Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/sched.c | |
3 | * | |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 LT |
34 | #include <linux/highmem.h> |
35 | #include <linux/smp_lock.h> | |
36 | #include <asm/mmu_context.h> | |
37 | #include <linux/interrupt.h> | |
c59ede7b | 38 | #include <linux/capability.h> |
1da177e4 LT |
39 | #include <linux/completion.h> |
40 | #include <linux/kernel_stat.h> | |
9a11b49a | 41 | #include <linux/debug_locks.h> |
1da177e4 LT |
42 | #include <linux/security.h> |
43 | #include <linux/notifier.h> | |
44 | #include <linux/profile.h> | |
7dfb7103 | 45 | #include <linux/freezer.h> |
198e2f18 | 46 | #include <linux/vmalloc.h> |
1da177e4 LT |
47 | #include <linux/blkdev.h> |
48 | #include <linux/delay.h> | |
b488893a | 49 | #include <linux/pid_namespace.h> |
1da177e4 LT |
50 | #include <linux/smp.h> |
51 | #include <linux/threads.h> | |
52 | #include <linux/timer.h> | |
53 | #include <linux/rcupdate.h> | |
54 | #include <linux/cpu.h> | |
55 | #include <linux/cpuset.h> | |
56 | #include <linux/percpu.h> | |
57 | #include <linux/kthread.h> | |
58 | #include <linux/seq_file.h> | |
e692ab53 | 59 | #include <linux/sysctl.h> |
1da177e4 LT |
60 | #include <linux/syscalls.h> |
61 | #include <linux/times.h> | |
8f0ab514 | 62 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 63 | #include <linux/kprobes.h> |
0ff92245 | 64 | #include <linux/delayacct.h> |
5517d86b | 65 | #include <linux/reciprocal_div.h> |
dff06c15 | 66 | #include <linux/unistd.h> |
f5ff8422 | 67 | #include <linux/pagemap.h> |
8f4d37ec | 68 | #include <linux/hrtimer.h> |
1da177e4 | 69 | |
5517d86b | 70 | #include <asm/tlb.h> |
838225b4 | 71 | #include <asm/irq_regs.h> |
1da177e4 | 72 | |
b035b6de AD |
73 | /* |
74 | * Scheduler clock - returns current time in nanosec units. | |
75 | * This is default implementation. | |
76 | * Architectures and sub-architectures can override this. | |
77 | */ | |
78 | unsigned long long __attribute__((weak)) sched_clock(void) | |
79 | { | |
d6322faf | 80 | return (unsigned long long)jiffies * (NSEC_PER_SEC / HZ); |
b035b6de AD |
81 | } |
82 | ||
1da177e4 LT |
83 | /* |
84 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
85 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
86 | * and back. | |
87 | */ | |
88 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
89 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
90 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
91 | ||
92 | /* | |
93 | * 'User priority' is the nice value converted to something we | |
94 | * can work with better when scaling various scheduler parameters, | |
95 | * it's a [ 0 ... 39 ] range. | |
96 | */ | |
97 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
98 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
99 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
100 | ||
101 | /* | |
d7876a08 | 102 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 103 | */ |
d6322faf | 104 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 105 | |
6aa645ea IM |
106 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
107 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
108 | ||
1da177e4 LT |
109 | /* |
110 | * These are the 'tuning knobs' of the scheduler: | |
111 | * | |
a4ec24b4 | 112 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
113 | * Timeslices get refilled after they expire. |
114 | */ | |
1da177e4 | 115 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 116 | |
5517d86b ED |
117 | #ifdef CONFIG_SMP |
118 | /* | |
119 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
120 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
121 | */ | |
122 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
123 | { | |
124 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
125 | } | |
126 | ||
127 | /* | |
128 | * Each time a sched group cpu_power is changed, | |
129 | * we must compute its reciprocal value | |
130 | */ | |
131 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
132 | { | |
133 | sg->__cpu_power += val; | |
134 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
135 | } | |
136 | #endif | |
137 | ||
e05606d3 IM |
138 | static inline int rt_policy(int policy) |
139 | { | |
140 | if (unlikely(policy == SCHED_FIFO) || unlikely(policy == SCHED_RR)) | |
141 | return 1; | |
142 | return 0; | |
143 | } | |
144 | ||
145 | static inline int task_has_rt_policy(struct task_struct *p) | |
146 | { | |
147 | return rt_policy(p->policy); | |
148 | } | |
149 | ||
1da177e4 | 150 | /* |
6aa645ea | 151 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 152 | */ |
6aa645ea IM |
153 | struct rt_prio_array { |
154 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
155 | struct list_head queue[MAX_RT_PRIO]; | |
156 | }; | |
157 | ||
052f1dc7 | 158 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 159 | |
68318b8e SV |
160 | #include <linux/cgroup.h> |
161 | ||
29f59db3 SV |
162 | struct cfs_rq; |
163 | ||
6f505b16 PZ |
164 | static LIST_HEAD(task_groups); |
165 | ||
29f59db3 | 166 | /* task group related information */ |
4cf86d77 | 167 | struct task_group { |
052f1dc7 | 168 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
169 | struct cgroup_subsys_state css; |
170 | #endif | |
052f1dc7 PZ |
171 | |
172 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
29f59db3 SV |
173 | /* schedulable entities of this group on each cpu */ |
174 | struct sched_entity **se; | |
175 | /* runqueue "owned" by this group on each cpu */ | |
176 | struct cfs_rq **cfs_rq; | |
177 | unsigned long shares; | |
052f1dc7 PZ |
178 | #endif |
179 | ||
180 | #ifdef CONFIG_RT_GROUP_SCHED | |
181 | struct sched_rt_entity **rt_se; | |
182 | struct rt_rq **rt_rq; | |
183 | ||
184 | u64 rt_runtime; | |
185 | #endif | |
6b2d7700 | 186 | |
ae8393e5 | 187 | struct rcu_head rcu; |
6f505b16 | 188 | struct list_head list; |
29f59db3 SV |
189 | }; |
190 | ||
052f1dc7 | 191 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
192 | /* Default task group's sched entity on each cpu */ |
193 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
194 | /* Default task group's cfs_rq on each cpu */ | |
195 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
196 | ||
9b5b7751 SV |
197 | static struct sched_entity *init_sched_entity_p[NR_CPUS]; |
198 | static struct cfs_rq *init_cfs_rq_p[NR_CPUS]; | |
052f1dc7 PZ |
199 | #endif |
200 | ||
201 | #ifdef CONFIG_RT_GROUP_SCHED | |
202 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
203 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
29f59db3 | 204 | |
6f505b16 PZ |
205 | static struct sched_rt_entity *init_sched_rt_entity_p[NR_CPUS]; |
206 | static struct rt_rq *init_rt_rq_p[NR_CPUS]; | |
052f1dc7 | 207 | #endif |
6f505b16 | 208 | |
8ed36996 | 209 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
210 | * a task group's cpu shares. |
211 | */ | |
8ed36996 | 212 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 213 | |
a1835615 SV |
214 | /* doms_cur_mutex serializes access to doms_cur[] array */ |
215 | static DEFINE_MUTEX(doms_cur_mutex); | |
216 | ||
052f1dc7 | 217 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
218 | #ifdef CONFIG_USER_SCHED |
219 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
220 | #else | |
221 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD | |
222 | #endif | |
223 | ||
052f1dc7 PZ |
224 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
225 | #endif | |
226 | ||
29f59db3 | 227 | /* Default task group. |
3a252015 | 228 | * Every task in system belong to this group at bootup. |
29f59db3 | 229 | */ |
4cf86d77 | 230 | struct task_group init_task_group = { |
052f1dc7 | 231 | #ifdef CONFIG_FAIR_GROUP_SCHED |
0eab9146 | 232 | .se = init_sched_entity_p, |
3a252015 | 233 | .cfs_rq = init_cfs_rq_p, |
052f1dc7 | 234 | #endif |
6f505b16 | 235 | |
052f1dc7 | 236 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
237 | .rt_se = init_sched_rt_entity_p, |
238 | .rt_rq = init_rt_rq_p, | |
24e377a8 | 239 | #endif |
052f1dc7 | 240 | }; |
29f59db3 SV |
241 | |
242 | /* return group to which a task belongs */ | |
4cf86d77 | 243 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 244 | { |
4cf86d77 | 245 | struct task_group *tg; |
9b5b7751 | 246 | |
052f1dc7 | 247 | #ifdef CONFIG_USER_SCHED |
24e377a8 | 248 | tg = p->user->tg; |
052f1dc7 | 249 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
250 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
251 | struct task_group, css); | |
24e377a8 | 252 | #else |
41a2d6cf | 253 | tg = &init_task_group; |
24e377a8 | 254 | #endif |
9b5b7751 | 255 | return tg; |
29f59db3 SV |
256 | } |
257 | ||
258 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 259 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 260 | { |
052f1dc7 | 261 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
262 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
263 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 264 | #endif |
6f505b16 | 265 | |
052f1dc7 | 266 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
267 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
268 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 269 | #endif |
29f59db3 SV |
270 | } |
271 | ||
a1835615 SV |
272 | static inline void lock_doms_cur(void) |
273 | { | |
274 | mutex_lock(&doms_cur_mutex); | |
275 | } | |
276 | ||
277 | static inline void unlock_doms_cur(void) | |
278 | { | |
279 | mutex_unlock(&doms_cur_mutex); | |
280 | } | |
281 | ||
29f59db3 SV |
282 | #else |
283 | ||
6f505b16 | 284 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
a1835615 SV |
285 | static inline void lock_doms_cur(void) { } |
286 | static inline void unlock_doms_cur(void) { } | |
29f59db3 | 287 | |
052f1dc7 | 288 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 289 | |
6aa645ea IM |
290 | /* CFS-related fields in a runqueue */ |
291 | struct cfs_rq { | |
292 | struct load_weight load; | |
293 | unsigned long nr_running; | |
294 | ||
6aa645ea | 295 | u64 exec_clock; |
e9acbff6 | 296 | u64 min_vruntime; |
6aa645ea IM |
297 | |
298 | struct rb_root tasks_timeline; | |
299 | struct rb_node *rb_leftmost; | |
300 | struct rb_node *rb_load_balance_curr; | |
6aa645ea IM |
301 | /* 'curr' points to currently running entity on this cfs_rq. |
302 | * It is set to NULL otherwise (i.e when none are currently running). | |
303 | */ | |
aa2ac252 | 304 | struct sched_entity *curr, *next; |
ddc97297 PZ |
305 | |
306 | unsigned long nr_spread_over; | |
307 | ||
62160e3f | 308 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
309 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
310 | ||
41a2d6cf IM |
311 | /* |
312 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
313 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
314 | * (like users, containers etc.) | |
315 | * | |
316 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
317 | * list is used during load balance. | |
318 | */ | |
41a2d6cf IM |
319 | struct list_head leaf_cfs_rq_list; |
320 | struct task_group *tg; /* group that "owns" this runqueue */ | |
6aa645ea IM |
321 | #endif |
322 | }; | |
1da177e4 | 323 | |
6aa645ea IM |
324 | /* Real-Time classes' related field in a runqueue: */ |
325 | struct rt_rq { | |
326 | struct rt_prio_array active; | |
63489e45 | 327 | unsigned long rt_nr_running; |
052f1dc7 | 328 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
329 | int highest_prio; /* highest queued rt task prio */ |
330 | #endif | |
fa85ae24 | 331 | #ifdef CONFIG_SMP |
73fe6aae | 332 | unsigned long rt_nr_migratory; |
a22d7fc1 | 333 | int overloaded; |
fa85ae24 | 334 | #endif |
6f505b16 | 335 | int rt_throttled; |
fa85ae24 | 336 | u64 rt_time; |
6f505b16 | 337 | |
052f1dc7 | 338 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
339 | unsigned long rt_nr_boosted; |
340 | ||
6f505b16 PZ |
341 | struct rq *rq; |
342 | struct list_head leaf_rt_rq_list; | |
343 | struct task_group *tg; | |
344 | struct sched_rt_entity *rt_se; | |
345 | #endif | |
6aa645ea IM |
346 | }; |
347 | ||
57d885fe GH |
348 | #ifdef CONFIG_SMP |
349 | ||
350 | /* | |
351 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
352 | * variables. Each exclusive cpuset essentially defines an island domain by |
353 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
354 | * exclusive cpuset is created, we also create and attach a new root-domain |
355 | * object. | |
356 | * | |
57d885fe GH |
357 | */ |
358 | struct root_domain { | |
359 | atomic_t refcount; | |
360 | cpumask_t span; | |
361 | cpumask_t online; | |
637f5085 | 362 | |
0eab9146 | 363 | /* |
637f5085 GH |
364 | * The "RT overload" flag: it gets set if a CPU has more than |
365 | * one runnable RT task. | |
366 | */ | |
367 | cpumask_t rto_mask; | |
0eab9146 | 368 | atomic_t rto_count; |
57d885fe GH |
369 | }; |
370 | ||
dc938520 GH |
371 | /* |
372 | * By default the system creates a single root-domain with all cpus as | |
373 | * members (mimicking the global state we have today). | |
374 | */ | |
57d885fe GH |
375 | static struct root_domain def_root_domain; |
376 | ||
377 | #endif | |
378 | ||
1da177e4 LT |
379 | /* |
380 | * This is the main, per-CPU runqueue data structure. | |
381 | * | |
382 | * Locking rule: those places that want to lock multiple runqueues | |
383 | * (such as the load balancing or the thread migration code), lock | |
384 | * acquire operations must be ordered by ascending &runqueue. | |
385 | */ | |
70b97a7f | 386 | struct rq { |
d8016491 IM |
387 | /* runqueue lock: */ |
388 | spinlock_t lock; | |
1da177e4 LT |
389 | |
390 | /* | |
391 | * nr_running and cpu_load should be in the same cacheline because | |
392 | * remote CPUs use both these fields when doing load calculation. | |
393 | */ | |
394 | unsigned long nr_running; | |
6aa645ea IM |
395 | #define CPU_LOAD_IDX_MAX 5 |
396 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 397 | unsigned char idle_at_tick; |
46cb4b7c SS |
398 | #ifdef CONFIG_NO_HZ |
399 | unsigned char in_nohz_recently; | |
400 | #endif | |
d8016491 IM |
401 | /* capture load from *all* tasks on this cpu: */ |
402 | struct load_weight load; | |
6aa645ea IM |
403 | unsigned long nr_load_updates; |
404 | u64 nr_switches; | |
405 | ||
406 | struct cfs_rq cfs; | |
6f505b16 PZ |
407 | struct rt_rq rt; |
408 | u64 rt_period_expire; | |
48d5e258 | 409 | int rt_throttled; |
6f505b16 | 410 | |
6aa645ea | 411 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
412 | /* list of leaf cfs_rq on this cpu: */ |
413 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
414 | #endif |
415 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 416 | struct list_head leaf_rt_rq_list; |
1da177e4 | 417 | #endif |
1da177e4 LT |
418 | |
419 | /* | |
420 | * This is part of a global counter where only the total sum | |
421 | * over all CPUs matters. A task can increase this counter on | |
422 | * one CPU and if it got migrated afterwards it may decrease | |
423 | * it on another CPU. Always updated under the runqueue lock: | |
424 | */ | |
425 | unsigned long nr_uninterruptible; | |
426 | ||
36c8b586 | 427 | struct task_struct *curr, *idle; |
c9819f45 | 428 | unsigned long next_balance; |
1da177e4 | 429 | struct mm_struct *prev_mm; |
6aa645ea | 430 | |
6aa645ea IM |
431 | u64 clock, prev_clock_raw; |
432 | s64 clock_max_delta; | |
433 | ||
cc203d24 | 434 | unsigned int clock_warps, clock_overflows, clock_underflows; |
2aa44d05 IM |
435 | u64 idle_clock; |
436 | unsigned int clock_deep_idle_events; | |
529c7726 | 437 | u64 tick_timestamp; |
6aa645ea | 438 | |
1da177e4 LT |
439 | atomic_t nr_iowait; |
440 | ||
441 | #ifdef CONFIG_SMP | |
0eab9146 | 442 | struct root_domain *rd; |
1da177e4 LT |
443 | struct sched_domain *sd; |
444 | ||
445 | /* For active balancing */ | |
446 | int active_balance; | |
447 | int push_cpu; | |
d8016491 IM |
448 | /* cpu of this runqueue: */ |
449 | int cpu; | |
1da177e4 | 450 | |
36c8b586 | 451 | struct task_struct *migration_thread; |
1da177e4 LT |
452 | struct list_head migration_queue; |
453 | #endif | |
454 | ||
8f4d37ec PZ |
455 | #ifdef CONFIG_SCHED_HRTICK |
456 | unsigned long hrtick_flags; | |
457 | ktime_t hrtick_expire; | |
458 | struct hrtimer hrtick_timer; | |
459 | #endif | |
460 | ||
1da177e4 LT |
461 | #ifdef CONFIG_SCHEDSTATS |
462 | /* latency stats */ | |
463 | struct sched_info rq_sched_info; | |
464 | ||
465 | /* sys_sched_yield() stats */ | |
480b9434 KC |
466 | unsigned int yld_exp_empty; |
467 | unsigned int yld_act_empty; | |
468 | unsigned int yld_both_empty; | |
469 | unsigned int yld_count; | |
1da177e4 LT |
470 | |
471 | /* schedule() stats */ | |
480b9434 KC |
472 | unsigned int sched_switch; |
473 | unsigned int sched_count; | |
474 | unsigned int sched_goidle; | |
1da177e4 LT |
475 | |
476 | /* try_to_wake_up() stats */ | |
480b9434 KC |
477 | unsigned int ttwu_count; |
478 | unsigned int ttwu_local; | |
b8efb561 IM |
479 | |
480 | /* BKL stats */ | |
480b9434 | 481 | unsigned int bkl_count; |
1da177e4 | 482 | #endif |
fcb99371 | 483 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
484 | }; |
485 | ||
f34e3b61 | 486 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 487 | |
dd41f596 IM |
488 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
489 | { | |
490 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
491 | } | |
492 | ||
0a2966b4 CL |
493 | static inline int cpu_of(struct rq *rq) |
494 | { | |
495 | #ifdef CONFIG_SMP | |
496 | return rq->cpu; | |
497 | #else | |
498 | return 0; | |
499 | #endif | |
500 | } | |
501 | ||
20d315d4 | 502 | /* |
b04a0f4c IM |
503 | * Update the per-runqueue clock, as finegrained as the platform can give |
504 | * us, but without assuming monotonicity, etc.: | |
20d315d4 | 505 | */ |
b04a0f4c | 506 | static void __update_rq_clock(struct rq *rq) |
20d315d4 IM |
507 | { |
508 | u64 prev_raw = rq->prev_clock_raw; | |
509 | u64 now = sched_clock(); | |
510 | s64 delta = now - prev_raw; | |
511 | u64 clock = rq->clock; | |
512 | ||
b04a0f4c IM |
513 | #ifdef CONFIG_SCHED_DEBUG |
514 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
515 | #endif | |
20d315d4 IM |
516 | /* |
517 | * Protect against sched_clock() occasionally going backwards: | |
518 | */ | |
519 | if (unlikely(delta < 0)) { | |
520 | clock++; | |
521 | rq->clock_warps++; | |
522 | } else { | |
523 | /* | |
524 | * Catch too large forward jumps too: | |
525 | */ | |
529c7726 IM |
526 | if (unlikely(clock + delta > rq->tick_timestamp + TICK_NSEC)) { |
527 | if (clock < rq->tick_timestamp + TICK_NSEC) | |
528 | clock = rq->tick_timestamp + TICK_NSEC; | |
529 | else | |
530 | clock++; | |
20d315d4 IM |
531 | rq->clock_overflows++; |
532 | } else { | |
533 | if (unlikely(delta > rq->clock_max_delta)) | |
534 | rq->clock_max_delta = delta; | |
535 | clock += delta; | |
536 | } | |
537 | } | |
538 | ||
539 | rq->prev_clock_raw = now; | |
540 | rq->clock = clock; | |
b04a0f4c | 541 | } |
20d315d4 | 542 | |
b04a0f4c IM |
543 | static void update_rq_clock(struct rq *rq) |
544 | { | |
545 | if (likely(smp_processor_id() == cpu_of(rq))) | |
546 | __update_rq_clock(rq); | |
20d315d4 IM |
547 | } |
548 | ||
674311d5 NP |
549 | /* |
550 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 551 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
552 | * |
553 | * The domain tree of any CPU may only be accessed from within | |
554 | * preempt-disabled sections. | |
555 | */ | |
48f24c4d IM |
556 | #define for_each_domain(cpu, __sd) \ |
557 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
558 | |
559 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
560 | #define this_rq() (&__get_cpu_var(runqueues)) | |
561 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
562 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
563 | ||
48d5e258 PZ |
564 | unsigned long rt_needs_cpu(int cpu) |
565 | { | |
566 | struct rq *rq = cpu_rq(cpu); | |
567 | u64 delta; | |
568 | ||
569 | if (!rq->rt_throttled) | |
570 | return 0; | |
571 | ||
572 | if (rq->clock > rq->rt_period_expire) | |
573 | return 1; | |
574 | ||
575 | delta = rq->rt_period_expire - rq->clock; | |
576 | do_div(delta, NSEC_PER_SEC / HZ); | |
577 | ||
578 | return (unsigned long)delta; | |
579 | } | |
580 | ||
bf5c91ba IM |
581 | /* |
582 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
583 | */ | |
584 | #ifdef CONFIG_SCHED_DEBUG | |
585 | # define const_debug __read_mostly | |
586 | #else | |
587 | # define const_debug static const | |
588 | #endif | |
589 | ||
590 | /* | |
591 | * Debugging: various feature bits | |
592 | */ | |
593 | enum { | |
bbdba7c0 | 594 | SCHED_FEAT_NEW_FAIR_SLEEPERS = 1, |
9612633a IM |
595 | SCHED_FEAT_WAKEUP_PREEMPT = 2, |
596 | SCHED_FEAT_START_DEBIT = 4, | |
2070ee01 PZ |
597 | SCHED_FEAT_HRTICK = 8, |
598 | SCHED_FEAT_DOUBLE_TICK = 16, | |
bf5c91ba IM |
599 | }; |
600 | ||
601 | const_debug unsigned int sysctl_sched_features = | |
8401f775 | 602 | SCHED_FEAT_NEW_FAIR_SLEEPERS * 1 | |
9612633a | 603 | SCHED_FEAT_WAKEUP_PREEMPT * 1 | |
8401f775 | 604 | SCHED_FEAT_START_DEBIT * 1 | |
8f4d37ec PZ |
605 | SCHED_FEAT_HRTICK * 1 | |
606 | SCHED_FEAT_DOUBLE_TICK * 0; | |
bf5c91ba IM |
607 | |
608 | #define sched_feat(x) (sysctl_sched_features & SCHED_FEAT_##x) | |
609 | ||
b82d9fdd PZ |
610 | /* |
611 | * Number of tasks to iterate in a single balance run. | |
612 | * Limited because this is done with IRQs disabled. | |
613 | */ | |
614 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
615 | ||
fa85ae24 | 616 | /* |
9f0c1e56 | 617 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
618 | * default: 1s |
619 | */ | |
9f0c1e56 | 620 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 621 | |
6892b75e IM |
622 | static __read_mostly int scheduler_running; |
623 | ||
9f0c1e56 PZ |
624 | /* |
625 | * part of the period that we allow rt tasks to run in us. | |
626 | * default: 0.95s | |
627 | */ | |
628 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 PZ |
629 | |
630 | /* | |
9f0c1e56 | 631 | * single value that denotes runtime == period, ie unlimited time. |
fa85ae24 | 632 | */ |
9f0c1e56 | 633 | #define RUNTIME_INF ((u64)~0ULL) |
fa85ae24 | 634 | |
27ec4407 IM |
635 | static const unsigned long long time_sync_thresh = 100000; |
636 | ||
637 | static DEFINE_PER_CPU(unsigned long long, time_offset); | |
638 | static DEFINE_PER_CPU(unsigned long long, prev_cpu_time); | |
639 | ||
e436d800 | 640 | /* |
27ec4407 IM |
641 | * Global lock which we take every now and then to synchronize |
642 | * the CPUs time. This method is not warp-safe, but it's good | |
643 | * enough to synchronize slowly diverging time sources and thus | |
644 | * it's good enough for tracing: | |
e436d800 | 645 | */ |
27ec4407 IM |
646 | static DEFINE_SPINLOCK(time_sync_lock); |
647 | static unsigned long long prev_global_time; | |
648 | ||
649 | static unsigned long long __sync_cpu_clock(cycles_t time, int cpu) | |
650 | { | |
651 | unsigned long flags; | |
652 | ||
653 | spin_lock_irqsave(&time_sync_lock, flags); | |
654 | ||
655 | if (time < prev_global_time) { | |
656 | per_cpu(time_offset, cpu) += prev_global_time - time; | |
657 | time = prev_global_time; | |
658 | } else { | |
659 | prev_global_time = time; | |
660 | } | |
661 | ||
662 | spin_unlock_irqrestore(&time_sync_lock, flags); | |
663 | ||
664 | return time; | |
665 | } | |
666 | ||
667 | static unsigned long long __cpu_clock(int cpu) | |
e436d800 | 668 | { |
e436d800 IM |
669 | unsigned long long now; |
670 | unsigned long flags; | |
b04a0f4c | 671 | struct rq *rq; |
e436d800 | 672 | |
8ced5f69 IM |
673 | /* |
674 | * Only call sched_clock() if the scheduler has already been | |
675 | * initialized (some code might call cpu_clock() very early): | |
676 | */ | |
6892b75e IM |
677 | if (unlikely(!scheduler_running)) |
678 | return 0; | |
679 | ||
680 | local_irq_save(flags); | |
681 | rq = cpu_rq(cpu); | |
682 | update_rq_clock(rq); | |
b04a0f4c | 683 | now = rq->clock; |
2cd4d0ea | 684 | local_irq_restore(flags); |
e436d800 IM |
685 | |
686 | return now; | |
687 | } | |
27ec4407 IM |
688 | |
689 | /* | |
690 | * For kernel-internal use: high-speed (but slightly incorrect) per-cpu | |
691 | * clock constructed from sched_clock(): | |
692 | */ | |
693 | unsigned long long cpu_clock(int cpu) | |
694 | { | |
695 | unsigned long long prev_cpu_time, time, delta_time; | |
696 | ||
697 | prev_cpu_time = per_cpu(prev_cpu_time, cpu); | |
698 | time = __cpu_clock(cpu) + per_cpu(time_offset, cpu); | |
699 | delta_time = time-prev_cpu_time; | |
700 | ||
701 | if (unlikely(delta_time > time_sync_thresh)) | |
702 | time = __sync_cpu_clock(time, cpu); | |
703 | ||
704 | return time; | |
705 | } | |
a58f6f25 | 706 | EXPORT_SYMBOL_GPL(cpu_clock); |
e436d800 | 707 | |
1da177e4 | 708 | #ifndef prepare_arch_switch |
4866cde0 NP |
709 | # define prepare_arch_switch(next) do { } while (0) |
710 | #endif | |
711 | #ifndef finish_arch_switch | |
712 | # define finish_arch_switch(prev) do { } while (0) | |
713 | #endif | |
714 | ||
051a1d1a DA |
715 | static inline int task_current(struct rq *rq, struct task_struct *p) |
716 | { | |
717 | return rq->curr == p; | |
718 | } | |
719 | ||
4866cde0 | 720 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 721 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 722 | { |
051a1d1a | 723 | return task_current(rq, p); |
4866cde0 NP |
724 | } |
725 | ||
70b97a7f | 726 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
727 | { |
728 | } | |
729 | ||
70b97a7f | 730 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 731 | { |
da04c035 IM |
732 | #ifdef CONFIG_DEBUG_SPINLOCK |
733 | /* this is a valid case when another task releases the spinlock */ | |
734 | rq->lock.owner = current; | |
735 | #endif | |
8a25d5de IM |
736 | /* |
737 | * If we are tracking spinlock dependencies then we have to | |
738 | * fix up the runqueue lock - which gets 'carried over' from | |
739 | * prev into current: | |
740 | */ | |
741 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
742 | ||
4866cde0 NP |
743 | spin_unlock_irq(&rq->lock); |
744 | } | |
745 | ||
746 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 747 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
748 | { |
749 | #ifdef CONFIG_SMP | |
750 | return p->oncpu; | |
751 | #else | |
051a1d1a | 752 | return task_current(rq, p); |
4866cde0 NP |
753 | #endif |
754 | } | |
755 | ||
70b97a7f | 756 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
757 | { |
758 | #ifdef CONFIG_SMP | |
759 | /* | |
760 | * We can optimise this out completely for !SMP, because the | |
761 | * SMP rebalancing from interrupt is the only thing that cares | |
762 | * here. | |
763 | */ | |
764 | next->oncpu = 1; | |
765 | #endif | |
766 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
767 | spin_unlock_irq(&rq->lock); | |
768 | #else | |
769 | spin_unlock(&rq->lock); | |
770 | #endif | |
771 | } | |
772 | ||
70b97a7f | 773 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
774 | { |
775 | #ifdef CONFIG_SMP | |
776 | /* | |
777 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
778 | * We must ensure this doesn't happen until the switch is completely | |
779 | * finished. | |
780 | */ | |
781 | smp_wmb(); | |
782 | prev->oncpu = 0; | |
783 | #endif | |
784 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
785 | local_irq_enable(); | |
1da177e4 | 786 | #endif |
4866cde0 NP |
787 | } |
788 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 789 | |
b29739f9 IM |
790 | /* |
791 | * __task_rq_lock - lock the runqueue a given task resides on. | |
792 | * Must be called interrupts disabled. | |
793 | */ | |
70b97a7f | 794 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
795 | __acquires(rq->lock) |
796 | { | |
3a5c359a AK |
797 | for (;;) { |
798 | struct rq *rq = task_rq(p); | |
799 | spin_lock(&rq->lock); | |
800 | if (likely(rq == task_rq(p))) | |
801 | return rq; | |
b29739f9 | 802 | spin_unlock(&rq->lock); |
b29739f9 | 803 | } |
b29739f9 IM |
804 | } |
805 | ||
1da177e4 LT |
806 | /* |
807 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 808 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
809 | * explicitly disabling preemption. |
810 | */ | |
70b97a7f | 811 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
812 | __acquires(rq->lock) |
813 | { | |
70b97a7f | 814 | struct rq *rq; |
1da177e4 | 815 | |
3a5c359a AK |
816 | for (;;) { |
817 | local_irq_save(*flags); | |
818 | rq = task_rq(p); | |
819 | spin_lock(&rq->lock); | |
820 | if (likely(rq == task_rq(p))) | |
821 | return rq; | |
1da177e4 | 822 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 823 | } |
1da177e4 LT |
824 | } |
825 | ||
a9957449 | 826 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
827 | __releases(rq->lock) |
828 | { | |
829 | spin_unlock(&rq->lock); | |
830 | } | |
831 | ||
70b97a7f | 832 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
833 | __releases(rq->lock) |
834 | { | |
835 | spin_unlock_irqrestore(&rq->lock, *flags); | |
836 | } | |
837 | ||
1da177e4 | 838 | /* |
cc2a73b5 | 839 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 840 | */ |
a9957449 | 841 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
842 | __acquires(rq->lock) |
843 | { | |
70b97a7f | 844 | struct rq *rq; |
1da177e4 LT |
845 | |
846 | local_irq_disable(); | |
847 | rq = this_rq(); | |
848 | spin_lock(&rq->lock); | |
849 | ||
850 | return rq; | |
851 | } | |
852 | ||
1b9f19c2 | 853 | /* |
2aa44d05 | 854 | * We are going deep-idle (irqs are disabled): |
1b9f19c2 | 855 | */ |
2aa44d05 | 856 | void sched_clock_idle_sleep_event(void) |
1b9f19c2 | 857 | { |
2aa44d05 IM |
858 | struct rq *rq = cpu_rq(smp_processor_id()); |
859 | ||
860 | spin_lock(&rq->lock); | |
861 | __update_rq_clock(rq); | |
862 | spin_unlock(&rq->lock); | |
863 | rq->clock_deep_idle_events++; | |
864 | } | |
865 | EXPORT_SYMBOL_GPL(sched_clock_idle_sleep_event); | |
866 | ||
867 | /* | |
868 | * We just idled delta nanoseconds (called with irqs disabled): | |
869 | */ | |
870 | void sched_clock_idle_wakeup_event(u64 delta_ns) | |
871 | { | |
872 | struct rq *rq = cpu_rq(smp_processor_id()); | |
873 | u64 now = sched_clock(); | |
1b9f19c2 | 874 | |
2aa44d05 IM |
875 | rq->idle_clock += delta_ns; |
876 | /* | |
877 | * Override the previous timestamp and ignore all | |
878 | * sched_clock() deltas that occured while we idled, | |
879 | * and use the PM-provided delta_ns to advance the | |
880 | * rq clock: | |
881 | */ | |
882 | spin_lock(&rq->lock); | |
883 | rq->prev_clock_raw = now; | |
884 | rq->clock += delta_ns; | |
885 | spin_unlock(&rq->lock); | |
782daeee | 886 | touch_softlockup_watchdog(); |
1b9f19c2 | 887 | } |
2aa44d05 | 888 | EXPORT_SYMBOL_GPL(sched_clock_idle_wakeup_event); |
1b9f19c2 | 889 | |
8f4d37ec PZ |
890 | static void __resched_task(struct task_struct *p, int tif_bit); |
891 | ||
892 | static inline void resched_task(struct task_struct *p) | |
893 | { | |
894 | __resched_task(p, TIF_NEED_RESCHED); | |
895 | } | |
896 | ||
897 | #ifdef CONFIG_SCHED_HRTICK | |
898 | /* | |
899 | * Use HR-timers to deliver accurate preemption points. | |
900 | * | |
901 | * Its all a bit involved since we cannot program an hrt while holding the | |
902 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
903 | * reschedule event. | |
904 | * | |
905 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
906 | * rq->lock. | |
907 | */ | |
908 | static inline void resched_hrt(struct task_struct *p) | |
909 | { | |
910 | __resched_task(p, TIF_HRTICK_RESCHED); | |
911 | } | |
912 | ||
913 | static inline void resched_rq(struct rq *rq) | |
914 | { | |
915 | unsigned long flags; | |
916 | ||
917 | spin_lock_irqsave(&rq->lock, flags); | |
918 | resched_task(rq->curr); | |
919 | spin_unlock_irqrestore(&rq->lock, flags); | |
920 | } | |
921 | ||
922 | enum { | |
923 | HRTICK_SET, /* re-programm hrtick_timer */ | |
924 | HRTICK_RESET, /* not a new slice */ | |
925 | }; | |
926 | ||
927 | /* | |
928 | * Use hrtick when: | |
929 | * - enabled by features | |
930 | * - hrtimer is actually high res | |
931 | */ | |
932 | static inline int hrtick_enabled(struct rq *rq) | |
933 | { | |
934 | if (!sched_feat(HRTICK)) | |
935 | return 0; | |
936 | return hrtimer_is_hres_active(&rq->hrtick_timer); | |
937 | } | |
938 | ||
939 | /* | |
940 | * Called to set the hrtick timer state. | |
941 | * | |
942 | * called with rq->lock held and irqs disabled | |
943 | */ | |
944 | static void hrtick_start(struct rq *rq, u64 delay, int reset) | |
945 | { | |
946 | assert_spin_locked(&rq->lock); | |
947 | ||
948 | /* | |
949 | * preempt at: now + delay | |
950 | */ | |
951 | rq->hrtick_expire = | |
952 | ktime_add_ns(rq->hrtick_timer.base->get_time(), delay); | |
953 | /* | |
954 | * indicate we need to program the timer | |
955 | */ | |
956 | __set_bit(HRTICK_SET, &rq->hrtick_flags); | |
957 | if (reset) | |
958 | __set_bit(HRTICK_RESET, &rq->hrtick_flags); | |
959 | ||
960 | /* | |
961 | * New slices are called from the schedule path and don't need a | |
962 | * forced reschedule. | |
963 | */ | |
964 | if (reset) | |
965 | resched_hrt(rq->curr); | |
966 | } | |
967 | ||
968 | static void hrtick_clear(struct rq *rq) | |
969 | { | |
970 | if (hrtimer_active(&rq->hrtick_timer)) | |
971 | hrtimer_cancel(&rq->hrtick_timer); | |
972 | } | |
973 | ||
974 | /* | |
975 | * Update the timer from the possible pending state. | |
976 | */ | |
977 | static void hrtick_set(struct rq *rq) | |
978 | { | |
979 | ktime_t time; | |
980 | int set, reset; | |
981 | unsigned long flags; | |
982 | ||
983 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
984 | ||
985 | spin_lock_irqsave(&rq->lock, flags); | |
986 | set = __test_and_clear_bit(HRTICK_SET, &rq->hrtick_flags); | |
987 | reset = __test_and_clear_bit(HRTICK_RESET, &rq->hrtick_flags); | |
988 | time = rq->hrtick_expire; | |
989 | clear_thread_flag(TIF_HRTICK_RESCHED); | |
990 | spin_unlock_irqrestore(&rq->lock, flags); | |
991 | ||
992 | if (set) { | |
993 | hrtimer_start(&rq->hrtick_timer, time, HRTIMER_MODE_ABS); | |
994 | if (reset && !hrtimer_active(&rq->hrtick_timer)) | |
995 | resched_rq(rq); | |
996 | } else | |
997 | hrtick_clear(rq); | |
998 | } | |
999 | ||
1000 | /* | |
1001 | * High-resolution timer tick. | |
1002 | * Runs from hardirq context with interrupts disabled. | |
1003 | */ | |
1004 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1005 | { | |
1006 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1007 | ||
1008 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1009 | ||
1010 | spin_lock(&rq->lock); | |
1011 | __update_rq_clock(rq); | |
1012 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); | |
1013 | spin_unlock(&rq->lock); | |
1014 | ||
1015 | return HRTIMER_NORESTART; | |
1016 | } | |
1017 | ||
1018 | static inline void init_rq_hrtick(struct rq *rq) | |
1019 | { | |
1020 | rq->hrtick_flags = 0; | |
1021 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
1022 | rq->hrtick_timer.function = hrtick; | |
1023 | rq->hrtick_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ; | |
1024 | } | |
1025 | ||
1026 | void hrtick_resched(void) | |
1027 | { | |
1028 | struct rq *rq; | |
1029 | unsigned long flags; | |
1030 | ||
1031 | if (!test_thread_flag(TIF_HRTICK_RESCHED)) | |
1032 | return; | |
1033 | ||
1034 | local_irq_save(flags); | |
1035 | rq = cpu_rq(smp_processor_id()); | |
1036 | hrtick_set(rq); | |
1037 | local_irq_restore(flags); | |
1038 | } | |
1039 | #else | |
1040 | static inline void hrtick_clear(struct rq *rq) | |
1041 | { | |
1042 | } | |
1043 | ||
1044 | static inline void hrtick_set(struct rq *rq) | |
1045 | { | |
1046 | } | |
1047 | ||
1048 | static inline void init_rq_hrtick(struct rq *rq) | |
1049 | { | |
1050 | } | |
1051 | ||
1052 | void hrtick_resched(void) | |
1053 | { | |
1054 | } | |
1055 | #endif | |
1056 | ||
c24d20db IM |
1057 | /* |
1058 | * resched_task - mark a task 'to be rescheduled now'. | |
1059 | * | |
1060 | * On UP this means the setting of the need_resched flag, on SMP it | |
1061 | * might also involve a cross-CPU call to trigger the scheduler on | |
1062 | * the target CPU. | |
1063 | */ | |
1064 | #ifdef CONFIG_SMP | |
1065 | ||
1066 | #ifndef tsk_is_polling | |
1067 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1068 | #endif | |
1069 | ||
8f4d37ec | 1070 | static void __resched_task(struct task_struct *p, int tif_bit) |
c24d20db IM |
1071 | { |
1072 | int cpu; | |
1073 | ||
1074 | assert_spin_locked(&task_rq(p)->lock); | |
1075 | ||
8f4d37ec | 1076 | if (unlikely(test_tsk_thread_flag(p, tif_bit))) |
c24d20db IM |
1077 | return; |
1078 | ||
8f4d37ec | 1079 | set_tsk_thread_flag(p, tif_bit); |
c24d20db IM |
1080 | |
1081 | cpu = task_cpu(p); | |
1082 | if (cpu == smp_processor_id()) | |
1083 | return; | |
1084 | ||
1085 | /* NEED_RESCHED must be visible before we test polling */ | |
1086 | smp_mb(); | |
1087 | if (!tsk_is_polling(p)) | |
1088 | smp_send_reschedule(cpu); | |
1089 | } | |
1090 | ||
