | /* | |
1408 | * wake_idle() will wake a task on an idle cpu if task->cpu is | |
1409 | * not idle and an idle cpu is available. The span of cpus to | |
1410 | * search starts with cpus closest then further out as needed, | |
1411 | * so we always favor a closer, idle cpu. | |
1412 | * | |
1413 | * Returns the CPU we should wake onto. | |
1414 | */ | |
1415 | #if defined(ARCH_HAS_SCHED_WAKE_IDLE) | |
36c8b586 | 1416 | static int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1417 | { |
1418 | cpumask_t tmp; | |
1419 | struct sched_domain *sd; | |
1420 | int i; | |
1421 | ||
4953198b SS |
1422 | /* |
1423 | * If it is idle, then it is the best cpu to run this task. | |
1424 | * | |
1425 | * This cpu is also the best, if it has more than one task already. | |
1426 | * Siblings must be also busy(in most cases) as they didn't already | |
1427 | * pickup the extra load from this cpu and hence we need not check | |
1428 | * sibling runqueue info. This will avoid the checks and cache miss | |
1429 | * penalities associated with that. | |
1430 | */ | |
1431 | if (idle_cpu(cpu) || cpu_rq(cpu)->nr_running > 1) | |
1da177e4 LT |
1432 | return cpu; |
1433 | ||
1434 | for_each_domain(cpu, sd) { | |
1435 | if (sd->flags & SD_WAKE_IDLE) { | |
e0f364f4 | 1436 | cpus_and(tmp, sd->span, p->cpus_allowed); |
1da177e4 | 1437 | for_each_cpu_mask(i, tmp) { |
cc367732 IM |
1438 | if (idle_cpu(i)) { |
1439 | if (i != task_cpu(p)) { | |
1440 | schedstat_inc(p, | |
1441 | se.nr_wakeups_idle); | |
1442 | } | |
1da177e4 | 1443 | return i; |
cc367732 | 1444 | } |
1da177e4 | 1445 | } |
9761eea8 | 1446 | } else { |
e0f364f4 | 1447 | break; |
9761eea8 | 1448 | } |
1da177e4 LT |
1449 | } |
1450 | return cpu; | |
1451 | } | |
1452 | #else | |
36c8b586 | 1453 | static inline int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1454 | { |
1455 | return cpu; | |
1456 | } | |
1457 | #endif | |
1458 | ||
1459 | /*** | |
1460 | * try_to_wake_up - wake up a thread | |
1461 | * @p: the to-be-woken-up thread | |
1462 | * @state: the mask of task states that can be woken | |
1463 | * @sync: do a synchronous wakeup? | |
1464 | * | |
1465 | * Put it on the run-queue if it's not already there. The "current" | |
1466 | * thread is always on the run-queue (except when the actual | |
1467 | * re-schedule is in progress), and as such you're allowed to do | |
1468 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1469 | * runnable without the overhead of this. | |
1470 | * | |
1471 | * returns failure only if the task is already active. | |
1472 | */ | |
36c8b586 | 1473 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 1474 | { |
cc367732 | 1475 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
1476 | unsigned long flags; |
1477 | long old_state; | |
70b97a7f | 1478 | struct rq *rq; |
1da177e4 | 1479 | #ifdef CONFIG_SMP |
7897986b | 1480 | struct sched_domain *sd, *this_sd = NULL; |
70b97a7f | 1481 | unsigned long load, this_load; |
1da177e4 LT |
1482 | int new_cpu; |
1483 | #endif | |
1484 | ||
1485 | rq = task_rq_lock(p, &flags); | |
1486 | old_state = p->state; | |
1487 | if (!(old_state & state)) | |
1488 | goto out; | |
1489 | ||
dd41f596 | 1490 | if (p->se.on_rq) |
1da177e4 LT |
1491 | goto out_running; |
1492 | ||
1493 | cpu = task_cpu(p); | |
cc367732 | 1494 | orig_cpu = cpu; |
1da177e4 LT |
1495 | this_cpu = smp_processor_id(); |
1496 | ||
1497 | #ifdef CONFIG_SMP | |
1498 | if (unlikely(task_running(rq, p))) | |
1499 | goto out_activate; | |
1500 | ||
7897986b NP |
1501 | new_cpu = cpu; |
1502 | ||
2d72376b | 1503 | schedstat_inc(rq, ttwu_count); |
1da177e4 LT |
1504 | if (cpu == this_cpu) { |
1505 | schedstat_inc(rq, ttwu_local); | |
7897986b NP |
1506 | goto out_set_cpu; |
1507 | } | |
1508 | ||
1509 | for_each_domain(this_cpu, sd) { | |
1510 | if (cpu_isset(cpu, sd->span)) { | |
1511 | schedstat_inc(sd, ttwu_wake_remote); | |
1512 | this_sd = sd; | |
1513 | break; | |
1da177e4 LT |
1514 | } |
1515 | } | |
1da177e4 | 1516 | |
7897986b | 1517 | if (unlikely(!cpu_isset(this_cpu, p->cpus_allowed))) |
1da177e4 LT |
1518 | goto out_set_cpu; |
1519 | ||
1da177e4 | 1520 | /* |
7897986b | 1521 | * Check for affine wakeup and passive balancing possibilities. |
1da177e4 | 1522 | */ |
7897986b NP |
1523 | if (this_sd) { |
1524 | int idx = this_sd->wake_idx; | |
1525 | unsigned int imbalance; | |
1da177e4 | 1526 | |
a3f21bce NP |
1527 | imbalance = 100 + (this_sd->imbalance_pct - 100) / 2; |
1528 | ||
7897986b NP |
1529 | load = source_load(cpu, idx); |
1530 | this_load = target_load(this_cpu, idx); | |
1da177e4 | 1531 | |
7897986b NP |
1532 | new_cpu = this_cpu; /* Wake to this CPU if we can */ |
1533 | ||
a3f21bce NP |
1534 | if (this_sd->flags & SD_WAKE_AFFINE) { |
1535 | unsigned long tl = this_load; | |
33859f7f MOS |
1536 | unsigned long tl_per_task; |
1537 | ||
71e20f18 IM |
1538 | /* |
1539 | * Attract cache-cold tasks on sync wakeups: | |
1540 | */ | |
1541 | if (sync && !task_hot(p, rq->clock, this_sd)) | |
1542 | goto out_set_cpu; | |
1543 | ||
cc367732 | 1544 | schedstat_inc(p, se.nr_wakeups_affine_attempts); |
33859f7f | 1545 | tl_per_task = cpu_avg_load_per_task(this_cpu); |
2dd73a4f | 1546 | |
1da177e4 | 1547 | /* |
a3f21bce NP |
1548 | * If sync wakeup then subtract the (maximum possible) |
1549 | * effect of the currently running task from the load | |
1550 | * of the current CPU: | |
1da177e4 | 1551 | */ |
a3f21bce | 1552 | if (sync) |
dd41f596 | 1553 | tl -= current->se.load.weight; |
a3f21bce NP |
1554 | |
1555 | if ((tl <= load && | |
2dd73a4f | 1556 | tl + target_load(cpu, idx) <= tl_per_task) || |
dd41f596 | 1557 | 100*(tl + p->se.load.weight) <= imbalance*load) { |
a3f21bce NP |
1558 | /* |
1559 | * This domain has SD_WAKE_AFFINE and | |
1560 | * p is cache cold in this domain, and | |
1561 | * there is no bad imbalance. | |
1562 | */ | |
1563 | schedstat_inc(this_sd, ttwu_move_affine); | |
cc367732 | 1564 | schedstat_inc(p, se.nr_wakeups_affine); |
a3f21bce NP |
1565 | goto out_set_cpu; |
1566 | } | |
1567 | } | |
1568 | ||
1569 | /* | |
1570 | * Start passive balancing when half the imbalance_pct | |
1571 | * limit is reached. | |
1572 | */ | |
1573 | if (this_sd->flags & SD_WAKE_BALANCE) { | |
1574 | if (imbalance*this_load <= 100*load) { | |
1575 | schedstat_inc(this_sd, ttwu_move_balance); | |
cc367732 | 1576 | schedstat_inc(p, se.nr_wakeups_passive); |
a3f21bce NP |
1577 | goto out_set_cpu; |
1578 | } | |
1da177e4 LT |
1579 | } |
1580 | } | |
1581 | ||
1582 | new_cpu = cpu; /* Could not wake to this_cpu. Wake to cpu instead */ | |
1583 | out_set_cpu: | |
1584 | new_cpu = wake_idle(new_cpu, p); | |
1585 | if (new_cpu != cpu) { | |
1586 | set_task_cpu(p, new_cpu); | |
1587 | task_rq_unlock(rq, &flags); | |
1588 | /* might preempt at this point */ | |
1589 | rq = task_rq_lock(p, &flags); | |
1590 | old_state = p->state; | |
1591 | if (!(old_state & state)) | |
1592 | goto out; | |
dd41f596 | 1593 | if (p->se.on_rq) |
1da177e4 LT |
1594 | goto out_running; |
1595 | ||
1596 | this_cpu = smp_processor_id(); | |
1597 | cpu = task_cpu(p); | |
1598 | } | |
1599 | ||
1600 | out_activate: | |
1601 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
1602 | schedstat_inc(p, se.nr_wakeups); |
1603 | if (sync) | |
1604 | schedstat_inc(p, se.nr_wakeups_sync); | |
1605 | if (orig_cpu != cpu) | |
1606 | schedstat_inc(p, se.nr_wakeups_migrate); | |
1607 | if (cpu == this_cpu) | |
1608 | schedstat_inc(p, se.nr_wakeups_local); | |
1609 | else | |
1610 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 1611 | update_rq_clock(rq); |
dd41f596 | 1612 | activate_task(rq, p, 1); |
9c63d9c0 | 1613 | check_preempt_curr(rq, p); |
1da177e4 LT |
1614 | success = 1; |
1615 | ||
1616 | out_running: | |
1617 | p->state = TASK_RUNNING; | |
1618 | out: | |
1619 | task_rq_unlock(rq, &flags); | |
1620 | ||
1621 | return success; | |
1622 | } | |
1623 | ||
36c8b586 | 1624 | int fastcall wake_up_process(struct task_struct *p) |
1da177e4 LT |
1625 | { |
1626 | return try_to_wake_up(p, TASK_STOPPED | TASK_TRACED | | |
1627 | TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE, 0); | |
1628 | } | |
1da177e4 LT |
1629 | EXPORT_SYMBOL(wake_up_process); |
1630 | ||
36c8b586 | 1631 | int fastcall wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1632 | { |
1633 | return try_to_wake_up(p, state, 0); | |
1634 | } | |
1635 | ||
1da177e4 LT |
1636 | /* |
1637 | * Perform scheduler related setup for a newly forked process p. | |
1638 | * p is forked by current. | |
dd41f596 IM |
1639 | * |
1640 | * __sched_fork() is basic setup used by init_idle() too: | |
1641 | */ | |
1642 | static void __sched_fork(struct task_struct *p) | |
1643 | { | |
dd41f596 IM |
1644 | p->se.exec_start = 0; |
1645 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1646 | p->se.prev_sum_exec_runtime = 0; |
6cfb0d5d IM |
1647 | |
1648 | #ifdef CONFIG_SCHEDSTATS | |
1649 | p->se.wait_start = 0; | |
dd41f596 IM |
1650 | p->se.sum_sleep_runtime = 0; |
1651 | p->se.sleep_start = 0; | |
dd41f596 IM |
1652 | p->se.block_start = 0; |
1653 | p->se.sleep_max = 0; | |
1654 | p->se.block_max = 0; | |
1655 | p->se.exec_max = 0; | |
eba1ed4b | 1656 | p->se.slice_max = 0; |
dd41f596 | 1657 | p->se.wait_max = 0; |
6cfb0d5d | 1658 | #endif |
476d139c | 1659 | |
dd41f596 IM |
1660 | INIT_LIST_HEAD(&p->run_list); |
1661 | p->se.on_rq = 0; | |
476d139c | 1662 | |
e107be36 AK |
1663 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1664 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1665 | #endif | |
1666 | ||
1da177e4 LT |
1667 | /* |
1668 | * We mark the process as running here, but have not actually | |
1669 | * inserted it onto the runqueue yet. This guarantees that | |
1670 | * nobody will actually run it, and a signal or other external | |
1671 | * event cannot wake it up and insert it on the runqueue either. | |
1672 | */ | |
1673 | p->state = TASK_RUNNING; | |
dd41f596 IM |
1674 | } |
1675 | ||
1676 | /* | |
1677 | * fork()/clone()-time setup: | |
1678 | */ | |
1679 | void sched_fork(struct task_struct *p, int clone_flags) | |
1680 | { | |
1681 | int cpu = get_cpu(); | |
1682 | ||
1683 | __sched_fork(p); | |
1684 | ||
1685 | #ifdef CONFIG_SMP | |
1686 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
1687 | #endif | |
02e4bac2 | 1688 | set_task_cpu(p, cpu); |
b29739f9 IM |
1689 | |
1690 | /* | |
1691 | * Make sure we do not leak PI boosting priority to the child: | |
1692 | */ | |
1693 | p->prio = current->normal_prio; | |
2ddbf952 HS |
1694 | if (!rt_prio(p->prio)) |
1695 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1696 | |
52f17b6c | 1697 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1698 | if (likely(sched_info_on())) |
52f17b6c | 1699 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1700 | #endif |
d6077cb8 | 1701 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
1702 | p->oncpu = 0; |
1703 | #endif | |
1da177e4 | 1704 | #ifdef CONFIG_PREEMPT |
4866cde0 | 1705 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1706 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1707 | #endif |
476d139c | 1708 | put_cpu(); |
1da177e4 LT |
1709 | } |
1710 | ||
1711 | /* | |
1712 | * wake_up_new_task - wake up a newly created task for the first time. | |
1713 | * | |
1714 | * This function will do some initial scheduler statistics housekeeping | |
1715 | * that must be done for every newly created context, then puts the task | |
1716 | * on the runqueue and wakes it. | |
1717 | */ | |
36c8b586 | 1718 | void fastcall wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
1719 | { |
1720 | unsigned long flags; | |
dd41f596 | 1721 | struct rq *rq; |
1da177e4 LT |
1722 | |
1723 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 1724 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 1725 | update_rq_clock(rq); |
1da177e4 LT |
1726 | |
1727 | p->prio = effective_prio(p); | |
1728 | ||
b9dca1e0 | 1729 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 1730 | activate_task(rq, p, 0); |
1da177e4 | 1731 | } else { |
1da177e4 | 1732 | /* |
dd41f596 IM |
1733 | * Let the scheduling class do new task startup |
1734 | * management (if any): | |
1da177e4 | 1735 | */ |
ee0827d8 | 1736 | p->sched_class->task_new(rq, p); |
e5fa2237 | 1737 | inc_nr_running(p, rq); |
1da177e4 | 1738 | } |
dd41f596 IM |
1739 | check_preempt_curr(rq, p); |
1740 | task_rq_unlock(rq, &flags); | |
1da177e4 LT |
1741 | } |
1742 | ||
e107be36 AK |
1743 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1744 | ||
1745 | /** | |
421cee29 RD |
1746 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
1747 | * @notifier: notifier struct to register | |
e107be36 AK |
1748 | */ |
1749 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1750 | { | |
1751 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1752 | } | |
1753 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1754 | ||
1755 | /** | |
1756 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1757 | * @notifier: notifier struct to unregister |
e107be36 AK |
1758 | * |
1759 | * This is safe to call from within a preemption notifier. | |
1760 | */ | |
1761 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1762 | { | |
1763 | hlist_del(¬ifier->link); | |
1764 | } | |
1765 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1766 | ||
1767 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1768 | { | |
1769 | struct preempt_notifier *notifier; | |
1770 | struct hlist_node *node; | |
1771 | ||
1772 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1773 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
1774 | } | |
1775 | ||
1776 | static void | |
1777 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1778 | struct task_struct *next) | |
1779 | { | |
1780 | struct preempt_notifier *notifier; | |
1781 | struct hlist_node *node; | |
1782 | ||
1783 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1784 | notifier->ops->sched_out(notifier, next); | |
1785 | } | |
1786 | ||
1787 | #else | |
1788 | ||
1789 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1790 | { | |
1791 | } | |
1792 | ||
1793 | static void | |
1794 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1795 | struct task_struct *next) | |
1796 | { | |
1797 | } | |
1798 | ||
1799 | #endif | |
1800 | ||
4866cde0 NP |
1801 | /** |
1802 | * prepare_task_switch - prepare to switch tasks | |
1803 | * @rq: the runqueue preparing to switch | |
421cee29 | 1804 | * @prev: the current task that is being switched out |
4866cde0 NP |
1805 | * @next: the task we are going to switch to. |
1806 | * | |
1807 | * This is called with the rq lock held and interrupts off. It must | |
1808 | * be paired with a subsequent finish_task_switch after the context | |
1809 | * switch. | |
1810 | * | |
1811 | * prepare_task_switch sets up locking and calls architecture specific | |
1812 | * hooks. | |
1813 | */ | |
e107be36 AK |
1814 | static inline void |
1815 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1816 | struct task_struct *next) | |
4866cde0 | 1817 | { |
e107be36 | 1818 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1819 | prepare_lock_switch(rq, next); |
1820 | prepare_arch_switch(next); | |
1821 | } | |
1822 | ||
1da177e4 LT |
1823 | /** |
1824 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1825 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1826 | * @prev: the thread we just switched away from. |
1827 | * | |
4866cde0 NP |
1828 | * finish_task_switch must be called after the context switch, paired |
1829 | * with a prepare_task_switch call before the context switch. | |
1830 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1831 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1832 | * |
1833 | * Note that we may have delayed dropping an mm in context_switch(). If | |
1834 | * so, we finish that here outside of the runqueue lock. (Doing it | |
1835 | * with the lock held can cause deadlocks; see schedule() for | |
1836 | * details.) | |
1837 | */ | |
a9957449 | 1838 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1839 | __releases(rq->lock) |
1840 | { | |
1da177e4 | 1841 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1842 | long prev_state; |
1da177e4 LT |
1843 | |
1844 | rq->prev_mm = NULL; | |
1845 | ||
1846 | /* | |
1847 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1848 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1849 | * schedule one last time. The schedule call will never return, and |
1850 | * the scheduled task must drop that reference. | |
c394cc9f | 1851 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1852 | * still held, otherwise prev could be scheduled on another cpu, die |
1853 | * there before we look at prev->state, and then the reference would | |
1854 | * be dropped twice. | |
1855 | * Manfred Spraul <manfred@colorfullife.com> | |
1856 | */ | |
55a101f8 | 1857 | prev_state = prev->state; |
4866cde0 NP |
1858 | finish_arch_switch(prev); |
1859 | finish_lock_switch(rq, prev); | |
e107be36 | 1860 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1861 | if (mm) |
1862 | mmdrop(mm); | |
c394cc9f | 1863 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1864 | /* |
1865 | * Remove function-return probe instances associated with this | |
1866 | * task and put them back on the free list. | |
9761eea8 | 1867 | */ |
c6fd91f0 | 1868 | kprobe_flush_task(prev); |
1da177e4 | 1869 | put_task_struct(prev); |
c6fd91f0 | 1870 | } |
1da177e4 LT |
1871 | } |
1872 | ||
1873 | /** | |
1874 | * schedule_tail - first thing a freshly forked thread must call. | |
1875 | * @prev: the thread we just switched away from. | |
1876 | */ | |
36c8b586 | 1877 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1878 | __releases(rq->lock) |
1879 | { | |
70b97a7f IM |
1880 | struct rq *rq = this_rq(); |
1881 | ||
4866cde0 NP |
1882 | finish_task_switch(rq, prev); |
1883 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
1884 | /* In this case, finish_task_switch does not reenable preemption */ | |
1885 | preempt_enable(); | |
1886 | #endif | |
1da177e4 | 1887 | if (current->set_child_tid) |
b488893a | 1888 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
1889 | } |
1890 | ||
1891 | /* | |
1892 | * context_switch - switch to the new MM and the new | |
1893 | * thread's register state. | |
1894 | */ | |
dd41f596 | 1895 | static inline void |
70b97a7f | 1896 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1897 | struct task_struct *next) |
1da177e4 | 1898 | { |
dd41f596 | 1899 | struct mm_struct *mm, *oldmm; |
1da177e4 | 1900 | |
e107be36 | 1901 | prepare_task_switch(rq, prev, next); |
dd41f596 IM |
1902 | mm = next->mm; |
1903 | oldmm = prev->active_mm; | |
9226d125 ZA |
1904 | /* |
1905 | * For paravirt, this is coupled with an exit in switch_to to | |
1906 | * combine the page table reload and the switch backend into | |
1907 | * one hypercall. | |
1908 | */ | |
1909 | arch_enter_lazy_cpu_mode(); | |
1910 | ||
dd41f596 | 1911 | if (unlikely(!mm)) { |
1da177e4 LT |
1912 | next->active_mm = oldmm; |
1913 | atomic_inc(&oldmm->mm_count); | |
1914 | enter_lazy_tlb(oldmm, next); | |
1915 | } else | |
1916 | switch_mm(oldmm, mm, next); | |
1917 | ||
dd41f596 | 1918 | if (unlikely(!prev->mm)) { |
1da177e4 | 1919 | prev->active_mm = NULL; |
1da177e4 LT |
1920 | rq->prev_mm = oldmm; |
1921 | } | |
3a5f5e48 IM |
1922 | /* |
1923 | * Since the runqueue lock will be released by the next | |
1924 | * task (which is an invalid locking op but in the case | |
1925 | * of the scheduler it's an obvious special-case), so we | |
1926 | * do an early lockdep release here: | |
1927 | */ | |
1928 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 1929 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 1930 | #endif |
1da177e4 LT |
1931 | |
1932 | /* Here we just switch the register state and the stack. */ | |
1933 | switch_to(prev, next, prev); | |
1934 | ||
dd41f596 IM |
1935 | barrier(); |
1936 | /* | |
1937 | * this_rq must be evaluated again because prev may have moved | |
1938 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
1939 | * frame will be invalid. | |
1940 | */ | |
1941 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
1942 | } |
1943 | ||
1944 | /* | |
1945 | * nr_running, nr_uninterruptible and nr_context_switches: | |
1946 | * | |
1947 | * externally visible scheduler statistics: current number of runnable | |
1948 | * threads, current number of uninterruptible-sleeping threads, total | |
1949 | * number of context switches performed since bootup. | |
1950 | */ | |
1951 | unsigned long nr_running(void) | |
1952 | { | |
1953 | unsigned long i, sum = 0; | |
1954 | ||
1955 | for_each_online_cpu(i) | |
1956 | sum += cpu_rq(i)->nr_running; | |
1957 | ||
1958 | return sum; | |
1959 | } | |
1960 | ||
1961 | unsigned long nr_uninterruptible(void) | |
1962 | { | |
1963 | unsigned long i, sum = 0; | |
1964 | ||
0a945022 | 1965 | for_each_possible_cpu(i) |
1da177e4 LT |
1966 | sum += cpu_rq(i)->nr_uninterruptible; |
1967 | ||
1968 | /* | |
1969 | * Since we read the counters lockless, it might be slightly | |
1970 | * inaccurate. Do not allow it to go below zero though: | |
1971 | */ | |
1972 | if (unlikely((long)sum < 0)) | |
1973 | sum = 0; | |
1974 | ||
1975 | return sum; | |
1976 | } | |
1977 | ||
1978 | unsigned long long nr_context_switches(void) | |
1979 | { | |
cc94abfc SR |
1980 | int i; |
1981 | unsigned long long sum = 0; | |
1da177e4 | 1982 | |
0a945022 | 1983 | for_each_possible_cpu(i) |
1da177e4 LT |
1984 | sum += cpu_rq(i)->nr_switches; |
1985 | ||
1986 | return sum; | |
1987 | } | |
1988 | ||
1989 | unsigned long nr_iowait(void) | |
1990 | { | |
1991 | unsigned long i, sum = 0; | |
1992 | ||
0a945022 | 1993 | for_each_possible_cpu(i) |
1da177e4 LT |
1994 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
1995 | ||
1996 | return sum; | |
1997 | } | |
1998 | ||
db1b1fef JS |
1999 | unsigned long nr_active(void) |
2000 | { | |
2001 | unsigned long i, running = 0, uninterruptible = 0; | |
2002 | ||
2003 | for_each_online_cpu(i) { | |
2004 | running += cpu_rq(i)->nr_running; | |
2005 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2006 | } | |
2007 | ||
2008 | if (unlikely((long)uninterruptible < 0)) | |
2009 | uninterruptible = 0; | |
2010 | ||
2011 | return running + uninterruptible; | |
2012 | } | |
2013 | ||
48f24c4d | 2014 | /* |
dd41f596 IM |
2015 | * Update rq->cpu_load[] statistics. This function is usually called every |
2016 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2017 | */ |
dd41f596 | 2018 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2019 | { |
495eca49 | 2020 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2021 | int i, scale; |
2022 | ||
2023 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2024 | |
2025 | /* Update our load: */ | |
2026 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2027 | unsigned long old_load, new_load; | |
2028 | ||
2029 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2030 | ||
2031 | old_load = this_rq->cpu_load[i]; | |
2032 | new_load = this_load; | |
a25707f3 IM |
2033 | /* |
2034 | * Round up the averaging division if load is increasing. This | |
2035 | * prevents us from getting stuck on 9 if the load is 10, for | |
2036 | * example. | |
2037 | */ | |
2038 | if (new_load > old_load) | |
2039 | new_load += scale-1; | |
dd41f596 IM |
2040 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2041 | } | |
48f24c4d IM |
2042 | } |
2043 | ||
dd41f596 IM |
2044 | #ifdef CONFIG_SMP |
2045 | ||
1da177e4 LT |
2046 | /* |
2047 | * double_rq_lock - safely lock two runqueues | |
2048 | * | |
2049 | * Note this does not disable interrupts like task_rq_lock, | |
2050 | * you need to do so manually before calling. | |
2051 | */ | |
70b97a7f | 2052 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2053 | __acquires(rq1->lock) |
2054 | __acquires(rq2->lock) | |
2055 | { | |
054b9108 | 2056 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2057 | if (rq1 == rq2) { |
2058 | spin_lock(&rq1->lock); | |
2059 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2060 | } else { | |
c96d145e | 2061 | if (rq1 < rq2) { |
1da177e4 LT |
2062 | spin_lock(&rq1->lock); |
2063 | spin_lock(&rq2->lock); | |
2064 | } else { | |
2065 | spin_lock(&rq2->lock); | |
2066 | spin_lock(&rq1->lock); | |
2067 | } | |
2068 | } | |
6e82a3be IM |
2069 | update_rq_clock(rq1); |
2070 | update_rq_clock(rq2); | |
1da177e4 LT |
2071 | } |
2072 | ||
2073 | /* | |
2074 | * double_rq_unlock - safely unlock two runqueues | |
2075 | * | |
2076 | * Note this does not restore interrupts like task_rq_unlock, | |
2077 | * you need to do so manually after calling. | |
2078 | */ | |
70b97a7f | 2079 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2080 | __releases(rq1->lock) |
2081 | __releases(rq2->lock) | |
2082 | { | |
2083 | spin_unlock(&rq1->lock); | |
2084 | if (rq1 != rq2) | |
2085 | spin_unlock(&rq2->lock); | |
2086 | else | |
2087 | __release(rq2->lock); | |
2088 | } | |
2089 | ||
2090 | /* | |
2091 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2092 | */ | |
70b97a7f | 2093 | static void double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2094 | __releases(this_rq->lock) |
2095 | __acquires(busiest->lock) | |
2096 | __acquires(this_rq->lock) | |
2097 | { | |
054b9108 KK |
2098 | if (unlikely(!irqs_disabled())) { |
2099 | /* printk() doesn't work good under rq->lock */ | |
2100 | spin_unlock(&this_rq->lock); | |
2101 | BUG_ON(1); | |
2102 | } | |
1da177e4 | 2103 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2104 | if (busiest < this_rq) { |
1da177e4 LT |
2105 | spin_unlock(&this_rq->lock); |
2106 | spin_lock(&busiest->lock); | |
2107 | spin_lock(&this_rq->lock); | |
2108 | } else | |
2109 | spin_lock(&busiest->lock); | |
2110 | } | |
2111 | } | |
2112 | ||
1da177e4 LT |
2113 | /* |
2114 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2115 | * This is accomplished by forcing the cpu_allowed mask to only | |
2116 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then | |
2117 | * the cpu_allowed mask is restored. | |
2118 | */ | |
36c8b586 | 2119 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2120 | { |
70b97a7f | 2121 | struct migration_req req; |
1da177e4 | 2122 | unsigned long flags; |
70b97a7f | 2123 | struct rq *rq; |
1da177e4 LT |
2124 | |
2125 | rq = task_rq_lock(p, &flags); | |
2126 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2127 | || unlikely(cpu_is_offline(dest_cpu))) | |
2128 | goto out; | |
2129 | ||
2130 | /* force the process onto the specified CPU */ | |
2131 | if (migrate_task(p, dest_cpu, &req)) { | |
2132 | /* Need to wait for migration thread (might exit: take ref). */ | |
2133 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2134 | |
1da177e4 LT |
