]>
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 | */ | |
304 | struct sched_entity *curr; | |
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, | |
41a2d6cf IM |
597 | SCHED_FEAT_TREE_AVG = 8, |
598 | SCHED_FEAT_APPROX_AVG = 16, | |
8f4d37ec PZ |
599 | SCHED_FEAT_HRTICK = 32, |
600 | SCHED_FEAT_DOUBLE_TICK = 64, | |
bf5c91ba IM |
601 | }; |
602 | ||
603 | const_debug unsigned int sysctl_sched_features = | |
8401f775 | 604 | SCHED_FEAT_NEW_FAIR_SLEEPERS * 1 | |
9612633a | 605 | SCHED_FEAT_WAKEUP_PREEMPT * 1 | |
8401f775 IM |
606 | SCHED_FEAT_START_DEBIT * 1 | |
607 | SCHED_FEAT_TREE_AVG * 0 | | |
8f4d37ec PZ |
608 | SCHED_FEAT_APPROX_AVG * 0 | |
609 | SCHED_FEAT_HRTICK * 1 | | |
610 | SCHED_FEAT_DOUBLE_TICK * 0; | |
bf5c91ba IM |
611 | |
612 | #define sched_feat(x) (sysctl_sched_features & SCHED_FEAT_##x) | |
613 | ||
b82d9fdd PZ |
614 | /* |
615 | * Number of tasks to iterate in a single balance run. | |
616 | * Limited because this is done with IRQs disabled. | |
617 | */ | |
618 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
619 | ||
fa85ae24 | 620 | /* |
9f0c1e56 | 621 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
622 | * default: 1s |
623 | */ | |
9f0c1e56 | 624 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 625 | |
6892b75e IM |
626 | static __read_mostly int scheduler_running; |
627 | ||
9f0c1e56 PZ |
628 | /* |
629 | * part of the period that we allow rt tasks to run in us. | |
630 | * default: 0.95s | |
631 | */ | |
632 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 PZ |
633 | |
634 | /* | |
9f0c1e56 | 635 | * single value that denotes runtime == period, ie unlimited time. |
fa85ae24 | 636 | */ |
9f0c1e56 | 637 | #define RUNTIME_INF ((u64)~0ULL) |
fa85ae24 | 638 | |
e436d800 IM |
639 | /* |
640 | * For kernel-internal use: high-speed (but slightly incorrect) per-cpu | |
641 | * clock constructed from sched_clock(): | |
642 | */ | |
643 | unsigned long long cpu_clock(int cpu) | |
644 | { | |
e436d800 IM |
645 | unsigned long long now; |
646 | unsigned long flags; | |
b04a0f4c | 647 | struct rq *rq; |
e436d800 | 648 | |
8ced5f69 IM |
649 | /* |
650 | * Only call sched_clock() if the scheduler has already been | |
651 | * initialized (some code might call cpu_clock() very early): | |
652 | */ | |
6892b75e IM |
653 | if (unlikely(!scheduler_running)) |
654 | return 0; | |
655 | ||
656 | local_irq_save(flags); | |
657 | rq = cpu_rq(cpu); | |
658 | update_rq_clock(rq); | |
b04a0f4c | 659 | now = rq->clock; |
2cd4d0ea | 660 | local_irq_restore(flags); |
e436d800 IM |
661 | |
662 | return now; | |
663 | } | |
a58f6f25 | 664 | EXPORT_SYMBOL_GPL(cpu_clock); |
e436d800 | 665 | |
1da177e4 | 666 | #ifndef prepare_arch_switch |
4866cde0 NP |
667 | # define prepare_arch_switch(next) do { } while (0) |
668 | #endif | |
669 | #ifndef finish_arch_switch | |
670 | # define finish_arch_switch(prev) do { } while (0) | |
671 | #endif | |
672 | ||
051a1d1a DA |
673 | static inline int task_current(struct rq *rq, struct task_struct *p) |
674 | { | |
675 | return rq->curr == p; | |
676 | } | |
677 | ||
4866cde0 | 678 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 679 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 680 | { |
051a1d1a | 681 | return task_current(rq, p); |
4866cde0 NP |
682 | } |
683 | ||
70b97a7f | 684 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
685 | { |
686 | } | |
687 | ||
70b97a7f | 688 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 689 | { |
da04c035 IM |
690 | #ifdef CONFIG_DEBUG_SPINLOCK |
691 | /* this is a valid case when another task releases the spinlock */ | |
692 | rq->lock.owner = current; | |
693 | #endif | |
8a25d5de IM |
694 | /* |
695 | * If we are tracking spinlock dependencies then we have to | |
696 | * fix up the runqueue lock - which gets 'carried over' from | |
697 | * prev into current: | |
698 | */ | |
699 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
700 | ||
4866cde0 NP |
701 | spin_unlock_irq(&rq->lock); |
702 | } | |
703 | ||
704 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 705 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
706 | { |
707 | #ifdef CONFIG_SMP | |
708 | return p->oncpu; | |
709 | #else | |
051a1d1a | 710 | return task_current(rq, p); |
4866cde0 NP |
711 | #endif |
712 | } | |
713 | ||
70b97a7f | 714 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
715 | { |
716 | #ifdef CONFIG_SMP | |
717 | /* | |
718 | * We can optimise this out completely for !SMP, because the | |
719 | * SMP rebalancing from interrupt is the only thing that cares | |
720 | * here. | |
721 | */ | |
722 | next->oncpu = 1; | |
723 | #endif | |
724 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
725 | spin_unlock_irq(&rq->lock); | |
726 | #else | |
727 | spin_unlock(&rq->lock); | |
728 | #endif | |
729 | } | |
730 | ||
70b97a7f | 731 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
732 | { |
733 | #ifdef CONFIG_SMP | |
734 | /* | |
735 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
736 | * We must ensure this doesn't happen until the switch is completely | |
737 | * finished. | |
738 | */ | |
739 | smp_wmb(); | |
740 | prev->oncpu = 0; | |
741 | #endif | |
742 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
743 | local_irq_enable(); | |
1da177e4 | 744 | #endif |
4866cde0 NP |
745 | } |
746 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 747 | |
b29739f9 IM |
748 | /* |
749 | * __task_rq_lock - lock the runqueue a given task resides on. | |
750 | * Must be called interrupts disabled. | |
751 | */ | |
70b97a7f | 752 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
753 | __acquires(rq->lock) |
754 | { | |
3a5c359a AK |
755 | for (;;) { |
756 | struct rq *rq = task_rq(p); | |
757 | spin_lock(&rq->lock); | |
758 | if (likely(rq == task_rq(p))) | |
759 | return rq; | |
b29739f9 | 760 | spin_unlock(&rq->lock); |
b29739f9 | 761 | } |
b29739f9 IM |
762 | } |
763 | ||
1da177e4 LT |
764 | /* |
765 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 766 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
767 | * explicitly disabling preemption. |
768 | */ | |
70b97a7f | 769 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
770 | __acquires(rq->lock) |
771 | { | |
70b97a7f | 772 | struct rq *rq; |
1da177e4 | 773 | |
3a5c359a AK |
774 | for (;;) { |
775 | local_irq_save(*flags); | |
776 | rq = task_rq(p); | |
777 | spin_lock(&rq->lock); | |
778 | if (likely(rq == task_rq(p))) | |
779 | return rq; | |
1da177e4 | 780 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 781 | } |
1da177e4 LT |
782 | } |
783 | ||
a9957449 | 784 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
785 | __releases(rq->lock) |
786 | { | |
787 | spin_unlock(&rq->lock); | |
788 | } | |
789 | ||
70b97a7f | 790 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
791 | __releases(rq->lock) |
792 | { | |
793 | spin_unlock_irqrestore(&rq->lock, *flags); | |
794 | } | |
795 | ||
1da177e4 | 796 | /* |
cc2a73b5 | 797 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 798 | */ |
a9957449 | 799 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
800 | __acquires(rq->lock) |
801 | { | |
70b97a7f | 802 | struct rq *rq; |
1da177e4 LT |
803 | |
804 | local_irq_disable(); | |
805 | rq = this_rq(); | |
806 | spin_lock(&rq->lock); | |
807 | ||
808 | return rq; | |
809 | } | |
810 | ||
1b9f19c2 | 811 | /* |
2aa44d05 | 812 | * We are going deep-idle (irqs are disabled): |
1b9f19c2 | 813 | */ |
2aa44d05 | 814 | void sched_clock_idle_sleep_event(void) |
1b9f19c2 | 815 | { |
2aa44d05 IM |
816 | struct rq *rq = cpu_rq(smp_processor_id()); |
817 | ||
818 | spin_lock(&rq->lock); | |
819 | __update_rq_clock(rq); | |
820 | spin_unlock(&rq->lock); | |
821 | rq->clock_deep_idle_events++; | |
822 | } | |
823 | EXPORT_SYMBOL_GPL(sched_clock_idle_sleep_event); | |
824 | ||
825 | /* | |
826 | * We just idled delta nanoseconds (called with irqs disabled): | |
827 | */ | |
828 | void sched_clock_idle_wakeup_event(u64 delta_ns) | |
829 | { | |
830 | struct rq *rq = cpu_rq(smp_processor_id()); | |
831 | u64 now = sched_clock(); | |
1b9f19c2 | 832 | |
2aa44d05 IM |
833 | rq->idle_clock += delta_ns; |
834 | /* | |
835 | * Override the previous timestamp and ignore all | |
836 | * sched_clock() deltas that occured while we idled, | |
837 | * and use the PM-provided delta_ns to advance the | |
838 | * rq clock: | |
839 | */ | |
840 | spin_lock(&rq->lock); | |
841 | rq->prev_clock_raw = now; | |
842 | rq->clock += delta_ns; | |
843 | spin_unlock(&rq->lock); | |
782daeee | 844 | touch_softlockup_watchdog(); |
1b9f19c2 | 845 | } |
2aa44d05 | 846 | EXPORT_SYMBOL_GPL(sched_clock_idle_wakeup_event); |
1b9f19c2 | 847 | |
8f4d37ec PZ |
848 | static void __resched_task(struct task_struct *p, int tif_bit); |
849 | ||
850 | static inline void resched_task(struct task_struct *p) | |
851 | { | |
852 | __resched_task(p, TIF_NEED_RESCHED); | |
853 | } | |
854 | ||
855 | #ifdef CONFIG_SCHED_HRTICK | |
856 | /* | |
857 | * Use HR-timers to deliver accurate preemption points. | |
858 | * | |
859 | * Its all a bit involved since we cannot program an hrt while holding the | |
860 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
861 | * reschedule event. | |
862 | * | |
863 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
864 | * rq->lock. | |
865 | */ | |
866 | static inline void resched_hrt(struct task_struct *p) | |
867 | { | |
868 | __resched_task(p, TIF_HRTICK_RESCHED); | |
869 | } | |
870 | ||
871 | static inline void resched_rq(struct rq *rq) | |
872 | { | |
873 | unsigned long flags; | |
874 | ||
875 | spin_lock_irqsave(&rq->lock, flags); | |
876 | resched_task(rq->curr); | |
877 | spin_unlock_irqrestore(&rq->lock, flags); | |
878 | } | |
879 | ||
880 | enum { | |
881 | HRTICK_SET, /* re-programm hrtick_timer */ | |
882 | HRTICK_RESET, /* not a new slice */ | |
883 | }; | |
884 | ||
885 | /* | |
886 | * Use hrtick when: | |
887 | * - enabled by features | |
888 | * - hrtimer is actually high res | |
889 | */ | |
890 | static inline int hrtick_enabled(struct rq *rq) | |
891 | { | |
892 | if (!sched_feat(HRTICK)) | |
893 | return 0; | |
894 | return hrtimer_is_hres_active(&rq->hrtick_timer); | |
895 | } | |
896 | ||
897 | /* | |
898 | * Called to set the hrtick timer state. | |
899 | * | |
900 | * called with rq->lock held and irqs disabled | |
901 | */ | |
902 | static void hrtick_start(struct rq *rq, u64 delay, int reset) | |
903 | { | |
904 | assert_spin_locked(&rq->lock); | |
905 | ||
906 | /* | |
907 | * preempt at: now + delay | |
908 | */ | |
909 | rq->hrtick_expire = | |
910 | ktime_add_ns(rq->hrtick_timer.base->get_time(), delay); | |
911 | /* | |
912 | * indicate we need to program the timer | |
913 | */ | |
914 | __set_bit(HRTICK_SET, &rq->hrtick_flags); | |
915 | if (reset) | |
916 | __set_bit(HRTICK_RESET, &rq->hrtick_flags); | |
917 | ||
918 | /* | |
919 | * New slices are called from the schedule path and don't need a | |
920 | * forced reschedule. | |
921 | */ | |
922 | if (reset) | |
923 | resched_hrt(rq->curr); | |
924 | } | |
925 | ||
926 | static void hrtick_clear(struct rq *rq) | |
927 | { | |
928 | if (hrtimer_active(&rq->hrtick_timer)) | |
929 | hrtimer_cancel(&rq->hrtick_timer); | |
930 | } | |
931 | ||
932 | /* | |
933 | * Update the timer from the possible pending state. | |
934 | */ | |
935 | static void hrtick_set(struct rq *rq) | |
936 | { | |
937 | ktime_t time; | |
938 | int set, reset; | |
939 | unsigned long flags; | |
940 | ||
941 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
942 | ||
943 | spin_lock_irqsave(&rq->lock, flags); | |
944 | set = __test_and_clear_bit(HRTICK_SET, &rq->hrtick_flags); | |
945 | reset = __test_and_clear_bit(HRTICK_RESET, &rq->hrtick_flags); | |
946 | time = rq->hrtick_expire; | |
947 | clear_thread_flag(TIF_HRTICK_RESCHED); | |
948 | spin_unlock_irqrestore(&rq->lock, flags); | |
949 | ||
950 | if (set) { | |
951 | hrtimer_start(&rq->hrtick_timer, time, HRTIMER_MODE_ABS); | |
952 | if (reset && !hrtimer_active(&rq->hrtick_timer)) | |
953 | resched_rq(rq); | |
954 | } else | |
955 | hrtick_clear(rq); | |
956 | } | |
957 | ||
958 | /* | |
959 | * High-resolution timer tick. | |
960 | * Runs from hardirq context with interrupts disabled. | |
961 | */ | |
962 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
963 | { | |
964 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
965 | ||
966 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
967 | ||
968 | spin_lock(&rq->lock); | |
969 | __update_rq_clock(rq); | |
970 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); | |
971 | spin_unlock(&rq->lock); | |
972 | ||
973 | return HRTIMER_NORESTART; | |
974 | } | |
975 | ||
976 | static inline void init_rq_hrtick(struct rq *rq) | |
977 | { | |
978 | rq->hrtick_flags = 0; | |
979 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
980 | rq->hrtick_timer.function = hrtick; | |
981 | rq->hrtick_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ; | |
982 | } | |
983 | ||
984 | void hrtick_resched(void) | |
985 | { | |
986 | struct rq *rq; | |
987 | unsigned long flags; | |
988 | ||
989 | if (!test_thread_flag(TIF_HRTICK_RESCHED)) | |
990 | return; | |
991 | ||
992 | local_irq_save(flags); | |
993 | rq = cpu_rq(smp_processor_id()); | |
994 | hrtick_set(rq); | |
995 | local_irq_restore(flags); | |
996 | } | |
997 | #else | |
998 | static inline void hrtick_clear(struct rq *rq) | |
999 | { | |
1000 | } | |
1001 | ||
1002 | static inline void hrtick_set(struct rq *rq) | |
1003 | { | |
1004 | } | |
1005 | ||
1006 | static inline void init_rq_hrtick(struct rq *rq) | |
1007 | { | |
1008 | } | |
1009 | ||
1010 | void hrtick_resched(void) | |
1011 | { | |
1012 | } | |
1013 | #endif | |
1014 | ||
c24d20db IM |
1015 | /* |
1016 | * resched_task - mark a task 'to be rescheduled now'. | |
1017 | * | |
1018 | * On UP this means the setting of the need_resched flag, on SMP it | |
1019 | * might also involve a cross-CPU call to trigger the scheduler on | |
1020 | * the target CPU. | |
1021 | */ | |
1022 | #ifdef CONFIG_SMP | |
1023 | ||
1024 | #ifndef tsk_is_polling | |
1025 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1026 | #endif | |
1027 | ||
8f4d37ec | 1028 | static void __resched_task(struct task_struct *p, int tif_bit) |
c24d20db IM |
1029 | { |
1030 | int cpu; | |
1031 | ||
1032 | assert_spin_locked(&task_rq(p)->lock); | |
1033 | ||
8f4d37ec | 1034 | if (unlikely(test_tsk_thread_flag(p, tif_bit))) |
c24d20db IM |
1035 | return; |
1036 | ||
8f4d37ec | 1037 | set_tsk_thread_flag(p, tif_bit); |
c24d20db IM |
1038 | |
1039 | cpu = task_cpu(p); | |
1040 | if (cpu == smp_processor_id()) | |
1041 | return; | |
1042 | ||
1043 | /* NEED_RESCHED must be visible before we test polling */ | |
1044 | smp_mb(); | |
1045 | if (!tsk_is_polling(p)) | |
1046 | smp_send_reschedule(cpu); | |
1047 | } | |
1048 | ||
1049 | static void resched_cpu(int cpu) | |
1050 | { | |
1051 | struct rq *rq = cpu_rq(cpu); | |
1052 | unsigned long flags; | |
1053 | ||
1054 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1055 | return; | |
1056 | resched_task(cpu_curr(cpu)); | |
1057 | spin_unlock_irqrestore(&rq->lock, flags); | |
1058 | } | |
1059 | #else | |
8f4d37ec | 1060 | static void __resched_task(struct task_struct *p, int tif_bit) |
c24d20db IM |
1061 | { |
1062 | assert_spin_locked(&task_rq(p)->lock); | |
8f4d37ec | 1063 | set_tsk_thread_flag(p, tif_bit); |
c24d20db IM |
1064 | } |
1065 | #endif | |
1066 | ||
45bf76df IM |
1067 | #if BITS_PER_LONG == 32 |
1068 | # define WMULT_CONST (~0UL) | |
1069 | #else | |
1070 | # define WMULT_CONST (1UL << 32) | |
1071 | #endif | |
1072 | ||
1073 | #define WMULT_SHIFT 32 | |
1074 | ||
194081eb IM |
1075 | /* |
1076 | * Shift right and round: | |
1077 | */ | |
cf2ab469 | 1078 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1079 | |
cb1c4fc9 | 1080 | static unsigned long |
45bf76df IM |
1081 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1082 | struct load_weight *lw) | |
1083 | { | |
1084 | u64 tmp; | |
1085 | ||
1086 | if (unlikely(!lw->inv_weight)) | |
194081eb | 1087 | lw->inv_weight = (WMULT_CONST - lw->weight/2) / lw->weight + 1; |
45bf76df IM |
1088 | |
1089 | tmp = (u64)delta_exec * weight; | |
1090 | /* | |
1091 | * Check whether we'd overflow the 64-bit multiplication: | |
1092 | */ | |
194081eb | 1093 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1094 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1095 | WMULT_SHIFT/2); |
1096 | else | |
cf2ab469 | 1097 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1098 | |
ecf691da | 1099 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1100 | } |
1101 | ||
1102 | static inline unsigned long | |
1103 | calc_delta_fair(unsigned long delta_exec, struct load_weight *lw) | |
1104 | { | |
1105 | return calc_delta_mine(delta_exec, NICE_0_LOAD, lw); | |
1106 | } | |
1107 | ||
1091985b | 1108 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1109 | { |
1110 | lw->weight += inc; | |
e89996ae | 1111 | lw->inv_weight = 0; |
45bf76df IM |
1112 | } |
1113 | ||
1091985b | 1114 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1115 | { |
1116 | lw->weight -= dec; | |
e89996ae | 1117 | lw->inv_weight = 0; |
45bf76df IM |
1118 | } |
1119 | ||
2dd73a4f PW |
1120 | /* |
1121 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1122 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1123 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1124 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1125 | * scaled version of the new time slice allocation that they receive on time |
1126 | * slice expiry etc. | |
1127 | */ | |
1128 | ||
dd41f596 IM |
1129 | #define WEIGHT_IDLEPRIO 2 |
1130 | #define WMULT_IDLEPRIO (1 << 31) | |
1131 | ||
1132 | /* | |
1133 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1134 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1135 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1136 | * that remained on nice 0. | |
1137 | * | |
1138 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1139 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1140 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1141 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1142 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1143 | */ |
1144 | static const int prio_to_weight[40] = { | |
254753dc IM |
1145 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1146 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1147 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1148 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1149 | /* 0 */ 1024, 820, 655, 526, 423, | |
1150 | /* 5 */ 335, 272, 215, 172, 137, | |
1151 | /* 10 */ 110, 87, 70, 56, 45, | |
1152 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1153 | }; |
1154 | ||
5714d2de IM |
1155 | /* |
1156 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1157 | * | |
1158 | * In cases where the weight does not change often, we can use the | |
1159 | * precalculated inverse to speed up arithmetics by turning divisions | |
1160 | * into multiplications: | |
1161 | */ | |
dd41f596 | 1162 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1163 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1164 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1165 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1166 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1167 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1168 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1169 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1170 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1171 | }; |
2dd73a4f | 1172 | |
dd41f596 IM |
1173 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1174 | ||
1175 | /* | |
1176 | * runqueue iterator, to support SMP load-balancing between different | |
1177 | * scheduling classes, without having to expose their internal data | |
1178 | * structures to the load-balancing proper: | |
1179 | */ | |
1180 | struct rq_iterator { | |
1181 | void *arg; | |
1182 | struct task_struct *(*start)(void *); | |
1183 | struct task_struct *(*next)(void *); | |
1184 | }; | |
1185 | ||
e1d1484f PW |
1186 | #ifdef CONFIG_SMP |
1187 | static unsigned long | |
1188 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1189 | unsigned long max_load_move, struct sched_domain *sd, | |
1190 | enum cpu_idle_type idle, int *all_pinned, | |
1191 | int *this_best_prio, struct rq_iterator *iterator); | |
1192 | ||
1193 | static int | |
1194 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1195 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1196 | struct rq_iterator *iterator); | |
e1d1484f | 1197 | #endif |
dd41f596 | 1198 | |
d842de87 SV |
1199 | #ifdef CONFIG_CGROUP_CPUACCT |
1200 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
1201 | #else | |
1202 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
1203 | #endif | |
1204 | ||
e7693a36 GH |
1205 | #ifdef CONFIG_SMP |
1206 | static unsigned long source_load(int cpu, int type); | |
1207 | static unsigned long target_load(int cpu, int type); | |
1208 | static unsigned long cpu_avg_load_per_task(int cpu); | |
1209 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1210 | #endif /* CONFIG_SMP */ | |
1211 | ||
dd41f596 | 1212 | #include "sched_stats.h" |
dd41f596 | 1213 | #include "sched_idletask.c" |
5522d5d5 IM |
1214 | #include "sched_fair.c" |
1215 | #include "sched_rt.c" | |
dd41f596 IM |
1216 | #ifdef CONFIG_SCHED_DEBUG |
1217 | # include "sched_debug.c" | |
1218 | #endif | |
1219 | ||
1220 | #define sched_class_highest (&rt_sched_class) | |
1221 | ||
62fb1851 PZ |
1222 | static inline void inc_load(struct rq *rq, const struct task_struct *p) |
1223 | { | |
1224 | update_load_add(&rq->load, p->se.load.weight); | |
1225 | } | |
1226 | ||
1227 | static inline void dec_load(struct rq *rq, const struct task_struct *p) | |
1228 | { | |
1229 | update_load_sub(&rq->load, p->se.load.weight); | |
1230 | } | |
1231 | ||
1232 | static void inc_nr_running(struct task_struct *p, struct rq *rq) | |
9c217245 IM |
1233 | { |
1234 | rq->nr_running++; | |
62fb1851 | 1235 | inc_load(rq, p); |
9c217245 IM |
1236 | } |
1237 | ||
62fb1851 | 1238 | static void dec_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
1239 | { |
1240 | rq->nr_running--; | |
62fb1851 | 1241 | dec_load(rq, p); |
9c217245 IM |
1242 | } |
1243 | ||
45bf76df IM |
1244 | static void set_load_weight(struct task_struct *p) |
1245 | { | |
1246 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1247 | p->se.load.weight = prio_to_weight[0] * 2; |
1248 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1249 | return; | |
1250 | } | |
45bf76df | 1251 | |
dd41f596 IM |
1252 | /* |
1253 | * SCHED_IDLE tasks get minimal weight: | |
1254 | */ | |
1255 | if (p->policy == SCHED_IDLE) { | |
1256 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1257 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1258 | return; | |
1259 | } | |
71f8bd46 | 1260 | |
dd41f596 IM |
1261 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1262 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1263 | } |
1264 | ||
8159f87e | 1265 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1266 | { |
dd41f596 | 1267 | sched_info_queued(p); |
fd390f6a | 1268 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1269 | p->se.on_rq = 1; |
71f8bd46 IM |
1270 | } |
1271 | ||
69be72c1 | 1272 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1273 | { |
f02231e5 | 1274 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1275 | p->se.on_rq = 0; |
71f8bd46 IM |
1276 | } |
1277 | ||
14531189 | 1278 | /* |
dd41f596 | 1279 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1280 | */ |
14531189 IM |
1281 | static inline int __normal_prio(struct task_struct *p) |
1282 | { | |
dd41f596 | 1283 | return p->static_prio; |
14531189 IM |
1284 | } |
1285 | ||
b29739f9 IM |
1286 | /* |
1287 | * Calculate the expected normal priority: i.e. priority | |
1288 | * without taking RT-inheritance into account. Might be | |
1289 | * boosted by interactivity modifiers. Changes upon fork, | |
1290 | * setprio syscalls, and whenever the interactivity | |
1291 | * estimator recalculates. | |
1292 | */ | |
36c8b586 | 1293 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1294 | { |
1295 | int prio; | |
1296 | ||
e05606d3 | 1297 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1298 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1299 | else | |
1300 | prio = __normal_prio(p); | |
1301 | return prio; | |
1302 | } | |
1303 | ||
1304 | /* | |
1305 | * Calculate the current priority, i.e. the priority | |
1306 | * taken into account by the scheduler. This value might | |
1307 | * be boosted by RT tasks, or might be boosted by | |
1308 | * interactivity modifiers. Will be RT if the task got | |
1309 | * RT-boosted. If not then it returns p->normal_prio. | |
1310 | */ | |
36c8b586 | 1311 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1312 | { |
1313 | p->normal_prio = normal_prio(p); | |
1314 | /* | |
1315 | * If we are RT tasks or we were boosted to RT priority, | |
1316 | * keep the priority unchanged. Otherwise, update priority | |
1317 | * to the normal priority: | |
1318 | */ | |
1319 | if (!rt_prio(p->prio)) | |
1320 | return p->normal_prio; | |
1321 | return p->prio; | |
1322 | } | |
1323 | ||
1da177e4 | 1324 | /* |
dd41f596 | 1325 | * activate_task - move a task to the runqueue. |
1da177e4 | 1326 | */ |
dd41f596 | 1327 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1328 | { |
d9514f6c | 1329 | if (task_contributes_to_load(p)) |
dd41f596 | 1330 | rq->nr_uninterruptible--; |
1da177e4 | 1331 | |
8159f87e | 1332 | enqueue_task(rq, p, wakeup); |
62fb1851 | 1333 | inc_nr_running(p, rq); |
1da177e4 LT |
1334 | } |
1335 | ||
1da177e4 LT |
1336 | /* |
1337 | * deactivate_task - remove a task from the runqueue. | |
1338 | */ | |
2e1cb74a | 1339 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1340 | { |
d9514f6c | 1341 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1342 | rq->nr_uninterruptible++; |
1343 | ||
69be72c1 | 1344 | dequeue_task(rq, p, sleep); |
62fb1851 | 1345 | dec_nr_running(p, rq); |
1da177e4 LT |
1346 | } |
1347 | ||
1da177e4 LT |
1348 | /** |
1349 | * task_curr - is this task currently executing on a CPU? | |
1350 | * @p: the task in question. | |
1351 | */ | |
36c8b586 | 1352 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1353 | { |
1354 | return cpu_curr(task_cpu(p)) == p; | |
1355 | } | |
1356 | ||
2dd73a4f PW |
1357 | /* Used instead of source_load when we know the type == 0 */ |
1358 | unsigned long weighted_cpuload(const int cpu) | |
1359 | { | |
495eca49 | 1360 | return cpu_rq(cpu)->load.weight; |
dd41f596 IM |
1361 | } |
1362 | ||
1363 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1364 | { | |
6f505b16 | 1365 | set_task_rq(p, cpu); |
dd41f596 | 1366 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1367 | /* |
1368 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1369 | * successfuly executed on another CPU. We must ensure that updates of | |
1370 | * per-task data have been completed by this moment. | |
1371 | */ | |
1372 | smp_wmb(); | |
dd41f596 | 1373 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1374 | #endif |
2dd73a4f PW |
1375 | } |
1376 | ||
cb469845 SR |
1377 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1378 | const struct sched_class *prev_class, | |
1379 | int oldprio, int running) | |
1380 | { | |
1381 | if (prev_class != p->sched_class) { | |
1382 | if (prev_class->switched_from) | |
1383 | prev_class->switched_from(rq, p, running); | |
1384 | p->sched_class->switched_to(rq, p, running); | |
1385 | } else | |
1386 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1387 | } | |
1388 | ||
1da177e4 | 1389 | #ifdef CONFIG_SMP |
c65cc870 | 1390 | |
cc367732 IM |
1391 | /* |
1392 | * Is this task likely cache-hot: | |
1393 | */ | |
e7693a36 | 1394 | static int |
cc367732 IM |
1395 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1396 | { | |
1397 | s64 delta; | |
1398 | ||
1399 | if (p->sched_class != &fair_sched_class) | |
1400 | return 0; | |
1401 | ||
6bc1665b IM |
1402 | if (sysctl_sched_migration_cost == -1) |
1403 | return 1; | |
1404 | if (sysctl_sched_migration_cost == 0) | |
1405 | return 0; | |
1406 | ||
cc367732 IM |
1407 | delta = now - p->se.exec_start; |
1408 | ||
1409 | return delta < (s64)sysctl_sched_migration_cost; | |
1410 | } | |
1411 | ||
1412 | ||
dd41f596 | 1413 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1414 | { |
dd41f596 IM |
1415 | int old_cpu = task_cpu(p); |
1416 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1417 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1418 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1419 | u64 clock_offset; |
dd41f596 IM |
1420 | |
1421 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1422 | |
1423 | #ifdef CONFIG_SCHEDSTATS | |
1424 | if (p->se.wait_start) | |
1425 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1426 | if (p->se.sleep_start) |
1427 | p->se.sleep_start -= clock_offset; | |
1428 | if (p->se.block_start) | |
1429 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1430 | if (old_cpu != new_cpu) { |
1431 | schedstat_inc(p, se.nr_migrations); | |
1432 | if (task_hot(p, old_rq->clock, NULL)) | |
1433 | schedstat_inc(p, se.nr_forced2_migrations); | |
1434 | } | |
6cfb0d5d | 1435 | #endif |
2830cf8c SV |
1436 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1437 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1438 | |
1439 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1440 | } |
1441 | ||
70b97a7f | 1442 | struct migration_req { |
1da177e4 | 1443 | struct list_head list; |
1da177e4 | 1444 | |
36c8b586 | 1445 | struct task_struct *task; |
1da177e4 LT |
1446 | int dest_cpu; |
1447 | ||
1da177e4 | 1448 | struct completion done; |
70b97a7f | 1449 | }; |
1da177e4 LT |
1450 | |
1451 | /* | |
1452 | * The task's runqueue lock must be held. | |
1453 | * Returns true if you have to wait for migration thread. | |
1454 | */ | |
36c8b586 | 1455 | static int |
70b97a7f | 1456 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1457 | { |
70b97a7f | 1458 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1459 | |
1460 | /* | |
1461 | * If the task is not on a runqueue (and not running), then | |
1462 | * it is sufficient to simply update the task's cpu field. | |
1463 | */ | |
dd41f596 | 1464 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1465 | set_task_cpu(p, dest_cpu); |
1466 | return 0; | |
1467 | } | |