1091 | static void resched_cpu(int cpu) | |
1092 | { | |
1093 | struct rq *rq = cpu_rq(cpu); | |
1094 | unsigned long flags; | |
1095 | ||
1096 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1097 | return; | |
1098 | resched_task(cpu_curr(cpu)); | |
1099 | spin_unlock_irqrestore(&rq->lock, flags); | |
1100 | } | |
06d8308c TG |
1101 | |
1102 | #ifdef CONFIG_NO_HZ | |
1103 | /* | |
1104 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1105 | * idle CPU then this timer might expire before the next timer event | |
1106 | * which is scheduled to wake up that CPU. In case of a completely | |
1107 | * idle system the next event might even be infinite time into the | |
1108 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1109 | * leaves the inner idle loop so the newly added timer is taken into | |
1110 | * account when the CPU goes back to idle and evaluates the timer | |
1111 | * wheel for the next timer event. | |
1112 | */ | |
1113 | void wake_up_idle_cpu(int cpu) | |
1114 | { | |
1115 | struct rq *rq = cpu_rq(cpu); | |
1116 | ||
1117 | if (cpu == smp_processor_id()) | |
1118 | return; | |
1119 | ||
1120 | /* | |
1121 | * This is safe, as this function is called with the timer | |
1122 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1123 | * to idle and has not yet set rq->curr to idle then it will | |
1124 | * be serialized on the timer wheel base lock and take the new | |
1125 | * timer into account automatically. | |
1126 | */ | |
1127 | if (rq->curr != rq->idle) | |
1128 | return; | |
1129 | ||
1130 | /* | |
1131 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1132 | * lockless. The worst case is that the other CPU runs the | |
1133 | * idle task through an additional NOOP schedule() | |
1134 | */ | |
1135 | set_tsk_thread_flag(rq->idle, TIF_NEED_RESCHED); | |
1136 | ||
1137 | /* NEED_RESCHED must be visible before we test polling */ | |
1138 | smp_mb(); | |
1139 | if (!tsk_is_polling(rq->idle)) | |
1140 | smp_send_reschedule(cpu); | |
1141 | } | |
1142 | #endif | |
1143 | ||
c24d20db | 1144 | #else |
8f4d37ec | 1145 | static void __resched_task(struct task_struct *p, int tif_bit) |
c24d20db IM |
1146 | { |
1147 | assert_spin_locked(&task_rq(p)->lock); | |
8f4d37ec | 1148 | set_tsk_thread_flag(p, tif_bit); |
c24d20db IM |
1149 | } |
1150 | #endif | |
1151 | ||
45bf76df IM |
1152 | #if BITS_PER_LONG == 32 |
1153 | # define WMULT_CONST (~0UL) | |
1154 | #else | |
1155 | # define WMULT_CONST (1UL << 32) | |
1156 | #endif | |
1157 | ||
1158 | #define WMULT_SHIFT 32 | |
1159 | ||
194081eb IM |
1160 | /* |
1161 | * Shift right and round: | |
1162 | */ | |
cf2ab469 | 1163 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1164 | |
cb1c4fc9 | 1165 | static unsigned long |
45bf76df IM |
1166 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1167 | struct load_weight *lw) | |
1168 | { | |
1169 | u64 tmp; | |
1170 | ||
1171 | if (unlikely(!lw->inv_weight)) | |
27d11726 | 1172 | lw->inv_weight = (WMULT_CONST-lw->weight/2) / (lw->weight+1); |
45bf76df IM |
1173 | |
1174 | tmp = (u64)delta_exec * weight; | |
1175 | /* | |
1176 | * Check whether we'd overflow the 64-bit multiplication: | |
1177 | */ | |
194081eb | 1178 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1179 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1180 | WMULT_SHIFT/2); |
1181 | else | |
cf2ab469 | 1182 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1183 | |
ecf691da | 1184 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1185 | } |
1186 | ||
1187 | static inline unsigned long | |
1188 | calc_delta_fair(unsigned long delta_exec, struct load_weight *lw) | |
1189 | { | |
1190 | return calc_delta_mine(delta_exec, NICE_0_LOAD, lw); | |
1191 | } | |
1192 | ||
1091985b | 1193 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1194 | { |
1195 | lw->weight += inc; | |
e89996ae | 1196 | lw->inv_weight = 0; |
45bf76df IM |
1197 | } |
1198 | ||
1091985b | 1199 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1200 | { |
1201 | lw->weight -= dec; | |
e89996ae | 1202 | lw->inv_weight = 0; |
45bf76df IM |
1203 | } |
1204 | ||
2dd73a4f PW |
1205 | /* |
1206 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1207 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1208 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1209 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1210 | * scaled version of the new time slice allocation that they receive on time |
1211 | * slice expiry etc. | |
1212 | */ | |
1213 | ||
dd41f596 IM |
1214 | #define WEIGHT_IDLEPRIO 2 |
1215 | #define WMULT_IDLEPRIO (1 << 31) | |
1216 | ||
1217 | /* | |
1218 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1219 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1220 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1221 | * that remained on nice 0. | |
1222 | * | |
1223 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1224 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1225 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1226 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1227 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1228 | */ |
1229 | static const int prio_to_weight[40] = { | |
254753dc IM |
1230 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1231 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1232 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1233 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1234 | /* 0 */ 1024, 820, 655, 526, 423, | |
1235 | /* 5 */ 335, 272, 215, 172, 137, | |
1236 | /* 10 */ 110, 87, 70, 56, 45, | |
1237 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1238 | }; |
1239 | ||
5714d2de IM |
1240 | /* |
1241 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1242 | * | |
1243 | * In cases where the weight does not change often, we can use the | |
1244 | * precalculated inverse to speed up arithmetics by turning divisions | |
1245 | * into multiplications: | |
1246 | */ | |
dd41f596 | 1247 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1248 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1249 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1250 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1251 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1252 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1253 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1254 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1255 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1256 | }; |
2dd73a4f | 1257 | |
dd41f596 IM |
1258 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1259 | ||
1260 | /* | |
1261 | * runqueue iterator, to support SMP load-balancing between different | |
1262 | * scheduling classes, without having to expose their internal data | |
1263 | * structures to the load-balancing proper: | |
1264 | */ | |
1265 | struct rq_iterator { | |
1266 | void *arg; | |
1267 | struct task_struct *(*start)(void *); | |
1268 | struct task_struct *(*next)(void *); | |
1269 | }; | |
1270 | ||
e1d1484f PW |
1271 | #ifdef CONFIG_SMP |
1272 | static unsigned long | |
1273 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1274 | unsigned long max_load_move, struct sched_domain *sd, | |
1275 | enum cpu_idle_type idle, int *all_pinned, | |
1276 | int *this_best_prio, struct rq_iterator *iterator); | |
1277 | ||
1278 | static int | |
1279 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1280 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1281 | struct rq_iterator *iterator); | |
e1d1484f | 1282 | #endif |
dd41f596 | 1283 | |
d842de87 SV |
1284 | #ifdef CONFIG_CGROUP_CPUACCT |
1285 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
1286 | #else | |
1287 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
1288 | #endif | |
1289 | ||
e7693a36 GH |
1290 | #ifdef CONFIG_SMP |
1291 | static unsigned long source_load(int cpu, int type); | |
1292 | static unsigned long target_load(int cpu, int type); | |
1293 | static unsigned long cpu_avg_load_per_task(int cpu); | |
1294 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1295 | #endif /* CONFIG_SMP */ | |
1296 | ||
dd41f596 | 1297 | #include "sched_stats.h" |
dd41f596 | 1298 | #include "sched_idletask.c" |
5522d5d5 IM |
1299 | #include "sched_fair.c" |
1300 | #include "sched_rt.c" | |
dd41f596 IM |
1301 | #ifdef CONFIG_SCHED_DEBUG |
1302 | # include "sched_debug.c" | |
1303 | #endif | |
1304 | ||
1305 | #define sched_class_highest (&rt_sched_class) | |
1306 | ||
62fb1851 PZ |
1307 | static inline void inc_load(struct rq *rq, const struct task_struct *p) |
1308 | { | |
1309 | update_load_add(&rq->load, p->se.load.weight); | |
1310 | } | |
1311 | ||
1312 | static inline void dec_load(struct rq *rq, const struct task_struct *p) | |
1313 | { | |
1314 | update_load_sub(&rq->load, p->se.load.weight); | |
1315 | } | |
1316 | ||
1317 | static void inc_nr_running(struct task_struct *p, struct rq *rq) | |
9c217245 IM |
1318 | { |
1319 | rq->nr_running++; | |
62fb1851 | 1320 | inc_load(rq, p); |
9c217245 IM |
1321 | } |
1322 | ||
62fb1851 | 1323 | static void dec_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
1324 | { |
1325 | rq->nr_running--; | |
62fb1851 | 1326 | dec_load(rq, p); |
9c217245 IM |
1327 | } |
1328 | ||
45bf76df IM |
1329 | static void set_load_weight(struct task_struct *p) |
1330 | { | |
1331 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1332 | p->se.load.weight = prio_to_weight[0] * 2; |
1333 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1334 | return; | |
1335 | } | |
45bf76df | 1336 | |
dd41f596 IM |
1337 | /* |
1338 | * SCHED_IDLE tasks get minimal weight: | |
1339 | */ | |
1340 | if (p->policy == SCHED_IDLE) { | |
1341 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1342 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1343 | return; | |
1344 | } | |
71f8bd46 | 1345 | |
dd41f596 IM |
1346 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1347 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1348 | } |
1349 | ||
8159f87e | 1350 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1351 | { |
dd41f596 | 1352 | sched_info_queued(p); |
fd390f6a | 1353 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1354 | p->se.on_rq = 1; |
71f8bd46 IM |
1355 | } |
1356 | ||
69be72c1 | 1357 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1358 | { |
f02231e5 | 1359 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1360 | p->se.on_rq = 0; |
71f8bd46 IM |
1361 | } |
1362 | ||
14531189 | 1363 | /* |
dd41f596 | 1364 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1365 | */ |
14531189 IM |
1366 | static inline int __normal_prio(struct task_struct *p) |
1367 | { | |
dd41f596 | 1368 | return p->static_prio; |
14531189 IM |
1369 | } |
1370 | ||
b29739f9 IM |
1371 | /* |
1372 | * Calculate the expected normal priority: i.e. priority | |
1373 | * without taking RT-inheritance into account. Might be | |
1374 | * boosted by interactivity modifiers. Changes upon fork, | |
1375 | * setprio syscalls, and whenever the interactivity | |
1376 | * estimator recalculates. | |
1377 | */ | |
36c8b586 | 1378 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1379 | { |
1380 | int prio; | |
1381 | ||
e05606d3 | 1382 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1383 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1384 | else | |
1385 | prio = __normal_prio(p); | |
1386 | return prio; | |
1387 | } | |
1388 | ||
1389 | /* | |
1390 | * Calculate the current priority, i.e. the priority | |
1391 | * taken into account by the scheduler. This value might | |
1392 | * be boosted by RT tasks, or might be boosted by | |
1393 | * interactivity modifiers. Will be RT if the task got | |
1394 | * RT-boosted. If not then it returns p->normal_prio. | |
1395 | */ | |
36c8b586 | 1396 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1397 | { |
1398 | p->normal_prio = normal_prio(p); | |
1399 | /* | |
1400 | * If we are RT tasks or we were boosted to RT priority, | |
1401 | * keep the priority unchanged. Otherwise, update priority | |
1402 | * to the normal priority: | |
1403 | */ | |
1404 | if (!rt_prio(p->prio)) | |
1405 | return p->normal_prio; | |
1406 | return p->prio; | |
1407 | } | |
1408 | ||
1da177e4 | 1409 | /* |
dd41f596 | 1410 | * activate_task - move a task to the runqueue. |
1da177e4 | 1411 | */ |
dd41f596 | 1412 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1413 | { |
d9514f6c | 1414 | if (task_contributes_to_load(p)) |
dd41f596 | 1415 | rq->nr_uninterruptible--; |
1da177e4 | 1416 | |
8159f87e | 1417 | enqueue_task(rq, p, wakeup); |
62fb1851 | 1418 | inc_nr_running(p, rq); |
1da177e4 LT |
1419 | } |
1420 | ||
1da177e4 LT |
1421 | /* |
1422 | * deactivate_task - remove a task from the runqueue. | |
1423 | */ | |
2e1cb74a | 1424 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1425 | { |
d9514f6c | 1426 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1427 | rq->nr_uninterruptible++; |
1428 | ||
69be72c1 | 1429 | dequeue_task(rq, p, sleep); |
62fb1851 | 1430 | dec_nr_running(p, rq); |
1da177e4 LT |
1431 | } |
1432 | ||
1da177e4 LT |
1433 | /** |
1434 | * task_curr - is this task currently executing on a CPU? | |
1435 | * @p: the task in question. | |
1436 | */ | |
36c8b586 | 1437 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1438 | { |
1439 | return cpu_curr(task_cpu(p)) == p; | |
1440 | } | |
1441 | ||
2dd73a4f PW |
1442 | /* Used instead of source_load when we know the type == 0 */ |
1443 | unsigned long weighted_cpuload(const int cpu) | |
1444 | { | |
495eca49 | 1445 | return cpu_rq(cpu)->load.weight; |
dd41f596 IM |
1446 | } |
1447 | ||
1448 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1449 | { | |
6f505b16 | 1450 | set_task_rq(p, cpu); |
dd41f596 | 1451 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1452 | /* |
1453 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1454 | * successfuly executed on another CPU. We must ensure that updates of | |
1455 | * per-task data have been completed by this moment. | |
1456 | */ | |
1457 | smp_wmb(); | |
dd41f596 | 1458 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1459 | #endif |
2dd73a4f PW |
1460 | } |
1461 | ||
cb469845 SR |
1462 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1463 | const struct sched_class *prev_class, | |
1464 | int oldprio, int running) | |
1465 | { | |
1466 | if (prev_class != p->sched_class) { | |
1467 | if (prev_class->switched_from) | |
1468 | prev_class->switched_from(rq, p, running); | |
1469 | p->sched_class->switched_to(rq, p, running); | |
1470 | } else | |
1471 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1472 | } | |
1473 | ||
1da177e4 | 1474 | #ifdef CONFIG_SMP |
c65cc870 | 1475 | |
cc367732 IM |
1476 | /* |
1477 | * Is this task likely cache-hot: | |
1478 | */ | |
e7693a36 | 1479 | static int |
cc367732 IM |
1480 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1481 | { | |
1482 | s64 delta; | |
1483 | ||
f540a608 IM |
1484 | /* |
1485 | * Buddy candidates are cache hot: | |
1486 | */ | |
1487 | if (&p->se == cfs_rq_of(&p->se)->next) | |
1488 | return 1; | |
1489 | ||
cc367732 IM |
1490 | if (p->sched_class != &fair_sched_class) |
1491 | return 0; | |
1492 | ||
6bc1665b IM |
1493 | if (sysctl_sched_migration_cost == -1) |
1494 | return 1; | |
1495 | if (sysctl_sched_migration_cost == 0) | |
1496 | return 0; | |
1497 | ||
cc367732 IM |
1498 | delta = now - p->se.exec_start; |
1499 | ||
1500 | return delta < (s64)sysctl_sched_migration_cost; | |
1501 | } | |
1502 | ||
1503 | ||
dd41f596 | 1504 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1505 | { |
dd41f596 IM |
1506 | int old_cpu = task_cpu(p); |
1507 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1508 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1509 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1510 | u64 clock_offset; |
dd41f596 IM |
1511 | |
1512 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1513 | |
1514 | #ifdef CONFIG_SCHEDSTATS | |
1515 | if (p->se.wait_start) | |
1516 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1517 | if (p->se.sleep_start) |
1518 | p->se.sleep_start -= clock_offset; | |
1519 | if (p->se.block_start) | |
1520 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1521 | if (old_cpu != new_cpu) { |
1522 | schedstat_inc(p, se.nr_migrations); | |
1523 | if (task_hot(p, old_rq->clock, NULL)) | |
1524 | schedstat_inc(p, se.nr_forced2_migrations); | |
1525 | } | |
6cfb0d5d | 1526 | #endif |
2830cf8c SV |
1527 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1528 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1529 | |
1530 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1531 | } |
1532 | ||
70b97a7f | 1533 | struct migration_req { |
1da177e4 | 1534 | struct list_head list; |
1da177e4 | 1535 | |
36c8b586 | 1536 | struct task_struct *task; |
1da177e4 LT |
1537 | int dest_cpu; |
1538 | ||
1da177e4 | 1539 | struct completion done; |
70b97a7f | 1540 | }; |
1da177e4 LT |
1541 | |
1542 | /* | |
1543 | * The task's runqueue lock must be held. | |
1544 | * Returns true if you have to wait for migration thread. | |
1545 | */ | |
36c8b586 | 1546 | static int |
70b97a7f | 1547 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1548 | { |
70b97a7f | 1549 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1550 | |
1551 | /* | |
1552 | * If the task is not on a runqueue (and not running), then | |
1553 | * it is sufficient to simply update the task's cpu field. | |
1554 | */ | |
dd41f596 | 1555 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1556 | set_task_cpu(p, dest_cpu); |
1557 | return 0; | |
1558 | } | |
1559 | ||
1560 | init_completion(&req->done); | |
1da177e4 LT |
1561 | req->task = p; |
1562 | req->dest_cpu = dest_cpu; | |
1563 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1564 | |
1da177e4 LT |
1565 | return 1; |
1566 | } | |
1567 | ||
1568 | /* | |
1569 | * wait_task_inactive - wait for a thread to unschedule. | |
1570 | * | |
1571 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1572 | * else this function might spin for a *long* time. This function can't | |
1573 | * be called with interrupts off, or it may introduce deadlock with | |
1574 | * smp_call_function() if an IPI is sent by the same process we are | |
1575 | * waiting to become inactive. | |
1576 | */ | |
36c8b586 | 1577 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1578 | { |
1579 | unsigned long flags; | |
dd41f596 | 1580 | int running, on_rq; |
70b97a7f | 1581 | struct rq *rq; |
1da177e4 | 1582 | |
3a5c359a AK |
1583 | for (;;) { |
1584 | /* | |
1585 | * We do the initial early heuristics without holding | |
1586 | * any task-queue locks at all. We'll only try to get | |
1587 | * the runqueue lock when things look like they will | |
1588 | * work out! | |
1589 | */ | |
1590 | rq = task_rq(p); | |
fa490cfd | 1591 | |
3a5c359a AK |
1592 | /* |
1593 | * If the task is actively running on another CPU | |
1594 | * still, just relax and busy-wait without holding | |
1595 | * any locks. | |
1596 | * | |
1597 | * NOTE! Since we don't hold any locks, it's not | |
1598 | * even sure that "rq" stays as the right runqueue! | |
1599 | * But we don't care, since "task_running()" will | |
1600 | * return false if the runqueue has changed and p | |
1601 | * is actually now running somewhere else! | |
1602 | */ | |
1603 | while (task_running(rq, p)) | |
1604 | cpu_relax(); | |
fa490cfd | 1605 | |
3a5c359a AK |
1606 | /* |
1607 | * Ok, time to look more closely! We need the rq | |
1608 | * lock now, to be *sure*. If we're wrong, we'll | |
1609 | * just go back and repeat. | |
1610 | */ | |
1611 | rq = task_rq_lock(p, &flags); | |
1612 | running = task_running(rq, p); | |
1613 | on_rq = p->se.on_rq; | |
1614 | task_rq_unlock(rq, &flags); | |
fa490cfd | 1615 | |
3a5c359a AK |
1616 | /* |
1617 | * Was it really running after all now that we | |
1618 | * checked with the proper locks actually held? | |
1619 | * | |
1620 | * Oops. Go back and try again.. | |
1621 | */ | |
1622 | if (unlikely(running)) { | |
1623 | cpu_relax(); | |
1624 | continue; | |
1625 | } | |
fa490cfd | 1626 | |
3a5c359a AK |
1627 | /* |
1628 | * It's not enough that it's not actively running, | |
1629 | * it must be off the runqueue _entirely_, and not | |
1630 | * preempted! | |
1631 | * | |
1632 | * So if it wa still runnable (but just not actively | |
1633 | * running right now), it's preempted, and we should | |
1634 | * yield - it could be a while. | |
1635 | */ | |
1636 | if (unlikely(on_rq)) { | |
1637 | schedule_timeout_uninterruptible(1); | |
1638 | continue; | |
1639 | } | |
fa490cfd | 1640 | |
3a5c359a AK |
1641 | /* |
1642 | * Ahh, all good. It wasn't running, and it wasn't | |
1643 | * runnable, which means that it will never become | |
1644 | * running in the future either. We're all done! | |
1645 | */ | |
1646 | break; | |
1647 | } | |
1da177e4 LT |
1648 | } |
1649 | ||
1650 | /*** | |
1651 | * kick_process - kick a running thread to enter/exit the kernel | |
1652 | * @p: the to-be-kicked thread | |
1653 | * | |
1654 | * Cause a process which is running on another CPU to enter | |
1655 | * kernel-mode, without any delay. (to get signals handled.) | |
1656 | * | |
1657 | * NOTE: this function doesnt have to take the runqueue lock, | |
1658 | * because all it wants to ensure is that the remote task enters | |
1659 | * the kernel. If the IPI races and the task has been migrated | |
1660 | * to another CPU then no harm is done and the purpose has been | |
1661 | * achieved as well. | |
1662 | */ | |
36c8b586 | 1663 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1664 | { |
1665 | int cpu; | |
1666 | ||
1667 | preempt_disable(); | |
1668 | cpu = task_cpu(p); | |
1669 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1670 | smp_send_reschedule(cpu); | |
1671 | preempt_enable(); | |
1672 | } | |
1673 | ||
1674 | /* | |
2dd73a4f PW |
1675 | * Return a low guess at the load of a migration-source cpu weighted |
1676 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1677 | * |
1678 | * We want to under-estimate the load of migration sources, to | |
1679 | * balance conservatively. | |
1680 | */ | |
a9957449 | 1681 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 1682 | { |
70b97a7f | 1683 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1684 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1685 | |
3b0bd9bc | 1686 | if (type == 0) |
dd41f596 | 1687 | return total; |
b910472d | 1688 | |
dd41f596 | 1689 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
1690 | } |
1691 | ||
1692 | /* | |
2dd73a4f PW |
1693 | * Return a high guess at the load of a migration-target cpu weighted |
1694 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1695 | */ |
a9957449 | 1696 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 1697 | { |
70b97a7f | 1698 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1699 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1700 | |
7897986b | 1701 | if (type == 0) |
dd41f596 | 1702 | return total; |
3b0bd9bc | 1703 | |
dd41f596 | 1704 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
1705 | } |
1706 | ||
1707 | /* | |
1708 | * Return the average load per task on the cpu's run queue | |
1709 | */ | |
e7693a36 | 1710 | static unsigned long cpu_avg_load_per_task(int cpu) |
2dd73a4f | 1711 | { |
70b97a7f | 1712 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1713 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f PW |
1714 | unsigned long n = rq->nr_running; |
1715 | ||
dd41f596 | 1716 | return n ? total / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1717 | } |
1718 | ||
147cbb4b NP |
1719 | /* |
1720 | * find_idlest_group finds and returns the least busy CPU group within the | |
1721 | * domain. | |
1722 | */ | |
1723 | static struct sched_group * | |
1724 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1725 | { | |
1726 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1727 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1728 | int load_idx = sd->forkexec_idx; | |
1729 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1730 | ||
1731 | do { | |
1732 | unsigned long load, avg_load; | |
1733 | int local_group; | |
1734 | int i; | |
1735 | ||
da5a5522 BD |
1736 | /* Skip over this group if it has no CPUs allowed */ |
1737 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 1738 | continue; |
da5a5522 | 1739 | |
147cbb4b | 1740 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1741 | |
1742 | /* Tally up the load of all CPUs in the group */ | |
1743 | avg_load = 0; | |
1744 | ||
1745 | for_each_cpu_mask(i, group->cpumask) { | |
1746 | /* Bias balancing toward cpus of our domain */ | |
1747 | if (local_group) | |
1748 | load = source_load(i, load_idx); | |
1749 | else | |
1750 | load = target_load(i, load_idx); | |
1751 | ||
1752 | avg_load += load; | |
1753 | } | |
1754 | ||
1755 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1756 | avg_load = sg_div_cpu_power(group, |
1757 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1758 | |
1759 | if (local_group) { | |
1760 | this_load = avg_load; | |
1761 | this = group; | |
1762 | } else if (avg_load < min_load) { | |
1763 | min_load = avg_load; | |
1764 | idlest = group; | |
1765 | } | |
3a5c359a | 1766 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
1767 | |
1768 | if (!idlest || 100*this_load < imbalance*min_load) | |
1769 | return NULL; | |
1770 | return idlest; | |
1771 | } | |
1772 | ||
1773 | /* | |
0feaece9 | 1774 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1775 | */ |
95cdf3b7 IM |
1776 | static int |
1777 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) | |
147cbb4b | 1778 | { |
da5a5522 | 1779 | cpumask_t tmp; |
147cbb4b NP |
1780 | unsigned long load, min_load = ULONG_MAX; |
1781 | int idlest = -1; | |
1782 | int i; | |
1783 | ||
da5a5522 BD |
1784 | /* Traverse only the allowed CPUs */ |
1785 | cpus_and(tmp, group->cpumask, p->cpus_allowed); | |
1786 | ||
1787 | for_each_cpu_mask(i, tmp) { | |
2dd73a4f | 1788 | load = weighted_cpuload(i); |
147cbb4b NP |
1789 | |
1790 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1791 | min_load = load; | |
1792 | idlest = i; | |
1793 | } | |
1794 | } | |
1795 | ||
1796 | return idlest; | |
1797 | } | |
1798 | ||
476d139c NP |
1799 | /* |
1800 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1801 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1802 | * SD_BALANCE_EXEC. | |
1803 | * | |
1804 | * Balance, ie. select the least loaded group. | |
1805 | * | |
1806 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1807 | * | |
1808 | * preempt must be disabled. | |
1809 | */ | |
1810 | static int sched_balance_self(int cpu, int flag) | |
1811 | { | |
1812 | struct task_struct *t = current; | |
1813 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1814 | |
c96d145e | 1815 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
1816 | /* |
1817 | * If power savings logic is enabled for a domain, stop there. | |
1818 | */ | |
5c45bf27 SS |
1819 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
1820 | break; | |
476d139c NP |
1821 | if (tmp->flags & flag) |
1822 | sd = tmp; | |
c96d145e | 1823 | } |
476d139c NP |
1824 | |
1825 | while (sd) { | |
1826 | cpumask_t span; | |
1827 | struct sched_group *group; | |
1a848870 SS |
1828 | int new_cpu, weight; |
1829 | ||
1830 | if (!(sd->flags & flag)) { | |
1831 | sd = sd->child; | |
1832 | continue; | |
1833 | } | |
476d139c NP |
1834 | |
1835 | span = sd->span; | |
1836 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1837 | if (!group) { |
1838 | sd = sd->child; | |
1839 | continue; | |
1840 | } | |
476d139c | 1841 | |
da5a5522 | 1842 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
1843 | if (new_cpu == -1 || new_cpu == cpu) { |
1844 | /* Now try balancing at a lower domain level of cpu */ | |
1845 | sd = sd->child; | |
1846 | continue; | |
1847 | } | |
476d139c | 1848 | |
1a848870 | 1849 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1850 | cpu = new_cpu; |
476d139c NP |
1851 | sd = NULL; |
1852 | weight = cpus_weight(span); | |
1853 | for_each_domain(cpu, tmp) { | |
1854 | if (weight <= cpus_weight(tmp->span)) | |
1855 | break; | |
1856 | if (tmp->flags & flag) | |
1857 | sd = tmp; | |
1858 | } | |
1859 | /* while loop will break here if sd == NULL */ | |
1860 | } | |
1861 | ||
1862 | return cpu; | |
1863 | } | |
1864 | ||
1865 | #endif /* CONFIG_SMP */ | |
1da177e4 | 1866 | |
1da177e4 LT |
1867 | /*** |
1868 | * try_to_wake_up - wake up a thread | |
1869 | * @p: the to-be-woken-up thread | |
1870 | * @state: the mask of task states that can be woken | |
1871 | * @sync: do a synchronous wakeup? | |
1872 | * | |
1873 | * Put it on the run-queue if it's not already there. The "current" | |
1874 | * thread is always on the run-queue (except when the actual | |
1875 | * re-schedule is in progress), and as such you're allowed to do | |
1876 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1877 | * runnable without the overhead of this. | |
1878 | * | |
1879 | * returns failure only if the task is already active. | |
1880 | */ | |
36c8b586 | 1881 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 1882 | { |
cc367732 | 1883 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
1884 | unsigned long flags; |
1885 | long old_state; | |
70b97a7f | 1886 | struct rq *rq; |
1da177e4 | 1887 | |
04e2f174 | 1888 | smp_wmb(); |
1da177e4 LT |
1889 | rq = task_rq_lock(p, &flags); |
1890 | old_state = p->state; | |
1891 | if (!(old_state & state)) | |
1892 | goto out; | |
1893 | ||
dd41f596 | 1894 | if (p->se.on_rq) |
1da177e4 LT |
1895 | goto out_running; |
1896 | ||
1897 | cpu = task_cpu(p); | |
cc367732 | 1898 | orig_cpu = cpu; |
1da177e4 LT |
1899 | this_cpu = smp_processor_id(); |
1900 | ||
1901 | #ifdef CONFIG_SMP | |
1902 | if (unlikely(task_running(rq, p))) | |
1903 | goto out_activate; | |
1904 | ||
5d2f5a61 DA |
1905 | cpu = p->sched_class->select_task_rq(p, sync); |
1906 | if (cpu != orig_cpu) { | |
1907 | set_task_cpu(p, cpu); | |
1da177e4 LT |
1908 | task_rq_unlock(rq, &flags); |
1909 | /* might preempt at this point */ | |
1910 | rq = task_rq_lock(p, &flags); | |
1911 | old_state = p->state; | |
1912 | if (!(old_state & state)) | |
1913 | goto out; | |
dd41f596 | 1914 | if (p->se.on_rq) |
1da177e4 LT |
1915 | goto out_running; |
1916 | ||
1917 | this_cpu = smp_processor_id(); | |
1918 | cpu = task_cpu(p); | |
1919 | } | |
1920 | ||
e7693a36 GH |
1921 | #ifdef CONFIG_SCHEDSTATS |
1922 | schedstat_inc(rq, ttwu_count); | |
1923 | if (cpu == this_cpu) | |
1924 | schedstat_inc(rq, ttwu_local); | |
1925 | else { | |
1926 | struct sched_domain *sd; | |
1927 | for_each_domain(this_cpu, sd) { | |
1928 | if (cpu_isset(cpu, sd->span)) { | |
1929 | schedstat_inc(sd, ttwu_wake_remote); | |
1930 | break; | |
1931 | } | |
1932 | } | |
1933 | } | |
e7693a36 GH |
1934 | #endif |
1935 | ||
1da177e4 LT |
1936 | out_activate: |
1937 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
1938 | schedstat_inc(p, se.nr_wakeups); |
1939 | if (sync) | |
1940 | schedstat_inc(p, se.nr_wakeups_sync); | |
1941 | if (orig_cpu != cpu) | |
1942 | schedstat_inc(p, se.nr_wakeups_migrate); | |
1943 | if (cpu == this_cpu) | |
1944 | schedstat_inc(p, se.nr_wakeups_local); | |
1945 | else | |
1946 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 1947 | update_rq_clock(rq); |
dd41f596 | 1948 | activate_task(rq, p, 1); |
1da177e4 LT |
1949 | success = 1; |
1950 | ||
1951 | out_running: | |
4ae7d5ce IM |
1952 | check_preempt_curr(rq, p); |
1953 | ||
1da177e4 | 1954 | p->state = TASK_RUNNING; |
9a897c5a SR |
1955 | #ifdef CONFIG_SMP |
1956 | if (p->sched_class->task_wake_up) | |
1957 | p->sched_class->task_wake_up(rq, p); | |
1958 | #endif | |
1da177e4 LT |
1959 | out: |
1960 | task_rq_unlock(rq, &flags); | |
1961 | ||
1962 | return success; | |
1963 | } | |
1964 | ||
7ad5b3a5 | 1965 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1966 | { |
d9514f6c | 1967 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 1968 | } |
1da177e4 LT |
1969 | EXPORT_SYMBOL(wake_up_process); |
1970 | ||
7ad5b3a5 | 1971 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1972 | { |
1973 | return try_to_wake_up(p, state, 0); | |
1974 | } | |
1975 | ||
1da177e4 LT |
1976 | /* |
1977 | * Perform scheduler related setup for a newly forked process p. | |
1978 | * p is forked by current. | |
dd41f596 IM |
1979 | * |
1980 | * __sched_fork() is basic setup used by init_idle() too: | |
1981 | */ | |
1982 | static void __sched_fork(struct task_struct *p) | |
1983 | { | |
dd41f596 IM |
1984 | p->se.exec_start = 0; |
1985 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1986 | p->se.prev_sum_exec_runtime = 0; |
4ae7d5ce IM |
1987 | p->se.last_wakeup = 0; |
1988 | p->se.avg_overlap = 0; | |
6cfb0d5d IM |
1989 | |
1990 | #ifdef CONFIG_SCHEDSTATS | |
1991 | p->se.wait_start = 0; | |
dd41f596 IM |
1992 | p->se.sum_sleep_runtime = 0; |
1993 | p->se.sleep_start = 0; | |
dd41f596 IM |
1994 | p->se.block_start = 0; |
1995 | p->se.sleep_max = 0; | |
1996 | p->se.block_max = 0; | |
1997 | p->se.exec_max = 0; | |
eba1ed4b | 1998 | p->se.slice_max = 0; |
dd41f596 | 1999 | p->se.wait_max = 0; |
6cfb0d5d | 2000 | #endif |
476d139c | 2001 | |
fa717060 | 2002 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2003 | p->se.on_rq = 0; |
476d139c | 2004 | |
e107be36 AK |
2005 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2006 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2007 | #endif | |
2008 | ||
1da177e4 LT |
2009 | /* |
2010 | * We mark the process as running here, but have not actually | |
2011 | * inserted it onto the runqueue yet. This guarantees that | |
2012 | * nobody will actually run it, and a signal or other external | |
2013 | * event cannot wake it up and insert it on the runqueue either. | |
2014 | */ | |
2015 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2016 | } |
2017 | ||
2018 | /* | |
2019 | * fork()/clone()-time setup: | |
2020 | */ | |
2021 | void sched_fork(struct task_struct *p, int clone_flags) | |
2022 | { | |
2023 | int cpu = get_cpu(); | |
2024 | ||
2025 | __sched_fork(p); | |
2026 | ||
2027 | #ifdef CONFIG_SMP | |
2028 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2029 | #endif | |
02e4bac2 | 2030 | set_task_cpu(p, cpu); |
b29739f9 IM |
2031 | |
2032 | /* | |
2033 | * Make sure we do not leak PI boosting priority to the child: | |
2034 | */ | |
2035 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2036 | if (!rt_prio(p->prio)) |
2037 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2038 | |
52f17b6c | 2039 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2040 | if (likely(sched_info_on())) |
52f17b6c | 2041 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2042 | #endif |
d6077cb8 | 2043 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2044 | p->oncpu = 0; |
2045 | #endif | |
1da177e4 | 2046 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2047 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2048 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2049 | #endif |
476d139c | 2050 | put_cpu(); |
1da177e4 LT |
2051 | } |
2052 | ||
2053 | /* | |
2054 | * wake_up_new_task - wake up a newly created task for the first time. | |
2055 | * | |
2056 | * This function will do some initial scheduler statistics housekeeping | |
2057 | * that must be done for every newly created context, then puts the task | |
2058 | * on the runqueue and wakes it. | |
2059 | */ | |
7ad5b3a5 | 2060 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2061 | { |
2062 | unsigned long flags; | |
dd41f596 | 2063 | struct rq *rq; |
1da177e4 LT |
2064 | |
2065 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2066 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2067 | update_rq_clock(rq); |
1da177e4 LT |
2068 | |
2069 | p->prio = effective_prio(p); | |
2070 | ||
b9dca1e0 | 2071 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2072 | activate_task(rq, p, 0); |
1da177e4 | 2073 | } else { |
1da177e4 | 2074 | /* |
dd41f596 IM |
2075 | * Let the scheduling class do new task startup |
2076 | * management (if any): | |
1da177e4 | 2077 | */ |
ee0827d8 | 2078 | p->sched_class->task_new(rq, p); |
62fb1851 | 2079 | inc_nr_running(p, rq); |
1da177e4 | 2080 | } |
dd41f596 | 2081 | check_preempt_curr(rq, p); |
9a897c5a SR |
2082 | #ifdef CONFIG_SMP |
2083 | if (p->sched_class->task_wake_up) | |
2084 | p->sched_class->task_wake_up(rq, p); | |
2085 | #endif | |
dd41f596 | 2086 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2087 | } |
2088 | ||
e107be36 AK |
2089 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2090 | ||
2091 | /** | |
421cee29 RD |
2092 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
2093 | * @notifier: notifier struct to register | |
e107be36 AK |
2094 | */ |
2095 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2096 | { | |
2097 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2098 | } | |