2135 | get_task_struct(mt); |
2136 | task_rq_unlock(rq, &flags); | |
2137 | wake_up_process(mt); | |
2138 | put_task_struct(mt); | |
2139 | wait_for_completion(&req.done); | |
36c8b586 | 2140 | |
1da177e4 LT |
2141 | return; |
2142 | } | |
2143 | out: | |
2144 | task_rq_unlock(rq, &flags); | |
2145 | } | |
2146 | ||
2147 | /* | |
476d139c NP |
2148 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2149 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2150 | */ |
2151 | void sched_exec(void) | |
2152 | { | |
1da177e4 | 2153 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2154 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2155 | put_cpu(); |
476d139c NP |
2156 | if (new_cpu != this_cpu) |
2157 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2158 | } |
2159 | ||
2160 | /* | |
2161 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2162 | * Both runqueues must be locked. | |
2163 | */ | |
dd41f596 IM |
2164 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2165 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2166 | { |
2e1cb74a | 2167 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2168 | set_task_cpu(p, this_cpu); |
dd41f596 | 2169 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2170 | /* |
2171 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2172 | * to be always true for them. | |
2173 | */ | |
dd41f596 | 2174 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2175 | } |
2176 | ||
2177 | /* | |
2178 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2179 | */ | |
858119e1 | 2180 | static |
70b97a7f | 2181 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2182 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2183 | int *all_pinned) |
1da177e4 LT |
2184 | { |
2185 | /* | |
2186 | * We do not migrate tasks that are: | |
2187 | * 1) running (obviously), or | |
2188 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2189 | * 3) are cache-hot on their current CPU. | |
2190 | */ | |
cc367732 IM |
2191 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2192 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2193 | return 0; |
cc367732 | 2194 | } |
81026794 NP |
2195 | *all_pinned = 0; |
2196 | ||
cc367732 IM |
2197 | if (task_running(rq, p)) { |
2198 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2199 | return 0; |
cc367732 | 2200 | } |
1da177e4 | 2201 | |
da84d961 IM |
2202 | /* |
2203 | * Aggressive migration if: | |
2204 | * 1) task is cache cold, or | |
2205 | * 2) too many balance attempts have failed. | |
2206 | */ | |
2207 | ||
6bc1665b IM |
2208 | if (!task_hot(p, rq->clock, sd) || |
2209 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2210 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2211 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2212 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2213 | schedstat_inc(p, se.nr_forced_migrations); |
2214 | } | |
da84d961 IM |
2215 | #endif |
2216 | return 1; | |
2217 | } | |
2218 | ||
cc367732 IM |
2219 | if (task_hot(p, rq->clock, sd)) { |
2220 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2221 | return 0; |
cc367732 | 2222 | } |
1da177e4 LT |
2223 | return 1; |
2224 | } | |
2225 | ||
dd41f596 | 2226 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, |
2dd73a4f | 2227 | unsigned long max_nr_move, unsigned long max_load_move, |
d15bcfdb | 2228 | struct sched_domain *sd, enum cpu_idle_type idle, |
dd41f596 | 2229 | int *all_pinned, unsigned long *load_moved, |
a4ac01c3 | 2230 | int *this_best_prio, struct rq_iterator *iterator) |
1da177e4 | 2231 | { |
dd41f596 IM |
2232 | int pulled = 0, pinned = 0, skip_for_load; |
2233 | struct task_struct *p; | |
2234 | long rem_load_move = max_load_move; | |
1da177e4 | 2235 | |
2dd73a4f | 2236 | if (max_nr_move == 0 || max_load_move == 0) |
1da177e4 LT |
2237 | goto out; |
2238 | ||
81026794 NP |
2239 | pinned = 1; |
2240 | ||
1da177e4 | 2241 | /* |
dd41f596 | 2242 | * Start the load-balancing iterator: |
1da177e4 | 2243 | */ |
dd41f596 IM |
2244 | p = iterator->start(iterator->arg); |
2245 | next: | |
2246 | if (!p) | |
1da177e4 | 2247 | goto out; |
50ddd969 PW |
2248 | /* |
2249 | * To help distribute high priority tasks accross CPUs we don't | |
2250 | * skip a task if it will be the highest priority task (i.e. smallest | |
2251 | * prio value) on its new queue regardless of its load weight | |
2252 | */ | |
dd41f596 IM |
2253 | skip_for_load = (p->se.load.weight >> 1) > rem_load_move + |
2254 | SCHED_LOAD_SCALE_FUZZ; | |
a4ac01c3 | 2255 | if ((skip_for_load && p->prio >= *this_best_prio) || |
dd41f596 | 2256 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2257 | p = iterator->next(iterator->arg); |
2258 | goto next; | |
1da177e4 LT |
2259 | } |
2260 | ||
dd41f596 | 2261 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2262 | pulled++; |
dd41f596 | 2263 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2264 | |
2dd73a4f PW |
2265 | /* |
2266 | * We only want to steal up to the prescribed number of tasks | |
2267 | * and the prescribed amount of weighted load. | |
2268 | */ | |
2269 | if (pulled < max_nr_move && rem_load_move > 0) { | |
a4ac01c3 PW |
2270 | if (p->prio < *this_best_prio) |
2271 | *this_best_prio = p->prio; | |
dd41f596 IM |
2272 | p = iterator->next(iterator->arg); |
2273 | goto next; | |
1da177e4 LT |
2274 | } |
2275 | out: | |
2276 | /* | |
2277 | * Right now, this is the only place pull_task() is called, | |
2278 | * so we can safely collect pull_task() stats here rather than | |
2279 | * inside pull_task(). | |
2280 | */ | |
2281 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2282 | |
2283 | if (all_pinned) | |
2284 | *all_pinned = pinned; | |
dd41f596 | 2285 | *load_moved = max_load_move - rem_load_move; |
1da177e4 LT |
2286 | return pulled; |
2287 | } | |
2288 | ||
dd41f596 | 2289 | /* |
43010659 PW |
2290 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2291 | * this_rq, as part of a balancing operation within domain "sd". | |
2292 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2293 | * |
2294 | * Called with both runqueues locked. | |
2295 | */ | |
2296 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2297 | unsigned long max_load_move, |
dd41f596 IM |
2298 | struct sched_domain *sd, enum cpu_idle_type idle, |
2299 | int *all_pinned) | |
2300 | { | |
5522d5d5 | 2301 | const struct sched_class *class = sched_class_highest; |
43010659 | 2302 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2303 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
2304 | |
2305 | do { | |
43010659 PW |
2306 | total_load_moved += |
2307 | class->load_balance(this_rq, this_cpu, busiest, | |
2308 | ULONG_MAX, max_load_move - total_load_moved, | |
a4ac01c3 | 2309 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 2310 | class = class->next; |
43010659 | 2311 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 2312 | |
43010659 PW |
2313 | return total_load_moved > 0; |
2314 | } | |
2315 | ||
2316 | /* | |
2317 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
2318 | * part of active balancing operations within "domain". | |
2319 | * Returns 1 if successful and 0 otherwise. | |
2320 | * | |
2321 | * Called with both runqueues locked. | |
2322 | */ | |
2323 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2324 | struct sched_domain *sd, enum cpu_idle_type idle) | |
2325 | { | |
5522d5d5 | 2326 | const struct sched_class *class; |
a4ac01c3 | 2327 | int this_best_prio = MAX_PRIO; |
43010659 PW |
2328 | |
2329 | for (class = sched_class_highest; class; class = class->next) | |
2330 | if (class->load_balance(this_rq, this_cpu, busiest, | |
a4ac01c3 PW |
2331 | 1, ULONG_MAX, sd, idle, NULL, |
2332 | &this_best_prio)) | |
43010659 PW |
2333 | return 1; |
2334 | ||
2335 | return 0; | |
dd41f596 IM |
2336 | } |
2337 | ||
1da177e4 LT |
2338 | /* |
2339 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2340 | * domain. It calculates and returns the amount of weighted load which |
2341 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2342 | */ |
2343 | static struct sched_group * | |
2344 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 IM |
2345 | unsigned long *imbalance, enum cpu_idle_type idle, |
2346 | int *sd_idle, cpumask_t *cpus, int *balance) | |
1da177e4 LT |
2347 | { |
2348 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2349 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2350 | unsigned long max_pull; |
2dd73a4f PW |
2351 | unsigned long busiest_load_per_task, busiest_nr_running; |
2352 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 2353 | int load_idx, group_imb = 0; |
5c45bf27 SS |
2354 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2355 | int power_savings_balance = 1; | |
2356 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2357 | unsigned long min_nr_running = ULONG_MAX; | |
2358 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2359 | #endif | |
1da177e4 LT |
2360 | |
2361 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2362 | busiest_load_per_task = busiest_nr_running = 0; |
2363 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2364 | if (idle == CPU_NOT_IDLE) |
7897986b | 2365 | load_idx = sd->busy_idx; |
d15bcfdb | 2366 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2367 | load_idx = sd->newidle_idx; |
2368 | else | |
2369 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2370 | |
2371 | do { | |
908a7c1b | 2372 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
2373 | int local_group; |
2374 | int i; | |
908a7c1b | 2375 | int __group_imb = 0; |
783609c6 | 2376 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2377 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2378 | |
2379 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2380 | ||
783609c6 SS |
2381 | if (local_group) |
2382 | balance_cpu = first_cpu(group->cpumask); | |
2383 | ||
1da177e4 | 2384 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2385 | sum_weighted_load = sum_nr_running = avg_load = 0; |
908a7c1b KC |
2386 | max_cpu_load = 0; |
2387 | min_cpu_load = ~0UL; | |
1da177e4 LT |
2388 | |
2389 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2390 | struct rq *rq; |
2391 | ||
2392 | if (!cpu_isset(i, *cpus)) | |
2393 | continue; | |
2394 | ||
2395 | rq = cpu_rq(i); | |
2dd73a4f | 2396 | |
9439aab8 | 2397 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
2398 | *sd_idle = 0; |
2399 | ||
1da177e4 | 2400 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2401 | if (local_group) { |
2402 | if (idle_cpu(i) && !first_idle_cpu) { | |
2403 | first_idle_cpu = 1; | |
2404 | balance_cpu = i; | |
2405 | } | |
2406 | ||
a2000572 | 2407 | load = target_load(i, load_idx); |
908a7c1b | 2408 | } else { |
a2000572 | 2409 | load = source_load(i, load_idx); |
908a7c1b KC |
2410 | if (load > max_cpu_load) |
2411 | max_cpu_load = load; | |
2412 | if (min_cpu_load > load) | |
2413 | min_cpu_load = load; | |
2414 | } | |
1da177e4 LT |
2415 | |
2416 | avg_load += load; | |
2dd73a4f | 2417 | sum_nr_running += rq->nr_running; |
dd41f596 | 2418 | sum_weighted_load += weighted_cpuload(i); |
1da177e4 LT |
2419 | } |
2420 | ||
783609c6 SS |
2421 | /* |
2422 | * First idle cpu or the first cpu(busiest) in this sched group | |
2423 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
2424 | * domains. In the newly idle case, we will allow all the cpu's |
2425 | * to do the newly idle load balance. | |
783609c6 | 2426 | */ |
9439aab8 SS |
2427 | if (idle != CPU_NEWLY_IDLE && local_group && |
2428 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
2429 | *balance = 0; |
2430 | goto ret; | |
2431 | } | |
2432 | ||
1da177e4 | 2433 | total_load += avg_load; |
5517d86b | 2434 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2435 | |
2436 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2437 | avg_load = sg_div_cpu_power(group, |
2438 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2439 | |
908a7c1b KC |
2440 | if ((max_cpu_load - min_cpu_load) > SCHED_LOAD_SCALE) |
2441 | __group_imb = 1; | |
2442 | ||
5517d86b | 2443 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2444 | |
1da177e4 LT |
2445 | if (local_group) { |
2446 | this_load = avg_load; | |
2447 | this = group; | |
2dd73a4f PW |
2448 | this_nr_running = sum_nr_running; |
2449 | this_load_per_task = sum_weighted_load; | |
2450 | } else if (avg_load > max_load && | |
908a7c1b | 2451 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
2452 | max_load = avg_load; |
2453 | busiest = group; | |
2dd73a4f PW |
2454 | busiest_nr_running = sum_nr_running; |
2455 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 2456 | group_imb = __group_imb; |
1da177e4 | 2457 | } |
5c45bf27 SS |
2458 | |
2459 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2460 | /* | |
2461 | * Busy processors will not participate in power savings | |
2462 | * balance. | |
2463 | */ | |
dd41f596 IM |
2464 | if (idle == CPU_NOT_IDLE || |
2465 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
2466 | goto group_next; | |
5c45bf27 SS |
2467 | |
2468 | /* | |
2469 | * If the local group is idle or completely loaded | |
2470 | * no need to do power savings balance at this domain | |
2471 | */ | |
2472 | if (local_group && (this_nr_running >= group_capacity || | |
2473 | !this_nr_running)) | |
2474 | power_savings_balance = 0; | |
2475 | ||
dd41f596 | 2476 | /* |
5c45bf27 SS |
2477 | * If a group is already running at full capacity or idle, |
2478 | * don't include that group in power savings calculations | |
dd41f596 IM |
2479 | */ |
2480 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 2481 | || !sum_nr_running) |
dd41f596 | 2482 | goto group_next; |
5c45bf27 | 2483 | |
dd41f596 | 2484 | /* |
5c45bf27 | 2485 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
2486 | * This is the group from where we need to pick up the load |
2487 | * for saving power | |
2488 | */ | |
2489 | if ((sum_nr_running < min_nr_running) || | |
2490 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
2491 | first_cpu(group->cpumask) < |
2492 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
2493 | group_min = group; |
2494 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
2495 | min_load_per_task = sum_weighted_load / |
2496 | sum_nr_running; | |
dd41f596 | 2497 | } |
5c45bf27 | 2498 | |
dd41f596 | 2499 | /* |
5c45bf27 | 2500 | * Calculate the group which is almost near its |
dd41f596 IM |
2501 | * capacity but still has some space to pick up some load |
2502 | * from other group and save more power | |
2503 | */ | |
2504 | if (sum_nr_running <= group_capacity - 1) { | |
2505 | if (sum_nr_running > leader_nr_running || | |
2506 | (sum_nr_running == leader_nr_running && | |
2507 | first_cpu(group->cpumask) > | |
2508 | first_cpu(group_leader->cpumask))) { | |
2509 | group_leader = group; | |
2510 | leader_nr_running = sum_nr_running; | |
2511 | } | |
48f24c4d | 2512 | } |
5c45bf27 SS |
2513 | group_next: |
2514 | #endif | |
1da177e4 LT |
2515 | group = group->next; |
2516 | } while (group != sd->groups); | |
2517 | ||
2dd73a4f | 2518 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
2519 | goto out_balanced; |
2520 | ||
2521 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
2522 | ||
2523 | if (this_load >= avg_load || | |
2524 | 100*max_load <= sd->imbalance_pct*this_load) | |
2525 | goto out_balanced; | |
2526 | ||
2dd73a4f | 2527 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
2528 | if (group_imb) |
2529 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
2530 | ||
1da177e4 LT |
2531 | /* |
2532 | * We're trying to get all the cpus to the average_load, so we don't | |
2533 | * want to push ourselves above the average load, nor do we wish to | |
2534 | * reduce the max loaded cpu below the average load, as either of these | |
2535 | * actions would just result in more rebalancing later, and ping-pong | |
2536 | * tasks around. Thus we look for the minimum possible imbalance. | |
2537 | * Negative imbalances (*we* are more loaded than anyone else) will | |
2538 | * be counted as no imbalance for these purposes -- we can't fix that | |
2539 | * by pulling tasks to us. Be careful of negative numbers as they'll | |
2540 | * appear as very large values with unsigned longs. | |
2541 | */ | |
2dd73a4f PW |
2542 | if (max_load <= busiest_load_per_task) |
2543 | goto out_balanced; | |
2544 | ||
2545 | /* | |
2546 | * In the presence of smp nice balancing, certain scenarios can have | |
2547 | * max load less than avg load(as we skip the groups at or below | |
2548 | * its cpu_power, while calculating max_load..) | |
2549 | */ | |
2550 | if (max_load < avg_load) { | |
2551 | *imbalance = 0; | |
2552 | goto small_imbalance; | |
2553 | } | |
0c117f1b SS |
2554 | |
2555 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 2556 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 2557 | |
1da177e4 | 2558 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
2559 | *imbalance = min(max_pull * busiest->__cpu_power, |
2560 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
2561 | / SCHED_LOAD_SCALE; |
2562 | ||
2dd73a4f PW |
2563 | /* |
2564 | * if *imbalance is less than the average load per runnable task | |
2565 | * there is no gaurantee that any tasks will be moved so we'll have | |
2566 | * a think about bumping its value to force at least one task to be | |
2567 | * moved | |
2568 | */ | |
7fd0d2dd | 2569 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 2570 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
2571 | unsigned int imbn; |
2572 | ||
2573 | small_imbalance: | |
2574 | pwr_move = pwr_now = 0; | |
2575 | imbn = 2; | |
2576 | if (this_nr_running) { | |
2577 | this_load_per_task /= this_nr_running; | |
2578 | if (busiest_load_per_task > this_load_per_task) | |
2579 | imbn = 1; | |
2580 | } else | |
2581 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 2582 | |
dd41f596 IM |
2583 | if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >= |
2584 | busiest_load_per_task * imbn) { | |
2dd73a4f | 2585 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2586 | return busiest; |
2587 | } | |
2588 | ||
2589 | /* | |
2590 | * OK, we don't have enough imbalance to justify moving tasks, | |
2591 | * however we may be able to increase total CPU power used by | |
2592 | * moving them. | |
2593 | */ | |
2594 | ||
5517d86b ED |
2595 | pwr_now += busiest->__cpu_power * |
2596 | min(busiest_load_per_task, max_load); | |
2597 | pwr_now += this->__cpu_power * | |
2598 | min(this_load_per_task, this_load); | |
1da177e4 LT |
2599 | pwr_now /= SCHED_LOAD_SCALE; |
2600 | ||
2601 | /* Amount of load we'd subtract */ | |
5517d86b ED |
2602 | tmp = sg_div_cpu_power(busiest, |
2603 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 2604 | if (max_load > tmp) |
5517d86b | 2605 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 2606 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
2607 | |
2608 | /* Amount of load we'd add */ | |
5517d86b | 2609 | if (max_load * busiest->__cpu_power < |
33859f7f | 2610 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
2611 | tmp = sg_div_cpu_power(this, |
2612 | max_load * busiest->__cpu_power); | |
1da177e4 | 2613 | else |
5517d86b ED |
2614 | tmp = sg_div_cpu_power(this, |
2615 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
2616 | pwr_move += this->__cpu_power * | |
2617 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
2618 | pwr_move /= SCHED_LOAD_SCALE; |
2619 | ||
2620 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
2621 | if (pwr_move > pwr_now) |
2622 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
2623 | } |
2624 | ||
1da177e4 LT |
2625 | return busiest; |
2626 | ||
2627 | out_balanced: | |
5c45bf27 | 2628 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 2629 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 2630 | goto ret; |
1da177e4 | 2631 | |
5c45bf27 SS |
2632 | if (this == group_leader && group_leader != group_min) { |
2633 | *imbalance = min_load_per_task; | |
2634 | return group_min; | |
2635 | } | |
5c45bf27 | 2636 | #endif |
783609c6 | 2637 | ret: |
1da177e4 LT |
2638 | *imbalance = 0; |
2639 | return NULL; | |
2640 | } | |
2641 | ||
2642 | /* | |
2643 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
2644 | */ | |
70b97a7f | 2645 | static struct rq * |
d15bcfdb | 2646 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
0a2966b4 | 2647 | unsigned long imbalance, cpumask_t *cpus) |
1da177e4 | 2648 | { |
70b97a7f | 2649 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 2650 | unsigned long max_load = 0; |
1da177e4 LT |
2651 | int i; |
2652 | ||
2653 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 2654 | unsigned long wl; |
0a2966b4 CL |
2655 | |
2656 | if (!cpu_isset(i, *cpus)) | |
2657 | continue; | |
2658 | ||
48f24c4d | 2659 | rq = cpu_rq(i); |
dd41f596 | 2660 | wl = weighted_cpuload(i); |
2dd73a4f | 2661 | |
dd41f596 | 2662 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 2663 | continue; |
1da177e4 | 2664 | |
dd41f596 IM |
2665 | if (wl > max_load) { |
2666 | max_load = wl; | |
48f24c4d | 2667 | busiest = rq; |
1da177e4 LT |
2668 | } |
2669 | } | |
2670 | ||
2671 | return busiest; | |
2672 | } | |
2673 | ||
77391d71 NP |
2674 | /* |
2675 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
2676 | * so long as it is large enough. | |
2677 | */ | |
2678 | #define MAX_PINNED_INTERVAL 512 | |
2679 | ||
1da177e4 LT |
2680 | /* |
2681 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2682 | * tasks if there is an imbalance. | |
1da177e4 | 2683 | */ |
70b97a7f | 2684 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 2685 | struct sched_domain *sd, enum cpu_idle_type idle, |
783609c6 | 2686 | int *balance) |
1da177e4 | 2687 | { |
43010659 | 2688 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 2689 | struct sched_group *group; |
1da177e4 | 2690 | unsigned long imbalance; |
70b97a7f | 2691 | struct rq *busiest; |
0a2966b4 | 2692 | cpumask_t cpus = CPU_MASK_ALL; |
fe2eea3f | 2693 | unsigned long flags; |
5969fe06 | 2694 | |
89c4710e SS |
2695 | /* |
2696 | * When power savings policy is enabled for the parent domain, idle | |
2697 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 2698 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 2699 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 2700 | */ |
d15bcfdb | 2701 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2702 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2703 | sd_idle = 1; |
1da177e4 | 2704 | |
2d72376b | 2705 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 2706 | |
0a2966b4 CL |
2707 | redo: |
2708 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
783609c6 SS |
2709 | &cpus, balance); |
2710 | ||
06066714 | 2711 | if (*balance == 0) |
783609c6 | 2712 | goto out_balanced; |
783609c6 | 2713 | |
1da177e4 LT |
2714 | if (!group) { |
2715 | schedstat_inc(sd, lb_nobusyg[idle]); | |
2716 | goto out_balanced; | |
2717 | } | |
2718 | ||
0a2966b4 | 2719 | busiest = find_busiest_queue(group, idle, imbalance, &cpus); |
1da177e4 LT |
2720 | if (!busiest) { |
2721 | schedstat_inc(sd, lb_nobusyq[idle]); | |
2722 | goto out_balanced; | |
2723 | } | |
2724 | ||
db935dbd | 2725 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
2726 | |
2727 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
2728 | ||
43010659 | 2729 | ld_moved = 0; |
1da177e4 LT |
2730 | if (busiest->nr_running > 1) { |
2731 | /* | |
2732 | * Attempt to move tasks. If find_busiest_group has found | |
2733 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 2734 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
2735 | * correctly treated as an imbalance. |
2736 | */ | |
fe2eea3f | 2737 | local_irq_save(flags); |
e17224bf | 2738 | double_rq_lock(this_rq, busiest); |
43010659 | 2739 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 2740 | imbalance, sd, idle, &all_pinned); |
e17224bf | 2741 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 2742 | local_irq_restore(flags); |
81026794 | 2743 | |
46cb4b7c SS |
2744 | /* |
2745 | * some other cpu did the load balance for us. | |
2746 | */ | |
43010659 | 2747 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
2748 | resched_cpu(this_cpu); |
2749 | ||
81026794 | 2750 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 CL |
2751 | if (unlikely(all_pinned)) { |
2752 | cpu_clear(cpu_of(busiest), cpus); | |
2753 | if (!cpus_empty(cpus)) | |
2754 | goto redo; | |
81026794 | 2755 | goto out_balanced; |
0a2966b4 | 2756 | } |
1da177e4 | 2757 | } |
81026794 | 2758 | |
43010659 | 2759 | if (!ld_moved) { |
1da177e4 LT |
2760 | schedstat_inc(sd, lb_failed[idle]); |
2761 | sd->nr_balance_failed++; | |
2762 | ||
2763 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 2764 | |
fe2eea3f | 2765 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
2766 | |
2767 | /* don't kick the migration_thread, if the curr | |
2768 | * task on busiest cpu can't be moved to this_cpu | |
2769 | */ | |
2770 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 2771 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
2772 | all_pinned = 1; |
2773 | goto out_one_pinned; | |
2774 | } | |
2775 | ||
1da177e4 LT |
2776 | if (!busiest->active_balance) { |
2777 | busiest->active_balance = 1; | |
2778 | busiest->push_cpu = this_cpu; | |
81026794 | 2779 | active_balance = 1; |
1da177e4 | 2780 | } |
fe2eea3f | 2781 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 2782 | if (active_balance) |
1da177e4 LT |
2783 | wake_up_process(busiest->migration_thread); |
2784 | ||
2785 | /* | |
2786 | * We've kicked active balancing, reset the failure | |
2787 | * counter. | |
2788 | */ | |
39507451 | 2789 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 2790 | } |
81026794 | 2791 | } else |
1da177e4 LT |
2792 | sd->nr_balance_failed = 0; |
2793 | ||
81026794 | 2794 | if (likely(!active_balance)) { |
1da177e4 LT |
2795 | /* We were unbalanced, so reset the balancing interval */ |
2796 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
2797 | } else { |
2798 | /* | |
2799 | * If we've begun active balancing, start to back off. This | |
2800 | * case may not be covered by the all_pinned logic if there | |
2801 | * is only 1 task on the busy runqueue (because we don't call | |
2802 | * move_tasks). | |
2803 | */ | |
2804 | if (sd->balance_interval < sd->max_interval) | |
2805 | sd->balance_interval *= 2; | |
1da177e4 LT |
2806 | } |
2807 | ||
43010659 | 2808 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2809 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2810 | return -1; |
43010659 | 2811 | return ld_moved; |
1da177e4 LT |
2812 | |
2813 | out_balanced: | |
1da177e4 LT |
2814 | schedstat_inc(sd, lb_balanced[idle]); |
2815 | ||
16cfb1c0 | 2816 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