1468 | ||
1469 | init_completion(&req->done); | |
1da177e4 LT |
1470 | req->task = p; |
1471 | req->dest_cpu = dest_cpu; | |
1472 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1473 | |
1da177e4 LT |
1474 | return 1; |
1475 | } | |
1476 | ||
1477 | /* | |
1478 | * wait_task_inactive - wait for a thread to unschedule. | |
1479 | * | |
1480 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1481 | * else this function might spin for a *long* time. This function can't | |
1482 | * be called with interrupts off, or it may introduce deadlock with | |
1483 | * smp_call_function() if an IPI is sent by the same process we are | |
1484 | * waiting to become inactive. | |
1485 | */ | |
36c8b586 | 1486 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1487 | { |
1488 | unsigned long flags; | |
dd41f596 | 1489 | int running, on_rq; |
70b97a7f | 1490 | struct rq *rq; |
1da177e4 | 1491 | |
3a5c359a AK |
1492 | for (;;) { |
1493 | /* | |
1494 | * We do the initial early heuristics without holding | |
1495 | * any task-queue locks at all. We'll only try to get | |
1496 | * the runqueue lock when things look like they will | |
1497 | * work out! | |
1498 | */ | |
1499 | rq = task_rq(p); | |
fa490cfd | 1500 | |
3a5c359a AK |
1501 | /* |
1502 | * If the task is actively running on another CPU | |
1503 | * still, just relax and busy-wait without holding | |
1504 | * any locks. | |
1505 | * | |
1506 | * NOTE! Since we don't hold any locks, it's not | |
1507 | * even sure that "rq" stays as the right runqueue! | |
1508 | * But we don't care, since "task_running()" will | |
1509 | * return false if the runqueue has changed and p | |
1510 | * is actually now running somewhere else! | |
1511 | */ | |
1512 | while (task_running(rq, p)) | |
1513 | cpu_relax(); | |
fa490cfd | 1514 | |
3a5c359a AK |
1515 | /* |
1516 | * Ok, time to look more closely! We need the rq | |
1517 | * lock now, to be *sure*. If we're wrong, we'll | |
1518 | * just go back and repeat. | |
1519 | */ | |
1520 | rq = task_rq_lock(p, &flags); | |
1521 | running = task_running(rq, p); | |
1522 | on_rq = p->se.on_rq; | |
1523 | task_rq_unlock(rq, &flags); | |
fa490cfd | 1524 | |
3a5c359a AK |
1525 | /* |
1526 | * Was it really running after all now that we | |
1527 | * checked with the proper locks actually held? | |
1528 | * | |
1529 | * Oops. Go back and try again.. | |
1530 | */ | |
1531 | if (unlikely(running)) { | |
1532 | cpu_relax(); | |
1533 | continue; | |
1534 | } | |
fa490cfd | 1535 | |
3a5c359a AK |
1536 | /* |
1537 | * It's not enough that it's not actively running, | |
1538 | * it must be off the runqueue _entirely_, and not | |
1539 | * preempted! | |
1540 | * | |
1541 | * So if it wa still runnable (but just not actively | |
1542 | * running right now), it's preempted, and we should | |
1543 | * yield - it could be a while. | |
1544 | */ | |
1545 | if (unlikely(on_rq)) { | |
1546 | schedule_timeout_uninterruptible(1); | |
1547 | continue; | |
1548 | } | |
fa490cfd | 1549 | |
3a5c359a AK |
1550 | /* |
1551 | * Ahh, all good. It wasn't running, and it wasn't | |
1552 | * runnable, which means that it will never become | |
1553 | * running in the future either. We're all done! | |
1554 | */ | |
1555 | break; | |
1556 | } | |
1da177e4 LT |
1557 | } |
1558 | ||
1559 | /*** | |
1560 | * kick_process - kick a running thread to enter/exit the kernel | |
1561 | * @p: the to-be-kicked thread | |
1562 | * | |
1563 | * Cause a process which is running on another CPU to enter | |
1564 | * kernel-mode, without any delay. (to get signals handled.) | |
1565 | * | |
1566 | * NOTE: this function doesnt have to take the runqueue lock, | |
1567 | * because all it wants to ensure is that the remote task enters | |
1568 | * the kernel. If the IPI races and the task has been migrated | |
1569 | * to another CPU then no harm is done and the purpose has been | |
1570 | * achieved as well. | |
1571 | */ | |
36c8b586 | 1572 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1573 | { |
1574 | int cpu; | |
1575 | ||
1576 | preempt_disable(); | |
1577 | cpu = task_cpu(p); | |
1578 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1579 | smp_send_reschedule(cpu); | |
1580 | preempt_enable(); | |
1581 | } | |
1582 | ||
1583 | /* | |
2dd73a4f PW |
1584 | * Return a low guess at the load of a migration-source cpu weighted |
1585 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1586 | * |
1587 | * We want to under-estimate the load of migration sources, to | |
1588 | * balance conservatively. | |
1589 | */ | |
a9957449 | 1590 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 1591 | { |
70b97a7f | 1592 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1593 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1594 | |
3b0bd9bc | 1595 | if (type == 0) |
dd41f596 | 1596 | return total; |
b910472d | 1597 | |
dd41f596 | 1598 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
1599 | } |
1600 | ||
1601 | /* | |
2dd73a4f PW |
1602 | * Return a high guess at the load of a migration-target cpu weighted |
1603 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1604 | */ |
a9957449 | 1605 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 1606 | { |
70b97a7f | 1607 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1608 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1609 | |
7897986b | 1610 | if (type == 0) |
dd41f596 | 1611 | return total; |
3b0bd9bc | 1612 | |
dd41f596 | 1613 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
1614 | } |
1615 | ||
1616 | /* | |
1617 | * Return the average load per task on the cpu's run queue | |
1618 | */ | |
e7693a36 | 1619 | static unsigned long cpu_avg_load_per_task(int cpu) |
2dd73a4f | 1620 | { |
70b97a7f | 1621 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1622 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f PW |
1623 | unsigned long n = rq->nr_running; |
1624 | ||
dd41f596 | 1625 | return n ? total / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1626 | } |
1627 | ||
147cbb4b NP |
1628 | /* |
1629 | * find_idlest_group finds and returns the least busy CPU group within the | |
1630 | * domain. | |
1631 | */ | |
1632 | static struct sched_group * | |
1633 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1634 | { | |
1635 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1636 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1637 | int load_idx = sd->forkexec_idx; | |
1638 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1639 | ||
1640 | do { | |
1641 | unsigned long load, avg_load; | |
1642 | int local_group; | |
1643 | int i; | |
1644 | ||
da5a5522 BD |
1645 | /* Skip over this group if it has no CPUs allowed */ |
1646 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 1647 | continue; |
da5a5522 | 1648 | |
147cbb4b | 1649 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1650 | |
1651 | /* Tally up the load of all CPUs in the group */ | |
1652 | avg_load = 0; | |
1653 | ||
1654 | for_each_cpu_mask(i, group->cpumask) { | |
1655 | /* Bias balancing toward cpus of our domain */ | |
1656 | if (local_group) | |
1657 | load = source_load(i, load_idx); | |
1658 | else | |
1659 | load = target_load(i, load_idx); | |
1660 | ||
1661 | avg_load += load; | |
1662 | } | |
1663 | ||
1664 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1665 | avg_load = sg_div_cpu_power(group, |
1666 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1667 | |
1668 | if (local_group) { | |
1669 | this_load = avg_load; | |
1670 | this = group; | |
1671 | } else if (avg_load < min_load) { | |
1672 | min_load = avg_load; | |
1673 | idlest = group; | |
1674 | } | |
3a5c359a | 1675 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
1676 | |
1677 | if (!idlest || 100*this_load < imbalance*min_load) | |
1678 | return NULL; | |
1679 | return idlest; | |
1680 | } | |
1681 | ||
1682 | /* | |
0feaece9 | 1683 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1684 | */ |
95cdf3b7 IM |
1685 | static int |
1686 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) | |
147cbb4b | 1687 | { |
da5a5522 | 1688 | cpumask_t tmp; |
147cbb4b NP |
1689 | unsigned long load, min_load = ULONG_MAX; |
1690 | int idlest = -1; | |
1691 | int i; | |
1692 | ||
da5a5522 BD |
1693 | /* Traverse only the allowed CPUs */ |
1694 | cpus_and(tmp, group->cpumask, p->cpus_allowed); | |
1695 | ||
1696 | for_each_cpu_mask(i, tmp) { | |
2dd73a4f | 1697 | load = weighted_cpuload(i); |
147cbb4b NP |
1698 | |
1699 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1700 | min_load = load; | |
1701 | idlest = i; | |
1702 | } | |
1703 | } | |
1704 | ||
1705 | return idlest; | |
1706 | } | |
1707 | ||
476d139c NP |
1708 | /* |
1709 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1710 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1711 | * SD_BALANCE_EXEC. | |
1712 | * | |
1713 | * Balance, ie. select the least loaded group. | |
1714 | * | |
1715 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1716 | * | |
1717 | * preempt must be disabled. | |
1718 | */ | |
1719 | static int sched_balance_self(int cpu, int flag) | |
1720 | { | |
1721 | struct task_struct *t = current; | |
1722 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1723 | |
c96d145e | 1724 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
1725 | /* |
1726 | * If power savings logic is enabled for a domain, stop there. | |
1727 | */ | |
5c45bf27 SS |
1728 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
1729 | break; | |
476d139c NP |
1730 | if (tmp->flags & flag) |
1731 | sd = tmp; | |
c96d145e | 1732 | } |
476d139c NP |
1733 | |
1734 | while (sd) { | |
1735 | cpumask_t span; | |
1736 | struct sched_group *group; | |
1a848870 SS |
1737 | int new_cpu, weight; |
1738 | ||
1739 | if (!(sd->flags & flag)) { | |
1740 | sd = sd->child; | |
1741 | continue; | |
1742 | } | |
476d139c NP |
1743 | |
1744 | span = sd->span; | |
1745 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1746 | if (!group) { |
1747 | sd = sd->child; | |
1748 | continue; | |
1749 | } | |
476d139c | 1750 | |
da5a5522 | 1751 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
1752 | if (new_cpu == -1 || new_cpu == cpu) { |
1753 | /* Now try balancing at a lower domain level of cpu */ | |
1754 | sd = sd->child; | |
1755 | continue; | |
1756 | } | |
476d139c | 1757 | |
1a848870 | 1758 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1759 | cpu = new_cpu; |
476d139c NP |
1760 | sd = NULL; |
1761 | weight = cpus_weight(span); | |
1762 | for_each_domain(cpu, tmp) { | |
1763 | if (weight <= cpus_weight(tmp->span)) | |
1764 | break; | |
1765 | if (tmp->flags & flag) | |
1766 | sd = tmp; | |
1767 | } | |
1768 | /* while loop will break here if sd == NULL */ | |
1769 | } | |
1770 | ||
1771 | return cpu; | |
1772 | } | |
1773 | ||
1774 | #endif /* CONFIG_SMP */ | |
1da177e4 | 1775 | |
1da177e4 LT |
1776 | /*** |
1777 | * try_to_wake_up - wake up a thread | |
1778 | * @p: the to-be-woken-up thread | |
1779 | * @state: the mask of task states that can be woken | |
1780 | * @sync: do a synchronous wakeup? | |
1781 | * | |
1782 | * Put it on the run-queue if it's not already there. The "current" | |
1783 | * thread is always on the run-queue (except when the actual | |
1784 | * re-schedule is in progress), and as such you're allowed to do | |
1785 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1786 | * runnable without the overhead of this. | |
1787 | * | |
1788 | * returns failure only if the task is already active. | |
1789 | */ | |
36c8b586 | 1790 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 1791 | { |
cc367732 | 1792 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
1793 | unsigned long flags; |
1794 | long old_state; | |
70b97a7f | 1795 | struct rq *rq; |
1da177e4 | 1796 | |
04e2f174 | 1797 | smp_wmb(); |
1da177e4 LT |
1798 | rq = task_rq_lock(p, &flags); |
1799 | old_state = p->state; | |
1800 | if (!(old_state & state)) | |
1801 | goto out; | |
1802 | ||
dd41f596 | 1803 | if (p->se.on_rq) |
1da177e4 LT |
1804 | goto out_running; |
1805 | ||
1806 | cpu = task_cpu(p); | |
cc367732 | 1807 | orig_cpu = cpu; |
1da177e4 LT |
1808 | this_cpu = smp_processor_id(); |
1809 | ||
1810 | #ifdef CONFIG_SMP | |
1811 | if (unlikely(task_running(rq, p))) | |
1812 | goto out_activate; | |
1813 | ||
5d2f5a61 DA |
1814 | cpu = p->sched_class->select_task_rq(p, sync); |
1815 | if (cpu != orig_cpu) { | |
1816 | set_task_cpu(p, cpu); | |
1da177e4 LT |
1817 | task_rq_unlock(rq, &flags); |
1818 | /* might preempt at this point */ | |
1819 | rq = task_rq_lock(p, &flags); | |
1820 | old_state = p->state; | |
1821 | if (!(old_state & state)) | |
1822 | goto out; | |
dd41f596 | 1823 | if (p->se.on_rq) |
1da177e4 LT |
1824 | goto out_running; |
1825 | ||
1826 | this_cpu = smp_processor_id(); | |
1827 | cpu = task_cpu(p); | |
1828 | } | |
1829 | ||
e7693a36 GH |
1830 | #ifdef CONFIG_SCHEDSTATS |
1831 | schedstat_inc(rq, ttwu_count); | |
1832 | if (cpu == this_cpu) | |
1833 | schedstat_inc(rq, ttwu_local); | |
1834 | else { | |
1835 | struct sched_domain *sd; | |
1836 | for_each_domain(this_cpu, sd) { | |
1837 | if (cpu_isset(cpu, sd->span)) { | |
1838 | schedstat_inc(sd, ttwu_wake_remote); | |
1839 | break; | |
1840 | } | |
1841 | } | |
1842 | } | |
e7693a36 GH |
1843 | #endif |
1844 | ||
1da177e4 LT |
1845 | out_activate: |
1846 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
1847 | schedstat_inc(p, se.nr_wakeups); |
1848 | if (sync) | |
1849 | schedstat_inc(p, se.nr_wakeups_sync); | |
1850 | if (orig_cpu != cpu) | |
1851 | schedstat_inc(p, se.nr_wakeups_migrate); | |
1852 | if (cpu == this_cpu) | |
1853 | schedstat_inc(p, se.nr_wakeups_local); | |
1854 | else | |
1855 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 1856 | update_rq_clock(rq); |
dd41f596 | 1857 | activate_task(rq, p, 1); |
9c63d9c0 | 1858 | check_preempt_curr(rq, p); |
1da177e4 LT |
1859 | success = 1; |
1860 | ||
1861 | out_running: | |
1862 | p->state = TASK_RUNNING; | |
9a897c5a SR |
1863 | #ifdef CONFIG_SMP |
1864 | if (p->sched_class->task_wake_up) | |
1865 | p->sched_class->task_wake_up(rq, p); | |
1866 | #endif | |
1da177e4 LT |
1867 | out: |
1868 | task_rq_unlock(rq, &flags); | |
1869 | ||
1870 | return success; | |
1871 | } | |
1872 | ||
7ad5b3a5 | 1873 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1874 | { |
d9514f6c | 1875 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 1876 | } |
1da177e4 LT |
1877 | EXPORT_SYMBOL(wake_up_process); |
1878 | ||
7ad5b3a5 | 1879 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1880 | { |
1881 | return try_to_wake_up(p, state, 0); | |
1882 | } | |
1883 | ||
1da177e4 LT |
1884 | /* |
1885 | * Perform scheduler related setup for a newly forked process p. | |
1886 | * p is forked by current. | |
dd41f596 IM |
1887 | * |
1888 | * __sched_fork() is basic setup used by init_idle() too: | |
1889 | */ | |
1890 | static void __sched_fork(struct task_struct *p) | |
1891 | { | |
dd41f596 IM |
1892 | p->se.exec_start = 0; |
1893 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1894 | p->se.prev_sum_exec_runtime = 0; |
6cfb0d5d IM |
1895 | |
1896 | #ifdef CONFIG_SCHEDSTATS | |
1897 | p->se.wait_start = 0; | |
dd41f596 IM |
1898 | p->se.sum_sleep_runtime = 0; |
1899 | p->se.sleep_start = 0; | |
dd41f596 IM |
1900 | p->se.block_start = 0; |
1901 | p->se.sleep_max = 0; | |
1902 | p->se.block_max = 0; | |
1903 | p->se.exec_max = 0; | |
eba1ed4b | 1904 | p->se.slice_max = 0; |
dd41f596 | 1905 | p->se.wait_max = 0; |
6cfb0d5d | 1906 | #endif |
476d139c | 1907 | |
fa717060 | 1908 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 1909 | p->se.on_rq = 0; |
476d139c | 1910 | |
e107be36 AK |
1911 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1912 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1913 | #endif | |
1914 | ||
1da177e4 LT |
1915 | /* |
1916 | * We mark the process as running here, but have not actually | |
1917 | * inserted it onto the runqueue yet. This guarantees that | |
1918 | * nobody will actually run it, and a signal or other external | |
1919 | * event cannot wake it up and insert it on the runqueue either. | |
1920 | */ | |
1921 | p->state = TASK_RUNNING; | |
dd41f596 IM |
1922 | } |
1923 | ||
1924 | /* | |
1925 | * fork()/clone()-time setup: | |
1926 | */ | |
1927 | void sched_fork(struct task_struct *p, int clone_flags) | |
1928 | { | |
1929 | int cpu = get_cpu(); | |
1930 | ||
1931 | __sched_fork(p); | |
1932 | ||
1933 | #ifdef CONFIG_SMP | |
1934 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
1935 | #endif | |
02e4bac2 | 1936 | set_task_cpu(p, cpu); |
b29739f9 IM |
1937 | |
1938 | /* | |
1939 | * Make sure we do not leak PI boosting priority to the child: | |
1940 | */ | |
1941 | p->prio = current->normal_prio; | |
2ddbf952 HS |
1942 | if (!rt_prio(p->prio)) |
1943 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1944 | |
52f17b6c | 1945 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1946 | if (likely(sched_info_on())) |
52f17b6c | 1947 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1948 | #endif |
d6077cb8 | 1949 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
1950 | p->oncpu = 0; |
1951 | #endif | |
1da177e4 | 1952 | #ifdef CONFIG_PREEMPT |
4866cde0 | 1953 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1954 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1955 | #endif |
476d139c | 1956 | put_cpu(); |
1da177e4 LT |
1957 | } |
1958 | ||
1959 | /* | |
1960 | * wake_up_new_task - wake up a newly created task for the first time. | |
1961 | * | |
1962 | * This function will do some initial scheduler statistics housekeeping | |
1963 | * that must be done for every newly created context, then puts the task | |
1964 | * on the runqueue and wakes it. | |
1965 | */ | |
7ad5b3a5 | 1966 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
1967 | { |
1968 | unsigned long flags; | |
dd41f596 | 1969 | struct rq *rq; |
1da177e4 LT |
1970 | |
1971 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 1972 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 1973 | update_rq_clock(rq); |
1da177e4 LT |
1974 | |
1975 | p->prio = effective_prio(p); | |
1976 | ||
b9dca1e0 | 1977 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 1978 | activate_task(rq, p, 0); |
1da177e4 | 1979 | } else { |
1da177e4 | 1980 | /* |
dd41f596 IM |
1981 | * Let the scheduling class do new task startup |
1982 | * management (if any): | |
1da177e4 | 1983 | */ |
ee0827d8 | 1984 | p->sched_class->task_new(rq, p); |
62fb1851 | 1985 | inc_nr_running(p, rq); |
1da177e4 | 1986 | } |
dd41f596 | 1987 | check_preempt_curr(rq, p); |
9a897c5a SR |
1988 | #ifdef CONFIG_SMP |
1989 | if (p->sched_class->task_wake_up) | |
1990 | p->sched_class->task_wake_up(rq, p); | |
1991 | #endif | |
dd41f596 | 1992 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
1993 | } |
1994 | ||
e107be36 AK |
1995 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1996 | ||
1997 | /** | |
421cee29 RD |
1998 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
1999 | * @notifier: notifier struct to register | |
e107be36 AK |
2000 | */ |
2001 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2002 | { | |
2003 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2004 | } | |
2005 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2006 | ||
2007 | /** | |
2008 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2009 | * @notifier: notifier struct to unregister |
e107be36 AK |
2010 | * |
2011 | * This is safe to call from within a preemption notifier. | |
2012 | */ | |
2013 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2014 | { | |
2015 | hlist_del(¬ifier->link); | |
2016 | } | |
2017 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2018 | ||
2019 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2020 | { | |
2021 | struct preempt_notifier *notifier; | |
2022 | struct hlist_node *node; | |
2023 | ||
2024 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2025 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2026 | } | |
2027 | ||
2028 | static void | |
2029 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2030 | struct task_struct *next) | |
2031 | { | |
2032 | struct preempt_notifier *notifier; | |
2033 | struct hlist_node *node; | |
2034 | ||
2035 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2036 | notifier->ops->sched_out(notifier, next); | |
2037 | } | |
2038 | ||
2039 | #else | |
2040 | ||
2041 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2042 | { | |
2043 | } | |
2044 | ||
2045 | static void | |
2046 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2047 | struct task_struct *next) | |
2048 | { | |
2049 | } | |
2050 | ||
2051 | #endif | |
2052 | ||
4866cde0 NP |
2053 | /** |
2054 | * prepare_task_switch - prepare to switch tasks | |
2055 | * @rq: the runqueue preparing to switch | |
421cee29 | 2056 | * @prev: the current task that is being switched out |
4866cde0 NP |
2057 | * @next: the task we are going to switch to. |
2058 | * | |
2059 | * This is called with the rq lock held and interrupts off. It must | |
2060 | * be paired with a subsequent finish_task_switch after the context | |
2061 | * switch. | |
2062 | * | |
2063 | * prepare_task_switch sets up locking and calls architecture specific | |
2064 | * hooks. | |
2065 | */ | |
e107be36 AK |
2066 | static inline void |
2067 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2068 | struct task_struct *next) | |
4866cde0 | 2069 | { |
e107be36 | 2070 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2071 | prepare_lock_switch(rq, next); |
2072 | prepare_arch_switch(next); | |
2073 | } | |
2074 | ||
1da177e4 LT |
2075 | /** |
2076 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2077 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2078 | * @prev: the thread we just switched away from. |
2079 | * | |
4866cde0 NP |
2080 | * finish_task_switch must be called after the context switch, paired |
2081 | * with a prepare_task_switch call before the context switch. | |
2082 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2083 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2084 | * |
2085 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2086 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2087 | * with the lock held can cause deadlocks; see schedule() for |
2088 | * details.) | |
2089 | */ | |
a9957449 | 2090 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2091 | __releases(rq->lock) |
2092 | { | |
1da177e4 | 2093 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2094 | long prev_state; |
1da177e4 LT |
2095 | |
2096 | rq->prev_mm = NULL; | |
2097 | ||
2098 | /* | |
2099 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2100 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2101 | * schedule one last time. The schedule call will never return, and |
2102 | * the scheduled task must drop that reference. | |
c394cc9f | 2103 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2104 | * still held, otherwise prev could be scheduled on another cpu, die |
2105 | * there before we look at prev->state, and then the reference would | |
2106 | * be dropped twice. | |
2107 | * Manfred Spraul <manfred@colorfullife.com> | |
2108 | */ | |
55a101f8 | 2109 | prev_state = prev->state; |
4866cde0 NP |
2110 | finish_arch_switch(prev); |
2111 | finish_lock_switch(rq, prev); | |
9a897c5a SR |
2112 | #ifdef CONFIG_SMP |
2113 | if (current->sched_class->post_schedule) | |
2114 | current->sched_class->post_schedule(rq); | |
2115 | #endif | |
e8fa1362 | 2116 | |
e107be36 | 2117 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2118 | if (mm) |
2119 | mmdrop(mm); | |
c394cc9f | 2120 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2121 | /* |
2122 | * Remove function-return probe instances associated with this | |
2123 | * task and put them back on the free list. | |
9761eea8 | 2124 | */ |
c6fd91f0 | 2125 | kprobe_flush_task(prev); |
1da177e4 | 2126 | put_task_struct(prev); |
c6fd91f0 | 2127 | } |
1da177e4 LT |
2128 | } |
2129 | ||
2130 | /** | |
2131 | * schedule_tail - first thing a freshly forked thread must call. | |
2132 | * @prev: the thread we just switched away from. | |
2133 | */ | |
36c8b586 | 2134 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2135 | __releases(rq->lock) |
2136 | { | |
70b97a7f IM |
2137 | struct rq *rq = this_rq(); |
2138 | ||
4866cde0 NP |
2139 | finish_task_switch(rq, prev); |
2140 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2141 | /* In this case, finish_task_switch does not reenable preemption */ | |
2142 | preempt_enable(); | |
2143 | #endif | |
1da177e4 | 2144 | if (current->set_child_tid) |
b488893a | 2145 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2146 | } |
2147 | ||
2148 | /* | |
2149 | * context_switch - switch to the new MM and the new | |
2150 | * thread's register state. | |
2151 | */ | |
dd41f596 | 2152 | static inline void |
70b97a7f | 2153 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2154 | struct task_struct *next) |
1da177e4 | 2155 | { |
dd41f596 | 2156 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2157 | |
e107be36 | 2158 | prepare_task_switch(rq, prev, next); |
dd41f596 IM |
2159 | mm = next->mm; |
2160 | oldmm = prev->active_mm; | |
9226d125 ZA |
2161 | /* |
2162 | * For paravirt, this is coupled with an exit in switch_to to | |
2163 | * combine the page table reload and the switch backend into | |
2164 | * one hypercall. | |
2165 | */ | |
2166 | arch_enter_lazy_cpu_mode(); | |
2167 | ||
dd41f596 | 2168 | if (unlikely(!mm)) { |
1da177e4 LT |
2169 | next->active_mm = oldmm; |
2170 | atomic_inc(&oldmm->mm_count); | |
2171 | enter_lazy_tlb(oldmm, next); | |
2172 | } else | |
2173 | switch_mm(oldmm, mm, next); | |
2174 | ||
dd41f596 | 2175 | if (unlikely(!prev->mm)) { |
1da177e4 | 2176 | prev->active_mm = NULL; |
1da177e4 LT |
2177 | rq->prev_mm = oldmm; |
2178 | } | |
3a5f5e48 IM |
2179 | /* |
2180 | * Since the runqueue lock will be released by the next | |
2181 | * task (which is an invalid locking op but in the case | |
2182 | * of the scheduler it's an obvious special-case), so we | |
2183 | * do an early lockdep release here: | |
2184 | */ | |
2185 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2186 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2187 | #endif |
1da177e4 LT |
2188 | |
2189 | /* Here we just switch the register state and the stack. */ | |
2190 | switch_to(prev, next, prev); | |
2191 | ||
dd41f596 IM |
2192 | barrier(); |
2193 | /* | |
2194 | * this_rq must be evaluated again because prev may have moved | |
2195 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2196 | * frame will be invalid. | |
2197 | */ | |
2198 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2199 | } |
2200 | ||
2201 | /* | |
2202 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2203 | * | |
2204 | * externally visible scheduler statistics: current number of runnable | |
2205 | * threads, current number of uninterruptible-sleeping threads, total | |
2206 | * number of context switches performed since bootup. | |
2207 | */ | |
2208 | unsigned long nr_running(void) | |
2209 | { | |
2210 | unsigned long i, sum = 0; | |
2211 | ||
2212 | for_each_online_cpu(i) | |
2213 | sum += cpu_rq(i)->nr_running; | |
2214 | ||
2215 | return sum; | |
2216 | } | |
2217 | ||
2218 | unsigned long nr_uninterruptible(void) | |
2219 | { | |
2220 | unsigned long i, sum = 0; | |
2221 | ||
0a945022 | 2222 | for_each_possible_cpu(i) |
1da177e4 LT |
2223 | sum += cpu_rq(i)->nr_uninterruptible; |
2224 | ||
2225 | /* | |
2226 | * Since we read the counters lockless, it might be slightly | |
2227 | * inaccurate. Do not allow it to go below zero though: | |
2228 | */ | |
2229 | if (unlikely((long)sum < 0)) | |
2230 | sum = 0; | |
2231 | ||
2232 | return sum; | |
2233 | } | |
2234 | ||
2235 | unsigned long long nr_context_switches(void) | |
2236 | { | |
cc94abfc SR |
2237 | int i; |
2238 | unsigned long long sum = 0; | |
1da177e4 | 2239 | |
0a945022 | 2240 | for_each_possible_cpu(i) |
1da177e4 LT |
2241 | sum += cpu_rq(i)->nr_switches; |
2242 | ||
2243 | return sum; | |
2244 | } | |
2245 | ||
2246 | unsigned long nr_iowait(void) | |
2247 | { | |
2248 | unsigned long i, sum = 0; | |
2249 | ||
0a945022 | 2250 | for_each_possible_cpu(i) |
1da177e4 LT |
2251 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2252 | ||
2253 | return sum; | |
2254 | } | |
2255 | ||
db1b1fef JS |
2256 | unsigned long nr_active(void) |
2257 | { | |
2258 | unsigned long i, running = 0, uninterruptible = 0; | |
2259 | ||
2260 | for_each_online_cpu(i) { | |
2261 | running += cpu_rq(i)->nr_running; | |
2262 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2263 | } | |
2264 | ||
2265 | if (unlikely((long)uninterruptible < 0)) | |
2266 | uninterruptible = 0; | |
2267 | ||
2268 | return running + uninterruptible; | |
2269 | } | |
2270 | ||
48f24c4d | 2271 | /* |
dd41f596 IM |
2272 | * Update rq->cpu_load[] statistics. This function is usually called every |
2273 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2274 | */ |
dd41f596 | 2275 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2276 | { |
495eca49 | 2277 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2278 | int i, scale; |
2279 | ||
2280 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2281 | |
2282 | /* Update our load: */ | |
2283 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2284 | unsigned long old_load, new_load; | |
2285 | ||
2286 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2287 | ||
2288 | old_load = this_rq->cpu_load[i]; | |
2289 | new_load = this_load; | |
a25707f3 IM |
2290 | /* |
2291 | * Round up the averaging division if load is increasing. This | |
2292 | * prevents us from getting stuck on 9 if the load is 10, for | |
2293 | * example. | |
2294 | */ | |
2295 | if (new_load > old_load) | |
2296 | new_load += scale-1; | |
dd41f596 IM |
2297 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2298 | } | |
48f24c4d IM |
2299 | } |
2300 | ||
dd41f596 IM |
2301 | #ifdef CONFIG_SMP |
2302 | ||
1da177e4 LT |
2303 | /* |
2304 | * double_rq_lock - safely lock two runqueues | |
2305 | * | |
2306 | * Note this does not disable interrupts like task_rq_lock, | |
2307 | * you need to do so manually before calling. | |
2308 | */ | |
70b97a7f | 2309 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2310 | __acquires(rq1->lock) |
2311 | __acquires(rq2->lock) | |
2312 | { | |
054b9108 | 2313 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2314 | if (rq1 == rq2) { |
2315 | spin_lock(&rq1->lock); | |