2099 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2100 | ||
2101 | /** | |
2102 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2103 | * @notifier: notifier struct to unregister |
e107be36 AK |
2104 | * |
2105 | * This is safe to call from within a preemption notifier. | |
2106 | */ | |
2107 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2108 | { | |
2109 | hlist_del(¬ifier->link); | |
2110 | } | |
2111 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2112 | ||
2113 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2114 | { | |
2115 | struct preempt_notifier *notifier; | |
2116 | struct hlist_node *node; | |
2117 | ||
2118 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2119 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2120 | } | |
2121 | ||
2122 | static void | |
2123 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2124 | struct task_struct *next) | |
2125 | { | |
2126 | struct preempt_notifier *notifier; | |
2127 | struct hlist_node *node; | |
2128 | ||
2129 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2130 | notifier->ops->sched_out(notifier, next); | |
2131 | } | |
2132 | ||
2133 | #else | |
2134 | ||
2135 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2136 | { | |
2137 | } | |
2138 | ||
2139 | static void | |
2140 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2141 | struct task_struct *next) | |
2142 | { | |
2143 | } | |
2144 | ||
2145 | #endif | |
2146 | ||
4866cde0 NP |
2147 | /** |
2148 | * prepare_task_switch - prepare to switch tasks | |
2149 | * @rq: the runqueue preparing to switch | |
421cee29 | 2150 | * @prev: the current task that is being switched out |
4866cde0 NP |
2151 | * @next: the task we are going to switch to. |
2152 | * | |
2153 | * This is called with the rq lock held and interrupts off. It must | |
2154 | * be paired with a subsequent finish_task_switch after the context | |
2155 | * switch. | |
2156 | * | |
2157 | * prepare_task_switch sets up locking and calls architecture specific | |
2158 | * hooks. | |
2159 | */ | |
e107be36 AK |
2160 | static inline void |
2161 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2162 | struct task_struct *next) | |
4866cde0 | 2163 | { |
e107be36 | 2164 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2165 | prepare_lock_switch(rq, next); |
2166 | prepare_arch_switch(next); | |
2167 | } | |
2168 | ||
1da177e4 LT |
2169 | /** |
2170 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2171 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2172 | * @prev: the thread we just switched away from. |
2173 | * | |
4866cde0 NP |
2174 | * finish_task_switch must be called after the context switch, paired |
2175 | * with a prepare_task_switch call before the context switch. | |
2176 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2177 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2178 | * |
2179 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2180 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2181 | * with the lock held can cause deadlocks; see schedule() for |
2182 | * details.) | |
2183 | */ | |
a9957449 | 2184 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2185 | __releases(rq->lock) |
2186 | { | |
1da177e4 | 2187 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2188 | long prev_state; |
1da177e4 LT |
2189 | |
2190 | rq->prev_mm = NULL; | |
2191 | ||
2192 | /* | |
2193 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2194 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2195 | * schedule one last time. The schedule call will never return, and |
2196 | * the scheduled task must drop that reference. | |
c394cc9f | 2197 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2198 | * still held, otherwise prev could be scheduled on another cpu, die |
2199 | * there before we look at prev->state, and then the reference would | |
2200 | * be dropped twice. | |
2201 | * Manfred Spraul <manfred@colorfullife.com> | |
2202 | */ | |
55a101f8 | 2203 | prev_state = prev->state; |
4866cde0 NP |
2204 | finish_arch_switch(prev); |
2205 | finish_lock_switch(rq, prev); | |
9a897c5a SR |
2206 | #ifdef CONFIG_SMP |
2207 | if (current->sched_class->post_schedule) | |
2208 | current->sched_class->post_schedule(rq); | |
2209 | #endif | |
e8fa1362 | 2210 | |
e107be36 | 2211 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2212 | if (mm) |
2213 | mmdrop(mm); | |
c394cc9f | 2214 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2215 | /* |
2216 | * Remove function-return probe instances associated with this | |
2217 | * task and put them back on the free list. | |
9761eea8 | 2218 | */ |
c6fd91f0 | 2219 | kprobe_flush_task(prev); |
1da177e4 | 2220 | put_task_struct(prev); |
c6fd91f0 | 2221 | } |
1da177e4 LT |
2222 | } |
2223 | ||
2224 | /** | |
2225 | * schedule_tail - first thing a freshly forked thread must call. | |
2226 | * @prev: the thread we just switched away from. | |
2227 | */ | |
36c8b586 | 2228 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2229 | __releases(rq->lock) |
2230 | { | |
70b97a7f IM |
2231 | struct rq *rq = this_rq(); |
2232 | ||
4866cde0 NP |
2233 | finish_task_switch(rq, prev); |
2234 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2235 | /* In this case, finish_task_switch does not reenable preemption */ | |
2236 | preempt_enable(); | |
2237 | #endif | |
1da177e4 | 2238 | if (current->set_child_tid) |
b488893a | 2239 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2240 | } |
2241 | ||
2242 | /* | |
2243 | * context_switch - switch to the new MM and the new | |
2244 | * thread's register state. | |
2245 | */ | |
dd41f596 | 2246 | static inline void |
70b97a7f | 2247 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2248 | struct task_struct *next) |
1da177e4 | 2249 | { |
dd41f596 | 2250 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2251 | |
e107be36 | 2252 | prepare_task_switch(rq, prev, next); |
dd41f596 IM |
2253 | mm = next->mm; |
2254 | oldmm = prev->active_mm; | |
9226d125 ZA |
2255 | /* |
2256 | * For paravirt, this is coupled with an exit in switch_to to | |
2257 | * combine the page table reload and the switch backend into | |
2258 | * one hypercall. | |
2259 | */ | |
2260 | arch_enter_lazy_cpu_mode(); | |
2261 | ||
dd41f596 | 2262 | if (unlikely(!mm)) { |
1da177e4 LT |
2263 | next->active_mm = oldmm; |
2264 | atomic_inc(&oldmm->mm_count); | |
2265 | enter_lazy_tlb(oldmm, next); | |
2266 | } else | |
2267 | switch_mm(oldmm, mm, next); | |
2268 | ||
dd41f596 | 2269 | if (unlikely(!prev->mm)) { |
1da177e4 | 2270 | prev->active_mm = NULL; |
1da177e4 LT |
2271 | rq->prev_mm = oldmm; |
2272 | } | |
3a5f5e48 IM |
2273 | /* |
2274 | * Since the runqueue lock will be released by the next | |
2275 | * task (which is an invalid locking op but in the case | |
2276 | * of the scheduler it's an obvious special-case), so we | |
2277 | * do an early lockdep release here: | |
2278 | */ | |
2279 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2280 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2281 | #endif |
1da177e4 LT |
2282 | |
2283 | /* Here we just switch the register state and the stack. */ | |
2284 | switch_to(prev, next, prev); | |
2285 | ||
dd41f596 IM |
2286 | barrier(); |
2287 | /* | |
2288 | * this_rq must be evaluated again because prev may have moved | |
2289 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2290 | * frame will be invalid. | |
2291 | */ | |
2292 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2293 | } |
2294 | ||
2295 | /* | |
2296 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2297 | * | |
2298 | * externally visible scheduler statistics: current number of runnable | |
2299 | * threads, current number of uninterruptible-sleeping threads, total | |
2300 | * number of context switches performed since bootup. | |
2301 | */ | |
2302 | unsigned long nr_running(void) | |
2303 | { | |
2304 | unsigned long i, sum = 0; | |
2305 | ||
2306 | for_each_online_cpu(i) | |
2307 | sum += cpu_rq(i)->nr_running; | |
2308 | ||
2309 | return sum; | |
2310 | } | |
2311 | ||
2312 | unsigned long nr_uninterruptible(void) | |
2313 | { | |
2314 | unsigned long i, sum = 0; | |
2315 | ||
0a945022 | 2316 | for_each_possible_cpu(i) |
1da177e4 LT |
2317 | sum += cpu_rq(i)->nr_uninterruptible; |
2318 | ||
2319 | /* | |
2320 | * Since we read the counters lockless, it might be slightly | |
2321 | * inaccurate. Do not allow it to go below zero though: | |
2322 | */ | |
2323 | if (unlikely((long)sum < 0)) | |
2324 | sum = 0; | |
2325 | ||
2326 | return sum; | |
2327 | } | |
2328 | ||
2329 | unsigned long long nr_context_switches(void) | |
2330 | { | |
cc94abfc SR |
2331 | int i; |
2332 | unsigned long long sum = 0; | |
1da177e4 | 2333 | |
0a945022 | 2334 | for_each_possible_cpu(i) |
1da177e4 LT |
2335 | sum += cpu_rq(i)->nr_switches; |
2336 | ||
2337 | return sum; | |
2338 | } | |
2339 | ||
2340 | unsigned long nr_iowait(void) | |
2341 | { | |
2342 | unsigned long i, sum = 0; | |
2343 | ||
0a945022 | 2344 | for_each_possible_cpu(i) |
1da177e4 LT |
2345 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2346 | ||
2347 | return sum; | |
2348 | } | |
2349 | ||
db1b1fef JS |
2350 | unsigned long nr_active(void) |
2351 | { | |
2352 | unsigned long i, running = 0, uninterruptible = 0; | |
2353 | ||
2354 | for_each_online_cpu(i) { | |
2355 | running += cpu_rq(i)->nr_running; | |
2356 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2357 | } | |
2358 | ||
2359 | if (unlikely((long)uninterruptible < 0)) | |
2360 | uninterruptible = 0; | |
2361 | ||
2362 | return running + uninterruptible; | |
2363 | } | |
2364 | ||
48f24c4d | 2365 | /* |
dd41f596 IM |
2366 | * Update rq->cpu_load[] statistics. This function is usually called every |
2367 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2368 | */ |
dd41f596 | 2369 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2370 | { |
495eca49 | 2371 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2372 | int i, scale; |
2373 | ||
2374 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2375 | |
2376 | /* Update our load: */ | |
2377 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2378 | unsigned long old_load, new_load; | |
2379 | ||
2380 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2381 | ||
2382 | old_load = this_rq->cpu_load[i]; | |
2383 | new_load = this_load; | |
a25707f3 IM |
2384 | /* |
2385 | * Round up the averaging division if load is increasing. This | |
2386 | * prevents us from getting stuck on 9 if the load is 10, for | |
2387 | * example. | |
2388 | */ | |
2389 | if (new_load > old_load) | |
2390 | new_load += scale-1; | |
dd41f596 IM |
2391 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2392 | } | |
48f24c4d IM |
2393 | } |
2394 | ||
dd41f596 IM |
2395 | #ifdef CONFIG_SMP |
2396 | ||
1da177e4 LT |
2397 | /* |
2398 | * double_rq_lock - safely lock two runqueues | |
2399 | * | |
2400 | * Note this does not disable interrupts like task_rq_lock, | |
2401 | * you need to do so manually before calling. | |
2402 | */ | |
70b97a7f | 2403 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2404 | __acquires(rq1->lock) |
2405 | __acquires(rq2->lock) | |
2406 | { | |
054b9108 | 2407 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2408 | if (rq1 == rq2) { |
2409 | spin_lock(&rq1->lock); | |
2410 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2411 | } else { | |
c96d145e | 2412 | if (rq1 < rq2) { |
1da177e4 LT |
2413 | spin_lock(&rq1->lock); |
2414 | spin_lock(&rq2->lock); | |
2415 | } else { | |
2416 | spin_lock(&rq2->lock); | |
2417 | spin_lock(&rq1->lock); | |
2418 | } | |
2419 | } | |
6e82a3be IM |
2420 | update_rq_clock(rq1); |
2421 | update_rq_clock(rq2); | |
1da177e4 LT |
2422 | } |
2423 | ||
2424 | /* | |
2425 | * double_rq_unlock - safely unlock two runqueues | |
2426 | * | |
2427 | * Note this does not restore interrupts like task_rq_unlock, | |
2428 | * you need to do so manually after calling. | |
2429 | */ | |
70b97a7f | 2430 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2431 | __releases(rq1->lock) |
2432 | __releases(rq2->lock) | |
2433 | { | |
2434 | spin_unlock(&rq1->lock); | |
2435 | if (rq1 != rq2) | |
2436 | spin_unlock(&rq2->lock); | |
2437 | else | |
2438 | __release(rq2->lock); | |
2439 | } | |
2440 | ||
2441 | /* | |
2442 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2443 | */ | |
e8fa1362 | 2444 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2445 | __releases(this_rq->lock) |
2446 | __acquires(busiest->lock) | |
2447 | __acquires(this_rq->lock) | |
2448 | { | |
e8fa1362 SR |
2449 | int ret = 0; |
2450 | ||
054b9108 KK |
2451 | if (unlikely(!irqs_disabled())) { |
2452 | /* printk() doesn't work good under rq->lock */ | |
2453 | spin_unlock(&this_rq->lock); | |
2454 | BUG_ON(1); | |
2455 | } | |
1da177e4 | 2456 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2457 | if (busiest < this_rq) { |
1da177e4 LT |
2458 | spin_unlock(&this_rq->lock); |
2459 | spin_lock(&busiest->lock); | |
2460 | spin_lock(&this_rq->lock); | |
e8fa1362 | 2461 | ret = 1; |
1da177e4 LT |
2462 | } else |
2463 | spin_lock(&busiest->lock); | |
2464 | } | |
e8fa1362 | 2465 | return ret; |
1da177e4 LT |
2466 | } |
2467 | ||
1da177e4 LT |
2468 | /* |
2469 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2470 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2471 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2472 | * the cpu_allowed mask is restored. |
2473 | */ | |
36c8b586 | 2474 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2475 | { |
70b97a7f | 2476 | struct migration_req req; |
1da177e4 | 2477 | unsigned long flags; |
70b97a7f | 2478 | struct rq *rq; |
1da177e4 LT |
2479 | |
2480 | rq = task_rq_lock(p, &flags); | |
2481 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2482 | || unlikely(cpu_is_offline(dest_cpu))) | |
2483 | goto out; | |
2484 | ||
2485 | /* force the process onto the specified CPU */ | |
2486 | if (migrate_task(p, dest_cpu, &req)) { | |
2487 | /* Need to wait for migration thread (might exit: take ref). */ | |
2488 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2489 | |
1da177e4 LT |
2490 | get_task_struct(mt); |
2491 | task_rq_unlock(rq, &flags); | |
2492 | wake_up_process(mt); | |
2493 | put_task_struct(mt); | |
2494 | wait_for_completion(&req.done); | |
36c8b586 | 2495 | |
1da177e4 LT |
2496 | return; |
2497 | } | |
2498 | out: | |
2499 | task_rq_unlock(rq, &flags); | |
2500 | } | |
2501 | ||
2502 | /* | |
476d139c NP |
2503 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2504 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2505 | */ |
2506 | void sched_exec(void) | |
2507 | { | |
1da177e4 | 2508 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2509 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2510 | put_cpu(); |
476d139c NP |
2511 | if (new_cpu != this_cpu) |
2512 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2513 | } |
2514 | ||
2515 | /* | |
2516 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2517 | * Both runqueues must be locked. | |
2518 | */ | |
dd41f596 IM |
2519 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2520 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2521 | { |
2e1cb74a | 2522 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2523 | set_task_cpu(p, this_cpu); |
dd41f596 | 2524 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2525 | /* |
2526 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2527 | * to be always true for them. | |
2528 | */ | |
dd41f596 | 2529 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2530 | } |
2531 | ||
2532 | /* | |
2533 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2534 | */ | |
858119e1 | 2535 | static |
70b97a7f | 2536 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2537 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2538 | int *all_pinned) |
1da177e4 LT |
2539 | { |
2540 | /* | |
2541 | * We do not migrate tasks that are: | |
2542 | * 1) running (obviously), or | |
2543 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2544 | * 3) are cache-hot on their current CPU. | |
2545 | */ | |
cc367732 IM |
2546 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2547 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2548 | return 0; |
cc367732 | 2549 | } |
81026794 NP |
2550 | *all_pinned = 0; |
2551 | ||
cc367732 IM |
2552 | if (task_running(rq, p)) { |
2553 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2554 | return 0; |
cc367732 | 2555 | } |
1da177e4 | 2556 | |
da84d961 IM |
2557 | /* |
2558 | * Aggressive migration if: | |
2559 | * 1) task is cache cold, or | |
2560 | * 2) too many balance attempts have failed. | |
2561 | */ | |
2562 | ||
6bc1665b IM |
2563 | if (!task_hot(p, rq->clock, sd) || |
2564 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2565 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2566 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2567 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2568 | schedstat_inc(p, se.nr_forced_migrations); |
2569 | } | |
da84d961 IM |
2570 | #endif |
2571 | return 1; | |
2572 | } | |
2573 | ||
cc367732 IM |
2574 | if (task_hot(p, rq->clock, sd)) { |
2575 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2576 | return 0; |
cc367732 | 2577 | } |
1da177e4 LT |
2578 | return 1; |
2579 | } | |
2580 | ||
e1d1484f PW |
2581 | static unsigned long |
2582 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2583 | unsigned long max_load_move, struct sched_domain *sd, | |
2584 | enum cpu_idle_type idle, int *all_pinned, | |
2585 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2586 | { |
b82d9fdd | 2587 | int loops = 0, pulled = 0, pinned = 0, skip_for_load; |
dd41f596 IM |
2588 | struct task_struct *p; |
2589 | long rem_load_move = max_load_move; | |
1da177e4 | 2590 | |
e1d1484f | 2591 | if (max_load_move == 0) |
1da177e4 LT |
2592 | goto out; |
2593 | ||
81026794 NP |
2594 | pinned = 1; |
2595 | ||
1da177e4 | 2596 | /* |
dd41f596 | 2597 | * Start the load-balancing iterator: |
1da177e4 | 2598 | */ |
dd41f596 IM |
2599 | p = iterator->start(iterator->arg); |
2600 | next: | |
b82d9fdd | 2601 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2602 | goto out; |
50ddd969 | 2603 | /* |
b82d9fdd | 2604 | * To help distribute high priority tasks across CPUs we don't |
50ddd969 PW |
2605 | * skip a task if it will be the highest priority task (i.e. smallest |
2606 | * prio value) on its new queue regardless of its load weight | |
2607 | */ | |
dd41f596 IM |
2608 | skip_for_load = (p->se.load.weight >> 1) > rem_load_move + |
2609 | SCHED_LOAD_SCALE_FUZZ; | |
a4ac01c3 | 2610 | if ((skip_for_load && p->prio >= *this_best_prio) || |
dd41f596 | 2611 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2612 | p = iterator->next(iterator->arg); |
2613 | goto next; | |
1da177e4 LT |
2614 | } |
2615 | ||
dd41f596 | 2616 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2617 | pulled++; |
dd41f596 | 2618 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2619 | |
2dd73a4f | 2620 | /* |
b82d9fdd | 2621 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 2622 | */ |
e1d1484f | 2623 | if (rem_load_move > 0) { |
a4ac01c3 PW |
2624 | if (p->prio < *this_best_prio) |
2625 | *this_best_prio = p->prio; | |
dd41f596 IM |
2626 | p = iterator->next(iterator->arg); |
2627 | goto next; | |
1da177e4 LT |
2628 | } |
2629 | out: | |
2630 | /* | |
e1d1484f | 2631 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
2632 | * so we can safely collect pull_task() stats here rather than |
2633 | * inside pull_task(). | |
2634 | */ | |
2635 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2636 | |
2637 | if (all_pinned) | |
2638 | *all_pinned = pinned; | |
e1d1484f PW |
2639 | |
2640 | return max_load_move - rem_load_move; | |
1da177e4 LT |
2641 | } |
2642 | ||
dd41f596 | 2643 | /* |
43010659 PW |
2644 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2645 | * this_rq, as part of a balancing operation within domain "sd". | |
2646 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2647 | * |
2648 | * Called with both runqueues locked. | |
2649 | */ | |
2650 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2651 | unsigned long max_load_move, |
dd41f596 IM |
2652 | struct sched_domain *sd, enum cpu_idle_type idle, |
2653 | int *all_pinned) | |
2654 | { | |
5522d5d5 | 2655 | const struct sched_class *class = sched_class_highest; |
43010659 | 2656 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2657 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
2658 | |
2659 | do { | |
43010659 PW |
2660 | total_load_moved += |
2661 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 2662 | max_load_move - total_load_moved, |
a4ac01c3 | 2663 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 2664 | class = class->next; |
43010659 | 2665 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 2666 | |
43010659 PW |
2667 | return total_load_moved > 0; |
2668 | } | |
2669 | ||
e1d1484f PW |
2670 | static int |
2671 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2672 | struct sched_domain *sd, enum cpu_idle_type idle, | |
2673 | struct rq_iterator *iterator) | |
2674 | { | |
2675 | struct task_struct *p = iterator->start(iterator->arg); | |
2676 | int pinned = 0; | |
2677 | ||
2678 | while (p) { | |
2679 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
2680 | pull_task(busiest, p, this_rq, this_cpu); | |
2681 | /* | |
2682 | * Right now, this is only the second place pull_task() | |
2683 | * is called, so we can safely collect pull_task() | |
2684 | * stats here rather than inside pull_task(). | |
2685 | */ | |
2686 | schedstat_inc(sd, lb_gained[idle]); | |
2687 | ||
2688 | return 1; | |
2689 | } | |
2690 | p = iterator->next(iterator->arg); | |
2691 | } | |
2692 | ||
2693 | return 0; | |
2694 | } | |
2695 | ||
43010659 PW |
2696 | /* |
2697 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
2698 | * part of active balancing operations within "domain". | |
2699 | * Returns 1 if successful and 0 otherwise. | |
2700 | * | |
2701 | * Called with both runqueues locked. | |
2702 | */ | |
2703 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2704 | struct sched_domain *sd, enum cpu_idle_type idle) | |
2705 | { | |
5522d5d5 | 2706 | const struct sched_class *class; |
43010659 PW |
2707 | |
2708 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 2709 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
2710 | return 1; |
2711 | ||
2712 | return 0; | |
dd41f596 IM |
2713 | } |
2714 | ||
1da177e4 LT |
2715 | /* |
2716 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2717 | * domain. It calculates and returns the amount of weighted load which |
2718 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2719 | */ |
2720 | static struct sched_group * | |
2721 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 IM |
2722 | unsigned long *imbalance, enum cpu_idle_type idle, |
2723 | int *sd_idle, cpumask_t *cpus, int *balance) | |
1da177e4 LT |
2724 | { |
2725 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2726 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2727 | unsigned long max_pull; |
2dd73a4f PW |
2728 | unsigned long busiest_load_per_task, busiest_nr_running; |
2729 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 2730 | int load_idx, group_imb = 0; |
5c45bf27 SS |
2731 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2732 | int power_savings_balance = 1; | |
2733 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2734 | unsigned long min_nr_running = ULONG_MAX; | |
2735 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2736 | #endif | |
1da177e4 LT |
2737 | |
2738 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2739 | busiest_load_per_task = busiest_nr_running = 0; |
2740 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2741 | if (idle == CPU_NOT_IDLE) |
7897986b | 2742 | load_idx = sd->busy_idx; |
d15bcfdb | 2743 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2744 | load_idx = sd->newidle_idx; |
2745 | else | |
2746 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2747 | |
2748 | do { | |
908a7c1b | 2749 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
2750 | int local_group; |
2751 | int i; | |
908a7c1b | 2752 | int __group_imb = 0; |
783609c6 | 2753 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2754 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2755 | |
2756 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2757 | ||
783609c6 SS |
2758 | if (local_group) |
2759 | balance_cpu = first_cpu(group->cpumask); | |
2760 | ||
1da177e4 | 2761 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2762 | sum_weighted_load = sum_nr_running = avg_load = 0; |
908a7c1b KC |
2763 | max_cpu_load = 0; |
2764 | min_cpu_load = ~0UL; | |
1da177e4 LT |
2765 | |
2766 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2767 | struct rq *rq; |
2768 | ||
2769 | if (!cpu_isset(i, *cpus)) | |
2770 | continue; | |
2771 | ||
2772 | rq = cpu_rq(i); | |
2dd73a4f | 2773 | |
9439aab8 | 2774 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
2775 | *sd_idle = 0; |
2776 | ||
1da177e4 | 2777 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2778 | if (local_group) { |
2779 | if (idle_cpu(i) && !first_idle_cpu) { | |
2780 | first_idle_cpu = 1; | |
2781 | balance_cpu = i; | |
2782 | } | |
2783 | ||
a2000572 | 2784 | load = target_load(i, load_idx); |
908a7c1b | 2785 | } else { |
a2000572 | 2786 | load = source_load(i, load_idx); |
908a7c1b KC |
2787 | if (load > max_cpu_load) |
2788 | max_cpu_load = load; | |
2789 | if (min_cpu_load > load) | |
2790 | min_cpu_load = load; | |
2791 | } | |
1da177e4 LT |
2792 | |
2793 | avg_load += load; | |
2dd73a4f | 2794 | sum_nr_running += rq->nr_running; |
dd41f596 | 2795 | sum_weighted_load += weighted_cpuload(i); |
1da177e4 LT |
2796 | } |
2797 | ||
783609c6 SS |
2798 | /* |
2799 | * First idle cpu or the first cpu(busiest) in this sched group | |
2800 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
2801 | * domains. In the newly idle case, we will allow all the cpu's |
2802 | * to do the newly idle load balance. | |
783609c6 | 2803 | */ |
9439aab8 SS |
2804 | if (idle != CPU_NEWLY_IDLE && local_group && |
2805 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
2806 | *balance = 0; |
2807 | goto ret; | |
2808 | } | |
2809 | ||
1da177e4 | 2810 | total_load += avg_load; |
5517d86b | 2811 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2812 | |
2813 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2814 | avg_load = sg_div_cpu_power(group, |
2815 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2816 | |
908a7c1b KC |
2817 | if ((max_cpu_load - min_cpu_load) > SCHED_LOAD_SCALE) |
2818 | __group_imb = 1; | |
2819 | ||
5517d86b | 2820 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2821 | |
1da177e4 LT |
2822 | if (local_group) { |
2823 | this_load = avg_load; | |
2824 | this = group; | |
2dd73a4f PW |
2825 | this_nr_running = sum_nr_running; |
2826 | this_load_per_task = sum_weighted_load; | |
2827 | } else if (avg_load > max_load && | |
908a7c1b | 2828 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
2829 | max_load = avg_load; |
2830 | busiest = group; | |
2dd73a4f PW |
2831 | busiest_nr_running = sum_nr_running; |
2832 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 2833 | group_imb = __group_imb; |
1da177e4 | 2834 | } |
5c45bf27 SS |
2835 | |
2836 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2837 | /* | |
2838 | * Busy processors will not participate in power savings | |
2839 | * balance. | |
2840 | */ | |
dd41f596 IM |
2841 | if (idle == CPU_NOT_IDLE || |
2842 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
2843 | goto group_next; | |
5c45bf27 SS |
2844 | |
2845 | /* | |
2846 | * If the local group is idle or completely loaded | |
2847 | * no need to do power savings balance at this domain | |
2848 | */ | |
2849 | if (local_group && (this_nr_running >= group_capacity || | |
2850 | !this_nr_running)) | |
2851 | power_savings_balance = 0; | |
2852 | ||
dd41f596 | 2853 | /* |
5c45bf27 SS |
2854 | * If a group is already running at full capacity or idle, |
2855 | * don't include that group in power savings calculations | |
dd41f596 IM |
2856 | */ |
2857 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 2858 | || !sum_nr_running) |
dd41f596 | 2859 | goto group_next; |
5c45bf27 | 2860 | |
dd41f596 | 2861 | /* |
5c45bf27 | 2862 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
2863 | * This is the group from where we need to pick up the load |
2864 | * for saving power | |
2865 | */ | |
2866 | if ((sum_nr_running < min_nr_running) || | |
2867 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
2868 | first_cpu(group->cpumask) < |
2869 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
2870 | group_min = group; |
2871 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
2872 | min_load_per_task = sum_weighted_load / |
2873 | sum_nr_running; | |
dd41f596 | 2874 | } |
5c45bf27 | 2875 | |
dd41f596 | 2876 | /* |
5c45bf27 | 2877 | * Calculate the group which is almost near its |
dd41f596 IM |
2878 | * capacity but still has some space to pick up some load |
2879 | * from other group and save more power | |
2880 | */ | |
2881 | if (sum_nr_running <= group_capacity - 1) { | |
2882 | if (sum_nr_running > leader_nr_running || | |
2883 | (sum_nr_running == leader_nr_running && | |
2884 | first_cpu(group->cpumask) > | |
2885 | first_cpu(group_leader->cpumask))) { | |
2886 | group_leader = group; | |
2887 | leader_nr_running = sum_nr_running; | |
2888 | } | |
48f24c4d | 2889 | } |
5c45bf27 SS |
2890 | group_next: |
2891 | #endif | |
1da177e4 LT |
2892 | group = group->next; |
2893 | } while (group != sd->groups); | |
2894 | ||
2dd73a4f | 2895 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
2896 | goto out_balanced; |
2897 | ||
2898 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
2899 | ||
2900 | if (this_load >= avg_load || | |
2901 | 100*max_load <= sd->imbalance_pct*this_load) | |
2902 | goto out_balanced; | |
2903 | ||
2dd73a4f | 2904 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
2905 | if (group_imb) |
2906 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
2907 | ||
1da177e4 LT |
2908 | /* |
2909 | * We're trying to get all the cpus to the average_load, so we don't | |
2910 | * want to push ourselves above the average load, nor do we wish to | |
2911 | * reduce the max loaded cpu below the average load, as either of these | |
2912 | * actions would just result in more rebalancing later, and ping-pong | |
2913 | * tasks around. Thus we look for the minimum possible imbalance. | |
2914 | * Negative imbalances (*we* are more loaded than anyone else) will | |
2915 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 2916 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
2917 | * appear as very large values with unsigned longs. |
2918 | */ | |
2dd73a4f PW |
2919 | if (max_load <= busiest_load_per_task) |
2920 | goto out_balanced; | |
2921 | ||
2922 | /* | |
2923 | * In the presence of smp nice balancing, certain scenarios can have | |
2924 | * max load less than avg load(as we skip the groups at or below | |
2925 | * its cpu_power, while calculating max_load..) | |
2926 | */ | |
2927 | if (max_load < avg_load) { | |
2928 | *imbalance = 0; | |
2929 | goto small_imbalance; | |
2930 | } | |
0c117f1b SS |
2931 | |
2932 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 2933 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 2934 | |
1da177e4 | 2935 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
2936 | *imbalance = min(max_pull * busiest->__cpu_power, |
2937 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
2938 | / SCHED_LOAD_SCALE; |
2939 | ||
2dd73a4f PW |
2940 | /* |
2941 | * if *imbalance is less than the average load per runnable task | |
2942 | * there is no gaurantee that any tasks will be moved so we'll have | |
2943 | * a think about bumping its value to force at least one task to be | |
2944 | * moved | |
2945 | */ | |
7fd0d2dd | 2946 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 2947 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
2948 | unsigned int imbn; |
2949 | ||
2950 | small_imbalance: | |
2951 | pwr_move = pwr_now = 0; | |
2952 | imbn = 2; | |
2953 | if (this_nr_running) { | |
2954 | this_load_per_task /= this_nr_running; | |
2955 | if (busiest_load_per_task > this_load_per_task) | |
2956 | imbn = 1; | |
2957 | } else | |
2958 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 2959 | |
dd41f596 IM |
2960 | if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >= |
2961 | busiest_load_per_task * imbn) { | |
2dd73a4f | 2962 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2963 | return busiest; |
2964 | } | |
2965 | ||
2966 | /* | |
2967 | * OK, we don't have enough imbalance to justify moving tasks, | |
2968 | * however we may be able to increase total CPU power used by | |
2969 | * moving them. | |
2970 | */ | |
2971 | ||
5517d86b ED |
2972 | pwr_now += busiest->__cpu_power * |
2973 | min(busiest_load_per_task, max_load); | |
2974 | pwr_now += this->__cpu_power * | |
2975 | min(this_load_per_task, this_load); | |
1da177e4 LT |
2976 | pwr_now /= SCHED_LOAD_SCALE; |
2977 | ||
2978 | /* Amount of load we'd subtract */ | |
5517d86b ED |
2979 | tmp = sg_div_cpu_power(busiest, |
2980 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 2981 | if (max_load > tmp) |
5517d86b | 2982 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 2983 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
2984 | |
2985 | /* Amount of load we'd add */ | |
5517d86b | 2986 | if (max_load * busiest->__cpu_power < |
33859f7f | 2987 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
2988 | tmp = sg_div_cpu_power(this, |
2989 | max_load * busiest->__cpu_power); | |
1da177e4 | 2990 | else |
5517d86b ED |
2991 | tmp = sg_div_cpu_power(this, |
2992 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
2993 | pwr_move += this->__cpu_power * | |
2994 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
2995 | pwr_move /= SCHED_LOAD_SCALE; |
2996 | ||
2997 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
2998 | if (pwr_move > pwr_now) |
2999 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
3000 | } |
3001 | ||
1da177e4 LT |
3002 | return busiest; |
3003 | ||
3004 | out_balanced: | |
5c45bf27 | 3005 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 3006 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 3007 | goto ret; |
1da177e4 | 3008 | |
5c45bf27 SS |
3009 | if (this == group_leader && group_leader != group_min) { |
3010 | *imbalance = min_load_per_task; | |
3011 | return group_min; | |
3012 | } | |
5c45bf27 | 3013 | #endif |
783609c6 | 3014 | ret: |
1da177e4 LT |
3015 | *imbalance = 0; |
3016 | return NULL; | |
3017 | } | |
3018 | ||
3019 | /* | |
3020 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3021 | */ | |
70b97a7f | 3022 | static struct rq * |
d15bcfdb | 3023 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
0a2966b4 | 3024 | unsigned long imbalance, cpumask_t *cpus) |
1da177e4 | 3025 | { |
70b97a7f | 3026 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3027 | unsigned long max_load = 0; |
1da177e4 LT |
3028 | int i; |
3029 | ||
3030 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 3031 | unsigned long wl; |
0a2966b4 CL |
3032 | |
3033 | if (!cpu_isset(i, *cpus)) | |
3034 | continue; | |
3035 | ||
48f24c4d | 3036 | rq = cpu_rq(i); |
dd41f596 | 3037 | wl = weighted_cpuload(i); |
2dd73a4f | 3038 | |
dd41f596 | 3039 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3040 | continue; |
1da177e4 | 3041 | |
dd41f596 IM |
3042 | if (wl > max_load) { |
3043 | max_load = wl; | |
48f24c4d | 3044 | busiest = rq; |
1da177e4 LT |
3045 | } |
3046 | } | |
3047 | ||
3048 | return busiest; | |
3049 | } | |
3050 | ||
77391d71 NP |
3051 | /* |
3052 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3053 | * so long as it is large enough. | |
3054 | */ | |
3055 | #define MAX_PINNED_INTERVAL 512 | |
3056 | ||
1da177e4 LT |
3057 | /* |
3058 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3059 | * tasks if there is an imbalance. | |
1da177e4 | 3060 | */ |
70b97a7f | 3061 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3062 | struct sched_domain *sd, enum cpu_idle_type idle, |
783609c6 | 3063 | int *balance) |
1da177e4 | 3064 | { |
43010659 | 3065 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3066 | struct sched_group *group; |
1da177e4 | 3067 | unsigned long imbalance; |
70b97a7f | 3068 | struct rq *busiest; |
0a2966b4 | 3069 | cpumask_t cpus = CPU_MASK_ALL; |
fe2eea3f | 3070 | unsigned long flags; |
5969fe06 | 3071 | |
89c4710e SS |
3072 | /* |
3073 | * When power savings policy is enabled for the parent domain, idle | |
3074 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3075 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3076 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3077 | */ |