2817 | |
2818 | out_one_pinned: | |
1da177e4 | 2819 | /* tune up the balancing interval */ |
77391d71 NP |
2820 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
2821 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
2822 | sd->balance_interval *= 2; |
2823 | ||
48f24c4d | 2824 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2825 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2826 | return -1; |
1da177e4 LT |
2827 | return 0; |
2828 | } | |
2829 | ||
2830 | /* | |
2831 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2832 | * tasks if there is an imbalance. | |
2833 | * | |
d15bcfdb | 2834 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
2835 | * this_rq is locked. |
2836 | */ | |
48f24c4d | 2837 | static int |
70b97a7f | 2838 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
2839 | { |
2840 | struct sched_group *group; | |
70b97a7f | 2841 | struct rq *busiest = NULL; |
1da177e4 | 2842 | unsigned long imbalance; |
43010659 | 2843 | int ld_moved = 0; |
5969fe06 | 2844 | int sd_idle = 0; |
969bb4e4 | 2845 | int all_pinned = 0; |
0a2966b4 | 2846 | cpumask_t cpus = CPU_MASK_ALL; |
5969fe06 | 2847 | |
89c4710e SS |
2848 | /* |
2849 | * When power savings policy is enabled for the parent domain, idle | |
2850 | * sibling can pick up load irrespective of busy siblings. In this case, | |
2851 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 2852 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
2853 | */ |
2854 | if (sd->flags & SD_SHARE_CPUPOWER && | |
2855 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 2856 | sd_idle = 1; |
1da177e4 | 2857 | |
2d72376b | 2858 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 2859 | redo: |
d15bcfdb | 2860 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
783609c6 | 2861 | &sd_idle, &cpus, NULL); |
1da177e4 | 2862 | if (!group) { |
d15bcfdb | 2863 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2864 | goto out_balanced; |
1da177e4 LT |
2865 | } |
2866 | ||
d15bcfdb | 2867 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, |
0a2966b4 | 2868 | &cpus); |
db935dbd | 2869 | if (!busiest) { |
d15bcfdb | 2870 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2871 | goto out_balanced; |
1da177e4 LT |
2872 | } |
2873 | ||
db935dbd NP |
2874 | BUG_ON(busiest == this_rq); |
2875 | ||
d15bcfdb | 2876 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 2877 | |
43010659 | 2878 | ld_moved = 0; |
d6d5cfaf NP |
2879 | if (busiest->nr_running > 1) { |
2880 | /* Attempt to move tasks */ | |
2881 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
2882 | /* this_rq->clock is already updated */ |
2883 | update_rq_clock(busiest); | |
43010659 | 2884 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
2885 | imbalance, sd, CPU_NEWLY_IDLE, |
2886 | &all_pinned); | |
d6d5cfaf | 2887 | spin_unlock(&busiest->lock); |
0a2966b4 | 2888 | |
969bb4e4 | 2889 | if (unlikely(all_pinned)) { |
0a2966b4 CL |
2890 | cpu_clear(cpu_of(busiest), cpus); |
2891 | if (!cpus_empty(cpus)) | |
2892 | goto redo; | |
2893 | } | |
d6d5cfaf NP |
2894 | } |
2895 | ||
43010659 | 2896 | if (!ld_moved) { |
d15bcfdb | 2897 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
2898 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
2899 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
2900 | return -1; |
2901 | } else | |
16cfb1c0 | 2902 | sd->nr_balance_failed = 0; |
1da177e4 | 2903 | |
43010659 | 2904 | return ld_moved; |
16cfb1c0 NP |
2905 | |
2906 | out_balanced: | |
d15bcfdb | 2907 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 2908 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2909 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2910 | return -1; |
16cfb1c0 | 2911 | sd->nr_balance_failed = 0; |
48f24c4d | 2912 | |
16cfb1c0 | 2913 | return 0; |
1da177e4 LT |
2914 | } |
2915 | ||
2916 | /* | |
2917 | * idle_balance is called by schedule() if this_cpu is about to become | |
2918 | * idle. Attempts to pull tasks from other CPUs. | |
2919 | */ | |
70b97a7f | 2920 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
2921 | { |
2922 | struct sched_domain *sd; | |
dd41f596 IM |
2923 | int pulled_task = -1; |
2924 | unsigned long next_balance = jiffies + HZ; | |
1da177e4 LT |
2925 | |
2926 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
2927 | unsigned long interval; |
2928 | ||
2929 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
2930 | continue; | |
2931 | ||
2932 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 2933 | /* If we've pulled tasks over stop searching: */ |
1bd77f2d | 2934 | pulled_task = load_balance_newidle(this_cpu, |
92c4ca5c CL |
2935 | this_rq, sd); |
2936 | ||
2937 | interval = msecs_to_jiffies(sd->balance_interval); | |
2938 | if (time_after(next_balance, sd->last_balance + interval)) | |
2939 | next_balance = sd->last_balance + interval; | |
2940 | if (pulled_task) | |
2941 | break; | |
1da177e4 | 2942 | } |
dd41f596 | 2943 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
2944 | /* |
2945 | * We are going idle. next_balance may be set based on | |
2946 | * a busy processor. So reset next_balance. | |
2947 | */ | |
2948 | this_rq->next_balance = next_balance; | |
dd41f596 | 2949 | } |
1da177e4 LT |
2950 | } |
2951 | ||
2952 | /* | |
2953 | * active_load_balance is run by migration threads. It pushes running tasks | |
2954 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
2955 | * running on each physical CPU where possible, and avoids physical / | |
2956 | * logical imbalances. | |
2957 | * | |
2958 | * Called with busiest_rq locked. | |
2959 | */ | |
70b97a7f | 2960 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 2961 | { |
39507451 | 2962 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
2963 | struct sched_domain *sd; |
2964 | struct rq *target_rq; | |
39507451 | 2965 | |
48f24c4d | 2966 | /* Is there any task to move? */ |
39507451 | 2967 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
2968 | return; |
2969 | ||
2970 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
2971 | |
2972 | /* | |
39507451 NP |
2973 | * This condition is "impossible", if it occurs |
2974 | * we need to fix it. Originally reported by | |
2975 | * Bjorn Helgaas on a 128-cpu setup. | |
1da177e4 | 2976 | */ |
39507451 | 2977 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 2978 | |
39507451 NP |
2979 | /* move a task from busiest_rq to target_rq */ |
2980 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
2981 | update_rq_clock(busiest_rq); |
2982 | update_rq_clock(target_rq); | |
39507451 NP |
2983 | |
2984 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 2985 | for_each_domain(target_cpu, sd) { |
39507451 | 2986 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 2987 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 2988 | break; |
c96d145e | 2989 | } |
39507451 | 2990 | |
48f24c4d | 2991 | if (likely(sd)) { |
2d72376b | 2992 | schedstat_inc(sd, alb_count); |
39507451 | 2993 | |
43010659 PW |
2994 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
2995 | sd, CPU_IDLE)) | |
48f24c4d IM |
2996 | schedstat_inc(sd, alb_pushed); |
2997 | else | |
2998 | schedstat_inc(sd, alb_failed); | |
2999 | } | |
39507451 | 3000 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
3001 | } |
3002 | ||
46cb4b7c SS |
3003 | #ifdef CONFIG_NO_HZ |
3004 | static struct { | |
3005 | atomic_t load_balancer; | |
3006 | cpumask_t cpu_mask; | |
3007 | } nohz ____cacheline_aligned = { | |
3008 | .load_balancer = ATOMIC_INIT(-1), | |
3009 | .cpu_mask = CPU_MASK_NONE, | |
3010 | }; | |
3011 | ||
7835b98b | 3012 | /* |
46cb4b7c SS |
3013 | * This routine will try to nominate the ilb (idle load balancing) |
3014 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3015 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3016 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3017 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3018 | * arrives... | |
3019 | * | |
3020 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3021 | * for idle load balancing. ilb owner will still be part of | |
3022 | * nohz.cpu_mask.. | |
7835b98b | 3023 | * |
46cb4b7c SS |
3024 | * While stopping the tick, this cpu will become the ilb owner if there |
3025 | * is no other owner. And will be the owner till that cpu becomes busy | |
3026 | * or if all cpus in the system stop their ticks at which point | |
3027 | * there is no need for ilb owner. | |
3028 | * | |
3029 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3030 | * next busy scheduler_tick() | |
3031 | */ | |
3032 | int select_nohz_load_balancer(int stop_tick) | |
3033 | { | |
3034 | int cpu = smp_processor_id(); | |
3035 | ||
3036 | if (stop_tick) { | |
3037 | cpu_set(cpu, nohz.cpu_mask); | |
3038 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3039 | ||
3040 | /* | |
3041 | * If we are going offline and still the leader, give up! | |
3042 | */ | |
3043 | if (cpu_is_offline(cpu) && | |
3044 | atomic_read(&nohz.load_balancer) == cpu) { | |
3045 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3046 | BUG(); | |
3047 | return 0; | |
3048 | } | |
3049 | ||
3050 | /* time for ilb owner also to sleep */ | |
3051 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3052 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3053 | atomic_set(&nohz.load_balancer, -1); | |
3054 | return 0; | |
3055 | } | |
3056 | ||
3057 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3058 | /* make me the ilb owner */ | |
3059 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3060 | return 1; | |
3061 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3062 | return 1; | |
3063 | } else { | |
3064 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3065 | return 0; | |
3066 | ||
3067 | cpu_clear(cpu, nohz.cpu_mask); | |
3068 | ||
3069 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3070 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3071 | BUG(); | |
3072 | } | |
3073 | return 0; | |
3074 | } | |
3075 | #endif | |
3076 | ||
3077 | static DEFINE_SPINLOCK(balancing); | |
3078 | ||
3079 | /* | |
7835b98b CL |
3080 | * It checks each scheduling domain to see if it is due to be balanced, |
3081 | * and initiates a balancing operation if so. | |
3082 | * | |
3083 | * Balancing parameters are set up in arch_init_sched_domains. | |
3084 | */ | |
a9957449 | 3085 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3086 | { |
46cb4b7c SS |
3087 | int balance = 1; |
3088 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3089 | unsigned long interval; |
3090 | struct sched_domain *sd; | |
46cb4b7c | 3091 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3092 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3093 | int update_next_balance = 0; |
1da177e4 | 3094 | |
46cb4b7c | 3095 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3096 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3097 | continue; | |
3098 | ||
3099 | interval = sd->balance_interval; | |
d15bcfdb | 3100 | if (idle != CPU_IDLE) |
1da177e4 LT |
3101 | interval *= sd->busy_factor; |
3102 | ||
3103 | /* scale ms to jiffies */ | |
3104 | interval = msecs_to_jiffies(interval); | |
3105 | if (unlikely(!interval)) | |
3106 | interval = 1; | |
dd41f596 IM |
3107 | if (interval > HZ*NR_CPUS/10) |
3108 | interval = HZ*NR_CPUS/10; | |
3109 | ||
1da177e4 | 3110 | |
08c183f3 CL |
3111 | if (sd->flags & SD_SERIALIZE) { |
3112 | if (!spin_trylock(&balancing)) | |
3113 | goto out; | |
3114 | } | |
3115 | ||
c9819f45 | 3116 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
46cb4b7c | 3117 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
3118 | /* |
3119 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3120 | * longer idle, or one of our SMT siblings is |
3121 | * not idle. | |
3122 | */ | |
d15bcfdb | 3123 | idle = CPU_NOT_IDLE; |
1da177e4 | 3124 | } |
1bd77f2d | 3125 | sd->last_balance = jiffies; |
1da177e4 | 3126 | } |
08c183f3 CL |
3127 | if (sd->flags & SD_SERIALIZE) |
3128 | spin_unlock(&balancing); | |
3129 | out: | |
f549da84 | 3130 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3131 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3132 | update_next_balance = 1; |
3133 | } | |
783609c6 SS |
3134 | |
3135 | /* | |
3136 | * Stop the load balance at this level. There is another | |
3137 | * CPU in our sched group which is doing load balancing more | |
3138 | * actively. | |
3139 | */ | |
3140 | if (!balance) | |
3141 | break; | |
1da177e4 | 3142 | } |
f549da84 SS |
3143 | |
3144 | /* | |
3145 | * next_balance will be updated only when there is a need. | |
3146 | * When the cpu is attached to null domain for ex, it will not be | |
3147 | * updated. | |
3148 | */ | |
3149 | if (likely(update_next_balance)) | |
3150 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3151 | } |
3152 | ||
3153 | /* | |
3154 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3155 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3156 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3157 | */ | |
3158 | static void run_rebalance_domains(struct softirq_action *h) | |
3159 | { | |
dd41f596 IM |
3160 | int this_cpu = smp_processor_id(); |
3161 | struct rq *this_rq = cpu_rq(this_cpu); | |
3162 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3163 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3164 | |
dd41f596 | 3165 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3166 | |
3167 | #ifdef CONFIG_NO_HZ | |
3168 | /* | |
3169 | * If this cpu is the owner for idle load balancing, then do the | |
3170 | * balancing on behalf of the other idle cpus whose ticks are | |
3171 | * stopped. | |
3172 | */ | |
dd41f596 IM |
3173 | if (this_rq->idle_at_tick && |
3174 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3175 | cpumask_t cpus = nohz.cpu_mask; |
3176 | struct rq *rq; | |
3177 | int balance_cpu; | |
3178 | ||
dd41f596 | 3179 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3180 | for_each_cpu_mask(balance_cpu, cpus) { |
3181 | /* | |
3182 | * If this cpu gets work to do, stop the load balancing | |
3183 | * work being done for other cpus. Next load | |
3184 | * balancing owner will pick it up. | |
3185 | */ | |
3186 | if (need_resched()) | |
3187 | break; | |
3188 | ||
de0cf899 | 3189 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3190 | |
3191 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3192 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3193 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3194 | } |
3195 | } | |
3196 | #endif | |
3197 | } | |
3198 | ||
3199 | /* | |
3200 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3201 | * | |
3202 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3203 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3204 | * if the whole system is idle. | |
3205 | */ | |
dd41f596 | 3206 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3207 | { |
46cb4b7c SS |
3208 | #ifdef CONFIG_NO_HZ |
3209 | /* | |
3210 | * If we were in the nohz mode recently and busy at the current | |
3211 | * scheduler tick, then check if we need to nominate new idle | |
3212 | * load balancer. | |
3213 | */ | |
3214 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3215 | rq->in_nohz_recently = 0; | |
3216 | ||
3217 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3218 | cpu_clear(cpu, nohz.cpu_mask); | |
3219 | atomic_set(&nohz.load_balancer, -1); | |
3220 | } | |
3221 | ||
3222 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3223 | /* | |
3224 | * simple selection for now: Nominate the | |
3225 | * first cpu in the nohz list to be the next | |
3226 | * ilb owner. | |
3227 | * | |
3228 | * TBD: Traverse the sched domains and nominate | |
3229 | * the nearest cpu in the nohz.cpu_mask. | |
3230 | */ | |
3231 | int ilb = first_cpu(nohz.cpu_mask); | |
3232 | ||
3233 | if (ilb != NR_CPUS) | |
3234 | resched_cpu(ilb); | |
3235 | } | |
3236 | } | |
3237 | ||
3238 | /* | |
3239 | * If this cpu is idle and doing idle load balancing for all the | |
3240 | * cpus with ticks stopped, is it time for that to stop? | |
3241 | */ | |
3242 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3243 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3244 | resched_cpu(cpu); | |
3245 | return; | |
3246 | } | |
3247 | ||
3248 | /* | |
3249 | * If this cpu is idle and the idle load balancing is done by | |
3250 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3251 | */ | |
3252 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3253 | cpu_isset(cpu, nohz.cpu_mask)) | |
3254 | return; | |
3255 | #endif | |
3256 | if (time_after_eq(jiffies, rq->next_balance)) | |
3257 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3258 | } |
dd41f596 IM |
3259 | |
3260 | #else /* CONFIG_SMP */ | |
3261 | ||
1da177e4 LT |
3262 | /* |
3263 | * on UP we do not need to balance between CPUs: | |
3264 | */ | |
70b97a7f | 3265 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3266 | { |
3267 | } | |
dd41f596 IM |
3268 | |
3269 | /* Avoid "used but not defined" warning on UP */ | |
3270 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3271 | unsigned long max_nr_move, unsigned long max_load_move, | |
3272 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3273 | int *all_pinned, unsigned long *load_moved, | |
a4ac01c3 | 3274 | int *this_best_prio, struct rq_iterator *iterator) |
dd41f596 IM |
3275 | { |
3276 | *load_moved = 0; | |
3277 | ||
3278 | return 0; | |
3279 | } | |
3280 | ||
1da177e4 LT |
3281 | #endif |
3282 | ||
1da177e4 LT |
3283 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3284 | ||
3285 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3286 | ||
3287 | /* | |
41b86e9c IM |
3288 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
3289 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 3290 | */ |
41b86e9c | 3291 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 3292 | { |
1da177e4 | 3293 | unsigned long flags; |
41b86e9c IM |
3294 | u64 ns, delta_exec; |
3295 | struct rq *rq; | |
48f24c4d | 3296 | |
41b86e9c IM |
3297 | rq = task_rq_lock(p, &flags); |
3298 | ns = p->se.sum_exec_runtime; | |
3299 | if (rq->curr == p) { | |
a8e504d2 IM |
3300 | update_rq_clock(rq); |
3301 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
3302 | if ((s64)delta_exec > 0) |
3303 | ns += delta_exec; | |
3304 | } | |
3305 | task_rq_unlock(rq, &flags); | |
48f24c4d | 3306 | |
1da177e4 LT |
3307 | return ns; |
3308 | } | |
3309 | ||
1da177e4 LT |
3310 | /* |
3311 | * Account user cpu time to a process. | |
3312 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
3313 | * @cputime: the cpu time spent in user space since the last update |
3314 | */ | |
3315 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3316 | { | |
3317 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3318 | cputime64_t tmp; | |
62d0df64 | 3319 | struct rq *rq = this_rq(); |
1da177e4 LT |
3320 | |
3321 | p->utime = cputime_add(p->utime, cputime); | |
3322 | ||
62d0df64 PM |
3323 | if (p != rq->idle) |
3324 | cpuacct_charge(p, cputime); | |
3325 | ||
1da177e4 LT |
3326 | /* Add user time to cpustat. */ |
3327 | tmp = cputime_to_cputime64(cputime); | |
3328 | if (TASK_NICE(p) > 0) | |
3329 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3330 | else | |
3331 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3332 | } | |
3333 | ||
94886b84 LV |
3334 | /* |
3335 | * Account guest cpu time to a process. | |
3336 | * @p: the process that the cpu time gets accounted to | |
3337 | * @cputime: the cpu time spent in virtual machine since the last update | |
3338 | */ | |
3339 | void account_guest_time(struct task_struct *p, cputime_t cputime) | |
3340 | { | |
3341 | cputime64_t tmp; | |
3342 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3343 | ||
3344 | tmp = cputime_to_cputime64(cputime); | |
3345 | ||
3346 | p->utime = cputime_add(p->utime, cputime); | |
3347 | p->gtime = cputime_add(p->gtime, cputime); | |
3348 | ||
3349 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3350 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3351 | } | |
3352 | ||
c66f08be MN |
3353 | /* |
3354 | * Account scaled user cpu time to a process. | |
3355 | * @p: the process that the cpu time gets accounted to | |
3356 | * @cputime: the cpu time spent in user space since the last update | |
3357 | */ | |
3358 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
3359 | { | |
3360 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
3361 | } | |
3362 | ||
1da177e4 LT |
3363 | /* |
3364 | * Account system cpu time to a process. | |
3365 | * @p: the process that the cpu time gets accounted to | |
3366 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3367 | * @cputime: the cpu time spent in kernel space since the last update | |
3368 | */ | |
3369 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3370 | cputime_t cputime) | |
3371 | { | |
3372 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3373 | struct rq *rq = this_rq(); |
1da177e4 LT |
3374 | cputime64_t tmp; |
3375 | ||
94886b84 LV |
3376 | if (p->flags & PF_VCPU) { |
3377 | account_guest_time(p, cputime); | |
94886b84 LV |
3378 | return; |
3379 | } | |
3380 | ||
1da177e4 LT |
3381 | p->stime = cputime_add(p->stime, cputime); |
3382 | ||
3383 | /* Add system time to cpustat. */ | |
3384 | tmp = cputime_to_cputime64(cputime); | |
3385 | if (hardirq_count() - hardirq_offset) | |
3386 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3387 | else if (softirq_count()) | |
3388 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
62d0df64 | 3389 | else if (p != rq->idle) { |
1da177e4 | 3390 | cpustat->system = cputime64_add(cpustat->system, tmp); |
62d0df64 PM |
3391 | cpuacct_charge(p, cputime); |
3392 | } else if (atomic_read(&rq->nr_iowait) > 0) | |
1da177e4 LT |
3393 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
3394 | else | |
3395 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3396 | /* Account for system time used */ | |
3397 | acct_update_integrals(p); | |
1da177e4 LT |
3398 | } |
3399 | ||
c66f08be MN |
3400 | /* |
3401 | * Account scaled system cpu time to a process. | |
3402 | * @p: the process that the cpu time gets accounted to | |
3403 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3404 | * @cputime: the cpu time spent in kernel space since the last update | |
3405 | */ | |
3406 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
3407 | { | |
3408 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
3409 | } | |
3410 | ||
1da177e4 LT |
3411 | /* |
3412 | * Account for involuntary wait time. | |
3413 | * @p: the process from which the cpu time has been stolen | |
3414 | * @steal: the cpu time spent in involuntary wait | |
3415 | */ | |
3416 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3417 | { | |
3418 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3419 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3420 | struct rq *rq = this_rq(); |
1da177e4 LT |
3421 | |
3422 | if (p == rq->idle) { | |
3423 | p->stime = cputime_add(p->stime, steal); | |
3424 | if (atomic_read(&rq->nr_iowait) > 0) | |
3425 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3426 | else | |
3427 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
62d0df64 | 3428 | } else { |
1da177e4 | 3429 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
62d0df64 PM |
3430 | cpuacct_charge(p, -tmp); |
3431 | } | |
1da177e4 LT |
3432 | } |
3433 | ||
7835b98b CL |
3434 | /* |
3435 | * This function gets called by the timer code, with HZ frequency. | |
3436 | * We call it with interrupts disabled. | |
3437 | * | |
3438 | * It also gets called by the fork code, when changing the parent's | |
3439 | * timeslices. | |
3440 | */ | |
3441 | void scheduler_tick(void) | |
3442 | { | |
7835b98b CL |
3443 | int cpu = smp_processor_id(); |
3444 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3445 | struct task_struct *curr = rq->curr; |
529c7726 | 3446 | u64 next_tick = rq->tick_timestamp + TICK_NSEC; |
dd41f596 IM |
3447 | |
3448 | spin_lock(&rq->lock); | |
546fe3c9 | 3449 | __update_rq_clock(rq); |
529c7726 IM |
3450 | /* |
3451 | * Let rq->clock advance by at least TICK_NSEC: | |
3452 | */ | |
3453 | if (unlikely(rq->clock < next_tick)) | |
3454 | rq->clock = next_tick; | |
3455 | rq->tick_timestamp = rq->clock; | |
f1a438d8 | 3456 | update_cpu_load(rq); |
dd41f596 IM |
3457 | if (curr != rq->idle) /* FIXME: needed? */ |
3458 | curr->sched_class->task_tick(rq, curr); | |
dd41f596 | 3459 | spin_unlock(&rq->lock); |
7835b98b | 3460 | |
e418e1c2 | 3461 | #ifdef CONFIG_SMP |
dd41f596 IM |
3462 | rq->idle_at_tick = idle_cpu(cpu); |
3463 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3464 | #endif |
1da177e4 LT |
3465 | } |
3466 | ||
1da177e4 LT |
3467 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3468 | ||
3469 | void fastcall add_preempt_count(int val) | |
3470 | { | |
3471 | /* | |
3472 | * Underflow? | |
3473 | */ | |
9a11b49a IM |
3474 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3475 | return; | |
1da177e4 LT |
3476 | preempt_count() += val; |
3477 | /* | |
3478 | * Spinlock count overflowing soon? | |
3479 | */ | |
33859f7f MOS |
3480 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3481 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3482 | } |
3483 | EXPORT_SYMBOL(add_preempt_count); | |
3484 | ||
3485 | void fastcall sub_preempt_count(int val) | |
3486 | { | |
3487 | /* | |
3488 | * Underflow? | |
3489 | */ | |
9a11b49a IM |
3490 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3491 | return; | |
1da177e4 LT |
3492 | /* |
3493 | * Is the spinlock portion underflowing? | |
3494 | */ | |
9a11b49a IM |
3495 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3496 | !(preempt_count() & PREEMPT_MASK))) | |
3497 | return; | |
3498 | ||
1da177e4 LT |
3499 | preempt_count() -= val; |
3500 | } | |
3501 | EXPORT_SYMBOL(sub_preempt_count); | |
3502 | ||
3503 | #endif | |
3504 | ||
3505 | /* | |
dd41f596 | 3506 | * Print scheduling while atomic bug: |
1da177e4 | 3507 | */ |
dd41f596 | 3508 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3509 | { |
dd41f596 | 3510 | printk(KERN_ERR "BUG: scheduling while atomic: %s/0x%08x/%d\n", |
ba25f9dc | 3511 | prev->comm, preempt_count(), task_pid_nr(prev)); |
dd41f596 IM |
3512 | debug_show_held_locks(prev); |
3513 | if (irqs_disabled()) | |
3514 | print_irqtrace_events(prev); | |
3515 | dump_stack(); | |
3516 | } | |
1da177e4 | 3517 | |
dd41f596 IM |
3518 | /* |
3519 | * Various schedule()-time debugging checks and statistics: | |
3520 | */ | |
3521 | static inline void schedule_debug(struct task_struct *prev) | |
3522 | { | |
1da177e4 LT |
3523 | /* |
3524 | * Test if we are atomic. Since do_exit() needs to call into | |
3525 | * schedule() atomically, we ignore that path for now. | |