2316 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2317 | } else { | |
c96d145e | 2318 | if (rq1 < rq2) { |
1da177e4 LT |
2319 | spin_lock(&rq1->lock); |
2320 | spin_lock(&rq2->lock); | |
2321 | } else { | |
2322 | spin_lock(&rq2->lock); | |
2323 | spin_lock(&rq1->lock); | |
2324 | } | |
2325 | } | |
6e82a3be IM |
2326 | update_rq_clock(rq1); |
2327 | update_rq_clock(rq2); | |
1da177e4 LT |
2328 | } |
2329 | ||
2330 | /* | |
2331 | * double_rq_unlock - safely unlock two runqueues | |
2332 | * | |
2333 | * Note this does not restore interrupts like task_rq_unlock, | |
2334 | * you need to do so manually after calling. | |
2335 | */ | |
70b97a7f | 2336 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2337 | __releases(rq1->lock) |
2338 | __releases(rq2->lock) | |
2339 | { | |
2340 | spin_unlock(&rq1->lock); | |
2341 | if (rq1 != rq2) | |
2342 | spin_unlock(&rq2->lock); | |
2343 | else | |
2344 | __release(rq2->lock); | |
2345 | } | |
2346 | ||
2347 | /* | |
2348 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2349 | */ | |
e8fa1362 | 2350 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2351 | __releases(this_rq->lock) |
2352 | __acquires(busiest->lock) | |
2353 | __acquires(this_rq->lock) | |
2354 | { | |
e8fa1362 SR |
2355 | int ret = 0; |
2356 | ||
054b9108 KK |
2357 | if (unlikely(!irqs_disabled())) { |
2358 | /* printk() doesn't work good under rq->lock */ | |
2359 | spin_unlock(&this_rq->lock); | |
2360 | BUG_ON(1); | |
2361 | } | |
1da177e4 | 2362 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2363 | if (busiest < this_rq) { |
1da177e4 LT |
2364 | spin_unlock(&this_rq->lock); |
2365 | spin_lock(&busiest->lock); | |
2366 | spin_lock(&this_rq->lock); | |
e8fa1362 | 2367 | ret = 1; |
1da177e4 LT |
2368 | } else |
2369 | spin_lock(&busiest->lock); | |
2370 | } | |
e8fa1362 | 2371 | return ret; |
1da177e4 LT |
2372 | } |
2373 | ||
1da177e4 LT |
2374 | /* |
2375 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2376 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2377 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2378 | * the cpu_allowed mask is restored. |
2379 | */ | |
36c8b586 | 2380 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2381 | { |
70b97a7f | 2382 | struct migration_req req; |
1da177e4 | 2383 | unsigned long flags; |
70b97a7f | 2384 | struct rq *rq; |
1da177e4 LT |
2385 | |
2386 | rq = task_rq_lock(p, &flags); | |
2387 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2388 | || unlikely(cpu_is_offline(dest_cpu))) | |
2389 | goto out; | |
2390 | ||
2391 | /* force the process onto the specified CPU */ | |
2392 | if (migrate_task(p, dest_cpu, &req)) { | |
2393 | /* Need to wait for migration thread (might exit: take ref). */ | |
2394 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2395 | |
1da177e4 LT |
2396 | get_task_struct(mt); |
2397 | task_rq_unlock(rq, &flags); | |
2398 | wake_up_process(mt); | |
2399 | put_task_struct(mt); | |
2400 | wait_for_completion(&req.done); | |
36c8b586 | 2401 | |
1da177e4 LT |
2402 | return; |
2403 | } | |
2404 | out: | |
2405 | task_rq_unlock(rq, &flags); | |
2406 | } | |
2407 | ||
2408 | /* | |
476d139c NP |
2409 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2410 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2411 | */ |
2412 | void sched_exec(void) | |
2413 | { | |
1da177e4 | 2414 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2415 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2416 | put_cpu(); |
476d139c NP |
2417 | if (new_cpu != this_cpu) |
2418 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2419 | } |
2420 | ||
2421 | /* | |
2422 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2423 | * Both runqueues must be locked. | |
2424 | */ | |
dd41f596 IM |
2425 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2426 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2427 | { |
2e1cb74a | 2428 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2429 | set_task_cpu(p, this_cpu); |
dd41f596 | 2430 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2431 | /* |
2432 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2433 | * to be always true for them. | |
2434 | */ | |
dd41f596 | 2435 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2436 | } |
2437 | ||
2438 | /* | |
2439 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2440 | */ | |
858119e1 | 2441 | static |
70b97a7f | 2442 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2443 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2444 | int *all_pinned) |
1da177e4 LT |
2445 | { |
2446 | /* | |
2447 | * We do not migrate tasks that are: | |
2448 | * 1) running (obviously), or | |
2449 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2450 | * 3) are cache-hot on their current CPU. | |
2451 | */ | |
cc367732 IM |
2452 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2453 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2454 | return 0; |
cc367732 | 2455 | } |
81026794 NP |
2456 | *all_pinned = 0; |
2457 | ||
cc367732 IM |
2458 | if (task_running(rq, p)) { |
2459 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2460 | return 0; |
cc367732 | 2461 | } |
1da177e4 | 2462 | |
da84d961 IM |
2463 | /* |
2464 | * Aggressive migration if: | |
2465 | * 1) task is cache cold, or | |
2466 | * 2) too many balance attempts have failed. | |
2467 | */ | |
2468 | ||
6bc1665b IM |
2469 | if (!task_hot(p, rq->clock, sd) || |
2470 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2471 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2472 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2473 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2474 | schedstat_inc(p, se.nr_forced_migrations); |
2475 | } | |
da84d961 IM |
2476 | #endif |
2477 | return 1; | |
2478 | } | |
2479 | ||
cc367732 IM |
2480 | if (task_hot(p, rq->clock, sd)) { |
2481 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2482 | return 0; |
cc367732 | 2483 | } |
1da177e4 LT |
2484 | return 1; |
2485 | } | |
2486 | ||
e1d1484f PW |
2487 | static unsigned long |
2488 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2489 | unsigned long max_load_move, struct sched_domain *sd, | |
2490 | enum cpu_idle_type idle, int *all_pinned, | |
2491 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2492 | { |
b82d9fdd | 2493 | int loops = 0, pulled = 0, pinned = 0, skip_for_load; |
dd41f596 IM |
2494 | struct task_struct *p; |
2495 | long rem_load_move = max_load_move; | |
1da177e4 | 2496 | |
e1d1484f | 2497 | if (max_load_move == 0) |
1da177e4 LT |
2498 | goto out; |
2499 | ||
81026794 NP |
2500 | pinned = 1; |
2501 | ||
1da177e4 | 2502 | /* |
dd41f596 | 2503 | * Start the load-balancing iterator: |
1da177e4 | 2504 | */ |
dd41f596 IM |
2505 | p = iterator->start(iterator->arg); |
2506 | next: | |
b82d9fdd | 2507 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2508 | goto out; |
50ddd969 | 2509 | /* |
b82d9fdd | 2510 | * To help distribute high priority tasks across CPUs we don't |
50ddd969 PW |
2511 | * skip a task if it will be the highest priority task (i.e. smallest |
2512 | * prio value) on its new queue regardless of its load weight | |
2513 | */ | |
dd41f596 IM |
2514 | skip_for_load = (p->se.load.weight >> 1) > rem_load_move + |
2515 | SCHED_LOAD_SCALE_FUZZ; | |
a4ac01c3 | 2516 | if ((skip_for_load && p->prio >= *this_best_prio) || |
dd41f596 | 2517 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2518 | p = iterator->next(iterator->arg); |
2519 | goto next; | |
1da177e4 LT |
2520 | } |
2521 | ||
dd41f596 | 2522 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2523 | pulled++; |
dd41f596 | 2524 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2525 | |
2dd73a4f | 2526 | /* |
b82d9fdd | 2527 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 2528 | */ |
e1d1484f | 2529 | if (rem_load_move > 0) { |
a4ac01c3 PW |
2530 | if (p->prio < *this_best_prio) |
2531 | *this_best_prio = p->prio; | |
dd41f596 IM |
2532 | p = iterator->next(iterator->arg); |
2533 | goto next; | |
1da177e4 LT |
2534 | } |
2535 | out: | |
2536 | /* | |
e1d1484f | 2537 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
2538 | * so we can safely collect pull_task() stats here rather than |
2539 | * inside pull_task(). | |
2540 | */ | |
2541 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2542 | |
2543 | if (all_pinned) | |
2544 | *all_pinned = pinned; | |
e1d1484f PW |
2545 | |
2546 | return max_load_move - rem_load_move; | |
1da177e4 LT |
2547 | } |
2548 | ||
dd41f596 | 2549 | /* |
43010659 PW |
2550 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2551 | * this_rq, as part of a balancing operation within domain "sd". | |
2552 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2553 | * |
2554 | * Called with both runqueues locked. | |
2555 | */ | |
2556 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2557 | unsigned long max_load_move, |
dd41f596 IM |
2558 | struct sched_domain *sd, enum cpu_idle_type idle, |
2559 | int *all_pinned) | |
2560 | { | |
5522d5d5 | 2561 | const struct sched_class *class = sched_class_highest; |
43010659 | 2562 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2563 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
2564 | |
2565 | do { | |
43010659 PW |
2566 | total_load_moved += |
2567 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 2568 | max_load_move - total_load_moved, |
a4ac01c3 | 2569 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 2570 | class = class->next; |
43010659 | 2571 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 2572 | |
43010659 PW |
2573 | return total_load_moved > 0; |
2574 | } | |
2575 | ||
e1d1484f PW |
2576 | static int |
2577 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2578 | struct sched_domain *sd, enum cpu_idle_type idle, | |
2579 | struct rq_iterator *iterator) | |
2580 | { | |
2581 | struct task_struct *p = iterator->start(iterator->arg); | |
2582 | int pinned = 0; | |
2583 | ||
2584 | while (p) { | |
2585 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
2586 | pull_task(busiest, p, this_rq, this_cpu); | |
2587 | /* | |
2588 | * Right now, this is only the second place pull_task() | |
2589 | * is called, so we can safely collect pull_task() | |
2590 | * stats here rather than inside pull_task(). | |
2591 | */ | |
2592 | schedstat_inc(sd, lb_gained[idle]); | |
2593 | ||
2594 | return 1; | |
2595 | } | |
2596 | p = iterator->next(iterator->arg); | |
2597 | } | |
2598 | ||
2599 | return 0; | |
2600 | } | |
2601 | ||
43010659 PW |
2602 | /* |
2603 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
2604 | * part of active balancing operations within "domain". | |
2605 | * Returns 1 if successful and 0 otherwise. | |
2606 | * | |
2607 | * Called with both runqueues locked. | |
2608 | */ | |
2609 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2610 | struct sched_domain *sd, enum cpu_idle_type idle) | |
2611 | { | |
5522d5d5 | 2612 | const struct sched_class *class; |
43010659 PW |
2613 | |
2614 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 2615 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
2616 | return 1; |
2617 | ||
2618 | return 0; | |
dd41f596 IM |
2619 | } |
2620 | ||
1da177e4 LT |
2621 | /* |
2622 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2623 | * domain. It calculates and returns the amount of weighted load which |
2624 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2625 | */ |
2626 | static struct sched_group * | |
2627 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 IM |
2628 | unsigned long *imbalance, enum cpu_idle_type idle, |
2629 | int *sd_idle, cpumask_t *cpus, int *balance) | |
1da177e4 LT |
2630 | { |
2631 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2632 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2633 | unsigned long max_pull; |
2dd73a4f PW |
2634 | unsigned long busiest_load_per_task, busiest_nr_running; |
2635 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 2636 | int load_idx, group_imb = 0; |
5c45bf27 SS |
2637 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2638 | int power_savings_balance = 1; | |
2639 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2640 | unsigned long min_nr_running = ULONG_MAX; | |
2641 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2642 | #endif | |
1da177e4 LT |
2643 | |
2644 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2645 | busiest_load_per_task = busiest_nr_running = 0; |
2646 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2647 | if (idle == CPU_NOT_IDLE) |
7897986b | 2648 | load_idx = sd->busy_idx; |
d15bcfdb | 2649 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2650 | load_idx = sd->newidle_idx; |
2651 | else | |
2652 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2653 | |
2654 | do { | |
908a7c1b | 2655 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
2656 | int local_group; |
2657 | int i; | |
908a7c1b | 2658 | int __group_imb = 0; |
783609c6 | 2659 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2660 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2661 | |
2662 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2663 | ||
783609c6 SS |
2664 | if (local_group) |
2665 | balance_cpu = first_cpu(group->cpumask); | |
2666 | ||
1da177e4 | 2667 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2668 | sum_weighted_load = sum_nr_running = avg_load = 0; |
908a7c1b KC |
2669 | max_cpu_load = 0; |
2670 | min_cpu_load = ~0UL; | |
1da177e4 LT |
2671 | |
2672 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2673 | struct rq *rq; |
2674 | ||
2675 | if (!cpu_isset(i, *cpus)) | |
2676 | continue; | |
2677 | ||
2678 | rq = cpu_rq(i); | |
2dd73a4f | 2679 | |
9439aab8 | 2680 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
2681 | *sd_idle = 0; |
2682 | ||
1da177e4 | 2683 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2684 | if (local_group) { |
2685 | if (idle_cpu(i) && !first_idle_cpu) { | |
2686 | first_idle_cpu = 1; | |
2687 | balance_cpu = i; | |
2688 | } | |
2689 | ||
a2000572 | 2690 | load = target_load(i, load_idx); |
908a7c1b | 2691 | } else { |
a2000572 | 2692 | load = source_load(i, load_idx); |
908a7c1b KC |
2693 | if (load > max_cpu_load) |
2694 | max_cpu_load = load; | |
2695 | if (min_cpu_load > load) | |
2696 | min_cpu_load = load; | |
2697 | } | |
1da177e4 LT |
2698 | |
2699 | avg_load += load; | |
2dd73a4f | 2700 | sum_nr_running += rq->nr_running; |
dd41f596 | 2701 | sum_weighted_load += weighted_cpuload(i); |
1da177e4 LT |
2702 | } |
2703 | ||
783609c6 SS |
2704 | /* |
2705 | * First idle cpu or the first cpu(busiest) in this sched group | |
2706 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
2707 | * domains. In the newly idle case, we will allow all the cpu's |
2708 | * to do the newly idle load balance. | |
783609c6 | 2709 | */ |
9439aab8 SS |
2710 | if (idle != CPU_NEWLY_IDLE && local_group && |
2711 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
2712 | *balance = 0; |
2713 | goto ret; | |
2714 | } | |
2715 | ||
1da177e4 | 2716 | total_load += avg_load; |
5517d86b | 2717 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2718 | |
2719 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2720 | avg_load = sg_div_cpu_power(group, |
2721 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2722 | |
908a7c1b KC |
2723 | if ((max_cpu_load - min_cpu_load) > SCHED_LOAD_SCALE) |
2724 | __group_imb = 1; | |
2725 | ||
5517d86b | 2726 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2727 | |
1da177e4 LT |
2728 | if (local_group) { |
2729 | this_load = avg_load; | |
2730 | this = group; | |
2dd73a4f PW |
2731 | this_nr_running = sum_nr_running; |
2732 | this_load_per_task = sum_weighted_load; | |
2733 | } else if (avg_load > max_load && | |
908a7c1b | 2734 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
2735 | max_load = avg_load; |
2736 | busiest = group; | |
2dd73a4f PW |
2737 | busiest_nr_running = sum_nr_running; |
2738 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 2739 | group_imb = __group_imb; |
1da177e4 | 2740 | } |
5c45bf27 SS |
2741 | |
2742 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2743 | /* | |
2744 | * Busy processors will not participate in power savings | |
2745 | * balance. | |
2746 | */ | |
dd41f596 IM |
2747 | if (idle == CPU_NOT_IDLE || |
2748 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
2749 | goto group_next; | |
5c45bf27 SS |
2750 | |
2751 | /* | |
2752 | * If the local group is idle or completely loaded | |
2753 | * no need to do power savings balance at this domain | |
2754 | */ | |
2755 | if (local_group && (this_nr_running >= group_capacity || | |
2756 | !this_nr_running)) | |
2757 | power_savings_balance = 0; | |
2758 | ||
dd41f596 | 2759 | /* |
5c45bf27 SS |
2760 | * If a group is already running at full capacity or idle, |
2761 | * don't include that group in power savings calculations | |
dd41f596 IM |
2762 | */ |
2763 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 2764 | || !sum_nr_running) |
dd41f596 | 2765 | goto group_next; |
5c45bf27 | 2766 | |
dd41f596 | 2767 | /* |
5c45bf27 | 2768 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
2769 | * This is the group from where we need to pick up the load |
2770 | * for saving power | |
2771 | */ | |
2772 | if ((sum_nr_running < min_nr_running) || | |
2773 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
2774 | first_cpu(group->cpumask) < |
2775 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
2776 | group_min = group; |
2777 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
2778 | min_load_per_task = sum_weighted_load / |
2779 | sum_nr_running; | |
dd41f596 | 2780 | } |
5c45bf27 | 2781 | |
dd41f596 | 2782 | /* |
5c45bf27 | 2783 | * Calculate the group which is almost near its |
dd41f596 IM |
2784 | * capacity but still has some space to pick up some load |
2785 | * from other group and save more power | |
2786 | */ | |
2787 | if (sum_nr_running <= group_capacity - 1) { | |
2788 | if (sum_nr_running > leader_nr_running || | |
2789 | (sum_nr_running == leader_nr_running && | |
2790 | first_cpu(group->cpumask) > | |
2791 | first_cpu(group_leader->cpumask))) { | |
2792 | group_leader = group; | |
2793 | leader_nr_running = sum_nr_running; | |
2794 | } | |
48f24c4d | 2795 | } |
5c45bf27 SS |
2796 | group_next: |
2797 | #endif | |
1da177e4 LT |
2798 | group = group->next; |
2799 | } while (group != sd->groups); | |
2800 | ||
2dd73a4f | 2801 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
2802 | goto out_balanced; |
2803 | ||
2804 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
2805 | ||
2806 | if (this_load >= avg_load || | |
2807 | 100*max_load <= sd->imbalance_pct*this_load) | |
2808 | goto out_balanced; | |
2809 | ||
2dd73a4f | 2810 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
2811 | if (group_imb) |
2812 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
2813 | ||
1da177e4 LT |
2814 | /* |
2815 | * We're trying to get all the cpus to the average_load, so we don't | |
2816 | * want to push ourselves above the average load, nor do we wish to | |
2817 | * reduce the max loaded cpu below the average load, as either of these | |
2818 | * actions would just result in more rebalancing later, and ping-pong | |
2819 | * tasks around. Thus we look for the minimum possible imbalance. | |
2820 | * Negative imbalances (*we* are more loaded than anyone else) will | |
2821 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 2822 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
2823 | * appear as very large values with unsigned longs. |
2824 | */ | |
2dd73a4f PW |
2825 | if (max_load <= busiest_load_per_task) |
2826 | goto out_balanced; | |
2827 | ||
2828 | /* | |
2829 | * In the presence of smp nice balancing, certain scenarios can have | |
2830 | * max load less than avg load(as we skip the groups at or below | |
2831 | * its cpu_power, while calculating max_load..) | |
2832 | */ | |
2833 | if (max_load < avg_load) { | |
2834 | *imbalance = 0; | |
2835 | goto small_imbalance; | |
2836 | } | |
0c117f1b SS |
2837 | |
2838 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 2839 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 2840 | |
1da177e4 | 2841 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
2842 | *imbalance = min(max_pull * busiest->__cpu_power, |
2843 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
2844 | / SCHED_LOAD_SCALE; |
2845 | ||
2dd73a4f PW |
2846 | /* |
2847 | * if *imbalance is less than the average load per runnable task | |
2848 | * there is no gaurantee that any tasks will be moved so we'll have | |
2849 | * a think about bumping its value to force at least one task to be | |
2850 | * moved | |
2851 | */ | |
7fd0d2dd | 2852 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 2853 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
2854 | unsigned int imbn; |
2855 | ||
2856 | small_imbalance: | |
2857 | pwr_move = pwr_now = 0; | |
2858 | imbn = 2; | |
2859 | if (this_nr_running) { | |
2860 | this_load_per_task /= this_nr_running; | |
2861 | if (busiest_load_per_task > this_load_per_task) | |
2862 | imbn = 1; | |
2863 | } else | |
2864 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 2865 | |
dd41f596 IM |
2866 | if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >= |
2867 | busiest_load_per_task * imbn) { | |
2dd73a4f | 2868 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2869 | return busiest; |
2870 | } | |
2871 | ||
2872 | /* | |
2873 | * OK, we don't have enough imbalance to justify moving tasks, | |
2874 | * however we may be able to increase total CPU power used by | |
2875 | * moving them. | |
2876 | */ | |
2877 | ||
5517d86b ED |
2878 | pwr_now += busiest->__cpu_power * |
2879 | min(busiest_load_per_task, max_load); | |
2880 | pwr_now += this->__cpu_power * | |
2881 | min(this_load_per_task, this_load); | |
1da177e4 LT |
2882 | pwr_now /= SCHED_LOAD_SCALE; |
2883 | ||
2884 | /* Amount of load we'd subtract */ | |
5517d86b ED |
2885 | tmp = sg_div_cpu_power(busiest, |
2886 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 2887 | if (max_load > tmp) |
5517d86b | 2888 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 2889 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
2890 | |
2891 | /* Amount of load we'd add */ | |
5517d86b | 2892 | if (max_load * busiest->__cpu_power < |
33859f7f | 2893 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
2894 | tmp = sg_div_cpu_power(this, |
2895 | max_load * busiest->__cpu_power); | |
1da177e4 | 2896 | else |
5517d86b ED |
2897 | tmp = sg_div_cpu_power(this, |
2898 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
2899 | pwr_move += this->__cpu_power * | |
2900 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
2901 | pwr_move /= SCHED_LOAD_SCALE; |
2902 | ||
2903 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
2904 | if (pwr_move > pwr_now) |
2905 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
2906 | } |
2907 | ||
1da177e4 LT |
2908 | return busiest; |
2909 | ||
2910 | out_balanced: | |
5c45bf27 | 2911 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 2912 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 2913 | goto ret; |
1da177e4 | 2914 | |
5c45bf27 SS |
2915 | if (this == group_leader && group_leader != group_min) { |
2916 | *imbalance = min_load_per_task; | |
2917 | return group_min; | |
2918 | } | |
5c45bf27 | 2919 | #endif |
783609c6 | 2920 | ret: |
1da177e4 LT |
2921 | *imbalance = 0; |
2922 | return NULL; | |
2923 | } | |
2924 | ||
2925 | /* | |
2926 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
2927 | */ | |
70b97a7f | 2928 | static struct rq * |
d15bcfdb | 2929 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
0a2966b4 | 2930 | unsigned long imbalance, cpumask_t *cpus) |
1da177e4 | 2931 | { |
70b97a7f | 2932 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 2933 | unsigned long max_load = 0; |
1da177e4 LT |
2934 | int i; |
2935 | ||
2936 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 2937 | unsigned long wl; |
0a2966b4 CL |
2938 | |
2939 | if (!cpu_isset(i, *cpus)) | |
2940 | continue; | |
2941 | ||
48f24c4d | 2942 | rq = cpu_rq(i); |
dd41f596 | 2943 | wl = weighted_cpuload(i); |
2dd73a4f | 2944 | |
dd41f596 | 2945 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 2946 | continue; |
1da177e4 | 2947 | |
dd41f596 IM |
2948 | if (wl > max_load) { |
2949 | max_load = wl; | |
48f24c4d | 2950 | busiest = rq; |
1da177e4 LT |
2951 | } |
2952 | } | |
2953 | ||
2954 | return busiest; | |
2955 | } | |
2956 | ||
77391d71 NP |
2957 | /* |
2958 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
2959 | * so long as it is large enough. | |
2960 | */ | |
2961 | #define MAX_PINNED_INTERVAL 512 | |
2962 | ||
1da177e4 LT |
2963 | /* |
2964 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2965 | * tasks if there is an imbalance. | |
1da177e4 | 2966 | */ |
70b97a7f | 2967 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 2968 | struct sched_domain *sd, enum cpu_idle_type idle, |
783609c6 | 2969 | int *balance) |
1da177e4 | 2970 | { |
43010659 | 2971 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 2972 | struct sched_group *group; |
1da177e4 | 2973 | unsigned long imbalance; |
70b97a7f | 2974 | struct rq *busiest; |
0a2966b4 | 2975 | cpumask_t cpus = CPU_MASK_ALL; |
fe2eea3f | 2976 | unsigned long flags; |
5969fe06 | 2977 | |
89c4710e SS |
2978 | /* |
2979 | * When power savings policy is enabled for the parent domain, idle | |
2980 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 2981 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 2982 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 2983 | */ |
d15bcfdb | 2984 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2985 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2986 | sd_idle = 1; |
1da177e4 | 2987 | |
2d72376b | 2988 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 2989 | |
0a2966b4 CL |
2990 | redo: |
2991 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
783609c6 SS |
2992 | &cpus, balance); |
2993 | ||
06066714 | 2994 | if (*balance == 0) |
783609c6 | 2995 | goto out_balanced; |
783609c6 | 2996 | |
1da177e4 LT |
2997 | if (!group) { |
2998 | schedstat_inc(sd, lb_nobusyg[idle]); | |
2999 | goto out_balanced; | |
3000 | } | |
3001 | ||
0a2966b4 | 3002 | busiest = find_busiest_queue(group, idle, imbalance, &cpus); |
1da177e4 LT |
3003 | if (!busiest) { |
3004 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3005 | goto out_balanced; | |
3006 | } | |
3007 | ||
db935dbd | 3008 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3009 | |
3010 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3011 | ||
43010659 | 3012 | ld_moved = 0; |
1da177e4 LT |
3013 | if (busiest->nr_running > 1) { |
3014 | /* | |
3015 | * Attempt to move tasks. If find_busiest_group has found | |
3016 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3017 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3018 | * correctly treated as an imbalance. |
3019 | */ | |
fe2eea3f | 3020 | local_irq_save(flags); |
e17224bf | 3021 | double_rq_lock(this_rq, busiest); |
43010659 | 3022 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3023 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3024 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3025 | local_irq_restore(flags); |
81026794 | 3026 | |
46cb4b7c SS |
3027 | /* |
3028 | * some other cpu did the load balance for us. | |
3029 | */ | |
43010659 | 3030 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3031 | resched_cpu(this_cpu); |
3032 | ||
81026794 | 3033 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 CL |
3034 | if (unlikely(all_pinned)) { |
3035 | cpu_clear(cpu_of(busiest), cpus); | |
3036 | if (!cpus_empty(cpus)) | |
3037 | goto redo; | |
81026794 | 3038 | goto out_balanced; |
0a2966b4 | 3039 | } |
1da177e4 | 3040 | } |
81026794 | 3041 | |
43010659 | 3042 | if (!ld_moved) { |
1da177e4 LT |
3043 | schedstat_inc(sd, lb_failed[idle]); |
3044 | sd->nr_balance_failed++; | |
3045 | ||
3046 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3047 | |
fe2eea3f | 3048 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3049 | |
3050 | /* don't kick the migration_thread, if the curr | |
3051 | * task on busiest cpu can't be moved to this_cpu | |
3052 | */ | |
3053 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3054 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3055 | all_pinned = 1; |
3056 | goto out_one_pinned; | |
3057 | } | |
3058 | ||
1da177e4 LT |
3059 | if (!busiest->active_balance) { |
3060 | busiest->active_balance = 1; | |
3061 | busiest->push_cpu = this_cpu; | |
81026794 | 3062 | active_balance = 1; |
1da177e4 | 3063 | } |
fe2eea3f | 3064 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3065 | if (active_balance) |
1da177e4 LT |
3066 | wake_up_process(busiest->migration_thread); |
3067 | ||
3068 | /* | |
3069 | * We've kicked active balancing, reset the failure | |
3070 | * counter. | |
3071 | */ | |
39507451 | 3072 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3073 | } |
81026794 | 3074 | } else |
1da177e4 LT |
3075 | sd->nr_balance_failed = 0; |
3076 | ||
81026794 | 3077 | if (likely(!active_balance)) { |
1da177e4 LT |
3078 | /* We were unbalanced, so reset the balancing interval */ |
3079 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
3080 | } else { |
3081 | /* | |
3082 | * If we've begun active balancing, start to back off. This | |
3083 | * case may not be covered by the all_pinned logic if there | |
3084 | * is only 1 task on the busy runqueue (because we don't call | |
3085 | * move_tasks). | |
3086 | */ | |
3087 | if (sd->balance_interval < sd->max_interval) | |
3088 | sd->balance_interval *= 2; | |
1da177e4 LT |
3089 | } |
3090 | ||
43010659 | 3091 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3092 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3093 | return -1; |
43010659 | 3094 | return ld_moved; |
1da177e4 LT |
3095 | |
3096 | out_balanced: | |
1da177e4 LT |
3097 | schedstat_inc(sd, lb_balanced[idle]); |
3098 | ||
16cfb1c0 | 3099 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
3100 | |
3101 | out_one_pinned: | |
1da177e4 | 3102 | /* tune up the balancing interval */ |
77391d71 NP |
3103 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
3104 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
3105 | sd->balance_interval *= 2; |
3106 | ||
48f24c4d | 3107 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3108 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3109 | return -1; |
1da177e4 LT |
3110 | return 0; |
3111 | } | |
3112 | ||
3113 | /* | |