d15bcfdb | 3078 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3079 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3080 | sd_idle = 1; |
1da177e4 | 3081 | |
2d72376b | 3082 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3083 | |
0a2966b4 CL |
3084 | redo: |
3085 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
783609c6 SS |
3086 | &cpus, balance); |
3087 | ||
06066714 | 3088 | if (*balance == 0) |
783609c6 | 3089 | goto out_balanced; |
783609c6 | 3090 | |
1da177e4 LT |
3091 | if (!group) { |
3092 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3093 | goto out_balanced; | |
3094 | } | |
3095 | ||
0a2966b4 | 3096 | busiest = find_busiest_queue(group, idle, imbalance, &cpus); |
1da177e4 LT |
3097 | if (!busiest) { |
3098 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3099 | goto out_balanced; | |
3100 | } | |
3101 | ||
db935dbd | 3102 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3103 | |
3104 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3105 | ||
43010659 | 3106 | ld_moved = 0; |
1da177e4 LT |
3107 | if (busiest->nr_running > 1) { |
3108 | /* | |
3109 | * Attempt to move tasks. If find_busiest_group has found | |
3110 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3111 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3112 | * correctly treated as an imbalance. |
3113 | */ | |
fe2eea3f | 3114 | local_irq_save(flags); |
e17224bf | 3115 | double_rq_lock(this_rq, busiest); |
43010659 | 3116 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3117 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3118 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3119 | local_irq_restore(flags); |
81026794 | 3120 | |
46cb4b7c SS |
3121 | /* |
3122 | * some other cpu did the load balance for us. | |
3123 | */ | |
43010659 | 3124 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3125 | resched_cpu(this_cpu); |
3126 | ||
81026794 | 3127 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 CL |
3128 | if (unlikely(all_pinned)) { |
3129 | cpu_clear(cpu_of(busiest), cpus); | |
3130 | if (!cpus_empty(cpus)) | |
3131 | goto redo; | |
81026794 | 3132 | goto out_balanced; |
0a2966b4 | 3133 | } |
1da177e4 | 3134 | } |
81026794 | 3135 | |
43010659 | 3136 | if (!ld_moved) { |
1da177e4 LT |
3137 | schedstat_inc(sd, lb_failed[idle]); |
3138 | sd->nr_balance_failed++; | |
3139 | ||
3140 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3141 | |
fe2eea3f | 3142 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3143 | |
3144 | /* don't kick the migration_thread, if the curr | |
3145 | * task on busiest cpu can't be moved to this_cpu | |
3146 | */ | |
3147 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3148 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3149 | all_pinned = 1; |
3150 | goto out_one_pinned; | |
3151 | } | |
3152 | ||
1da177e4 LT |
3153 | if (!busiest->active_balance) { |
3154 | busiest->active_balance = 1; | |
3155 | busiest->push_cpu = this_cpu; | |
81026794 | 3156 | active_balance = 1; |
1da177e4 | 3157 | } |
fe2eea3f | 3158 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3159 | if (active_balance) |
1da177e4 LT |
3160 | wake_up_process(busiest->migration_thread); |
3161 | ||
3162 | /* | |
3163 | * We've kicked active balancing, reset the failure | |
3164 | * counter. | |
3165 | */ | |
39507451 | 3166 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3167 | } |
81026794 | 3168 | } else |
1da177e4 LT |
3169 | sd->nr_balance_failed = 0; |
3170 | ||
81026794 | 3171 | if (likely(!active_balance)) { |
1da177e4 LT |
3172 | /* We were unbalanced, so reset the balancing interval */ |
3173 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
3174 | } else { |
3175 | /* | |
3176 | * If we've begun active balancing, start to back off. This | |
3177 | * case may not be covered by the all_pinned logic if there | |
3178 | * is only 1 task on the busy runqueue (because we don't call | |
3179 | * move_tasks). | |
3180 | */ | |
3181 | if (sd->balance_interval < sd->max_interval) | |
3182 | sd->balance_interval *= 2; | |
1da177e4 LT |
3183 | } |
3184 | ||
43010659 | 3185 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3186 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3187 | return -1; |
43010659 | 3188 | return ld_moved; |
1da177e4 LT |
3189 | |
3190 | out_balanced: | |
1da177e4 LT |
3191 | schedstat_inc(sd, lb_balanced[idle]); |
3192 | ||
16cfb1c0 | 3193 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
3194 | |
3195 | out_one_pinned: | |
1da177e4 | 3196 | /* tune up the balancing interval */ |
77391d71 NP |
3197 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
3198 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
3199 | sd->balance_interval *= 2; |
3200 | ||
48f24c4d | 3201 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3202 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3203 | return -1; |
1da177e4 LT |
3204 | return 0; |
3205 | } | |
3206 | ||
3207 | /* | |
3208 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3209 | * tasks if there is an imbalance. | |
3210 | * | |
d15bcfdb | 3211 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
3212 | * this_rq is locked. |
3213 | */ | |
48f24c4d | 3214 | static int |
70b97a7f | 3215 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
3216 | { |
3217 | struct sched_group *group; | |
70b97a7f | 3218 | struct rq *busiest = NULL; |
1da177e4 | 3219 | unsigned long imbalance; |
43010659 | 3220 | int ld_moved = 0; |
5969fe06 | 3221 | int sd_idle = 0; |
969bb4e4 | 3222 | int all_pinned = 0; |
0a2966b4 | 3223 | cpumask_t cpus = CPU_MASK_ALL; |
5969fe06 | 3224 | |
89c4710e SS |
3225 | /* |
3226 | * When power savings policy is enabled for the parent domain, idle | |
3227 | * sibling can pick up load irrespective of busy siblings. In this case, | |
3228 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 3229 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
3230 | */ |
3231 | if (sd->flags & SD_SHARE_CPUPOWER && | |
3232 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3233 | sd_idle = 1; |
1da177e4 | 3234 | |
2d72376b | 3235 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 3236 | redo: |
d15bcfdb | 3237 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
783609c6 | 3238 | &sd_idle, &cpus, NULL); |
1da177e4 | 3239 | if (!group) { |
d15bcfdb | 3240 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3241 | goto out_balanced; |
1da177e4 LT |
3242 | } |
3243 | ||
d15bcfdb | 3244 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, |
0a2966b4 | 3245 | &cpus); |
db935dbd | 3246 | if (!busiest) { |
d15bcfdb | 3247 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3248 | goto out_balanced; |
1da177e4 LT |
3249 | } |
3250 | ||
db935dbd NP |
3251 | BUG_ON(busiest == this_rq); |
3252 | ||
d15bcfdb | 3253 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 3254 | |
43010659 | 3255 | ld_moved = 0; |
d6d5cfaf NP |
3256 | if (busiest->nr_running > 1) { |
3257 | /* Attempt to move tasks */ | |
3258 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
3259 | /* this_rq->clock is already updated */ |
3260 | update_rq_clock(busiest); | |
43010659 | 3261 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
3262 | imbalance, sd, CPU_NEWLY_IDLE, |
3263 | &all_pinned); | |
d6d5cfaf | 3264 | spin_unlock(&busiest->lock); |
0a2966b4 | 3265 | |
969bb4e4 | 3266 | if (unlikely(all_pinned)) { |
0a2966b4 CL |
3267 | cpu_clear(cpu_of(busiest), cpus); |
3268 | if (!cpus_empty(cpus)) | |
3269 | goto redo; | |
3270 | } | |
d6d5cfaf NP |
3271 | } |
3272 | ||
43010659 | 3273 | if (!ld_moved) { |
d15bcfdb | 3274 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
3275 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
3276 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
3277 | return -1; |
3278 | } else | |
16cfb1c0 | 3279 | sd->nr_balance_failed = 0; |
1da177e4 | 3280 | |
43010659 | 3281 | return ld_moved; |
16cfb1c0 NP |
3282 | |
3283 | out_balanced: | |
d15bcfdb | 3284 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 3285 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3286 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3287 | return -1; |
16cfb1c0 | 3288 | sd->nr_balance_failed = 0; |
48f24c4d | 3289 | |
16cfb1c0 | 3290 | return 0; |
1da177e4 LT |
3291 | } |
3292 | ||
3293 | /* | |
3294 | * idle_balance is called by schedule() if this_cpu is about to become | |
3295 | * idle. Attempts to pull tasks from other CPUs. | |
3296 | */ | |
70b97a7f | 3297 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
3298 | { |
3299 | struct sched_domain *sd; | |
dd41f596 IM |
3300 | int pulled_task = -1; |
3301 | unsigned long next_balance = jiffies + HZ; | |
1da177e4 LT |
3302 | |
3303 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
3304 | unsigned long interval; |
3305 | ||
3306 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
3307 | continue; | |
3308 | ||
3309 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 3310 | /* If we've pulled tasks over stop searching: */ |
1bd77f2d | 3311 | pulled_task = load_balance_newidle(this_cpu, |
92c4ca5c CL |
3312 | this_rq, sd); |
3313 | ||
3314 | interval = msecs_to_jiffies(sd->balance_interval); | |
3315 | if (time_after(next_balance, sd->last_balance + interval)) | |
3316 | next_balance = sd->last_balance + interval; | |
3317 | if (pulled_task) | |
3318 | break; | |
1da177e4 | 3319 | } |
dd41f596 | 3320 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
3321 | /* |
3322 | * We are going idle. next_balance may be set based on | |
3323 | * a busy processor. So reset next_balance. | |
3324 | */ | |
3325 | this_rq->next_balance = next_balance; | |
dd41f596 | 3326 | } |
1da177e4 LT |
3327 | } |
3328 | ||
3329 | /* | |
3330 | * active_load_balance is run by migration threads. It pushes running tasks | |
3331 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
3332 | * running on each physical CPU where possible, and avoids physical / | |
3333 | * logical imbalances. | |
3334 | * | |
3335 | * Called with busiest_rq locked. | |
3336 | */ | |
70b97a7f | 3337 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3338 | { |
39507451 | 3339 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3340 | struct sched_domain *sd; |
3341 | struct rq *target_rq; | |
39507451 | 3342 | |
48f24c4d | 3343 | /* Is there any task to move? */ |
39507451 | 3344 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3345 | return; |
3346 | ||
3347 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3348 | |
3349 | /* | |
39507451 | 3350 | * This condition is "impossible", if it occurs |
41a2d6cf | 3351 | * we need to fix it. Originally reported by |
39507451 | 3352 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3353 | */ |
39507451 | 3354 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3355 | |
39507451 NP |
3356 | /* move a task from busiest_rq to target_rq */ |
3357 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3358 | update_rq_clock(busiest_rq); |
3359 | update_rq_clock(target_rq); | |
39507451 NP |
3360 | |
3361 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3362 | for_each_domain(target_cpu, sd) { |
39507451 | 3363 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 3364 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 3365 | break; |
c96d145e | 3366 | } |
39507451 | 3367 | |
48f24c4d | 3368 | if (likely(sd)) { |
2d72376b | 3369 | schedstat_inc(sd, alb_count); |
39507451 | 3370 | |
43010659 PW |
3371 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3372 | sd, CPU_IDLE)) | |
48f24c4d IM |
3373 | schedstat_inc(sd, alb_pushed); |
3374 | else | |
3375 | schedstat_inc(sd, alb_failed); | |
3376 | } | |
39507451 | 3377 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
3378 | } |
3379 | ||
46cb4b7c SS |
3380 | #ifdef CONFIG_NO_HZ |
3381 | static struct { | |
3382 | atomic_t load_balancer; | |
41a2d6cf | 3383 | cpumask_t cpu_mask; |
46cb4b7c SS |
3384 | } nohz ____cacheline_aligned = { |
3385 | .load_balancer = ATOMIC_INIT(-1), | |
3386 | .cpu_mask = CPU_MASK_NONE, | |
3387 | }; | |
3388 | ||
7835b98b | 3389 | /* |
46cb4b7c SS |
3390 | * This routine will try to nominate the ilb (idle load balancing) |
3391 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3392 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3393 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3394 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3395 | * arrives... | |
3396 | * | |
3397 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3398 | * for idle load balancing. ilb owner will still be part of | |
3399 | * nohz.cpu_mask.. | |
7835b98b | 3400 | * |
46cb4b7c SS |
3401 | * While stopping the tick, this cpu will become the ilb owner if there |
3402 | * is no other owner. And will be the owner till that cpu becomes busy | |
3403 | * or if all cpus in the system stop their ticks at which point | |
3404 | * there is no need for ilb owner. | |
3405 | * | |
3406 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3407 | * next busy scheduler_tick() | |
3408 | */ | |
3409 | int select_nohz_load_balancer(int stop_tick) | |
3410 | { | |
3411 | int cpu = smp_processor_id(); | |
3412 | ||
3413 | if (stop_tick) { | |
3414 | cpu_set(cpu, nohz.cpu_mask); | |
3415 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3416 | ||
3417 | /* | |
3418 | * If we are going offline and still the leader, give up! | |
3419 | */ | |
3420 | if (cpu_is_offline(cpu) && | |
3421 | atomic_read(&nohz.load_balancer) == cpu) { | |
3422 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3423 | BUG(); | |
3424 | return 0; | |
3425 | } | |
3426 | ||
3427 | /* time for ilb owner also to sleep */ | |
3428 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3429 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3430 | atomic_set(&nohz.load_balancer, -1); | |
3431 | return 0; | |
3432 | } | |
3433 | ||
3434 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3435 | /* make me the ilb owner */ | |
3436 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3437 | return 1; | |
3438 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3439 | return 1; | |
3440 | } else { | |
3441 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3442 | return 0; | |
3443 | ||
3444 | cpu_clear(cpu, nohz.cpu_mask); | |
3445 | ||
3446 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3447 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3448 | BUG(); | |
3449 | } | |
3450 | return 0; | |
3451 | } | |
3452 | #endif | |
3453 | ||
3454 | static DEFINE_SPINLOCK(balancing); | |
3455 | ||
3456 | /* | |
7835b98b CL |
3457 | * It checks each scheduling domain to see if it is due to be balanced, |
3458 | * and initiates a balancing operation if so. | |
3459 | * | |
3460 | * Balancing parameters are set up in arch_init_sched_domains. | |
3461 | */ | |
a9957449 | 3462 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3463 | { |
46cb4b7c SS |
3464 | int balance = 1; |
3465 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3466 | unsigned long interval; |
3467 | struct sched_domain *sd; | |
46cb4b7c | 3468 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3469 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3470 | int update_next_balance = 0; |
1da177e4 | 3471 | |
46cb4b7c | 3472 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3473 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3474 | continue; | |
3475 | ||
3476 | interval = sd->balance_interval; | |
d15bcfdb | 3477 | if (idle != CPU_IDLE) |
1da177e4 LT |
3478 | interval *= sd->busy_factor; |
3479 | ||
3480 | /* scale ms to jiffies */ | |
3481 | interval = msecs_to_jiffies(interval); | |
3482 | if (unlikely(!interval)) | |
3483 | interval = 1; | |
dd41f596 IM |
3484 | if (interval > HZ*NR_CPUS/10) |
3485 | interval = HZ*NR_CPUS/10; | |
3486 | ||
1da177e4 | 3487 | |
08c183f3 CL |
3488 | if (sd->flags & SD_SERIALIZE) { |
3489 | if (!spin_trylock(&balancing)) | |
3490 | goto out; | |
3491 | } | |
3492 | ||
c9819f45 | 3493 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
46cb4b7c | 3494 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
3495 | /* |
3496 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3497 | * longer idle, or one of our SMT siblings is |
3498 | * not idle. | |
3499 | */ | |
d15bcfdb | 3500 | idle = CPU_NOT_IDLE; |
1da177e4 | 3501 | } |
1bd77f2d | 3502 | sd->last_balance = jiffies; |
1da177e4 | 3503 | } |
08c183f3 CL |
3504 | if (sd->flags & SD_SERIALIZE) |
3505 | spin_unlock(&balancing); | |
3506 | out: | |
f549da84 | 3507 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3508 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3509 | update_next_balance = 1; |
3510 | } | |
783609c6 SS |
3511 | |
3512 | /* | |
3513 | * Stop the load balance at this level. There is another | |
3514 | * CPU in our sched group which is doing load balancing more | |
3515 | * actively. | |
3516 | */ | |
3517 | if (!balance) | |
3518 | break; | |
1da177e4 | 3519 | } |
f549da84 SS |
3520 | |
3521 | /* | |
3522 | * next_balance will be updated only when there is a need. | |
3523 | * When the cpu is attached to null domain for ex, it will not be | |
3524 | * updated. | |
3525 | */ | |
3526 | if (likely(update_next_balance)) | |
3527 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3528 | } |
3529 | ||
3530 | /* | |
3531 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3532 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3533 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3534 | */ | |
3535 | static void run_rebalance_domains(struct softirq_action *h) | |
3536 | { | |
dd41f596 IM |
3537 | int this_cpu = smp_processor_id(); |
3538 | struct rq *this_rq = cpu_rq(this_cpu); | |
3539 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3540 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3541 | |
dd41f596 | 3542 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3543 | |
3544 | #ifdef CONFIG_NO_HZ | |
3545 | /* | |
3546 | * If this cpu is the owner for idle load balancing, then do the | |
3547 | * balancing on behalf of the other idle cpus whose ticks are | |
3548 | * stopped. | |
3549 | */ | |
dd41f596 IM |
3550 | if (this_rq->idle_at_tick && |
3551 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3552 | cpumask_t cpus = nohz.cpu_mask; |
3553 | struct rq *rq; | |
3554 | int balance_cpu; | |
3555 | ||
dd41f596 | 3556 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3557 | for_each_cpu_mask(balance_cpu, cpus) { |
3558 | /* | |
3559 | * If this cpu gets work to do, stop the load balancing | |
3560 | * work being done for other cpus. Next load | |
3561 | * balancing owner will pick it up. | |
3562 | */ | |
3563 | if (need_resched()) | |
3564 | break; | |
3565 | ||
de0cf899 | 3566 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3567 | |
3568 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3569 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3570 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3571 | } |
3572 | } | |
3573 | #endif | |
3574 | } | |
3575 | ||
3576 | /* | |
3577 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3578 | * | |
3579 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3580 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3581 | * if the whole system is idle. | |
3582 | */ | |
dd41f596 | 3583 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3584 | { |
46cb4b7c SS |
3585 | #ifdef CONFIG_NO_HZ |
3586 | /* | |
3587 | * If we were in the nohz mode recently and busy at the current | |
3588 | * scheduler tick, then check if we need to nominate new idle | |
3589 | * load balancer. | |
3590 | */ | |
3591 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3592 | rq->in_nohz_recently = 0; | |
3593 | ||
3594 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3595 | cpu_clear(cpu, nohz.cpu_mask); | |
3596 | atomic_set(&nohz.load_balancer, -1); | |
3597 | } | |
3598 | ||
3599 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3600 | /* | |
3601 | * simple selection for now: Nominate the | |
3602 | * first cpu in the nohz list to be the next | |
3603 | * ilb owner. | |
3604 | * | |
3605 | * TBD: Traverse the sched domains and nominate | |
3606 | * the nearest cpu in the nohz.cpu_mask. | |
3607 | */ | |
3608 | int ilb = first_cpu(nohz.cpu_mask); | |
3609 | ||
3610 | if (ilb != NR_CPUS) | |
3611 | resched_cpu(ilb); | |
3612 | } | |
3613 | } | |
3614 | ||
3615 | /* | |
3616 | * If this cpu is idle and doing idle load balancing for all the | |
3617 | * cpus with ticks stopped, is it time for that to stop? | |
3618 | */ | |
3619 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3620 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3621 | resched_cpu(cpu); | |
3622 | return; | |
3623 | } | |
3624 | ||
3625 | /* | |
3626 | * If this cpu is idle and the idle load balancing is done by | |
3627 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3628 | */ | |
3629 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3630 | cpu_isset(cpu, nohz.cpu_mask)) | |
3631 | return; | |
3632 | #endif | |
3633 | if (time_after_eq(jiffies, rq->next_balance)) | |
3634 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3635 | } |
dd41f596 IM |
3636 | |
3637 | #else /* CONFIG_SMP */ | |
3638 | ||
1da177e4 LT |
3639 | /* |
3640 | * on UP we do not need to balance between CPUs: | |
3641 | */ | |
70b97a7f | 3642 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3643 | { |
3644 | } | |
dd41f596 | 3645 | |
1da177e4 LT |
3646 | #endif |
3647 | ||
1da177e4 LT |
3648 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3649 | ||
3650 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3651 | ||
3652 | /* | |
41b86e9c IM |
3653 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
3654 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 3655 | */ |
41b86e9c | 3656 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 3657 | { |
1da177e4 | 3658 | unsigned long flags; |
41b86e9c IM |
3659 | u64 ns, delta_exec; |
3660 | struct rq *rq; | |
48f24c4d | 3661 | |
41b86e9c IM |
3662 | rq = task_rq_lock(p, &flags); |
3663 | ns = p->se.sum_exec_runtime; | |
051a1d1a | 3664 | if (task_current(rq, p)) { |
a8e504d2 IM |
3665 | update_rq_clock(rq); |
3666 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
3667 | if ((s64)delta_exec > 0) |
3668 | ns += delta_exec; | |
3669 | } | |
3670 | task_rq_unlock(rq, &flags); | |
48f24c4d | 3671 | |
1da177e4 LT |
3672 | return ns; |
3673 | } | |
3674 | ||
1da177e4 LT |
3675 | /* |
3676 | * Account user cpu time to a process. | |
3677 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
3678 | * @cputime: the cpu time spent in user space since the last update |
3679 | */ | |
3680 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3681 | { | |
3682 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3683 | cputime64_t tmp; | |
3684 | ||
3685 | p->utime = cputime_add(p->utime, cputime); | |
3686 | ||
3687 | /* Add user time to cpustat. */ | |
3688 | tmp = cputime_to_cputime64(cputime); | |
3689 | if (TASK_NICE(p) > 0) | |
3690 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3691 | else | |
3692 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3693 | } | |
3694 | ||
94886b84 LV |
3695 | /* |
3696 | * Account guest cpu time to a process. | |
3697 | * @p: the process that the cpu time gets accounted to | |
3698 | * @cputime: the cpu time spent in virtual machine since the last update | |
3699 | */ | |
f7402e03 | 3700 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
3701 | { |
3702 | cputime64_t tmp; | |
3703 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3704 | ||
3705 | tmp = cputime_to_cputime64(cputime); | |
3706 | ||
3707 | p->utime = cputime_add(p->utime, cputime); | |
3708 | p->gtime = cputime_add(p->gtime, cputime); | |
3709 | ||
3710 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3711 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3712 | } | |
3713 | ||
c66f08be MN |
3714 | /* |
3715 | * Account scaled user cpu time to a process. | |
3716 | * @p: the process that the cpu time gets accounted to | |
3717 | * @cputime: the cpu time spent in user space since the last update | |
3718 | */ | |
3719 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
3720 | { | |
3721 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
3722 | } | |
3723 | ||
1da177e4 LT |
3724 | /* |
3725 | * Account system cpu time to a process. | |
3726 | * @p: the process that the cpu time gets accounted to | |
3727 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3728 | * @cputime: the cpu time spent in kernel space since the last update | |
3729 | */ | |
3730 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3731 | cputime_t cputime) | |
3732 | { | |
3733 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3734 | struct rq *rq = this_rq(); |
1da177e4 LT |
3735 | cputime64_t tmp; |
3736 | ||
9778385d CB |
3737 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) |
3738 | return account_guest_time(p, cputime); | |
94886b84 | 3739 | |
1da177e4 LT |
3740 | p->stime = cputime_add(p->stime, cputime); |
3741 | ||
3742 | /* Add system time to cpustat. */ | |
3743 | tmp = cputime_to_cputime64(cputime); | |
3744 | if (hardirq_count() - hardirq_offset) | |
3745 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3746 | else if (softirq_count()) | |
3747 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 3748 | else if (p != rq->idle) |
1da177e4 | 3749 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 3750 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
3751 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
3752 | else | |
3753 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3754 | /* Account for system time used */ | |
3755 | acct_update_integrals(p); | |
1da177e4 LT |
3756 | } |
3757 | ||
c66f08be MN |
3758 | /* |
3759 | * Account scaled system cpu time to a process. | |
3760 | * @p: the process that the cpu time gets accounted to | |
3761 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3762 | * @cputime: the cpu time spent in kernel space since the last update | |
3763 | */ | |
3764 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
3765 | { | |
3766 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
3767 | } | |
3768 | ||
1da177e4 LT |
3769 | /* |
3770 | * Account for involuntary wait time. | |
3771 | * @p: the process from which the cpu time has been stolen | |
3772 | * @steal: the cpu time spent in involuntary wait | |
3773 | */ | |
3774 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3775 | { | |
3776 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3777 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3778 | struct rq *rq = this_rq(); |
1da177e4 LT |
3779 | |
3780 | if (p == rq->idle) { | |
3781 | p->stime = cputime_add(p->stime, steal); | |
3782 | if (atomic_read(&rq->nr_iowait) > 0) | |
3783 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3784 | else | |
3785 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 3786 | } else |
1da177e4 LT |
3787 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
3788 | } | |
3789 | ||
7835b98b CL |
3790 | /* |
3791 | * This function gets called by the timer code, with HZ frequency. | |
3792 | * We call it with interrupts disabled. | |
3793 | * | |
3794 | * It also gets called by the fork code, when changing the parent's | |
3795 | * timeslices. | |
3796 | */ | |
3797 | void scheduler_tick(void) | |
3798 | { | |
7835b98b CL |
3799 | int cpu = smp_processor_id(); |
3800 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3801 | struct task_struct *curr = rq->curr; |
529c7726 | 3802 | u64 next_tick = rq->tick_timestamp + TICK_NSEC; |
dd41f596 IM |
3803 | |
3804 | spin_lock(&rq->lock); | |
546fe3c9 | 3805 | __update_rq_clock(rq); |
529c7726 IM |
3806 | /* |
3807 | * Let rq->clock advance by at least TICK_NSEC: | |
3808 | */ | |
cc203d24 | 3809 | if (unlikely(rq->clock < next_tick)) { |
529c7726 | 3810 | rq->clock = next_tick; |
cc203d24 GC |
3811 | rq->clock_underflows++; |
3812 | } | |
529c7726 | 3813 | rq->tick_timestamp = rq->clock; |
f1a438d8 | 3814 | update_cpu_load(rq); |
fa85ae24 PZ |
3815 | curr->sched_class->task_tick(rq, curr, 0); |
3816 | update_sched_rt_period(rq); | |
dd41f596 | 3817 | spin_unlock(&rq->lock); |
7835b98b | 3818 | |
e418e1c2 | 3819 | #ifdef CONFIG_SMP |
dd41f596 IM |
3820 | rq->idle_at_tick = idle_cpu(cpu); |
3821 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3822 | #endif |
1da177e4 LT |
3823 | } |
3824 | ||
1da177e4 LT |
3825 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3826 | ||
43627582 | 3827 | void __kprobes add_preempt_count(int val) |
1da177e4 LT |
3828 | { |
3829 | /* | |
3830 | * Underflow? | |
3831 | */ | |
9a11b49a IM |
3832 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3833 | return; | |
1da177e4 LT |
3834 | preempt_count() += val; |
3835 | /* | |
3836 | * Spinlock count overflowing soon? | |
3837 | */ | |
33859f7f MOS |
3838 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3839 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3840 | } |
3841 | EXPORT_SYMBOL(add_preempt_count); | |
3842 | ||
43627582 | 3843 | void __kprobes sub_preempt_count(int val) |
1da177e4 LT |
3844 | { |
3845 | /* | |
3846 | * Underflow? | |
3847 | */ | |
9a11b49a IM |
3848 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3849 | return; | |
1da177e4 LT |
3850 | /* |
3851 | * Is the spinlock portion underflowing? | |
3852 | */ | |
9a11b49a IM |
3853 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3854 | !(preempt_count() & PREEMPT_MASK))) | |
3855 | return; | |
3856 | ||
1da177e4 LT |
3857 | preempt_count() -= val; |
3858 | } | |
3859 | EXPORT_SYMBOL(sub_preempt_count); | |
3860 | ||
3861 | #endif | |
3862 | ||
3863 | /* | |
dd41f596 | 3864 | * Print scheduling while atomic bug: |
1da177e4 | 3865 | */ |
dd41f596 | 3866 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3867 | { |
838225b4 SS |
3868 | struct pt_regs *regs = get_irq_regs(); |
3869 | ||
3870 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
3871 | prev->comm, prev->pid, preempt_count()); | |
3872 | ||
dd41f596 IM |
3873 | debug_show_held_locks(prev); |
3874 | if (irqs_disabled()) | |
3875 | print_irqtrace_events(prev); | |
838225b4 SS |
3876 | |
3877 | if (regs) | |
3878 | show_regs(regs); | |
3879 | else | |
3880 | dump_stack(); | |
dd41f596 | 3881 | } |
1da177e4 | 3882 | |
dd41f596 IM |
3883 | /* |
3884 | * Various schedule()-time debugging checks and statistics: | |
3885 | */ | |
3886 | static inline void schedule_debug(struct task_struct *prev) | |
3887 | { | |
1da177e4 | 3888 | /* |
41a2d6cf | 3889 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3890 | * schedule() atomically, we ignore that path for now. |
3891 | * Otherwise, whine if we are scheduling when we should not be. | |
3892 | */ | |
dd41f596 IM |
3893 | if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state)) |
3894 | __schedule_bug(prev); | |
3895 | ||
1da177e4 LT |
3896 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3897 | ||
2d72376b | 3898 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3899 | #ifdef CONFIG_SCHEDSTATS |
3900 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3901 | schedstat_inc(this_rq(), bkl_count); |
3902 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3903 | } |
3904 | #endif | |
dd41f596 IM |
3905 | } |
3906 | ||
3907 | /* | |
3908 | * Pick up the highest-prio task: | |
3909 | */ | |
3910 | static inline struct task_struct * | |
ff95f3df | 3911 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 3912 | { |
5522d5d5 | 3913 | const struct sched_class *class; |
dd41f596 | 3914 | struct task_struct *p; |
1da177e4 LT |
3915 | |
3916 | /* | |
dd41f596 IM |
3917 | * Optimization: we know that if all tasks are in |
3918 | * the fair class we can call that function directly: | |
1da177e4 | 3919 | */ |
dd41f596 | 3920 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3921 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3922 | if (likely(p)) |
3923 | return p; | |
1da177e4 LT |
3924 | } |
3925 | ||
dd41f596 IM |
3926 | class = sched_class_highest; |
3927 | for ( ; ; ) { | |
fb8d4724 | 3928 | p = class->pick_next_task(rq); |
dd41f596 IM |
3929 | if (p) |
3930 | return p; | |
3931 | /* | |
3932 | * Will never be NULL as the idle class always | |
3933 | * returns a non-NULL p: | |
3934 | */ | |
3935 | class = class->next; | |
3936 | } | |
3937 | } | |
1da177e4 | 3938 | |
dd41f596 IM |
3939 | /* |
3940 | * schedule() is the main scheduler function. | |
3941 | */ | |
3942 | asmlinkage void __sched schedule(void) | |
3943 | { | |
3944 | struct task_struct *prev, *next; | |
67ca7bde | 3945 | unsigned long *switch_count; |
dd41f596 | 3946 | struct rq *rq; |
dd41f596 IM |
3947 | int cpu; |
3948 | ||
3949 | need_resched: | |
3950 | preempt_disable(); | |
3951 | cpu = smp_processor_id(); | |
3952 | rq = cpu_rq(cpu); | |
3953 | rcu_qsctr_inc(cpu); | |
3954 | prev = rq->curr; | |
3955 | switch_count = &prev->nivcsw; | |
3956 | ||
3957 | release_kernel_lock(prev); | |
3958 | need_resched_nonpreemptible: | |
3959 | ||
3960 | schedule_debug(prev); | |
1da177e4 | 3961 | |
8f4d37ec PZ |
3962 | hrtick_clear(rq); |
3963 | ||
1e819950 IM |
3964 | /* |
3965 | * Do the rq-clock update outside the rq lock: | |
3966 | */ | |
3967 | local_irq_disable(); | |
c1b3da3e | 3968 | __update_rq_clock(rq); |
1e819950 IM |
3969 | spin_lock(&rq->lock); |
3970 | clear_tsk_need_resched(prev); | |
1da177e4 | 3971 | |
1da177e4 | 3972 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
1da177e4 | 3973 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && |
23e3c3cd | 3974 | signal_pending(prev))) { |
1da177e4 | 3975 | prev->state = TASK_RUNNING; |
dd41f596 | 3976 | } else { |
2e1cb74a | 3977 | deactivate_task(rq, prev, 1); |
1da177e4 | 3978 | } |
dd41f596 | 3979 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3980 | } |
3981 | ||
9a897c5a SR |
3982 | #ifdef CONFIG_SMP |
3983 | if (prev->sched_class->pre_schedule) | |
3984 | prev->sched_class->pre_schedule(rq, prev); | |
3985 | #endif | |
f65eda4f | 3986 | |
dd41f596 | 3987 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3988 | idle_balance(cpu, rq); |
1da177e4 | 3989 | |
31ee529c | 3990 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 3991 | next = pick_next_task(rq, prev); |
1da177e4 LT |
3992 | |
3993 | sched_info_switch(prev, next); | |
dd41f596 | 3994 | |
1da177e4 | 3995 | if (likely(prev != next)) { |
1da177e4 LT |
3996 | rq->nr_switches++; |
3997 | rq->curr = next; | |
3998 | ++*switch_count; | |
3999 | ||
dd41f596 | 4000 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
4001 | /* |
4002 | * the context switch might have flipped the stack from under | |
4003 | * us, hence refresh the local variables. | |
4004 | */ | |
4005 | cpu = smp_processor_id(); | |
4006 | rq = cpu_rq(cpu); | |
1da177e4 LT |
4007 | } else |
4008 | spin_unlock_irq(&rq->lock); | |
4009 | ||
8f4d37ec PZ |
4010 | hrtick_set(rq); |
4011 | ||
4012 | if (unlikely(reacquire_kernel_lock(current) < 0)) | |
1da177e4 | 4013 | goto need_resched_nonpreemptible; |
8f4d37ec | 4014 | |
1da177e4 LT |
4015 | preempt_enable_no_resched(); |
4016 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
4017 | goto need_resched; | |
4018 | } | |
1da177e4 LT |
4019 | EXPORT_SYMBOL(schedule); |
4020 | ||
4021 | #ifdef CONFIG_PREEMPT | |
4022 | /* | |
2ed6e34f | 4023 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4024 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4025 | * occur there and call schedule directly. |
4026 | */ | |
4027 | asmlinkage void __sched preempt_schedule(void) | |
4028 | { | |
4029 | struct thread_info *ti = current_thread_info(); | |
1da177e4 LT |
4030 | struct task_struct *task = current; |
4031 | int saved_lock_depth; | |
6478d880 | 4032 | |
1da177e4 LT |
4033 | /* |
4034 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4035 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4036 | */ |
beed33a8 | 4037 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4038 | return; |
4039 | ||
3a5c359a AK |
4040 | do { |
4041 | add_preempt_count(PREEMPT_ACTIVE); | |
4042 | ||
4043 | /* | |
4044 | * We keep the big kernel semaphore locked, but we | |
4045 | * clear ->lock_depth so that schedule() doesnt | |
4046 | * auto-release the semaphore: | |
4047 | */ | |
3a5c359a AK |
4048 | saved_lock_depth = task->lock_depth; |
4049 | task->lock_depth = -1; | |
3a5c359a | 4050 | schedule(); |
3a5c359a | 4051 | task->lock_depth = saved_lock_depth; |
3a5c359a | 4052 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4053 | |
3a5c359a AK |
4054 | /* |
4055 | * Check again in case we missed a preemption opportunity | |
4056 | * between schedule and now. | |
4057 | */ | |
4058 | barrier(); | |
4059 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 4060 | } |
1da177e4 LT |
4061 | EXPORT_SYMBOL(preempt_schedule); |
4062 | ||
4063 | /* | |