3526 | * Otherwise, whine if we are scheduling when we should not be. | |
3527 | */ | |
dd41f596 IM |
3528 | if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state)) |
3529 | __schedule_bug(prev); | |
3530 | ||
1da177e4 LT |
3531 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3532 | ||
2d72376b | 3533 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3534 | #ifdef CONFIG_SCHEDSTATS |
3535 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3536 | schedstat_inc(this_rq(), bkl_count); |
3537 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3538 | } |
3539 | #endif | |
dd41f596 IM |
3540 | } |
3541 | ||
3542 | /* | |
3543 | * Pick up the highest-prio task: | |
3544 | */ | |
3545 | static inline struct task_struct * | |
ff95f3df | 3546 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 3547 | { |
5522d5d5 | 3548 | const struct sched_class *class; |
dd41f596 | 3549 | struct task_struct *p; |
1da177e4 LT |
3550 | |
3551 | /* | |
dd41f596 IM |
3552 | * Optimization: we know that if all tasks are in |
3553 | * the fair class we can call that function directly: | |
1da177e4 | 3554 | */ |
dd41f596 | 3555 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3556 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3557 | if (likely(p)) |
3558 | return p; | |
1da177e4 LT |
3559 | } |
3560 | ||
dd41f596 IM |
3561 | class = sched_class_highest; |
3562 | for ( ; ; ) { | |
fb8d4724 | 3563 | p = class->pick_next_task(rq); |
dd41f596 IM |
3564 | if (p) |
3565 | return p; | |
3566 | /* | |
3567 | * Will never be NULL as the idle class always | |
3568 | * returns a non-NULL p: | |
3569 | */ | |
3570 | class = class->next; | |
3571 | } | |
3572 | } | |
1da177e4 | 3573 | |
dd41f596 IM |
3574 | /* |
3575 | * schedule() is the main scheduler function. | |
3576 | */ | |
3577 | asmlinkage void __sched schedule(void) | |
3578 | { | |
3579 | struct task_struct *prev, *next; | |
3580 | long *switch_count; | |
3581 | struct rq *rq; | |
dd41f596 IM |
3582 | int cpu; |
3583 | ||
3584 | need_resched: | |
3585 | preempt_disable(); | |
3586 | cpu = smp_processor_id(); | |
3587 | rq = cpu_rq(cpu); | |
3588 | rcu_qsctr_inc(cpu); | |
3589 | prev = rq->curr; | |
3590 | switch_count = &prev->nivcsw; | |
3591 | ||
3592 | release_kernel_lock(prev); | |
3593 | need_resched_nonpreemptible: | |
3594 | ||
3595 | schedule_debug(prev); | |
1da177e4 | 3596 | |
1e819950 IM |
3597 | /* |
3598 | * Do the rq-clock update outside the rq lock: | |
3599 | */ | |
3600 | local_irq_disable(); | |
c1b3da3e | 3601 | __update_rq_clock(rq); |
1e819950 IM |
3602 | spin_lock(&rq->lock); |
3603 | clear_tsk_need_resched(prev); | |
1da177e4 | 3604 | |
1da177e4 | 3605 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
1da177e4 | 3606 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && |
dd41f596 | 3607 | unlikely(signal_pending(prev)))) { |
1da177e4 | 3608 | prev->state = TASK_RUNNING; |
dd41f596 | 3609 | } else { |
2e1cb74a | 3610 | deactivate_task(rq, prev, 1); |
1da177e4 | 3611 | } |
dd41f596 | 3612 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3613 | } |
3614 | ||
dd41f596 | 3615 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3616 | idle_balance(cpu, rq); |
1da177e4 | 3617 | |
31ee529c | 3618 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 3619 | next = pick_next_task(rq, prev); |
1da177e4 LT |
3620 | |
3621 | sched_info_switch(prev, next); | |
dd41f596 | 3622 | |
1da177e4 | 3623 | if (likely(prev != next)) { |
1da177e4 LT |
3624 | rq->nr_switches++; |
3625 | rq->curr = next; | |
3626 | ++*switch_count; | |
3627 | ||
dd41f596 | 3628 | context_switch(rq, prev, next); /* unlocks the rq */ |
1da177e4 LT |
3629 | } else |
3630 | spin_unlock_irq(&rq->lock); | |
3631 | ||
dd41f596 IM |
3632 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
3633 | cpu = smp_processor_id(); | |
3634 | rq = cpu_rq(cpu); | |
1da177e4 | 3635 | goto need_resched_nonpreemptible; |
dd41f596 | 3636 | } |
1da177e4 LT |
3637 | preempt_enable_no_resched(); |
3638 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3639 | goto need_resched; | |
3640 | } | |
1da177e4 LT |
3641 | EXPORT_SYMBOL(schedule); |
3642 | ||
3643 | #ifdef CONFIG_PREEMPT | |
3644 | /* | |
2ed6e34f | 3645 | * this is the entry point to schedule() from in-kernel preemption |
1da177e4 LT |
3646 | * off of preempt_enable. Kernel preemptions off return from interrupt |
3647 | * occur there and call schedule directly. | |
3648 | */ | |
3649 | asmlinkage void __sched preempt_schedule(void) | |
3650 | { | |
3651 | struct thread_info *ti = current_thread_info(); | |
3652 | #ifdef CONFIG_PREEMPT_BKL | |
3653 | struct task_struct *task = current; | |
3654 | int saved_lock_depth; | |
3655 | #endif | |
3656 | /* | |
3657 | * If there is a non-zero preempt_count or interrupts are disabled, | |
3658 | * we do not want to preempt the current task. Just return.. | |
3659 | */ | |
beed33a8 | 3660 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3661 | return; |
3662 | ||
3a5c359a AK |
3663 | do { |
3664 | add_preempt_count(PREEMPT_ACTIVE); | |
3665 | ||
3666 | /* | |
3667 | * We keep the big kernel semaphore locked, but we | |
3668 | * clear ->lock_depth so that schedule() doesnt | |
3669 | * auto-release the semaphore: | |
3670 | */ | |
1da177e4 | 3671 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a AK |
3672 | saved_lock_depth = task->lock_depth; |
3673 | task->lock_depth = -1; | |
1da177e4 | 3674 | #endif |
3a5c359a | 3675 | schedule(); |
1da177e4 | 3676 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a | 3677 | task->lock_depth = saved_lock_depth; |
1da177e4 | 3678 | #endif |
3a5c359a | 3679 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3680 | |
3a5c359a AK |
3681 | /* |
3682 | * Check again in case we missed a preemption opportunity | |
3683 | * between schedule and now. | |
3684 | */ | |
3685 | barrier(); | |
3686 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 3687 | } |
1da177e4 LT |
3688 | EXPORT_SYMBOL(preempt_schedule); |
3689 | ||
3690 | /* | |
2ed6e34f | 3691 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3692 | * off of irq context. |
3693 | * Note, that this is called and return with irqs disabled. This will | |
3694 | * protect us against recursive calling from irq. | |
3695 | */ | |
3696 | asmlinkage void __sched preempt_schedule_irq(void) | |
3697 | { | |
3698 | struct thread_info *ti = current_thread_info(); | |
3699 | #ifdef CONFIG_PREEMPT_BKL | |
3700 | struct task_struct *task = current; | |
3701 | int saved_lock_depth; | |
3702 | #endif | |
2ed6e34f | 3703 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3704 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3705 | ||
3a5c359a AK |
3706 | do { |
3707 | add_preempt_count(PREEMPT_ACTIVE); | |
3708 | ||
3709 | /* | |
3710 | * We keep the big kernel semaphore locked, but we | |
3711 | * clear ->lock_depth so that schedule() doesnt | |
3712 | * auto-release the semaphore: | |
3713 | */ | |
1da177e4 | 3714 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a AK |
3715 | saved_lock_depth = task->lock_depth; |
3716 | task->lock_depth = -1; | |
1da177e4 | 3717 | #endif |
3a5c359a AK |
3718 | local_irq_enable(); |
3719 | schedule(); | |
3720 | local_irq_disable(); | |
1da177e4 | 3721 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a | 3722 | task->lock_depth = saved_lock_depth; |
1da177e4 | 3723 | #endif |
3a5c359a | 3724 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3725 | |
3a5c359a AK |
3726 | /* |
3727 | * Check again in case we missed a preemption opportunity | |
3728 | * between schedule and now. | |
3729 | */ | |
3730 | barrier(); | |
3731 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
3732 | } |
3733 | ||
3734 | #endif /* CONFIG_PREEMPT */ | |
3735 | ||
95cdf3b7 IM |
3736 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
3737 | void *key) | |
1da177e4 | 3738 | { |
48f24c4d | 3739 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 3740 | } |
1da177e4 LT |
3741 | EXPORT_SYMBOL(default_wake_function); |
3742 | ||
3743 | /* | |
3744 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just | |
3745 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
3746 | * number) then we wake all the non-exclusive tasks and one exclusive task. | |
3747 | * | |
3748 | * There are circumstances in which we can try to wake a task which has already | |
3749 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns | |
3750 | * zero in this (rare) case, and we handle it by continuing to scan the queue. | |
3751 | */ | |
3752 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
3753 | int nr_exclusive, int sync, void *key) | |
3754 | { | |
2e45874c | 3755 | wait_queue_t *curr, *next; |
1da177e4 | 3756 | |
2e45874c | 3757 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3758 | unsigned flags = curr->flags; |
3759 | ||
1da177e4 | 3760 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 3761 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3762 | break; |
3763 | } | |
3764 | } | |
3765 | ||
3766 | /** | |
3767 | * __wake_up - wake up threads blocked on a waitqueue. | |
3768 | * @q: the waitqueue | |
3769 | * @mode: which threads | |
3770 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3771 | * @key: is directly passed to the wakeup function |
1da177e4 LT |
3772 | */ |
3773 | void fastcall __wake_up(wait_queue_head_t *q, unsigned int mode, | |
95cdf3b7 | 3774 | int nr_exclusive, void *key) |
1da177e4 LT |
3775 | { |
3776 | unsigned long flags; | |
3777 | ||
3778 | spin_lock_irqsave(&q->lock, flags); | |
3779 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3780 | spin_unlock_irqrestore(&q->lock, flags); | |
3781 | } | |
1da177e4 LT |
3782 | EXPORT_SYMBOL(__wake_up); |
3783 | ||
3784 | /* | |
3785 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3786 | */ | |
3787 | void fastcall __wake_up_locked(wait_queue_head_t *q, unsigned int mode) | |
3788 | { | |
3789 | __wake_up_common(q, mode, 1, 0, NULL); | |
3790 | } | |
3791 | ||
3792 | /** | |
67be2dd1 | 3793 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3794 | * @q: the waitqueue |
3795 | * @mode: which threads | |
3796 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
3797 | * | |
3798 | * The sync wakeup differs that the waker knows that it will schedule | |
3799 | * away soon, so while the target thread will be woken up, it will not | |
3800 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3801 | * with each other. This can prevent needless bouncing between CPUs. | |
3802 | * | |
3803 | * On UP it can prevent extra preemption. | |
3804 | */ | |
95cdf3b7 IM |
3805 | void fastcall |
3806 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
1da177e4 LT |
3807 | { |
3808 | unsigned long flags; | |
3809 | int sync = 1; | |
3810 | ||
3811 | if (unlikely(!q)) | |
3812 | return; | |
3813 | ||
3814 | if (unlikely(!nr_exclusive)) | |
3815 | sync = 0; | |
3816 | ||
3817 | spin_lock_irqsave(&q->lock, flags); | |
3818 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
3819 | spin_unlock_irqrestore(&q->lock, flags); | |
3820 | } | |
3821 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
3822 | ||
3823 | void fastcall complete(struct completion *x) | |
3824 | { | |
3825 | unsigned long flags; | |
3826 | ||
3827 | spin_lock_irqsave(&x->wait.lock, flags); | |
3828 | x->done++; | |
3829 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3830 | 1, 0, NULL); | |
3831 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3832 | } | |
3833 | EXPORT_SYMBOL(complete); | |
3834 | ||
3835 | void fastcall complete_all(struct completion *x) | |
3836 | { | |
3837 | unsigned long flags; | |
3838 | ||
3839 | spin_lock_irqsave(&x->wait.lock, flags); | |
3840 | x->done += UINT_MAX/2; | |
3841 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3842 | 0, 0, NULL); | |
3843 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3844 | } | |
3845 | EXPORT_SYMBOL(complete_all); | |
3846 | ||
8cbbe86d AK |
3847 | static inline long __sched |
3848 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3849 | { |
1da177e4 LT |
3850 | if (!x->done) { |
3851 | DECLARE_WAITQUEUE(wait, current); | |
3852 | ||
3853 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3854 | __add_wait_queue_tail(&x->wait, &wait); | |
3855 | do { | |
8cbbe86d AK |
3856 | if (state == TASK_INTERRUPTIBLE && |
3857 | signal_pending(current)) { | |
3858 | __remove_wait_queue(&x->wait, &wait); | |
3859 | return -ERESTARTSYS; | |
3860 | } | |
3861 | __set_current_state(state); | |
1da177e4 LT |
3862 | spin_unlock_irq(&x->wait.lock); |
3863 | timeout = schedule_timeout(timeout); | |
3864 | spin_lock_irq(&x->wait.lock); | |
3865 | if (!timeout) { | |
3866 | __remove_wait_queue(&x->wait, &wait); | |
8cbbe86d | 3867 | return timeout; |
1da177e4 LT |
3868 | } |
3869 | } while (!x->done); | |
3870 | __remove_wait_queue(&x->wait, &wait); | |
3871 | } | |
3872 | x->done--; | |
1da177e4 LT |
3873 | return timeout; |
3874 | } | |
1da177e4 | 3875 | |
8cbbe86d AK |
3876 | static long __sched |
3877 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3878 | { |
1da177e4 LT |
3879 | might_sleep(); |
3880 | ||
3881 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 3882 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 3883 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3884 | return timeout; |
3885 | } | |
1da177e4 | 3886 | |
8cbbe86d AK |
3887 | void fastcall __sched wait_for_completion(struct completion *x) |
3888 | { | |
3889 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 3890 | } |
8cbbe86d | 3891 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 LT |
3892 | |
3893 | unsigned long fastcall __sched | |
8cbbe86d | 3894 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 3895 | { |
8cbbe86d | 3896 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 3897 | } |
8cbbe86d | 3898 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 3899 | |
8cbbe86d | 3900 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 3901 | { |
51e97990 AK |
3902 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
3903 | if (t == -ERESTARTSYS) | |
3904 | return t; | |
3905 | return 0; | |
0fec171c | 3906 | } |
8cbbe86d | 3907 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 3908 | |
8cbbe86d AK |
3909 | unsigned long fastcall __sched |
3910 | wait_for_completion_interruptible_timeout(struct completion *x, | |
3911 | unsigned long timeout) | |
0fec171c | 3912 | { |
8cbbe86d | 3913 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 3914 | } |
8cbbe86d | 3915 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 3916 | |
8cbbe86d AK |
3917 | static long __sched |
3918 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 3919 | { |
0fec171c IM |
3920 | unsigned long flags; |
3921 | wait_queue_t wait; | |
3922 | ||
3923 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 3924 | |
8cbbe86d | 3925 | __set_current_state(state); |
1da177e4 | 3926 | |
8cbbe86d AK |
3927 | spin_lock_irqsave(&q->lock, flags); |
3928 | __add_wait_queue(q, &wait); | |
3929 | spin_unlock(&q->lock); | |
3930 | timeout = schedule_timeout(timeout); | |
3931 | spin_lock_irq(&q->lock); | |
3932 | __remove_wait_queue(q, &wait); | |
3933 | spin_unlock_irqrestore(&q->lock, flags); | |
3934 | ||
3935 | return timeout; | |
3936 | } | |
3937 | ||
3938 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
3939 | { | |
3940 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 3941 | } |
1da177e4 LT |
3942 | EXPORT_SYMBOL(interruptible_sleep_on); |
3943 | ||
0fec171c | 3944 | long __sched |
95cdf3b7 | 3945 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3946 | { |
8cbbe86d | 3947 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 3948 | } |
1da177e4 LT |
3949 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3950 | ||
0fec171c | 3951 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3952 | { |
8cbbe86d | 3953 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 3954 | } |
1da177e4 LT |
3955 | EXPORT_SYMBOL(sleep_on); |
3956 | ||
0fec171c | 3957 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3958 | { |
8cbbe86d | 3959 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 3960 | } |
1da177e4 LT |
3961 | EXPORT_SYMBOL(sleep_on_timeout); |
3962 | ||
b29739f9 IM |
3963 | #ifdef CONFIG_RT_MUTEXES |
3964 | ||
3965 | /* | |
3966 | * rt_mutex_setprio - set the current priority of a task | |
3967 | * @p: task | |
3968 | * @prio: prio value (kernel-internal form) | |
3969 | * | |
3970 | * This function changes the 'effective' priority of a task. It does | |
3971 | * not touch ->normal_prio like __setscheduler(). | |
3972 | * | |
3973 | * Used by the rt_mutex code to implement priority inheritance logic. | |
3974 | */ | |
36c8b586 | 3975 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
3976 | { |
3977 | unsigned long flags; | |
83b699ed | 3978 | int oldprio, on_rq, running; |
70b97a7f | 3979 | struct rq *rq; |
b29739f9 IM |
3980 | |
3981 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
3982 | ||
3983 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 3984 | update_rq_clock(rq); |
b29739f9 | 3985 | |
d5f9f942 | 3986 | oldprio = p->prio; |
dd41f596 | 3987 | on_rq = p->se.on_rq; |
83b699ed SV |
3988 | running = task_running(rq, p); |
3989 | if (on_rq) { | |
69be72c1 | 3990 | dequeue_task(rq, p, 0); |
83b699ed SV |
3991 | if (running) |
3992 | p->sched_class->put_prev_task(rq, p); | |
3993 | } | |
dd41f596 IM |
3994 | |
3995 | if (rt_prio(prio)) | |
3996 | p->sched_class = &rt_sched_class; | |
3997 | else | |
3998 | p->sched_class = &fair_sched_class; | |
3999 | ||
b29739f9 IM |
4000 | p->prio = prio; |
4001 | ||
dd41f596 | 4002 | if (on_rq) { |
83b699ed SV |
4003 | if (running) |
4004 | p->sched_class->set_curr_task(rq); | |
8159f87e | 4005 | enqueue_task(rq, p, 0); |
b29739f9 IM |
4006 | /* |
4007 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4008 | * our priority decreased, or if we are not currently running on |
4009 | * this runqueue and our priority is higher than the current's | |
b29739f9 | 4010 | */ |
83b699ed | 4011 | if (running) { |
d5f9f942 AM |
4012 | if (p->prio > oldprio) |
4013 | resched_task(rq->curr); | |
dd41f596 IM |
4014 | } else { |
4015 | check_preempt_curr(rq, p); | |
4016 | } | |
b29739f9 IM |
4017 | } |
4018 | task_rq_unlock(rq, &flags); | |
4019 | } | |
4020 | ||
4021 | #endif | |
4022 | ||
36c8b586 | 4023 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4024 | { |
dd41f596 | 4025 | int old_prio, delta, on_rq; |
1da177e4 | 4026 | unsigned long flags; |
70b97a7f | 4027 | struct rq *rq; |
1da177e4 LT |
4028 | |
4029 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4030 | return; | |
4031 | /* | |
4032 | * We have to be careful, if called from sys_setpriority(), | |
4033 | * the task might be in the middle of scheduling on another CPU. | |
4034 | */ | |
4035 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4036 | update_rq_clock(rq); |
1da177e4 LT |
4037 | /* |
4038 | * The RT priorities are set via sched_setscheduler(), but we still | |
4039 | * allow the 'normal' nice value to be set - but as expected | |
4040 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4041 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4042 | */ |
e05606d3 | 4043 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4044 | p->static_prio = NICE_TO_PRIO(nice); |
4045 | goto out_unlock; | |
4046 | } | |
dd41f596 IM |
4047 | on_rq = p->se.on_rq; |
4048 | if (on_rq) { | |
69be72c1 | 4049 | dequeue_task(rq, p, 0); |
79b5dddf | 4050 | dec_load(rq, p); |
2dd73a4f | 4051 | } |
1da177e4 | 4052 | |
1da177e4 | 4053 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4054 | set_load_weight(p); |
b29739f9 IM |
4055 | old_prio = p->prio; |
4056 | p->prio = effective_prio(p); | |
4057 | delta = p->prio - old_prio; | |
1da177e4 | 4058 | |
dd41f596 | 4059 | if (on_rq) { |
8159f87e | 4060 | enqueue_task(rq, p, 0); |
29b4b623 | 4061 | inc_load(rq, p); |
1da177e4 | 4062 | /* |
d5f9f942 AM |
4063 | * If the task increased its priority or is running and |
4064 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4065 | */ |
d5f9f942 | 4066 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4067 | resched_task(rq->curr); |
4068 | } | |
4069 | out_unlock: | |
4070 | task_rq_unlock(rq, &flags); | |
4071 | } | |
1da177e4 LT |
4072 | EXPORT_SYMBOL(set_user_nice); |
4073 | ||
e43379f1 MM |
4074 | /* |
4075 | * can_nice - check if a task can reduce its nice value | |
4076 | * @p: task | |
4077 | * @nice: nice value | |
4078 | */ | |
36c8b586 | 4079 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4080 | { |
024f4747 MM |
4081 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4082 | int nice_rlim = 20 - nice; | |
48f24c4d | 4083 | |
e43379f1 MM |
4084 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4085 | capable(CAP_SYS_NICE)); | |
4086 | } | |
4087 | ||
1da177e4 LT |
4088 | #ifdef __ARCH_WANT_SYS_NICE |
4089 | ||
4090 | /* | |
4091 | * sys_nice - change the priority of the current process. | |
4092 | * @increment: priority increment | |
4093 | * | |
4094 | * sys_setpriority is a more generic, but much slower function that | |
4095 | * does similar things. | |
4096 | */ | |
4097 | asmlinkage long sys_nice(int increment) | |
4098 | { | |
48f24c4d | 4099 | long nice, retval; |
1da177e4 LT |
4100 | |
4101 | /* | |
4102 | * Setpriority might change our priority at the same moment. | |
4103 | * We don't have to worry. Conceptually one call occurs first | |
4104 | * and we have a single winner. | |
4105 | */ | |
e43379f1 MM |
4106 | if (increment < -40) |
4107 | increment = -40; | |
1da177e4 LT |
4108 | if (increment > 40) |
4109 | increment = 40; | |
4110 | ||
4111 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4112 | if (nice < -20) | |
4113 | nice = -20; | |
4114 | if (nice > 19) | |
4115 | nice = 19; | |
4116 | ||
e43379f1 MM |
4117 | if (increment < 0 && !can_nice(current, nice)) |
4118 | return -EPERM; | |
4119 | ||
1da177e4 LT |
4120 | retval = security_task_setnice(current, nice); |
4121 | if (retval) | |
4122 | return retval; | |
4123 | ||
4124 | set_user_nice(current, nice); | |
4125 | return 0; | |
4126 | } | |
4127 | ||
4128 | #endif | |
4129 | ||
4130 | /** | |
4131 | * task_prio - return the priority value of a given task. | |
4132 | * @p: the task in question. | |
4133 | * | |
4134 | * This is the priority value as seen by users in /proc. | |
4135 | * RT tasks are offset by -200. Normal tasks are centered | |
4136 | * around 0, value goes from -16 to +15. | |
4137 | */ | |
36c8b586 | 4138 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4139 | { |
4140 | return p->prio - MAX_RT_PRIO; | |
4141 | } | |
4142 | ||
4143 | /** | |
4144 | * task_nice - return the nice value of a given task. | |
4145 | * @p: the task in question. | |
4146 | */ | |
36c8b586 | 4147 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4148 | { |
4149 | return TASK_NICE(p); | |
4150 | } | |
1da177e4 | 4151 | EXPORT_SYMBOL_GPL(task_nice); |
1da177e4 LT |
4152 | |
4153 | /** | |
4154 | * idle_cpu - is a given cpu idle currently? | |
4155 | * @cpu: the processor in question. | |
4156 | */ | |
4157 | int idle_cpu(int cpu) | |
4158 | { | |
4159 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4160 | } | |
4161 | ||
1da177e4 LT |
4162 | /** |
4163 | * idle_task - return the idle task for a given cpu. | |
4164 | * @cpu: the processor in question. | |
4165 | */ | |
36c8b586 | 4166 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4167 | { |
4168 | return cpu_rq(cpu)->idle; | |
4169 | } | |
4170 | ||
4171 | /** | |
4172 | * find_process_by_pid - find a process with a matching PID value. | |
4173 | * @pid: the pid in question. | |
4174 | */ | |
a9957449 | 4175 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4176 | { |
228ebcbe | 4177 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4178 | } |
4179 | ||
4180 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4181 | static void |
4182 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4183 | { |
dd41f596 | 4184 | BUG_ON(p->se.on_rq); |
48f24c4d | 4185 | |
1da177e4 | 4186 | p->policy = policy; |
dd41f596 IM |
4187 | switch (p->policy) { |
4188 | case SCHED_NORMAL: | |
4189 | case SCHED_BATCH: | |
4190 | case SCHED_IDLE: | |
4191 | p->sched_class = &fair_sched_class; | |
4192 | break; | |
4193 | case SCHED_FIFO: | |
4194 | case SCHED_RR: | |
4195 | p->sched_class = &rt_sched_class; | |
4196 | break; | |
4197 | } | |
4198 | ||
1da177e4 | 4199 | p->rt_priority = prio; |
b29739f9 IM |
4200 | p->normal_prio = normal_prio(p); |
4201 | /* we are holding p->pi_lock already */ | |
4202 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4203 | set_load_weight(p); |
1da177e4 LT |
4204 | } |
4205 | ||
4206 | /** | |
72fd4a35 | 4207 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4208 | * @p: the task in question. |
4209 | * @policy: new policy. | |
4210 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4211 | * |
72fd4a35 | 4212 | * NOTE that the task may be already dead. |
1da177e4 | 4213 | */ |
95cdf3b7 IM |
4214 | int sched_setscheduler(struct task_struct *p, int policy, |
4215 | struct sched_param *param) | |
1da177e4 | 4216 | { |
83b699ed | 4217 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4218 | unsigned long flags; |
70b97a7f | 4219 | struct rq *rq; |
1da177e4 | 4220 | |
66e5393a SR |
4221 | /* may grab non-irq protected spin_locks */ |
4222 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4223 | recheck: |
4224 | /* double check policy once rq lock held */ | |
4225 | if (policy < 0) | |
4226 | policy = oldpolicy = p->policy; | |
4227 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4228 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4229 | policy != SCHED_IDLE) | |
b0a9499c | 4230 | return -EINVAL; |
1da177e4 LT |
4231 | /* |
4232 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4233 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4234 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4235 | */ |
4236 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4237 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4238 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4239 | return -EINVAL; |
e05606d3 | 4240 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4241 | return -EINVAL; |
4242 | ||
37e4ab3f OC |
4243 | /* |
4244 | * Allow unprivileged RT tasks to decrease priority: | |
4245 | */ | |
4246 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4247 | if (rt_policy(policy)) { |
8dc3e909 | 4248 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4249 | |
4250 | if (!lock_task_sighand(p, &flags)) | |
4251 | return -ESRCH; | |
4252 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4253 | unlock_task_sighand(p, &flags); | |
4254 | ||