3114 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3115 | * tasks if there is an imbalance. | |
3116 | * | |
d15bcfdb | 3117 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
3118 | * this_rq is locked. |
3119 | */ | |
48f24c4d | 3120 | static int |
70b97a7f | 3121 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
3122 | { |
3123 | struct sched_group *group; | |
70b97a7f | 3124 | struct rq *busiest = NULL; |
1da177e4 | 3125 | unsigned long imbalance; |
43010659 | 3126 | int ld_moved = 0; |
5969fe06 | 3127 | int sd_idle = 0; |
969bb4e4 | 3128 | int all_pinned = 0; |
0a2966b4 | 3129 | cpumask_t cpus = CPU_MASK_ALL; |
5969fe06 | 3130 | |
89c4710e SS |
3131 | /* |
3132 | * When power savings policy is enabled for the parent domain, idle | |
3133 | * sibling can pick up load irrespective of busy siblings. In this case, | |
3134 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 3135 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
3136 | */ |
3137 | if (sd->flags & SD_SHARE_CPUPOWER && | |
3138 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3139 | sd_idle = 1; |
1da177e4 | 3140 | |
2d72376b | 3141 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 3142 | redo: |
d15bcfdb | 3143 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
783609c6 | 3144 | &sd_idle, &cpus, NULL); |
1da177e4 | 3145 | if (!group) { |
d15bcfdb | 3146 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3147 | goto out_balanced; |
1da177e4 LT |
3148 | } |
3149 | ||
d15bcfdb | 3150 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, |
0a2966b4 | 3151 | &cpus); |
db935dbd | 3152 | if (!busiest) { |
d15bcfdb | 3153 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3154 | goto out_balanced; |
1da177e4 LT |
3155 | } |
3156 | ||
db935dbd NP |
3157 | BUG_ON(busiest == this_rq); |
3158 | ||
d15bcfdb | 3159 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 3160 | |
43010659 | 3161 | ld_moved = 0; |
d6d5cfaf NP |
3162 | if (busiest->nr_running > 1) { |
3163 | /* Attempt to move tasks */ | |
3164 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
3165 | /* this_rq->clock is already updated */ |
3166 | update_rq_clock(busiest); | |
43010659 | 3167 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
3168 | imbalance, sd, CPU_NEWLY_IDLE, |
3169 | &all_pinned); | |
d6d5cfaf | 3170 | spin_unlock(&busiest->lock); |
0a2966b4 | 3171 | |
969bb4e4 | 3172 | if (unlikely(all_pinned)) { |
0a2966b4 CL |
3173 | cpu_clear(cpu_of(busiest), cpus); |
3174 | if (!cpus_empty(cpus)) | |
3175 | goto redo; | |
3176 | } | |
d6d5cfaf NP |
3177 | } |
3178 | ||
43010659 | 3179 | if (!ld_moved) { |
d15bcfdb | 3180 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
3181 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
3182 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
3183 | return -1; |
3184 | } else | |
16cfb1c0 | 3185 | sd->nr_balance_failed = 0; |
1da177e4 | 3186 | |
43010659 | 3187 | return ld_moved; |
16cfb1c0 NP |
3188 | |
3189 | out_balanced: | |
d15bcfdb | 3190 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 3191 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3192 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3193 | return -1; |
16cfb1c0 | 3194 | sd->nr_balance_failed = 0; |
48f24c4d | 3195 | |
16cfb1c0 | 3196 | return 0; |
1da177e4 LT |
3197 | } |
3198 | ||
3199 | /* | |
3200 | * idle_balance is called by schedule() if this_cpu is about to become | |
3201 | * idle. Attempts to pull tasks from other CPUs. | |
3202 | */ | |
70b97a7f | 3203 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
3204 | { |
3205 | struct sched_domain *sd; | |
dd41f596 IM |
3206 | int pulled_task = -1; |
3207 | unsigned long next_balance = jiffies + HZ; | |
1da177e4 LT |
3208 | |
3209 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
3210 | unsigned long interval; |
3211 | ||
3212 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
3213 | continue; | |
3214 | ||
3215 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 3216 | /* If we've pulled tasks over stop searching: */ |
1bd77f2d | 3217 | pulled_task = load_balance_newidle(this_cpu, |
92c4ca5c CL |
3218 | this_rq, sd); |
3219 | ||
3220 | interval = msecs_to_jiffies(sd->balance_interval); | |
3221 | if (time_after(next_balance, sd->last_balance + interval)) | |
3222 | next_balance = sd->last_balance + interval; | |
3223 | if (pulled_task) | |
3224 | break; | |
1da177e4 | 3225 | } |
dd41f596 | 3226 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
3227 | /* |
3228 | * We are going idle. next_balance may be set based on | |
3229 | * a busy processor. So reset next_balance. | |
3230 | */ | |
3231 | this_rq->next_balance = next_balance; | |
dd41f596 | 3232 | } |
1da177e4 LT |
3233 | } |
3234 | ||
3235 | /* | |
3236 | * active_load_balance is run by migration threads. It pushes running tasks | |
3237 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
3238 | * running on each physical CPU where possible, and avoids physical / | |
3239 | * logical imbalances. | |
3240 | * | |
3241 | * Called with busiest_rq locked. | |
3242 | */ | |
70b97a7f | 3243 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3244 | { |
39507451 | 3245 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3246 | struct sched_domain *sd; |
3247 | struct rq *target_rq; | |
39507451 | 3248 | |
48f24c4d | 3249 | /* Is there any task to move? */ |
39507451 | 3250 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3251 | return; |
3252 | ||
3253 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3254 | |
3255 | /* | |
39507451 | 3256 | * This condition is "impossible", if it occurs |
41a2d6cf | 3257 | * we need to fix it. Originally reported by |
39507451 | 3258 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3259 | */ |
39507451 | 3260 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3261 | |
39507451 NP |
3262 | /* move a task from busiest_rq to target_rq */ |
3263 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3264 | update_rq_clock(busiest_rq); |
3265 | update_rq_clock(target_rq); | |
39507451 NP |
3266 | |
3267 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3268 | for_each_domain(target_cpu, sd) { |
39507451 | 3269 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 3270 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 3271 | break; |
c96d145e | 3272 | } |
39507451 | 3273 | |
48f24c4d | 3274 | if (likely(sd)) { |
2d72376b | 3275 | schedstat_inc(sd, alb_count); |
39507451 | 3276 | |
43010659 PW |
3277 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3278 | sd, CPU_IDLE)) | |
48f24c4d IM |
3279 | schedstat_inc(sd, alb_pushed); |
3280 | else | |
3281 | schedstat_inc(sd, alb_failed); | |
3282 | } | |
39507451 | 3283 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
3284 | } |
3285 | ||
46cb4b7c SS |
3286 | #ifdef CONFIG_NO_HZ |
3287 | static struct { | |
3288 | atomic_t load_balancer; | |
41a2d6cf | 3289 | cpumask_t cpu_mask; |
46cb4b7c SS |
3290 | } nohz ____cacheline_aligned = { |
3291 | .load_balancer = ATOMIC_INIT(-1), | |
3292 | .cpu_mask = CPU_MASK_NONE, | |
3293 | }; | |
3294 | ||
7835b98b | 3295 | /* |
46cb4b7c SS |
3296 | * This routine will try to nominate the ilb (idle load balancing) |
3297 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3298 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3299 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3300 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3301 | * arrives... | |
3302 | * | |
3303 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3304 | * for idle load balancing. ilb owner will still be part of | |
3305 | * nohz.cpu_mask.. | |
7835b98b | 3306 | * |
46cb4b7c SS |
3307 | * While stopping the tick, this cpu will become the ilb owner if there |
3308 | * is no other owner. And will be the owner till that cpu becomes busy | |
3309 | * or if all cpus in the system stop their ticks at which point | |
3310 | * there is no need for ilb owner. | |
3311 | * | |
3312 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3313 | * next busy scheduler_tick() | |
3314 | */ | |
3315 | int select_nohz_load_balancer(int stop_tick) | |
3316 | { | |
3317 | int cpu = smp_processor_id(); | |
3318 | ||
3319 | if (stop_tick) { | |
3320 | cpu_set(cpu, nohz.cpu_mask); | |
3321 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3322 | ||
3323 | /* | |
3324 | * If we are going offline and still the leader, give up! | |
3325 | */ | |
3326 | if (cpu_is_offline(cpu) && | |
3327 | atomic_read(&nohz.load_balancer) == cpu) { | |
3328 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3329 | BUG(); | |
3330 | return 0; | |
3331 | } | |
3332 | ||
3333 | /* time for ilb owner also to sleep */ | |
3334 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3335 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3336 | atomic_set(&nohz.load_balancer, -1); | |
3337 | return 0; | |
3338 | } | |
3339 | ||
3340 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3341 | /* make me the ilb owner */ | |
3342 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3343 | return 1; | |
3344 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3345 | return 1; | |
3346 | } else { | |
3347 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3348 | return 0; | |
3349 | ||
3350 | cpu_clear(cpu, nohz.cpu_mask); | |
3351 | ||
3352 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3353 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3354 | BUG(); | |
3355 | } | |
3356 | return 0; | |
3357 | } | |
3358 | #endif | |
3359 | ||
3360 | static DEFINE_SPINLOCK(balancing); | |
3361 | ||
3362 | /* | |
7835b98b CL |
3363 | * It checks each scheduling domain to see if it is due to be balanced, |
3364 | * and initiates a balancing operation if so. | |
3365 | * | |
3366 | * Balancing parameters are set up in arch_init_sched_domains. | |
3367 | */ | |
a9957449 | 3368 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3369 | { |
46cb4b7c SS |
3370 | int balance = 1; |
3371 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3372 | unsigned long interval; |
3373 | struct sched_domain *sd; | |
46cb4b7c | 3374 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3375 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3376 | int update_next_balance = 0; |
1da177e4 | 3377 | |
46cb4b7c | 3378 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3379 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3380 | continue; | |
3381 | ||
3382 | interval = sd->balance_interval; | |
d15bcfdb | 3383 | if (idle != CPU_IDLE) |
1da177e4 LT |
3384 | interval *= sd->busy_factor; |
3385 | ||
3386 | /* scale ms to jiffies */ | |
3387 | interval = msecs_to_jiffies(interval); | |
3388 | if (unlikely(!interval)) | |
3389 | interval = 1; | |
dd41f596 IM |
3390 | if (interval > HZ*NR_CPUS/10) |
3391 | interval = HZ*NR_CPUS/10; | |
3392 | ||
1da177e4 | 3393 | |
08c183f3 CL |
3394 | if (sd->flags & SD_SERIALIZE) { |
3395 | if (!spin_trylock(&balancing)) | |
3396 | goto out; | |
3397 | } | |
3398 | ||
c9819f45 | 3399 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
46cb4b7c | 3400 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
3401 | /* |
3402 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3403 | * longer idle, or one of our SMT siblings is |
3404 | * not idle. | |
3405 | */ | |
d15bcfdb | 3406 | idle = CPU_NOT_IDLE; |
1da177e4 | 3407 | } |
1bd77f2d | 3408 | sd->last_balance = jiffies; |
1da177e4 | 3409 | } |
08c183f3 CL |
3410 | if (sd->flags & SD_SERIALIZE) |
3411 | spin_unlock(&balancing); | |
3412 | out: | |
f549da84 | 3413 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3414 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3415 | update_next_balance = 1; |
3416 | } | |
783609c6 SS |
3417 | |
3418 | /* | |
3419 | * Stop the load balance at this level. There is another | |
3420 | * CPU in our sched group which is doing load balancing more | |
3421 | * actively. | |
3422 | */ | |
3423 | if (!balance) | |
3424 | break; | |
1da177e4 | 3425 | } |
f549da84 SS |
3426 | |
3427 | /* | |
3428 | * next_balance will be updated only when there is a need. | |
3429 | * When the cpu is attached to null domain for ex, it will not be | |
3430 | * updated. | |
3431 | */ | |
3432 | if (likely(update_next_balance)) | |
3433 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3434 | } |
3435 | ||
3436 | /* | |
3437 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3438 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3439 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3440 | */ | |
3441 | static void run_rebalance_domains(struct softirq_action *h) | |
3442 | { | |
dd41f596 IM |
3443 | int this_cpu = smp_processor_id(); |
3444 | struct rq *this_rq = cpu_rq(this_cpu); | |
3445 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3446 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3447 | |
dd41f596 | 3448 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3449 | |
3450 | #ifdef CONFIG_NO_HZ | |
3451 | /* | |
3452 | * If this cpu is the owner for idle load balancing, then do the | |
3453 | * balancing on behalf of the other idle cpus whose ticks are | |
3454 | * stopped. | |
3455 | */ | |
dd41f596 IM |
3456 | if (this_rq->idle_at_tick && |
3457 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3458 | cpumask_t cpus = nohz.cpu_mask; |
3459 | struct rq *rq; | |
3460 | int balance_cpu; | |
3461 | ||
dd41f596 | 3462 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3463 | for_each_cpu_mask(balance_cpu, cpus) { |
3464 | /* | |
3465 | * If this cpu gets work to do, stop the load balancing | |
3466 | * work being done for other cpus. Next load | |
3467 | * balancing owner will pick it up. | |
3468 | */ | |
3469 | if (need_resched()) | |
3470 | break; | |
3471 | ||
de0cf899 | 3472 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3473 | |
3474 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3475 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3476 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3477 | } |
3478 | } | |
3479 | #endif | |
3480 | } | |
3481 | ||
3482 | /* | |
3483 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3484 | * | |
3485 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3486 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3487 | * if the whole system is idle. | |
3488 | */ | |
dd41f596 | 3489 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3490 | { |
46cb4b7c SS |
3491 | #ifdef CONFIG_NO_HZ |
3492 | /* | |
3493 | * If we were in the nohz mode recently and busy at the current | |
3494 | * scheduler tick, then check if we need to nominate new idle | |
3495 | * load balancer. | |
3496 | */ | |
3497 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3498 | rq->in_nohz_recently = 0; | |
3499 | ||
3500 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3501 | cpu_clear(cpu, nohz.cpu_mask); | |
3502 | atomic_set(&nohz.load_balancer, -1); | |
3503 | } | |
3504 | ||
3505 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3506 | /* | |
3507 | * simple selection for now: Nominate the | |
3508 | * first cpu in the nohz list to be the next | |
3509 | * ilb owner. | |
3510 | * | |
3511 | * TBD: Traverse the sched domains and nominate | |
3512 | * the nearest cpu in the nohz.cpu_mask. | |
3513 | */ | |
3514 | int ilb = first_cpu(nohz.cpu_mask); | |
3515 | ||
3516 | if (ilb != NR_CPUS) | |
3517 | resched_cpu(ilb); | |
3518 | } | |
3519 | } | |
3520 | ||
3521 | /* | |
3522 | * If this cpu is idle and doing idle load balancing for all the | |
3523 | * cpus with ticks stopped, is it time for that to stop? | |
3524 | */ | |
3525 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3526 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3527 | resched_cpu(cpu); | |
3528 | return; | |
3529 | } | |
3530 | ||
3531 | /* | |
3532 | * If this cpu is idle and the idle load balancing is done by | |
3533 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3534 | */ | |
3535 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3536 | cpu_isset(cpu, nohz.cpu_mask)) | |
3537 | return; | |
3538 | #endif | |
3539 | if (time_after_eq(jiffies, rq->next_balance)) | |
3540 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3541 | } |
dd41f596 IM |
3542 | |
3543 | #else /* CONFIG_SMP */ | |
3544 | ||
1da177e4 LT |
3545 | /* |
3546 | * on UP we do not need to balance between CPUs: | |
3547 | */ | |
70b97a7f | 3548 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3549 | { |
3550 | } | |
dd41f596 | 3551 | |
1da177e4 LT |
3552 | #endif |
3553 | ||
1da177e4 LT |
3554 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3555 | ||
3556 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3557 | ||
3558 | /* | |
41b86e9c IM |
3559 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
3560 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 3561 | */ |
41b86e9c | 3562 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 3563 | { |
1da177e4 | 3564 | unsigned long flags; |
41b86e9c IM |
3565 | u64 ns, delta_exec; |
3566 | struct rq *rq; | |
48f24c4d | 3567 | |
41b86e9c IM |
3568 | rq = task_rq_lock(p, &flags); |
3569 | ns = p->se.sum_exec_runtime; | |
051a1d1a | 3570 | if (task_current(rq, p)) { |
a8e504d2 IM |
3571 | update_rq_clock(rq); |
3572 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
3573 | if ((s64)delta_exec > 0) |
3574 | ns += delta_exec; | |
3575 | } | |
3576 | task_rq_unlock(rq, &flags); | |
48f24c4d | 3577 | |
1da177e4 LT |
3578 | return ns; |
3579 | } | |
3580 | ||
1da177e4 LT |
3581 | /* |
3582 | * Account user cpu time to a process. | |
3583 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
3584 | * @cputime: the cpu time spent in user space since the last update |
3585 | */ | |
3586 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3587 | { | |
3588 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3589 | cputime64_t tmp; | |
3590 | ||
3591 | p->utime = cputime_add(p->utime, cputime); | |
3592 | ||
3593 | /* Add user time to cpustat. */ | |
3594 | tmp = cputime_to_cputime64(cputime); | |
3595 | if (TASK_NICE(p) > 0) | |
3596 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3597 | else | |
3598 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3599 | } | |
3600 | ||
94886b84 LV |
3601 | /* |
3602 | * Account guest cpu time to a process. | |
3603 | * @p: the process that the cpu time gets accounted to | |
3604 | * @cputime: the cpu time spent in virtual machine since the last update | |
3605 | */ | |
f7402e03 | 3606 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
3607 | { |
3608 | cputime64_t tmp; | |
3609 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3610 | ||
3611 | tmp = cputime_to_cputime64(cputime); | |
3612 | ||
3613 | p->utime = cputime_add(p->utime, cputime); | |
3614 | p->gtime = cputime_add(p->gtime, cputime); | |
3615 | ||
3616 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3617 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3618 | } | |
3619 | ||
c66f08be MN |
3620 | /* |
3621 | * Account scaled user cpu time to a process. | |
3622 | * @p: the process that the cpu time gets accounted to | |
3623 | * @cputime: the cpu time spent in user space since the last update | |
3624 | */ | |
3625 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
3626 | { | |
3627 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
3628 | } | |
3629 | ||
1da177e4 LT |
3630 | /* |
3631 | * Account system cpu time to a process. | |
3632 | * @p: the process that the cpu time gets accounted to | |
3633 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3634 | * @cputime: the cpu time spent in kernel space since the last update | |
3635 | */ | |
3636 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3637 | cputime_t cputime) | |
3638 | { | |
3639 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3640 | struct rq *rq = this_rq(); |
1da177e4 LT |
3641 | cputime64_t tmp; |
3642 | ||
9778385d CB |
3643 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) |
3644 | return account_guest_time(p, cputime); | |
94886b84 | 3645 | |
1da177e4 LT |
3646 | p->stime = cputime_add(p->stime, cputime); |
3647 | ||
3648 | /* Add system time to cpustat. */ | |
3649 | tmp = cputime_to_cputime64(cputime); | |
3650 | if (hardirq_count() - hardirq_offset) | |
3651 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3652 | else if (softirq_count()) | |
3653 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 3654 | else if (p != rq->idle) |
1da177e4 | 3655 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 3656 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
3657 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
3658 | else | |
3659 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3660 | /* Account for system time used */ | |
3661 | acct_update_integrals(p); | |
1da177e4 LT |
3662 | } |
3663 | ||
c66f08be MN |
3664 | /* |
3665 | * Account scaled system cpu time to a process. | |
3666 | * @p: the process that the cpu time gets accounted to | |
3667 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3668 | * @cputime: the cpu time spent in kernel space since the last update | |
3669 | */ | |
3670 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
3671 | { | |
3672 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
3673 | } | |
3674 | ||
1da177e4 LT |
3675 | /* |
3676 | * Account for involuntary wait time. | |
3677 | * @p: the process from which the cpu time has been stolen | |
3678 | * @steal: the cpu time spent in involuntary wait | |
3679 | */ | |
3680 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3681 | { | |
3682 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3683 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3684 | struct rq *rq = this_rq(); |
1da177e4 LT |
3685 | |
3686 | if (p == rq->idle) { | |
3687 | p->stime = cputime_add(p->stime, steal); | |
3688 | if (atomic_read(&rq->nr_iowait) > 0) | |
3689 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3690 | else | |
3691 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 3692 | } else |
1da177e4 LT |
3693 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
3694 | } | |
3695 | ||
7835b98b CL |
3696 | /* |
3697 | * This function gets called by the timer code, with HZ frequency. | |
3698 | * We call it with interrupts disabled. | |
3699 | * | |
3700 | * It also gets called by the fork code, when changing the parent's | |
3701 | * timeslices. | |
3702 | */ | |
3703 | void scheduler_tick(void) | |
3704 | { | |
7835b98b CL |
3705 | int cpu = smp_processor_id(); |
3706 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3707 | struct task_struct *curr = rq->curr; |
529c7726 | 3708 | u64 next_tick = rq->tick_timestamp + TICK_NSEC; |
dd41f596 IM |
3709 | |
3710 | spin_lock(&rq->lock); | |
546fe3c9 | 3711 | __update_rq_clock(rq); |
529c7726 IM |
3712 | /* |
3713 | * Let rq->clock advance by at least TICK_NSEC: | |
3714 | */ | |
cc203d24 | 3715 | if (unlikely(rq->clock < next_tick)) { |
529c7726 | 3716 | rq->clock = next_tick; |
cc203d24 GC |
3717 | rq->clock_underflows++; |
3718 | } | |
529c7726 | 3719 | rq->tick_timestamp = rq->clock; |
f1a438d8 | 3720 | update_cpu_load(rq); |
fa85ae24 PZ |
3721 | curr->sched_class->task_tick(rq, curr, 0); |
3722 | update_sched_rt_period(rq); | |
dd41f596 | 3723 | spin_unlock(&rq->lock); |
7835b98b | 3724 | |
e418e1c2 | 3725 | #ifdef CONFIG_SMP |
dd41f596 IM |
3726 | rq->idle_at_tick = idle_cpu(cpu); |
3727 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3728 | #endif |
1da177e4 LT |
3729 | } |
3730 | ||
1da177e4 LT |
3731 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3732 | ||
43627582 | 3733 | void __kprobes add_preempt_count(int val) |
1da177e4 LT |
3734 | { |
3735 | /* | |
3736 | * Underflow? | |
3737 | */ | |
9a11b49a IM |
3738 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3739 | return; | |
1da177e4 LT |
3740 | preempt_count() += val; |
3741 | /* | |
3742 | * Spinlock count overflowing soon? | |
3743 | */ | |
33859f7f MOS |
3744 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3745 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3746 | } |
3747 | EXPORT_SYMBOL(add_preempt_count); | |
3748 | ||
43627582 | 3749 | void __kprobes sub_preempt_count(int val) |
1da177e4 LT |
3750 | { |
3751 | /* | |
3752 | * Underflow? | |
3753 | */ | |
9a11b49a IM |
3754 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3755 | return; | |
1da177e4 LT |
3756 | /* |
3757 | * Is the spinlock portion underflowing? | |
3758 | */ | |
9a11b49a IM |
3759 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3760 | !(preempt_count() & PREEMPT_MASK))) | |
3761 | return; | |
3762 | ||
1da177e4 LT |
3763 | preempt_count() -= val; |
3764 | } | |
3765 | EXPORT_SYMBOL(sub_preempt_count); | |
3766 | ||
3767 | #endif | |
3768 | ||
3769 | /* | |
dd41f596 | 3770 | * Print scheduling while atomic bug: |
1da177e4 | 3771 | */ |
dd41f596 | 3772 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3773 | { |
838225b4 SS |
3774 | struct pt_regs *regs = get_irq_regs(); |
3775 | ||
3776 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
3777 | prev->comm, prev->pid, preempt_count()); | |
3778 | ||
dd41f596 IM |
3779 | debug_show_held_locks(prev); |
3780 | if (irqs_disabled()) | |
3781 | print_irqtrace_events(prev); | |
838225b4 SS |
3782 | |
3783 | if (regs) | |
3784 | show_regs(regs); | |
3785 | else | |
3786 | dump_stack(); | |
dd41f596 | 3787 | } |
1da177e4 | 3788 | |
dd41f596 IM |
3789 | /* |
3790 | * Various schedule()-time debugging checks and statistics: | |
3791 | */ | |
3792 | static inline void schedule_debug(struct task_struct *prev) | |
3793 | { | |
1da177e4 | 3794 | /* |
41a2d6cf | 3795 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3796 | * schedule() atomically, we ignore that path for now. |
3797 | * Otherwise, whine if we are scheduling when we should not be. | |
3798 | */ | |
dd41f596 IM |
3799 | if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state)) |
3800 | __schedule_bug(prev); | |
3801 | ||
1da177e4 LT |
3802 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3803 | ||
2d72376b | 3804 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3805 | #ifdef CONFIG_SCHEDSTATS |
3806 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3807 | schedstat_inc(this_rq(), bkl_count); |
3808 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3809 | } |
3810 | #endif | |
dd41f596 IM |
3811 | } |
3812 | ||
3813 | /* | |
3814 | * Pick up the highest-prio task: | |
3815 | */ | |
3816 | static inline struct task_struct * | |
ff95f3df | 3817 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 3818 | { |
5522d5d5 | 3819 | const struct sched_class *class; |
dd41f596 | 3820 | struct task_struct *p; |
1da177e4 LT |
3821 | |
3822 | /* | |
dd41f596 IM |
3823 | * Optimization: we know that if all tasks are in |
3824 | * the fair class we can call that function directly: | |
1da177e4 | 3825 | */ |
dd41f596 | 3826 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3827 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3828 | if (likely(p)) |
3829 | return p; | |
1da177e4 LT |
3830 | } |
3831 | ||
dd41f596 IM |
3832 | class = sched_class_highest; |
3833 | for ( ; ; ) { | |
fb8d4724 | 3834 | p = class->pick_next_task(rq); |
dd41f596 IM |
3835 | if (p) |
3836 | return p; | |
3837 | /* | |
3838 | * Will never be NULL as the idle class always | |
3839 | * returns a non-NULL p: | |
3840 | */ | |
3841 | class = class->next; | |
3842 | } | |
3843 | } | |
1da177e4 | 3844 | |
dd41f596 IM |
3845 | /* |
3846 | * schedule() is the main scheduler function. | |
3847 | */ | |
3848 | asmlinkage void __sched schedule(void) | |
3849 | { | |
3850 | struct task_struct *prev, *next; | |
67ca7bde | 3851 | unsigned long *switch_count; |
dd41f596 | 3852 | struct rq *rq; |
dd41f596 IM |
3853 | int cpu; |
3854 | ||
3855 | need_resched: | |
3856 | preempt_disable(); | |
3857 | cpu = smp_processor_id(); | |
3858 | rq = cpu_rq(cpu); | |
3859 | rcu_qsctr_inc(cpu); | |
3860 | prev = rq->curr; | |
3861 | switch_count = &prev->nivcsw; | |
3862 | ||
3863 | release_kernel_lock(prev); | |
3864 | need_resched_nonpreemptible: | |
3865 | ||
3866 | schedule_debug(prev); | |
1da177e4 | 3867 | |
8f4d37ec PZ |
3868 | hrtick_clear(rq); |
3869 | ||
1e819950 IM |
3870 | /* |
3871 | * Do the rq-clock update outside the rq lock: | |
3872 | */ | |
3873 | local_irq_disable(); | |
c1b3da3e | 3874 | __update_rq_clock(rq); |
1e819950 IM |
3875 | spin_lock(&rq->lock); |
3876 | clear_tsk_need_resched(prev); | |
1da177e4 | 3877 | |
1da177e4 | 3878 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
1da177e4 | 3879 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && |
dd41f596 | 3880 | unlikely(signal_pending(prev)))) { |
1da177e4 | 3881 | prev->state = TASK_RUNNING; |
dd41f596 | 3882 | } else { |
2e1cb74a | 3883 | deactivate_task(rq, prev, 1); |
1da177e4 | 3884 | } |
dd41f596 | 3885 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3886 | } |
3887 | ||
9a897c5a SR |
3888 | #ifdef CONFIG_SMP |
3889 | if (prev->sched_class->pre_schedule) | |
3890 | prev->sched_class->pre_schedule(rq, prev); | |
3891 | #endif | |
f65eda4f | 3892 | |
dd41f596 | 3893 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3894 | idle_balance(cpu, rq); |
1da177e4 | 3895 | |
31ee529c | 3896 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 3897 | next = pick_next_task(rq, prev); |
1da177e4 LT |
3898 | |
3899 | sched_info_switch(prev, next); | |
dd41f596 | 3900 | |
1da177e4 | 3901 | if (likely(prev != next)) { |
1da177e4 LT |
3902 | rq->nr_switches++; |
3903 | rq->curr = next; | |
3904 | ++*switch_count; | |
3905 | ||
dd41f596 | 3906 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
3907 | /* |
3908 | * the context switch might have flipped the stack from under | |
3909 | * us, hence refresh the local variables. | |
3910 | */ | |
3911 | cpu = smp_processor_id(); | |
3912 | rq = cpu_rq(cpu); | |
1da177e4 LT |
3913 | } else |
3914 | spin_unlock_irq(&rq->lock); | |
3915 | ||
8f4d37ec PZ |
3916 | hrtick_set(rq); |
3917 | ||
3918 | if (unlikely(reacquire_kernel_lock(current) < 0)) | |
1da177e4 | 3919 | goto need_resched_nonpreemptible; |
8f4d37ec | 3920 | |
1da177e4 LT |
3921 | preempt_enable_no_resched(); |
3922 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3923 | goto need_resched; | |
3924 | } | |
1da177e4 LT |
3925 | EXPORT_SYMBOL(schedule); |
3926 | ||
3927 | #ifdef CONFIG_PREEMPT | |
3928 | /* | |
2ed6e34f | 3929 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3930 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3931 | * occur there and call schedule directly. |
3932 | */ | |