2ed6e34f | 4064 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4065 | * off of irq context. |
4066 | * Note, that this is called and return with irqs disabled. This will | |
4067 | * protect us against recursive calling from irq. | |
4068 | */ | |
4069 | asmlinkage void __sched preempt_schedule_irq(void) | |
4070 | { | |
4071 | struct thread_info *ti = current_thread_info(); | |
1da177e4 LT |
4072 | struct task_struct *task = current; |
4073 | int saved_lock_depth; | |
6478d880 | 4074 | |
2ed6e34f | 4075 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4076 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4077 | ||
3a5c359a AK |
4078 | do { |
4079 | add_preempt_count(PREEMPT_ACTIVE); | |
4080 | ||
4081 | /* | |
4082 | * We keep the big kernel semaphore locked, but we | |
4083 | * clear ->lock_depth so that schedule() doesnt | |
4084 | * auto-release the semaphore: | |
4085 | */ | |
3a5c359a AK |
4086 | saved_lock_depth = task->lock_depth; |
4087 | task->lock_depth = -1; | |
3a5c359a AK |
4088 | local_irq_enable(); |
4089 | schedule(); | |
4090 | local_irq_disable(); | |
3a5c359a | 4091 | task->lock_depth = saved_lock_depth; |
3a5c359a | 4092 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4093 | |
3a5c359a AK |
4094 | /* |
4095 | * Check again in case we missed a preemption opportunity | |
4096 | * between schedule and now. | |
4097 | */ | |
4098 | barrier(); | |
4099 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
4100 | } |
4101 | ||
4102 | #endif /* CONFIG_PREEMPT */ | |
4103 | ||
95cdf3b7 IM |
4104 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
4105 | void *key) | |
1da177e4 | 4106 | { |
48f24c4d | 4107 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 4108 | } |
1da177e4 LT |
4109 | EXPORT_SYMBOL(default_wake_function); |
4110 | ||
4111 | /* | |
41a2d6cf IM |
4112 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4113 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4114 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4115 | * | |
4116 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4117 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4118 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4119 | */ | |
4120 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
4121 | int nr_exclusive, int sync, void *key) | |
4122 | { | |
2e45874c | 4123 | wait_queue_t *curr, *next; |
1da177e4 | 4124 | |
2e45874c | 4125 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4126 | unsigned flags = curr->flags; |
4127 | ||
1da177e4 | 4128 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 4129 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4130 | break; |
4131 | } | |
4132 | } | |
4133 | ||
4134 | /** | |
4135 | * __wake_up - wake up threads blocked on a waitqueue. | |
4136 | * @q: the waitqueue | |
4137 | * @mode: which threads | |
4138 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4139 | * @key: is directly passed to the wakeup function |
1da177e4 | 4140 | */ |
7ad5b3a5 | 4141 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4142 | int nr_exclusive, void *key) |
1da177e4 LT |
4143 | { |
4144 | unsigned long flags; | |
4145 | ||
4146 | spin_lock_irqsave(&q->lock, flags); | |
4147 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4148 | spin_unlock_irqrestore(&q->lock, flags); | |
4149 | } | |
1da177e4 LT |
4150 | EXPORT_SYMBOL(__wake_up); |
4151 | ||
4152 | /* | |
4153 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4154 | */ | |
7ad5b3a5 | 4155 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4156 | { |
4157 | __wake_up_common(q, mode, 1, 0, NULL); | |
4158 | } | |
4159 | ||
4160 | /** | |
67be2dd1 | 4161 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4162 | * @q: the waitqueue |
4163 | * @mode: which threads | |
4164 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4165 | * | |
4166 | * The sync wakeup differs that the waker knows that it will schedule | |
4167 | * away soon, so while the target thread will be woken up, it will not | |
4168 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4169 | * with each other. This can prevent needless bouncing between CPUs. | |
4170 | * | |
4171 | * On UP it can prevent extra preemption. | |
4172 | */ | |
7ad5b3a5 | 4173 | void |
95cdf3b7 | 4174 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
1da177e4 LT |
4175 | { |
4176 | unsigned long flags; | |
4177 | int sync = 1; | |
4178 | ||
4179 | if (unlikely(!q)) | |
4180 | return; | |
4181 | ||
4182 | if (unlikely(!nr_exclusive)) | |
4183 | sync = 0; | |
4184 | ||
4185 | spin_lock_irqsave(&q->lock, flags); | |
4186 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
4187 | spin_unlock_irqrestore(&q->lock, flags); | |
4188 | } | |
4189 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
4190 | ||
b15136e9 | 4191 | void complete(struct completion *x) |
1da177e4 LT |
4192 | { |
4193 | unsigned long flags; | |
4194 | ||
4195 | spin_lock_irqsave(&x->wait.lock, flags); | |
4196 | x->done++; | |
d9514f6c | 4197 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4198 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4199 | } | |
4200 | EXPORT_SYMBOL(complete); | |
4201 | ||
b15136e9 | 4202 | void complete_all(struct completion *x) |
1da177e4 LT |
4203 | { |
4204 | unsigned long flags; | |
4205 | ||
4206 | spin_lock_irqsave(&x->wait.lock, flags); | |
4207 | x->done += UINT_MAX/2; | |
d9514f6c | 4208 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4209 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4210 | } | |
4211 | EXPORT_SYMBOL(complete_all); | |
4212 | ||
8cbbe86d AK |
4213 | static inline long __sched |
4214 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4215 | { |
1da177e4 LT |
4216 | if (!x->done) { |
4217 | DECLARE_WAITQUEUE(wait, current); | |
4218 | ||
4219 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
4220 | __add_wait_queue_tail(&x->wait, &wait); | |
4221 | do { | |
009e577e MW |
4222 | if ((state == TASK_INTERRUPTIBLE && |
4223 | signal_pending(current)) || | |
4224 | (state == TASK_KILLABLE && | |
4225 | fatal_signal_pending(current))) { | |
8cbbe86d AK |
4226 | __remove_wait_queue(&x->wait, &wait); |
4227 | return -ERESTARTSYS; | |
4228 | } | |
4229 | __set_current_state(state); | |
1da177e4 LT |
4230 | spin_unlock_irq(&x->wait.lock); |
4231 | timeout = schedule_timeout(timeout); | |
4232 | spin_lock_irq(&x->wait.lock); | |
4233 | if (!timeout) { | |
4234 | __remove_wait_queue(&x->wait, &wait); | |
8cbbe86d | 4235 | return timeout; |
1da177e4 LT |
4236 | } |
4237 | } while (!x->done); | |
4238 | __remove_wait_queue(&x->wait, &wait); | |
4239 | } | |
4240 | x->done--; | |
1da177e4 LT |
4241 | return timeout; |
4242 | } | |
1da177e4 | 4243 | |
8cbbe86d AK |
4244 | static long __sched |
4245 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4246 | { |
1da177e4 LT |
4247 | might_sleep(); |
4248 | ||
4249 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4250 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4251 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4252 | return timeout; |
4253 | } | |
1da177e4 | 4254 | |
b15136e9 | 4255 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4256 | { |
4257 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4258 | } |
8cbbe86d | 4259 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4260 | |
b15136e9 | 4261 | unsigned long __sched |
8cbbe86d | 4262 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4263 | { |
8cbbe86d | 4264 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4265 | } |
8cbbe86d | 4266 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4267 | |
8cbbe86d | 4268 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4269 | { |
51e97990 AK |
4270 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4271 | if (t == -ERESTARTSYS) | |
4272 | return t; | |
4273 | return 0; | |
0fec171c | 4274 | } |
8cbbe86d | 4275 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4276 | |
b15136e9 | 4277 | unsigned long __sched |
8cbbe86d AK |
4278 | wait_for_completion_interruptible_timeout(struct completion *x, |
4279 | unsigned long timeout) | |
0fec171c | 4280 | { |
8cbbe86d | 4281 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4282 | } |
8cbbe86d | 4283 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4284 | |
009e577e MW |
4285 | int __sched wait_for_completion_killable(struct completion *x) |
4286 | { | |
4287 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4288 | if (t == -ERESTARTSYS) | |
4289 | return t; | |
4290 | return 0; | |
4291 | } | |
4292 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4293 | ||
8cbbe86d AK |
4294 | static long __sched |
4295 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4296 | { |
0fec171c IM |
4297 | unsigned long flags; |
4298 | wait_queue_t wait; | |
4299 | ||
4300 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4301 | |
8cbbe86d | 4302 | __set_current_state(state); |
1da177e4 | 4303 | |
8cbbe86d AK |
4304 | spin_lock_irqsave(&q->lock, flags); |
4305 | __add_wait_queue(q, &wait); | |
4306 | spin_unlock(&q->lock); | |
4307 | timeout = schedule_timeout(timeout); | |
4308 | spin_lock_irq(&q->lock); | |
4309 | __remove_wait_queue(q, &wait); | |
4310 | spin_unlock_irqrestore(&q->lock, flags); | |
4311 | ||
4312 | return timeout; | |
4313 | } | |
4314 | ||
4315 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4316 | { | |
4317 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4318 | } |
1da177e4 LT |
4319 | EXPORT_SYMBOL(interruptible_sleep_on); |
4320 | ||
0fec171c | 4321 | long __sched |
95cdf3b7 | 4322 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4323 | { |
8cbbe86d | 4324 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4325 | } |
1da177e4 LT |
4326 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4327 | ||
0fec171c | 4328 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4329 | { |
8cbbe86d | 4330 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4331 | } |
1da177e4 LT |
4332 | EXPORT_SYMBOL(sleep_on); |
4333 | ||
0fec171c | 4334 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4335 | { |
8cbbe86d | 4336 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4337 | } |
1da177e4 LT |
4338 | EXPORT_SYMBOL(sleep_on_timeout); |
4339 | ||
b29739f9 IM |
4340 | #ifdef CONFIG_RT_MUTEXES |
4341 | ||
4342 | /* | |
4343 | * rt_mutex_setprio - set the current priority of a task | |
4344 | * @p: task | |
4345 | * @prio: prio value (kernel-internal form) | |
4346 | * | |
4347 | * This function changes the 'effective' priority of a task. It does | |
4348 | * not touch ->normal_prio like __setscheduler(). | |
4349 | * | |
4350 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4351 | */ | |
36c8b586 | 4352 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4353 | { |
4354 | unsigned long flags; | |
83b699ed | 4355 | int oldprio, on_rq, running; |
70b97a7f | 4356 | struct rq *rq; |
cb469845 | 4357 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
4358 | |
4359 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4360 | ||
4361 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4362 | update_rq_clock(rq); |
b29739f9 | 4363 | |
d5f9f942 | 4364 | oldprio = p->prio; |
dd41f596 | 4365 | on_rq = p->se.on_rq; |
051a1d1a | 4366 | running = task_current(rq, p); |
0e1f3483 | 4367 | if (on_rq) |
69be72c1 | 4368 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4369 | if (running) |
4370 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4371 | |
4372 | if (rt_prio(prio)) | |
4373 | p->sched_class = &rt_sched_class; | |
4374 | else | |
4375 | p->sched_class = &fair_sched_class; | |
4376 | ||
b29739f9 IM |
4377 | p->prio = prio; |
4378 | ||
0e1f3483 HS |
4379 | if (running) |
4380 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4381 | if (on_rq) { |
8159f87e | 4382 | enqueue_task(rq, p, 0); |
cb469845 SR |
4383 | |
4384 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4385 | } |
4386 | task_rq_unlock(rq, &flags); | |
4387 | } | |
4388 | ||
4389 | #endif | |
4390 | ||
36c8b586 | 4391 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4392 | { |
dd41f596 | 4393 | int old_prio, delta, on_rq; |
1da177e4 | 4394 | unsigned long flags; |
70b97a7f | 4395 | struct rq *rq; |
1da177e4 LT |
4396 | |
4397 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4398 | return; | |
4399 | /* | |
4400 | * We have to be careful, if called from sys_setpriority(), | |
4401 | * the task might be in the middle of scheduling on another CPU. | |
4402 | */ | |
4403 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4404 | update_rq_clock(rq); |
1da177e4 LT |
4405 | /* |
4406 | * The RT priorities are set via sched_setscheduler(), but we still | |
4407 | * allow the 'normal' nice value to be set - but as expected | |
4408 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4409 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4410 | */ |
e05606d3 | 4411 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4412 | p->static_prio = NICE_TO_PRIO(nice); |
4413 | goto out_unlock; | |
4414 | } | |
dd41f596 | 4415 | on_rq = p->se.on_rq; |
62fb1851 | 4416 | if (on_rq) { |
69be72c1 | 4417 | dequeue_task(rq, p, 0); |
62fb1851 PZ |
4418 | dec_load(rq, p); |
4419 | } | |
1da177e4 | 4420 | |
1da177e4 | 4421 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4422 | set_load_weight(p); |
b29739f9 IM |
4423 | old_prio = p->prio; |
4424 | p->prio = effective_prio(p); | |
4425 | delta = p->prio - old_prio; | |
1da177e4 | 4426 | |
dd41f596 | 4427 | if (on_rq) { |
8159f87e | 4428 | enqueue_task(rq, p, 0); |
62fb1851 | 4429 | inc_load(rq, p); |
1da177e4 | 4430 | /* |
d5f9f942 AM |
4431 | * If the task increased its priority or is running and |
4432 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4433 | */ |
d5f9f942 | 4434 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4435 | resched_task(rq->curr); |
4436 | } | |
4437 | out_unlock: | |
4438 | task_rq_unlock(rq, &flags); | |
4439 | } | |
1da177e4 LT |
4440 | EXPORT_SYMBOL(set_user_nice); |
4441 | ||
e43379f1 MM |
4442 | /* |
4443 | * can_nice - check if a task can reduce its nice value | |
4444 | * @p: task | |
4445 | * @nice: nice value | |
4446 | */ | |
36c8b586 | 4447 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4448 | { |
024f4747 MM |
4449 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4450 | int nice_rlim = 20 - nice; | |
48f24c4d | 4451 | |
e43379f1 MM |
4452 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4453 | capable(CAP_SYS_NICE)); | |
4454 | } | |
4455 | ||
1da177e4 LT |
4456 | #ifdef __ARCH_WANT_SYS_NICE |
4457 | ||
4458 | /* | |
4459 | * sys_nice - change the priority of the current process. | |
4460 | * @increment: priority increment | |
4461 | * | |
4462 | * sys_setpriority is a more generic, but much slower function that | |
4463 | * does similar things. | |
4464 | */ | |
4465 | asmlinkage long sys_nice(int increment) | |
4466 | { | |
48f24c4d | 4467 | long nice, retval; |
1da177e4 LT |
4468 | |
4469 | /* | |
4470 | * Setpriority might change our priority at the same moment. | |
4471 | * We don't have to worry. Conceptually one call occurs first | |
4472 | * and we have a single winner. | |
4473 | */ | |
e43379f1 MM |
4474 | if (increment < -40) |
4475 | increment = -40; | |
1da177e4 LT |
4476 | if (increment > 40) |
4477 | increment = 40; | |
4478 | ||
4479 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4480 | if (nice < -20) | |
4481 | nice = -20; | |
4482 | if (nice > 19) | |
4483 | nice = 19; | |
4484 | ||
e43379f1 MM |
4485 | if (increment < 0 && !can_nice(current, nice)) |
4486 | return -EPERM; | |
4487 | ||
1da177e4 LT |
4488 | retval = security_task_setnice(current, nice); |
4489 | if (retval) | |
4490 | return retval; | |
4491 | ||
4492 | set_user_nice(current, nice); | |
4493 | return 0; | |
4494 | } | |
4495 | ||
4496 | #endif | |
4497 | ||
4498 | /** | |
4499 | * task_prio - return the priority value of a given task. | |
4500 | * @p: the task in question. | |
4501 | * | |
4502 | * This is the priority value as seen by users in /proc. | |
4503 | * RT tasks are offset by -200. Normal tasks are centered | |
4504 | * around 0, value goes from -16 to +15. | |
4505 | */ | |
36c8b586 | 4506 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4507 | { |
4508 | return p->prio - MAX_RT_PRIO; | |
4509 | } | |
4510 | ||
4511 | /** | |
4512 | * task_nice - return the nice value of a given task. | |
4513 | * @p: the task in question. | |
4514 | */ | |
36c8b586 | 4515 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4516 | { |
4517 | return TASK_NICE(p); | |
4518 | } | |
150d8bed | 4519 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4520 | |
4521 | /** | |
4522 | * idle_cpu - is a given cpu idle currently? | |
4523 | * @cpu: the processor in question. | |
4524 | */ | |
4525 | int idle_cpu(int cpu) | |
4526 | { | |
4527 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4528 | } | |
4529 | ||
1da177e4 LT |
4530 | /** |
4531 | * idle_task - return the idle task for a given cpu. | |
4532 | * @cpu: the processor in question. | |
4533 | */ | |
36c8b586 | 4534 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4535 | { |
4536 | return cpu_rq(cpu)->idle; | |
4537 | } | |
4538 | ||
4539 | /** | |
4540 | * find_process_by_pid - find a process with a matching PID value. | |
4541 | * @pid: the pid in question. | |
4542 | */ | |
a9957449 | 4543 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4544 | { |
228ebcbe | 4545 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4546 | } |
4547 | ||
4548 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4549 | static void |
4550 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4551 | { |
dd41f596 | 4552 | BUG_ON(p->se.on_rq); |
48f24c4d | 4553 | |
1da177e4 | 4554 | p->policy = policy; |
dd41f596 IM |
4555 | switch (p->policy) { |
4556 | case SCHED_NORMAL: | |
4557 | case SCHED_BATCH: | |
4558 | case SCHED_IDLE: | |
4559 | p->sched_class = &fair_sched_class; | |
4560 | break; | |
4561 | case SCHED_FIFO: | |
4562 | case SCHED_RR: | |
4563 | p->sched_class = &rt_sched_class; | |
4564 | break; | |
4565 | } | |
4566 | ||
1da177e4 | 4567 | p->rt_priority = prio; |
b29739f9 IM |
4568 | p->normal_prio = normal_prio(p); |
4569 | /* we are holding p->pi_lock already */ | |
4570 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4571 | set_load_weight(p); |
1da177e4 LT |
4572 | } |
4573 | ||
4574 | /** | |
72fd4a35 | 4575 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4576 | * @p: the task in question. |
4577 | * @policy: new policy. | |
4578 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4579 | * |
72fd4a35 | 4580 | * NOTE that the task may be already dead. |
1da177e4 | 4581 | */ |
95cdf3b7 IM |
4582 | int sched_setscheduler(struct task_struct *p, int policy, |
4583 | struct sched_param *param) | |
1da177e4 | 4584 | { |
83b699ed | 4585 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4586 | unsigned long flags; |
cb469845 | 4587 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 4588 | struct rq *rq; |
1da177e4 | 4589 | |
66e5393a SR |
4590 | /* may grab non-irq protected spin_locks */ |
4591 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4592 | recheck: |
4593 | /* double check policy once rq lock held */ | |
4594 | if (policy < 0) | |
4595 | policy = oldpolicy = p->policy; | |
4596 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4597 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4598 | policy != SCHED_IDLE) | |
b0a9499c | 4599 | return -EINVAL; |
1da177e4 LT |
4600 | /* |
4601 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4602 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4603 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4604 | */ |
4605 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4606 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4607 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4608 | return -EINVAL; |
e05606d3 | 4609 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4610 | return -EINVAL; |
4611 | ||
37e4ab3f OC |
4612 | /* |
4613 | * Allow unprivileged RT tasks to decrease priority: | |
4614 | */ | |
4615 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4616 | if (rt_policy(policy)) { |
8dc3e909 | 4617 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4618 | |
4619 | if (!lock_task_sighand(p, &flags)) | |
4620 | return -ESRCH; | |
4621 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4622 | unlock_task_sighand(p, &flags); | |
4623 | ||
4624 | /* can't set/change the rt policy */ | |
4625 | if (policy != p->policy && !rlim_rtprio) | |
4626 | return -EPERM; | |
4627 | ||
4628 | /* can't increase priority */ | |
4629 | if (param->sched_priority > p->rt_priority && | |
4630 | param->sched_priority > rlim_rtprio) | |
4631 | return -EPERM; | |
4632 | } | |
dd41f596 IM |
4633 | /* |
4634 | * Like positive nice levels, dont allow tasks to | |
4635 | * move out of SCHED_IDLE either: | |
4636 | */ | |
4637 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4638 | return -EPERM; | |
5fe1d75f | 4639 | |
37e4ab3f OC |
4640 | /* can't change other user's priorities */ |
4641 | if ((current->euid != p->euid) && | |
4642 | (current->euid != p->uid)) | |
4643 | return -EPERM; | |
4644 | } | |
1da177e4 | 4645 | |
b68aa230 PZ |
4646 | #ifdef CONFIG_RT_GROUP_SCHED |
4647 | /* | |
4648 | * Do not allow realtime tasks into groups that have no runtime | |
4649 | * assigned. | |
4650 | */ | |
4651 | if (rt_policy(policy) && task_group(p)->rt_runtime == 0) | |
4652 | return -EPERM; | |
4653 | #endif | |
4654 | ||
1da177e4 LT |
4655 | retval = security_task_setscheduler(p, policy, param); |
4656 | if (retval) | |
4657 | return retval; | |
b29739f9 IM |
4658 | /* |
4659 | * make sure no PI-waiters arrive (or leave) while we are | |
4660 | * changing the priority of the task: | |
4661 | */ | |
4662 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4663 | /* |
4664 | * To be able to change p->policy safely, the apropriate | |
4665 | * runqueue lock must be held. | |
4666 | */ | |
b29739f9 | 4667 | rq = __task_rq_lock(p); |
1da177e4 LT |
4668 | /* recheck policy now with rq lock held */ |
4669 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4670 | policy = oldpolicy = -1; | |
b29739f9 IM |
4671 | __task_rq_unlock(rq); |
4672 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4673 | goto recheck; |
4674 | } | |
2daa3577 | 4675 | update_rq_clock(rq); |
dd41f596 | 4676 | on_rq = p->se.on_rq; |
051a1d1a | 4677 | running = task_current(rq, p); |
0e1f3483 | 4678 | if (on_rq) |
2e1cb74a | 4679 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
4680 | if (running) |
4681 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4682 | |
1da177e4 | 4683 | oldprio = p->prio; |
dd41f596 | 4684 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4685 | |
0e1f3483 HS |
4686 | if (running) |
4687 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
4688 | if (on_rq) { |
4689 | activate_task(rq, p, 0); | |
cb469845 SR |
4690 | |
4691 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 4692 | } |
b29739f9 IM |
4693 | __task_rq_unlock(rq); |
4694 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4695 | ||
95e02ca9 TG |
4696 | rt_mutex_adjust_pi(p); |
4697 | ||
1da177e4 LT |
4698 | return 0; |
4699 | } | |
4700 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4701 | ||
95cdf3b7 IM |
4702 | static int |
4703 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4704 | { |
1da177e4 LT |
4705 | struct sched_param lparam; |
4706 | struct task_struct *p; | |
36c8b586 | 4707 | int retval; |
1da177e4 LT |
4708 | |
4709 | if (!param || pid < 0) | |
4710 | return -EINVAL; | |
4711 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4712 | return -EFAULT; | |
5fe1d75f ON |
4713 | |
4714 | rcu_read_lock(); | |
4715 | retval = -ESRCH; | |
1da177e4 | 4716 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4717 | if (p != NULL) |
4718 | retval = sched_setscheduler(p, policy, &lparam); | |
4719 | rcu_read_unlock(); | |
36c8b586 | 4720 | |
1da177e4 LT |
4721 | return retval; |
4722 | } | |
4723 | ||
4724 | /** | |
4725 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4726 | * @pid: the pid in question. | |
4727 | * @policy: new policy. | |
4728 | * @param: structure containing the new RT priority. | |
4729 | */ | |
41a2d6cf IM |
4730 | asmlinkage long |
4731 | sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4732 | { |
c21761f1 JB |
4733 | /* negative values for policy are not valid */ |
4734 | if (policy < 0) | |
4735 | return -EINVAL; | |
4736 | ||
1da177e4 LT |
4737 | return do_sched_setscheduler(pid, policy, param); |
4738 | } | |
4739 | ||
4740 | /** | |
4741 | * sys_sched_setparam - set/change the RT priority of a thread | |
4742 | * @pid: the pid in question. | |
4743 | * @param: structure containing the new RT priority. | |
4744 | */ | |
4745 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4746 | { | |
4747 | return do_sched_setscheduler(pid, -1, param); | |
4748 | } | |
4749 | ||
4750 | /** | |
4751 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4752 | * @pid: the pid in question. | |
4753 | */ | |
4754 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4755 | { | |
36c8b586 | 4756 | struct task_struct *p; |
3a5c359a | 4757 | int retval; |
1da177e4 LT |
4758 | |
4759 | if (pid < 0) | |
3a5c359a | 4760 | return -EINVAL; |
1da177e4 LT |
4761 | |
4762 | retval = -ESRCH; | |
4763 | read_lock(&tasklist_lock); | |
4764 | p = find_process_by_pid(pid); | |
4765 | if (p) { | |
4766 | retval = security_task_getscheduler(p); | |
4767 | if (!retval) | |
4768 | retval = p->policy; | |
4769 | } | |
4770 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
4771 | return retval; |
4772 | } | |
4773 | ||
4774 | /** | |
4775 | * sys_sched_getscheduler - get the RT priority of a thread | |
4776 | * @pid: the pid in question. | |
4777 | * @param: structure containing the RT priority. | |
4778 | */ | |
4779 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
4780 | { | |
4781 | struct sched_param lp; | |
36c8b586 | 4782 | struct task_struct *p; |
3a5c359a | 4783 | int retval; |
1da177e4 LT |
4784 | |
4785 | if (!param || pid < 0) | |
3a5c359a | 4786 | return -EINVAL; |
1da177e4 LT |
4787 | |
4788 | read_lock(&tasklist_lock); | |
4789 | p = find_process_by_pid(pid); | |
4790 | retval = -ESRCH; | |
4791 | if (!p) | |
4792 | goto out_unlock; | |
4793 | ||
4794 | retval = security_task_getscheduler(p); | |
4795 | if (retval) | |
4796 | goto out_unlock; | |
4797 | ||
4798 | lp.sched_priority = p->rt_priority; | |
4799 | read_unlock(&tasklist_lock); | |
4800 | ||
4801 | /* | |
4802 | * This one might sleep, we cannot do it with a spinlock held ... | |
4803 | */ | |
4804 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4805 | ||
1da177e4 LT |
4806 | return retval; |
4807 | ||
4808 | out_unlock: | |
4809 | read_unlock(&tasklist_lock); | |
4810 | return retval; | |
4811 | } | |
4812 | ||
4813 | long sched_setaffinity(pid_t pid, cpumask_t new_mask) | |
4814 | { | |
1da177e4 | 4815 | cpumask_t cpus_allowed; |
36c8b586 IM |
4816 | struct task_struct *p; |
4817 | int retval; | |
1da177e4 | 4818 | |
95402b38 | 4819 | get_online_cpus(); |
1da177e4 LT |
4820 | read_lock(&tasklist_lock); |
4821 | ||
4822 | p = find_process_by_pid(pid); | |
4823 | if (!p) { | |
4824 | read_unlock(&tasklist_lock); | |
95402b38 | 4825 | put_online_cpus(); |
1da177e4 LT |
4826 | return -ESRCH; |
4827 | } | |
4828 | ||
4829 | /* | |
4830 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 4831 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
4832 | * usage count and then drop tasklist_lock. |
4833 | */ | |
4834 | get_task_struct(p); | |
4835 | read_unlock(&tasklist_lock); | |
4836 | ||
4837 | retval = -EPERM; | |
4838 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
4839 | !capable(CAP_SYS_NICE)) | |
4840 | goto out_unlock; | |
4841 | ||
e7834f8f DQ |
4842 | retval = security_task_setscheduler(p, 0, NULL); |
4843 | if (retval) | |
4844 | goto out_unlock; | |
4845 | ||
1da177e4 LT |
4846 | cpus_allowed = cpuset_cpus_allowed(p); |
4847 | cpus_and(new_mask, new_mask, cpus_allowed); | |
8707d8b8 | 4848 | again: |
1da177e4 LT |
4849 | retval = set_cpus_allowed(p, new_mask); |
4850 | ||
8707d8b8 PM |
4851 | if (!retval) { |
4852 | cpus_allowed = cpuset_cpus_allowed(p); | |
4853 | if (!cpus_subset(new_mask, cpus_allowed)) { | |
4854 | /* | |
4855 | * We must have raced with a concurrent cpuset | |
4856 | * update. Just reset the cpus_allowed to the | |
4857 | * cpuset's cpus_allowed | |
4858 | */ | |
4859 | new_mask = cpus_allowed; | |
4860 | goto again; | |
4861 | } | |
4862 | } | |
1da177e4 LT |
4863 | out_unlock: |
4864 | put_task_struct(p); | |
95402b38 | 4865 | put_online_cpus(); |
1da177e4 LT |
4866 | return retval; |
4867 | } | |
4868 | ||
4869 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
4870 | cpumask_t *new_mask) | |
4871 | { | |
4872 | if (len < sizeof(cpumask_t)) { | |
4873 | memset(new_mask, 0, sizeof(cpumask_t)); | |
4874 | } else if (len > sizeof(cpumask_t)) { | |
4875 | len = sizeof(cpumask_t); | |
4876 | } | |
4877 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
4878 | } | |
4879 | ||
4880 | /** | |
4881 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4882 | * @pid: pid of the process | |
4883 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4884 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4885 | */ | |
4886 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
4887 | unsigned long __user *user_mask_ptr) | |
4888 | { | |
4889 | cpumask_t new_mask; | |
4890 | int retval; | |
4891 | ||
4892 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
4893 | if (retval) | |
4894 | return retval; | |
4895 | ||
4896 | return sched_setaffinity(pid, new_mask); | |
4897 | } | |
4898 | ||
4899 | /* | |
4900 | * Represents all cpu's present in the system | |
4901 | * In systems capable of hotplug, this map could dynamically grow | |
4902 | * as new cpu's are detected in the system via any platform specific | |
4903 | * method, such as ACPI for e.g. | |
4904 | */ | |
4905 | ||
4cef0c61 | 4906 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
4907 | EXPORT_SYMBOL(cpu_present_map); |
4908 | ||
4909 | #ifndef CONFIG_SMP | |
4cef0c61 | 4910 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
4911 | EXPORT_SYMBOL(cpu_online_map); |
4912 | ||
4cef0c61 | 4913 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 4914 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
4915 | #endif |
4916 | ||
4917 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
4918 | { | |
36c8b586 | 4919 | struct task_struct *p; |
1da177e4 | 4920 | int retval; |
1da177e4 | 4921 | |
95402b38 | 4922 | get_online_cpus(); |
1da177e4 LT |
4923 | read_lock(&tasklist_lock); |
4924 | ||
4925 | retval = -ESRCH; | |
4926 | p = find_process_by_pid(pid); | |
4927 | if (!p) | |
4928 | goto out_unlock; | |
4929 | ||
e7834f8f DQ |
4930 | retval = security_task_getscheduler(p); |
4931 | if (retval) | |
4932 | goto out_unlock; | |
4933 | ||
2f7016d9 | 4934 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
4935 | |
4936 | out_unlock: | |
4937 | read_unlock(&tasklist_lock); | |
95402b38 | 4938 | put_online_cpus(); |
1da177e4 | 4939 | |
9531b62f | 4940 | return retval; |
1da177e4 LT |
4941 | } |
4942 | ||
4943 | /** | |
4944 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4945 | * @pid: pid of the process | |
4946 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4947 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4948 | */ | |
4949 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
4950 | unsigned long __user *user_mask_ptr) | |
4951 | { | |
4952 | int ret; | |
4953 | cpumask_t mask; | |
4954 | ||
4955 | if (len < sizeof(cpumask_t)) | |
4956 | return -EINVAL; | |
4957 | ||
4958 | ret = sched_getaffinity(pid, &mask); | |
4959 | if (ret < 0) | |
4960 | return ret; | |
4961 | ||
4962 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
4963 | return -EFAULT; | |
4964 | ||
4965 | return sizeof(cpumask_t); | |
4966 | } | |
4967 | ||
4968 | /** | |
4969 | * sys_sched_yield - yield the current processor to other threads. | |
4970 | * | |
dd41f596 IM |
4971 | * This function yields the current CPU to other tasks. If there are no |
4972 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
4973 | */ |
4974 | asmlinkage long sys_sched_yield(void) | |
4975 | { | |
70b97a7f | 4976 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4977 | |
2d72376b | 4978 | schedstat_inc(rq, yld_count); |
4530d7ab | 4979 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4980 | |
4981 | /* | |
4982 | * Since we are going to call schedule() anyway, there's | |
4983 | * no need to preempt or enable interrupts: | |
4984 | */ | |
4985 | __release(rq->lock); | |
8a25d5de | 4986 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4987 | _raw_spin_unlock(&rq->lock); |
4988 | preempt_enable_no_resched(); | |
4989 | ||
4990 | schedule(); | |
4991 | ||
4992 | return 0; | |
4993 | } | |
4994 | ||
e7b38404 | 4995 | static void __cond_resched(void) |
1da177e4 | 4996 | { |
8e0a43d8 IM |
4997 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
4998 | __might_sleep(__FILE__, __LINE__); | |
4999 | #endif | |
5bbcfd90 IM |
5000 | /* |
5001 | * The BKS might be reacquired before we have dropped | |
5002 | * PREEMPT_ACTIVE, which could trigger a second | |
5003 | * cond_resched() call. | |
5004 | */ | |
1da177e4 LT |
5005 | do { |
5006 | add_preempt_count(PREEMPT_ACTIVE); | |
5007 | schedule(); | |
5008 | sub_preempt_count(PREEMPT_ACTIVE); | |
5009 | } while (need_resched()); | |
5010 | } | |
5011 | ||
02b67cc3 HX |
5012 | #if !defined(CONFIG_PREEMPT) || defined(CONFIG_PREEMPT_VOLUNTARY) |
5013 | int __sched _cond_resched(void) | |
1da177e4 | 5014 | { |
9414232f IM |
5015 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
5016 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
5017 | __cond_resched(); |
5018 | return 1; | |
5019 | } | |
5020 | return 0; | |
5021 | } | |
02b67cc3 HX |
5022 | EXPORT_SYMBOL(_cond_resched); |
5023 | #endif | |
1da177e4 LT |
5024 | |
5025 | /* | |
5026 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
5027 | * call schedule, and on return reacquire the lock. | |
5028 | * | |
41a2d6cf | 5029 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5030 | * operations here to prevent schedule() from being called twice (once via |
5031 | * spin_unlock(), once by hand). | |
5032 | */ | |
95cdf3b7 | 5033 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5034 | { |
95c354fe | 5035 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
5036 | int ret = 0; |
5037 | ||
95c354fe | 5038 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5039 | spin_unlock(lock); |
95c354fe NP |
5040 | if (resched && need_resched()) |
5041 | __cond_resched(); | |
5042 | else | |
5043 | cpu_relax(); | |
6df3cecb | 5044 | ret = 1; |
1da177e4 | 5045 | spin_lock(lock); |
1da177e4 | 5046 | } |
6df3cecb | 5047 | return ret; |
1da177e4 | 5048 | } |
1da177e4 LT |
5049 | EXPORT_SYMBOL(cond_resched_lock); |
5050 | ||
5051 | int __sched cond_resched_softirq(void) | |
5052 | { | |
5053 | BUG_ON(!in_softirq()); | |
5054 | ||
9414232f | 5055 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 5056 | local_bh_enable(); |
1da177e4 LT |
5057 | __cond_resched(); |
5058 | local_bh_disable(); | |
5059 | return 1; | |
5060 | } | |
5061 | return 0; | |
5062 | } | |
1da177e4 LT |
5063 | EXPORT_SYMBOL(cond_resched_softirq); |
5064 | ||
1da177e4 LT |
5065 | /** |
5066 | * yield - yield the current processor to other threads. | |
5067 | * | |
72fd4a35 | 5068 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5069 | * thread runnable and calls sys_sched_yield(). |
5070 | */ | |
5071 | void __sched yield(void) | |
5072 | { | |
5073 | set_current_state(TASK_RUNNING); | |
5074 | sys_sched_yield(); | |
5075 | } | |
1da177e4 LT |
5076 | EXPORT_SYMBOL(yield); |
5077 | ||
5078 | /* | |
41a2d6cf | 5079 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
5080 | * that process accounting knows that this is a task in IO wait state. |
5081 | * | |
5082 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
5083 | * has set its backing_dev_info: the queue against which it should throttle) | |
5084 | */ | |
5085 | void __sched io_schedule(void) | |
5086 | { | |
70b97a7f | 5087 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 5088 | |
0ff92245 | 5089 | delayacct_blkio_start(); |
1da177e4 LT |
5090 | atomic_inc(&rq->nr_iowait); |
5091 | schedule(); | |
5092 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5093 | delayacct_blkio_end(); |
1da177e4 | 5094 | } |
1da177e4 LT |
5095 | EXPORT_SYMBOL(io_schedule); |
5096 | ||
5097 | long __sched io_schedule_timeout(long timeout) | |
5098 | { | |
70b97a7f | 5099 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
5100 | long ret; |
5101 | ||
0ff92245 | 5102 | delayacct_blkio_start(); |
1da177e4 LT |
5103 | atomic_inc(&rq->nr_iowait); |