4255 | /* can't set/change the rt policy */ | |
4256 | if (policy != p->policy && !rlim_rtprio) | |
4257 | return -EPERM; | |
4258 | ||
4259 | /* can't increase priority */ | |
4260 | if (param->sched_priority > p->rt_priority && | |
4261 | param->sched_priority > rlim_rtprio) | |
4262 | return -EPERM; | |
4263 | } | |
dd41f596 IM |
4264 | /* |
4265 | * Like positive nice levels, dont allow tasks to | |
4266 | * move out of SCHED_IDLE either: | |
4267 | */ | |
4268 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4269 | return -EPERM; | |
5fe1d75f | 4270 | |
37e4ab3f OC |
4271 | /* can't change other user's priorities */ |
4272 | if ((current->euid != p->euid) && | |
4273 | (current->euid != p->uid)) | |
4274 | return -EPERM; | |
4275 | } | |
1da177e4 LT |
4276 | |
4277 | retval = security_task_setscheduler(p, policy, param); | |
4278 | if (retval) | |
4279 | return retval; | |
b29739f9 IM |
4280 | /* |
4281 | * make sure no PI-waiters arrive (or leave) while we are | |
4282 | * changing the priority of the task: | |
4283 | */ | |
4284 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4285 | /* |
4286 | * To be able to change p->policy safely, the apropriate | |
4287 | * runqueue lock must be held. | |
4288 | */ | |
b29739f9 | 4289 | rq = __task_rq_lock(p); |
1da177e4 LT |
4290 | /* recheck policy now with rq lock held */ |
4291 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4292 | policy = oldpolicy = -1; | |
b29739f9 IM |
4293 | __task_rq_unlock(rq); |
4294 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4295 | goto recheck; |
4296 | } | |
2daa3577 | 4297 | update_rq_clock(rq); |
dd41f596 | 4298 | on_rq = p->se.on_rq; |
83b699ed SV |
4299 | running = task_running(rq, p); |
4300 | if (on_rq) { | |
2e1cb74a | 4301 | deactivate_task(rq, p, 0); |
83b699ed SV |
4302 | if (running) |
4303 | p->sched_class->put_prev_task(rq, p); | |
4304 | } | |
f6b53205 | 4305 | |
1da177e4 | 4306 | oldprio = p->prio; |
dd41f596 | 4307 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4308 | |
dd41f596 | 4309 | if (on_rq) { |
83b699ed SV |
4310 | if (running) |
4311 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4312 | activate_task(rq, p, 0); |
1da177e4 LT |
4313 | /* |
4314 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4315 | * our priority decreased, or if we are not currently running on |
4316 | * this runqueue and our priority is higher than the current's | |
1da177e4 | 4317 | */ |
83b699ed | 4318 | if (running) { |
d5f9f942 AM |
4319 | if (p->prio > oldprio) |
4320 | resched_task(rq->curr); | |
dd41f596 IM |
4321 | } else { |
4322 | check_preempt_curr(rq, p); | |
4323 | } | |
1da177e4 | 4324 | } |
b29739f9 IM |
4325 | __task_rq_unlock(rq); |
4326 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4327 | ||
95e02ca9 TG |
4328 | rt_mutex_adjust_pi(p); |
4329 | ||
1da177e4 LT |
4330 | return 0; |
4331 | } | |
4332 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4333 | ||
95cdf3b7 IM |
4334 | static int |
4335 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4336 | { |
1da177e4 LT |
4337 | struct sched_param lparam; |
4338 | struct task_struct *p; | |
36c8b586 | 4339 | int retval; |
1da177e4 LT |
4340 | |
4341 | if (!param || pid < 0) | |
4342 | return -EINVAL; | |
4343 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4344 | return -EFAULT; | |
5fe1d75f ON |
4345 | |
4346 | rcu_read_lock(); | |
4347 | retval = -ESRCH; | |
1da177e4 | 4348 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4349 | if (p != NULL) |
4350 | retval = sched_setscheduler(p, policy, &lparam); | |
4351 | rcu_read_unlock(); | |
36c8b586 | 4352 | |
1da177e4 LT |
4353 | return retval; |
4354 | } | |
4355 | ||
4356 | /** | |
4357 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4358 | * @pid: the pid in question. | |
4359 | * @policy: new policy. | |
4360 | * @param: structure containing the new RT priority. | |
4361 | */ | |
4362 | asmlinkage long sys_sched_setscheduler(pid_t pid, int policy, | |
4363 | struct sched_param __user *param) | |
4364 | { | |
c21761f1 JB |
4365 | /* negative values for policy are not valid */ |
4366 | if (policy < 0) | |
4367 | return -EINVAL; | |
4368 | ||
1da177e4 LT |
4369 | return do_sched_setscheduler(pid, policy, param); |
4370 | } | |
4371 | ||
4372 | /** | |
4373 | * sys_sched_setparam - set/change the RT priority of a thread | |
4374 | * @pid: the pid in question. | |
4375 | * @param: structure containing the new RT priority. | |
4376 | */ | |
4377 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4378 | { | |
4379 | return do_sched_setscheduler(pid, -1, param); | |
4380 | } | |
4381 | ||
4382 | /** | |
4383 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4384 | * @pid: the pid in question. | |
4385 | */ | |
4386 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4387 | { | |
36c8b586 | 4388 | struct task_struct *p; |
3a5c359a | 4389 | int retval; |
1da177e4 LT |
4390 | |
4391 | if (pid < 0) | |
3a5c359a | 4392 | return -EINVAL; |
1da177e4 LT |
4393 | |
4394 | retval = -ESRCH; | |
4395 | read_lock(&tasklist_lock); | |
4396 | p = find_process_by_pid(pid); | |
4397 | if (p) { | |
4398 | retval = security_task_getscheduler(p); | |
4399 | if (!retval) | |
4400 | retval = p->policy; | |
4401 | } | |
4402 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
4403 | return retval; |
4404 | } | |
4405 | ||
4406 | /** | |
4407 | * sys_sched_getscheduler - get the RT priority of a thread | |
4408 | * @pid: the pid in question. | |
4409 | * @param: structure containing the RT priority. | |
4410 | */ | |
4411 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
4412 | { | |
4413 | struct sched_param lp; | |
36c8b586 | 4414 | struct task_struct *p; |
3a5c359a | 4415 | int retval; |
1da177e4 LT |
4416 | |
4417 | if (!param || pid < 0) | |
3a5c359a | 4418 | return -EINVAL; |
1da177e4 LT |
4419 | |
4420 | read_lock(&tasklist_lock); | |
4421 | p = find_process_by_pid(pid); | |
4422 | retval = -ESRCH; | |
4423 | if (!p) | |
4424 | goto out_unlock; | |
4425 | ||
4426 | retval = security_task_getscheduler(p); | |
4427 | if (retval) | |
4428 | goto out_unlock; | |
4429 | ||
4430 | lp.sched_priority = p->rt_priority; | |
4431 | read_unlock(&tasklist_lock); | |
4432 | ||
4433 | /* | |
4434 | * This one might sleep, we cannot do it with a spinlock held ... | |
4435 | */ | |
4436 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4437 | ||
1da177e4 LT |
4438 | return retval; |
4439 | ||
4440 | out_unlock: | |
4441 | read_unlock(&tasklist_lock); | |
4442 | return retval; | |
4443 | } | |
4444 | ||
4445 | long sched_setaffinity(pid_t pid, cpumask_t new_mask) | |
4446 | { | |
1da177e4 | 4447 | cpumask_t cpus_allowed; |
36c8b586 IM |
4448 | struct task_struct *p; |
4449 | int retval; | |
1da177e4 | 4450 | |
5be9361c | 4451 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4452 | read_lock(&tasklist_lock); |
4453 | ||
4454 | p = find_process_by_pid(pid); | |
4455 | if (!p) { | |
4456 | read_unlock(&tasklist_lock); | |
5be9361c | 4457 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4458 | return -ESRCH; |
4459 | } | |
4460 | ||
4461 | /* | |
4462 | * It is not safe to call set_cpus_allowed with the | |
4463 | * tasklist_lock held. We will bump the task_struct's | |
4464 | * usage count and then drop tasklist_lock. | |
4465 | */ | |
4466 | get_task_struct(p); | |
4467 | read_unlock(&tasklist_lock); | |
4468 | ||
4469 | retval = -EPERM; | |
4470 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
4471 | !capable(CAP_SYS_NICE)) | |
4472 | goto out_unlock; | |
4473 | ||
e7834f8f DQ |
4474 | retval = security_task_setscheduler(p, 0, NULL); |
4475 | if (retval) | |
4476 | goto out_unlock; | |
4477 | ||
1da177e4 LT |
4478 | cpus_allowed = cpuset_cpus_allowed(p); |
4479 | cpus_and(new_mask, new_mask, cpus_allowed); | |
8707d8b8 | 4480 | again: |
1da177e4 LT |
4481 | retval = set_cpus_allowed(p, new_mask); |
4482 | ||
8707d8b8 PM |
4483 | if (!retval) { |
4484 | cpus_allowed = cpuset_cpus_allowed(p); | |
4485 | if (!cpus_subset(new_mask, cpus_allowed)) { | |
4486 | /* | |
4487 | * We must have raced with a concurrent cpuset | |
4488 | * update. Just reset the cpus_allowed to the | |
4489 | * cpuset's cpus_allowed | |
4490 | */ | |
4491 | new_mask = cpus_allowed; | |
4492 | goto again; | |
4493 | } | |
4494 | } | |
1da177e4 LT |
4495 | out_unlock: |
4496 | put_task_struct(p); | |
5be9361c | 4497 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4498 | return retval; |
4499 | } | |
4500 | ||
4501 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
4502 | cpumask_t *new_mask) | |
4503 | { | |
4504 | if (len < sizeof(cpumask_t)) { | |
4505 | memset(new_mask, 0, sizeof(cpumask_t)); | |
4506 | } else if (len > sizeof(cpumask_t)) { | |
4507 | len = sizeof(cpumask_t); | |
4508 | } | |
4509 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
4510 | } | |
4511 | ||
4512 | /** | |
4513 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4514 | * @pid: pid of the process | |
4515 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4516 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4517 | */ | |
4518 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
4519 | unsigned long __user *user_mask_ptr) | |
4520 | { | |
4521 | cpumask_t new_mask; | |
4522 | int retval; | |
4523 | ||
4524 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
4525 | if (retval) | |
4526 | return retval; | |
4527 | ||
4528 | return sched_setaffinity(pid, new_mask); | |
4529 | } | |
4530 | ||
4531 | /* | |
4532 | * Represents all cpu's present in the system | |
4533 | * In systems capable of hotplug, this map could dynamically grow | |
4534 | * as new cpu's are detected in the system via any platform specific | |
4535 | * method, such as ACPI for e.g. | |
4536 | */ | |
4537 | ||
4cef0c61 | 4538 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
4539 | EXPORT_SYMBOL(cpu_present_map); |
4540 | ||
4541 | #ifndef CONFIG_SMP | |
4cef0c61 | 4542 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
4543 | EXPORT_SYMBOL(cpu_online_map); |
4544 | ||
4cef0c61 | 4545 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 4546 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
4547 | #endif |
4548 | ||
4549 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
4550 | { | |
36c8b586 | 4551 | struct task_struct *p; |
1da177e4 | 4552 | int retval; |
1da177e4 | 4553 | |
5be9361c | 4554 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4555 | read_lock(&tasklist_lock); |
4556 | ||
4557 | retval = -ESRCH; | |
4558 | p = find_process_by_pid(pid); | |
4559 | if (!p) | |
4560 | goto out_unlock; | |
4561 | ||
e7834f8f DQ |
4562 | retval = security_task_getscheduler(p); |
4563 | if (retval) | |
4564 | goto out_unlock; | |
4565 | ||
2f7016d9 | 4566 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
4567 | |
4568 | out_unlock: | |
4569 | read_unlock(&tasklist_lock); | |
5be9361c | 4570 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 | 4571 | |
9531b62f | 4572 | return retval; |
1da177e4 LT |
4573 | } |
4574 | ||
4575 | /** | |
4576 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4577 | * @pid: pid of the process | |
4578 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4579 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4580 | */ | |
4581 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
4582 | unsigned long __user *user_mask_ptr) | |
4583 | { | |
4584 | int ret; | |
4585 | cpumask_t mask; | |
4586 | ||
4587 | if (len < sizeof(cpumask_t)) | |
4588 | return -EINVAL; | |
4589 | ||
4590 | ret = sched_getaffinity(pid, &mask); | |
4591 | if (ret < 0) | |
4592 | return ret; | |
4593 | ||
4594 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
4595 | return -EFAULT; | |
4596 | ||
4597 | return sizeof(cpumask_t); | |
4598 | } | |
4599 | ||
4600 | /** | |
4601 | * sys_sched_yield - yield the current processor to other threads. | |
4602 | * | |
dd41f596 IM |
4603 | * This function yields the current CPU to other tasks. If there are no |
4604 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
4605 | */ |
4606 | asmlinkage long sys_sched_yield(void) | |
4607 | { | |
70b97a7f | 4608 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4609 | |
2d72376b | 4610 | schedstat_inc(rq, yld_count); |
4530d7ab | 4611 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4612 | |
4613 | /* | |
4614 | * Since we are going to call schedule() anyway, there's | |
4615 | * no need to preempt or enable interrupts: | |
4616 | */ | |
4617 | __release(rq->lock); | |
8a25d5de | 4618 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4619 | _raw_spin_unlock(&rq->lock); |
4620 | preempt_enable_no_resched(); | |
4621 | ||
4622 | schedule(); | |
4623 | ||
4624 | return 0; | |
4625 | } | |
4626 | ||
e7b38404 | 4627 | static void __cond_resched(void) |
1da177e4 | 4628 | { |
8e0a43d8 IM |
4629 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
4630 | __might_sleep(__FILE__, __LINE__); | |
4631 | #endif | |
5bbcfd90 IM |
4632 | /* |
4633 | * The BKS might be reacquired before we have dropped | |
4634 | * PREEMPT_ACTIVE, which could trigger a second | |
4635 | * cond_resched() call. | |
4636 | */ | |
1da177e4 LT |
4637 | do { |
4638 | add_preempt_count(PREEMPT_ACTIVE); | |
4639 | schedule(); | |
4640 | sub_preempt_count(PREEMPT_ACTIVE); | |
4641 | } while (need_resched()); | |
4642 | } | |
4643 | ||
4644 | int __sched cond_resched(void) | |
4645 | { | |
9414232f IM |
4646 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
4647 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
4648 | __cond_resched(); |
4649 | return 1; | |
4650 | } | |
4651 | return 0; | |
4652 | } | |
1da177e4 LT |
4653 | EXPORT_SYMBOL(cond_resched); |
4654 | ||
4655 | /* | |
4656 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
4657 | * call schedule, and on return reacquire the lock. | |
4658 | * | |
4659 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level | |
4660 | * operations here to prevent schedule() from being called twice (once via | |
4661 | * spin_unlock(), once by hand). | |
4662 | */ | |
95cdf3b7 | 4663 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4664 | { |
6df3cecb JK |
4665 | int ret = 0; |
4666 | ||
1da177e4 LT |
4667 | if (need_lockbreak(lock)) { |
4668 | spin_unlock(lock); | |
4669 | cpu_relax(); | |
6df3cecb | 4670 | ret = 1; |
1da177e4 LT |
4671 | spin_lock(lock); |
4672 | } | |
9414232f | 4673 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
8a25d5de | 4674 | spin_release(&lock->dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4675 | _raw_spin_unlock(lock); |
4676 | preempt_enable_no_resched(); | |
4677 | __cond_resched(); | |
6df3cecb | 4678 | ret = 1; |
1da177e4 | 4679 | spin_lock(lock); |
1da177e4 | 4680 | } |
6df3cecb | 4681 | return ret; |
1da177e4 | 4682 | } |
1da177e4 LT |
4683 | EXPORT_SYMBOL(cond_resched_lock); |
4684 | ||
4685 | int __sched cond_resched_softirq(void) | |
4686 | { | |
4687 | BUG_ON(!in_softirq()); | |
4688 | ||
9414232f | 4689 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 4690 | local_bh_enable(); |
1da177e4 LT |
4691 | __cond_resched(); |
4692 | local_bh_disable(); | |
4693 | return 1; | |
4694 | } | |
4695 | return 0; | |
4696 | } | |
1da177e4 LT |
4697 | EXPORT_SYMBOL(cond_resched_softirq); |
4698 | ||
1da177e4 LT |
4699 | /** |
4700 | * yield - yield the current processor to other threads. | |
4701 | * | |
72fd4a35 | 4702 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4703 | * thread runnable and calls sys_sched_yield(). |
4704 | */ | |
4705 | void __sched yield(void) | |
4706 | { | |
4707 | set_current_state(TASK_RUNNING); | |
4708 | sys_sched_yield(); | |
4709 | } | |
1da177e4 LT |
4710 | EXPORT_SYMBOL(yield); |
4711 | ||
4712 | /* | |
4713 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so | |
4714 | * that process accounting knows that this is a task in IO wait state. | |
4715 | * | |
4716 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
4717 | * has set its backing_dev_info: the queue against which it should throttle) | |
4718 | */ | |
4719 | void __sched io_schedule(void) | |
4720 | { | |
70b97a7f | 4721 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 4722 | |
0ff92245 | 4723 | delayacct_blkio_start(); |
1da177e4 LT |
4724 | atomic_inc(&rq->nr_iowait); |
4725 | schedule(); | |
4726 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4727 | delayacct_blkio_end(); |
1da177e4 | 4728 | } |
1da177e4 LT |
4729 | EXPORT_SYMBOL(io_schedule); |
4730 | ||
4731 | long __sched io_schedule_timeout(long timeout) | |
4732 | { | |
70b97a7f | 4733 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
4734 | long ret; |
4735 | ||
0ff92245 | 4736 | delayacct_blkio_start(); |
1da177e4 LT |
4737 | atomic_inc(&rq->nr_iowait); |
4738 | ret = schedule_timeout(timeout); | |
4739 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4740 | delayacct_blkio_end(); |
1da177e4 LT |
4741 | return ret; |
4742 | } | |
4743 | ||
4744 | /** | |
4745 | * sys_sched_get_priority_max - return maximum RT priority. | |
4746 | * @policy: scheduling class. | |
4747 | * | |
4748 | * this syscall returns the maximum rt_priority that can be used | |
4749 | * by a given scheduling class. | |
4750 | */ | |
4751 | asmlinkage long sys_sched_get_priority_max(int policy) | |
4752 | { | |
4753 | int ret = -EINVAL; | |
4754 | ||
4755 | switch (policy) { | |
4756 | case SCHED_FIFO: | |
4757 | case SCHED_RR: | |
4758 | ret = MAX_USER_RT_PRIO-1; | |
4759 | break; | |
4760 | case SCHED_NORMAL: | |
b0a9499c | 4761 | case SCHED_BATCH: |
dd41f596 | 4762 | case SCHED_IDLE: |
1da177e4 LT |
4763 | ret = 0; |
4764 | break; | |
4765 | } | |
4766 | return ret; | |
4767 | } | |
4768 | ||
4769 | /** | |
4770 | * sys_sched_get_priority_min - return minimum RT priority. | |
4771 | * @policy: scheduling class. | |
4772 | * | |
4773 | * this syscall returns the minimum rt_priority that can be used | |
4774 | * by a given scheduling class. | |
4775 | */ | |
4776 | asmlinkage long sys_sched_get_priority_min(int policy) | |
4777 | { | |
4778 | int ret = -EINVAL; | |
4779 | ||
4780 | switch (policy) { | |
4781 | case SCHED_FIFO: | |
4782 | case SCHED_RR: | |
4783 | ret = 1; | |
4784 | break; | |
4785 | case SCHED_NORMAL: | |
b0a9499c | 4786 | case SCHED_BATCH: |
dd41f596 | 4787 | case SCHED_IDLE: |
1da177e4 LT |
4788 | ret = 0; |
4789 | } | |
4790 | return ret; | |
4791 | } | |
4792 | ||
4793 | /** | |
4794 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4795 | * @pid: pid of the process. | |
4796 | * @interval: userspace pointer to the timeslice value. | |
4797 | * | |
4798 | * this syscall writes the default timeslice value of a given process | |
4799 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4800 | */ | |
4801 | asmlinkage | |
4802 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
4803 | { | |
36c8b586 | 4804 | struct task_struct *p; |
a4ec24b4 | 4805 | unsigned int time_slice; |
3a5c359a | 4806 | int retval; |
1da177e4 | 4807 | struct timespec t; |
1da177e4 LT |
4808 | |
4809 | if (pid < 0) | |
3a5c359a | 4810 | return -EINVAL; |
1da177e4 LT |
4811 | |
4812 | retval = -ESRCH; | |
4813 | read_lock(&tasklist_lock); | |
4814 | p = find_process_by_pid(pid); | |
4815 | if (!p) | |
4816 | goto out_unlock; | |
4817 | ||
4818 | retval = security_task_getscheduler(p); | |
4819 | if (retval) | |
4820 | goto out_unlock; | |
4821 | ||
a4ec24b4 DA |
4822 | if (p->policy == SCHED_FIFO) |
4823 | time_slice = 0; | |
4824 | else if (p->policy == SCHED_RR) | |
4825 | time_slice = DEF_TIMESLICE; | |
4826 | else { | |
4827 | struct sched_entity *se = &p->se; | |
4828 | unsigned long flags; | |
4829 | struct rq *rq; | |
4830 | ||
4831 | rq = task_rq_lock(p, &flags); | |
4832 | time_slice = NS_TO_JIFFIES(sched_slice(cfs_rq_of(se), se)); | |
4833 | task_rq_unlock(rq, &flags); | |
4834 | } | |
1da177e4 | 4835 | read_unlock(&tasklist_lock); |
a4ec24b4 | 4836 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4837 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4838 | return retval; |
3a5c359a | 4839 | |
1da177e4 LT |
4840 | out_unlock: |
4841 | read_unlock(&tasklist_lock); | |
4842 | return retval; | |
4843 | } | |
4844 | ||
2ed6e34f | 4845 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 IM |
4846 | |
4847 | static void show_task(struct task_struct *p) | |
1da177e4 | 4848 | { |
1da177e4 | 4849 | unsigned long free = 0; |
36c8b586 | 4850 | unsigned state; |
1da177e4 | 4851 | |
1da177e4 | 4852 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 4853 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 4854 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4855 | #if BITS_PER_LONG == 32 |
1da177e4 | 4856 | if (state == TASK_RUNNING) |
cc4ea795 | 4857 | printk(KERN_CONT " running "); |
1da177e4 | 4858 | else |
cc4ea795 | 4859 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4860 | #else |
4861 | if (state == TASK_RUNNING) | |
cc4ea795 | 4862 | printk(KERN_CONT " running task "); |
1da177e4 | 4863 | else |
cc4ea795 | 4864 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4865 | #endif |
4866 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
4867 | { | |
10ebffde | 4868 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
4869 | while (!*n) |
4870 | n++; | |
10ebffde | 4871 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
4872 | } |
4873 | #endif | |
ba25f9dc PE |
4874 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
4875 | task_pid_nr(p), task_pid_nr(p->parent)); | |
1da177e4 LT |
4876 | |
4877 | if (state != TASK_RUNNING) | |
4878 | show_stack(p, NULL); | |
4879 | } | |
4880 | ||
e59e2ae2 | 4881 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4882 | { |
36c8b586 | 4883 | struct task_struct *g, *p; |
1da177e4 | 4884 | |
4bd77321 IM |
4885 | #if BITS_PER_LONG == 32 |
4886 | printk(KERN_INFO | |
4887 | " task PC stack pid father\n"); | |
1da177e4 | 4888 | #else |
4bd77321 IM |
4889 | printk(KERN_INFO |
4890 | " task PC stack pid father\n"); | |
1da177e4 LT |
4891 | #endif |
4892 | read_lock(&tasklist_lock); | |
4893 | do_each_thread(g, p) { | |
4894 | /* | |
4895 | * reset the NMI-timeout, listing all files on a slow | |
4896 | * console might take alot of time: | |
4897 | */ | |
4898 | touch_nmi_watchdog(); | |
39bc89fd | 4899 | if (!state_filter || (p->state & state_filter)) |
e59e2ae2 | 4900 | show_task(p); |
1da177e4 LT |
4901 | } while_each_thread(g, p); |
4902 | ||
04c9167f JF |
4903 | touch_all_softlockup_watchdogs(); |
4904 | ||
dd41f596 IM |
4905 | #ifdef CONFIG_SCHED_DEBUG |
4906 | sysrq_sched_debug_show(); | |
4907 | #endif | |
1da177e4 | 4908 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
4909 | /* |
4910 | * Only show locks if all tasks are dumped: | |
4911 | */ | |
4912 | if (state_filter == -1) | |
4913 | debug_show_all_locks(); | |
1da177e4 LT |
4914 | } |
4915 | ||
1df21055 IM |
4916 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4917 | { | |
dd41f596 | 4918 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4919 | } |
4920 | ||
f340c0d1 IM |
4921 | /** |
4922 | * init_idle - set up an idle thread for a given CPU | |
4923 | * @idle: task in question | |
4924 | * @cpu: cpu the idle task belongs to | |
4925 | * | |
4926 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4927 | * flag, to make booting more robust. | |
4928 | */ | |
5c1e1767 | 4929 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4930 | { |
70b97a7f | 4931 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4932 | unsigned long flags; |
4933 | ||
dd41f596 IM |
4934 | __sched_fork(idle); |
4935 | idle->se.exec_start = sched_clock(); | |
4936 | ||
b29739f9 | 4937 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 4938 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 4939 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
4940 | |
4941 | spin_lock_irqsave(&rq->lock, flags); | |
4942 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
4943 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4944 | idle->oncpu = 1; | |
4945 | #endif | |
1da177e4 LT |
4946 | spin_unlock_irqrestore(&rq->lock, flags); |
4947 | ||
4948 | /* Set the preempt count _outside_ the spinlocks! */ | |
4949 | #if defined(CONFIG_PREEMPT) && !defined(CONFIG_PREEMPT_BKL) | |
a1261f54 | 4950 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); |
1da177e4 | 4951 | #else |
a1261f54 | 4952 | task_thread_info(idle)->preempt_count = 0; |
1da177e4 | 4953 | #endif |
dd41f596 IM |
4954 | /* |
4955 | * The idle tasks have their own, simple scheduling class: | |
4956 | */ | |
4957 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
4958 | } |
4959 | ||
4960 | /* | |
4961 | * In a system that switches off the HZ timer nohz_cpu_mask | |
4962 | * indicates which cpus entered this state. This is used | |
4963 | * in the rcu update to wait only for active cpus. For system | |
4964 | * which do not switch off the HZ timer nohz_cpu_mask should | |
4965 | * always be CPU_MASK_NONE. | |
4966 | */ | |
4967 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
4968 | ||
4969 | #ifdef CONFIG_SMP | |
4970 | /* | |
4971 | * This is how migration works: | |
4972 | * | |
70b97a7f | 4973 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
4974 | * runqueue and wake up that CPU's migration thread. |
4975 | * 2) we down() the locked semaphore => thread blocks. | |
4976 | * 3) migration thread wakes up (implicitly it forces the migrated | |
4977 | * thread off the CPU) | |
4978 | * 4) it gets the migration request and checks whether the migrated | |
4979 | * task is still in the wrong runqueue. | |
4980 | * 5) if it's in the wrong runqueue then the migration thread removes | |
4981 | * it and puts it into the right queue. | |
4982 | * 6) migration thread up()s the semaphore. | |
4983 | * 7) we wake up and the migration is done. | |
4984 | */ | |
4985 | ||
4986 | /* | |
4987 | * Change a given task's CPU affinity. Migrate the thread to a | |
4988 | * proper CPU and schedule it away if the CPU it's executing on | |
4989 | * is removed from the allowed bitmask. | |
4990 | * | |
4991 | * NOTE: the caller must have a valid reference to the task, the | |
4992 | * task must not exit() & deallocate itself prematurely. The | |
4993 | * call is not atomic; no spinlocks may be held. | |
4994 | */ | |
36c8b586 | 4995 | int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask) |
1da177e4 | 4996 | { |
70b97a7f | 4997 | struct migration_req req; |
1da177e4 | 4998 | unsigned long flags; |
70b97a7f | 4999 | struct rq *rq; |
48f24c4d | 5000 | int ret = 0; |
1da177e4 LT |
5001 | |
5002 | rq = task_rq_lock(p, &flags); | |
5003 | if (!cpus_intersects(new_mask, cpu_online_map)) { | |
5004 | ret = -EINVAL; | |
5005 | goto out; | |
5006 | } | |
5007 | ||
5008 | p->cpus_allowed = new_mask; | |