3933 | asmlinkage void __sched preempt_schedule(void) | |
3934 | { | |
3935 | struct thread_info *ti = current_thread_info(); | |
1da177e4 LT |
3936 | struct task_struct *task = current; |
3937 | int saved_lock_depth; | |
6478d880 | 3938 | |
1da177e4 LT |
3939 | /* |
3940 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3941 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3942 | */ |
beed33a8 | 3943 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3944 | return; |
3945 | ||
3a5c359a AK |
3946 | do { |
3947 | add_preempt_count(PREEMPT_ACTIVE); | |
3948 | ||
3949 | /* | |
3950 | * We keep the big kernel semaphore locked, but we | |
3951 | * clear ->lock_depth so that schedule() doesnt | |
3952 | * auto-release the semaphore: | |
3953 | */ | |
3a5c359a AK |
3954 | saved_lock_depth = task->lock_depth; |
3955 | task->lock_depth = -1; | |
3a5c359a | 3956 | schedule(); |
3a5c359a | 3957 | task->lock_depth = saved_lock_depth; |
3a5c359a | 3958 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3959 | |
3a5c359a AK |
3960 | /* |
3961 | * Check again in case we missed a preemption opportunity | |
3962 | * between schedule and now. | |
3963 | */ | |
3964 | barrier(); | |
3965 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 3966 | } |
1da177e4 LT |
3967 | EXPORT_SYMBOL(preempt_schedule); |
3968 | ||
3969 | /* | |
2ed6e34f | 3970 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3971 | * off of irq context. |
3972 | * Note, that this is called and return with irqs disabled. This will | |
3973 | * protect us against recursive calling from irq. | |
3974 | */ | |
3975 | asmlinkage void __sched preempt_schedule_irq(void) | |
3976 | { | |
3977 | struct thread_info *ti = current_thread_info(); | |
1da177e4 LT |
3978 | struct task_struct *task = current; |
3979 | int saved_lock_depth; | |
6478d880 | 3980 | |
2ed6e34f | 3981 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3982 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3983 | ||
3a5c359a AK |
3984 | do { |
3985 | add_preempt_count(PREEMPT_ACTIVE); | |
3986 | ||
3987 | /* | |
3988 | * We keep the big kernel semaphore locked, but we | |
3989 | * clear ->lock_depth so that schedule() doesnt | |
3990 | * auto-release the semaphore: | |
3991 | */ | |
3a5c359a AK |
3992 | saved_lock_depth = task->lock_depth; |
3993 | task->lock_depth = -1; | |
3a5c359a AK |
3994 | local_irq_enable(); |
3995 | schedule(); | |
3996 | local_irq_disable(); | |
3a5c359a | 3997 | task->lock_depth = saved_lock_depth; |
3a5c359a | 3998 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3999 | |
3a5c359a AK |
4000 | /* |
4001 | * Check again in case we missed a preemption opportunity | |
4002 | * between schedule and now. | |
4003 | */ | |
4004 | barrier(); | |
4005 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
4006 | } |
4007 | ||
4008 | #endif /* CONFIG_PREEMPT */ | |
4009 | ||
95cdf3b7 IM |
4010 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
4011 | void *key) | |
1da177e4 | 4012 | { |
48f24c4d | 4013 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 4014 | } |
1da177e4 LT |
4015 | EXPORT_SYMBOL(default_wake_function); |
4016 | ||
4017 | /* | |
41a2d6cf IM |
4018 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4019 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4020 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4021 | * | |
4022 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4023 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4024 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4025 | */ | |
4026 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
4027 | int nr_exclusive, int sync, void *key) | |
4028 | { | |
2e45874c | 4029 | wait_queue_t *curr, *next; |
1da177e4 | 4030 | |
2e45874c | 4031 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4032 | unsigned flags = curr->flags; |
4033 | ||
1da177e4 | 4034 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 4035 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4036 | break; |
4037 | } | |
4038 | } | |
4039 | ||
4040 | /** | |
4041 | * __wake_up - wake up threads blocked on a waitqueue. | |
4042 | * @q: the waitqueue | |
4043 | * @mode: which threads | |
4044 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4045 | * @key: is directly passed to the wakeup function |
1da177e4 | 4046 | */ |
7ad5b3a5 | 4047 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4048 | int nr_exclusive, void *key) |
1da177e4 LT |
4049 | { |
4050 | unsigned long flags; | |
4051 | ||
4052 | spin_lock_irqsave(&q->lock, flags); | |
4053 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4054 | spin_unlock_irqrestore(&q->lock, flags); | |
4055 | } | |
1da177e4 LT |
4056 | EXPORT_SYMBOL(__wake_up); |
4057 | ||
4058 | /* | |
4059 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4060 | */ | |
7ad5b3a5 | 4061 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4062 | { |
4063 | __wake_up_common(q, mode, 1, 0, NULL); | |
4064 | } | |
4065 | ||
4066 | /** | |
67be2dd1 | 4067 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4068 | * @q: the waitqueue |
4069 | * @mode: which threads | |
4070 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4071 | * | |
4072 | * The sync wakeup differs that the waker knows that it will schedule | |
4073 | * away soon, so while the target thread will be woken up, it will not | |
4074 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4075 | * with each other. This can prevent needless bouncing between CPUs. | |
4076 | * | |
4077 | * On UP it can prevent extra preemption. | |
4078 | */ | |
7ad5b3a5 | 4079 | void |
95cdf3b7 | 4080 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
1da177e4 LT |
4081 | { |
4082 | unsigned long flags; | |
4083 | int sync = 1; | |
4084 | ||
4085 | if (unlikely(!q)) | |
4086 | return; | |
4087 | ||
4088 | if (unlikely(!nr_exclusive)) | |
4089 | sync = 0; | |
4090 | ||
4091 | spin_lock_irqsave(&q->lock, flags); | |
4092 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
4093 | spin_unlock_irqrestore(&q->lock, flags); | |
4094 | } | |
4095 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
4096 | ||
b15136e9 | 4097 | void complete(struct completion *x) |
1da177e4 LT |
4098 | { |
4099 | unsigned long flags; | |
4100 | ||
4101 | spin_lock_irqsave(&x->wait.lock, flags); | |
4102 | x->done++; | |
d9514f6c | 4103 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4104 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4105 | } | |
4106 | EXPORT_SYMBOL(complete); | |
4107 | ||
b15136e9 | 4108 | void complete_all(struct completion *x) |
1da177e4 LT |
4109 | { |
4110 | unsigned long flags; | |
4111 | ||
4112 | spin_lock_irqsave(&x->wait.lock, flags); | |
4113 | x->done += UINT_MAX/2; | |
d9514f6c | 4114 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4115 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4116 | } | |
4117 | EXPORT_SYMBOL(complete_all); | |
4118 | ||
8cbbe86d AK |
4119 | static inline long __sched |
4120 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4121 | { |
1da177e4 LT |
4122 | if (!x->done) { |
4123 | DECLARE_WAITQUEUE(wait, current); | |
4124 | ||
4125 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
4126 | __add_wait_queue_tail(&x->wait, &wait); | |
4127 | do { | |
009e577e MW |
4128 | if ((state == TASK_INTERRUPTIBLE && |
4129 | signal_pending(current)) || | |
4130 | (state == TASK_KILLABLE && | |
4131 | fatal_signal_pending(current))) { | |
8cbbe86d AK |
4132 | __remove_wait_queue(&x->wait, &wait); |
4133 | return -ERESTARTSYS; | |
4134 | } | |
4135 | __set_current_state(state); | |
1da177e4 LT |
4136 | spin_unlock_irq(&x->wait.lock); |
4137 | timeout = schedule_timeout(timeout); | |
4138 | spin_lock_irq(&x->wait.lock); | |
4139 | if (!timeout) { | |
4140 | __remove_wait_queue(&x->wait, &wait); | |
8cbbe86d | 4141 | return timeout; |
1da177e4 LT |
4142 | } |
4143 | } while (!x->done); | |
4144 | __remove_wait_queue(&x->wait, &wait); | |
4145 | } | |
4146 | x->done--; | |
1da177e4 LT |
4147 | return timeout; |
4148 | } | |
1da177e4 | 4149 | |
8cbbe86d AK |
4150 | static long __sched |
4151 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4152 | { |
1da177e4 LT |
4153 | might_sleep(); |
4154 | ||
4155 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4156 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4157 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4158 | return timeout; |
4159 | } | |
1da177e4 | 4160 | |
b15136e9 | 4161 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4162 | { |
4163 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4164 | } |
8cbbe86d | 4165 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4166 | |
b15136e9 | 4167 | unsigned long __sched |
8cbbe86d | 4168 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4169 | { |
8cbbe86d | 4170 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4171 | } |
8cbbe86d | 4172 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4173 | |
8cbbe86d | 4174 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4175 | { |
51e97990 AK |
4176 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4177 | if (t == -ERESTARTSYS) | |
4178 | return t; | |
4179 | return 0; | |
0fec171c | 4180 | } |
8cbbe86d | 4181 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4182 | |
b15136e9 | 4183 | unsigned long __sched |
8cbbe86d AK |
4184 | wait_for_completion_interruptible_timeout(struct completion *x, |
4185 | unsigned long timeout) | |
0fec171c | 4186 | { |
8cbbe86d | 4187 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4188 | } |
8cbbe86d | 4189 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4190 | |
009e577e MW |
4191 | int __sched wait_for_completion_killable(struct completion *x) |
4192 | { | |
4193 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4194 | if (t == -ERESTARTSYS) | |
4195 | return t; | |
4196 | return 0; | |
4197 | } | |
4198 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4199 | ||
8cbbe86d AK |
4200 | static long __sched |
4201 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4202 | { |
0fec171c IM |
4203 | unsigned long flags; |
4204 | wait_queue_t wait; | |
4205 | ||
4206 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4207 | |
8cbbe86d | 4208 | __set_current_state(state); |
1da177e4 | 4209 | |
8cbbe86d AK |
4210 | spin_lock_irqsave(&q->lock, flags); |
4211 | __add_wait_queue(q, &wait); | |
4212 | spin_unlock(&q->lock); | |
4213 | timeout = schedule_timeout(timeout); | |
4214 | spin_lock_irq(&q->lock); | |
4215 | __remove_wait_queue(q, &wait); | |
4216 | spin_unlock_irqrestore(&q->lock, flags); | |
4217 | ||
4218 | return timeout; | |
4219 | } | |
4220 | ||
4221 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4222 | { | |
4223 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4224 | } |
1da177e4 LT |
4225 | EXPORT_SYMBOL(interruptible_sleep_on); |
4226 | ||
0fec171c | 4227 | long __sched |
95cdf3b7 | 4228 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4229 | { |
8cbbe86d | 4230 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4231 | } |
1da177e4 LT |
4232 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4233 | ||
0fec171c | 4234 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4235 | { |
8cbbe86d | 4236 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4237 | } |
1da177e4 LT |
4238 | EXPORT_SYMBOL(sleep_on); |
4239 | ||
0fec171c | 4240 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4241 | { |
8cbbe86d | 4242 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4243 | } |
1da177e4 LT |
4244 | EXPORT_SYMBOL(sleep_on_timeout); |
4245 | ||
b29739f9 IM |
4246 | #ifdef CONFIG_RT_MUTEXES |
4247 | ||
4248 | /* | |
4249 | * rt_mutex_setprio - set the current priority of a task | |
4250 | * @p: task | |
4251 | * @prio: prio value (kernel-internal form) | |
4252 | * | |
4253 | * This function changes the 'effective' priority of a task. It does | |
4254 | * not touch ->normal_prio like __setscheduler(). | |
4255 | * | |
4256 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4257 | */ | |
36c8b586 | 4258 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4259 | { |
4260 | unsigned long flags; | |
83b699ed | 4261 | int oldprio, on_rq, running; |
70b97a7f | 4262 | struct rq *rq; |
cb469845 | 4263 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
4264 | |
4265 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4266 | ||
4267 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4268 | update_rq_clock(rq); |
b29739f9 | 4269 | |
d5f9f942 | 4270 | oldprio = p->prio; |
dd41f596 | 4271 | on_rq = p->se.on_rq; |
051a1d1a | 4272 | running = task_current(rq, p); |
0e1f3483 | 4273 | if (on_rq) |
69be72c1 | 4274 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4275 | if (running) |
4276 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4277 | |
4278 | if (rt_prio(prio)) | |
4279 | p->sched_class = &rt_sched_class; | |
4280 | else | |
4281 | p->sched_class = &fair_sched_class; | |
4282 | ||
b29739f9 IM |
4283 | p->prio = prio; |
4284 | ||
0e1f3483 HS |
4285 | if (running) |
4286 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4287 | if (on_rq) { |
8159f87e | 4288 | enqueue_task(rq, p, 0); |
cb469845 SR |
4289 | |
4290 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4291 | } |
4292 | task_rq_unlock(rq, &flags); | |
4293 | } | |
4294 | ||
4295 | #endif | |
4296 | ||
36c8b586 | 4297 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4298 | { |
dd41f596 | 4299 | int old_prio, delta, on_rq; |
1da177e4 | 4300 | unsigned long flags; |
70b97a7f | 4301 | struct rq *rq; |
1da177e4 LT |
4302 | |
4303 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4304 | return; | |
4305 | /* | |
4306 | * We have to be careful, if called from sys_setpriority(), | |
4307 | * the task might be in the middle of scheduling on another CPU. | |
4308 | */ | |
4309 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4310 | update_rq_clock(rq); |
1da177e4 LT |
4311 | /* |
4312 | * The RT priorities are set via sched_setscheduler(), but we still | |
4313 | * allow the 'normal' nice value to be set - but as expected | |
4314 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4315 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4316 | */ |
e05606d3 | 4317 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4318 | p->static_prio = NICE_TO_PRIO(nice); |
4319 | goto out_unlock; | |
4320 | } | |
dd41f596 | 4321 | on_rq = p->se.on_rq; |
62fb1851 | 4322 | if (on_rq) { |
69be72c1 | 4323 | dequeue_task(rq, p, 0); |
62fb1851 PZ |
4324 | dec_load(rq, p); |
4325 | } | |
1da177e4 | 4326 | |
1da177e4 | 4327 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4328 | set_load_weight(p); |
b29739f9 IM |
4329 | old_prio = p->prio; |
4330 | p->prio = effective_prio(p); | |
4331 | delta = p->prio - old_prio; | |
1da177e4 | 4332 | |
dd41f596 | 4333 | if (on_rq) { |
8159f87e | 4334 | enqueue_task(rq, p, 0); |
62fb1851 | 4335 | inc_load(rq, p); |
1da177e4 | 4336 | /* |
d5f9f942 AM |
4337 | * If the task increased its priority or is running and |
4338 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4339 | */ |
d5f9f942 | 4340 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4341 | resched_task(rq->curr); |
4342 | } | |
4343 | out_unlock: | |
4344 | task_rq_unlock(rq, &flags); | |
4345 | } | |
1da177e4 LT |
4346 | EXPORT_SYMBOL(set_user_nice); |
4347 | ||
e43379f1 MM |
4348 | /* |
4349 | * can_nice - check if a task can reduce its nice value | |
4350 | * @p: task | |
4351 | * @nice: nice value | |
4352 | */ | |
36c8b586 | 4353 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4354 | { |
024f4747 MM |
4355 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4356 | int nice_rlim = 20 - nice; | |
48f24c4d | 4357 | |
e43379f1 MM |
4358 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4359 | capable(CAP_SYS_NICE)); | |
4360 | } | |
4361 | ||
1da177e4 LT |
4362 | #ifdef __ARCH_WANT_SYS_NICE |
4363 | ||
4364 | /* | |
4365 | * sys_nice - change the priority of the current process. | |
4366 | * @increment: priority increment | |
4367 | * | |
4368 | * sys_setpriority is a more generic, but much slower function that | |
4369 | * does similar things. | |
4370 | */ | |
4371 | asmlinkage long sys_nice(int increment) | |
4372 | { | |
48f24c4d | 4373 | long nice, retval; |
1da177e4 LT |
4374 | |
4375 | /* | |
4376 | * Setpriority might change our priority at the same moment. | |
4377 | * We don't have to worry. Conceptually one call occurs first | |
4378 | * and we have a single winner. | |
4379 | */ | |
e43379f1 MM |
4380 | if (increment < -40) |
4381 | increment = -40; | |
1da177e4 LT |
4382 | if (increment > 40) |
4383 | increment = 40; | |
4384 | ||
4385 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4386 | if (nice < -20) | |
4387 | nice = -20; | |
4388 | if (nice > 19) | |
4389 | nice = 19; | |
4390 | ||
e43379f1 MM |
4391 | if (increment < 0 && !can_nice(current, nice)) |
4392 | return -EPERM; | |
4393 | ||
1da177e4 LT |
4394 | retval = security_task_setnice(current, nice); |
4395 | if (retval) | |
4396 | return retval; | |
4397 | ||
4398 | set_user_nice(current, nice); | |
4399 | return 0; | |
4400 | } | |
4401 | ||
4402 | #endif | |
4403 | ||
4404 | /** | |
4405 | * task_prio - return the priority value of a given task. | |
4406 | * @p: the task in question. | |
4407 | * | |
4408 | * This is the priority value as seen by users in /proc. | |
4409 | * RT tasks are offset by -200. Normal tasks are centered | |
4410 | * around 0, value goes from -16 to +15. | |
4411 | */ | |
36c8b586 | 4412 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4413 | { |
4414 | return p->prio - MAX_RT_PRIO; | |
4415 | } | |
4416 | ||
4417 | /** | |
4418 | * task_nice - return the nice value of a given task. | |
4419 | * @p: the task in question. | |
4420 | */ | |
36c8b586 | 4421 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4422 | { |
4423 | return TASK_NICE(p); | |
4424 | } | |
150d8bed | 4425 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4426 | |
4427 | /** | |
4428 | * idle_cpu - is a given cpu idle currently? | |
4429 | * @cpu: the processor in question. | |
4430 | */ | |
4431 | int idle_cpu(int cpu) | |
4432 | { | |
4433 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4434 | } | |
4435 | ||
1da177e4 LT |
4436 | /** |
4437 | * idle_task - return the idle task for a given cpu. | |
4438 | * @cpu: the processor in question. | |
4439 | */ | |
36c8b586 | 4440 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4441 | { |
4442 | return cpu_rq(cpu)->idle; | |
4443 | } | |
4444 | ||
4445 | /** | |
4446 | * find_process_by_pid - find a process with a matching PID value. | |
4447 | * @pid: the pid in question. | |
4448 | */ | |
a9957449 | 4449 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4450 | { |
228ebcbe | 4451 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4452 | } |
4453 | ||
4454 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4455 | static void |
4456 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4457 | { |
dd41f596 | 4458 | BUG_ON(p->se.on_rq); |
48f24c4d | 4459 | |
1da177e4 | 4460 | p->policy = policy; |
dd41f596 IM |
4461 | switch (p->policy) { |
4462 | case SCHED_NORMAL: | |
4463 | case SCHED_BATCH: | |
4464 | case SCHED_IDLE: | |
4465 | p->sched_class = &fair_sched_class; | |
4466 | break; | |
4467 | case SCHED_FIFO: | |
4468 | case SCHED_RR: | |
4469 | p->sched_class = &rt_sched_class; | |
4470 | break; | |
4471 | } | |
4472 | ||
1da177e4 | 4473 | p->rt_priority = prio; |
b29739f9 IM |
4474 | p->normal_prio = normal_prio(p); |
4475 | /* we are holding p->pi_lock already */ | |
4476 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4477 | set_load_weight(p); |
1da177e4 LT |
4478 | } |
4479 | ||
4480 | /** | |
72fd4a35 | 4481 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4482 | * @p: the task in question. |
4483 | * @policy: new policy. | |
4484 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4485 | * |
72fd4a35 | 4486 | * NOTE that the task may be already dead. |
1da177e4 | 4487 | */ |
95cdf3b7 IM |
4488 | int sched_setscheduler(struct task_struct *p, int policy, |
4489 | struct sched_param *param) | |
1da177e4 | 4490 | { |
83b699ed | 4491 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4492 | unsigned long flags; |
cb469845 | 4493 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 4494 | struct rq *rq; |
1da177e4 | 4495 | |
66e5393a SR |
4496 | /* may grab non-irq protected spin_locks */ |
4497 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4498 | recheck: |
4499 | /* double check policy once rq lock held */ | |
4500 | if (policy < 0) | |
4501 | policy = oldpolicy = p->policy; | |
4502 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4503 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4504 | policy != SCHED_IDLE) | |
b0a9499c | 4505 | return -EINVAL; |
1da177e4 LT |
4506 | /* |
4507 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4508 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4509 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4510 | */ |
4511 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4512 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4513 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4514 | return -EINVAL; |
e05606d3 | 4515 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4516 | return -EINVAL; |
4517 | ||
37e4ab3f OC |
4518 | /* |
4519 | * Allow unprivileged RT tasks to decrease priority: | |
4520 | */ | |
4521 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4522 | if (rt_policy(policy)) { |
8dc3e909 | 4523 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4524 | |
4525 | if (!lock_task_sighand(p, &flags)) | |
4526 | return -ESRCH; | |
4527 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4528 | unlock_task_sighand(p, &flags); | |
4529 | ||
4530 | /* can't set/change the rt policy */ | |
4531 | if (policy != p->policy && !rlim_rtprio) | |
4532 | return -EPERM; | |
4533 | ||
4534 | /* can't increase priority */ | |
4535 | if (param->sched_priority > p->rt_priority && | |
4536 | param->sched_priority > rlim_rtprio) | |
4537 | return -EPERM; | |
4538 | } | |
dd41f596 IM |
4539 | /* |
4540 | * Like positive nice levels, dont allow tasks to | |
4541 | * move out of SCHED_IDLE either: | |
4542 | */ | |
4543 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4544 | return -EPERM; | |
5fe1d75f | 4545 | |
37e4ab3f OC |
4546 | /* can't change other user's priorities */ |
4547 | if ((current->euid != p->euid) && | |
4548 | (current->euid != p->uid)) | |
4549 | return -EPERM; | |
4550 | } | |
1da177e4 | 4551 | |
b68aa230 PZ |
4552 | #ifdef CONFIG_RT_GROUP_SCHED |
4553 | /* | |
4554 | * Do not allow realtime tasks into groups that have no runtime | |
4555 | * assigned. | |
4556 | */ | |
4557 | if (rt_policy(policy) && task_group(p)->rt_runtime == 0) | |
4558 | return -EPERM; | |
4559 | #endif | |
4560 | ||
1da177e4 LT |
4561 | retval = security_task_setscheduler(p, policy, param); |
4562 | if (retval) | |
4563 | return retval; | |
b29739f9 IM |
4564 | /* |
4565 | * make sure no PI-waiters arrive (or leave) while we are | |
4566 | * changing the priority of the task: | |
4567 | */ | |
4568 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4569 | /* |
4570 | * To be able to change p->policy safely, the apropriate | |
4571 | * runqueue lock must be held. | |
4572 | */ | |
b29739f9 | 4573 | rq = __task_rq_lock(p); |
1da177e4 LT |
4574 | /* recheck policy now with rq lock held */ |
4575 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4576 | policy = oldpolicy = -1; | |
b29739f9 IM |
4577 | __task_rq_unlock(rq); |
4578 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4579 | goto recheck; |
4580 | } | |
2daa3577 | 4581 | update_rq_clock(rq); |
dd41f596 | 4582 | on_rq = p->se.on_rq; |
051a1d1a | 4583 | running = task_current(rq, p); |
0e1f3483 | 4584 | if (on_rq) |
2e1cb74a | 4585 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
4586 | if (running) |
4587 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4588 | |
1da177e4 | 4589 | oldprio = p->prio; |
dd41f596 | 4590 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4591 | |
0e1f3483 HS |
4592 | if (running) |
4593 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
4594 | if (on_rq) { |
4595 | activate_task(rq, p, 0); | |
cb469845 SR |
4596 | |
4597 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 4598 | } |
b29739f9 IM |
4599 | __task_rq_unlock(rq); |
4600 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4601 | ||
95e02ca9 TG |
4602 | rt_mutex_adjust_pi(p); |
4603 | ||
1da177e4 LT |
4604 | return 0; |
4605 | } | |
4606 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4607 | ||
95cdf3b7 IM |
4608 | static int |
4609 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4610 | { |
1da177e4 LT |
4611 | struct sched_param lparam; |
4612 | struct task_struct *p; | |
36c8b586 | 4613 | int retval; |
1da177e4 LT |
4614 | |
4615 | if (!param || pid < 0) | |
4616 | return -EINVAL; | |
4617 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4618 | return -EFAULT; | |
5fe1d75f ON |
4619 | |
4620 | rcu_read_lock(); | |
4621 | retval = -ESRCH; | |
1da177e4 | 4622 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4623 | if (p != NULL) |
4624 | retval = sched_setscheduler(p, policy, &lparam); | |
4625 | rcu_read_unlock(); | |
36c8b586 | 4626 | |
1da177e4 LT |
4627 | return retval; |
4628 | } | |
4629 | ||
4630 | /** | |
4631 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4632 | * @pid: the pid in question. | |
4633 | * @policy: new policy. | |
4634 | * @param: structure containing the new RT priority. | |
4635 | */ | |
41a2d6cf IM |
4636 | asmlinkage long |
4637 | sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4638 | { |
c21761f1 JB |
4639 | /* negative values for policy are not valid */ |
4640 | if (policy < 0) | |
4641 | return -EINVAL; | |
4642 | ||
1da177e4 LT |
4643 | return do_sched_setscheduler(pid, policy, param); |
4644 | } | |
4645 | ||
4646 | /** | |
4647 | * sys_sched_setparam - set/change the RT priority of a thread | |
4648 | * @pid: the pid in question. | |
4649 | * @param: structure containing the new RT priority. | |
4650 | */ | |
4651 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4652 | { | |
4653 | return do_sched_setscheduler(pid, -1, param); | |
4654 | } | |
4655 | ||
4656 | /** | |
4657 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4658 | * @pid: the pid in question. | |
4659 | */ | |
4660 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4661 | { | |
36c8b586 | 4662 | struct task_struct *p; |
3a5c359a | 4663 | int retval; |
1da177e4 LT |
4664 | |
4665 | if (pid < 0) | |
3a5c359a | 4666 | return -EINVAL; |
1da177e4 LT |
4667 | |
4668 | retval = -ESRCH; | |
4669 | read_lock(&tasklist_lock); | |
4670 | p = find_process_by_pid(pid); | |
4671 | if (p) { | |
4672 | retval = security_task_getscheduler(p); | |
4673 | if (!retval) | |
4674 | retval = p->policy; | |
4675 | } | |
4676 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
4677 | return retval; |
4678 | } | |
4679 | ||
4680 | /** | |
4681 | * sys_sched_getscheduler - get the RT priority of a thread | |
4682 | * @pid: the pid in question. | |
4683 | * @param: structure containing the RT priority. | |
4684 | */ | |
4685 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
4686 | { | |
4687 | struct sched_param lp; | |
36c8b586 | 4688 | struct task_struct *p; |
3a5c359a | 4689 | int retval; |
1da177e4 LT |
4690 | |
4691 | if (!param || pid < 0) | |
3a5c359a | 4692 | return -EINVAL; |
1da177e4 LT |
4693 | |
4694 | read_lock(&tasklist_lock); | |
4695 | p = find_process_by_pid(pid); | |
4696 | retval = -ESRCH; | |
4697 | if (!p) | |
4698 | goto out_unlock; | |
4699 | ||
4700 | retval = security_task_getscheduler(p); | |
4701 | if (retval) | |
4702 | goto out_unlock; | |
4703 | ||
4704 | lp.sched_priority = p->rt_priority; | |
4705 | read_unlock(&tasklist_lock); | |
4706 | ||
4707 | /* | |
4708 | * This one might sleep, we cannot do it with a spinlock held ... | |
4709 | */ | |
4710 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4711 | ||
1da177e4 LT |
4712 | return retval; |
4713 | ||
4714 | out_unlock: | |
4715 | read_unlock(&tasklist_lock); | |
4716 | return retval; | |
4717 | } | |
4718 | ||
4719 | long sched_setaffinity(pid_t pid, cpumask_t new_mask) | |
4720 | { | |
1da177e4 | 4721 | cpumask_t cpus_allowed; |
36c8b586 IM |
4722 | struct task_struct *p; |
4723 | int retval; | |
1da177e4 | 4724 | |
95402b38 | 4725 | get_online_cpus(); |
1da177e4 LT |
4726 | read_lock(&tasklist_lock); |
4727 | ||
4728 | p = find_process_by_pid(pid); | |
4729 | if (!p) { | |
4730 | read_unlock(&tasklist_lock); | |
95402b38 | 4731 | put_online_cpus(); |
1da177e4 LT |
4732 | return -ESRCH; |
4733 | } | |
4734 | ||
4735 | /* | |
4736 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 4737 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
4738 | * usage count and then drop tasklist_lock. |
4739 | */ | |
4740 | get_task_struct(p); | |
4741 | read_unlock(&tasklist_lock); | |
4742 | ||
4743 | retval = -EPERM; | |
4744 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
4745 | !capable(CAP_SYS_NICE)) | |
4746 | goto out_unlock; | |
4747 | ||
e7834f8f DQ |
4748 | retval = security_task_setscheduler(p, 0, NULL); |
4749 | if (retval) | |
4750 | goto out_unlock; | |
4751 | ||
1da177e4 LT |
4752 | cpus_allowed = cpuset_cpus_allowed(p); |
4753 | cpus_and(new_mask, new_mask, cpus_allowed); | |
8707d8b8 | 4754 | again: |
1da177e4 LT |
4755 | retval = set_cpus_allowed(p, new_mask); |
4756 | ||
8707d8b8 PM |
4757 | if (!retval) { |
4758 | cpus_allowed = cpuset_cpus_allowed(p); | |
4759 | if (!cpus_subset(new_mask, cpus_allowed)) { | |
4760 | /* | |
4761 | * We must have raced with a concurrent cpuset | |
4762 | * update. Just reset the cpus_allowed to the | |
4763 | * cpuset's cpus_allowed | |
4764 | */ | |
4765 | new_mask = cpus_allowed; | |
4766 | goto again; | |
4767 | } | |
4768 | } | |
1da177e4 LT |
4769 | out_unlock: |
4770 | put_task_struct(p); | |
95402b38 | 4771 | put_online_cpus(); |
1da177e4 LT |
4772 | return retval; |
4773 | } | |
4774 | ||
4775 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
4776 | cpumask_t *new_mask) | |
4777 | { | |
4778 | if (len < sizeof(cpumask_t)) { | |
4779 | memset(new_mask, 0, sizeof(cpumask_t)); | |
4780 | } else if (len > sizeof(cpumask_t)) { | |
4781 | len = sizeof(cpumask_t); | |
4782 | } | |
4783 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
4784 | } | |
4785 | ||
4786 | /** | |