5104 | ret = schedule_timeout(timeout); | |
5105 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5106 | delayacct_blkio_end(); |
1da177e4 LT |
5107 | return ret; |
5108 | } | |
5109 | ||
5110 | /** | |
5111 | * sys_sched_get_priority_max - return maximum RT priority. | |
5112 | * @policy: scheduling class. | |
5113 | * | |
5114 | * this syscall returns the maximum rt_priority that can be used | |
5115 | * by a given scheduling class. | |
5116 | */ | |
5117 | asmlinkage long sys_sched_get_priority_max(int policy) | |
5118 | { | |
5119 | int ret = -EINVAL; | |
5120 | ||
5121 | switch (policy) { | |
5122 | case SCHED_FIFO: | |
5123 | case SCHED_RR: | |
5124 | ret = MAX_USER_RT_PRIO-1; | |
5125 | break; | |
5126 | case SCHED_NORMAL: | |
b0a9499c | 5127 | case SCHED_BATCH: |
dd41f596 | 5128 | case SCHED_IDLE: |
1da177e4 LT |
5129 | ret = 0; |
5130 | break; | |
5131 | } | |
5132 | return ret; | |
5133 | } | |
5134 | ||
5135 | /** | |
5136 | * sys_sched_get_priority_min - return minimum RT priority. | |
5137 | * @policy: scheduling class. | |
5138 | * | |
5139 | * this syscall returns the minimum rt_priority that can be used | |
5140 | * by a given scheduling class. | |
5141 | */ | |
5142 | asmlinkage long sys_sched_get_priority_min(int policy) | |
5143 | { | |
5144 | int ret = -EINVAL; | |
5145 | ||
5146 | switch (policy) { | |
5147 | case SCHED_FIFO: | |
5148 | case SCHED_RR: | |
5149 | ret = 1; | |
5150 | break; | |
5151 | case SCHED_NORMAL: | |
b0a9499c | 5152 | case SCHED_BATCH: |
dd41f596 | 5153 | case SCHED_IDLE: |
1da177e4 LT |
5154 | ret = 0; |
5155 | } | |
5156 | return ret; | |
5157 | } | |
5158 | ||
5159 | /** | |
5160 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5161 | * @pid: pid of the process. | |
5162 | * @interval: userspace pointer to the timeslice value. | |
5163 | * | |
5164 | * this syscall writes the default timeslice value of a given process | |
5165 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5166 | */ | |
5167 | asmlinkage | |
5168 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
5169 | { | |
36c8b586 | 5170 | struct task_struct *p; |
a4ec24b4 | 5171 | unsigned int time_slice; |
3a5c359a | 5172 | int retval; |
1da177e4 | 5173 | struct timespec t; |
1da177e4 LT |
5174 | |
5175 | if (pid < 0) | |
3a5c359a | 5176 | return -EINVAL; |
1da177e4 LT |
5177 | |
5178 | retval = -ESRCH; | |
5179 | read_lock(&tasklist_lock); | |
5180 | p = find_process_by_pid(pid); | |
5181 | if (!p) | |
5182 | goto out_unlock; | |
5183 | ||
5184 | retval = security_task_getscheduler(p); | |
5185 | if (retval) | |
5186 | goto out_unlock; | |
5187 | ||
77034937 IM |
5188 | /* |
5189 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
5190 | * tasks that are on an otherwise idle runqueue: | |
5191 | */ | |
5192 | time_slice = 0; | |
5193 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 5194 | time_slice = DEF_TIMESLICE; |
1868f958 | 5195 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
5196 | struct sched_entity *se = &p->se; |
5197 | unsigned long flags; | |
5198 | struct rq *rq; | |
5199 | ||
5200 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
5201 | if (rq->cfs.load.weight) |
5202 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
5203 | task_rq_unlock(rq, &flags); |
5204 | } | |
1da177e4 | 5205 | read_unlock(&tasklist_lock); |
a4ec24b4 | 5206 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5207 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5208 | return retval; |
3a5c359a | 5209 | |
1da177e4 LT |
5210 | out_unlock: |
5211 | read_unlock(&tasklist_lock); | |
5212 | return retval; | |
5213 | } | |
5214 | ||
2ed6e34f | 5215 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 | 5216 | |
82a1fcb9 | 5217 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5218 | { |
1da177e4 | 5219 | unsigned long free = 0; |
36c8b586 | 5220 | unsigned state; |
1da177e4 | 5221 | |
1da177e4 | 5222 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 5223 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5224 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5225 | #if BITS_PER_LONG == 32 |
1da177e4 | 5226 | if (state == TASK_RUNNING) |
cc4ea795 | 5227 | printk(KERN_CONT " running "); |
1da177e4 | 5228 | else |
cc4ea795 | 5229 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5230 | #else |
5231 | if (state == TASK_RUNNING) | |
cc4ea795 | 5232 | printk(KERN_CONT " running task "); |
1da177e4 | 5233 | else |
cc4ea795 | 5234 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5235 | #endif |
5236 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
5237 | { | |
10ebffde | 5238 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
5239 | while (!*n) |
5240 | n++; | |
10ebffde | 5241 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
5242 | } |
5243 | #endif | |
ba25f9dc | 5244 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 5245 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 5246 | |
5fb5e6de | 5247 | show_stack(p, NULL); |
1da177e4 LT |
5248 | } |
5249 | ||
e59e2ae2 | 5250 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5251 | { |
36c8b586 | 5252 | struct task_struct *g, *p; |
1da177e4 | 5253 | |
4bd77321 IM |
5254 | #if BITS_PER_LONG == 32 |
5255 | printk(KERN_INFO | |
5256 | " task PC stack pid father\n"); | |
1da177e4 | 5257 | #else |
4bd77321 IM |
5258 | printk(KERN_INFO |
5259 | " task PC stack pid father\n"); | |
1da177e4 LT |
5260 | #endif |
5261 | read_lock(&tasklist_lock); | |
5262 | do_each_thread(g, p) { | |
5263 | /* | |
5264 | * reset the NMI-timeout, listing all files on a slow | |
5265 | * console might take alot of time: | |
5266 | */ | |
5267 | touch_nmi_watchdog(); | |
39bc89fd | 5268 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5269 | sched_show_task(p); |
1da177e4 LT |
5270 | } while_each_thread(g, p); |
5271 | ||
04c9167f JF |
5272 | touch_all_softlockup_watchdogs(); |
5273 | ||
dd41f596 IM |
5274 | #ifdef CONFIG_SCHED_DEBUG |
5275 | sysrq_sched_debug_show(); | |
5276 | #endif | |
1da177e4 | 5277 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5278 | /* |
5279 | * Only show locks if all tasks are dumped: | |
5280 | */ | |
5281 | if (state_filter == -1) | |
5282 | debug_show_all_locks(); | |
1da177e4 LT |
5283 | } |
5284 | ||
1df21055 IM |
5285 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5286 | { | |
dd41f596 | 5287 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5288 | } |
5289 | ||
f340c0d1 IM |
5290 | /** |
5291 | * init_idle - set up an idle thread for a given CPU | |
5292 | * @idle: task in question | |
5293 | * @cpu: cpu the idle task belongs to | |
5294 | * | |
5295 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5296 | * flag, to make booting more robust. | |
5297 | */ | |
5c1e1767 | 5298 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5299 | { |
70b97a7f | 5300 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5301 | unsigned long flags; |
5302 | ||
dd41f596 IM |
5303 | __sched_fork(idle); |
5304 | idle->se.exec_start = sched_clock(); | |
5305 | ||
b29739f9 | 5306 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 5307 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 5308 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
5309 | |
5310 | spin_lock_irqsave(&rq->lock, flags); | |
5311 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
5312 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5313 | idle->oncpu = 1; | |
5314 | #endif | |
1da177e4 LT |
5315 | spin_unlock_irqrestore(&rq->lock, flags); |
5316 | ||
5317 | /* Set the preempt count _outside_ the spinlocks! */ | |
a1261f54 | 5318 | task_thread_info(idle)->preempt_count = 0; |
6478d880 | 5319 | |
dd41f596 IM |
5320 | /* |
5321 | * The idle tasks have their own, simple scheduling class: | |
5322 | */ | |
5323 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5324 | } |
5325 | ||
5326 | /* | |
5327 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5328 | * indicates which cpus entered this state. This is used | |
5329 | * in the rcu update to wait only for active cpus. For system | |
5330 | * which do not switch off the HZ timer nohz_cpu_mask should | |
5331 | * always be CPU_MASK_NONE. | |
5332 | */ | |
5333 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
5334 | ||
19978ca6 IM |
5335 | /* |
5336 | * Increase the granularity value when there are more CPUs, | |
5337 | * because with more CPUs the 'effective latency' as visible | |
5338 | * to users decreases. But the relationship is not linear, | |
5339 | * so pick a second-best guess by going with the log2 of the | |
5340 | * number of CPUs. | |
5341 | * | |
5342 | * This idea comes from the SD scheduler of Con Kolivas: | |
5343 | */ | |
5344 | static inline void sched_init_granularity(void) | |
5345 | { | |
5346 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
5347 | const unsigned long limit = 200000000; | |
5348 | ||
5349 | sysctl_sched_min_granularity *= factor; | |
5350 | if (sysctl_sched_min_granularity > limit) | |
5351 | sysctl_sched_min_granularity = limit; | |
5352 | ||
5353 | sysctl_sched_latency *= factor; | |
5354 | if (sysctl_sched_latency > limit) | |
5355 | sysctl_sched_latency = limit; | |
5356 | ||
5357 | sysctl_sched_wakeup_granularity *= factor; | |
5358 | sysctl_sched_batch_wakeup_granularity *= factor; | |
5359 | } | |
5360 | ||
1da177e4 LT |
5361 | #ifdef CONFIG_SMP |
5362 | /* | |
5363 | * This is how migration works: | |
5364 | * | |
70b97a7f | 5365 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5366 | * runqueue and wake up that CPU's migration thread. |
5367 | * 2) we down() the locked semaphore => thread blocks. | |
5368 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5369 | * thread off the CPU) | |
5370 | * 4) it gets the migration request and checks whether the migrated | |
5371 | * task is still in the wrong runqueue. | |
5372 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5373 | * it and puts it into the right queue. | |
5374 | * 6) migration thread up()s the semaphore. | |
5375 | * 7) we wake up and the migration is done. | |
5376 | */ | |
5377 | ||
5378 | /* | |
5379 | * Change a given task's CPU affinity. Migrate the thread to a | |
5380 | * proper CPU and schedule it away if the CPU it's executing on | |
5381 | * is removed from the allowed bitmask. | |
5382 | * | |
5383 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5384 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5385 | * call is not atomic; no spinlocks may be held. |
5386 | */ | |
36c8b586 | 5387 | int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask) |
1da177e4 | 5388 | { |
70b97a7f | 5389 | struct migration_req req; |
1da177e4 | 5390 | unsigned long flags; |
70b97a7f | 5391 | struct rq *rq; |
48f24c4d | 5392 | int ret = 0; |
1da177e4 LT |
5393 | |
5394 | rq = task_rq_lock(p, &flags); | |
5395 | if (!cpus_intersects(new_mask, cpu_online_map)) { | |
5396 | ret = -EINVAL; | |
5397 | goto out; | |
5398 | } | |
5399 | ||
73fe6aae GH |
5400 | if (p->sched_class->set_cpus_allowed) |
5401 | p->sched_class->set_cpus_allowed(p, &new_mask); | |
5402 | else { | |
0eab9146 | 5403 | p->cpus_allowed = new_mask; |
6f505b16 | 5404 | p->rt.nr_cpus_allowed = cpus_weight(new_mask); |
73fe6aae GH |
5405 | } |
5406 | ||
1da177e4 LT |
5407 | /* Can the task run on the task's current CPU? If so, we're done */ |
5408 | if (cpu_isset(task_cpu(p), new_mask)) | |
5409 | goto out; | |
5410 | ||
5411 | if (migrate_task(p, any_online_cpu(new_mask), &req)) { | |
5412 | /* Need help from migration thread: drop lock and wait. */ | |
5413 | task_rq_unlock(rq, &flags); | |
5414 | wake_up_process(rq->migration_thread); | |
5415 | wait_for_completion(&req.done); | |
5416 | tlb_migrate_finish(p->mm); | |
5417 | return 0; | |
5418 | } | |
5419 | out: | |
5420 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5421 | |
1da177e4 LT |
5422 | return ret; |
5423 | } | |
1da177e4 LT |
5424 | EXPORT_SYMBOL_GPL(set_cpus_allowed); |
5425 | ||
5426 | /* | |
41a2d6cf | 5427 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5428 | * this because either it can't run here any more (set_cpus_allowed() |
5429 | * away from this CPU, or CPU going down), or because we're | |
5430 | * attempting to rebalance this task on exec (sched_exec). | |
5431 | * | |
5432 | * So we race with normal scheduler movements, but that's OK, as long | |
5433 | * as the task is no longer on this CPU. | |
efc30814 KK |
5434 | * |
5435 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5436 | */ |
efc30814 | 5437 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5438 | { |
70b97a7f | 5439 | struct rq *rq_dest, *rq_src; |
dd41f596 | 5440 | int ret = 0, on_rq; |
1da177e4 LT |
5441 | |
5442 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 5443 | return ret; |
1da177e4 LT |
5444 | |
5445 | rq_src = cpu_rq(src_cpu); | |
5446 | rq_dest = cpu_rq(dest_cpu); | |
5447 | ||
5448 | double_rq_lock(rq_src, rq_dest); | |
5449 | /* Already moved. */ | |
5450 | if (task_cpu(p) != src_cpu) | |
5451 | goto out; | |
5452 | /* Affinity changed (again). */ | |
5453 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
5454 | goto out; | |
5455 | ||
dd41f596 | 5456 | on_rq = p->se.on_rq; |
6e82a3be | 5457 | if (on_rq) |
2e1cb74a | 5458 | deactivate_task(rq_src, p, 0); |
6e82a3be | 5459 | |
1da177e4 | 5460 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
5461 | if (on_rq) { |
5462 | activate_task(rq_dest, p, 0); | |
5463 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 5464 | } |
efc30814 | 5465 | ret = 1; |
1da177e4 LT |
5466 | out: |
5467 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 5468 | return ret; |
1da177e4 LT |
5469 | } |
5470 | ||
5471 | /* | |
5472 | * migration_thread - this is a highprio system thread that performs | |
5473 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5474 | * another runqueue. | |
5475 | */ | |
95cdf3b7 | 5476 | static int migration_thread(void *data) |
1da177e4 | 5477 | { |
1da177e4 | 5478 | int cpu = (long)data; |
70b97a7f | 5479 | struct rq *rq; |
1da177e4 LT |
5480 | |
5481 | rq = cpu_rq(cpu); | |
5482 | BUG_ON(rq->migration_thread != current); | |
5483 | ||
5484 | set_current_state(TASK_INTERRUPTIBLE); | |
5485 | while (!kthread_should_stop()) { | |
70b97a7f | 5486 | struct migration_req *req; |
1da177e4 | 5487 | struct list_head *head; |
1da177e4 | 5488 | |
1da177e4 LT |
5489 | spin_lock_irq(&rq->lock); |
5490 | ||
5491 | if (cpu_is_offline(cpu)) { | |
5492 | spin_unlock_irq(&rq->lock); | |
5493 | goto wait_to_die; | |
5494 | } | |
5495 | ||
5496 | if (rq->active_balance) { | |
5497 | active_load_balance(rq, cpu); | |
5498 | rq->active_balance = 0; | |
5499 | } | |
5500 | ||
5501 | head = &rq->migration_queue; | |
5502 | ||
5503 | if (list_empty(head)) { | |
5504 | spin_unlock_irq(&rq->lock); | |
5505 | schedule(); | |
5506 | set_current_state(TASK_INTERRUPTIBLE); | |
5507 | continue; | |
5508 | } | |
70b97a7f | 5509 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5510 | list_del_init(head->next); |
5511 | ||
674311d5 NP |
5512 | spin_unlock(&rq->lock); |
5513 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5514 | local_irq_enable(); | |
1da177e4 LT |
5515 | |
5516 | complete(&req->done); | |
5517 | } | |
5518 | __set_current_state(TASK_RUNNING); | |
5519 | return 0; | |
5520 | ||
5521 | wait_to_die: | |
5522 | /* Wait for kthread_stop */ | |
5523 | set_current_state(TASK_INTERRUPTIBLE); | |
5524 | while (!kthread_should_stop()) { | |
5525 | schedule(); | |
5526 | set_current_state(TASK_INTERRUPTIBLE); | |
5527 | } | |
5528 | __set_current_state(TASK_RUNNING); | |
5529 | return 0; | |
5530 | } | |
5531 | ||
5532 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5533 | |
5534 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5535 | { | |
5536 | int ret; | |
5537 | ||
5538 | local_irq_disable(); | |
5539 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5540 | local_irq_enable(); | |
5541 | return ret; | |
5542 | } | |
5543 | ||
054b9108 | 5544 | /* |
3a4fa0a2 | 5545 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
5546 | * NOTE: interrupts should be disabled by the caller |
5547 | */ | |
48f24c4d | 5548 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5549 | { |
efc30814 | 5550 | unsigned long flags; |
1da177e4 | 5551 | cpumask_t mask; |
70b97a7f IM |
5552 | struct rq *rq; |
5553 | int dest_cpu; | |
1da177e4 | 5554 | |
3a5c359a AK |
5555 | do { |
5556 | /* On same node? */ | |
5557 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
5558 | cpus_and(mask, mask, p->cpus_allowed); | |
5559 | dest_cpu = any_online_cpu(mask); | |
5560 | ||
5561 | /* On any allowed CPU? */ | |
5562 | if (dest_cpu == NR_CPUS) | |
5563 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
5564 | ||
5565 | /* No more Mr. Nice Guy. */ | |
5566 | if (dest_cpu == NR_CPUS) { | |
470fd646 CW |
5567 | cpumask_t cpus_allowed = cpuset_cpus_allowed_locked(p); |
5568 | /* | |
5569 | * Try to stay on the same cpuset, where the | |
5570 | * current cpuset may be a subset of all cpus. | |
5571 | * The cpuset_cpus_allowed_locked() variant of | |
41a2d6cf | 5572 | * cpuset_cpus_allowed() will not block. It must be |
470fd646 CW |
5573 | * called within calls to cpuset_lock/cpuset_unlock. |
5574 | */ | |
3a5c359a | 5575 | rq = task_rq_lock(p, &flags); |
470fd646 | 5576 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
5577 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5578 | task_rq_unlock(rq, &flags); | |
1da177e4 | 5579 | |
3a5c359a AK |
5580 | /* |
5581 | * Don't tell them about moving exiting tasks or | |
5582 | * kernel threads (both mm NULL), since they never | |
5583 | * leave kernel. | |
5584 | */ | |
41a2d6cf | 5585 | if (p->mm && printk_ratelimit()) { |
3a5c359a AK |
5586 | printk(KERN_INFO "process %d (%s) no " |
5587 | "longer affine to cpu%d\n", | |
41a2d6cf IM |
5588 | task_pid_nr(p), p->comm, dead_cpu); |
5589 | } | |
3a5c359a | 5590 | } |
f7b4cddc | 5591 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
5592 | } |
5593 | ||
5594 | /* | |
5595 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5596 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5597 | * for performance reasons the counter is not stricly tracking tasks to | |
5598 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5599 | * to keep the global sum constant after CPU-down: | |
5600 | */ | |
70b97a7f | 5601 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5602 | { |
70b97a7f | 5603 | struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL)); |
1da177e4 LT |
5604 | unsigned long flags; |
5605 | ||
5606 | local_irq_save(flags); | |
5607 | double_rq_lock(rq_src, rq_dest); | |
5608 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5609 | rq_src->nr_uninterruptible = 0; | |
5610 | double_rq_unlock(rq_src, rq_dest); | |
5611 | local_irq_restore(flags); | |
5612 | } | |
5613 | ||
5614 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5615 | static void migrate_live_tasks(int src_cpu) | |
5616 | { | |
48f24c4d | 5617 | struct task_struct *p, *t; |
1da177e4 | 5618 | |
f7b4cddc | 5619 | read_lock(&tasklist_lock); |
1da177e4 | 5620 | |
48f24c4d IM |
5621 | do_each_thread(t, p) { |
5622 | if (p == current) | |
1da177e4 LT |
5623 | continue; |
5624 | ||
48f24c4d IM |
5625 | if (task_cpu(p) == src_cpu) |
5626 | move_task_off_dead_cpu(src_cpu, p); | |
5627 | } while_each_thread(t, p); | |
1da177e4 | 5628 | |
f7b4cddc | 5629 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5630 | } |
5631 | ||
dd41f596 IM |
5632 | /* |
5633 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5634 | * It does so by boosting its priority to highest possible. |
5635 | * Used by CPU offline code. | |
1da177e4 LT |
5636 | */ |
5637 | void sched_idle_next(void) | |
5638 | { | |
48f24c4d | 5639 | int this_cpu = smp_processor_id(); |
70b97a7f | 5640 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5641 | struct task_struct *p = rq->idle; |
5642 | unsigned long flags; | |
5643 | ||
5644 | /* cpu has to be offline */ | |
48f24c4d | 5645 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5646 | |
48f24c4d IM |
5647 | /* |
5648 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5649 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5650 | */ |
5651 | spin_lock_irqsave(&rq->lock, flags); | |
5652 | ||
dd41f596 | 5653 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 5654 | |
94bc9a7b DA |
5655 | update_rq_clock(rq); |
5656 | activate_task(rq, p, 0); | |
1da177e4 LT |
5657 | |
5658 | spin_unlock_irqrestore(&rq->lock, flags); | |
5659 | } | |
5660 | ||
48f24c4d IM |
5661 | /* |
5662 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5663 | * offline. |
5664 | */ | |
5665 | void idle_task_exit(void) | |
5666 | { | |
5667 | struct mm_struct *mm = current->active_mm; | |
5668 | ||
5669 | BUG_ON(cpu_online(smp_processor_id())); | |
5670 | ||
5671 | if (mm != &init_mm) | |
5672 | switch_mm(mm, &init_mm, current); | |
5673 | mmdrop(mm); | |
5674 | } | |
5675 | ||
054b9108 | 5676 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5677 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5678 | { |
70b97a7f | 5679 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5680 | |
5681 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5682 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5683 | |
5684 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5685 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5686 | |
48f24c4d | 5687 | get_task_struct(p); |
1da177e4 LT |
5688 | |
5689 | /* | |
5690 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 5691 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
5692 | * fine. |
5693 | */ | |
f7b4cddc | 5694 | spin_unlock_irq(&rq->lock); |
48f24c4d | 5695 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 5696 | spin_lock_irq(&rq->lock); |
1da177e4 | 5697 | |
48f24c4d | 5698 | put_task_struct(p); |
1da177e4 LT |
5699 | } |
5700 | ||
5701 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5702 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5703 | { | |
70b97a7f | 5704 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5705 | struct task_struct *next; |
48f24c4d | 5706 | |
dd41f596 IM |
5707 | for ( ; ; ) { |
5708 | if (!rq->nr_running) | |
5709 | break; | |
a8e504d2 | 5710 | update_rq_clock(rq); |
ff95f3df | 5711 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
5712 | if (!next) |
5713 | break; | |
5714 | migrate_dead(dead_cpu, next); | |
e692ab53 | 5715 | |
1da177e4 LT |
5716 | } |
5717 | } | |
5718 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5719 | ||
e692ab53 NP |
5720 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5721 | ||
5722 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5723 | { |
5724 | .procname = "sched_domain", | |
c57baf1e | 5725 | .mode = 0555, |
e0361851 | 5726 | }, |
38605cae | 5727 | {0, }, |
e692ab53 NP |
5728 | }; |
5729 | ||
5730 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 5731 | { |
c57baf1e | 5732 | .ctl_name = CTL_KERN, |
e0361851 | 5733 | .procname = "kernel", |
c57baf1e | 5734 | .mode = 0555, |
e0361851 AD |
5735 | .child = sd_ctl_dir, |
5736 | }, | |
38605cae | 5737 | {0, }, |
e692ab53 NP |
5738 | }; |
5739 | ||
5740 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5741 | { | |
5742 | struct ctl_table *entry = | |
5cf9f062 | 5743 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5744 | |
e692ab53 NP |
5745 | return entry; |
5746 | } | |
5747 | ||
6382bc90 MM |
5748 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5749 | { | |
cd790076 | 5750 | struct ctl_table *entry; |
6382bc90 | 5751 | |
cd790076 MM |
5752 | /* |
5753 | * In the intermediate directories, both the child directory and | |
5754 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5755 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5756 | * static strings and all have proc handlers. |
5757 | */ | |
5758 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5759 | if (entry->child) |
5760 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5761 | if (entry->proc_handler == NULL) |
5762 | kfree(entry->procname); | |
5763 | } | |
6382bc90 MM |
5764 | |
5765 | kfree(*tablep); | |
5766 | *tablep = NULL; | |
5767 | } | |
5768 | ||
e692ab53 | 5769 | static void |
e0361851 | 5770 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5771 | const char *procname, void *data, int maxlen, |
5772 | mode_t mode, proc_handler *proc_handler) | |
5773 | { | |
e692ab53 NP |
5774 | entry->procname = procname; |
5775 | entry->data = data; | |
5776 | entry->maxlen = maxlen; | |
5777 | entry->mode = mode; | |
5778 | entry->proc_handler = proc_handler; | |
5779 | } | |
5780 | ||
5781 | static struct ctl_table * | |
5782 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5783 | { | |
ace8b3d6 | 5784 | struct ctl_table *table = sd_alloc_ctl_entry(12); |
e692ab53 | 5785 | |
ad1cdc1d MM |
5786 | if (table == NULL) |
5787 | return NULL; | |
5788 | ||
e0361851 | 5789 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5790 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5791 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5792 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5793 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5794 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5795 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5796 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5797 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5798 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5799 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5800 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5801 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5802 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5803 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5804 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5805 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5806 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5807 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5808 | &sd->cache_nice_tries, |
5809 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5810 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5811 | sizeof(int), 0644, proc_dointvec_minmax); |
6323469f | 5812 | /* &table[11] is terminator */ |
e692ab53 NP |
5813 | |
5814 | return table; | |
5815 | } | |
5816 | ||
9a4e7159 | 5817 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5818 | { |
5819 | struct ctl_table *entry, *table; | |
5820 | struct sched_domain *sd; | |
5821 | int domain_num = 0, i; | |
5822 | char buf[32]; | |
5823 | ||
5824 | for_each_domain(cpu, sd) | |
5825 | domain_num++; | |
5826 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5827 | if (table == NULL) |
5828 | return NULL; | |
e692ab53 NP |
5829 | |
5830 | i = 0; | |
5831 | for_each_domain(cpu, sd) { | |
5832 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5833 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5834 | entry->mode = 0555; |
e692ab53 NP |
5835 | entry->child = sd_alloc_ctl_domain_table(sd); |
5836 | entry++; | |
5837 | i++; | |
5838 | } | |
5839 | return table; | |
5840 | } | |
5841 | ||
5842 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5843 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
5844 | { |
5845 | int i, cpu_num = num_online_cpus(); | |
5846 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
5847 | char buf[32]; | |
5848 | ||
7378547f MM |
5849 | WARN_ON(sd_ctl_dir[0].child); |
5850 | sd_ctl_dir[0].child = entry; | |
5851 | ||
ad1cdc1d MM |
5852 | if (entry == NULL) |
5853 | return; | |
5854 | ||
97b6ea7b | 5855 | for_each_online_cpu(i) { |
e692ab53 | 5856 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5857 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5858 | entry->mode = 0555; |
e692ab53 | 5859 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5860 | entry++; |
e692ab53 | 5861 | } |
7378547f MM |
5862 | |
5863 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5864 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5865 | } | |
6382bc90 | 5866 | |
7378547f | 5867 | /* may be called multiple times per register */ |
6382bc90 MM |
5868 | static void unregister_sched_domain_sysctl(void) |
5869 | { | |
7378547f MM |
5870 | if (sd_sysctl_header) |
5871 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5872 | sd_sysctl_header = NULL; |
7378547f MM |
5873 | if (sd_ctl_dir[0].child) |
5874 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5875 | } |
e692ab53 | 5876 | #else |
6382bc90 MM |
5877 | static void register_sched_domain_sysctl(void) |
5878 | { | |
5879 | } | |
5880 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5881 | { |
5882 | } | |
5883 | #endif | |
5884 | ||
1da177e4 LT |
5885 | /* |
5886 | * migration_call - callback that gets triggered when a CPU is added. | |
5887 | * Here we can start up the necessary migration thread for the new CPU. | |
5888 | */ | |
48f24c4d IM |
5889 | static int __cpuinit |
5890 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5891 | { |
1da177e4 | 5892 | struct task_struct *p; |
48f24c4d | 5893 | int cpu = (long)hcpu; |
1da177e4 | 5894 | unsigned long flags; |
70b97a7f | 5895 | struct rq *rq; |
1da177e4 LT |
5896 | |
5897 | switch (action) { | |
5be9361c | 5898 | |
1da177e4 | 5899 | case CPU_UP_PREPARE: |
8bb78442 | 5900 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 5901 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
5902 | if (IS_ERR(p)) |
5903 | return NOTIFY_BAD; | |
1da177e4 LT |
5904 | kthread_bind(p, cpu); |
5905 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5906 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 5907 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
5908 | task_rq_unlock(rq, &flags); |
5909 | cpu_rq(cpu)->migration_thread = p; | |
5910 | break; | |
48f24c4d | 5911 | |
1da177e4 | 5912 | case CPU_ONLINE: |
8bb78442 | 5913 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 5914 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 5915 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
5916 | |
5917 | /* Update our root-domain */ | |
5918 | rq = cpu_rq(cpu); | |
5919 | spin_lock_irqsave(&rq->lock, flags); | |
5920 | if (rq->rd) { | |
5921 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
5922 | cpu_set(cpu, rq->rd->online); | |
5923 | } | |
5924 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 5925 | break; |
48f24c4d | 5926 | |
1da177e4 LT |
5927 | #ifdef CONFIG_HOTPLUG_CPU |
5928 | case CPU_UP_CANCELED: | |
8bb78442 | 5929 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5930 | if (!cpu_rq(cpu)->migration_thread) |
5931 | break; | |
41a2d6cf | 5932 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
5933 | kthread_bind(cpu_rq(cpu)->migration_thread, |
5934 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
5935 | kthread_stop(cpu_rq(cpu)->migration_thread); |
5936 | cpu_rq(cpu)->migration_thread = NULL; | |
5937 | break; | |
48f24c4d | 5938 | |
1da177e4 | 5939 | case CPU_DEAD: |
8bb78442 | 5940 | case CPU_DEAD_FROZEN: |
470fd646 | 5941 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
5942 | migrate_live_tasks(cpu); |
5943 | rq = cpu_rq(cpu); | |
5944 | kthread_stop(rq->migration_thread); | |
5945 | rq->migration_thread = NULL; | |
5946 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 5947 | spin_lock_irq(&rq->lock); |
a8e504d2 | 5948 | update_rq_clock(rq); |
2e1cb74a | 5949 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 5950 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
5951 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5952 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 5953 | migrate_dead_tasks(cpu); |
d2da272a | 5954 | spin_unlock_irq(&rq->lock); |
470fd646 | 5955 | cpuset_unlock(); |
1da177e4 LT |
5956 | migrate_nr_uninterruptible(rq); |
5957 | BUG_ON(rq->nr_running != 0); | |
5958 | ||
41a2d6cf IM |
5959 | /* |
5960 | * No need to migrate the tasks: it was best-effort if | |
5961 | * they didn't take sched_hotcpu_mutex. Just wake up | |
5962 | * the requestors. | |
5963 | */ | |
1da177e4 LT |
5964 | spin_lock_irq(&rq->lock); |
5965 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
5966 | struct migration_req *req; |
5967 | ||
1da177e4 | 5968 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5969 | struct migration_req, list); |
1da177e4 LT |
5970 | list_del_init(&req->list); |
5971 | complete(&req->done); | |
5972 | } | |
5973 | spin_unlock_irq(&rq->lock); | |
5974 | break; | |
57d885fe | 5975 | |
08f503b0 GH |
5976 | case CPU_DYING: |
5977 | case CPU_DYING_FROZEN: | |
57d885fe GH |
5978 | /* Update our root-domain */ |
5979 | rq = cpu_rq(cpu); | |
5980 | spin_lock_irqsave(&rq->lock, flags); | |
5981 | if (rq->rd) { | |
5982 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
5983 | cpu_clear(cpu, rq->rd->online); | |
5984 | } | |
5985 | spin_unlock_irqrestore(&rq->lock, flags); | |
5986 | break; | |
1da177e4 LT |
5987 | #endif |
5988 | } | |
5989 | return NOTIFY_OK; | |
5990 | } | |
5991 | ||
5992 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
5993 | * happens before everything else. | |
5994 | */ | |
26c2143b | 5995 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5996 | .notifier_call = migration_call, |
5997 | .priority = 10 | |
5998 | }; | |
5999 | ||
e6fe6649 | 6000 | void __init migration_init(void) |
1da177e4 LT |
6001 | { |
6002 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6003 | int err; |
48f24c4d IM |
6004 | |
6005 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
6006 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6007 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6008 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6009 | register_cpu_notifier(&migration_notifier); | |
1da177e4 LT |
6010 | } |
6011 | #endif | |
6012 | ||
6013 | #ifdef CONFIG_SMP | |
476f3534 CL |
6014 | |
6015 | /* Number of possible processor ids */ | |
6016 | int nr_cpu_ids __read_mostly = NR_CPUS; | |
6017 | EXPORT_SYMBOL(nr_cpu_ids); | |
6018 | ||
3e9830dc | 6019 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff IM |
6020 | |
6021 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level) | |
1da177e4 | 6022 | { |
4dcf6aff IM |
6023 | struct sched_group *group = sd->groups; |
6024 | cpumask_t groupmask; | |
6025 | char str[NR_CPUS]; | |
1da177e4 | 6026 | |
4dcf6aff IM |
6027 | cpumask_scnprintf(str, NR_CPUS, sd->span); |
6028 | cpus_clear(groupmask); | |
6029 | ||
6030 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6031 | ||
6032 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
6033 | printk("does not load-balance\n"); | |
6034 | if (sd->parent) | |
6035 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
6036 | " has parent"); | |
6037 | return -1; | |
41c7ce9a NP |
6038 | } |
6039 | ||
4dcf6aff IM |
6040 | printk(KERN_CONT "span %s\n", str); |
6041 | ||
6042 | if (!cpu_isset(cpu, sd->span)) { | |
6043 | printk(KERN_ERR "ERROR: domain->span does not contain " | |
6044 | "CPU%d\n", cpu); | |
6045 | } | |
6046 | if (!cpu_isset(cpu, group->cpumask)) { | |
6047 | printk(KERN_ERR "ERROR: domain->groups does not contain" | |
6048 | " CPU%d\n", cpu); | |
6049 | } | |
1da177e4 | 6050 | |
4dcf6aff | 6051 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6052 | do { |
4dcf6aff IM |
6053 | if (!group) { |
6054 | printk("\n"); | |
6055 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6056 | break; |
6057 | } | |
6058 | ||
4dcf6aff IM |
6059 | if (!group->__cpu_power) { |
6060 | printk(KERN_CONT "\n"); | |
6061 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6062 | "set\n"); | |
6063 | break; | |
6064 | } | |
1da177e4 | 6065 | |
4dcf6aff IM |
6066 | if (!cpus_weight(group->cpumask)) { |
6067 | printk(KERN_CONT "\n"); | |
6068 | printk(KERN_ERR "ERROR: empty group\n"); | |
6069 | break; | |
6070 | } | |
1da177e4 | 6071 | |
4dcf6aff IM |
6072 | if (cpus_intersects(groupmask, group->cpumask)) { |
6073 | printk(KERN_CONT "\n"); | |
6074 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
6075 | break; | |
6076 | } | |
1da177e4 | 6077 | |
4dcf6aff | 6078 | cpus_or(groupmask, groupmask, group->cpumask); |
1da177e4 | 6079 | |
4dcf6aff IM |
6080 | cpumask_scnprintf(str, NR_CPUS, group->cpumask); |
6081 | printk(KERN_CONT " %s", str); | |
1da177e4 | 6082 | |
4dcf6aff IM |
6083 | group = group->next; |
6084 | } while (group != sd->groups); | |
6085 | printk(KERN_CONT "\n"); | |
1da177e4 | 6086 | |
4dcf6aff IM |
6087 | if (!cpus_equal(sd->span, groupmask)) |
6088 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); | |
1da177e4 | 6089 | |
4dcf6aff IM |
6090 | if (sd->parent && !cpus_subset(groupmask, sd->parent->span)) |
6091 | printk(KERN_ERR "ERROR: parent span is not a superset " | |
6092 | "of domain->span\n"); | |
6093 | return 0; | |
6094 | } | |
1da177e4 | 6095 | |
4dcf6aff IM |
6096 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6097 | { | |
6098 | int level = 0; | |
1da177e4 | 6099 | |
4dcf6aff IM |
6100 | if (!sd) { |