5009 | /* Can the task run on the task's current CPU? If so, we're done */ | |
5010 | if (cpu_isset(task_cpu(p), new_mask)) | |
5011 | goto out; | |
5012 | ||
5013 | if (migrate_task(p, any_online_cpu(new_mask), &req)) { | |
5014 | /* Need help from migration thread: drop lock and wait. */ | |
5015 | task_rq_unlock(rq, &flags); | |
5016 | wake_up_process(rq->migration_thread); | |
5017 | wait_for_completion(&req.done); | |
5018 | tlb_migrate_finish(p->mm); | |
5019 | return 0; | |
5020 | } | |
5021 | out: | |
5022 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5023 | |
1da177e4 LT |
5024 | return ret; |
5025 | } | |
1da177e4 LT |
5026 | EXPORT_SYMBOL_GPL(set_cpus_allowed); |
5027 | ||
5028 | /* | |
5029 | * Move (not current) task off this cpu, onto dest cpu. We're doing | |
5030 | * this because either it can't run here any more (set_cpus_allowed() | |
5031 | * away from this CPU, or CPU going down), or because we're | |
5032 | * attempting to rebalance this task on exec (sched_exec). | |
5033 | * | |
5034 | * So we race with normal scheduler movements, but that's OK, as long | |
5035 | * as the task is no longer on this CPU. | |
efc30814 KK |
5036 | * |
5037 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5038 | */ |
efc30814 | 5039 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5040 | { |
70b97a7f | 5041 | struct rq *rq_dest, *rq_src; |
dd41f596 | 5042 | int ret = 0, on_rq; |
1da177e4 LT |
5043 | |
5044 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 5045 | return ret; |
1da177e4 LT |
5046 | |
5047 | rq_src = cpu_rq(src_cpu); | |
5048 | rq_dest = cpu_rq(dest_cpu); | |
5049 | ||
5050 | double_rq_lock(rq_src, rq_dest); | |
5051 | /* Already moved. */ | |
5052 | if (task_cpu(p) != src_cpu) | |
5053 | goto out; | |
5054 | /* Affinity changed (again). */ | |
5055 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
5056 | goto out; | |
5057 | ||
dd41f596 | 5058 | on_rq = p->se.on_rq; |
6e82a3be | 5059 | if (on_rq) |
2e1cb74a | 5060 | deactivate_task(rq_src, p, 0); |
6e82a3be | 5061 | |
1da177e4 | 5062 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
5063 | if (on_rq) { |
5064 | activate_task(rq_dest, p, 0); | |
5065 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 5066 | } |
efc30814 | 5067 | ret = 1; |
1da177e4 LT |
5068 | out: |
5069 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 5070 | return ret; |
1da177e4 LT |
5071 | } |
5072 | ||
5073 | /* | |
5074 | * migration_thread - this is a highprio system thread that performs | |
5075 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5076 | * another runqueue. | |
5077 | */ | |
95cdf3b7 | 5078 | static int migration_thread(void *data) |
1da177e4 | 5079 | { |
1da177e4 | 5080 | int cpu = (long)data; |
70b97a7f | 5081 | struct rq *rq; |
1da177e4 LT |
5082 | |
5083 | rq = cpu_rq(cpu); | |
5084 | BUG_ON(rq->migration_thread != current); | |
5085 | ||
5086 | set_current_state(TASK_INTERRUPTIBLE); | |
5087 | while (!kthread_should_stop()) { | |
70b97a7f | 5088 | struct migration_req *req; |
1da177e4 | 5089 | struct list_head *head; |
1da177e4 | 5090 | |
1da177e4 LT |
5091 | spin_lock_irq(&rq->lock); |
5092 | ||
5093 | if (cpu_is_offline(cpu)) { | |
5094 | spin_unlock_irq(&rq->lock); | |
5095 | goto wait_to_die; | |
5096 | } | |
5097 | ||
5098 | if (rq->active_balance) { | |
5099 | active_load_balance(rq, cpu); | |
5100 | rq->active_balance = 0; | |
5101 | } | |
5102 | ||
5103 | head = &rq->migration_queue; | |
5104 | ||
5105 | if (list_empty(head)) { | |
5106 | spin_unlock_irq(&rq->lock); | |
5107 | schedule(); | |
5108 | set_current_state(TASK_INTERRUPTIBLE); | |
5109 | continue; | |
5110 | } | |
70b97a7f | 5111 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5112 | list_del_init(head->next); |
5113 | ||
674311d5 NP |
5114 | spin_unlock(&rq->lock); |
5115 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5116 | local_irq_enable(); | |
1da177e4 LT |
5117 | |
5118 | complete(&req->done); | |
5119 | } | |
5120 | __set_current_state(TASK_RUNNING); | |
5121 | return 0; | |
5122 | ||
5123 | wait_to_die: | |
5124 | /* Wait for kthread_stop */ | |
5125 | set_current_state(TASK_INTERRUPTIBLE); | |
5126 | while (!kthread_should_stop()) { | |
5127 | schedule(); | |
5128 | set_current_state(TASK_INTERRUPTIBLE); | |
5129 | } | |
5130 | __set_current_state(TASK_RUNNING); | |
5131 | return 0; | |
5132 | } | |
5133 | ||
5134 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5135 | |
5136 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5137 | { | |
5138 | int ret; | |
5139 | ||
5140 | local_irq_disable(); | |
5141 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5142 | local_irq_enable(); | |
5143 | return ret; | |
5144 | } | |
5145 | ||
054b9108 | 5146 | /* |
3a4fa0a2 | 5147 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
5148 | * NOTE: interrupts should be disabled by the caller |
5149 | */ | |
48f24c4d | 5150 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5151 | { |
efc30814 | 5152 | unsigned long flags; |
1da177e4 | 5153 | cpumask_t mask; |
70b97a7f IM |
5154 | struct rq *rq; |
5155 | int dest_cpu; | |
1da177e4 | 5156 | |
3a5c359a AK |
5157 | do { |
5158 | /* On same node? */ | |
5159 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
5160 | cpus_and(mask, mask, p->cpus_allowed); | |
5161 | dest_cpu = any_online_cpu(mask); | |
5162 | ||
5163 | /* On any allowed CPU? */ | |
5164 | if (dest_cpu == NR_CPUS) | |
5165 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
5166 | ||
5167 | /* No more Mr. Nice Guy. */ | |
5168 | if (dest_cpu == NR_CPUS) { | |
470fd646 CW |
5169 | cpumask_t cpus_allowed = cpuset_cpus_allowed_locked(p); |
5170 | /* | |
5171 | * Try to stay on the same cpuset, where the | |
5172 | * current cpuset may be a subset of all cpus. | |
5173 | * The cpuset_cpus_allowed_locked() variant of | |
5174 | * cpuset_cpus_allowed() will not block. It must be | |
5175 | * called within calls to cpuset_lock/cpuset_unlock. | |
5176 | */ | |
3a5c359a | 5177 | rq = task_rq_lock(p, &flags); |
470fd646 | 5178 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
5179 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5180 | task_rq_unlock(rq, &flags); | |
1da177e4 | 5181 | |
3a5c359a AK |
5182 | /* |
5183 | * Don't tell them about moving exiting tasks or | |
5184 | * kernel threads (both mm NULL), since they never | |
5185 | * leave kernel. | |
5186 | */ | |
5187 | if (p->mm && printk_ratelimit()) | |
5188 | printk(KERN_INFO "process %d (%s) no " | |
5189 | "longer affine to cpu%d\n", | |
ba25f9dc | 5190 | task_pid_nr(p), p->comm, dead_cpu); |
3a5c359a | 5191 | } |
f7b4cddc | 5192 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
5193 | } |
5194 | ||
5195 | /* | |
5196 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5197 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5198 | * for performance reasons the counter is not stricly tracking tasks to | |
5199 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5200 | * to keep the global sum constant after CPU-down: | |
5201 | */ | |
70b97a7f | 5202 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5203 | { |
70b97a7f | 5204 | struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL)); |
1da177e4 LT |
5205 | unsigned long flags; |
5206 | ||
5207 | local_irq_save(flags); | |
5208 | double_rq_lock(rq_src, rq_dest); | |
5209 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5210 | rq_src->nr_uninterruptible = 0; | |
5211 | double_rq_unlock(rq_src, rq_dest); | |
5212 | local_irq_restore(flags); | |
5213 | } | |
5214 | ||
5215 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5216 | static void migrate_live_tasks(int src_cpu) | |
5217 | { | |
48f24c4d | 5218 | struct task_struct *p, *t; |
1da177e4 | 5219 | |
f7b4cddc | 5220 | read_lock(&tasklist_lock); |
1da177e4 | 5221 | |
48f24c4d IM |
5222 | do_each_thread(t, p) { |
5223 | if (p == current) | |
1da177e4 LT |
5224 | continue; |
5225 | ||
48f24c4d IM |
5226 | if (task_cpu(p) == src_cpu) |
5227 | move_task_off_dead_cpu(src_cpu, p); | |
5228 | } while_each_thread(t, p); | |
1da177e4 | 5229 | |
f7b4cddc | 5230 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5231 | } |
5232 | ||
a9957449 AD |
5233 | /* |
5234 | * activate_idle_task - move idle task to the _front_ of runqueue. | |
5235 | */ | |
5236 | static void activate_idle_task(struct task_struct *p, struct rq *rq) | |
5237 | { | |
5238 | update_rq_clock(rq); | |
5239 | ||
5240 | if (p->state == TASK_UNINTERRUPTIBLE) | |
5241 | rq->nr_uninterruptible--; | |
5242 | ||
5243 | enqueue_task(rq, p, 0); | |
5244 | inc_nr_running(p, rq); | |
5245 | } | |
5246 | ||
dd41f596 IM |
5247 | /* |
5248 | * Schedules idle task to be the next runnable task on current CPU. | |
1da177e4 | 5249 | * It does so by boosting its priority to highest possible and adding it to |
48f24c4d | 5250 | * the _front_ of the runqueue. Used by CPU offline code. |
1da177e4 LT |
5251 | */ |
5252 | void sched_idle_next(void) | |
5253 | { | |
48f24c4d | 5254 | int this_cpu = smp_processor_id(); |
70b97a7f | 5255 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5256 | struct task_struct *p = rq->idle; |
5257 | unsigned long flags; | |
5258 | ||
5259 | /* cpu has to be offline */ | |
48f24c4d | 5260 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5261 | |
48f24c4d IM |
5262 | /* |
5263 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5264 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5265 | */ |
5266 | spin_lock_irqsave(&rq->lock, flags); | |
5267 | ||
dd41f596 | 5268 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d IM |
5269 | |
5270 | /* Add idle task to the _front_ of its priority queue: */ | |
dd41f596 | 5271 | activate_idle_task(p, rq); |
1da177e4 LT |
5272 | |
5273 | spin_unlock_irqrestore(&rq->lock, flags); | |
5274 | } | |
5275 | ||
48f24c4d IM |
5276 | /* |
5277 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5278 | * offline. |
5279 | */ | |
5280 | void idle_task_exit(void) | |
5281 | { | |
5282 | struct mm_struct *mm = current->active_mm; | |
5283 | ||
5284 | BUG_ON(cpu_online(smp_processor_id())); | |
5285 | ||
5286 | if (mm != &init_mm) | |
5287 | switch_mm(mm, &init_mm, current); | |
5288 | mmdrop(mm); | |
5289 | } | |
5290 | ||
054b9108 | 5291 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5292 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5293 | { |
70b97a7f | 5294 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5295 | |
5296 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5297 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5298 | |
5299 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5300 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5301 | |
48f24c4d | 5302 | get_task_struct(p); |
1da177e4 LT |
5303 | |
5304 | /* | |
5305 | * Drop lock around migration; if someone else moves it, | |
5306 | * that's OK. No task can be added to this CPU, so iteration is | |
5307 | * fine. | |
5308 | */ | |
f7b4cddc | 5309 | spin_unlock_irq(&rq->lock); |
48f24c4d | 5310 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 5311 | spin_lock_irq(&rq->lock); |
1da177e4 | 5312 | |
48f24c4d | 5313 | put_task_struct(p); |
1da177e4 LT |
5314 | } |
5315 | ||
5316 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5317 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5318 | { | |
70b97a7f | 5319 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5320 | struct task_struct *next; |
48f24c4d | 5321 | |
dd41f596 IM |
5322 | for ( ; ; ) { |
5323 | if (!rq->nr_running) | |
5324 | break; | |
a8e504d2 | 5325 | update_rq_clock(rq); |
ff95f3df | 5326 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
5327 | if (!next) |
5328 | break; | |
5329 | migrate_dead(dead_cpu, next); | |
e692ab53 | 5330 | |
1da177e4 LT |
5331 | } |
5332 | } | |
5333 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5334 | ||
e692ab53 NP |
5335 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5336 | ||
5337 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5338 | { |
5339 | .procname = "sched_domain", | |
c57baf1e | 5340 | .mode = 0555, |
e0361851 | 5341 | }, |
e692ab53 NP |
5342 | {0,}, |
5343 | }; | |
5344 | ||
5345 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 5346 | { |
c57baf1e | 5347 | .ctl_name = CTL_KERN, |
e0361851 | 5348 | .procname = "kernel", |
c57baf1e | 5349 | .mode = 0555, |
e0361851 AD |
5350 | .child = sd_ctl_dir, |
5351 | }, | |
e692ab53 NP |
5352 | {0,}, |
5353 | }; | |
5354 | ||
5355 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5356 | { | |
5357 | struct ctl_table *entry = | |
5cf9f062 | 5358 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5359 | |
e692ab53 NP |
5360 | return entry; |
5361 | } | |
5362 | ||
6382bc90 MM |
5363 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5364 | { | |
cd790076 | 5365 | struct ctl_table *entry; |
6382bc90 | 5366 | |
cd790076 MM |
5367 | /* |
5368 | * In the intermediate directories, both the child directory and | |
5369 | * procname are dynamically allocated and could fail but the mode | |
5370 | * will always be set. In the lowest directory the names are | |
5371 | * static strings and all have proc handlers. | |
5372 | */ | |
5373 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5374 | if (entry->child) |
5375 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5376 | if (entry->proc_handler == NULL) |
5377 | kfree(entry->procname); | |
5378 | } | |
6382bc90 MM |
5379 | |
5380 | kfree(*tablep); | |
5381 | *tablep = NULL; | |
5382 | } | |
5383 | ||
e692ab53 | 5384 | static void |
e0361851 | 5385 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5386 | const char *procname, void *data, int maxlen, |
5387 | mode_t mode, proc_handler *proc_handler) | |
5388 | { | |
e692ab53 NP |
5389 | entry->procname = procname; |
5390 | entry->data = data; | |
5391 | entry->maxlen = maxlen; | |
5392 | entry->mode = mode; | |
5393 | entry->proc_handler = proc_handler; | |
5394 | } | |
5395 | ||
5396 | static struct ctl_table * | |
5397 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5398 | { | |
ace8b3d6 | 5399 | struct ctl_table *table = sd_alloc_ctl_entry(12); |
e692ab53 | 5400 | |
ad1cdc1d MM |
5401 | if (table == NULL) |
5402 | return NULL; | |
5403 | ||
e0361851 | 5404 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5405 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5406 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5407 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5408 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5409 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5410 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5411 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5412 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5413 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5414 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5415 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5416 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5417 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5418 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5419 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5420 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5421 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5422 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5423 | &sd->cache_nice_tries, |
5424 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5425 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5426 | sizeof(int), 0644, proc_dointvec_minmax); |
6323469f | 5427 | /* &table[11] is terminator */ |
e692ab53 NP |
5428 | |
5429 | return table; | |
5430 | } | |
5431 | ||
8401f775 | 5432 | static ctl_table * sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5433 | { |
5434 | struct ctl_table *entry, *table; | |
5435 | struct sched_domain *sd; | |
5436 | int domain_num = 0, i; | |
5437 | char buf[32]; | |
5438 | ||
5439 | for_each_domain(cpu, sd) | |
5440 | domain_num++; | |
5441 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5442 | if (table == NULL) |
5443 | return NULL; | |
e692ab53 NP |
5444 | |
5445 | i = 0; | |
5446 | for_each_domain(cpu, sd) { | |
5447 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5448 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5449 | entry->mode = 0555; |
e692ab53 NP |
5450 | entry->child = sd_alloc_ctl_domain_table(sd); |
5451 | entry++; | |
5452 | i++; | |
5453 | } | |
5454 | return table; | |
5455 | } | |
5456 | ||
5457 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5458 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
5459 | { |
5460 | int i, cpu_num = num_online_cpus(); | |
5461 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
5462 | char buf[32]; | |
5463 | ||
ad1cdc1d MM |
5464 | if (entry == NULL) |
5465 | return; | |
5466 | ||
e692ab53 NP |
5467 | sd_ctl_dir[0].child = entry; |
5468 | ||
97b6ea7b | 5469 | for_each_online_cpu(i) { |
e692ab53 | 5470 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5471 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5472 | entry->mode = 0555; |
e692ab53 | 5473 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5474 | entry++; |
e692ab53 NP |
5475 | } |
5476 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); | |
5477 | } | |
6382bc90 MM |
5478 | |
5479 | static void unregister_sched_domain_sysctl(void) | |
5480 | { | |
5481 | unregister_sysctl_table(sd_sysctl_header); | |
5482 | sd_sysctl_header = NULL; | |
5483 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
5484 | } | |
e692ab53 | 5485 | #else |
6382bc90 MM |
5486 | static void register_sched_domain_sysctl(void) |
5487 | { | |
5488 | } | |
5489 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5490 | { |
5491 | } | |
5492 | #endif | |
5493 | ||
1da177e4 LT |
5494 | /* |
5495 | * migration_call - callback that gets triggered when a CPU is added. | |
5496 | * Here we can start up the necessary migration thread for the new CPU. | |
5497 | */ | |
48f24c4d IM |
5498 | static int __cpuinit |
5499 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5500 | { |
1da177e4 | 5501 | struct task_struct *p; |
48f24c4d | 5502 | int cpu = (long)hcpu; |
1da177e4 | 5503 | unsigned long flags; |
70b97a7f | 5504 | struct rq *rq; |
1da177e4 LT |
5505 | |
5506 | switch (action) { | |
5be9361c GS |
5507 | case CPU_LOCK_ACQUIRE: |
5508 | mutex_lock(&sched_hotcpu_mutex); | |
5509 | break; | |
5510 | ||
1da177e4 | 5511 | case CPU_UP_PREPARE: |
8bb78442 | 5512 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 5513 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
5514 | if (IS_ERR(p)) |
5515 | return NOTIFY_BAD; | |
1da177e4 LT |
5516 | kthread_bind(p, cpu); |
5517 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5518 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 5519 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
5520 | task_rq_unlock(rq, &flags); |
5521 | cpu_rq(cpu)->migration_thread = p; | |
5522 | break; | |
48f24c4d | 5523 | |
1da177e4 | 5524 | case CPU_ONLINE: |
8bb78442 | 5525 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 5526 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 LT |
5527 | wake_up_process(cpu_rq(cpu)->migration_thread); |
5528 | break; | |
48f24c4d | 5529 | |
1da177e4 LT |
5530 | #ifdef CONFIG_HOTPLUG_CPU |
5531 | case CPU_UP_CANCELED: | |
8bb78442 | 5532 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5533 | if (!cpu_rq(cpu)->migration_thread) |
5534 | break; | |
1da177e4 | 5535 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
5536 | kthread_bind(cpu_rq(cpu)->migration_thread, |
5537 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
5538 | kthread_stop(cpu_rq(cpu)->migration_thread); |
5539 | cpu_rq(cpu)->migration_thread = NULL; | |
5540 | break; | |
48f24c4d | 5541 | |
1da177e4 | 5542 | case CPU_DEAD: |
8bb78442 | 5543 | case CPU_DEAD_FROZEN: |
470fd646 | 5544 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
5545 | migrate_live_tasks(cpu); |
5546 | rq = cpu_rq(cpu); | |
5547 | kthread_stop(rq->migration_thread); | |
5548 | rq->migration_thread = NULL; | |
5549 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 5550 | spin_lock_irq(&rq->lock); |
a8e504d2 | 5551 | update_rq_clock(rq); |
2e1cb74a | 5552 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 5553 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
5554 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5555 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 5556 | migrate_dead_tasks(cpu); |
d2da272a | 5557 | spin_unlock_irq(&rq->lock); |
470fd646 | 5558 | cpuset_unlock(); |
1da177e4 LT |
5559 | migrate_nr_uninterruptible(rq); |
5560 | BUG_ON(rq->nr_running != 0); | |
5561 | ||
5562 | /* No need to migrate the tasks: it was best-effort if | |
5be9361c | 5563 | * they didn't take sched_hotcpu_mutex. Just wake up |
1da177e4 LT |
5564 | * the requestors. */ |
5565 | spin_lock_irq(&rq->lock); | |
5566 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
5567 | struct migration_req *req; |
5568 | ||
1da177e4 | 5569 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5570 | struct migration_req, list); |
1da177e4 LT |
5571 | list_del_init(&req->list); |
5572 | complete(&req->done); | |
5573 | } | |
5574 | spin_unlock_irq(&rq->lock); | |
5575 | break; | |
5576 | #endif | |
5be9361c GS |
5577 | case CPU_LOCK_RELEASE: |
5578 | mutex_unlock(&sched_hotcpu_mutex); | |
5579 | break; | |
1da177e4 LT |
5580 | } |
5581 | return NOTIFY_OK; | |
5582 | } | |
5583 | ||
5584 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
5585 | * happens before everything else. | |
5586 | */ | |
26c2143b | 5587 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5588 | .notifier_call = migration_call, |
5589 | .priority = 10 | |
5590 | }; | |
5591 | ||
5592 | int __init migration_init(void) | |
5593 | { | |
5594 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5595 | int err; |
48f24c4d IM |
5596 | |
5597 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5598 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5599 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5600 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5601 | register_cpu_notifier(&migration_notifier); | |
48f24c4d | 5602 | |
1da177e4 LT |
5603 | return 0; |
5604 | } | |
5605 | #endif | |
5606 | ||
5607 | #ifdef CONFIG_SMP | |
476f3534 CL |
5608 | |
5609 | /* Number of possible processor ids */ | |
5610 | int nr_cpu_ids __read_mostly = NR_CPUS; | |
5611 | EXPORT_SYMBOL(nr_cpu_ids); | |
5612 | ||
3e9830dc | 5613 | #ifdef CONFIG_SCHED_DEBUG |
1da177e4 LT |
5614 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5615 | { | |
5616 | int level = 0; | |
5617 | ||
41c7ce9a NP |
5618 | if (!sd) { |
5619 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5620 | return; | |
5621 | } | |
5622 | ||
1da177e4 LT |
5623 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5624 | ||
5625 | do { | |
5626 | int i; | |
5627 | char str[NR_CPUS]; | |
5628 | struct sched_group *group = sd->groups; | |
5629 | cpumask_t groupmask; | |
5630 | ||
5631 | cpumask_scnprintf(str, NR_CPUS, sd->span); | |
5632 | cpus_clear(groupmask); | |
5633 | ||
5634 | printk(KERN_DEBUG); | |
5635 | for (i = 0; i < level + 1; i++) | |
5636 | printk(" "); | |
5637 | printk("domain %d: ", level); | |
5638 | ||
5639 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
5640 | printk("does not load-balance\n"); | |
5641 | if (sd->parent) | |
33859f7f MOS |
5642 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5643 | " has parent"); | |
1da177e4 LT |
5644 | break; |
5645 | } | |
5646 | ||
5647 | printk("span %s\n", str); | |
5648 | ||
5649 | if (!cpu_isset(cpu, sd->span)) | |
33859f7f MOS |
5650 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5651 | "CPU%d\n", cpu); | |
1da177e4 | 5652 | if (!cpu_isset(cpu, group->cpumask)) |
33859f7f MOS |
5653 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5654 | " CPU%d\n", cpu); | |
1da177e4 LT |
5655 | |
5656 | printk(KERN_DEBUG); | |
5657 | for (i = 0; i < level + 2; i++) | |
5658 | printk(" "); | |
5659 | printk("groups:"); | |
5660 | do { | |
5661 | if (!group) { | |
5662 | printk("\n"); | |
5663 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
5664 | break; | |
5665 | } | |
5666 | ||
5517d86b | 5667 | if (!group->__cpu_power) { |
cc4ea795 | 5668 | printk(KERN_CONT "\n"); |
33859f7f MOS |
5669 | printk(KERN_ERR "ERROR: domain->cpu_power not " |
5670 | "set\n"); | |
26797a34 | 5671 | break; |
1da177e4 LT |
5672 | } |
5673 | ||
5674 | if (!cpus_weight(group->cpumask)) { | |
cc4ea795 | 5675 | printk(KERN_CONT "\n"); |
1da177e4 | 5676 | printk(KERN_ERR "ERROR: empty group\n"); |
26797a34 | 5677 | break; |
1da177e4 LT |
5678 | } |
5679 | ||
5680 | if (cpus_intersects(groupmask, group->cpumask)) { | |
cc4ea795 | 5681 | printk(KERN_CONT "\n"); |
1da177e4 | 5682 | printk(KERN_ERR "ERROR: repeated CPUs\n"); |
26797a34 | 5683 | break; |
1da177e4 LT |
5684 | } |
5685 | ||
5686 | cpus_or(groupmask, groupmask, group->cpumask); | |
5687 | ||
5688 | cpumask_scnprintf(str, NR_CPUS, group->cpumask); | |
cc4ea795 | 5689 | printk(KERN_CONT " %s", str); |
1da177e4 LT |
5690 | |
5691 | group = group->next; | |
5692 | } while (group != sd->groups); | |
cc4ea795 | 5693 | printk(KERN_CONT "\n"); |
1da177e4 LT |
5694 | |
5695 | if (!cpus_equal(sd->span, groupmask)) | |
33859f7f MOS |
5696 | printk(KERN_ERR "ERROR: groups don't span " |
5697 | "domain->span\n"); | |
1da177e4 LT |
5698 | |
5699 | level++; | |
5700 | sd = sd->parent; | |
33859f7f MOS |
5701 | if (!sd) |
5702 | continue; | |
1da177e4 | 5703 | |
33859f7f MOS |
5704 | if (!cpus_subset(groupmask, sd->span)) |
5705 | printk(KERN_ERR "ERROR: parent span is not a superset " | |
5706 | "of domain->span\n"); | |
1da177e4 LT |
5707 | |
5708 | } while (sd); | |
5709 | } | |
5710 | #else | |
48f24c4d | 5711 | # define sched_domain_debug(sd, cpu) do { } while (0) |
1da177e4 LT |
5712 | #endif |
5713 | ||
1a20ff27 | 5714 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
5715 | { |
5716 | if (cpus_weight(sd->span) == 1) | |
5717 | return 1; | |
5718 | ||
5719 | /* Following flags need at least 2 groups */ | |
5720 | if (sd->flags & (SD_LOAD_BALANCE | | |
5721 | SD_BALANCE_NEWIDLE | | |
5722 | SD_BALANCE_FORK | | |
89c4710e SS |
5723 | SD_BALANCE_EXEC | |
5724 | SD_SHARE_CPUPOWER | | |
5725 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5726 | if (sd->groups != sd->groups->next) |
5727 | return 0; | |
5728 | } | |
5729 | ||
5730 | /* Following flags don't use groups */ | |
5731 | if (sd->flags & (SD_WAKE_IDLE | | |
5732 | SD_WAKE_AFFINE | | |
5733 | SD_WAKE_BALANCE)) | |
5734 | return 0; | |
5735 | ||
5736 | return 1; | |
5737 | } | |
5738 | ||
48f24c4d IM |
5739 | static int |
5740 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5741 | { |