4787 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4788 | * @pid: pid of the process | |
4789 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4790 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4791 | */ | |
4792 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
4793 | unsigned long __user *user_mask_ptr) | |
4794 | { | |
4795 | cpumask_t new_mask; | |
4796 | int retval; | |
4797 | ||
4798 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
4799 | if (retval) | |
4800 | return retval; | |
4801 | ||
4802 | return sched_setaffinity(pid, new_mask); | |
4803 | } | |
4804 | ||
4805 | /* | |
4806 | * Represents all cpu's present in the system | |
4807 | * In systems capable of hotplug, this map could dynamically grow | |
4808 | * as new cpu's are detected in the system via any platform specific | |
4809 | * method, such as ACPI for e.g. | |
4810 | */ | |
4811 | ||
4cef0c61 | 4812 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
4813 | EXPORT_SYMBOL(cpu_present_map); |
4814 | ||
4815 | #ifndef CONFIG_SMP | |
4cef0c61 | 4816 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
4817 | EXPORT_SYMBOL(cpu_online_map); |
4818 | ||
4cef0c61 | 4819 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 4820 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
4821 | #endif |
4822 | ||
4823 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
4824 | { | |
36c8b586 | 4825 | struct task_struct *p; |
1da177e4 | 4826 | int retval; |
1da177e4 | 4827 | |
95402b38 | 4828 | get_online_cpus(); |
1da177e4 LT |
4829 | read_lock(&tasklist_lock); |
4830 | ||
4831 | retval = -ESRCH; | |
4832 | p = find_process_by_pid(pid); | |
4833 | if (!p) | |
4834 | goto out_unlock; | |
4835 | ||
e7834f8f DQ |
4836 | retval = security_task_getscheduler(p); |
4837 | if (retval) | |
4838 | goto out_unlock; | |
4839 | ||
2f7016d9 | 4840 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
4841 | |
4842 | out_unlock: | |
4843 | read_unlock(&tasklist_lock); | |
95402b38 | 4844 | put_online_cpus(); |
1da177e4 | 4845 | |
9531b62f | 4846 | return retval; |
1da177e4 LT |
4847 | } |
4848 | ||
4849 | /** | |
4850 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4851 | * @pid: pid of the process | |
4852 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4853 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4854 | */ | |
4855 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
4856 | unsigned long __user *user_mask_ptr) | |
4857 | { | |
4858 | int ret; | |
4859 | cpumask_t mask; | |
4860 | ||
4861 | if (len < sizeof(cpumask_t)) | |
4862 | return -EINVAL; | |
4863 | ||
4864 | ret = sched_getaffinity(pid, &mask); | |
4865 | if (ret < 0) | |
4866 | return ret; | |
4867 | ||
4868 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
4869 | return -EFAULT; | |
4870 | ||
4871 | return sizeof(cpumask_t); | |
4872 | } | |
4873 | ||
4874 | /** | |
4875 | * sys_sched_yield - yield the current processor to other threads. | |
4876 | * | |
dd41f596 IM |
4877 | * This function yields the current CPU to other tasks. If there are no |
4878 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
4879 | */ |
4880 | asmlinkage long sys_sched_yield(void) | |
4881 | { | |
70b97a7f | 4882 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4883 | |
2d72376b | 4884 | schedstat_inc(rq, yld_count); |
4530d7ab | 4885 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4886 | |
4887 | /* | |
4888 | * Since we are going to call schedule() anyway, there's | |
4889 | * no need to preempt or enable interrupts: | |
4890 | */ | |
4891 | __release(rq->lock); | |
8a25d5de | 4892 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4893 | _raw_spin_unlock(&rq->lock); |
4894 | preempt_enable_no_resched(); | |
4895 | ||
4896 | schedule(); | |
4897 | ||
4898 | return 0; | |
4899 | } | |
4900 | ||
e7b38404 | 4901 | static void __cond_resched(void) |
1da177e4 | 4902 | { |
8e0a43d8 IM |
4903 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
4904 | __might_sleep(__FILE__, __LINE__); | |
4905 | #endif | |
5bbcfd90 IM |
4906 | /* |
4907 | * The BKS might be reacquired before we have dropped | |
4908 | * PREEMPT_ACTIVE, which could trigger a second | |
4909 | * cond_resched() call. | |
4910 | */ | |
1da177e4 LT |
4911 | do { |
4912 | add_preempt_count(PREEMPT_ACTIVE); | |
4913 | schedule(); | |
4914 | sub_preempt_count(PREEMPT_ACTIVE); | |
4915 | } while (need_resched()); | |
4916 | } | |
4917 | ||
02b67cc3 HX |
4918 | #if !defined(CONFIG_PREEMPT) || defined(CONFIG_PREEMPT_VOLUNTARY) |
4919 | int __sched _cond_resched(void) | |
1da177e4 | 4920 | { |
9414232f IM |
4921 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
4922 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
4923 | __cond_resched(); |
4924 | return 1; | |
4925 | } | |
4926 | return 0; | |
4927 | } | |
02b67cc3 HX |
4928 | EXPORT_SYMBOL(_cond_resched); |
4929 | #endif | |
1da177e4 LT |
4930 | |
4931 | /* | |
4932 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
4933 | * call schedule, and on return reacquire the lock. | |
4934 | * | |
41a2d6cf | 4935 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4936 | * operations here to prevent schedule() from being called twice (once via |
4937 | * spin_unlock(), once by hand). | |
4938 | */ | |
95cdf3b7 | 4939 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4940 | { |
95c354fe | 4941 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
4942 | int ret = 0; |
4943 | ||
95c354fe | 4944 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4945 | spin_unlock(lock); |
95c354fe NP |
4946 | if (resched && need_resched()) |
4947 | __cond_resched(); | |
4948 | else | |
4949 | cpu_relax(); | |
6df3cecb | 4950 | ret = 1; |
1da177e4 | 4951 | spin_lock(lock); |
1da177e4 | 4952 | } |
6df3cecb | 4953 | return ret; |
1da177e4 | 4954 | } |
1da177e4 LT |
4955 | EXPORT_SYMBOL(cond_resched_lock); |
4956 | ||
4957 | int __sched cond_resched_softirq(void) | |
4958 | { | |
4959 | BUG_ON(!in_softirq()); | |
4960 | ||
9414232f | 4961 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 4962 | local_bh_enable(); |
1da177e4 LT |
4963 | __cond_resched(); |
4964 | local_bh_disable(); | |
4965 | return 1; | |
4966 | } | |
4967 | return 0; | |
4968 | } | |
1da177e4 LT |
4969 | EXPORT_SYMBOL(cond_resched_softirq); |
4970 | ||
1da177e4 LT |
4971 | /** |
4972 | * yield - yield the current processor to other threads. | |
4973 | * | |
72fd4a35 | 4974 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4975 | * thread runnable and calls sys_sched_yield(). |
4976 | */ | |
4977 | void __sched yield(void) | |
4978 | { | |
4979 | set_current_state(TASK_RUNNING); | |
4980 | sys_sched_yield(); | |
4981 | } | |
1da177e4 LT |
4982 | EXPORT_SYMBOL(yield); |
4983 | ||
4984 | /* | |
41a2d6cf | 4985 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
4986 | * that process accounting knows that this is a task in IO wait state. |
4987 | * | |
4988 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
4989 | * has set its backing_dev_info: the queue against which it should throttle) | |
4990 | */ | |
4991 | void __sched io_schedule(void) | |
4992 | { | |
70b97a7f | 4993 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 4994 | |
0ff92245 | 4995 | delayacct_blkio_start(); |
1da177e4 LT |
4996 | atomic_inc(&rq->nr_iowait); |
4997 | schedule(); | |
4998 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4999 | delayacct_blkio_end(); |
1da177e4 | 5000 | } |
1da177e4 LT |
5001 | EXPORT_SYMBOL(io_schedule); |
5002 | ||
5003 | long __sched io_schedule_timeout(long timeout) | |
5004 | { | |
70b97a7f | 5005 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
5006 | long ret; |
5007 | ||
0ff92245 | 5008 | delayacct_blkio_start(); |
1da177e4 LT |
5009 | atomic_inc(&rq->nr_iowait); |
5010 | ret = schedule_timeout(timeout); | |
5011 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5012 | delayacct_blkio_end(); |
1da177e4 LT |
5013 | return ret; |
5014 | } | |
5015 | ||
5016 | /** | |
5017 | * sys_sched_get_priority_max - return maximum RT priority. | |
5018 | * @policy: scheduling class. | |
5019 | * | |
5020 | * this syscall returns the maximum rt_priority that can be used | |
5021 | * by a given scheduling class. | |
5022 | */ | |
5023 | asmlinkage long sys_sched_get_priority_max(int policy) | |
5024 | { | |
5025 | int ret = -EINVAL; | |
5026 | ||
5027 | switch (policy) { | |
5028 | case SCHED_FIFO: | |
5029 | case SCHED_RR: | |
5030 | ret = MAX_USER_RT_PRIO-1; | |
5031 | break; | |
5032 | case SCHED_NORMAL: | |
b0a9499c | 5033 | case SCHED_BATCH: |
dd41f596 | 5034 | case SCHED_IDLE: |
1da177e4 LT |
5035 | ret = 0; |
5036 | break; | |
5037 | } | |
5038 | return ret; | |
5039 | } | |
5040 | ||
5041 | /** | |
5042 | * sys_sched_get_priority_min - return minimum RT priority. | |
5043 | * @policy: scheduling class. | |
5044 | * | |
5045 | * this syscall returns the minimum rt_priority that can be used | |
5046 | * by a given scheduling class. | |
5047 | */ | |
5048 | asmlinkage long sys_sched_get_priority_min(int policy) | |
5049 | { | |
5050 | int ret = -EINVAL; | |
5051 | ||
5052 | switch (policy) { | |
5053 | case SCHED_FIFO: | |
5054 | case SCHED_RR: | |
5055 | ret = 1; | |
5056 | break; | |
5057 | case SCHED_NORMAL: | |
b0a9499c | 5058 | case SCHED_BATCH: |
dd41f596 | 5059 | case SCHED_IDLE: |
1da177e4 LT |
5060 | ret = 0; |
5061 | } | |
5062 | return ret; | |
5063 | } | |
5064 | ||
5065 | /** | |
5066 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5067 | * @pid: pid of the process. | |
5068 | * @interval: userspace pointer to the timeslice value. | |
5069 | * | |
5070 | * this syscall writes the default timeslice value of a given process | |
5071 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5072 | */ | |
5073 | asmlinkage | |
5074 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
5075 | { | |
36c8b586 | 5076 | struct task_struct *p; |
a4ec24b4 | 5077 | unsigned int time_slice; |
3a5c359a | 5078 | int retval; |
1da177e4 | 5079 | struct timespec t; |
1da177e4 LT |
5080 | |
5081 | if (pid < 0) | |
3a5c359a | 5082 | return -EINVAL; |
1da177e4 LT |
5083 | |
5084 | retval = -ESRCH; | |
5085 | read_lock(&tasklist_lock); | |
5086 | p = find_process_by_pid(pid); | |
5087 | if (!p) | |
5088 | goto out_unlock; | |
5089 | ||
5090 | retval = security_task_getscheduler(p); | |
5091 | if (retval) | |
5092 | goto out_unlock; | |
5093 | ||
77034937 IM |
5094 | /* |
5095 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
5096 | * tasks that are on an otherwise idle runqueue: | |
5097 | */ | |
5098 | time_slice = 0; | |
5099 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 5100 | time_slice = DEF_TIMESLICE; |
1868f958 | 5101 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
5102 | struct sched_entity *se = &p->se; |
5103 | unsigned long flags; | |
5104 | struct rq *rq; | |
5105 | ||
5106 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
5107 | if (rq->cfs.load.weight) |
5108 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
5109 | task_rq_unlock(rq, &flags); |
5110 | } | |
1da177e4 | 5111 | read_unlock(&tasklist_lock); |
a4ec24b4 | 5112 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5113 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5114 | return retval; |
3a5c359a | 5115 | |
1da177e4 LT |
5116 | out_unlock: |
5117 | read_unlock(&tasklist_lock); | |
5118 | return retval; | |
5119 | } | |
5120 | ||
2ed6e34f | 5121 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 | 5122 | |
82a1fcb9 | 5123 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5124 | { |
1da177e4 | 5125 | unsigned long free = 0; |
36c8b586 | 5126 | unsigned state; |
1da177e4 | 5127 | |
1da177e4 | 5128 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 5129 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5130 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5131 | #if BITS_PER_LONG == 32 |
1da177e4 | 5132 | if (state == TASK_RUNNING) |
cc4ea795 | 5133 | printk(KERN_CONT " running "); |
1da177e4 | 5134 | else |
cc4ea795 | 5135 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5136 | #else |
5137 | if (state == TASK_RUNNING) | |
cc4ea795 | 5138 | printk(KERN_CONT " running task "); |
1da177e4 | 5139 | else |
cc4ea795 | 5140 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5141 | #endif |
5142 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
5143 | { | |
10ebffde | 5144 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
5145 | while (!*n) |
5146 | n++; | |
10ebffde | 5147 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
5148 | } |
5149 | #endif | |
ba25f9dc | 5150 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 5151 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 5152 | |
5fb5e6de | 5153 | show_stack(p, NULL); |
1da177e4 LT |
5154 | } |
5155 | ||
e59e2ae2 | 5156 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5157 | { |
36c8b586 | 5158 | struct task_struct *g, *p; |
1da177e4 | 5159 | |
4bd77321 IM |
5160 | #if BITS_PER_LONG == 32 |
5161 | printk(KERN_INFO | |
5162 | " task PC stack pid father\n"); | |
1da177e4 | 5163 | #else |
4bd77321 IM |
5164 | printk(KERN_INFO |
5165 | " task PC stack pid father\n"); | |
1da177e4 LT |
5166 | #endif |
5167 | read_lock(&tasklist_lock); | |
5168 | do_each_thread(g, p) { | |
5169 | /* | |
5170 | * reset the NMI-timeout, listing all files on a slow | |
5171 | * console might take alot of time: | |
5172 | */ | |
5173 | touch_nmi_watchdog(); | |
39bc89fd | 5174 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5175 | sched_show_task(p); |
1da177e4 LT |
5176 | } while_each_thread(g, p); |
5177 | ||
04c9167f JF |
5178 | touch_all_softlockup_watchdogs(); |
5179 | ||
dd41f596 IM |
5180 | #ifdef CONFIG_SCHED_DEBUG |
5181 | sysrq_sched_debug_show(); | |
5182 | #endif | |
1da177e4 | 5183 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5184 | /* |
5185 | * Only show locks if all tasks are dumped: | |
5186 | */ | |
5187 | if (state_filter == -1) | |
5188 | debug_show_all_locks(); | |
1da177e4 LT |
5189 | } |
5190 | ||
1df21055 IM |
5191 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5192 | { | |
dd41f596 | 5193 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5194 | } |
5195 | ||
f340c0d1 IM |
5196 | /** |
5197 | * init_idle - set up an idle thread for a given CPU | |
5198 | * @idle: task in question | |
5199 | * @cpu: cpu the idle task belongs to | |
5200 | * | |
5201 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5202 | * flag, to make booting more robust. | |
5203 | */ | |
5c1e1767 | 5204 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5205 | { |
70b97a7f | 5206 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5207 | unsigned long flags; |
5208 | ||
dd41f596 IM |
5209 | __sched_fork(idle); |
5210 | idle->se.exec_start = sched_clock(); | |
5211 | ||
b29739f9 | 5212 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 5213 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 5214 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
5215 | |
5216 | spin_lock_irqsave(&rq->lock, flags); | |
5217 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
5218 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5219 | idle->oncpu = 1; | |
5220 | #endif | |
1da177e4 LT |
5221 | spin_unlock_irqrestore(&rq->lock, flags); |
5222 | ||
5223 | /* Set the preempt count _outside_ the spinlocks! */ | |
a1261f54 | 5224 | task_thread_info(idle)->preempt_count = 0; |
6478d880 | 5225 | |
dd41f596 IM |
5226 | /* |
5227 | * The idle tasks have their own, simple scheduling class: | |
5228 | */ | |
5229 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5230 | } |
5231 | ||
5232 | /* | |
5233 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5234 | * indicates which cpus entered this state. This is used | |
5235 | * in the rcu update to wait only for active cpus. For system | |
5236 | * which do not switch off the HZ timer nohz_cpu_mask should | |
5237 | * always be CPU_MASK_NONE. | |
5238 | */ | |
5239 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
5240 | ||
19978ca6 IM |
5241 | /* |
5242 | * Increase the granularity value when there are more CPUs, | |
5243 | * because with more CPUs the 'effective latency' as visible | |
5244 | * to users decreases. But the relationship is not linear, | |
5245 | * so pick a second-best guess by going with the log2 of the | |
5246 | * number of CPUs. | |
5247 | * | |
5248 | * This idea comes from the SD scheduler of Con Kolivas: | |
5249 | */ | |
5250 | static inline void sched_init_granularity(void) | |
5251 | { | |
5252 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
5253 | const unsigned long limit = 200000000; | |
5254 | ||
5255 | sysctl_sched_min_granularity *= factor; | |
5256 | if (sysctl_sched_min_granularity > limit) | |
5257 | sysctl_sched_min_granularity = limit; | |
5258 | ||
5259 | sysctl_sched_latency *= factor; | |
5260 | if (sysctl_sched_latency > limit) | |
5261 | sysctl_sched_latency = limit; | |
5262 | ||
5263 | sysctl_sched_wakeup_granularity *= factor; | |
5264 | sysctl_sched_batch_wakeup_granularity *= factor; | |
5265 | } | |
5266 | ||
1da177e4 LT |
5267 | #ifdef CONFIG_SMP |
5268 | /* | |
5269 | * This is how migration works: | |
5270 | * | |
70b97a7f | 5271 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5272 | * runqueue and wake up that CPU's migration thread. |
5273 | * 2) we down() the locked semaphore => thread blocks. | |
5274 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5275 | * thread off the CPU) | |
5276 | * 4) it gets the migration request and checks whether the migrated | |
5277 | * task is still in the wrong runqueue. | |
5278 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5279 | * it and puts it into the right queue. | |
5280 | * 6) migration thread up()s the semaphore. | |
5281 | * 7) we wake up and the migration is done. | |
5282 | */ | |
5283 | ||
5284 | /* | |
5285 | * Change a given task's CPU affinity. Migrate the thread to a | |
5286 | * proper CPU and schedule it away if the CPU it's executing on | |
5287 | * is removed from the allowed bitmask. | |
5288 | * | |
5289 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5290 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5291 | * call is not atomic; no spinlocks may be held. |
5292 | */ | |
36c8b586 | 5293 | int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask) |
1da177e4 | 5294 | { |
70b97a7f | 5295 | struct migration_req req; |
1da177e4 | 5296 | unsigned long flags; |
70b97a7f | 5297 | struct rq *rq; |
48f24c4d | 5298 | int ret = 0; |
1da177e4 LT |
5299 | |
5300 | rq = task_rq_lock(p, &flags); | |
5301 | if (!cpus_intersects(new_mask, cpu_online_map)) { | |
5302 | ret = -EINVAL; | |
5303 | goto out; | |
5304 | } | |
5305 | ||
73fe6aae GH |
5306 | if (p->sched_class->set_cpus_allowed) |
5307 | p->sched_class->set_cpus_allowed(p, &new_mask); | |
5308 | else { | |
0eab9146 | 5309 | p->cpus_allowed = new_mask; |
6f505b16 | 5310 | p->rt.nr_cpus_allowed = cpus_weight(new_mask); |
73fe6aae GH |
5311 | } |
5312 | ||
1da177e4 LT |
5313 | /* Can the task run on the task's current CPU? If so, we're done */ |
5314 | if (cpu_isset(task_cpu(p), new_mask)) | |
5315 | goto out; | |
5316 | ||
5317 | if (migrate_task(p, any_online_cpu(new_mask), &req)) { | |
5318 | /* Need help from migration thread: drop lock and wait. */ | |
5319 | task_rq_unlock(rq, &flags); | |
5320 | wake_up_process(rq->migration_thread); | |
5321 | wait_for_completion(&req.done); | |
5322 | tlb_migrate_finish(p->mm); | |
5323 | return 0; | |
5324 | } | |
5325 | out: | |
5326 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5327 | |
1da177e4 LT |
5328 | return ret; |
5329 | } | |
1da177e4 LT |
5330 | EXPORT_SYMBOL_GPL(set_cpus_allowed); |
5331 | ||
5332 | /* | |
41a2d6cf | 5333 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5334 | * this because either it can't run here any more (set_cpus_allowed() |
5335 | * away from this CPU, or CPU going down), or because we're | |
5336 | * attempting to rebalance this task on exec (sched_exec). | |
5337 | * | |
5338 | * So we race with normal scheduler movements, but that's OK, as long | |
5339 | * as the task is no longer on this CPU. | |
efc30814 KK |
5340 | * |
5341 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5342 | */ |
efc30814 | 5343 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5344 | { |
70b97a7f | 5345 | struct rq *rq_dest, *rq_src; |
dd41f596 | 5346 | int ret = 0, on_rq; |
1da177e4 LT |
5347 | |
5348 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 5349 | return ret; |
1da177e4 LT |
5350 | |
5351 | rq_src = cpu_rq(src_cpu); | |
5352 | rq_dest = cpu_rq(dest_cpu); | |
5353 | ||
5354 | double_rq_lock(rq_src, rq_dest); | |
5355 | /* Already moved. */ | |
5356 | if (task_cpu(p) != src_cpu) | |
5357 | goto out; | |
5358 | /* Affinity changed (again). */ | |
5359 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
5360 | goto out; | |
5361 | ||
dd41f596 | 5362 | on_rq = p->se.on_rq; |
6e82a3be | 5363 | if (on_rq) |
2e1cb74a | 5364 | deactivate_task(rq_src, p, 0); |
6e82a3be | 5365 | |
1da177e4 | 5366 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
5367 | if (on_rq) { |
5368 | activate_task(rq_dest, p, 0); | |
5369 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 5370 | } |
efc30814 | 5371 | ret = 1; |
1da177e4 LT |
5372 | out: |
5373 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 5374 | return ret; |
1da177e4 LT |
5375 | } |
5376 | ||
5377 | /* | |
5378 | * migration_thread - this is a highprio system thread that performs | |
5379 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5380 | * another runqueue. | |
5381 | */ | |
95cdf3b7 | 5382 | static int migration_thread(void *data) |
1da177e4 | 5383 | { |
1da177e4 | 5384 | int cpu = (long)data; |
70b97a7f | 5385 | struct rq *rq; |
1da177e4 LT |
5386 | |
5387 | rq = cpu_rq(cpu); | |
5388 | BUG_ON(rq->migration_thread != current); | |
5389 | ||
5390 | set_current_state(TASK_INTERRUPTIBLE); | |
5391 | while (!kthread_should_stop()) { | |
70b97a7f | 5392 | struct migration_req *req; |
1da177e4 | 5393 | struct list_head *head; |
1da177e4 | 5394 | |
1da177e4 LT |
5395 | spin_lock_irq(&rq->lock); |
5396 | ||
5397 | if (cpu_is_offline(cpu)) { | |
5398 | spin_unlock_irq(&rq->lock); | |
5399 | goto wait_to_die; | |
5400 | } | |
5401 | ||
5402 | if (rq->active_balance) { | |
5403 | active_load_balance(rq, cpu); | |
5404 | rq->active_balance = 0; | |
5405 | } | |
5406 | ||
5407 | head = &rq->migration_queue; | |
5408 | ||
5409 | if (list_empty(head)) { | |
5410 | spin_unlock_irq(&rq->lock); | |
5411 | schedule(); | |
5412 | set_current_state(TASK_INTERRUPTIBLE); | |
5413 | continue; | |
5414 | } | |
70b97a7f | 5415 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5416 | list_del_init(head->next); |
5417 | ||
674311d5 NP |
5418 | spin_unlock(&rq->lock); |
5419 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5420 | local_irq_enable(); | |
1da177e4 LT |
5421 | |
5422 | complete(&req->done); | |
5423 | } | |
5424 | __set_current_state(TASK_RUNNING); | |
5425 | return 0; | |
5426 | ||
5427 | wait_to_die: | |
5428 | /* Wait for kthread_stop */ | |
5429 | set_current_state(TASK_INTERRUPTIBLE); | |
5430 | while (!kthread_should_stop()) { | |
5431 | schedule(); | |
5432 | set_current_state(TASK_INTERRUPTIBLE); | |
5433 | } | |
5434 | __set_current_state(TASK_RUNNING); | |
5435 | return 0; | |
5436 | } | |
5437 | ||
5438 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5439 | |
5440 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5441 | { | |
5442 | int ret; | |
5443 | ||
5444 | local_irq_disable(); | |
5445 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5446 | local_irq_enable(); | |
5447 | return ret; | |
5448 | } | |
5449 | ||
054b9108 | 5450 | /* |
3a4fa0a2 | 5451 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
5452 | * NOTE: interrupts should be disabled by the caller |
5453 | */ | |
48f24c4d | 5454 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5455 | { |
efc30814 | 5456 | unsigned long flags; |
1da177e4 | 5457 | cpumask_t mask; |
70b97a7f IM |
5458 | struct rq *rq; |
5459 | int dest_cpu; | |
1da177e4 | 5460 | |
3a5c359a AK |
5461 | do { |
5462 | /* On same node? */ | |
5463 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
5464 | cpus_and(mask, mask, p->cpus_allowed); | |
5465 | dest_cpu = any_online_cpu(mask); | |
5466 | ||
5467 | /* On any allowed CPU? */ | |
5468 | if (dest_cpu == NR_CPUS) | |
5469 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
5470 | ||
5471 | /* No more Mr. Nice Guy. */ | |
5472 | if (dest_cpu == NR_CPUS) { | |
470fd646 CW |
5473 | cpumask_t cpus_allowed = cpuset_cpus_allowed_locked(p); |
5474 | /* | |
5475 | * Try to stay on the same cpuset, where the | |
5476 | * current cpuset may be a subset of all cpus. | |
5477 | * The cpuset_cpus_allowed_locked() variant of | |
41a2d6cf | 5478 | * cpuset_cpus_allowed() will not block. It must be |
470fd646 CW |
5479 | * called within calls to cpuset_lock/cpuset_unlock. |
5480 | */ | |
3a5c359a | 5481 | rq = task_rq_lock(p, &flags); |
470fd646 | 5482 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
5483 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5484 | task_rq_unlock(rq, &flags); | |
1da177e4 | 5485 | |
3a5c359a AK |
5486 | /* |
5487 | * Don't tell them about moving exiting tasks or | |
5488 | * kernel threads (both mm NULL), since they never | |
5489 | * leave kernel. | |
5490 | */ | |
41a2d6cf | 5491 | if (p->mm && printk_ratelimit()) { |
3a5c359a AK |
5492 | printk(KERN_INFO "process %d (%s) no " |
5493 | "longer affine to cpu%d\n", | |
41a2d6cf IM |
5494 | task_pid_nr(p), p->comm, dead_cpu); |
5495 | } | |
3a5c359a | 5496 | } |
f7b4cddc | 5497 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
5498 | } |
5499 | ||
5500 | /* | |
5501 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5502 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5503 | * for performance reasons the counter is not stricly tracking tasks to | |
5504 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5505 | * to keep the global sum constant after CPU-down: | |
5506 | */ | |
70b97a7f | 5507 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5508 | { |
70b97a7f | 5509 | struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL)); |
1da177e4 LT |
5510 | unsigned long flags; |
5511 | ||
5512 | local_irq_save(flags); | |
5513 | double_rq_lock(rq_src, rq_dest); | |
5514 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5515 | rq_src->nr_uninterruptible = 0; | |
5516 | double_rq_unlock(rq_src, rq_dest); | |
5517 | local_irq_restore(flags); | |
5518 | } | |
5519 | ||
5520 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5521 | static void migrate_live_tasks(int src_cpu) | |
5522 | { | |
48f24c4d | 5523 | struct task_struct *p, *t; |
1da177e4 | 5524 | |
f7b4cddc | 5525 | read_lock(&tasklist_lock); |
1da177e4 | 5526 | |
48f24c4d IM |
5527 | do_each_thread(t, p) { |
5528 | if (p == current) | |
1da177e4 LT |
5529 | continue; |
5530 | ||
48f24c4d IM |
5531 | if (task_cpu(p) == src_cpu) |
5532 | move_task_off_dead_cpu(src_cpu, p); | |
5533 | } while_each_thread(t, p); | |
1da177e4 | 5534 | |
f7b4cddc | 5535 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5536 | } |
5537 | ||
dd41f596 IM |
5538 | /* |
5539 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5540 | * It does so by boosting its priority to highest possible. |
5541 | * Used by CPU offline code. | |
1da177e4 LT |
5542 | */ |
5543 | void sched_idle_next(void) | |
5544 | { | |
48f24c4d | 5545 | int this_cpu = smp_processor_id(); |
70b97a7f | 5546 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5547 | struct task_struct *p = rq->idle; |
5548 | unsigned long flags; | |
5549 | ||
5550 | /* cpu has to be offline */ | |
48f24c4d | 5551 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5552 | |
48f24c4d IM |
5553 | /* |
5554 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5555 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5556 | */ |
5557 | spin_lock_irqsave(&rq->lock, flags); | |
5558 | ||
dd41f596 | 5559 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 5560 | |
94bc9a7b DA |
5561 | update_rq_clock(rq); |
5562 | activate_task(rq, p, 0); | |
1da177e4 LT |
5563 | |
5564 | spin_unlock_irqrestore(&rq->lock, flags); | |
5565 | } | |
5566 | ||
48f24c4d IM |
5567 | /* |
5568 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5569 | * offline. |
5570 | */ | |
5571 | void idle_task_exit(void) | |
5572 | { | |
5573 | struct mm_struct *mm = current->active_mm; | |
5574 | ||
5575 | BUG_ON(cpu_online(smp_processor_id())); | |
5576 | ||
5577 | if (mm != &init_mm) | |
5578 | switch_mm(mm, &init_mm, current); | |
5579 | mmdrop(mm); | |
5580 | } | |
5581 | ||
054b9108 | 5582 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5583 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5584 | { |
70b97a7f | 5585 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5586 | |
5587 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5588 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5589 | |
5590 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5591 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5592 | |
48f24c4d | 5593 | get_task_struct(p); |
1da177e4 LT |
5594 | |
5595 | /* | |
5596 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 5597 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
5598 | * fine. |
5599 | */ | |
f7b4cddc | 5600 | spin_unlock_irq(&rq->lock); |
48f24c4d | 5601 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 5602 | spin_lock_irq(&rq->lock); |
1da177e4 | 5603 | |
48f24c4d | 5604 | put_task_struct(p); |
1da177e4 LT |
5605 | } |
5606 | ||
5607 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5608 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5609 | { | |
70b97a7f | 5610 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5611 | struct task_struct *next; |
48f24c4d | 5612 | |
dd41f596 IM |
5613 | for ( ; ; ) { |
5614 | if (!rq->nr_running) | |
5615 | break; | |
a8e504d2 | 5616 | update_rq_clock(rq); |
ff95f3df | 5617 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
5618 | if (!next) |
5619 | break; | |
5620 | migrate_dead(dead_cpu, next); | |
e692ab53 | 5621 | |
1da177e4 LT |
5622 | } |
5623 | } | |
5624 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5625 | ||
e692ab53 NP |
5626 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5627 | ||
5628 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5629 | { |
5630 | .procname = "sched_domain", | |
c57baf1e | 5631 | .mode = 0555, |
e0361851 | 5632 | }, |
38605cae | 5633 | {0, }, |
e692ab53 NP |