6101 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6102 | return; | |
6103 | } | |
1da177e4 | 6104 | |
4dcf6aff IM |
6105 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6106 | ||
6107 | for (;;) { | |
6108 | if (sched_domain_debug_one(sd, cpu, level)) | |
6109 | break; | |
1da177e4 LT |
6110 | level++; |
6111 | sd = sd->parent; | |
33859f7f | 6112 | if (!sd) |
4dcf6aff IM |
6113 | break; |
6114 | } | |
1da177e4 LT |
6115 | } |
6116 | #else | |
48f24c4d | 6117 | # define sched_domain_debug(sd, cpu) do { } while (0) |
1da177e4 LT |
6118 | #endif |
6119 | ||
1a20ff27 | 6120 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
6121 | { |
6122 | if (cpus_weight(sd->span) == 1) | |
6123 | return 1; | |
6124 | ||
6125 | /* Following flags need at least 2 groups */ | |
6126 | if (sd->flags & (SD_LOAD_BALANCE | | |
6127 | SD_BALANCE_NEWIDLE | | |
6128 | SD_BALANCE_FORK | | |
89c4710e SS |
6129 | SD_BALANCE_EXEC | |
6130 | SD_SHARE_CPUPOWER | | |
6131 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6132 | if (sd->groups != sd->groups->next) |
6133 | return 0; | |
6134 | } | |
6135 | ||
6136 | /* Following flags don't use groups */ | |
6137 | if (sd->flags & (SD_WAKE_IDLE | | |
6138 | SD_WAKE_AFFINE | | |
6139 | SD_WAKE_BALANCE)) | |
6140 | return 0; | |
6141 | ||
6142 | return 1; | |
6143 | } | |
6144 | ||
48f24c4d IM |
6145 | static int |
6146 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6147 | { |
6148 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6149 | ||
6150 | if (sd_degenerate(parent)) | |
6151 | return 1; | |
6152 | ||
6153 | if (!cpus_equal(sd->span, parent->span)) | |
6154 | return 0; | |
6155 | ||
6156 | /* Does parent contain flags not in child? */ | |
6157 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
6158 | if (cflags & SD_WAKE_AFFINE) | |
6159 | pflags &= ~SD_WAKE_BALANCE; | |
6160 | /* Flags needing groups don't count if only 1 group in parent */ | |
6161 | if (parent->groups == parent->groups->next) { | |
6162 | pflags &= ~(SD_LOAD_BALANCE | | |
6163 | SD_BALANCE_NEWIDLE | | |
6164 | SD_BALANCE_FORK | | |
89c4710e SS |
6165 | SD_BALANCE_EXEC | |
6166 | SD_SHARE_CPUPOWER | | |
6167 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
6168 | } |
6169 | if (~cflags & pflags) | |
6170 | return 0; | |
6171 | ||
6172 | return 1; | |
6173 | } | |
6174 | ||
57d885fe GH |
6175 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6176 | { | |
6177 | unsigned long flags; | |
6178 | const struct sched_class *class; | |
6179 | ||
6180 | spin_lock_irqsave(&rq->lock, flags); | |
6181 | ||
6182 | if (rq->rd) { | |
6183 | struct root_domain *old_rd = rq->rd; | |
6184 | ||
0eab9146 | 6185 | for (class = sched_class_highest; class; class = class->next) { |
57d885fe GH |
6186 | if (class->leave_domain) |
6187 | class->leave_domain(rq); | |
0eab9146 | 6188 | } |
57d885fe | 6189 | |
dc938520 GH |
6190 | cpu_clear(rq->cpu, old_rd->span); |
6191 | cpu_clear(rq->cpu, old_rd->online); | |
6192 | ||
57d885fe GH |
6193 | if (atomic_dec_and_test(&old_rd->refcount)) |
6194 | kfree(old_rd); | |
6195 | } | |
6196 | ||
6197 | atomic_inc(&rd->refcount); | |
6198 | rq->rd = rd; | |
6199 | ||
dc938520 | 6200 | cpu_set(rq->cpu, rd->span); |
1f94ef59 GH |
6201 | if (cpu_isset(rq->cpu, cpu_online_map)) |
6202 | cpu_set(rq->cpu, rd->online); | |
dc938520 | 6203 | |
0eab9146 | 6204 | for (class = sched_class_highest; class; class = class->next) { |
57d885fe GH |
6205 | if (class->join_domain) |
6206 | class->join_domain(rq); | |
0eab9146 | 6207 | } |
57d885fe GH |
6208 | |
6209 | spin_unlock_irqrestore(&rq->lock, flags); | |
6210 | } | |
6211 | ||
dc938520 | 6212 | static void init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6213 | { |
6214 | memset(rd, 0, sizeof(*rd)); | |
6215 | ||
dc938520 GH |
6216 | cpus_clear(rd->span); |
6217 | cpus_clear(rd->online); | |
57d885fe GH |
6218 | } |
6219 | ||
6220 | static void init_defrootdomain(void) | |
6221 | { | |
dc938520 | 6222 | init_rootdomain(&def_root_domain); |
57d885fe GH |
6223 | atomic_set(&def_root_domain.refcount, 1); |
6224 | } | |
6225 | ||
dc938520 | 6226 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6227 | { |
6228 | struct root_domain *rd; | |
6229 | ||
6230 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6231 | if (!rd) | |
6232 | return NULL; | |
6233 | ||
dc938520 | 6234 | init_rootdomain(rd); |
57d885fe GH |
6235 | |
6236 | return rd; | |
6237 | } | |
6238 | ||
1da177e4 | 6239 | /* |
0eab9146 | 6240 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6241 | * hold the hotplug lock. |
6242 | */ | |
0eab9146 IM |
6243 | static void |
6244 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6245 | { |
70b97a7f | 6246 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6247 | struct sched_domain *tmp; |
6248 | ||
6249 | /* Remove the sched domains which do not contribute to scheduling. */ | |
6250 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
6251 | struct sched_domain *parent = tmp->parent; | |
6252 | if (!parent) | |
6253 | break; | |
1a848870 | 6254 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6255 | tmp->parent = parent->parent; |
1a848870 SS |
6256 | if (parent->parent) |
6257 | parent->parent->child = tmp; | |
6258 | } | |
245af2c7 SS |
6259 | } |
6260 | ||
1a848870 | 6261 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6262 | sd = sd->parent; |
1a848870 SS |
6263 | if (sd) |
6264 | sd->child = NULL; | |
6265 | } | |
1da177e4 LT |
6266 | |
6267 | sched_domain_debug(sd, cpu); | |
6268 | ||
57d885fe | 6269 | rq_attach_root(rq, rd); |
674311d5 | 6270 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6271 | } |
6272 | ||
6273 | /* cpus with isolated domains */ | |
67af63a6 | 6274 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
6275 | |
6276 | /* Setup the mask of cpus configured for isolated domains */ | |
6277 | static int __init isolated_cpu_setup(char *str) | |
6278 | { | |
6279 | int ints[NR_CPUS], i; | |
6280 | ||
6281 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
6282 | cpus_clear(cpu_isolated_map); | |
6283 | for (i = 1; i <= ints[0]; i++) | |
6284 | if (ints[i] < NR_CPUS) | |
6285 | cpu_set(ints[i], cpu_isolated_map); | |
6286 | return 1; | |
6287 | } | |
6288 | ||
8927f494 | 6289 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6290 | |
6291 | /* | |
6711cab4 SS |
6292 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6293 | * to a function which identifies what group(along with sched group) a CPU | |
6294 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
6295 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
6296 | * |
6297 | * init_sched_build_groups will build a circular linked list of the groups | |
6298 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6299 | * and ->cpu_power to 0. | |
6300 | */ | |
a616058b | 6301 | static void |
6711cab4 SS |
6302 | init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map, |
6303 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, | |
6304 | struct sched_group **sg)) | |
1da177e4 LT |
6305 | { |
6306 | struct sched_group *first = NULL, *last = NULL; | |
6307 | cpumask_t covered = CPU_MASK_NONE; | |
6308 | int i; | |
6309 | ||
6310 | for_each_cpu_mask(i, span) { | |
6711cab4 SS |
6311 | struct sched_group *sg; |
6312 | int group = group_fn(i, cpu_map, &sg); | |
1da177e4 LT |
6313 | int j; |
6314 | ||
6315 | if (cpu_isset(i, covered)) | |
6316 | continue; | |
6317 | ||
6318 | sg->cpumask = CPU_MASK_NONE; | |
5517d86b | 6319 | sg->__cpu_power = 0; |
1da177e4 LT |
6320 | |
6321 | for_each_cpu_mask(j, span) { | |
6711cab4 | 6322 | if (group_fn(j, cpu_map, NULL) != group) |
1da177e4 LT |
6323 | continue; |
6324 | ||
6325 | cpu_set(j, covered); | |
6326 | cpu_set(j, sg->cpumask); | |
6327 | } | |
6328 | if (!first) | |
6329 | first = sg; | |
6330 | if (last) | |
6331 | last->next = sg; | |
6332 | last = sg; | |
6333 | } | |
6334 | last->next = first; | |
6335 | } | |
6336 | ||
9c1cfda2 | 6337 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6338 | |
9c1cfda2 | 6339 | #ifdef CONFIG_NUMA |
198e2f18 | 6340 | |
9c1cfda2 JH |
6341 | /** |
6342 | * find_next_best_node - find the next node to include in a sched_domain | |
6343 | * @node: node whose sched_domain we're building | |
6344 | * @used_nodes: nodes already in the sched_domain | |
6345 | * | |
41a2d6cf | 6346 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6347 | * finds the closest node not already in the @used_nodes map. |
6348 | * | |
6349 | * Should use nodemask_t. | |
6350 | */ | |
6351 | static int find_next_best_node(int node, unsigned long *used_nodes) | |
6352 | { | |
6353 | int i, n, val, min_val, best_node = 0; | |
6354 | ||
6355 | min_val = INT_MAX; | |
6356 | ||
6357 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6358 | /* Start at @node */ | |
6359 | n = (node + i) % MAX_NUMNODES; | |
6360 | ||
6361 | if (!nr_cpus_node(n)) | |
6362 | continue; | |
6363 | ||
6364 | /* Skip already used nodes */ | |
6365 | if (test_bit(n, used_nodes)) | |
6366 | continue; | |
6367 | ||
6368 | /* Simple min distance search */ | |
6369 | val = node_distance(node, n); | |
6370 | ||
6371 | if (val < min_val) { | |
6372 | min_val = val; | |
6373 | best_node = n; | |
6374 | } | |
6375 | } | |
6376 | ||
6377 | set_bit(best_node, used_nodes); | |
6378 | return best_node; | |
6379 | } | |
6380 | ||
6381 | /** | |
6382 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6383 | * @node: node whose cpumask we're constructing | |
6384 | * @size: number of nodes to include in this span | |
6385 | * | |
41a2d6cf | 6386 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6387 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6388 | * out optimally. | |
6389 | */ | |
6390 | static cpumask_t sched_domain_node_span(int node) | |
6391 | { | |
9c1cfda2 | 6392 | DECLARE_BITMAP(used_nodes, MAX_NUMNODES); |
48f24c4d IM |
6393 | cpumask_t span, nodemask; |
6394 | int i; | |
9c1cfda2 JH |
6395 | |
6396 | cpus_clear(span); | |
6397 | bitmap_zero(used_nodes, MAX_NUMNODES); | |
6398 | ||
6399 | nodemask = node_to_cpumask(node); | |
6400 | cpus_or(span, span, nodemask); | |
6401 | set_bit(node, used_nodes); | |
6402 | ||
6403 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
6404 | int next_node = find_next_best_node(node, used_nodes); | |
48f24c4d | 6405 | |
9c1cfda2 JH |
6406 | nodemask = node_to_cpumask(next_node); |
6407 | cpus_or(span, span, nodemask); | |
6408 | } | |
6409 | ||
6410 | return span; | |
6411 | } | |
6412 | #endif | |
6413 | ||
5c45bf27 | 6414 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6415 | |
9c1cfda2 | 6416 | /* |
48f24c4d | 6417 | * SMT sched-domains: |
9c1cfda2 | 6418 | */ |
1da177e4 LT |
6419 | #ifdef CONFIG_SCHED_SMT |
6420 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 6421 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 6422 | |
41a2d6cf IM |
6423 | static int |
6424 | cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1da177e4 | 6425 | { |
6711cab4 SS |
6426 | if (sg) |
6427 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
6428 | return cpu; |
6429 | } | |
6430 | #endif | |
6431 | ||
48f24c4d IM |
6432 | /* |
6433 | * multi-core sched-domains: | |
6434 | */ | |
1e9f28fa SS |
6435 | #ifdef CONFIG_SCHED_MC |
6436 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 6437 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
6438 | #endif |
6439 | ||
6440 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf IM |
6441 | static int |
6442 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1e9f28fa | 6443 | { |
6711cab4 | 6444 | int group; |
d5a7430d | 6445 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 6446 | cpus_and(mask, mask, *cpu_map); |
6711cab4 SS |
6447 | group = first_cpu(mask); |
6448 | if (sg) | |
6449 | *sg = &per_cpu(sched_group_core, group); | |
6450 | return group; | |
1e9f28fa SS |
6451 | } |
6452 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf IM |
6453 | static int |
6454 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1e9f28fa | 6455 | { |
6711cab4 SS |
6456 | if (sg) |
6457 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
6458 | return cpu; |
6459 | } | |
6460 | #endif | |
6461 | ||
1da177e4 | 6462 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 6463 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 6464 | |
41a2d6cf IM |
6465 | static int |
6466 | cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1da177e4 | 6467 | { |
6711cab4 | 6468 | int group; |
48f24c4d | 6469 | #ifdef CONFIG_SCHED_MC |
1e9f28fa | 6470 | cpumask_t mask = cpu_coregroup_map(cpu); |
a616058b | 6471 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 6472 | group = first_cpu(mask); |
1e9f28fa | 6473 | #elif defined(CONFIG_SCHED_SMT) |
d5a7430d | 6474 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 6475 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 6476 | group = first_cpu(mask); |
1da177e4 | 6477 | #else |
6711cab4 | 6478 | group = cpu; |
1da177e4 | 6479 | #endif |
6711cab4 SS |
6480 | if (sg) |
6481 | *sg = &per_cpu(sched_group_phys, group); | |
6482 | return group; | |
1da177e4 LT |
6483 | } |
6484 | ||
6485 | #ifdef CONFIG_NUMA | |
1da177e4 | 6486 | /* |
9c1cfda2 JH |
6487 | * The init_sched_build_groups can't handle what we want to do with node |
6488 | * groups, so roll our own. Now each node has its own list of groups which | |
6489 | * gets dynamically allocated. | |
1da177e4 | 6490 | */ |
9c1cfda2 | 6491 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
d1b55138 | 6492 | static struct sched_group **sched_group_nodes_bycpu[NR_CPUS]; |
1da177e4 | 6493 | |
9c1cfda2 | 6494 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 6495 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 6496 | |
6711cab4 SS |
6497 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
6498 | struct sched_group **sg) | |
9c1cfda2 | 6499 | { |
6711cab4 SS |
6500 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu)); |
6501 | int group; | |
6502 | ||
6503 | cpus_and(nodemask, nodemask, *cpu_map); | |
6504 | group = first_cpu(nodemask); | |
6505 | ||
6506 | if (sg) | |
6507 | *sg = &per_cpu(sched_group_allnodes, group); | |
6508 | return group; | |
1da177e4 | 6509 | } |
6711cab4 | 6510 | |
08069033 SS |
6511 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6512 | { | |
6513 | struct sched_group *sg = group_head; | |
6514 | int j; | |
6515 | ||
6516 | if (!sg) | |
6517 | return; | |
3a5c359a AK |
6518 | do { |
6519 | for_each_cpu_mask(j, sg->cpumask) { | |
6520 | struct sched_domain *sd; | |
08069033 | 6521 | |
3a5c359a AK |
6522 | sd = &per_cpu(phys_domains, j); |
6523 | if (j != first_cpu(sd->groups->cpumask)) { | |
6524 | /* | |
6525 | * Only add "power" once for each | |
6526 | * physical package. | |
6527 | */ | |
6528 | continue; | |
6529 | } | |
08069033 | 6530 | |
3a5c359a AK |
6531 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
6532 | } | |
6533 | sg = sg->next; | |
6534 | } while (sg != group_head); | |
08069033 | 6535 | } |
1da177e4 LT |
6536 | #endif |
6537 | ||
a616058b | 6538 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6539 | /* Free memory allocated for various sched_group structures */ |
6540 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6541 | { | |
a616058b | 6542 | int cpu, i; |
51888ca2 SV |
6543 | |
6544 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
6545 | struct sched_group **sched_group_nodes |
6546 | = sched_group_nodes_bycpu[cpu]; | |
6547 | ||
51888ca2 SV |
6548 | if (!sched_group_nodes) |
6549 | continue; | |
6550 | ||
6551 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6552 | cpumask_t nodemask = node_to_cpumask(i); | |
6553 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; | |
6554 | ||
6555 | cpus_and(nodemask, nodemask, *cpu_map); | |
6556 | if (cpus_empty(nodemask)) | |
6557 | continue; | |
6558 | ||
6559 | if (sg == NULL) | |
6560 | continue; | |
6561 | sg = sg->next; | |
6562 | next_sg: | |
6563 | oldsg = sg; | |
6564 | sg = sg->next; | |
6565 | kfree(oldsg); | |
6566 | if (oldsg != sched_group_nodes[i]) | |
6567 | goto next_sg; | |
6568 | } | |
6569 | kfree(sched_group_nodes); | |
6570 | sched_group_nodes_bycpu[cpu] = NULL; | |
6571 | } | |
51888ca2 | 6572 | } |
a616058b SS |
6573 | #else |
6574 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6575 | { | |
6576 | } | |
6577 | #endif | |
51888ca2 | 6578 | |
89c4710e SS |
6579 | /* |
6580 | * Initialize sched groups cpu_power. | |
6581 | * | |
6582 | * cpu_power indicates the capacity of sched group, which is used while | |
6583 | * distributing the load between different sched groups in a sched domain. | |
6584 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6585 | * there are asymmetries in the topology. If there are asymmetries, group | |
6586 | * having more cpu_power will pickup more load compared to the group having | |
6587 | * less cpu_power. | |
6588 | * | |
6589 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
6590 | * the maximum number of tasks a group can handle in the presence of other idle | |
6591 | * or lightly loaded groups in the same sched domain. | |
6592 | */ | |
6593 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6594 | { | |
6595 | struct sched_domain *child; | |
6596 | struct sched_group *group; | |
6597 | ||
6598 | WARN_ON(!sd || !sd->groups); | |
6599 | ||
6600 | if (cpu != first_cpu(sd->groups->cpumask)) | |
6601 | return; | |
6602 | ||
6603 | child = sd->child; | |
6604 | ||
5517d86b ED |
6605 | sd->groups->__cpu_power = 0; |
6606 | ||
89c4710e SS |
6607 | /* |
6608 | * For perf policy, if the groups in child domain share resources | |
6609 | * (for example cores sharing some portions of the cache hierarchy | |
6610 | * or SMT), then set this domain groups cpu_power such that each group | |
6611 | * can handle only one task, when there are other idle groups in the | |
6612 | * same sched domain. | |
6613 | */ | |
6614 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
6615 | (child->flags & | |
6616 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 6617 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
6618 | return; |
6619 | } | |
6620 | ||
89c4710e SS |
6621 | /* |
6622 | * add cpu_power of each child group to this groups cpu_power | |
6623 | */ | |
6624 | group = child->groups; | |
6625 | do { | |
5517d86b | 6626 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
6627 | group = group->next; |
6628 | } while (group != child->groups); | |
6629 | } | |
6630 | ||
1da177e4 | 6631 | /* |
1a20ff27 DG |
6632 | * Build sched domains for a given set of cpus and attach the sched domains |
6633 | * to the individual cpus | |
1da177e4 | 6634 | */ |
51888ca2 | 6635 | static int build_sched_domains(const cpumask_t *cpu_map) |
1da177e4 LT |
6636 | { |
6637 | int i; | |
57d885fe | 6638 | struct root_domain *rd; |
d1b55138 JH |
6639 | #ifdef CONFIG_NUMA |
6640 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 6641 | int sd_allnodes = 0; |
d1b55138 JH |
6642 | |
6643 | /* | |
6644 | * Allocate the per-node list of sched groups | |
6645 | */ | |
5cf9f062 | 6646 | sched_group_nodes = kcalloc(MAX_NUMNODES, sizeof(struct sched_group *), |
41a2d6cf | 6647 | GFP_KERNEL); |
d1b55138 JH |
6648 | if (!sched_group_nodes) { |
6649 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 6650 | return -ENOMEM; |
d1b55138 JH |
6651 | } |
6652 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
6653 | #endif | |
1da177e4 | 6654 | |
dc938520 | 6655 | rd = alloc_rootdomain(); |
57d885fe GH |
6656 | if (!rd) { |
6657 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
6658 | return -ENOMEM; | |
6659 | } | |
6660 | ||
1da177e4 | 6661 | /* |
1a20ff27 | 6662 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 6663 | */ |
1a20ff27 | 6664 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6665 | struct sched_domain *sd = NULL, *p; |
6666 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(i)); | |
6667 | ||
1a20ff27 | 6668 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6669 | |
6670 | #ifdef CONFIG_NUMA | |
dd41f596 IM |
6671 | if (cpus_weight(*cpu_map) > |
6672 | SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) { | |
9c1cfda2 JH |
6673 | sd = &per_cpu(allnodes_domains, i); |
6674 | *sd = SD_ALLNODES_INIT; | |
6675 | sd->span = *cpu_map; | |
6711cab4 | 6676 | cpu_to_allnodes_group(i, cpu_map, &sd->groups); |
9c1cfda2 | 6677 | p = sd; |
6711cab4 | 6678 | sd_allnodes = 1; |
9c1cfda2 JH |
6679 | } else |
6680 | p = NULL; | |
6681 | ||
1da177e4 | 6682 | sd = &per_cpu(node_domains, i); |
1da177e4 | 6683 | *sd = SD_NODE_INIT; |
9c1cfda2 JH |
6684 | sd->span = sched_domain_node_span(cpu_to_node(i)); |
6685 | sd->parent = p; | |
1a848870 SS |
6686 | if (p) |
6687 | p->child = sd; | |
9c1cfda2 | 6688 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
6689 | #endif |
6690 | ||
6691 | p = sd; | |
6692 | sd = &per_cpu(phys_domains, i); | |
1da177e4 LT |
6693 | *sd = SD_CPU_INIT; |
6694 | sd->span = nodemask; | |
6695 | sd->parent = p; | |
1a848870 SS |
6696 | if (p) |
6697 | p->child = sd; | |
6711cab4 | 6698 | cpu_to_phys_group(i, cpu_map, &sd->groups); |
1da177e4 | 6699 | |
1e9f28fa SS |
6700 | #ifdef CONFIG_SCHED_MC |
6701 | p = sd; | |
6702 | sd = &per_cpu(core_domains, i); | |
1e9f28fa SS |
6703 | *sd = SD_MC_INIT; |
6704 | sd->span = cpu_coregroup_map(i); | |
6705 | cpus_and(sd->span, sd->span, *cpu_map); | |
6706 | sd->parent = p; | |
1a848870 | 6707 | p->child = sd; |
6711cab4 | 6708 | cpu_to_core_group(i, cpu_map, &sd->groups); |
1e9f28fa SS |
6709 | #endif |
6710 | ||
1da177e4 LT |
6711 | #ifdef CONFIG_SCHED_SMT |
6712 | p = sd; | |
6713 | sd = &per_cpu(cpu_domains, i); | |
1da177e4 | 6714 | *sd = SD_SIBLING_INIT; |
d5a7430d | 6715 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6716 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 6717 | sd->parent = p; |
1a848870 | 6718 | p->child = sd; |
6711cab4 | 6719 | cpu_to_cpu_group(i, cpu_map, &sd->groups); |
1da177e4 LT |
6720 | #endif |
6721 | } | |
6722 | ||
6723 | #ifdef CONFIG_SCHED_SMT | |
6724 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 6725 | for_each_cpu_mask(i, *cpu_map) { |
d5a7430d | 6726 | cpumask_t this_sibling_map = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6727 | cpus_and(this_sibling_map, this_sibling_map, *cpu_map); |
1da177e4 LT |
6728 | if (i != first_cpu(this_sibling_map)) |
6729 | continue; | |
6730 | ||
dd41f596 IM |
6731 | init_sched_build_groups(this_sibling_map, cpu_map, |
6732 | &cpu_to_cpu_group); | |
1da177e4 LT |
6733 | } |
6734 | #endif | |
6735 | ||
1e9f28fa SS |
6736 | #ifdef CONFIG_SCHED_MC |
6737 | /* Set up multi-core groups */ | |
6738 | for_each_cpu_mask(i, *cpu_map) { | |
6739 | cpumask_t this_core_map = cpu_coregroup_map(i); | |
6740 | cpus_and(this_core_map, this_core_map, *cpu_map); | |
6741 | if (i != first_cpu(this_core_map)) | |
6742 | continue; | |
dd41f596 IM |
6743 | init_sched_build_groups(this_core_map, cpu_map, |
6744 | &cpu_to_core_group); | |
1e9f28fa SS |
6745 | } |
6746 | #endif | |
6747 | ||
1da177e4 LT |
6748 | /* Set up physical groups */ |
6749 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6750 | cpumask_t nodemask = node_to_cpumask(i); | |
6751 | ||
1a20ff27 | 6752 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6753 | if (cpus_empty(nodemask)) |
6754 | continue; | |
6755 | ||
6711cab4 | 6756 | init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group); |
1da177e4 LT |
6757 | } |
6758 | ||
6759 | #ifdef CONFIG_NUMA | |
6760 | /* Set up node groups */ | |
6711cab4 | 6761 | if (sd_allnodes) |
dd41f596 IM |
6762 | init_sched_build_groups(*cpu_map, cpu_map, |
6763 | &cpu_to_allnodes_group); | |
9c1cfda2 JH |
6764 | |
6765 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6766 | /* Set up node groups */ | |
6767 | struct sched_group *sg, *prev; | |
6768 | cpumask_t nodemask = node_to_cpumask(i); | |
6769 | cpumask_t domainspan; | |
6770 | cpumask_t covered = CPU_MASK_NONE; | |
6771 | int j; | |
6772 | ||
6773 | cpus_and(nodemask, nodemask, *cpu_map); | |
d1b55138 JH |
6774 | if (cpus_empty(nodemask)) { |
6775 | sched_group_nodes[i] = NULL; | |
9c1cfda2 | 6776 | continue; |
d1b55138 | 6777 | } |
9c1cfda2 JH |
6778 | |
6779 | domainspan = sched_domain_node_span(i); | |
6780 | cpus_and(domainspan, domainspan, *cpu_map); | |
6781 | ||
15f0b676 | 6782 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
6783 | if (!sg) { |
6784 | printk(KERN_WARNING "Can not alloc domain group for " | |
6785 | "node %d\n", i); | |
6786 | goto error; | |
6787 | } | |
9c1cfda2 JH |
6788 | sched_group_nodes[i] = sg; |
6789 | for_each_cpu_mask(j, nodemask) { | |
6790 | struct sched_domain *sd; | |
9761eea8 | 6791 | |
9c1cfda2 JH |
6792 | sd = &per_cpu(node_domains, j); |
6793 | sd->groups = sg; | |
9c1cfda2 | 6794 | } |
5517d86b | 6795 | sg->__cpu_power = 0; |
9c1cfda2 | 6796 | sg->cpumask = nodemask; |
51888ca2 | 6797 | sg->next = sg; |
9c1cfda2 JH |
6798 | cpus_or(covered, covered, nodemask); |
6799 | prev = sg; | |
6800 | ||
6801 | for (j = 0; j < MAX_NUMNODES; j++) { | |
6802 | cpumask_t tmp, notcovered; | |
6803 | int n = (i + j) % MAX_NUMNODES; | |
6804 | ||
6805 | cpus_complement(notcovered, covered); | |
6806 | cpus_and(tmp, notcovered, *cpu_map); | |
6807 | cpus_and(tmp, tmp, domainspan); | |
6808 | if (cpus_empty(tmp)) | |
6809 | break; | |
6810 | ||
6811 | nodemask = node_to_cpumask(n); | |
6812 | cpus_and(tmp, tmp, nodemask); | |
6813 | if (cpus_empty(tmp)) | |
6814 | continue; | |
6815 | ||
15f0b676 SV |
6816 | sg = kmalloc_node(sizeof(struct sched_group), |
6817 | GFP_KERNEL, i); | |
9c1cfda2 JH |
6818 | if (!sg) { |
6819 | printk(KERN_WARNING | |
6820 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 6821 | goto error; |
9c1cfda2 | 6822 | } |
5517d86b | 6823 | sg->__cpu_power = 0; |
9c1cfda2 | 6824 | sg->cpumask = tmp; |
51888ca2 | 6825 | sg->next = prev->next; |
9c1cfda2 JH |
6826 | cpus_or(covered, covered, tmp); |
6827 | prev->next = sg; | |
6828 | prev = sg; | |
6829 | } | |
9c1cfda2 | 6830 | } |
1da177e4 LT |
6831 | #endif |
6832 | ||
6833 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 6834 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 6835 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6836 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
6837 | ||
89c4710e | 6838 | init_sched_groups_power(i, sd); |
5c45bf27 | 6839 | } |
1da177e4 | 6840 | #endif |
1e9f28fa | 6841 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 6842 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6843 | struct sched_domain *sd = &per_cpu(core_domains, i); |
6844 | ||
89c4710e | 6845 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
6846 | } |
6847 | #endif | |
1e9f28fa | 6848 | |
5c45bf27 | 6849 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6850 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
6851 | ||
89c4710e | 6852 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6853 | } |
6854 | ||
9c1cfda2 | 6855 | #ifdef CONFIG_NUMA |
08069033 SS |
6856 | for (i = 0; i < MAX_NUMNODES; i++) |
6857 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 6858 | |
6711cab4 SS |
6859 | if (sd_allnodes) { |
6860 | struct sched_group *sg; | |
f712c0c7 | 6861 | |
6711cab4 | 6862 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg); |
f712c0c7 SS |
6863 | init_numa_sched_groups_power(sg); |
6864 | } | |
9c1cfda2 JH |
6865 | #endif |
6866 | ||
1da177e4 | 6867 | /* Attach the domains */ |
1a20ff27 | 6868 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6869 | struct sched_domain *sd; |
6870 | #ifdef CONFIG_SCHED_SMT | |
6871 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
6872 | #elif defined(CONFIG_SCHED_MC) |
6873 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
6874 | #else |
6875 | sd = &per_cpu(phys_domains, i); | |
6876 | #endif | |
57d885fe | 6877 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 6878 | } |
51888ca2 SV |
6879 | |
6880 | return 0; | |
6881 | ||
a616058b | 6882 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6883 | error: |
6884 | free_sched_groups(cpu_map); | |
6885 | return -ENOMEM; | |
a616058b | 6886 | #endif |
1da177e4 | 6887 | } |
029190c5 PJ |
6888 | |
6889 | static cpumask_t *doms_cur; /* current sched domains */ | |
6890 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
6891 | ||
6892 | /* | |
6893 | * Special case: If a kmalloc of a doms_cur partition (array of | |
6894 | * cpumask_t) fails, then fallback to a single sched domain, | |
6895 | * as determined by the single cpumask_t fallback_doms. | |
6896 | */ | |
6897 | static cpumask_t fallback_doms; | |
6898 | ||
22e52b07 HC |
6899 | void __attribute__((weak)) arch_update_cpu_topology(void) |
6900 | { | |
6901 | } | |
6902 | ||
1a20ff27 | 6903 | /* |
41a2d6cf | 6904 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6905 | * For now this just excludes isolated cpus, but could be used to |
6906 | * exclude other special cases in the future. | |
1a20ff27 | 6907 | */ |
51888ca2 | 6908 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 6909 | { |
7378547f MM |
6910 | int err; |
6911 | ||
22e52b07 | 6912 | arch_update_cpu_topology(); |
029190c5 PJ |
6913 | ndoms_cur = 1; |
6914 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
6915 | if (!doms_cur) | |
6916 | doms_cur = &fallback_doms; | |
6917 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
7378547f | 6918 | err = build_sched_domains(doms_cur); |
6382bc90 | 6919 | register_sched_domain_sysctl(); |
7378547f MM |
6920 | |
6921 | return err; | |
1a20ff27 DG |
6922 | } |
6923 | ||
6924 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map) | |
1da177e4 | 6925 | { |
51888ca2 | 6926 | free_sched_groups(cpu_map); |
9c1cfda2 | 6927 | } |
1da177e4 | 6928 | |
1a20ff27 DG |
6929 | /* |
6930 | * Detach sched domains from a group of cpus specified in cpu_map | |
6931 | * These cpus will now be attached to the NULL domain | |
6932 | */ | |
858119e1 | 6933 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6934 | { |
6935 | int i; | |
6936 | ||
6382bc90 MM |
6937 | unregister_sched_domain_sysctl(); |
6938 | ||
1a20ff27 | 6939 | for_each_cpu_mask(i, *cpu_map) |
57d885fe | 6940 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 DG |
6941 | synchronize_sched(); |
6942 | arch_destroy_sched_domains(cpu_map); | |
6943 | } | |
6944 | ||
029190c5 PJ |
6945 | /* |
6946 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6947 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6948 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6949 | * It destroys each deleted domain and builds each new domain. | |
6950 | * | |
6951 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
41a2d6cf IM |
6952 | * The masks don't intersect (don't overlap.) We should setup one |
6953 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6954 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6955 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6956 | * it as it is. | |
6957 | * | |
41a2d6cf IM |
6958 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
6959 | * ownership of it and will kfree it when done with it. If the caller | |
029190c5 PJ |
6960 | * failed the kmalloc call, then it can pass in doms_new == NULL, |
6961 | * and partition_sched_domains() will fallback to the single partition | |
6962 | * 'fallback_doms'. | |
6963 | * | |
6964 | * Call with hotplug lock held | |
6965 | */ | |
6966 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new) | |
6967 | { | |
6968 | int i, j; | |
6969 | ||
a1835615 SV |
6970 | lock_doms_cur(); |
6971 | ||
7378547f MM |
6972 | /* always unregister in case we don't destroy any domains */ |
6973 | unregister_sched_domain_sysctl(); | |
6974 | ||
029190c5 PJ |
6975 | if (doms_new == NULL) { |
6976 | ndoms_new = 1; | |
6977 | doms_new = &fallback_doms; | |
6978 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
6979 | } | |
6980 | ||
6981 | /* Destroy deleted domains */ | |
6982 | for (i = 0; i < ndoms_cur; i++) { | |
6983 | for (j = 0; j < ndoms_new; j++) { | |
6984 | if (cpus_equal(doms_cur[i], doms_new[j])) | |
6985 | goto match1; | |
6986 | } | |
6987 | /* no match - a current sched domain not in new doms_new[] */ | |
6988 | detach_destroy_domains(doms_cur + i); | |
6989 | match1: | |
6990 | ; | |
6991 | } | |
6992 | ||
6993 | /* Build new domains */ | |
6994 | for (i = 0; i < ndoms_new; i++) { | |
6995 | for (j = 0; j < ndoms_cur; j++) { | |
6996 | if (cpus_equal(doms_new[i], doms_cur[j])) | |
6997 | goto match2; | |
6998 | } | |
6999 | /* no match - add a new doms_new */ | |
7000 | build_sched_domains(doms_new + i); | |
7001 | match2: | |
7002 | ; | |
7003 | } | |
7004 | ||
7005 | /* Remember the new sched domains */ | |
7006 | if (doms_cur != &fallback_doms) | |
7007 | kfree(doms_cur); | |
7008 | doms_cur = doms_new; | |
7009 | ndoms_cur = ndoms_new; | |
7378547f MM |
7010 | |
7011 | register_sched_domain_sysctl(); | |
a1835615 SV |
7012 | |
7013 | unlock_doms_cur(); | |
029190c5 PJ |
7014 | } |
7015 | ||
5c45bf27 | 7016 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
9aefd0ab | 7017 | int arch_reinit_sched_domains(void) |
5c45bf27 SS |
7018 | { |
7019 | int err; | |
7020 | ||
95402b38 | 7021 | get_online_cpus(); |
5c45bf27 SS |
7022 | detach_destroy_domains(&cpu_online_map); |
7023 | err = arch_init_sched_domains(&cpu_online_map); | |
95402b38 | 7024 | put_online_cpus(); |
5c45bf27 SS |
7025 | |
7026 | return err; | |
7027 | } | |
7028 | ||
7029 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7030 | { | |
7031 | int ret; | |
7032 | ||
7033 | if (buf[0] != '0' && buf[0] != '1') | |
7034 | return -EINVAL; | |
7035 | ||
7036 | if (smt) | |
7037 | sched_smt_power_savings = (buf[0] == '1'); | |
7038 | else | |
7039 | sched_mc_power_savings = (buf[0] == '1'); | |
7040 | ||
7041 | ret = arch_reinit_sched_domains(); | |
7042 | ||
7043 | return ret ? ret : count; | |
7044 | } | |
7045 | ||
5c45bf27 SS |
7046 | #ifdef CONFIG_SCHED_MC |
7047 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
7048 | { | |
7049 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7050 | } | |
48f24c4d IM |
7051 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
7052 | const char *buf, size_t count) | |
5c45bf27 SS |
7053 | { |
7054 | return sched_power_savings_store(buf, count, 0); | |
7055 | } | |
6707de00 AB |
7056 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
7057 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7058 | #endif |
7059 | ||
7060 | #ifdef CONFIG_SCHED_SMT | |
7061 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
7062 | { | |
7063 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7064 | } | |
48f24c4d IM |
7065 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
7066 | const char *buf, size_t count) | |
5c45bf27 SS |
7067 | { |
7068 | return sched_power_savings_store(buf, count, 1); | |
7069 | } | |
6707de00 AB |
7070 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
7071 | sched_smt_power_savings_store); | |
7072 | #endif | |
7073 | ||
7074 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
7075 | { | |
7076 | int err = 0; | |
7077 | ||
7078 | #ifdef CONFIG_SCHED_SMT | |
7079 | if (smt_capable()) | |
7080 | err = sysfs_create_file(&cls->kset.kobj, | |
7081 | &attr_sched_smt_power_savings.attr); | |
7082 | #endif | |
7083 | #ifdef CONFIG_SCHED_MC | |
7084 | if (!err && mc_capable()) | |
7085 | err = sysfs_create_file(&cls->kset.kobj, | |
7086 | &attr_sched_mc_power_savings.attr); | |
7087 | #endif | |
7088 | return err; | |
7089 | } | |
5c45bf27 SS |
7090 | #endif |
7091 | ||
1da177e4 | 7092 | /* |
41a2d6cf | 7093 | * Force a reinitialization of the sched domains hierarchy. The domains |
1da177e4 | 7094 | * and groups cannot be updated in place without racing with the balancing |
41c7ce9a | 7095 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
7096 | * which will prevent rebalancing while the sched domains are recalculated. |
7097 | */ | |
7098 | static int update_sched_domains(struct notifier_block *nfb, | |
7099 | unsigned long action, void *hcpu) | |
7100 | { | |
1da177e4 LT |
7101 | switch (action) { |
7102 | case CPU_UP_PREPARE: | |
8bb78442 | 7103 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 7104 | case CPU_DOWN_PREPARE: |
8bb78442 | 7105 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 7106 | detach_destroy_domains(&cpu_online_map); |
1da177e4 LT |
7107 | return NOTIFY_OK; |
7108 | ||
7109 | case CPU_UP_CANCELED: | |
8bb78442 | 7110 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 7111 | case CPU_DOWN_FAILED: |
8bb78442 | 7112 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7113 | case CPU_ONLINE: |
8bb78442 | 7114 | case CPU_ONLINE_FROZEN: |
1da177e4 | 7115 | case CPU_DEAD: |
8bb78442 | 7116 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
7117 | /* |
7118 | * Fall through and re-initialise the domains. | |
7119 | */ | |
7120 | break; | |
7121 | default: | |
7122 | return NOTIFY_DONE; | |
7123 | } | |
7124 | ||