5742 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5743 | ||
5744 | if (sd_degenerate(parent)) | |
5745 | return 1; | |
5746 | ||
5747 | if (!cpus_equal(sd->span, parent->span)) | |
5748 | return 0; | |
5749 | ||
5750 | /* Does parent contain flags not in child? */ | |
5751 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
5752 | if (cflags & SD_WAKE_AFFINE) | |
5753 | pflags &= ~SD_WAKE_BALANCE; | |
5754 | /* Flags needing groups don't count if only 1 group in parent */ | |
5755 | if (parent->groups == parent->groups->next) { | |
5756 | pflags &= ~(SD_LOAD_BALANCE | | |
5757 | SD_BALANCE_NEWIDLE | | |
5758 | SD_BALANCE_FORK | | |
89c4710e SS |
5759 | SD_BALANCE_EXEC | |
5760 | SD_SHARE_CPUPOWER | | |
5761 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
5762 | } |
5763 | if (~cflags & pflags) | |
5764 | return 0; | |
5765 | ||
5766 | return 1; | |
5767 | } | |
5768 | ||
1da177e4 LT |
5769 | /* |
5770 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must | |
5771 | * hold the hotplug lock. | |
5772 | */ | |
9c1cfda2 | 5773 | static void cpu_attach_domain(struct sched_domain *sd, int cpu) |
1da177e4 | 5774 | { |
70b97a7f | 5775 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5776 | struct sched_domain *tmp; |
5777 | ||
5778 | /* Remove the sched domains which do not contribute to scheduling. */ | |
5779 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
5780 | struct sched_domain *parent = tmp->parent; | |
5781 | if (!parent) | |
5782 | break; | |
1a848870 | 5783 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5784 | tmp->parent = parent->parent; |
1a848870 SS |
5785 | if (parent->parent) |
5786 | parent->parent->child = tmp; | |
5787 | } | |
245af2c7 SS |
5788 | } |
5789 | ||
1a848870 | 5790 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 5791 | sd = sd->parent; |
1a848870 SS |
5792 | if (sd) |
5793 | sd->child = NULL; | |
5794 | } | |
1da177e4 LT |
5795 | |
5796 | sched_domain_debug(sd, cpu); | |
5797 | ||
674311d5 | 5798 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
5799 | } |
5800 | ||
5801 | /* cpus with isolated domains */ | |
67af63a6 | 5802 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
5803 | |
5804 | /* Setup the mask of cpus configured for isolated domains */ | |
5805 | static int __init isolated_cpu_setup(char *str) | |
5806 | { | |
5807 | int ints[NR_CPUS], i; | |
5808 | ||
5809 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
5810 | cpus_clear(cpu_isolated_map); | |
5811 | for (i = 1; i <= ints[0]; i++) | |
5812 | if (ints[i] < NR_CPUS) | |
5813 | cpu_set(ints[i], cpu_isolated_map); | |
5814 | return 1; | |
5815 | } | |
5816 | ||
8927f494 | 5817 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
5818 | |
5819 | /* | |
6711cab4 SS |
5820 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
5821 | * to a function which identifies what group(along with sched group) a CPU | |
5822 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
5823 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
5824 | * |
5825 | * init_sched_build_groups will build a circular linked list of the groups | |
5826 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5827 | * and ->cpu_power to 0. | |
5828 | */ | |
a616058b | 5829 | static void |
6711cab4 SS |
5830 | init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map, |
5831 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, | |
5832 | struct sched_group **sg)) | |
1da177e4 LT |
5833 | { |
5834 | struct sched_group *first = NULL, *last = NULL; | |
5835 | cpumask_t covered = CPU_MASK_NONE; | |
5836 | int i; | |
5837 | ||
5838 | for_each_cpu_mask(i, span) { | |
6711cab4 SS |
5839 | struct sched_group *sg; |
5840 | int group = group_fn(i, cpu_map, &sg); | |
1da177e4 LT |
5841 | int j; |
5842 | ||
5843 | if (cpu_isset(i, covered)) | |
5844 | continue; | |
5845 | ||
5846 | sg->cpumask = CPU_MASK_NONE; | |
5517d86b | 5847 | sg->__cpu_power = 0; |
1da177e4 LT |
5848 | |
5849 | for_each_cpu_mask(j, span) { | |
6711cab4 | 5850 | if (group_fn(j, cpu_map, NULL) != group) |
1da177e4 LT |
5851 | continue; |
5852 | ||
5853 | cpu_set(j, covered); | |
5854 | cpu_set(j, sg->cpumask); | |
5855 | } | |
5856 | if (!first) | |
5857 | first = sg; | |
5858 | if (last) | |
5859 | last->next = sg; | |
5860 | last = sg; | |
5861 | } | |
5862 | last->next = first; | |
5863 | } | |
5864 | ||
9c1cfda2 | 5865 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 5866 | |
9c1cfda2 | 5867 | #ifdef CONFIG_NUMA |
198e2f18 | 5868 | |
9c1cfda2 JH |
5869 | /** |
5870 | * find_next_best_node - find the next node to include in a sched_domain | |
5871 | * @node: node whose sched_domain we're building | |
5872 | * @used_nodes: nodes already in the sched_domain | |
5873 | * | |
5874 | * Find the next node to include in a given scheduling domain. Simply | |
5875 | * finds the closest node not already in the @used_nodes map. | |
5876 | * | |
5877 | * Should use nodemask_t. | |
5878 | */ | |
5879 | static int find_next_best_node(int node, unsigned long *used_nodes) | |
5880 | { | |
5881 | int i, n, val, min_val, best_node = 0; | |
5882 | ||
5883 | min_val = INT_MAX; | |
5884 | ||
5885 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5886 | /* Start at @node */ | |
5887 | n = (node + i) % MAX_NUMNODES; | |
5888 | ||
5889 | if (!nr_cpus_node(n)) | |
5890 | continue; | |
5891 | ||
5892 | /* Skip already used nodes */ | |
5893 | if (test_bit(n, used_nodes)) | |
5894 | continue; | |
5895 | ||
5896 | /* Simple min distance search */ | |
5897 | val = node_distance(node, n); | |
5898 | ||
5899 | if (val < min_val) { | |
5900 | min_val = val; | |
5901 | best_node = n; | |
5902 | } | |
5903 | } | |
5904 | ||
5905 | set_bit(best_node, used_nodes); | |
5906 | return best_node; | |
5907 | } | |
5908 | ||
5909 | /** | |
5910 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
5911 | * @node: node whose cpumask we're constructing | |
5912 | * @size: number of nodes to include in this span | |
5913 | * | |
5914 | * Given a node, construct a good cpumask for its sched_domain to span. It | |
5915 | * should be one that prevents unnecessary balancing, but also spreads tasks | |
5916 | * out optimally. | |
5917 | */ | |
5918 | static cpumask_t sched_domain_node_span(int node) | |
5919 | { | |
9c1cfda2 | 5920 | DECLARE_BITMAP(used_nodes, MAX_NUMNODES); |
48f24c4d IM |
5921 | cpumask_t span, nodemask; |
5922 | int i; | |
9c1cfda2 JH |
5923 | |
5924 | cpus_clear(span); | |
5925 | bitmap_zero(used_nodes, MAX_NUMNODES); | |
5926 | ||
5927 | nodemask = node_to_cpumask(node); | |
5928 | cpus_or(span, span, nodemask); | |
5929 | set_bit(node, used_nodes); | |
5930 | ||
5931 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
5932 | int next_node = find_next_best_node(node, used_nodes); | |
48f24c4d | 5933 | |
9c1cfda2 JH |
5934 | nodemask = node_to_cpumask(next_node); |
5935 | cpus_or(span, span, nodemask); | |
5936 | } | |
5937 | ||
5938 | return span; | |
5939 | } | |
5940 | #endif | |
5941 | ||
5c45bf27 | 5942 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 5943 | |
9c1cfda2 | 5944 | /* |
48f24c4d | 5945 | * SMT sched-domains: |
9c1cfda2 | 5946 | */ |
1da177e4 LT |
5947 | #ifdef CONFIG_SCHED_SMT |
5948 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 5949 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 5950 | |
6711cab4 SS |
5951 | static int cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, |
5952 | struct sched_group **sg) | |
1da177e4 | 5953 | { |
6711cab4 SS |
5954 | if (sg) |
5955 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
5956 | return cpu; |
5957 | } | |
5958 | #endif | |
5959 | ||
48f24c4d IM |
5960 | /* |
5961 | * multi-core sched-domains: | |
5962 | */ | |
1e9f28fa SS |
5963 | #ifdef CONFIG_SCHED_MC |
5964 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 5965 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
5966 | #endif |
5967 | ||
5968 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
6711cab4 SS |
5969 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5970 | struct sched_group **sg) | |
1e9f28fa | 5971 | { |
6711cab4 | 5972 | int group; |
d5a7430d | 5973 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 5974 | cpus_and(mask, mask, *cpu_map); |
6711cab4 SS |
5975 | group = first_cpu(mask); |
5976 | if (sg) | |
5977 | *sg = &per_cpu(sched_group_core, group); | |
5978 | return group; | |
1e9f28fa SS |
5979 | } |
5980 | #elif defined(CONFIG_SCHED_MC) | |
6711cab4 SS |
5981 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5982 | struct sched_group **sg) | |
1e9f28fa | 5983 | { |
6711cab4 SS |
5984 | if (sg) |
5985 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
5986 | return cpu; |
5987 | } | |
5988 | #endif | |
5989 | ||
1da177e4 | 5990 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 5991 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 5992 | |
6711cab4 SS |
5993 | static int cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, |
5994 | struct sched_group **sg) | |
1da177e4 | 5995 | { |
6711cab4 | 5996 | int group; |
48f24c4d | 5997 | #ifdef CONFIG_SCHED_MC |
1e9f28fa | 5998 | cpumask_t mask = cpu_coregroup_map(cpu); |
a616058b | 5999 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 6000 | group = first_cpu(mask); |
1e9f28fa | 6001 | #elif defined(CONFIG_SCHED_SMT) |
d5a7430d | 6002 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 6003 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 6004 | group = first_cpu(mask); |
1da177e4 | 6005 | #else |
6711cab4 | 6006 | group = cpu; |
1da177e4 | 6007 | #endif |
6711cab4 SS |
6008 | if (sg) |
6009 | *sg = &per_cpu(sched_group_phys, group); | |
6010 | return group; | |
1da177e4 LT |
6011 | } |
6012 | ||
6013 | #ifdef CONFIG_NUMA | |
1da177e4 | 6014 | /* |
9c1cfda2 JH |
6015 | * The init_sched_build_groups can't handle what we want to do with node |
6016 | * groups, so roll our own. Now each node has its own list of groups which | |
6017 | * gets dynamically allocated. | |
1da177e4 | 6018 | */ |
9c1cfda2 | 6019 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
d1b55138 | 6020 | static struct sched_group **sched_group_nodes_bycpu[NR_CPUS]; |
1da177e4 | 6021 | |
9c1cfda2 | 6022 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 6023 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 6024 | |
6711cab4 SS |
6025 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
6026 | struct sched_group **sg) | |
9c1cfda2 | 6027 | { |
6711cab4 SS |
6028 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu)); |
6029 | int group; | |
6030 | ||
6031 | cpus_and(nodemask, nodemask, *cpu_map); | |
6032 | group = first_cpu(nodemask); | |
6033 | ||
6034 | if (sg) | |
6035 | *sg = &per_cpu(sched_group_allnodes, group); | |
6036 | return group; | |
1da177e4 | 6037 | } |
6711cab4 | 6038 | |
08069033 SS |
6039 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6040 | { | |
6041 | struct sched_group *sg = group_head; | |
6042 | int j; | |
6043 | ||
6044 | if (!sg) | |
6045 | return; | |
3a5c359a AK |
6046 | do { |
6047 | for_each_cpu_mask(j, sg->cpumask) { | |
6048 | struct sched_domain *sd; | |
08069033 | 6049 | |
3a5c359a AK |
6050 | sd = &per_cpu(phys_domains, j); |
6051 | if (j != first_cpu(sd->groups->cpumask)) { | |
6052 | /* | |
6053 | * Only add "power" once for each | |
6054 | * physical package. | |
6055 | */ | |
6056 | continue; | |
6057 | } | |
08069033 | 6058 | |
3a5c359a AK |
6059 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
6060 | } | |
6061 | sg = sg->next; | |
6062 | } while (sg != group_head); | |
08069033 | 6063 | } |
1da177e4 LT |
6064 | #endif |
6065 | ||
a616058b | 6066 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6067 | /* Free memory allocated for various sched_group structures */ |
6068 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6069 | { | |
a616058b | 6070 | int cpu, i; |
51888ca2 SV |
6071 | |
6072 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
6073 | struct sched_group **sched_group_nodes |
6074 | = sched_group_nodes_bycpu[cpu]; | |
6075 | ||
51888ca2 SV |
6076 | if (!sched_group_nodes) |
6077 | continue; | |
6078 | ||
6079 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6080 | cpumask_t nodemask = node_to_cpumask(i); | |
6081 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; | |
6082 | ||
6083 | cpus_and(nodemask, nodemask, *cpu_map); | |
6084 | if (cpus_empty(nodemask)) | |
6085 | continue; | |
6086 | ||
6087 | if (sg == NULL) | |
6088 | continue; | |
6089 | sg = sg->next; | |
6090 | next_sg: | |
6091 | oldsg = sg; | |
6092 | sg = sg->next; | |
6093 | kfree(oldsg); | |
6094 | if (oldsg != sched_group_nodes[i]) | |
6095 | goto next_sg; | |
6096 | } | |
6097 | kfree(sched_group_nodes); | |
6098 | sched_group_nodes_bycpu[cpu] = NULL; | |
6099 | } | |
51888ca2 | 6100 | } |
a616058b SS |
6101 | #else |
6102 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6103 | { | |
6104 | } | |
6105 | #endif | |
51888ca2 | 6106 | |
89c4710e SS |
6107 | /* |
6108 | * Initialize sched groups cpu_power. | |
6109 | * | |
6110 | * cpu_power indicates the capacity of sched group, which is used while | |
6111 | * distributing the load between different sched groups in a sched domain. | |
6112 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6113 | * there are asymmetries in the topology. If there are asymmetries, group | |
6114 | * having more cpu_power will pickup more load compared to the group having | |
6115 | * less cpu_power. | |
6116 | * | |
6117 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
6118 | * the maximum number of tasks a group can handle in the presence of other idle | |
6119 | * or lightly loaded groups in the same sched domain. | |
6120 | */ | |
6121 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6122 | { | |
6123 | struct sched_domain *child; | |
6124 | struct sched_group *group; | |
6125 | ||
6126 | WARN_ON(!sd || !sd->groups); | |
6127 | ||
6128 | if (cpu != first_cpu(sd->groups->cpumask)) | |
6129 | return; | |
6130 | ||
6131 | child = sd->child; | |
6132 | ||
5517d86b ED |
6133 | sd->groups->__cpu_power = 0; |
6134 | ||
89c4710e SS |
6135 | /* |
6136 | * For perf policy, if the groups in child domain share resources | |
6137 | * (for example cores sharing some portions of the cache hierarchy | |
6138 | * or SMT), then set this domain groups cpu_power such that each group | |
6139 | * can handle only one task, when there are other idle groups in the | |
6140 | * same sched domain. | |
6141 | */ | |
6142 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
6143 | (child->flags & | |
6144 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 6145 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
6146 | return; |
6147 | } | |
6148 | ||
89c4710e SS |
6149 | /* |
6150 | * add cpu_power of each child group to this groups cpu_power | |
6151 | */ | |
6152 | group = child->groups; | |
6153 | do { | |
5517d86b | 6154 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
6155 | group = group->next; |
6156 | } while (group != child->groups); | |
6157 | } | |
6158 | ||
1da177e4 | 6159 | /* |
1a20ff27 DG |
6160 | * Build sched domains for a given set of cpus and attach the sched domains |
6161 | * to the individual cpus | |
1da177e4 | 6162 | */ |
51888ca2 | 6163 | static int build_sched_domains(const cpumask_t *cpu_map) |
1da177e4 LT |
6164 | { |
6165 | int i; | |
d1b55138 JH |
6166 | #ifdef CONFIG_NUMA |
6167 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 6168 | int sd_allnodes = 0; |
d1b55138 JH |
6169 | |
6170 | /* | |
6171 | * Allocate the per-node list of sched groups | |
6172 | */ | |
5cf9f062 | 6173 | sched_group_nodes = kcalloc(MAX_NUMNODES, sizeof(struct sched_group *), |
d3a5aa98 | 6174 | GFP_KERNEL); |
d1b55138 JH |
6175 | if (!sched_group_nodes) { |
6176 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 6177 | return -ENOMEM; |
d1b55138 JH |
6178 | } |
6179 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
6180 | #endif | |
1da177e4 LT |
6181 | |
6182 | /* | |
1a20ff27 | 6183 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 6184 | */ |
1a20ff27 | 6185 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6186 | struct sched_domain *sd = NULL, *p; |
6187 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(i)); | |
6188 | ||
1a20ff27 | 6189 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6190 | |
6191 | #ifdef CONFIG_NUMA | |
dd41f596 IM |
6192 | if (cpus_weight(*cpu_map) > |
6193 | SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) { | |
9c1cfda2 JH |
6194 | sd = &per_cpu(allnodes_domains, i); |
6195 | *sd = SD_ALLNODES_INIT; | |
6196 | sd->span = *cpu_map; | |
6711cab4 | 6197 | cpu_to_allnodes_group(i, cpu_map, &sd->groups); |
9c1cfda2 | 6198 | p = sd; |
6711cab4 | 6199 | sd_allnodes = 1; |
9c1cfda2 JH |
6200 | } else |
6201 | p = NULL; | |
6202 | ||
1da177e4 | 6203 | sd = &per_cpu(node_domains, i); |
1da177e4 | 6204 | *sd = SD_NODE_INIT; |
9c1cfda2 JH |
6205 | sd->span = sched_domain_node_span(cpu_to_node(i)); |
6206 | sd->parent = p; | |
1a848870 SS |
6207 | if (p) |
6208 | p->child = sd; | |
9c1cfda2 | 6209 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
6210 | #endif |
6211 | ||
6212 | p = sd; | |
6213 | sd = &per_cpu(phys_domains, i); | |
1da177e4 LT |
6214 | *sd = SD_CPU_INIT; |
6215 | sd->span = nodemask; | |
6216 | sd->parent = p; | |
1a848870 SS |
6217 | if (p) |
6218 | p->child = sd; | |
6711cab4 | 6219 | cpu_to_phys_group(i, cpu_map, &sd->groups); |
1da177e4 | 6220 | |
1e9f28fa SS |
6221 | #ifdef CONFIG_SCHED_MC |
6222 | p = sd; | |
6223 | sd = &per_cpu(core_domains, i); | |
1e9f28fa SS |
6224 | *sd = SD_MC_INIT; |
6225 | sd->span = cpu_coregroup_map(i); | |
6226 | cpus_and(sd->span, sd->span, *cpu_map); | |
6227 | sd->parent = p; | |
1a848870 | 6228 | p->child = sd; |
6711cab4 | 6229 | cpu_to_core_group(i, cpu_map, &sd->groups); |
1e9f28fa SS |
6230 | #endif |
6231 | ||
1da177e4 LT |
6232 | #ifdef CONFIG_SCHED_SMT |
6233 | p = sd; | |
6234 | sd = &per_cpu(cpu_domains, i); | |
1da177e4 | 6235 | *sd = SD_SIBLING_INIT; |
d5a7430d | 6236 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6237 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 6238 | sd->parent = p; |
1a848870 | 6239 | p->child = sd; |
6711cab4 | 6240 | cpu_to_cpu_group(i, cpu_map, &sd->groups); |
1da177e4 LT |
6241 | #endif |
6242 | } | |
6243 | ||
6244 | #ifdef CONFIG_SCHED_SMT | |
6245 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 6246 | for_each_cpu_mask(i, *cpu_map) { |
d5a7430d | 6247 | cpumask_t this_sibling_map = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6248 | cpus_and(this_sibling_map, this_sibling_map, *cpu_map); |
1da177e4 LT |
6249 | if (i != first_cpu(this_sibling_map)) |
6250 | continue; | |
6251 | ||
dd41f596 IM |
6252 | init_sched_build_groups(this_sibling_map, cpu_map, |
6253 | &cpu_to_cpu_group); | |
1da177e4 LT |
6254 | } |
6255 | #endif | |
6256 | ||
1e9f28fa SS |
6257 | #ifdef CONFIG_SCHED_MC |
6258 | /* Set up multi-core groups */ | |
6259 | for_each_cpu_mask(i, *cpu_map) { | |
6260 | cpumask_t this_core_map = cpu_coregroup_map(i); | |
6261 | cpus_and(this_core_map, this_core_map, *cpu_map); | |
6262 | if (i != first_cpu(this_core_map)) | |
6263 | continue; | |
dd41f596 IM |
6264 | init_sched_build_groups(this_core_map, cpu_map, |
6265 | &cpu_to_core_group); | |
1e9f28fa SS |
6266 | } |
6267 | #endif | |
6268 | ||
1da177e4 LT |
6269 | /* Set up physical groups */ |
6270 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6271 | cpumask_t nodemask = node_to_cpumask(i); | |
6272 | ||
1a20ff27 | 6273 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6274 | if (cpus_empty(nodemask)) |
6275 | continue; | |
6276 | ||
6711cab4 | 6277 | init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group); |
1da177e4 LT |
6278 | } |
6279 | ||
6280 | #ifdef CONFIG_NUMA | |
6281 | /* Set up node groups */ | |
6711cab4 | 6282 | if (sd_allnodes) |
dd41f596 IM |
6283 | init_sched_build_groups(*cpu_map, cpu_map, |
6284 | &cpu_to_allnodes_group); | |
9c1cfda2 JH |
6285 | |
6286 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6287 | /* Set up node groups */ | |
6288 | struct sched_group *sg, *prev; | |
6289 | cpumask_t nodemask = node_to_cpumask(i); | |
6290 | cpumask_t domainspan; | |
6291 | cpumask_t covered = CPU_MASK_NONE; | |
6292 | int j; | |
6293 | ||
6294 | cpus_and(nodemask, nodemask, *cpu_map); | |
d1b55138 JH |
6295 | if (cpus_empty(nodemask)) { |
6296 | sched_group_nodes[i] = NULL; | |
9c1cfda2 | 6297 | continue; |
d1b55138 | 6298 | } |
9c1cfda2 JH |
6299 | |
6300 | domainspan = sched_domain_node_span(i); | |
6301 | cpus_and(domainspan, domainspan, *cpu_map); | |
6302 | ||
15f0b676 | 6303 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
6304 | if (!sg) { |
6305 | printk(KERN_WARNING "Can not alloc domain group for " | |
6306 | "node %d\n", i); | |
6307 | goto error; | |
6308 | } | |
9c1cfda2 JH |
6309 | sched_group_nodes[i] = sg; |
6310 | for_each_cpu_mask(j, nodemask) { | |
6311 | struct sched_domain *sd; | |
9761eea8 | 6312 | |
9c1cfda2 JH |
6313 | sd = &per_cpu(node_domains, j); |
6314 | sd->groups = sg; | |
9c1cfda2 | 6315 | } |
5517d86b | 6316 | sg->__cpu_power = 0; |
9c1cfda2 | 6317 | sg->cpumask = nodemask; |
51888ca2 | 6318 | sg->next = sg; |
9c1cfda2 JH |
6319 | cpus_or(covered, covered, nodemask); |
6320 | prev = sg; | |
6321 | ||
6322 | for (j = 0; j < MAX_NUMNODES; j++) { | |
6323 | cpumask_t tmp, notcovered; | |
6324 | int n = (i + j) % MAX_NUMNODES; | |
6325 | ||
6326 | cpus_complement(notcovered, covered); | |
6327 | cpus_and(tmp, notcovered, *cpu_map); | |
6328 | cpus_and(tmp, tmp, domainspan); | |
6329 | if (cpus_empty(tmp)) | |
6330 | break; | |
6331 | ||
6332 | nodemask = node_to_cpumask(n); | |
6333 | cpus_and(tmp, tmp, nodemask); | |
6334 | if (cpus_empty(tmp)) | |
6335 | continue; | |
6336 | ||
15f0b676 SV |
6337 | sg = kmalloc_node(sizeof(struct sched_group), |
6338 | GFP_KERNEL, i); | |
9c1cfda2 JH |
6339 | if (!sg) { |
6340 | printk(KERN_WARNING | |
6341 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 6342 | goto error; |
9c1cfda2 | 6343 | } |
5517d86b | 6344 | sg->__cpu_power = 0; |
9c1cfda2 | 6345 | sg->cpumask = tmp; |
51888ca2 | 6346 | sg->next = prev->next; |
9c1cfda2 JH |
6347 | cpus_or(covered, covered, tmp); |
6348 | prev->next = sg; | |
6349 | prev = sg; | |
6350 | } | |
9c1cfda2 | 6351 | } |
1da177e4 LT |
6352 | #endif |
6353 | ||
6354 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 6355 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 6356 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6357 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
6358 | ||
89c4710e | 6359 | init_sched_groups_power(i, sd); |
5c45bf27 | 6360 | } |
1da177e4 | 6361 | #endif |
1e9f28fa | 6362 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 6363 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6364 | struct sched_domain *sd = &per_cpu(core_domains, i); |
6365 | ||
89c4710e | 6366 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
6367 | } |
6368 | #endif | |
1e9f28fa | 6369 | |
5c45bf27 | 6370 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6371 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
6372 | ||
89c4710e | 6373 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6374 | } |
6375 | ||
9c1cfda2 | 6376 | #ifdef CONFIG_NUMA |
08069033 SS |
6377 | for (i = 0; i < MAX_NUMNODES; i++) |
6378 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 6379 | |
6711cab4 SS |
6380 | if (sd_allnodes) { |
6381 | struct sched_group *sg; | |
f712c0c7 | 6382 | |
6711cab4 | 6383 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg); |
f712c0c7 SS |
6384 | init_numa_sched_groups_power(sg); |
6385 | } | |
9c1cfda2 JH |
6386 | #endif |
6387 | ||
1da177e4 | 6388 | /* Attach the domains */ |
1a20ff27 | 6389 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6390 | struct sched_domain *sd; |
6391 | #ifdef CONFIG_SCHED_SMT | |
6392 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
6393 | #elif defined(CONFIG_SCHED_MC) |
6394 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
6395 | #else |
6396 | sd = &per_cpu(phys_domains, i); | |
6397 | #endif | |
6398 | cpu_attach_domain(sd, i); | |
6399 | } | |
51888ca2 SV |
6400 | |
6401 | return 0; | |
6402 | ||
a616058b | 6403 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6404 | error: |
6405 | free_sched_groups(cpu_map); | |
6406 | return -ENOMEM; | |
a616058b | 6407 | #endif |
1da177e4 | 6408 | } |
029190c5 PJ |
6409 | |
6410 | static cpumask_t *doms_cur; /* current sched domains */ | |
6411 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
6412 | ||
6413 | /* | |
6414 | * Special case: If a kmalloc of a doms_cur partition (array of | |
6415 | * cpumask_t) fails, then fallback to a single sched domain, | |
6416 | * as determined by the single cpumask_t fallback_doms. | |
6417 | */ | |
6418 | static cpumask_t fallback_doms; | |
6419 | ||
1a20ff27 DG |
6420 | /* |
6421 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. | |
029190c5 PJ |
6422 | * For now this just excludes isolated cpus, but could be used to |
6423 | * exclude other special cases in the future. | |
1a20ff27 | 6424 | */ |
51888ca2 | 6425 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 6426 | { |
029190c5 PJ |
6427 | ndoms_cur = 1; |
6428 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
6429 | if (!doms_cur) | |
6430 | doms_cur = &fallback_doms; | |
6431 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
6382bc90 | 6432 | register_sched_domain_sysctl(); |
029190c5 | 6433 | return build_sched_domains(doms_cur); |
1a20ff27 DG |
6434 | } |
6435 | ||
6436 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map) | |
1da177e4 | 6437 | { |
51888ca2 | 6438 | free_sched_groups(cpu_map); |
9c1cfda2 | 6439 | } |
1da177e4 | 6440 | |
1a20ff27 DG |
6441 | /* |
6442 | * Detach sched domains from a group of cpus specified in cpu_map | |
6443 | * These cpus will now be attached to the NULL domain | |
6444 | */ | |
858119e1 | 6445 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6446 | { |
6447 | int i; | |
6448 | ||
6382bc90 MM |
6449 | unregister_sched_domain_sysctl(); |
6450 | ||
1a20ff27 DG |
6451 | for_each_cpu_mask(i, *cpu_map) |
6452 | cpu_attach_domain(NULL, i); | |
6453 | synchronize_sched(); | |
6454 | arch_destroy_sched_domains(cpu_map); | |
6455 | } | |
6456 | ||
029190c5 PJ |
6457 | /* |
6458 | * Partition sched domains as specified by the 'ndoms_new' | |
6459 | * cpumasks in the array doms_new[] of cpumasks. This compares | |
6460 | * doms_new[] to the current sched domain partitioning, doms_cur[]. | |
6461 | * It destroys each deleted domain and builds each new domain. | |
6462 | * | |
6463 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