5634 | }; |
5635 | ||
5636 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 5637 | { |
c57baf1e | 5638 | .ctl_name = CTL_KERN, |
e0361851 | 5639 | .procname = "kernel", |
c57baf1e | 5640 | .mode = 0555, |
e0361851 AD |
5641 | .child = sd_ctl_dir, |
5642 | }, | |
38605cae | 5643 | {0, }, |
e692ab53 NP |
5644 | }; |
5645 | ||
5646 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5647 | { | |
5648 | struct ctl_table *entry = | |
5cf9f062 | 5649 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5650 | |
e692ab53 NP |
5651 | return entry; |
5652 | } | |
5653 | ||
6382bc90 MM |
5654 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5655 | { | |
cd790076 | 5656 | struct ctl_table *entry; |
6382bc90 | 5657 | |
cd790076 MM |
5658 | /* |
5659 | * In the intermediate directories, both the child directory and | |
5660 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5661 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5662 | * static strings and all have proc handlers. |
5663 | */ | |
5664 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5665 | if (entry->child) |
5666 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5667 | if (entry->proc_handler == NULL) |
5668 | kfree(entry->procname); | |
5669 | } | |
6382bc90 MM |
5670 | |
5671 | kfree(*tablep); | |
5672 | *tablep = NULL; | |
5673 | } | |
5674 | ||
e692ab53 | 5675 | static void |
e0361851 | 5676 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5677 | const char *procname, void *data, int maxlen, |
5678 | mode_t mode, proc_handler *proc_handler) | |
5679 | { | |
e692ab53 NP |
5680 | entry->procname = procname; |
5681 | entry->data = data; | |
5682 | entry->maxlen = maxlen; | |
5683 | entry->mode = mode; | |
5684 | entry->proc_handler = proc_handler; | |
5685 | } | |
5686 | ||
5687 | static struct ctl_table * | |
5688 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5689 | { | |
ace8b3d6 | 5690 | struct ctl_table *table = sd_alloc_ctl_entry(12); |
e692ab53 | 5691 | |
ad1cdc1d MM |
5692 | if (table == NULL) |
5693 | return NULL; | |
5694 | ||
e0361851 | 5695 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5696 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5697 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5698 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5699 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5700 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5701 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5702 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5703 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5704 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5705 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5706 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5707 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5708 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5709 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5710 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5711 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5712 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5713 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5714 | &sd->cache_nice_tries, |
5715 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5716 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5717 | sizeof(int), 0644, proc_dointvec_minmax); |
6323469f | 5718 | /* &table[11] is terminator */ |
e692ab53 NP |
5719 | |
5720 | return table; | |
5721 | } | |
5722 | ||
9a4e7159 | 5723 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5724 | { |
5725 | struct ctl_table *entry, *table; | |
5726 | struct sched_domain *sd; | |
5727 | int domain_num = 0, i; | |
5728 | char buf[32]; | |
5729 | ||
5730 | for_each_domain(cpu, sd) | |
5731 | domain_num++; | |
5732 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5733 | if (table == NULL) |
5734 | return NULL; | |
e692ab53 NP |
5735 | |
5736 | i = 0; | |
5737 | for_each_domain(cpu, sd) { | |
5738 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5739 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5740 | entry->mode = 0555; |
e692ab53 NP |
5741 | entry->child = sd_alloc_ctl_domain_table(sd); |
5742 | entry++; | |
5743 | i++; | |
5744 | } | |
5745 | return table; | |
5746 | } | |
5747 | ||
5748 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5749 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
5750 | { |
5751 | int i, cpu_num = num_online_cpus(); | |
5752 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
5753 | char buf[32]; | |
5754 | ||
7378547f MM |
5755 | WARN_ON(sd_ctl_dir[0].child); |
5756 | sd_ctl_dir[0].child = entry; | |
5757 | ||
ad1cdc1d MM |
5758 | if (entry == NULL) |
5759 | return; | |
5760 | ||
97b6ea7b | 5761 | for_each_online_cpu(i) { |
e692ab53 | 5762 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5763 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5764 | entry->mode = 0555; |
e692ab53 | 5765 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5766 | entry++; |
e692ab53 | 5767 | } |
7378547f MM |
5768 | |
5769 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5770 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5771 | } | |
6382bc90 | 5772 | |
7378547f | 5773 | /* may be called multiple times per register */ |
6382bc90 MM |
5774 | static void unregister_sched_domain_sysctl(void) |
5775 | { | |
7378547f MM |
5776 | if (sd_sysctl_header) |
5777 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5778 | sd_sysctl_header = NULL; |
7378547f MM |
5779 | if (sd_ctl_dir[0].child) |
5780 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5781 | } |
e692ab53 | 5782 | #else |
6382bc90 MM |
5783 | static void register_sched_domain_sysctl(void) |
5784 | { | |
5785 | } | |
5786 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5787 | { |
5788 | } | |
5789 | #endif | |
5790 | ||
1da177e4 LT |
5791 | /* |
5792 | * migration_call - callback that gets triggered when a CPU is added. | |
5793 | * Here we can start up the necessary migration thread for the new CPU. | |
5794 | */ | |
48f24c4d IM |
5795 | static int __cpuinit |
5796 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5797 | { |
1da177e4 | 5798 | struct task_struct *p; |
48f24c4d | 5799 | int cpu = (long)hcpu; |
1da177e4 | 5800 | unsigned long flags; |
70b97a7f | 5801 | struct rq *rq; |
1da177e4 LT |
5802 | |
5803 | switch (action) { | |
5be9361c | 5804 | |
1da177e4 | 5805 | case CPU_UP_PREPARE: |
8bb78442 | 5806 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 5807 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
5808 | if (IS_ERR(p)) |
5809 | return NOTIFY_BAD; | |
1da177e4 LT |
5810 | kthread_bind(p, cpu); |
5811 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5812 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 5813 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
5814 | task_rq_unlock(rq, &flags); |
5815 | cpu_rq(cpu)->migration_thread = p; | |
5816 | break; | |
48f24c4d | 5817 | |
1da177e4 | 5818 | case CPU_ONLINE: |
8bb78442 | 5819 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 5820 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 5821 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
5822 | |
5823 | /* Update our root-domain */ | |
5824 | rq = cpu_rq(cpu); | |
5825 | spin_lock_irqsave(&rq->lock, flags); | |
5826 | if (rq->rd) { | |
5827 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
5828 | cpu_set(cpu, rq->rd->online); | |
5829 | } | |
5830 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 5831 | break; |
48f24c4d | 5832 | |
1da177e4 LT |
5833 | #ifdef CONFIG_HOTPLUG_CPU |
5834 | case CPU_UP_CANCELED: | |
8bb78442 | 5835 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5836 | if (!cpu_rq(cpu)->migration_thread) |
5837 | break; | |
41a2d6cf | 5838 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
5839 | kthread_bind(cpu_rq(cpu)->migration_thread, |
5840 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
5841 | kthread_stop(cpu_rq(cpu)->migration_thread); |
5842 | cpu_rq(cpu)->migration_thread = NULL; | |
5843 | break; | |
48f24c4d | 5844 | |
1da177e4 | 5845 | case CPU_DEAD: |
8bb78442 | 5846 | case CPU_DEAD_FROZEN: |
470fd646 | 5847 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
5848 | migrate_live_tasks(cpu); |
5849 | rq = cpu_rq(cpu); | |
5850 | kthread_stop(rq->migration_thread); | |
5851 | rq->migration_thread = NULL; | |
5852 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 5853 | spin_lock_irq(&rq->lock); |
a8e504d2 | 5854 | update_rq_clock(rq); |
2e1cb74a | 5855 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 5856 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
5857 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5858 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 5859 | migrate_dead_tasks(cpu); |
d2da272a | 5860 | spin_unlock_irq(&rq->lock); |
470fd646 | 5861 | cpuset_unlock(); |
1da177e4 LT |
5862 | migrate_nr_uninterruptible(rq); |
5863 | BUG_ON(rq->nr_running != 0); | |
5864 | ||
41a2d6cf IM |
5865 | /* |
5866 | * No need to migrate the tasks: it was best-effort if | |
5867 | * they didn't take sched_hotcpu_mutex. Just wake up | |
5868 | * the requestors. | |
5869 | */ | |
1da177e4 LT |
5870 | spin_lock_irq(&rq->lock); |
5871 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
5872 | struct migration_req *req; |
5873 | ||
1da177e4 | 5874 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5875 | struct migration_req, list); |
1da177e4 LT |
5876 | list_del_init(&req->list); |
5877 | complete(&req->done); | |
5878 | } | |
5879 | spin_unlock_irq(&rq->lock); | |
5880 | break; | |
57d885fe | 5881 | |
08f503b0 GH |
5882 | case CPU_DYING: |
5883 | case CPU_DYING_FROZEN: | |
57d885fe GH |
5884 | /* Update our root-domain */ |
5885 | rq = cpu_rq(cpu); | |
5886 | spin_lock_irqsave(&rq->lock, flags); | |
5887 | if (rq->rd) { | |
5888 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
5889 | cpu_clear(cpu, rq->rd->online); | |
5890 | } | |
5891 | spin_unlock_irqrestore(&rq->lock, flags); | |
5892 | break; | |
1da177e4 LT |
5893 | #endif |
5894 | } | |
5895 | return NOTIFY_OK; | |
5896 | } | |
5897 | ||
5898 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
5899 | * happens before everything else. | |
5900 | */ | |
26c2143b | 5901 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5902 | .notifier_call = migration_call, |
5903 | .priority = 10 | |
5904 | }; | |
5905 | ||
e6fe6649 | 5906 | void __init migration_init(void) |
1da177e4 LT |
5907 | { |
5908 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5909 | int err; |
48f24c4d IM |
5910 | |
5911 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5912 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5913 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5914 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5915 | register_cpu_notifier(&migration_notifier); | |
1da177e4 LT |
5916 | } |
5917 | #endif | |
5918 | ||
5919 | #ifdef CONFIG_SMP | |
476f3534 CL |
5920 | |
5921 | /* Number of possible processor ids */ | |
5922 | int nr_cpu_ids __read_mostly = NR_CPUS; | |
5923 | EXPORT_SYMBOL(nr_cpu_ids); | |
5924 | ||
3e9830dc | 5925 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff IM |
5926 | |
5927 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level) | |
1da177e4 | 5928 | { |
4dcf6aff IM |
5929 | struct sched_group *group = sd->groups; |
5930 | cpumask_t groupmask; | |
5931 | char str[NR_CPUS]; | |
1da177e4 | 5932 | |
4dcf6aff IM |
5933 | cpumask_scnprintf(str, NR_CPUS, sd->span); |
5934 | cpus_clear(groupmask); | |
5935 | ||
5936 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5937 | ||
5938 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
5939 | printk("does not load-balance\n"); | |
5940 | if (sd->parent) | |
5941 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
5942 | " has parent"); | |
5943 | return -1; | |
41c7ce9a NP |
5944 | } |
5945 | ||
4dcf6aff IM |
5946 | printk(KERN_CONT "span %s\n", str); |
5947 | ||
5948 | if (!cpu_isset(cpu, sd->span)) { | |
5949 | printk(KERN_ERR "ERROR: domain->span does not contain " | |
5950 | "CPU%d\n", cpu); | |
5951 | } | |
5952 | if (!cpu_isset(cpu, group->cpumask)) { | |
5953 | printk(KERN_ERR "ERROR: domain->groups does not contain" | |
5954 | " CPU%d\n", cpu); | |
5955 | } | |
1da177e4 | 5956 | |
4dcf6aff | 5957 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5958 | do { |
4dcf6aff IM |
5959 | if (!group) { |
5960 | printk("\n"); | |
5961 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5962 | break; |
5963 | } | |
5964 | ||
4dcf6aff IM |
5965 | if (!group->__cpu_power) { |
5966 | printk(KERN_CONT "\n"); | |
5967 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5968 | "set\n"); | |
5969 | break; | |
5970 | } | |
1da177e4 | 5971 | |
4dcf6aff IM |
5972 | if (!cpus_weight(group->cpumask)) { |
5973 | printk(KERN_CONT "\n"); | |
5974 | printk(KERN_ERR "ERROR: empty group\n"); | |
5975 | break; | |
5976 | } | |
1da177e4 | 5977 | |
4dcf6aff IM |
5978 | if (cpus_intersects(groupmask, group->cpumask)) { |
5979 | printk(KERN_CONT "\n"); | |
5980 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
5981 | break; | |
5982 | } | |
1da177e4 | 5983 | |
4dcf6aff | 5984 | cpus_or(groupmask, groupmask, group->cpumask); |
1da177e4 | 5985 | |
4dcf6aff IM |
5986 | cpumask_scnprintf(str, NR_CPUS, group->cpumask); |
5987 | printk(KERN_CONT " %s", str); | |
1da177e4 | 5988 | |
4dcf6aff IM |
5989 | group = group->next; |
5990 | } while (group != sd->groups); | |
5991 | printk(KERN_CONT "\n"); | |
1da177e4 | 5992 | |
4dcf6aff IM |
5993 | if (!cpus_equal(sd->span, groupmask)) |
5994 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); | |
1da177e4 | 5995 | |
4dcf6aff IM |
5996 | if (sd->parent && !cpus_subset(groupmask, sd->parent->span)) |
5997 | printk(KERN_ERR "ERROR: parent span is not a superset " | |
5998 | "of domain->span\n"); | |
5999 | return 0; | |
6000 | } | |
1da177e4 | 6001 | |
4dcf6aff IM |
6002 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6003 | { | |
6004 | int level = 0; | |
1da177e4 | 6005 | |
4dcf6aff IM |
6006 | if (!sd) { |
6007 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6008 | return; | |
6009 | } | |
1da177e4 | 6010 | |
4dcf6aff IM |
6011 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6012 | ||
6013 | for (;;) { | |
6014 | if (sched_domain_debug_one(sd, cpu, level)) | |
6015 | break; | |
1da177e4 LT |
6016 | level++; |
6017 | sd = sd->parent; | |
33859f7f | 6018 | if (!sd) |
4dcf6aff IM |
6019 | break; |
6020 | } | |
1da177e4 LT |
6021 | } |
6022 | #else | |
48f24c4d | 6023 | # define sched_domain_debug(sd, cpu) do { } while (0) |
1da177e4 LT |
6024 | #endif |
6025 | ||
1a20ff27 | 6026 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
6027 | { |
6028 | if (cpus_weight(sd->span) == 1) | |
6029 | return 1; | |
6030 | ||
6031 | /* Following flags need at least 2 groups */ | |
6032 | if (sd->flags & (SD_LOAD_BALANCE | | |
6033 | SD_BALANCE_NEWIDLE | | |
6034 | SD_BALANCE_FORK | | |
89c4710e SS |
6035 | SD_BALANCE_EXEC | |
6036 | SD_SHARE_CPUPOWER | | |
6037 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6038 | if (sd->groups != sd->groups->next) |
6039 | return 0; | |
6040 | } | |
6041 | ||
6042 | /* Following flags don't use groups */ | |
6043 | if (sd->flags & (SD_WAKE_IDLE | | |
6044 | SD_WAKE_AFFINE | | |
6045 | SD_WAKE_BALANCE)) | |
6046 | return 0; | |
6047 | ||
6048 | return 1; | |
6049 | } | |
6050 | ||
48f24c4d IM |
6051 | static int |
6052 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6053 | { |
6054 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6055 | ||
6056 | if (sd_degenerate(parent)) | |
6057 | return 1; | |
6058 | ||
6059 | if (!cpus_equal(sd->span, parent->span)) | |
6060 | return 0; | |
6061 | ||
6062 | /* Does parent contain flags not in child? */ | |
6063 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
6064 | if (cflags & SD_WAKE_AFFINE) | |
6065 | pflags &= ~SD_WAKE_BALANCE; | |
6066 | /* Flags needing groups don't count if only 1 group in parent */ | |
6067 | if (parent->groups == parent->groups->next) { | |
6068 | pflags &= ~(SD_LOAD_BALANCE | | |
6069 | SD_BALANCE_NEWIDLE | | |
6070 | SD_BALANCE_FORK | | |
89c4710e SS |
6071 | SD_BALANCE_EXEC | |
6072 | SD_SHARE_CPUPOWER | | |
6073 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
6074 | } |
6075 | if (~cflags & pflags) | |
6076 | return 0; | |
6077 | ||
6078 | return 1; | |
6079 | } | |
6080 | ||
57d885fe GH |
6081 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6082 | { | |
6083 | unsigned long flags; | |
6084 | const struct sched_class *class; | |
6085 | ||
6086 | spin_lock_irqsave(&rq->lock, flags); | |
6087 | ||
6088 | if (rq->rd) { | |
6089 | struct root_domain *old_rd = rq->rd; | |
6090 | ||
0eab9146 | 6091 | for (class = sched_class_highest; class; class = class->next) { |
57d885fe GH |
6092 | if (class->leave_domain) |
6093 | class->leave_domain(rq); | |
0eab9146 | 6094 | } |
57d885fe | 6095 | |
dc938520 GH |
6096 | cpu_clear(rq->cpu, old_rd->span); |
6097 | cpu_clear(rq->cpu, old_rd->online); | |
6098 | ||
57d885fe GH |
6099 | if (atomic_dec_and_test(&old_rd->refcount)) |
6100 | kfree(old_rd); | |
6101 | } | |
6102 | ||
6103 | atomic_inc(&rd->refcount); | |
6104 | rq->rd = rd; | |
6105 | ||
dc938520 | 6106 | cpu_set(rq->cpu, rd->span); |
1f94ef59 GH |
6107 | if (cpu_isset(rq->cpu, cpu_online_map)) |
6108 | cpu_set(rq->cpu, rd->online); | |
dc938520 | 6109 | |
0eab9146 | 6110 | for (class = sched_class_highest; class; class = class->next) { |
57d885fe GH |
6111 | if (class->join_domain) |
6112 | class->join_domain(rq); | |
0eab9146 | 6113 | } |
57d885fe GH |
6114 | |
6115 | spin_unlock_irqrestore(&rq->lock, flags); | |
6116 | } | |
6117 | ||
dc938520 | 6118 | static void init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6119 | { |
6120 | memset(rd, 0, sizeof(*rd)); | |
6121 | ||
dc938520 GH |
6122 | cpus_clear(rd->span); |
6123 | cpus_clear(rd->online); | |
57d885fe GH |
6124 | } |
6125 | ||
6126 | static void init_defrootdomain(void) | |
6127 | { | |
dc938520 | 6128 | init_rootdomain(&def_root_domain); |
57d885fe GH |
6129 | atomic_set(&def_root_domain.refcount, 1); |
6130 | } | |
6131 | ||
dc938520 | 6132 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6133 | { |
6134 | struct root_domain *rd; | |
6135 | ||
6136 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6137 | if (!rd) | |
6138 | return NULL; | |
6139 | ||
dc938520 | 6140 | init_rootdomain(rd); |
57d885fe GH |
6141 | |
6142 | return rd; | |
6143 | } | |
6144 | ||
1da177e4 | 6145 | /* |
0eab9146 | 6146 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6147 | * hold the hotplug lock. |
6148 | */ | |
0eab9146 IM |
6149 | static void |
6150 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6151 | { |
70b97a7f | 6152 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6153 | struct sched_domain *tmp; |
6154 | ||
6155 | /* Remove the sched domains which do not contribute to scheduling. */ | |
6156 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
6157 | struct sched_domain *parent = tmp->parent; | |
6158 | if (!parent) | |
6159 | break; | |
1a848870 | 6160 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6161 | tmp->parent = parent->parent; |
1a848870 SS |
6162 | if (parent->parent) |
6163 | parent->parent->child = tmp; | |
6164 | } | |
245af2c7 SS |
6165 | } |
6166 | ||
1a848870 | 6167 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6168 | sd = sd->parent; |
1a848870 SS |
6169 | if (sd) |
6170 | sd->child = NULL; | |
6171 | } | |
1da177e4 LT |
6172 | |
6173 | sched_domain_debug(sd, cpu); | |
6174 | ||
57d885fe | 6175 | rq_attach_root(rq, rd); |
674311d5 | 6176 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6177 | } |
6178 | ||
6179 | /* cpus with isolated domains */ | |
67af63a6 | 6180 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
6181 | |
6182 | /* Setup the mask of cpus configured for isolated domains */ | |
6183 | static int __init isolated_cpu_setup(char *str) | |
6184 | { | |
6185 | int ints[NR_CPUS], i; | |
6186 | ||
6187 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
6188 | cpus_clear(cpu_isolated_map); | |
6189 | for (i = 1; i <= ints[0]; i++) | |
6190 | if (ints[i] < NR_CPUS) | |
6191 | cpu_set(ints[i], cpu_isolated_map); | |
6192 | return 1; | |
6193 | } | |
6194 | ||
8927f494 | 6195 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6196 | |
6197 | /* | |
6711cab4 SS |
6198 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6199 | * to a function which identifies what group(along with sched group) a CPU | |
6200 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
6201 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
6202 | * |
6203 | * init_sched_build_groups will build a circular linked list of the groups | |
6204 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6205 | * and ->cpu_power to 0. | |
6206 | */ | |
a616058b | 6207 | static void |
6711cab4 SS |
6208 | init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map, |
6209 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, | |
6210 | struct sched_group **sg)) | |
1da177e4 LT |
6211 | { |
6212 | struct sched_group *first = NULL, *last = NULL; | |
6213 | cpumask_t covered = CPU_MASK_NONE; | |
6214 | int i; | |
6215 | ||
6216 | for_each_cpu_mask(i, span) { | |
6711cab4 SS |
6217 | struct sched_group *sg; |
6218 | int group = group_fn(i, cpu_map, &sg); | |
1da177e4 LT |
6219 | int j; |
6220 | ||
6221 | if (cpu_isset(i, covered)) | |
6222 | continue; | |
6223 | ||
6224 | sg->cpumask = CPU_MASK_NONE; | |
5517d86b | 6225 | sg->__cpu_power = 0; |
1da177e4 LT |
6226 | |
6227 | for_each_cpu_mask(j, span) { | |
6711cab4 | 6228 | if (group_fn(j, cpu_map, NULL) != group) |
1da177e4 LT |
6229 | continue; |
6230 | ||
6231 | cpu_set(j, covered); | |
6232 | cpu_set(j, sg->cpumask); | |
6233 | } | |
6234 | if (!first) | |
6235 | first = sg; | |
6236 | if (last) | |
6237 | last->next = sg; | |
6238 | last = sg; | |
6239 | } | |
6240 | last->next = first; | |
6241 | } | |
6242 | ||
9c1cfda2 | 6243 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6244 | |
9c1cfda2 | 6245 | #ifdef CONFIG_NUMA |
198e2f18 | 6246 | |
9c1cfda2 JH |
6247 | /** |
6248 | * find_next_best_node - find the next node to include in a sched_domain | |
6249 | * @node: node whose sched_domain we're building | |
6250 | * @used_nodes: nodes already in the sched_domain | |
6251 | * | |
41a2d6cf | 6252 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6253 | * finds the closest node not already in the @used_nodes map. |
6254 | * | |
6255 | * Should use nodemask_t. | |
6256 | */ | |
6257 | static int find_next_best_node(int node, unsigned long *used_nodes) | |
6258 | { | |
6259 | int i, n, val, min_val, best_node = 0; | |
6260 | ||
6261 | min_val = INT_MAX; | |
6262 | ||
6263 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6264 | /* Start at @node */ | |
6265 | n = (node + i) % MAX_NUMNODES; | |
6266 | ||
6267 | if (!nr_cpus_node(n)) | |
6268 | continue; | |
6269 | ||
6270 | /* Skip already used nodes */ | |
6271 | if (test_bit(n, used_nodes)) | |
6272 | continue; | |
6273 | ||
6274 | /* Simple min distance search */ | |
6275 | val = node_distance(node, n); | |
6276 | ||
6277 | if (val < min_val) { | |
6278 | min_val = val; | |
6279 | best_node = n; | |
6280 | } | |
6281 | } | |
6282 | ||
6283 | set_bit(best_node, used_nodes); | |
6284 | return best_node; | |
6285 | } | |
6286 | ||
6287 | /** | |
6288 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6289 | * @node: node whose cpumask we're constructing | |
6290 | * @size: number of nodes to include in this span | |
6291 | * | |
41a2d6cf | 6292 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6293 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6294 | * out optimally. | |
6295 | */ | |
6296 | static cpumask_t sched_domain_node_span(int node) | |
6297 | { | |
9c1cfda2 | 6298 | DECLARE_BITMAP(used_nodes, MAX_NUMNODES); |
48f24c4d IM |
6299 | cpumask_t span, nodemask; |
6300 | int i; | |
9c1cfda2 JH |
6301 | |
6302 | cpus_clear(span); | |
6303 | bitmap_zero(used_nodes, MAX_NUMNODES); | |
6304 | ||
6305 | nodemask = node_to_cpumask(node); | |
6306 | cpus_or(span, span, nodemask); | |
6307 | set_bit(node, used_nodes); | |
6308 | ||
6309 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
6310 | int next_node = find_next_best_node(node, used_nodes); | |
48f24c4d | 6311 | |
9c1cfda2 JH |
6312 | nodemask = node_to_cpumask(next_node); |
6313 | cpus_or(span, span, nodemask); | |
6314 | } | |
6315 | ||
6316 | return span; | |
6317 | } | |
6318 | #endif | |
6319 | ||
5c45bf27 | 6320 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6321 | |
9c1cfda2 | 6322 | /* |
48f24c4d | 6323 | * SMT sched-domains: |
9c1cfda2 | 6324 | */ |
1da177e4 LT |
6325 | #ifdef CONFIG_SCHED_SMT |
6326 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 6327 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 6328 | |
41a2d6cf IM |
6329 | static int |
6330 | cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1da177e4 | 6331 | { |
6711cab4 SS |
6332 | if (sg) |
6333 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
6334 | return cpu; |
6335 | } | |
6336 | #endif | |
6337 | ||
48f24c4d IM |
6338 | /* |
6339 | * multi-core sched-domains: | |
6340 | */ | |
1e9f28fa SS |
6341 | #ifdef CONFIG_SCHED_MC |
6342 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 6343 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
6344 | #endif |
6345 | ||
6346 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf IM |
6347 | static int |
6348 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1e9f28fa | 6349 | { |
6711cab4 | 6350 | int group; |
d5a7430d | 6351 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 6352 | cpus_and(mask, mask, *cpu_map); |
6711cab4 SS |
6353 | group = first_cpu(mask); |
6354 | if (sg) | |
6355 | *sg = &per_cpu(sched_group_core, group); | |
6356 | return group; | |
1e9f28fa SS |
6357 | } |
6358 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf IM |
6359 | static int |
6360 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1e9f28fa | 6361 | { |
6711cab4 SS |
6362 | if (sg) |
6363 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
6364 | return cpu; |
6365 | } | |
6366 | #endif | |
6367 | ||
1da177e4 | 6368 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 6369 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 6370 | |
41a2d6cf IM |
6371 | static int |
6372 | cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1da177e4 | 6373 | { |
6711cab4 | 6374 | int group; |
48f24c4d | 6375 | #ifdef CONFIG_SCHED_MC |
1e9f28fa | 6376 | cpumask_t mask = cpu_coregroup_map(cpu); |
a616058b | 6377 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 6378 | group = first_cpu(mask); |
1e9f28fa | 6379 | #elif defined(CONFIG_SCHED_SMT) |
d5a7430d | 6380 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 6381 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 6382 | group = first_cpu(mask); |
1da177e4 | 6383 | #else |
6711cab4 | 6384 | group = cpu; |
1da177e4 | 6385 | #endif |
6711cab4 SS |
6386 | if (sg) |
6387 | *sg = &per_cpu(sched_group_phys, group); | |
6388 | return group; | |
1da177e4 LT |
6389 | } |
6390 | ||
6391 | #ifdef CONFIG_NUMA | |
1da177e4 | 6392 | /* |
9c1cfda2 JH |
6393 | * The init_sched_build_groups can't handle what we want to do with node |
6394 | * groups, so roll our own. Now each node has its own list of groups which | |
6395 | * gets dynamically allocated. | |
1da177e4 | 6396 | */ |
9c1cfda2 | 6397 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
d1b55138 | 6398 | static struct sched_group **sched_group_nodes_bycpu[NR_CPUS]; |
1da177e4 | 6399 | |
9c1cfda2 | 6400 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 6401 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 6402 | |
6711cab4 SS |
6403 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
6404 | struct sched_group **sg) | |
9c1cfda2 | 6405 | { |
6711cab4 SS |
6406 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu)); |
6407 | int group; | |
6408 | ||
6409 | cpus_and(nodemask, nodemask, *cpu_map); | |
6410 | group = first_cpu(nodemask); | |
6411 | ||
6412 | if (sg) | |
6413 | *sg = &per_cpu(sched_group_allnodes, group); | |
6414 | return group; | |
1da177e4 | 6415 | } |
6711cab4 | 6416 | |
08069033 SS |
6417 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6418 | { | |
6419 | struct sched_group *sg = group_head; | |
6420 | int j; | |
6421 | ||
6422 | if (!sg) | |
6423 | return; | |
3a5c359a AK |
6424 | do { |
6425 | for_each_cpu_mask(j, sg->cpumask) { | |
6426 | struct sched_domain *sd; | |
08069033 | 6427 | |
3a5c359a AK |
6428 | sd = &per_cpu(phys_domains, j); |
6429 | if (j != first_cpu(sd->groups->cpumask)) { | |
6430 | /* | |
6431 | * Only add "power" once for each | |
6432 | * physical package. | |
6433 | */ | |
6434 | continue; | |
6435 | } | |
08069033 | 6436 | |
3a5c359a AK |
6437 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
6438 | } | |
6439 | sg = sg->next; | |
6440 | } while (sg != group_head); | |
08069033 | 6441 | } |
1da177e4 LT |
6442 | #endif |
6443 | ||
a616058b | 6444 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6445 | /* Free memory allocated for various sched_group structures */ |
6446 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6447 | { | |
a616058b | 6448 | int cpu, i; |
51888ca2 SV |
6449 | |
6450 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
6451 | struct sched_group **sched_group_nodes |
6452 | = sched_group_nodes_bycpu[cpu]; | |
6453 | ||
51888ca2 SV |
6454 | if (!sched_group_nodes) |
6455 | continue; | |
6456 | ||
6457 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6458 | cpumask_t nodemask = node_to_cpumask(i); | |
6459 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; | |
6460 | ||
6461 | cpus_and(nodemask, nodemask, *cpu_map); | |
6462 | if (cpus_empty(nodemask)) | |
6463 | continue; | |
6464 | ||
6465 | if (sg == NULL) | |
6466 | continue; | |
6467 | sg = sg->next; | |
6468 | next_sg: | |
6469 | oldsg = sg; | |
6470 | sg = sg->next; | |
6471 | kfree(oldsg); | |
6472 | if (oldsg != sched_group_nodes[i]) | |
6473 | goto next_sg; | |
6474 | } | |
6475 | kfree(sched_group_nodes); | |
6476 | sched_group_nodes_bycpu[cpu] = NULL; | |
6477 | } | |
51888ca2 | 6478 | } |
a616058b SS |
6479 | #else |
6480 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6481 | { | |
6482 | } | |
6483 | #endif | |
51888ca2 | 6484 | |
89c4710e SS |
6485 | /* |
6486 | * Initialize sched groups cpu_power. | |
6487 | * | |