7125 | /* The hotplug lock is already held by cpu_up/cpu_down */ | |
1a20ff27 | 7126 | arch_init_sched_domains(&cpu_online_map); |
1da177e4 LT |
7127 | |
7128 | return NOTIFY_OK; | |
7129 | } | |
1da177e4 LT |
7130 | |
7131 | void __init sched_init_smp(void) | |
7132 | { | |
5c1e1767 NP |
7133 | cpumask_t non_isolated_cpus; |
7134 | ||
95402b38 | 7135 | get_online_cpus(); |
1a20ff27 | 7136 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 7137 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
7138 | if (cpus_empty(non_isolated_cpus)) |
7139 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
95402b38 | 7140 | put_online_cpus(); |
1da177e4 LT |
7141 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7142 | hotcpu_notifier(update_sched_domains, 0); | |
5c1e1767 NP |
7143 | |
7144 | /* Move init over to a non-isolated CPU */ | |
7145 | if (set_cpus_allowed(current, non_isolated_cpus) < 0) | |
7146 | BUG(); | |
19978ca6 | 7147 | sched_init_granularity(); |
1da177e4 LT |
7148 | } |
7149 | #else | |
7150 | void __init sched_init_smp(void) | |
7151 | { | |
19978ca6 | 7152 | sched_init_granularity(); |
1da177e4 LT |
7153 | } |
7154 | #endif /* CONFIG_SMP */ | |
7155 | ||
7156 | int in_sched_functions(unsigned long addr) | |
7157 | { | |
1da177e4 LT |
7158 | return in_lock_functions(addr) || |
7159 | (addr >= (unsigned long)__sched_text_start | |
7160 | && addr < (unsigned long)__sched_text_end); | |
7161 | } | |
7162 | ||
a9957449 | 7163 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7164 | { |
7165 | cfs_rq->tasks_timeline = RB_ROOT; | |
dd41f596 IM |
7166 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7167 | cfs_rq->rq = rq; | |
7168 | #endif | |
67e9fb2a | 7169 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7170 | } |
7171 | ||
fa85ae24 PZ |
7172 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7173 | { | |
7174 | struct rt_prio_array *array; | |
7175 | int i; | |
7176 | ||
7177 | array = &rt_rq->active; | |
7178 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7179 | INIT_LIST_HEAD(array->queue + i); | |
7180 | __clear_bit(i, array->bitmap); | |
7181 | } | |
7182 | /* delimiter for bitsearch: */ | |
7183 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7184 | ||
052f1dc7 | 7185 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
48d5e258 PZ |
7186 | rt_rq->highest_prio = MAX_RT_PRIO; |
7187 | #endif | |
fa85ae24 PZ |
7188 | #ifdef CONFIG_SMP |
7189 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 PZ |
7190 | rt_rq->overloaded = 0; |
7191 | #endif | |
7192 | ||
7193 | rt_rq->rt_time = 0; | |
7194 | rt_rq->rt_throttled = 0; | |
6f505b16 | 7195 | |
052f1dc7 | 7196 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7197 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7198 | rt_rq->rq = rq; |
7199 | #endif | |
fa85ae24 PZ |
7200 | } |
7201 | ||
6f505b16 PZ |
7202 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7203 | static void init_tg_cfs_entry(struct rq *rq, struct task_group *tg, | |
7204 | struct cfs_rq *cfs_rq, struct sched_entity *se, | |
7205 | int cpu, int add) | |
7206 | { | |
7207 | tg->cfs_rq[cpu] = cfs_rq; | |
7208 | init_cfs_rq(cfs_rq, rq); | |
7209 | cfs_rq->tg = tg; | |
7210 | if (add) | |
7211 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7212 | ||
7213 | tg->se[cpu] = se; | |
7214 | se->cfs_rq = &rq->cfs; | |
7215 | se->my_q = cfs_rq; | |
7216 | se->load.weight = tg->shares; | |
7217 | se->load.inv_weight = div64_64(1ULL<<32, se->load.weight); | |
7218 | se->parent = NULL; | |
7219 | } | |
052f1dc7 | 7220 | #endif |
6f505b16 | 7221 | |
052f1dc7 | 7222 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
7223 | static void init_tg_rt_entry(struct rq *rq, struct task_group *tg, |
7224 | struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, | |
7225 | int cpu, int add) | |
7226 | { | |
7227 | tg->rt_rq[cpu] = rt_rq; | |
7228 | init_rt_rq(rt_rq, rq); | |
7229 | rt_rq->tg = tg; | |
7230 | rt_rq->rt_se = rt_se; | |
7231 | if (add) | |
7232 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
7233 | ||
7234 | tg->rt_se[cpu] = rt_se; | |
7235 | rt_se->rt_rq = &rq->rt; | |
7236 | rt_se->my_q = rt_rq; | |
7237 | rt_se->parent = NULL; | |
7238 | INIT_LIST_HEAD(&rt_se->run_list); | |
7239 | } | |
7240 | #endif | |
7241 | ||
1da177e4 LT |
7242 | void __init sched_init(void) |
7243 | { | |
476f3534 | 7244 | int highest_cpu = 0; |
dd41f596 IM |
7245 | int i, j; |
7246 | ||
57d885fe GH |
7247 | #ifdef CONFIG_SMP |
7248 | init_defrootdomain(); | |
7249 | #endif | |
7250 | ||
052f1dc7 | 7251 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 PZ |
7252 | list_add(&init_task_group.list, &task_groups); |
7253 | #endif | |
7254 | ||
0a945022 | 7255 | for_each_possible_cpu(i) { |
70b97a7f | 7256 | struct rq *rq; |
1da177e4 LT |
7257 | |
7258 | rq = cpu_rq(i); | |
7259 | spin_lock_init(&rq->lock); | |
fcb99371 | 7260 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 7261 | rq->nr_running = 0; |
dd41f596 IM |
7262 | rq->clock = 1; |
7263 | init_cfs_rq(&rq->cfs, rq); | |
6f505b16 | 7264 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7265 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 7266 | init_task_group.shares = init_task_group_load; |
6f505b16 PZ |
7267 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
7268 | init_tg_cfs_entry(rq, &init_task_group, | |
7269 | &per_cpu(init_cfs_rq, i), | |
7270 | &per_cpu(init_sched_entity, i), i, 1); | |
7271 | ||
052f1dc7 PZ |
7272 | #endif |
7273 | #ifdef CONFIG_RT_GROUP_SCHED | |
9f0c1e56 PZ |
7274 | init_task_group.rt_runtime = |
7275 | sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
6f505b16 PZ |
7276 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
7277 | init_tg_rt_entry(rq, &init_task_group, | |
7278 | &per_cpu(init_rt_rq, i), | |
7279 | &per_cpu(init_sched_rt_entity, i), i, 1); | |
dd41f596 | 7280 | #endif |
fa85ae24 | 7281 | rq->rt_period_expire = 0; |
48d5e258 | 7282 | rq->rt_throttled = 0; |
1da177e4 | 7283 | |
dd41f596 IM |
7284 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7285 | rq->cpu_load[j] = 0; | |
1da177e4 | 7286 | #ifdef CONFIG_SMP |
41c7ce9a | 7287 | rq->sd = NULL; |
57d885fe | 7288 | rq->rd = NULL; |
1da177e4 | 7289 | rq->active_balance = 0; |
dd41f596 | 7290 | rq->next_balance = jiffies; |
1da177e4 | 7291 | rq->push_cpu = 0; |
0a2966b4 | 7292 | rq->cpu = i; |
1da177e4 LT |
7293 | rq->migration_thread = NULL; |
7294 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 7295 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 7296 | #endif |
8f4d37ec | 7297 | init_rq_hrtick(rq); |
1da177e4 | 7298 | atomic_set(&rq->nr_iowait, 0); |
476f3534 | 7299 | highest_cpu = i; |
1da177e4 LT |
7300 | } |
7301 | ||
2dd73a4f | 7302 | set_load_weight(&init_task); |
b50f60ce | 7303 | |
e107be36 AK |
7304 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7305 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7306 | #endif | |
7307 | ||
c9819f45 | 7308 | #ifdef CONFIG_SMP |
476f3534 | 7309 | nr_cpu_ids = highest_cpu + 1; |
c9819f45 CL |
7310 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); |
7311 | #endif | |
7312 | ||
b50f60ce HC |
7313 | #ifdef CONFIG_RT_MUTEXES |
7314 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
7315 | #endif | |
7316 | ||
1da177e4 LT |
7317 | /* |
7318 | * The boot idle thread does lazy MMU switching as well: | |
7319 | */ | |
7320 | atomic_inc(&init_mm.mm_count); | |
7321 | enter_lazy_tlb(&init_mm, current); | |
7322 | ||
7323 | /* | |
7324 | * Make us the idle thread. Technically, schedule() should not be | |
7325 | * called from this thread, however somewhere below it might be, | |
7326 | * but because we are the idle thread, we just pick up running again | |
7327 | * when this runqueue becomes "idle". | |
7328 | */ | |
7329 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
7330 | /* |
7331 | * During early bootup we pretend to be a normal task: | |
7332 | */ | |
7333 | current->sched_class = &fair_sched_class; | |
6892b75e IM |
7334 | |
7335 | scheduler_running = 1; | |
1da177e4 LT |
7336 | } |
7337 | ||
7338 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
7339 | void __might_sleep(char *file, int line) | |
7340 | { | |
48f24c4d | 7341 | #ifdef in_atomic |
1da177e4 LT |
7342 | static unsigned long prev_jiffy; /* ratelimiting */ |
7343 | ||
7344 | if ((in_atomic() || irqs_disabled()) && | |
7345 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
7346 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7347 | return; | |
7348 | prev_jiffy = jiffies; | |
91368d73 | 7349 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
7350 | " context at %s:%d\n", file, line); |
7351 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
7352 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 7353 | debug_show_held_locks(current); |
3117df04 IM |
7354 | if (irqs_disabled()) |
7355 | print_irqtrace_events(current); | |
1da177e4 LT |
7356 | dump_stack(); |
7357 | } | |
7358 | #endif | |
7359 | } | |
7360 | EXPORT_SYMBOL(__might_sleep); | |
7361 | #endif | |
7362 | ||
7363 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7364 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7365 | { | |
7366 | int on_rq; | |
7367 | update_rq_clock(rq); | |
7368 | on_rq = p->se.on_rq; | |
7369 | if (on_rq) | |
7370 | deactivate_task(rq, p, 0); | |
7371 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
7372 | if (on_rq) { | |
7373 | activate_task(rq, p, 0); | |
7374 | resched_task(rq->curr); | |
7375 | } | |
7376 | } | |
7377 | ||
1da177e4 LT |
7378 | void normalize_rt_tasks(void) |
7379 | { | |
a0f98a1c | 7380 | struct task_struct *g, *p; |
1da177e4 | 7381 | unsigned long flags; |
70b97a7f | 7382 | struct rq *rq; |
1da177e4 | 7383 | |
4cf5d77a | 7384 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7385 | do_each_thread(g, p) { |
178be793 IM |
7386 | /* |
7387 | * Only normalize user tasks: | |
7388 | */ | |
7389 | if (!p->mm) | |
7390 | continue; | |
7391 | ||
6cfb0d5d | 7392 | p->se.exec_start = 0; |
6cfb0d5d | 7393 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 7394 | p->se.wait_start = 0; |
dd41f596 | 7395 | p->se.sleep_start = 0; |
dd41f596 | 7396 | p->se.block_start = 0; |
6cfb0d5d | 7397 | #endif |
dd41f596 IM |
7398 | task_rq(p)->clock = 0; |
7399 | ||
7400 | if (!rt_task(p)) { | |
7401 | /* | |
7402 | * Renice negative nice level userspace | |
7403 | * tasks back to 0: | |
7404 | */ | |
7405 | if (TASK_NICE(p) < 0 && p->mm) | |
7406 | set_user_nice(p, 0); | |
1da177e4 | 7407 | continue; |
dd41f596 | 7408 | } |
1da177e4 | 7409 | |
4cf5d77a | 7410 | spin_lock(&p->pi_lock); |
b29739f9 | 7411 | rq = __task_rq_lock(p); |
1da177e4 | 7412 | |
178be793 | 7413 | normalize_task(rq, p); |
3a5e4dc1 | 7414 | |
b29739f9 | 7415 | __task_rq_unlock(rq); |
4cf5d77a | 7416 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7417 | } while_each_thread(g, p); |
7418 | ||
4cf5d77a | 7419 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7420 | } |
7421 | ||
7422 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
7423 | |
7424 | #ifdef CONFIG_IA64 | |
7425 | /* | |
7426 | * These functions are only useful for the IA64 MCA handling. | |
7427 | * | |
7428 | * They can only be called when the whole system has been | |
7429 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7430 | * activity can take place. Using them for anything else would | |
7431 | * be a serious bug, and as a result, they aren't even visible | |
7432 | * under any other configuration. | |
7433 | */ | |
7434 | ||
7435 | /** | |
7436 | * curr_task - return the current task for a given cpu. | |
7437 | * @cpu: the processor in question. | |
7438 | * | |
7439 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7440 | */ | |
36c8b586 | 7441 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7442 | { |
7443 | return cpu_curr(cpu); | |
7444 | } | |
7445 | ||
7446 | /** | |
7447 | * set_curr_task - set the current task for a given cpu. | |
7448 | * @cpu: the processor in question. | |
7449 | * @p: the task pointer to set. | |
7450 | * | |
7451 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7452 | * are serviced on a separate stack. It allows the architecture to switch the |
7453 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7454 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7455 | * and caller must save the original value of the current task (see | |
7456 | * curr_task() above) and restore that value before reenabling interrupts and | |
7457 | * re-starting the system. | |
7458 | * | |
7459 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7460 | */ | |
36c8b586 | 7461 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7462 | { |
7463 | cpu_curr(cpu) = p; | |
7464 | } | |
7465 | ||
7466 | #endif | |
29f59db3 | 7467 | |
052f1dc7 | 7468 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 7469 | |
bccbe08a PZ |
7470 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7471 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
7472 | { |
7473 | int i; | |
7474 | ||
7475 | for_each_possible_cpu(i) { | |
7476 | if (tg->cfs_rq) | |
7477 | kfree(tg->cfs_rq[i]); | |
7478 | if (tg->se) | |
7479 | kfree(tg->se[i]); | |
6f505b16 PZ |
7480 | } |
7481 | ||
7482 | kfree(tg->cfs_rq); | |
7483 | kfree(tg->se); | |
6f505b16 PZ |
7484 | } |
7485 | ||
bccbe08a | 7486 | static int alloc_fair_sched_group(struct task_group *tg) |
29f59db3 | 7487 | { |
29f59db3 SV |
7488 | struct cfs_rq *cfs_rq; |
7489 | struct sched_entity *se; | |
9b5b7751 | 7490 | struct rq *rq; |
29f59db3 SV |
7491 | int i; |
7492 | ||
9b5b7751 | 7493 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
7494 | if (!tg->cfs_rq) |
7495 | goto err; | |
9b5b7751 | 7496 | tg->se = kzalloc(sizeof(se) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
7497 | if (!tg->se) |
7498 | goto err; | |
052f1dc7 PZ |
7499 | |
7500 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
7501 | |
7502 | for_each_possible_cpu(i) { | |
9b5b7751 | 7503 | rq = cpu_rq(i); |
29f59db3 | 7504 | |
6f505b16 PZ |
7505 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), |
7506 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
7507 | if (!cfs_rq) |
7508 | goto err; | |
7509 | ||
6f505b16 PZ |
7510 | se = kmalloc_node(sizeof(struct sched_entity), |
7511 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
7512 | if (!se) |
7513 | goto err; | |
7514 | ||
052f1dc7 | 7515 | init_tg_cfs_entry(rq, tg, cfs_rq, se, i, 0); |
bccbe08a PZ |
7516 | } |
7517 | ||
7518 | return 1; | |
7519 | ||
7520 | err: | |
7521 | return 0; | |
7522 | } | |
7523 | ||
7524 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
7525 | { | |
7526 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
7527 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
7528 | } | |
7529 | ||
7530 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
7531 | { | |
7532 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
7533 | } | |
7534 | #else | |
7535 | static inline void free_fair_sched_group(struct task_group *tg) | |
7536 | { | |
7537 | } | |
7538 | ||
7539 | static inline int alloc_fair_sched_group(struct task_group *tg) | |
7540 | { | |
7541 | return 1; | |
7542 | } | |
7543 | ||
7544 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
7545 | { | |
7546 | } | |
7547 | ||
7548 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
7549 | { | |
7550 | } | |
052f1dc7 PZ |
7551 | #endif |
7552 | ||
7553 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
7554 | static void free_rt_sched_group(struct task_group *tg) |
7555 | { | |
7556 | int i; | |
7557 | ||
7558 | for_each_possible_cpu(i) { | |
7559 | if (tg->rt_rq) | |
7560 | kfree(tg->rt_rq[i]); | |
7561 | if (tg->rt_se) | |
7562 | kfree(tg->rt_se[i]); | |
7563 | } | |
7564 | ||
7565 | kfree(tg->rt_rq); | |
7566 | kfree(tg->rt_se); | |
7567 | } | |
7568 | ||
7569 | static int alloc_rt_sched_group(struct task_group *tg) | |
7570 | { | |
7571 | struct rt_rq *rt_rq; | |
7572 | struct sched_rt_entity *rt_se; | |
7573 | struct rq *rq; | |
7574 | int i; | |
7575 | ||
7576 | tg->rt_rq = kzalloc(sizeof(rt_rq) * NR_CPUS, GFP_KERNEL); | |
7577 | if (!tg->rt_rq) | |
7578 | goto err; | |
7579 | tg->rt_se = kzalloc(sizeof(rt_se) * NR_CPUS, GFP_KERNEL); | |
7580 | if (!tg->rt_se) | |
7581 | goto err; | |
7582 | ||
7583 | tg->rt_runtime = 0; | |
7584 | ||
7585 | for_each_possible_cpu(i) { | |
7586 | rq = cpu_rq(i); | |
7587 | ||
6f505b16 PZ |
7588 | rt_rq = kmalloc_node(sizeof(struct rt_rq), |
7589 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
7590 | if (!rt_rq) | |
7591 | goto err; | |
29f59db3 | 7592 | |
6f505b16 PZ |
7593 | rt_se = kmalloc_node(sizeof(struct sched_rt_entity), |
7594 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
7595 | if (!rt_se) | |
7596 | goto err; | |
29f59db3 | 7597 | |
6f505b16 | 7598 | init_tg_rt_entry(rq, tg, rt_rq, rt_se, i, 0); |
29f59db3 SV |
7599 | } |
7600 | ||
bccbe08a PZ |
7601 | return 1; |
7602 | ||
7603 | err: | |
7604 | return 0; | |
7605 | } | |
7606 | ||
7607 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
7608 | { | |
7609 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
7610 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
7611 | } | |
7612 | ||
7613 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
7614 | { | |
7615 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
7616 | } | |
7617 | #else | |
7618 | static inline void free_rt_sched_group(struct task_group *tg) | |
7619 | { | |
7620 | } | |
7621 | ||
7622 | static inline int alloc_rt_sched_group(struct task_group *tg) | |
7623 | { | |
7624 | return 1; | |
7625 | } | |
7626 | ||
7627 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
7628 | { | |
7629 | } | |
7630 | ||
7631 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
7632 | { | |
7633 | } | |
7634 | #endif | |
7635 | ||
7636 | static void free_sched_group(struct task_group *tg) | |
7637 | { | |
7638 | free_fair_sched_group(tg); | |
7639 | free_rt_sched_group(tg); | |
7640 | kfree(tg); | |
7641 | } | |
7642 | ||
7643 | /* allocate runqueue etc for a new task group */ | |
7644 | struct task_group *sched_create_group(void) | |
7645 | { | |
7646 | struct task_group *tg; | |
7647 | unsigned long flags; | |
7648 | int i; | |
7649 | ||
7650 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7651 | if (!tg) | |
7652 | return ERR_PTR(-ENOMEM); | |
7653 | ||
7654 | if (!alloc_fair_sched_group(tg)) | |
7655 | goto err; | |
7656 | ||
7657 | if (!alloc_rt_sched_group(tg)) | |
7658 | goto err; | |
7659 | ||
8ed36996 | 7660 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 7661 | for_each_possible_cpu(i) { |
bccbe08a PZ |
7662 | register_fair_sched_group(tg, i); |
7663 | register_rt_sched_group(tg, i); | |
9b5b7751 | 7664 | } |
6f505b16 | 7665 | list_add_rcu(&tg->list, &task_groups); |
8ed36996 | 7666 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 7667 | |
9b5b7751 | 7668 | return tg; |
29f59db3 SV |
7669 | |
7670 | err: | |
6f505b16 | 7671 | free_sched_group(tg); |
29f59db3 SV |
7672 | return ERR_PTR(-ENOMEM); |
7673 | } | |
7674 | ||
9b5b7751 | 7675 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7676 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7677 | { |
29f59db3 | 7678 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7679 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7680 | } |
7681 | ||
9b5b7751 | 7682 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7683 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 7684 | { |
8ed36996 | 7685 | unsigned long flags; |
9b5b7751 | 7686 | int i; |
29f59db3 | 7687 | |
8ed36996 | 7688 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 7689 | for_each_possible_cpu(i) { |
bccbe08a PZ |
7690 | unregister_fair_sched_group(tg, i); |
7691 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 7692 | } |
6f505b16 | 7693 | list_del_rcu(&tg->list); |
8ed36996 | 7694 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 7695 | |
9b5b7751 | 7696 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 7697 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
7698 | } |
7699 | ||
9b5b7751 | 7700 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7701 | * The caller of this function should have put the task in its new group |
7702 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7703 | * reflect its new group. | |
9b5b7751 SV |
7704 | */ |
7705 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
7706 | { |
7707 | int on_rq, running; | |
7708 | unsigned long flags; | |
7709 | struct rq *rq; | |
7710 | ||
7711 | rq = task_rq_lock(tsk, &flags); | |
7712 | ||
29f59db3 SV |
7713 | update_rq_clock(rq); |
7714 | ||
051a1d1a | 7715 | running = task_current(rq, tsk); |
29f59db3 SV |
7716 | on_rq = tsk->se.on_rq; |
7717 | ||
0e1f3483 | 7718 | if (on_rq) |
29f59db3 | 7719 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
7720 | if (unlikely(running)) |
7721 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 7722 | |
6f505b16 | 7723 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 7724 | |
810b3817 PZ |
7725 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7726 | if (tsk->sched_class->moved_group) | |
7727 | tsk->sched_class->moved_group(tsk); | |
7728 | #endif | |
7729 | ||
0e1f3483 HS |
7730 | if (unlikely(running)) |
7731 | tsk->sched_class->set_curr_task(rq); | |
7732 | if (on_rq) | |
7074badb | 7733 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7734 | |
29f59db3 SV |
7735 | task_rq_unlock(rq, &flags); |
7736 | } | |
7737 | ||
052f1dc7 | 7738 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
7739 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
7740 | { | |
7741 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
7742 | struct rq *rq = cfs_rq->rq; | |
7743 | int on_rq; | |
7744 | ||
62fb1851 | 7745 | spin_lock_irq(&rq->lock); |
29f59db3 SV |
7746 | |
7747 | on_rq = se->on_rq; | |
62fb1851 | 7748 | if (on_rq) |
29f59db3 SV |
7749 | dequeue_entity(cfs_rq, se, 0); |
7750 | ||
7751 | se->load.weight = shares; | |
7752 | se->load.inv_weight = div64_64((1ULL<<32), shares); | |
7753 | ||
62fb1851 | 7754 | if (on_rq) |
29f59db3 | 7755 | enqueue_entity(cfs_rq, se, 0); |
62fb1851 PZ |
7756 | |
7757 | spin_unlock_irq(&rq->lock); | |
29f59db3 SV |
7758 | } |
7759 | ||
8ed36996 PZ |
7760 | static DEFINE_MUTEX(shares_mutex); |
7761 | ||
4cf86d77 | 7762 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
7763 | { |
7764 | int i; | |
8ed36996 | 7765 | unsigned long flags; |
c61935fd | 7766 | |
62fb1851 PZ |
7767 | /* |
7768 | * A weight of 0 or 1 can cause arithmetics problems. | |
7769 | * (The default weight is 1024 - so there's no practical | |
7770 | * limitation from this.) | |
7771 | */ | |
7772 | if (shares < 2) | |
7773 | shares = 2; | |
7774 | ||
8ed36996 | 7775 | mutex_lock(&shares_mutex); |
9b5b7751 | 7776 | if (tg->shares == shares) |
5cb350ba | 7777 | goto done; |
29f59db3 | 7778 | |
8ed36996 | 7779 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
7780 | for_each_possible_cpu(i) |
7781 | unregister_fair_sched_group(tg, i); | |
8ed36996 | 7782 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
7783 | |
7784 | /* wait for any ongoing reference to this group to finish */ | |
7785 | synchronize_sched(); | |
7786 | ||
7787 | /* | |
7788 | * Now we are free to modify the group's share on each cpu | |
7789 | * w/o tripping rebalance_share or load_balance_fair. | |
7790 | */ | |
9b5b7751 | 7791 | tg->shares = shares; |
62fb1851 | 7792 | for_each_possible_cpu(i) |
9b5b7751 | 7793 | set_se_shares(tg->se[i], shares); |
29f59db3 | 7794 | |
6b2d7700 SV |
7795 | /* |
7796 | * Enable load balance activity on this group, by inserting it back on | |
7797 | * each cpu's rq->leaf_cfs_rq_list. | |
7798 | */ | |
8ed36996 | 7799 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
7800 | for_each_possible_cpu(i) |
7801 | register_fair_sched_group(tg, i); | |
8ed36996 | 7802 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 7803 | done: |
8ed36996 | 7804 | mutex_unlock(&shares_mutex); |
9b5b7751 | 7805 | return 0; |
29f59db3 SV |
7806 | } |
7807 | ||
5cb350ba DG |
7808 | unsigned long sched_group_shares(struct task_group *tg) |
7809 | { | |
7810 | return tg->shares; | |
7811 | } | |
052f1dc7 | 7812 | #endif |
5cb350ba | 7813 | |
052f1dc7 | 7814 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7815 | /* |
9f0c1e56 | 7816 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 7817 | */ |
9f0c1e56 PZ |
7818 | static DEFINE_MUTEX(rt_constraints_mutex); |
7819 | ||
7820 | static unsigned long to_ratio(u64 period, u64 runtime) | |
7821 | { | |
7822 | if (runtime == RUNTIME_INF) | |
7823 | return 1ULL << 16; | |
7824 | ||
2692a240 | 7825 | return div64_64(runtime << 16, period); |
9f0c1e56 PZ |
7826 | } |
7827 | ||
7828 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) | |
6f505b16 PZ |
7829 | { |
7830 | struct task_group *tgi; | |
7831 | unsigned long total = 0; | |
9f0c1e56 PZ |
7832 | unsigned long global_ratio = |
7833 | to_ratio(sysctl_sched_rt_period, | |
7834 | sysctl_sched_rt_runtime < 0 ? | |
7835 | RUNTIME_INF : sysctl_sched_rt_runtime); | |
6f505b16 PZ |
7836 | |
7837 | rcu_read_lock(); | |
9f0c1e56 PZ |
7838 | list_for_each_entry_rcu(tgi, &task_groups, list) { |
7839 | if (tgi == tg) | |
7840 | continue; | |
6f505b16 | 7841 | |
9f0c1e56 PZ |
7842 | total += to_ratio(period, tgi->rt_runtime); |
7843 | } | |
7844 | rcu_read_unlock(); | |
6f505b16 | 7845 | |
9f0c1e56 | 7846 | return total + to_ratio(period, runtime) < global_ratio; |
6f505b16 PZ |
7847 | } |
7848 | ||
521f1a24 DG |
7849 | /* Must be called with tasklist_lock held */ |
7850 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
7851 | { | |
7852 | struct task_struct *g, *p; | |
7853 | do_each_thread(g, p) { | |
7854 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
7855 | return 1; | |
7856 | } while_each_thread(g, p); | |
7857 | return 0; | |
7858 | } | |
7859 | ||
9f0c1e56 | 7860 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
6f505b16 | 7861 | { |
9f0c1e56 PZ |
7862 | u64 rt_runtime, rt_period; |
7863 | int err = 0; | |
7864 | ||
2692a240 | 7865 | rt_period = (u64)sysctl_sched_rt_period * NSEC_PER_USEC; |
9f0c1e56 PZ |
7866 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; |
7867 | if (rt_runtime_us == -1) | |
2692a240 | 7868 | rt_runtime = RUNTIME_INF; |
9f0c1e56 PZ |
7869 | |
7870 | mutex_lock(&rt_constraints_mutex); | |
521f1a24 DG |
7871 | read_lock(&tasklist_lock); |
7872 | if (rt_runtime_us == 0 && tg_has_rt_tasks(tg)) { | |
7873 | err = -EBUSY; | |
7874 | goto unlock; | |
7875 | } | |
9f0c1e56 PZ |
7876 | if (!__rt_schedulable(tg, rt_period, rt_runtime)) { |
7877 | err = -EINVAL; | |
7878 | goto unlock; | |
7879 | } | |
9f0c1e56 PZ |
7880 | tg->rt_runtime = rt_runtime; |
7881 | unlock: | |
521f1a24 | 7882 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7883 | mutex_unlock(&rt_constraints_mutex); |
7884 | ||
7885 | return err; | |
6f505b16 PZ |
7886 | } |
7887 | ||
9f0c1e56 PZ |
7888 | long sched_group_rt_runtime(struct task_group *tg) |
7889 | { | |
7890 | u64 rt_runtime_us; | |
7891 | ||
7892 | if (tg->rt_runtime == RUNTIME_INF) | |
7893 | return -1; | |
7894 | ||
7895 | rt_runtime_us = tg->rt_runtime; | |
7896 | do_div(rt_runtime_us, NSEC_PER_USEC); | |
7897 | return rt_runtime_us; | |
7898 | } | |
052f1dc7 PZ |
7899 | #endif |
7900 | #endif /* CONFIG_GROUP_SCHED */ | |
68318b8e | 7901 | |
052f1dc7 | 7902 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
7903 | |
7904 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 7905 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 7906 | { |
2b01dfe3 PM |
7907 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
7908 | struct task_group, css); | |
68318b8e SV |
7909 | } |
7910 | ||
7911 | static struct cgroup_subsys_state * | |
2b01dfe3 | 7912 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e SV |
7913 | { |
7914 | struct task_group *tg; | |
7915 | ||
2b01dfe3 | 7916 | if (!cgrp->parent) { |
68318b8e | 7917 | /* This is early initialization for the top cgroup */ |
2b01dfe3 | 7918 | init_task_group.css.cgroup = cgrp; |
68318b8e SV |
7919 | return &init_task_group.css; |
7920 | } | |
7921 | ||
7922 | /* we support only 1-level deep hierarchical scheduler atm */ | |
2b01dfe3 | 7923 | if (cgrp->parent->parent) |
68318b8e SV |
7924 | return ERR_PTR(-EINVAL); |
7925 | ||
7926 | tg = sched_create_group(); | |
7927 | if (IS_ERR(tg)) | |
7928 | return ERR_PTR(-ENOMEM); | |
7929 | ||
7930 | /* Bind the cgroup to task_group object we just created */ | |
2b01dfe3 | 7931 | tg->css.cgroup = cgrp; |
68318b8e SV |
7932 | |
7933 | return &tg->css; | |
7934 | } | |
7935 | ||
41a2d6cf IM |
7936 | static void |
7937 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 7938 | { |
2b01dfe3 | 7939 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7940 | |
7941 | sched_destroy_group(tg); | |
7942 | } | |
7943 | ||
41a2d6cf IM |
7944 | static int |
7945 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
7946 | struct task_struct *tsk) | |
68318b8e | 7947 | { |
b68aa230 PZ |
7948 | #ifdef CONFIG_RT_GROUP_SCHED |
7949 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7950 | if (rt_task(tsk) && cgroup_tg(cgrp)->rt_runtime == 0) | |
7951 | return -EINVAL; | |
7952 | #else | |
68318b8e SV |
7953 | /* We don't support RT-tasks being in separate groups */ |
7954 | if (tsk->sched_class != &fair_sched_class) | |
7955 | return -EINVAL; | |
b68aa230 | 7956 | #endif |
68318b8e SV |
7957 | |
7958 | return 0; | |
7959 | } | |
7960 | ||
7961 | static void | |
2b01dfe3 | 7962 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
7963 | struct cgroup *old_cont, struct task_struct *tsk) |
7964 | { | |
7965 | sched_move_task(tsk); | |
7966 | } | |
7967 | ||
052f1dc7 | 7968 | #ifdef CONFIG_FAIR_GROUP_SCHED |
2b01dfe3 PM |
7969 | static int cpu_shares_write_uint(struct cgroup *cgrp, struct cftype *cftype, |
7970 | u64 shareval) | |
68318b8e | 7971 | { |
2b01dfe3 | 7972 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
7973 | } |
7974 | ||
2b01dfe3 | 7975 | static u64 cpu_shares_read_uint(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 7976 | { |
2b01dfe3 | 7977 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7978 | |
7979 | return (u64) tg->shares; | |
7980 | } | |
052f1dc7 | 7981 | #endif |
68318b8e | 7982 | |
052f1dc7 | 7983 | #ifdef CONFIG_RT_GROUP_SCHED |
9f0c1e56 PZ |
7984 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
7985 | struct file *file, | |
7986 | const char __user *userbuf, | |
7987 | size_t nbytes, loff_t *unused_ppos) | |
6f505b16 | 7988 | { |
9f0c1e56 PZ |
7989 | char buffer[64]; |
7990 | int retval = 0; | |
7991 | s64 val; | |
7992 | char *end; | |
7993 | ||
7994 | if (!nbytes) | |
7995 | return -EINVAL; | |
7996 | if (nbytes >= sizeof(buffer)) | |
7997 | return -E2BIG; | |
7998 | if (copy_from_user(buffer, userbuf, nbytes)) | |
7999 | return -EFAULT; | |
8000 | ||
8001 | buffer[nbytes] = 0; /* nul-terminate */ | |
8002 | ||
8003 | /* strip newline if necessary */ | |
8004 | if (nbytes && (buffer[nbytes-1] == '\n')) | |
8005 | buffer[nbytes-1] = 0; | |
8006 | val = simple_strtoll(buffer, &end, 0); | |
8007 | if (*end) | |
8008 | return -EINVAL; | |
8009 | ||
8010 | /* Pass to subsystem */ | |
8011 | retval = sched_group_set_rt_runtime(cgroup_tg(cgrp), val); | |
8012 | if (!retval) | |
8013 | retval = nbytes; | |
8014 | return retval; | |
6f505b16 PZ |
8015 | } |
8016 | ||
9f0c1e56 PZ |
8017 | static ssize_t cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft, |
8018 | struct file *file, | |
8019 | char __user *buf, size_t nbytes, | |
8020 | loff_t *ppos) | |
6f505b16 | 8021 | { |
9f0c1e56 PZ |
8022 | char tmp[64]; |
8023 | long val = sched_group_rt_runtime(cgroup_tg(cgrp)); | |
8024 | int len = sprintf(tmp, "%ld\n", val); | |
6f505b16 | 8025 | |
9f0c1e56 | 8026 | return simple_read_from_buffer(buf, nbytes, ppos, tmp, len); |
6f505b16 | 8027 | } |
052f1dc7 | 8028 | #endif |
6f505b16 | 8029 | |
fe5c7cc2 | 8030 | static struct cftype cpu_files[] = { |
052f1dc7 | 8031 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8032 | { |
8033 | .name = "shares", | |
8034 | .read_uint = cpu_shares_read_uint, | |
8035 | .write_uint = cpu_shares_write_uint, | |
8036 | }, | |
052f1dc7 PZ |
8037 | #endif |
8038 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8039 | { |
9f0c1e56 PZ |
8040 | .name = "rt_runtime_us", |
8041 | .read = cpu_rt_runtime_read, | |
8042 | .write = cpu_rt_runtime_write, | |
6f505b16 | 8043 | }, |
052f1dc7 | 8044 | #endif |
68318b8e SV |
8045 | }; |
8046 | ||
8047 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8048 | { | |
fe5c7cc2 | 8049 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8050 | } |
8051 | ||
8052 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8053 | .name = "cpu", |
8054 | .create = cpu_cgroup_create, | |
8055 | .destroy = cpu_cgroup_destroy, | |
8056 | .can_attach = cpu_cgroup_can_attach, | |
8057 | .attach = cpu_cgroup_attach, | |
8058 | .populate = cpu_cgroup_populate, | |
8059 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8060 | .early_init = 1, |
8061 | }; | |
8062 | ||
052f1dc7 | 8063 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8064 | |
8065 | #ifdef CONFIG_CGROUP_CPUACCT | |
8066 | ||
8067 | /* | |
8068 | * CPU accounting code for task groups. | |
8069 | * | |
8070 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8071 | * (balbir@in.ibm.com). | |
8072 | */ | |
8073 | ||
8074 | /* track cpu usage of a group of tasks */ | |
8075 | struct cpuacct { | |
8076 | struct cgroup_subsys_state css; | |
8077 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
8078 | u64 *cpuusage; | |
8079 | }; | |
8080 | ||
8081 | struct cgroup_subsys cpuacct_subsys; | |
8082 | ||
8083 | /* return cpu accounting group corresponding to this container */ | |
8084 | static inline struct cpuacct *cgroup_ca(struct cgroup *cont) | |
8085 | { | |
8086 | return container_of(cgroup_subsys_state(cont, cpuacct_subsys_id), | |
8087 | struct cpuacct, css); | |
8088 | } | |
8089 | ||
8090 | /* return cpu accounting group to which this task belongs */ | |
8091 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8092 | { | |
8093 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8094 | struct cpuacct, css); | |
8095 | } | |
8096 | ||
8097 | /* create a new cpu accounting group */ | |
8098 | static struct cgroup_subsys_state *cpuacct_create( | |
8099 | struct cgroup_subsys *ss, struct cgroup *cont) | |
8100 | { | |
8101 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
8102 | ||
8103 | if (!ca) | |
8104 | return ERR_PTR(-ENOMEM); | |
8105 | ||
8106 | ca->cpuusage = alloc_percpu(u64); | |
8107 | if (!ca->cpuusage) { | |
8108 | kfree(ca); | |
8109 | return ERR_PTR(-ENOMEM); | |
8110 | } | |
8111 | ||
8112 | return &ca->css; | |
8113 | } | |
8114 | ||
8115 | /* destroy an existing cpu accounting group */ | |
41a2d6cf IM |
8116 | static void |
8117 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cont) | |
d842de87 SV |
8118 | { |
8119 | struct cpuacct *ca = cgroup_ca(cont); | |
8120 | ||
8121 | free_percpu(ca->cpuusage); | |
8122 | kfree(ca); | |
8123 | } | |
8124 | ||
8125 | /* return total cpu usage (in nanoseconds) of a group */ | |
8126 | static u64 cpuusage_read(struct cgroup *cont, struct cftype *cft) | |
8127 | { | |
8128 | struct cpuacct *ca = cgroup_ca(cont); | |
8129 | u64 totalcpuusage = 0; | |
8130 | int i; | |
8131 | ||
8132 | for_each_possible_cpu(i) { | |
8133 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
8134 | ||
8135 | /* | |
8136 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
8137 | * platforms. | |
8138 | */ | |
8139 | spin_lock_irq(&cpu_rq(i)->lock); | |
8140 | totalcpuusage += *cpuusage; | |
8141 | spin_unlock_irq(&cpu_rq(i)->lock); | |
8142 | } | |
8143 | ||
8144 | return totalcpuusage; | |
8145 | } | |
8146 | ||
8147 | static struct cftype files[] = { | |
8148 | { | |
8149 | .name = "usage", | |
8150 | .read_uint = cpuusage_read, | |
8151 | }, | |
8152 | }; | |
8153 | ||
8154 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8155 | { | |
8156 | return cgroup_add_files(cont, ss, files, ARRAY_SIZE(files)); | |
8157 | } | |
8158 | ||
8159 | /* | |
8160 | * charge this task's execution time to its accounting group. | |
8161 | * | |
8162 | * called with rq->lock held. | |
8163 | */ | |
8164 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
8165 | { | |
8166 | struct cpuacct *ca; | |
8167 | ||
8168 | if (!cpuacct_subsys.active) | |
8169 | return; | |
8170 | ||
8171 | ca = task_ca(tsk); | |
8172 | if (ca) { | |
8173 | u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk)); | |
8174 | ||
8175 | *cpuusage += cputime; | |
8176 | } | |
8177 | } | |
8178 | ||
8179 | struct cgroup_subsys cpuacct_subsys = { | |
8180 | .name = "cpuacct", | |
8181 | .create = cpuacct_create, | |
8182 | .destroy = cpuacct_destroy, | |
8183 | .populate = cpuacct_populate, | |
8184 | .subsys_id = cpuacct_subsys_id, | |
8185 | }; | |
8186 | #endif /* CONFIG_CGROUP_CPUACCT */ |