6464 | * The masks don't intersect (don't overlap.) We should setup one | |
6465 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6466 | * not be load balanced. If the same cpumask appears both in the | |
6467 | * current 'doms_cur' domains and in the new 'doms_new', we can leave | |
6468 | * it as it is. | |
6469 | * | |
6470 | * The passed in 'doms_new' should be kmalloc'd. This routine takes | |
6471 | * ownership of it and will kfree it when done with it. If the caller | |
6472 | * failed the kmalloc call, then it can pass in doms_new == NULL, | |
6473 | * and partition_sched_domains() will fallback to the single partition | |
6474 | * 'fallback_doms'. | |
6475 | * | |
6476 | * Call with hotplug lock held | |
6477 | */ | |
6478 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new) | |
6479 | { | |
6480 | int i, j; | |
6481 | ||
6482 | if (doms_new == NULL) { | |
6483 | ndoms_new = 1; | |
6484 | doms_new = &fallback_doms; | |
6485 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
6486 | } | |
6487 | ||
6488 | /* Destroy deleted domains */ | |
6489 | for (i = 0; i < ndoms_cur; i++) { | |
6490 | for (j = 0; j < ndoms_new; j++) { | |
6491 | if (cpus_equal(doms_cur[i], doms_new[j])) | |
6492 | goto match1; | |
6493 | } | |
6494 | /* no match - a current sched domain not in new doms_new[] */ | |
6495 | detach_destroy_domains(doms_cur + i); | |
6496 | match1: | |
6497 | ; | |
6498 | } | |
6499 | ||
6500 | /* Build new domains */ | |
6501 | for (i = 0; i < ndoms_new; i++) { | |
6502 | for (j = 0; j < ndoms_cur; j++) { | |
6503 | if (cpus_equal(doms_new[i], doms_cur[j])) | |
6504 | goto match2; | |
6505 | } | |
6506 | /* no match - add a new doms_new */ | |
6507 | build_sched_domains(doms_new + i); | |
6508 | match2: | |
6509 | ; | |
6510 | } | |
6511 | ||
6512 | /* Remember the new sched domains */ | |
6513 | if (doms_cur != &fallback_doms) | |
6514 | kfree(doms_cur); | |
6515 | doms_cur = doms_new; | |
6516 | ndoms_cur = ndoms_new; | |
6517 | } | |
6518 | ||
5c45bf27 | 6519 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
6707de00 | 6520 | static int arch_reinit_sched_domains(void) |
5c45bf27 SS |
6521 | { |
6522 | int err; | |
6523 | ||
5be9361c | 6524 | mutex_lock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6525 | detach_destroy_domains(&cpu_online_map); |
6526 | err = arch_init_sched_domains(&cpu_online_map); | |
5be9361c | 6527 | mutex_unlock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6528 | |
6529 | return err; | |
6530 | } | |
6531 | ||
6532 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6533 | { | |
6534 | int ret; | |
6535 | ||
6536 | if (buf[0] != '0' && buf[0] != '1') | |
6537 | return -EINVAL; | |
6538 | ||
6539 | if (smt) | |
6540 | sched_smt_power_savings = (buf[0] == '1'); | |
6541 | else | |
6542 | sched_mc_power_savings = (buf[0] == '1'); | |
6543 | ||
6544 | ret = arch_reinit_sched_domains(); | |
6545 | ||
6546 | return ret ? ret : count; | |
6547 | } | |
6548 | ||
5c45bf27 SS |
6549 | #ifdef CONFIG_SCHED_MC |
6550 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
6551 | { | |
6552 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
6553 | } | |
48f24c4d IM |
6554 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
6555 | const char *buf, size_t count) | |
5c45bf27 SS |
6556 | { |
6557 | return sched_power_savings_store(buf, count, 0); | |
6558 | } | |
6707de00 AB |
6559 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
6560 | sched_mc_power_savings_store); | |
5c45bf27 SS |
6561 | #endif |
6562 | ||
6563 | #ifdef CONFIG_SCHED_SMT | |
6564 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
6565 | { | |
6566 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
6567 | } | |
48f24c4d IM |
6568 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
6569 | const char *buf, size_t count) | |
5c45bf27 SS |
6570 | { |
6571 | return sched_power_savings_store(buf, count, 1); | |
6572 | } | |
6707de00 AB |
6573 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
6574 | sched_smt_power_savings_store); | |
6575 | #endif | |
6576 | ||
6577 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
6578 | { | |
6579 | int err = 0; | |
6580 | ||
6581 | #ifdef CONFIG_SCHED_SMT | |
6582 | if (smt_capable()) | |
6583 | err = sysfs_create_file(&cls->kset.kobj, | |
6584 | &attr_sched_smt_power_savings.attr); | |
6585 | #endif | |
6586 | #ifdef CONFIG_SCHED_MC | |
6587 | if (!err && mc_capable()) | |
6588 | err = sysfs_create_file(&cls->kset.kobj, | |
6589 | &attr_sched_mc_power_savings.attr); | |
6590 | #endif | |
6591 | return err; | |
6592 | } | |
5c45bf27 SS |
6593 | #endif |
6594 | ||
1da177e4 LT |
6595 | /* |
6596 | * Force a reinitialization of the sched domains hierarchy. The domains | |
6597 | * and groups cannot be updated in place without racing with the balancing | |
41c7ce9a | 6598 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
6599 | * which will prevent rebalancing while the sched domains are recalculated. |
6600 | */ | |
6601 | static int update_sched_domains(struct notifier_block *nfb, | |
6602 | unsigned long action, void *hcpu) | |
6603 | { | |
1da177e4 LT |
6604 | switch (action) { |
6605 | case CPU_UP_PREPARE: | |
8bb78442 | 6606 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 6607 | case CPU_DOWN_PREPARE: |
8bb78442 | 6608 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 6609 | detach_destroy_domains(&cpu_online_map); |
1da177e4 LT |
6610 | return NOTIFY_OK; |
6611 | ||
6612 | case CPU_UP_CANCELED: | |
8bb78442 | 6613 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 6614 | case CPU_DOWN_FAILED: |
8bb78442 | 6615 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 6616 | case CPU_ONLINE: |
8bb78442 | 6617 | case CPU_ONLINE_FROZEN: |
1da177e4 | 6618 | case CPU_DEAD: |
8bb78442 | 6619 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
6620 | /* |
6621 | * Fall through and re-initialise the domains. | |
6622 | */ | |
6623 | break; | |
6624 | default: | |
6625 | return NOTIFY_DONE; | |
6626 | } | |
6627 | ||
6628 | /* The hotplug lock is already held by cpu_up/cpu_down */ | |
1a20ff27 | 6629 | arch_init_sched_domains(&cpu_online_map); |
1da177e4 LT |
6630 | |
6631 | return NOTIFY_OK; | |
6632 | } | |
1da177e4 LT |
6633 | |
6634 | void __init sched_init_smp(void) | |
6635 | { | |
5c1e1767 NP |
6636 | cpumask_t non_isolated_cpus; |
6637 | ||
5be9361c | 6638 | mutex_lock(&sched_hotcpu_mutex); |
1a20ff27 | 6639 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 6640 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
6641 | if (cpus_empty(non_isolated_cpus)) |
6642 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
5be9361c | 6643 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
6644 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
6645 | hotcpu_notifier(update_sched_domains, 0); | |
5c1e1767 NP |
6646 | |
6647 | /* Move init over to a non-isolated CPU */ | |
6648 | if (set_cpus_allowed(current, non_isolated_cpus) < 0) | |
6649 | BUG(); | |
1da177e4 LT |
6650 | } |
6651 | #else | |
6652 | void __init sched_init_smp(void) | |
6653 | { | |
6654 | } | |
6655 | #endif /* CONFIG_SMP */ | |
6656 | ||
6657 | int in_sched_functions(unsigned long addr) | |
6658 | { | |
6659 | /* Linker adds these: start and end of __sched functions */ | |
6660 | extern char __sched_text_start[], __sched_text_end[]; | |
48f24c4d | 6661 | |
1da177e4 LT |
6662 | return in_lock_functions(addr) || |
6663 | (addr >= (unsigned long)__sched_text_start | |
6664 | && addr < (unsigned long)__sched_text_end); | |
6665 | } | |
6666 | ||
a9957449 | 6667 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
6668 | { |
6669 | cfs_rq->tasks_timeline = RB_ROOT; | |
dd41f596 IM |
6670 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6671 | cfs_rq->rq = rq; | |
6672 | #endif | |
67e9fb2a | 6673 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
6674 | } |
6675 | ||
1da177e4 LT |
6676 | void __init sched_init(void) |
6677 | { | |
476f3534 | 6678 | int highest_cpu = 0; |
dd41f596 IM |
6679 | int i, j; |
6680 | ||
0a945022 | 6681 | for_each_possible_cpu(i) { |
dd41f596 | 6682 | struct rt_prio_array *array; |
70b97a7f | 6683 | struct rq *rq; |
1da177e4 LT |
6684 | |
6685 | rq = cpu_rq(i); | |
6686 | spin_lock_init(&rq->lock); | |
fcb99371 | 6687 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 6688 | rq->nr_running = 0; |
dd41f596 IM |
6689 | rq->clock = 1; |
6690 | init_cfs_rq(&rq->cfs, rq); | |
6691 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6692 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); | |
3a252015 IM |
6693 | { |
6694 | struct cfs_rq *cfs_rq = &per_cpu(init_cfs_rq, i); | |
6695 | struct sched_entity *se = | |
6696 | &per_cpu(init_sched_entity, i); | |
6697 | ||
6698 | init_cfs_rq_p[i] = cfs_rq; | |
6699 | init_cfs_rq(cfs_rq, rq); | |
4cf86d77 | 6700 | cfs_rq->tg = &init_task_group; |
3a252015 | 6701 | list_add(&cfs_rq->leaf_cfs_rq_list, |
29f59db3 SV |
6702 | &rq->leaf_cfs_rq_list); |
6703 | ||
3a252015 IM |
6704 | init_sched_entity_p[i] = se; |
6705 | se->cfs_rq = &rq->cfs; | |
6706 | se->my_q = cfs_rq; | |
4cf86d77 | 6707 | se->load.weight = init_task_group_load; |
9b5b7751 | 6708 | se->load.inv_weight = |
4cf86d77 | 6709 | div64_64(1ULL<<32, init_task_group_load); |
3a252015 IM |
6710 | se->parent = NULL; |
6711 | } | |
4cf86d77 | 6712 | init_task_group.shares = init_task_group_load; |
5cb350ba | 6713 | spin_lock_init(&init_task_group.lock); |
dd41f596 | 6714 | #endif |
1da177e4 | 6715 | |
dd41f596 IM |
6716 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6717 | rq->cpu_load[j] = 0; | |
1da177e4 | 6718 | #ifdef CONFIG_SMP |
41c7ce9a | 6719 | rq->sd = NULL; |
1da177e4 | 6720 | rq->active_balance = 0; |
dd41f596 | 6721 | rq->next_balance = jiffies; |
1da177e4 | 6722 | rq->push_cpu = 0; |
0a2966b4 | 6723 | rq->cpu = i; |
1da177e4 LT |
6724 | rq->migration_thread = NULL; |
6725 | INIT_LIST_HEAD(&rq->migration_queue); | |
6726 | #endif | |
6727 | atomic_set(&rq->nr_iowait, 0); | |
6728 | ||
dd41f596 IM |
6729 | array = &rq->rt.active; |
6730 | for (j = 0; j < MAX_RT_PRIO; j++) { | |
6731 | INIT_LIST_HEAD(array->queue + j); | |
6732 | __clear_bit(j, array->bitmap); | |
1da177e4 | 6733 | } |
476f3534 | 6734 | highest_cpu = i; |
dd41f596 IM |
6735 | /* delimiter for bitsearch: */ |
6736 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
1da177e4 LT |
6737 | } |
6738 | ||
2dd73a4f | 6739 | set_load_weight(&init_task); |
b50f60ce | 6740 | |
e107be36 AK |
6741 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6742 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6743 | #endif | |
6744 | ||
c9819f45 | 6745 | #ifdef CONFIG_SMP |
476f3534 | 6746 | nr_cpu_ids = highest_cpu + 1; |
c9819f45 CL |
6747 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); |
6748 | #endif | |
6749 | ||
b50f60ce HC |
6750 | #ifdef CONFIG_RT_MUTEXES |
6751 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
6752 | #endif | |
6753 | ||
1da177e4 LT |
6754 | /* |
6755 | * The boot idle thread does lazy MMU switching as well: | |
6756 | */ | |
6757 | atomic_inc(&init_mm.mm_count); | |
6758 | enter_lazy_tlb(&init_mm, current); | |
6759 | ||
6760 | /* | |
6761 | * Make us the idle thread. Technically, schedule() should not be | |
6762 | * called from this thread, however somewhere below it might be, | |
6763 | * but because we are the idle thread, we just pick up running again | |
6764 | * when this runqueue becomes "idle". | |
6765 | */ | |
6766 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
6767 | /* |
6768 | * During early bootup we pretend to be a normal task: | |
6769 | */ | |
6770 | current->sched_class = &fair_sched_class; | |
1da177e4 LT |
6771 | } |
6772 | ||
6773 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
6774 | void __might_sleep(char *file, int line) | |
6775 | { | |
48f24c4d | 6776 | #ifdef in_atomic |
1da177e4 LT |
6777 | static unsigned long prev_jiffy; /* ratelimiting */ |
6778 | ||
6779 | if ((in_atomic() || irqs_disabled()) && | |
6780 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
6781 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6782 | return; | |
6783 | prev_jiffy = jiffies; | |
91368d73 | 6784 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
6785 | " context at %s:%d\n", file, line); |
6786 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
6787 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 6788 | debug_show_held_locks(current); |
3117df04 IM |
6789 | if (irqs_disabled()) |
6790 | print_irqtrace_events(current); | |
1da177e4 LT |
6791 | dump_stack(); |
6792 | } | |
6793 | #endif | |
6794 | } | |
6795 | EXPORT_SYMBOL(__might_sleep); | |
6796 | #endif | |
6797 | ||
6798 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
6799 | static void normalize_task(struct rq *rq, struct task_struct *p) |
6800 | { | |
6801 | int on_rq; | |
6802 | update_rq_clock(rq); | |
6803 | on_rq = p->se.on_rq; | |
6804 | if (on_rq) | |
6805 | deactivate_task(rq, p, 0); | |
6806 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
6807 | if (on_rq) { | |
6808 | activate_task(rq, p, 0); | |
6809 | resched_task(rq->curr); | |
6810 | } | |
6811 | } | |
6812 | ||
1da177e4 LT |
6813 | void normalize_rt_tasks(void) |
6814 | { | |
a0f98a1c | 6815 | struct task_struct *g, *p; |
1da177e4 | 6816 | unsigned long flags; |
70b97a7f | 6817 | struct rq *rq; |
1da177e4 LT |
6818 | |
6819 | read_lock_irq(&tasklist_lock); | |
a0f98a1c | 6820 | do_each_thread(g, p) { |
178be793 IM |
6821 | /* |
6822 | * Only normalize user tasks: | |
6823 | */ | |
6824 | if (!p->mm) | |
6825 | continue; | |
6826 | ||
6cfb0d5d | 6827 | p->se.exec_start = 0; |
6cfb0d5d | 6828 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 6829 | p->se.wait_start = 0; |
dd41f596 | 6830 | p->se.sleep_start = 0; |
dd41f596 | 6831 | p->se.block_start = 0; |
6cfb0d5d | 6832 | #endif |
dd41f596 IM |
6833 | task_rq(p)->clock = 0; |
6834 | ||
6835 | if (!rt_task(p)) { | |
6836 | /* | |
6837 | * Renice negative nice level userspace | |
6838 | * tasks back to 0: | |
6839 | */ | |
6840 | if (TASK_NICE(p) < 0 && p->mm) | |
6841 | set_user_nice(p, 0); | |
1da177e4 | 6842 | continue; |
dd41f596 | 6843 | } |
1da177e4 | 6844 | |
b29739f9 IM |
6845 | spin_lock_irqsave(&p->pi_lock, flags); |
6846 | rq = __task_rq_lock(p); | |
1da177e4 | 6847 | |
178be793 | 6848 | normalize_task(rq, p); |
3a5e4dc1 | 6849 | |
b29739f9 IM |
6850 | __task_rq_unlock(rq); |
6851 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
a0f98a1c IM |
6852 | } while_each_thread(g, p); |
6853 | ||
1da177e4 LT |
6854 | read_unlock_irq(&tasklist_lock); |
6855 | } | |
6856 | ||
6857 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
6858 | |
6859 | #ifdef CONFIG_IA64 | |
6860 | /* | |
6861 | * These functions are only useful for the IA64 MCA handling. | |
6862 | * | |
6863 | * They can only be called when the whole system has been | |
6864 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6865 | * activity can take place. Using them for anything else would | |
6866 | * be a serious bug, and as a result, they aren't even visible | |
6867 | * under any other configuration. | |
6868 | */ | |
6869 | ||
6870 | /** | |
6871 | * curr_task - return the current task for a given cpu. | |
6872 | * @cpu: the processor in question. | |
6873 | * | |
6874 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6875 | */ | |
36c8b586 | 6876 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
6877 | { |
6878 | return cpu_curr(cpu); | |
6879 | } | |
6880 | ||
6881 | /** | |
6882 | * set_curr_task - set the current task for a given cpu. | |
6883 | * @cpu: the processor in question. | |
6884 | * @p: the task pointer to set. | |
6885 | * | |
6886 | * Description: This function must only be used when non-maskable interrupts | |
6887 | * are serviced on a separate stack. It allows the architecture to switch the | |
6888 | * notion of the current task on a cpu in a non-blocking manner. This function | |
6889 | * must be called with all CPU's synchronized, and interrupts disabled, the | |
6890 | * and caller must save the original value of the current task (see | |
6891 | * curr_task() above) and restore that value before reenabling interrupts and | |
6892 | * re-starting the system. | |
6893 | * | |
6894 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6895 | */ | |
36c8b586 | 6896 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
6897 | { |
6898 | cpu_curr(cpu) = p; | |
6899 | } | |
6900 | ||
6901 | #endif | |
29f59db3 SV |
6902 | |
6903 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6904 | ||
29f59db3 | 6905 | /* allocate runqueue etc for a new task group */ |
4cf86d77 | 6906 | struct task_group *sched_create_group(void) |
29f59db3 | 6907 | { |
4cf86d77 | 6908 | struct task_group *tg; |
29f59db3 SV |
6909 | struct cfs_rq *cfs_rq; |
6910 | struct sched_entity *se; | |
9b5b7751 | 6911 | struct rq *rq; |
29f59db3 SV |
6912 | int i; |
6913 | ||
29f59db3 SV |
6914 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); |
6915 | if (!tg) | |
6916 | return ERR_PTR(-ENOMEM); | |
6917 | ||
9b5b7751 | 6918 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
6919 | if (!tg->cfs_rq) |
6920 | goto err; | |
9b5b7751 | 6921 | tg->se = kzalloc(sizeof(se) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
6922 | if (!tg->se) |
6923 | goto err; | |
6924 | ||
6925 | for_each_possible_cpu(i) { | |
9b5b7751 | 6926 | rq = cpu_rq(i); |
29f59db3 SV |
6927 | |
6928 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), GFP_KERNEL, | |
6929 | cpu_to_node(i)); | |
6930 | if (!cfs_rq) | |
6931 | goto err; | |
6932 | ||
6933 | se = kmalloc_node(sizeof(struct sched_entity), GFP_KERNEL, | |
6934 | cpu_to_node(i)); | |
6935 | if (!se) | |
6936 | goto err; | |
6937 | ||
6938 | memset(cfs_rq, 0, sizeof(struct cfs_rq)); | |
6939 | memset(se, 0, sizeof(struct sched_entity)); | |
6940 | ||
6941 | tg->cfs_rq[i] = cfs_rq; | |
6942 | init_cfs_rq(cfs_rq, rq); | |
6943 | cfs_rq->tg = tg; | |
29f59db3 SV |
6944 | |
6945 | tg->se[i] = se; | |
6946 | se->cfs_rq = &rq->cfs; | |
6947 | se->my_q = cfs_rq; | |
6948 | se->load.weight = NICE_0_LOAD; | |
6949 | se->load.inv_weight = div64_64(1ULL<<32, NICE_0_LOAD); | |
6950 | se->parent = NULL; | |
6951 | } | |
6952 | ||
9b5b7751 SV |
6953 | for_each_possible_cpu(i) { |
6954 | rq = cpu_rq(i); | |
6955 | cfs_rq = tg->cfs_rq[i]; | |
6956 | list_add_rcu(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
6957 | } | |
29f59db3 | 6958 | |
9b5b7751 | 6959 | tg->shares = NICE_0_LOAD; |
5cb350ba | 6960 | spin_lock_init(&tg->lock); |
29f59db3 | 6961 | |
9b5b7751 | 6962 | return tg; |
29f59db3 SV |
6963 | |
6964 | err: | |
6965 | for_each_possible_cpu(i) { | |
a65914b3 | 6966 | if (tg->cfs_rq) |
29f59db3 | 6967 | kfree(tg->cfs_rq[i]); |
a65914b3 | 6968 | if (tg->se) |
29f59db3 SV |
6969 | kfree(tg->se[i]); |
6970 | } | |
a65914b3 IM |
6971 | kfree(tg->cfs_rq); |
6972 | kfree(tg->se); | |
6973 | kfree(tg); | |
29f59db3 SV |
6974 | |
6975 | return ERR_PTR(-ENOMEM); | |
6976 | } | |
6977 | ||
9b5b7751 SV |
6978 | /* rcu callback to free various structures associated with a task group */ |
6979 | static void free_sched_group(struct rcu_head *rhp) | |
29f59db3 | 6980 | { |
9b5b7751 | 6981 | struct cfs_rq *cfs_rq = container_of(rhp, struct cfs_rq, rcu); |
4cf86d77 | 6982 | struct task_group *tg = cfs_rq->tg; |
29f59db3 SV |
6983 | struct sched_entity *se; |
6984 | int i; | |
6985 | ||
29f59db3 SV |
6986 | /* now it should be safe to free those cfs_rqs */ |
6987 | for_each_possible_cpu(i) { | |
6988 | cfs_rq = tg->cfs_rq[i]; | |
6989 | kfree(cfs_rq); | |
6990 | ||
6991 | se = tg->se[i]; | |
6992 | kfree(se); | |
6993 | } | |
6994 | ||
6995 | kfree(tg->cfs_rq); | |
6996 | kfree(tg->se); | |
6997 | kfree(tg); | |
6998 | } | |
6999 | ||
9b5b7751 | 7000 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7001 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 7002 | { |
9b5b7751 SV |
7003 | struct cfs_rq *cfs_rq; |
7004 | int i; | |
29f59db3 | 7005 | |
9b5b7751 SV |
7006 | for_each_possible_cpu(i) { |
7007 | cfs_rq = tg->cfs_rq[i]; | |
7008 | list_del_rcu(&cfs_rq->leaf_cfs_rq_list); | |
7009 | } | |
7010 | ||
7011 | cfs_rq = tg->cfs_rq[0]; | |
7012 | ||
7013 | /* wait for possible concurrent references to cfs_rqs complete */ | |
7014 | call_rcu(&cfs_rq->rcu, free_sched_group); | |
29f59db3 SV |
7015 | } |
7016 | ||
9b5b7751 | 7017 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7018 | * The caller of this function should have put the task in its new group |
7019 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7020 | * reflect its new group. | |
9b5b7751 SV |
7021 | */ |
7022 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
7023 | { |
7024 | int on_rq, running; | |
7025 | unsigned long flags; | |
7026 | struct rq *rq; | |
7027 | ||
7028 | rq = task_rq_lock(tsk, &flags); | |
7029 | ||
7030 | if (tsk->sched_class != &fair_sched_class) | |
7031 | goto done; | |
7032 | ||
7033 | update_rq_clock(rq); | |
7034 | ||
7035 | running = task_running(rq, tsk); | |
7036 | on_rq = tsk->se.on_rq; | |
7037 | ||
83b699ed | 7038 | if (on_rq) { |
29f59db3 | 7039 | dequeue_task(rq, tsk, 0); |
83b699ed SV |
7040 | if (unlikely(running)) |
7041 | tsk->sched_class->put_prev_task(rq, tsk); | |
7042 | } | |
29f59db3 SV |
7043 | |
7044 | set_task_cfs_rq(tsk); | |
7045 | ||
83b699ed SV |
7046 | if (on_rq) { |
7047 | if (unlikely(running)) | |
7048 | tsk->sched_class->set_curr_task(rq); | |
7074badb | 7049 | enqueue_task(rq, tsk, 0); |
83b699ed | 7050 | } |
29f59db3 SV |
7051 | |
7052 | done: | |
7053 | task_rq_unlock(rq, &flags); | |
7054 | } | |
7055 | ||
7056 | static void set_se_shares(struct sched_entity *se, unsigned long shares) | |
7057 | { | |
7058 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
7059 | struct rq *rq = cfs_rq->rq; | |
7060 | int on_rq; | |
7061 | ||
7062 | spin_lock_irq(&rq->lock); | |
7063 | ||
7064 | on_rq = se->on_rq; | |
7065 | if (on_rq) | |
7066 | dequeue_entity(cfs_rq, se, 0); | |
7067 | ||
7068 | se->load.weight = shares; | |
7069 | se->load.inv_weight = div64_64((1ULL<<32), shares); | |
7070 | ||
7071 | if (on_rq) | |
7072 | enqueue_entity(cfs_rq, se, 0); | |
7073 | ||
7074 | spin_unlock_irq(&rq->lock); | |
7075 | } | |
7076 | ||
4cf86d77 | 7077 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
7078 | { |
7079 | int i; | |
29f59db3 | 7080 | |
5cb350ba | 7081 | spin_lock(&tg->lock); |
9b5b7751 | 7082 | if (tg->shares == shares) |
5cb350ba | 7083 | goto done; |
29f59db3 | 7084 | |
9b5b7751 | 7085 | tg->shares = shares; |
29f59db3 | 7086 | for_each_possible_cpu(i) |
9b5b7751 | 7087 | set_se_shares(tg->se[i], shares); |
29f59db3 | 7088 | |
5cb350ba DG |
7089 | done: |
7090 | spin_unlock(&tg->lock); | |
9b5b7751 | 7091 | return 0; |
29f59db3 SV |
7092 | } |
7093 | ||
5cb350ba DG |
7094 | unsigned long sched_group_shares(struct task_group *tg) |
7095 | { | |
7096 | return tg->shares; | |
7097 | } | |
7098 | ||
3a252015 | 7099 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e SV |
7100 | |
7101 | #ifdef CONFIG_FAIR_CGROUP_SCHED | |
7102 | ||
7103 | /* return corresponding task_group object of a cgroup */ | |
7104 | static inline struct task_group *cgroup_tg(struct cgroup *cont) | |
7105 | { | |
7106 | return container_of(cgroup_subsys_state(cont, cpu_cgroup_subsys_id), | |
7107 | struct task_group, css); | |
7108 | } | |
7109 | ||
7110 | static struct cgroup_subsys_state * | |
7111 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) | |
7112 | { | |
7113 | struct task_group *tg; | |
7114 | ||
7115 | if (!cont->parent) { | |
7116 | /* This is early initialization for the top cgroup */ | |
7117 | init_task_group.css.cgroup = cont; | |
7118 | return &init_task_group.css; | |
7119 | } | |
7120 | ||
7121 | /* we support only 1-level deep hierarchical scheduler atm */ | |
7122 | if (cont->parent->parent) | |
7123 | return ERR_PTR(-EINVAL); | |
7124 | ||
7125 | tg = sched_create_group(); | |
7126 | if (IS_ERR(tg)) | |
7127 | return ERR_PTR(-ENOMEM); | |
7128 | ||
7129 | /* Bind the cgroup to task_group object we just created */ | |
7130 | tg->css.cgroup = cont; | |
7131 | ||
7132 | return &tg->css; | |
7133 | } | |
7134 | ||
7135 | static void cpu_cgroup_destroy(struct cgroup_subsys *ss, | |
7136 | struct cgroup *cont) | |
7137 | { | |
7138 | struct task_group *tg = cgroup_tg(cont); | |
7139 | ||
7140 | sched_destroy_group(tg); | |
7141 | } | |
7142 | ||
7143 | static int cpu_cgroup_can_attach(struct cgroup_subsys *ss, | |
7144 | struct cgroup *cont, struct task_struct *tsk) | |
7145 | { | |
7146 | /* We don't support RT-tasks being in separate groups */ | |
7147 | if (tsk->sched_class != &fair_sched_class) | |
7148 | return -EINVAL; | |
7149 | ||
7150 | return 0; | |
7151 | } | |
7152 | ||
7153 | static void | |
7154 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cont, | |
7155 | struct cgroup *old_cont, struct task_struct *tsk) | |
7156 | { | |
7157 | sched_move_task(tsk); | |
7158 | } | |
7159 | ||
7160 | static ssize_t cpu_shares_write(struct cgroup *cont, struct cftype *cftype, | |
7161 | struct file *file, const char __user *userbuf, | |
7162 | size_t nbytes, loff_t *ppos) | |
7163 | { | |
7164 | unsigned long shareval; | |
7165 | struct task_group *tg = cgroup_tg(cont); | |
7166 | char buffer[2*sizeof(unsigned long) + 1]; | |
7167 | int rc; | |
7168 | ||
7169 | if (nbytes > 2*sizeof(unsigned long)) /* safety check */ | |
7170 | return -E2BIG; | |
7171 | ||
7172 | if (copy_from_user(buffer, userbuf, nbytes)) | |
7173 | return -EFAULT; | |
7174 | ||
7175 | buffer[nbytes] = 0; /* nul-terminate */ | |
7176 | shareval = simple_strtoul(buffer, NULL, 10); | |
7177 | ||
7178 | rc = sched_group_set_shares(tg, shareval); | |
7179 | ||
7180 | return (rc < 0 ? rc : nbytes); | |
7181 | } | |
7182 | ||
7183 | static u64 cpu_shares_read_uint(struct cgroup *cont, struct cftype *cft) | |
7184 | { | |
7185 | struct task_group *tg = cgroup_tg(cont); | |
7186 | ||
7187 | return (u64) tg->shares; | |
7188 | } | |
7189 | ||
7190 | static struct cftype cpu_shares = { | |
7191 | .name = "shares", | |
7192 | .read_uint = cpu_shares_read_uint, | |
7193 | .write = cpu_shares_write, | |
7194 | }; | |
7195 | ||
7196 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
7197 | { | |
7198 | return cgroup_add_file(cont, ss, &cpu_shares); | |
7199 | } | |
7200 | ||
7201 | struct cgroup_subsys cpu_cgroup_subsys = { | |
7202 | .name = "cpu", | |
7203 | .create = cpu_cgroup_create, | |
7204 | .destroy = cpu_cgroup_destroy, | |
7205 | .can_attach = cpu_cgroup_can_attach, | |
7206 | .attach = cpu_cgroup_attach, | |
7207 | .populate = cpu_cgroup_populate, | |
7208 | .subsys_id = cpu_cgroup_subsys_id, | |
7209 | .early_init = 1, | |
7210 | }; | |
7211 | ||
7212 | #endif /* CONFIG_FAIR_CGROUP_SCHED */ |