6488 | * cpu_power indicates the capacity of sched group, which is used while | |
6489 | * distributing the load between different sched groups in a sched domain. | |
6490 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6491 | * there are asymmetries in the topology. If there are asymmetries, group | |
6492 | * having more cpu_power will pickup more load compared to the group having | |
6493 | * less cpu_power. | |
6494 | * | |
6495 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
6496 | * the maximum number of tasks a group can handle in the presence of other idle | |
6497 | * or lightly loaded groups in the same sched domain. | |
6498 | */ | |
6499 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6500 | { | |
6501 | struct sched_domain *child; | |
6502 | struct sched_group *group; | |
6503 | ||
6504 | WARN_ON(!sd || !sd->groups); | |
6505 | ||
6506 | if (cpu != first_cpu(sd->groups->cpumask)) | |
6507 | return; | |
6508 | ||
6509 | child = sd->child; | |
6510 | ||
5517d86b ED |
6511 | sd->groups->__cpu_power = 0; |
6512 | ||
89c4710e SS |
6513 | /* |
6514 | * For perf policy, if the groups in child domain share resources | |
6515 | * (for example cores sharing some portions of the cache hierarchy | |
6516 | * or SMT), then set this domain groups cpu_power such that each group | |
6517 | * can handle only one task, when there are other idle groups in the | |
6518 | * same sched domain. | |
6519 | */ | |
6520 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
6521 | (child->flags & | |
6522 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 6523 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
6524 | return; |
6525 | } | |
6526 | ||
89c4710e SS |
6527 | /* |
6528 | * add cpu_power of each child group to this groups cpu_power | |
6529 | */ | |
6530 | group = child->groups; | |
6531 | do { | |
5517d86b | 6532 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
6533 | group = group->next; |
6534 | } while (group != child->groups); | |
6535 | } | |
6536 | ||
1da177e4 | 6537 | /* |
1a20ff27 DG |
6538 | * Build sched domains for a given set of cpus and attach the sched domains |
6539 | * to the individual cpus | |
1da177e4 | 6540 | */ |
51888ca2 | 6541 | static int build_sched_domains(const cpumask_t *cpu_map) |
1da177e4 LT |
6542 | { |
6543 | int i; | |
57d885fe | 6544 | struct root_domain *rd; |
d1b55138 JH |
6545 | #ifdef CONFIG_NUMA |
6546 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 6547 | int sd_allnodes = 0; |
d1b55138 JH |
6548 | |
6549 | /* | |
6550 | * Allocate the per-node list of sched groups | |
6551 | */ | |
5cf9f062 | 6552 | sched_group_nodes = kcalloc(MAX_NUMNODES, sizeof(struct sched_group *), |
41a2d6cf | 6553 | GFP_KERNEL); |
d1b55138 JH |
6554 | if (!sched_group_nodes) { |
6555 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 6556 | return -ENOMEM; |
d1b55138 JH |
6557 | } |
6558 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
6559 | #endif | |
1da177e4 | 6560 | |
dc938520 | 6561 | rd = alloc_rootdomain(); |
57d885fe GH |
6562 | if (!rd) { |
6563 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
6564 | return -ENOMEM; | |
6565 | } | |
6566 | ||
1da177e4 | 6567 | /* |
1a20ff27 | 6568 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 6569 | */ |
1a20ff27 | 6570 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6571 | struct sched_domain *sd = NULL, *p; |
6572 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(i)); | |
6573 | ||
1a20ff27 | 6574 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6575 | |
6576 | #ifdef CONFIG_NUMA | |
dd41f596 IM |
6577 | if (cpus_weight(*cpu_map) > |
6578 | SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) { | |
9c1cfda2 JH |
6579 | sd = &per_cpu(allnodes_domains, i); |
6580 | *sd = SD_ALLNODES_INIT; | |
6581 | sd->span = *cpu_map; | |
6711cab4 | 6582 | cpu_to_allnodes_group(i, cpu_map, &sd->groups); |
9c1cfda2 | 6583 | p = sd; |
6711cab4 | 6584 | sd_allnodes = 1; |
9c1cfda2 JH |
6585 | } else |
6586 | p = NULL; | |
6587 | ||
1da177e4 | 6588 | sd = &per_cpu(node_domains, i); |
1da177e4 | 6589 | *sd = SD_NODE_INIT; |
9c1cfda2 JH |
6590 | sd->span = sched_domain_node_span(cpu_to_node(i)); |
6591 | sd->parent = p; | |
1a848870 SS |
6592 | if (p) |
6593 | p->child = sd; | |
9c1cfda2 | 6594 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
6595 | #endif |
6596 | ||
6597 | p = sd; | |
6598 | sd = &per_cpu(phys_domains, i); | |
1da177e4 LT |
6599 | *sd = SD_CPU_INIT; |
6600 | sd->span = nodemask; | |
6601 | sd->parent = p; | |
1a848870 SS |
6602 | if (p) |
6603 | p->child = sd; | |
6711cab4 | 6604 | cpu_to_phys_group(i, cpu_map, &sd->groups); |
1da177e4 | 6605 | |
1e9f28fa SS |
6606 | #ifdef CONFIG_SCHED_MC |
6607 | p = sd; | |
6608 | sd = &per_cpu(core_domains, i); | |
1e9f28fa SS |
6609 | *sd = SD_MC_INIT; |
6610 | sd->span = cpu_coregroup_map(i); | |
6611 | cpus_and(sd->span, sd->span, *cpu_map); | |
6612 | sd->parent = p; | |
1a848870 | 6613 | p->child = sd; |
6711cab4 | 6614 | cpu_to_core_group(i, cpu_map, &sd->groups); |
1e9f28fa SS |
6615 | #endif |
6616 | ||
1da177e4 LT |
6617 | #ifdef CONFIG_SCHED_SMT |
6618 | p = sd; | |
6619 | sd = &per_cpu(cpu_domains, i); | |
1da177e4 | 6620 | *sd = SD_SIBLING_INIT; |
d5a7430d | 6621 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6622 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 6623 | sd->parent = p; |
1a848870 | 6624 | p->child = sd; |
6711cab4 | 6625 | cpu_to_cpu_group(i, cpu_map, &sd->groups); |
1da177e4 LT |
6626 | #endif |
6627 | } | |
6628 | ||
6629 | #ifdef CONFIG_SCHED_SMT | |
6630 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 6631 | for_each_cpu_mask(i, *cpu_map) { |
d5a7430d | 6632 | cpumask_t this_sibling_map = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6633 | cpus_and(this_sibling_map, this_sibling_map, *cpu_map); |
1da177e4 LT |
6634 | if (i != first_cpu(this_sibling_map)) |
6635 | continue; | |
6636 | ||
dd41f596 IM |
6637 | init_sched_build_groups(this_sibling_map, cpu_map, |
6638 | &cpu_to_cpu_group); | |
1da177e4 LT |
6639 | } |
6640 | #endif | |
6641 | ||
1e9f28fa SS |
6642 | #ifdef CONFIG_SCHED_MC |
6643 | /* Set up multi-core groups */ | |
6644 | for_each_cpu_mask(i, *cpu_map) { | |
6645 | cpumask_t this_core_map = cpu_coregroup_map(i); | |
6646 | cpus_and(this_core_map, this_core_map, *cpu_map); | |
6647 | if (i != first_cpu(this_core_map)) | |
6648 | continue; | |
dd41f596 IM |
6649 | init_sched_build_groups(this_core_map, cpu_map, |
6650 | &cpu_to_core_group); | |
1e9f28fa SS |
6651 | } |
6652 | #endif | |
6653 | ||
1da177e4 LT |
6654 | /* Set up physical groups */ |
6655 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6656 | cpumask_t nodemask = node_to_cpumask(i); | |
6657 | ||
1a20ff27 | 6658 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6659 | if (cpus_empty(nodemask)) |
6660 | continue; | |
6661 | ||
6711cab4 | 6662 | init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group); |
1da177e4 LT |
6663 | } |
6664 | ||
6665 | #ifdef CONFIG_NUMA | |
6666 | /* Set up node groups */ | |
6711cab4 | 6667 | if (sd_allnodes) |
dd41f596 IM |
6668 | init_sched_build_groups(*cpu_map, cpu_map, |
6669 | &cpu_to_allnodes_group); | |
9c1cfda2 JH |
6670 | |
6671 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6672 | /* Set up node groups */ | |
6673 | struct sched_group *sg, *prev; | |
6674 | cpumask_t nodemask = node_to_cpumask(i); | |
6675 | cpumask_t domainspan; | |
6676 | cpumask_t covered = CPU_MASK_NONE; | |
6677 | int j; | |
6678 | ||
6679 | cpus_and(nodemask, nodemask, *cpu_map); | |
d1b55138 JH |
6680 | if (cpus_empty(nodemask)) { |
6681 | sched_group_nodes[i] = NULL; | |
9c1cfda2 | 6682 | continue; |
d1b55138 | 6683 | } |
9c1cfda2 JH |
6684 | |
6685 | domainspan = sched_domain_node_span(i); | |
6686 | cpus_and(domainspan, domainspan, *cpu_map); | |
6687 | ||
15f0b676 | 6688 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
6689 | if (!sg) { |
6690 | printk(KERN_WARNING "Can not alloc domain group for " | |
6691 | "node %d\n", i); | |
6692 | goto error; | |
6693 | } | |
9c1cfda2 JH |
6694 | sched_group_nodes[i] = sg; |
6695 | for_each_cpu_mask(j, nodemask) { | |
6696 | struct sched_domain *sd; | |
9761eea8 | 6697 | |
9c1cfda2 JH |
6698 | sd = &per_cpu(node_domains, j); |
6699 | sd->groups = sg; | |
9c1cfda2 | 6700 | } |
5517d86b | 6701 | sg->__cpu_power = 0; |
9c1cfda2 | 6702 | sg->cpumask = nodemask; |
51888ca2 | 6703 | sg->next = sg; |
9c1cfda2 JH |
6704 | cpus_or(covered, covered, nodemask); |
6705 | prev = sg; | |
6706 | ||
6707 | for (j = 0; j < MAX_NUMNODES; j++) { | |
6708 | cpumask_t tmp, notcovered; | |
6709 | int n = (i + j) % MAX_NUMNODES; | |
6710 | ||
6711 | cpus_complement(notcovered, covered); | |
6712 | cpus_and(tmp, notcovered, *cpu_map); | |
6713 | cpus_and(tmp, tmp, domainspan); | |
6714 | if (cpus_empty(tmp)) | |
6715 | break; | |
6716 | ||
6717 | nodemask = node_to_cpumask(n); | |
6718 | cpus_and(tmp, tmp, nodemask); | |
6719 | if (cpus_empty(tmp)) | |
6720 | continue; | |
6721 | ||
15f0b676 SV |
6722 | sg = kmalloc_node(sizeof(struct sched_group), |
6723 | GFP_KERNEL, i); | |
9c1cfda2 JH |
6724 | if (!sg) { |
6725 | printk(KERN_WARNING | |
6726 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 6727 | goto error; |
9c1cfda2 | 6728 | } |
5517d86b | 6729 | sg->__cpu_power = 0; |
9c1cfda2 | 6730 | sg->cpumask = tmp; |
51888ca2 | 6731 | sg->next = prev->next; |
9c1cfda2 JH |
6732 | cpus_or(covered, covered, tmp); |
6733 | prev->next = sg; | |
6734 | prev = sg; | |
6735 | } | |
9c1cfda2 | 6736 | } |
1da177e4 LT |
6737 | #endif |
6738 | ||
6739 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 6740 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 6741 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6742 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
6743 | ||
89c4710e | 6744 | init_sched_groups_power(i, sd); |
5c45bf27 | 6745 | } |
1da177e4 | 6746 | #endif |
1e9f28fa | 6747 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 6748 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6749 | struct sched_domain *sd = &per_cpu(core_domains, i); |
6750 | ||
89c4710e | 6751 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
6752 | } |
6753 | #endif | |
1e9f28fa | 6754 | |
5c45bf27 | 6755 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6756 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
6757 | ||
89c4710e | 6758 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6759 | } |
6760 | ||
9c1cfda2 | 6761 | #ifdef CONFIG_NUMA |
08069033 SS |
6762 | for (i = 0; i < MAX_NUMNODES; i++) |
6763 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 6764 | |
6711cab4 SS |
6765 | if (sd_allnodes) { |
6766 | struct sched_group *sg; | |
f712c0c7 | 6767 | |
6711cab4 | 6768 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg); |
f712c0c7 SS |
6769 | init_numa_sched_groups_power(sg); |
6770 | } | |
9c1cfda2 JH |
6771 | #endif |
6772 | ||
1da177e4 | 6773 | /* Attach the domains */ |
1a20ff27 | 6774 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6775 | struct sched_domain *sd; |
6776 | #ifdef CONFIG_SCHED_SMT | |
6777 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
6778 | #elif defined(CONFIG_SCHED_MC) |
6779 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
6780 | #else |
6781 | sd = &per_cpu(phys_domains, i); | |
6782 | #endif | |
57d885fe | 6783 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 6784 | } |
51888ca2 SV |
6785 | |
6786 | return 0; | |
6787 | ||
a616058b | 6788 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6789 | error: |
6790 | free_sched_groups(cpu_map); | |
6791 | return -ENOMEM; | |
a616058b | 6792 | #endif |
1da177e4 | 6793 | } |
029190c5 PJ |
6794 | |
6795 | static cpumask_t *doms_cur; /* current sched domains */ | |
6796 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
6797 | ||
6798 | /* | |
6799 | * Special case: If a kmalloc of a doms_cur partition (array of | |
6800 | * cpumask_t) fails, then fallback to a single sched domain, | |
6801 | * as determined by the single cpumask_t fallback_doms. | |
6802 | */ | |
6803 | static cpumask_t fallback_doms; | |
6804 | ||
1a20ff27 | 6805 | /* |
41a2d6cf | 6806 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6807 | * For now this just excludes isolated cpus, but could be used to |
6808 | * exclude other special cases in the future. | |
1a20ff27 | 6809 | */ |
51888ca2 | 6810 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 6811 | { |
7378547f MM |
6812 | int err; |
6813 | ||
029190c5 PJ |
6814 | ndoms_cur = 1; |
6815 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
6816 | if (!doms_cur) | |
6817 | doms_cur = &fallback_doms; | |
6818 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
7378547f | 6819 | err = build_sched_domains(doms_cur); |
6382bc90 | 6820 | register_sched_domain_sysctl(); |
7378547f MM |
6821 | |
6822 | return err; | |
1a20ff27 DG |
6823 | } |
6824 | ||
6825 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map) | |
1da177e4 | 6826 | { |
51888ca2 | 6827 | free_sched_groups(cpu_map); |
9c1cfda2 | 6828 | } |
1da177e4 | 6829 | |
1a20ff27 DG |
6830 | /* |
6831 | * Detach sched domains from a group of cpus specified in cpu_map | |
6832 | * These cpus will now be attached to the NULL domain | |
6833 | */ | |
858119e1 | 6834 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6835 | { |
6836 | int i; | |
6837 | ||
6382bc90 MM |
6838 | unregister_sched_domain_sysctl(); |
6839 | ||
1a20ff27 | 6840 | for_each_cpu_mask(i, *cpu_map) |
57d885fe | 6841 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 DG |
6842 | synchronize_sched(); |
6843 | arch_destroy_sched_domains(cpu_map); | |
6844 | } | |
6845 | ||
029190c5 PJ |
6846 | /* |
6847 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6848 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6849 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6850 | * It destroys each deleted domain and builds each new domain. | |
6851 | * | |
6852 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
41a2d6cf IM |
6853 | * The masks don't intersect (don't overlap.) We should setup one |
6854 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6855 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6856 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6857 | * it as it is. | |
6858 | * | |
41a2d6cf IM |
6859 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
6860 | * ownership of it and will kfree it when done with it. If the caller | |
029190c5 PJ |
6861 | * failed the kmalloc call, then it can pass in doms_new == NULL, |
6862 | * and partition_sched_domains() will fallback to the single partition | |
6863 | * 'fallback_doms'. | |
6864 | * | |
6865 | * Call with hotplug lock held | |
6866 | */ | |
6867 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new) | |
6868 | { | |
6869 | int i, j; | |
6870 | ||
a1835615 SV |
6871 | lock_doms_cur(); |
6872 | ||
7378547f MM |
6873 | /* always unregister in case we don't destroy any domains */ |
6874 | unregister_sched_domain_sysctl(); | |
6875 | ||
029190c5 PJ |
6876 | if (doms_new == NULL) { |
6877 | ndoms_new = 1; | |
6878 | doms_new = &fallback_doms; | |
6879 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
6880 | } | |
6881 | ||
6882 | /* Destroy deleted domains */ | |
6883 | for (i = 0; i < ndoms_cur; i++) { | |
6884 | for (j = 0; j < ndoms_new; j++) { | |
6885 | if (cpus_equal(doms_cur[i], doms_new[j])) | |
6886 | goto match1; | |
6887 | } | |
6888 | /* no match - a current sched domain not in new doms_new[] */ | |
6889 | detach_destroy_domains(doms_cur + i); | |
6890 | match1: | |
6891 | ; | |
6892 | } | |
6893 | ||
6894 | /* Build new domains */ | |
6895 | for (i = 0; i < ndoms_new; i++) { | |
6896 | for (j = 0; j < ndoms_cur; j++) { | |
6897 | if (cpus_equal(doms_new[i], doms_cur[j])) | |
6898 | goto match2; | |
6899 | } | |
6900 | /* no match - add a new doms_new */ | |
6901 | build_sched_domains(doms_new + i); | |
6902 | match2: | |
6903 | ; | |
6904 | } | |
6905 | ||
6906 | /* Remember the new sched domains */ | |
6907 | if (doms_cur != &fallback_doms) | |
6908 | kfree(doms_cur); | |
6909 | doms_cur = doms_new; | |
6910 | ndoms_cur = ndoms_new; | |
7378547f MM |
6911 | |
6912 | register_sched_domain_sysctl(); | |
a1835615 SV |
6913 | |
6914 | unlock_doms_cur(); | |
029190c5 PJ |
6915 | } |
6916 | ||
5c45bf27 | 6917 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
6707de00 | 6918 | static int arch_reinit_sched_domains(void) |
5c45bf27 SS |
6919 | { |
6920 | int err; | |
6921 | ||
95402b38 | 6922 | get_online_cpus(); |
5c45bf27 SS |
6923 | detach_destroy_domains(&cpu_online_map); |
6924 | err = arch_init_sched_domains(&cpu_online_map); | |
95402b38 | 6925 | put_online_cpus(); |
5c45bf27 SS |
6926 | |
6927 | return err; | |
6928 | } | |
6929 | ||
6930 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6931 | { | |
6932 | int ret; | |
6933 | ||
6934 | if (buf[0] != '0' && buf[0] != '1') | |
6935 | return -EINVAL; | |
6936 | ||
6937 | if (smt) | |
6938 | sched_smt_power_savings = (buf[0] == '1'); | |
6939 | else | |
6940 | sched_mc_power_savings = (buf[0] == '1'); | |
6941 | ||
6942 | ret = arch_reinit_sched_domains(); | |
6943 | ||
6944 | return ret ? ret : count; | |
6945 | } | |
6946 | ||
5c45bf27 SS |
6947 | #ifdef CONFIG_SCHED_MC |
6948 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
6949 | { | |
6950 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
6951 | } | |
48f24c4d IM |
6952 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
6953 | const char *buf, size_t count) | |
5c45bf27 SS |
6954 | { |
6955 | return sched_power_savings_store(buf, count, 0); | |
6956 | } | |
6707de00 AB |
6957 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
6958 | sched_mc_power_savings_store); | |
5c45bf27 SS |
6959 | #endif |
6960 | ||
6961 | #ifdef CONFIG_SCHED_SMT | |
6962 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
6963 | { | |
6964 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
6965 | } | |
48f24c4d IM |
6966 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
6967 | const char *buf, size_t count) | |
5c45bf27 SS |
6968 | { |
6969 | return sched_power_savings_store(buf, count, 1); | |
6970 | } | |
6707de00 AB |
6971 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
6972 | sched_smt_power_savings_store); | |
6973 | #endif | |
6974 | ||
6975 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
6976 | { | |
6977 | int err = 0; | |
6978 | ||
6979 | #ifdef CONFIG_SCHED_SMT | |
6980 | if (smt_capable()) | |
6981 | err = sysfs_create_file(&cls->kset.kobj, | |
6982 | &attr_sched_smt_power_savings.attr); | |
6983 | #endif | |
6984 | #ifdef CONFIG_SCHED_MC | |
6985 | if (!err && mc_capable()) | |
6986 | err = sysfs_create_file(&cls->kset.kobj, | |
6987 | &attr_sched_mc_power_savings.attr); | |
6988 | #endif | |
6989 | return err; | |
6990 | } | |
5c45bf27 SS |
6991 | #endif |
6992 | ||
1da177e4 | 6993 | /* |
41a2d6cf | 6994 | * Force a reinitialization of the sched domains hierarchy. The domains |
1da177e4 | 6995 | * and groups cannot be updated in place without racing with the balancing |
41c7ce9a | 6996 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
6997 | * which will prevent rebalancing while the sched domains are recalculated. |
6998 | */ | |
6999 | static int update_sched_domains(struct notifier_block *nfb, | |
7000 | unsigned long action, void *hcpu) | |
7001 | { | |
1da177e4 LT |
7002 | switch (action) { |
7003 | case CPU_UP_PREPARE: | |
8bb78442 | 7004 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 7005 | case CPU_DOWN_PREPARE: |
8bb78442 | 7006 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 7007 | detach_destroy_domains(&cpu_online_map); |
1da177e4 LT |
7008 | return NOTIFY_OK; |
7009 | ||
7010 | case CPU_UP_CANCELED: | |
8bb78442 | 7011 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 7012 | case CPU_DOWN_FAILED: |
8bb78442 | 7013 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7014 | case CPU_ONLINE: |
8bb78442 | 7015 | case CPU_ONLINE_FROZEN: |
1da177e4 | 7016 | case CPU_DEAD: |
8bb78442 | 7017 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
7018 | /* |
7019 | * Fall through and re-initialise the domains. | |
7020 | */ | |
7021 | break; | |
7022 | default: | |
7023 | return NOTIFY_DONE; | |
7024 | } | |
7025 | ||
7026 | /* The hotplug lock is already held by cpu_up/cpu_down */ | |
1a20ff27 | 7027 | arch_init_sched_domains(&cpu_online_map); |
1da177e4 LT |
7028 | |
7029 | return NOTIFY_OK; | |
7030 | } | |
1da177e4 LT |
7031 | |
7032 | void __init sched_init_smp(void) | |
7033 | { | |
5c1e1767 NP |
7034 | cpumask_t non_isolated_cpus; |
7035 | ||
95402b38 | 7036 | get_online_cpus(); |
1a20ff27 | 7037 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 7038 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
7039 | if (cpus_empty(non_isolated_cpus)) |
7040 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
95402b38 | 7041 | put_online_cpus(); |
1da177e4 LT |
7042 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7043 | hotcpu_notifier(update_sched_domains, 0); | |
5c1e1767 NP |
7044 | |
7045 | /* Move init over to a non-isolated CPU */ | |
7046 | if (set_cpus_allowed(current, non_isolated_cpus) < 0) | |
7047 | BUG(); | |
19978ca6 | 7048 | sched_init_granularity(); |
1da177e4 LT |
7049 | } |
7050 | #else | |
7051 | void __init sched_init_smp(void) | |
7052 | { | |
19978ca6 | 7053 | sched_init_granularity(); |
1da177e4 LT |
7054 | } |
7055 | #endif /* CONFIG_SMP */ | |
7056 | ||
7057 | int in_sched_functions(unsigned long addr) | |
7058 | { | |
1da177e4 LT |
7059 | return in_lock_functions(addr) || |
7060 | (addr >= (unsigned long)__sched_text_start | |
7061 | && addr < (unsigned long)__sched_text_end); | |
7062 | } | |
7063 | ||
a9957449 | 7064 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7065 | { |
7066 | cfs_rq->tasks_timeline = RB_ROOT; | |
dd41f596 IM |
7067 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7068 | cfs_rq->rq = rq; | |
7069 | #endif | |
67e9fb2a | 7070 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7071 | } |
7072 | ||
fa85ae24 PZ |
7073 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7074 | { | |
7075 | struct rt_prio_array *array; | |
7076 | int i; | |
7077 | ||
7078 | array = &rt_rq->active; | |
7079 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7080 | INIT_LIST_HEAD(array->queue + i); | |
7081 | __clear_bit(i, array->bitmap); | |
7082 | } | |
7083 | /* delimiter for bitsearch: */ | |
7084 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7085 | ||
052f1dc7 | 7086 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
48d5e258 PZ |
7087 | rt_rq->highest_prio = MAX_RT_PRIO; |
7088 | #endif | |
fa85ae24 PZ |
7089 | #ifdef CONFIG_SMP |
7090 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 PZ |
7091 | rt_rq->overloaded = 0; |
7092 | #endif | |
7093 | ||
7094 | rt_rq->rt_time = 0; | |
7095 | rt_rq->rt_throttled = 0; | |
6f505b16 | 7096 | |
052f1dc7 | 7097 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7098 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7099 | rt_rq->rq = rq; |
7100 | #endif | |
fa85ae24 PZ |
7101 | } |
7102 | ||
6f505b16 PZ |
7103 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7104 | static void init_tg_cfs_entry(struct rq *rq, struct task_group *tg, | |
7105 | struct cfs_rq *cfs_rq, struct sched_entity *se, | |
7106 | int cpu, int add) | |
7107 | { | |
7108 | tg->cfs_rq[cpu] = cfs_rq; | |
7109 | init_cfs_rq(cfs_rq, rq); | |
7110 | cfs_rq->tg = tg; | |
7111 | if (add) | |
7112 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7113 | ||
7114 | tg->se[cpu] = se; | |
7115 | se->cfs_rq = &rq->cfs; | |
7116 | se->my_q = cfs_rq; | |
7117 | se->load.weight = tg->shares; | |
7118 | se->load.inv_weight = div64_64(1ULL<<32, se->load.weight); | |
7119 | se->parent = NULL; | |
7120 | } | |
052f1dc7 | 7121 | #endif |
6f505b16 | 7122 | |
052f1dc7 | 7123 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
7124 | static void init_tg_rt_entry(struct rq *rq, struct task_group *tg, |
7125 | struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, | |
7126 | int cpu, int add) | |
7127 | { | |
7128 | tg->rt_rq[cpu] = rt_rq; | |
7129 | init_rt_rq(rt_rq, rq); | |
7130 | rt_rq->tg = tg; | |
7131 | rt_rq->rt_se = rt_se; | |
7132 | if (add) | |
7133 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
7134 | ||
7135 | tg->rt_se[cpu] = rt_se; | |
7136 | rt_se->rt_rq = &rq->rt; | |
7137 | rt_se->my_q = rt_rq; | |
7138 | rt_se->parent = NULL; | |
7139 | INIT_LIST_HEAD(&rt_se->run_list); | |
7140 | } | |
7141 | #endif | |
7142 | ||
1da177e4 LT |
7143 | void __init sched_init(void) |
7144 | { | |
476f3534 | 7145 | int highest_cpu = 0; |
dd41f596 IM |
7146 | int i, j; |
7147 | ||
57d885fe GH |
7148 | #ifdef CONFIG_SMP |
7149 | init_defrootdomain(); | |
7150 | #endif | |
7151 | ||
052f1dc7 | 7152 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 PZ |
7153 | list_add(&init_task_group.list, &task_groups); |
7154 | #endif | |
7155 | ||
0a945022 | 7156 | for_each_possible_cpu(i) { |
70b97a7f | 7157 | struct rq *rq; |
1da177e4 LT |
7158 | |
7159 | rq = cpu_rq(i); | |
7160 | spin_lock_init(&rq->lock); | |
fcb99371 | 7161 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 7162 | rq->nr_running = 0; |
dd41f596 IM |
7163 | rq->clock = 1; |
7164 | init_cfs_rq(&rq->cfs, rq); | |
6f505b16 | 7165 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7166 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 7167 | init_task_group.shares = init_task_group_load; |
6f505b16 PZ |
7168 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
7169 | init_tg_cfs_entry(rq, &init_task_group, | |
7170 | &per_cpu(init_cfs_rq, i), | |
7171 | &per_cpu(init_sched_entity, i), i, 1); | |
7172 | ||
052f1dc7 PZ |
7173 | #endif |
7174 | #ifdef CONFIG_RT_GROUP_SCHED | |
9f0c1e56 PZ |
7175 | init_task_group.rt_runtime = |
7176 | sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
6f505b16 PZ |
7177 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
7178 | init_tg_rt_entry(rq, &init_task_group, | |
7179 | &per_cpu(init_rt_rq, i), | |
7180 | &per_cpu(init_sched_rt_entity, i), i, 1); | |
dd41f596 | 7181 | #endif |
fa85ae24 | 7182 | rq->rt_period_expire = 0; |
48d5e258 | 7183 | rq->rt_throttled = 0; |
1da177e4 | 7184 | |
dd41f596 IM |
7185 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7186 | rq->cpu_load[j] = 0; | |
1da177e4 | 7187 | #ifdef CONFIG_SMP |
41c7ce9a | 7188 | rq->sd = NULL; |
57d885fe | 7189 | rq->rd = NULL; |
1da177e4 | 7190 | rq->active_balance = 0; |
dd41f596 | 7191 | rq->next_balance = jiffies; |
1da177e4 | 7192 | rq->push_cpu = 0; |
0a2966b4 | 7193 | rq->cpu = i; |
1da177e4 LT |
7194 | rq->migration_thread = NULL; |
7195 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 7196 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 7197 | #endif |
8f4d37ec | 7198 | init_rq_hrtick(rq); |
1da177e4 | 7199 | atomic_set(&rq->nr_iowait, 0); |
476f3534 | 7200 | highest_cpu = i; |
1da177e4 LT |
7201 | } |
7202 | ||
2dd73a4f | 7203 | set_load_weight(&init_task); |
b50f60ce | 7204 | |
e107be36 AK |
7205 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7206 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7207 | #endif | |
7208 | ||
c9819f45 | 7209 | #ifdef CONFIG_SMP |
476f3534 | 7210 | nr_cpu_ids = highest_cpu + 1; |
c9819f45 CL |
7211 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); |
7212 | #endif | |
7213 | ||
b50f60ce HC |
7214 | #ifdef CONFIG_RT_MUTEXES |
7215 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
7216 | #endif | |
7217 | ||
1da177e4 LT |
7218 | /* |
7219 | * The boot idle thread does lazy MMU switching as well: | |
7220 | */ | |
7221 | atomic_inc(&init_mm.mm_count); | |
7222 | enter_lazy_tlb(&init_mm, current); | |
7223 | ||
7224 | /* | |
7225 | * Make us the idle thread. Technically, schedule() should not be | |
7226 | * called from this thread, however somewhere below it might be, | |
7227 | * but because we are the idle thread, we just pick up running again | |
7228 | * when this runqueue becomes "idle". | |
7229 | */ | |
7230 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
7231 | /* |
7232 | * During early bootup we pretend to be a normal task: | |
7233 | */ | |
7234 | current->sched_class = &fair_sched_class; | |
6892b75e IM |
7235 | |
7236 | scheduler_running = 1; | |
1da177e4 LT |
7237 | } |
7238 | ||
7239 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
7240 | void __might_sleep(char *file, int line) | |
7241 | { | |
48f24c4d | 7242 | #ifdef in_atomic |
1da177e4 LT |
7243 | static unsigned long prev_jiffy; /* ratelimiting */ |
7244 | ||
7245 | if ((in_atomic() || irqs_disabled()) && | |
7246 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
7247 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7248 | return; | |
7249 | prev_jiffy = jiffies; | |
91368d73 | 7250 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
7251 | " context at %s:%d\n", file, line); |
7252 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
7253 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 7254 | debug_show_held_locks(current); |
3117df04 IM |
7255 | if (irqs_disabled()) |
7256 | print_irqtrace_events(current); | |
1da177e4 LT |
7257 | dump_stack(); |
7258 | } | |
7259 | #endif | |
7260 | } | |
7261 | EXPORT_SYMBOL(__might_sleep); | |
7262 | #endif | |
7263 | ||
7264 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7265 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7266 | { | |
7267 | int on_rq; | |
7268 | update_rq_clock(rq); | |
7269 | on_rq = p->se.on_rq; | |
7270 | if (on_rq) | |
7271 | deactivate_task(rq, p, 0); | |
7272 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
7273 | if (on_rq) { | |
7274 | activate_task(rq, p, 0); | |
7275 | resched_task(rq->curr); | |
7276 | } | |
7277 | } | |
7278 | ||
1da177e4 LT |
7279 | void normalize_rt_tasks(void) |
7280 | { | |
a0f98a1c | 7281 | struct task_struct *g, *p; |
1da177e4 | 7282 | unsigned long flags; |
70b97a7f | 7283 | struct rq *rq; |
1da177e4 | 7284 | |
4cf5d77a | 7285 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7286 | do_each_thread(g, p) { |
178be793 IM |
7287 | /* |
7288 | * Only normalize user tasks: | |
7289 | */ | |
7290 | if (!p->mm) | |
7291 | continue; | |
7292 | ||
6cfb0d5d | 7293 | p->se.exec_start = 0; |
6cfb0d5d | 7294 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 7295 | p->se.wait_start = 0; |
dd41f596 | 7296 | p->se.sleep_start = 0; |
dd41f596 | 7297 | p->se.block_start = 0; |
6cfb0d5d | 7298 | #endif |
dd41f596 IM |
7299 | task_rq(p)->clock = 0; |
7300 | ||
7301 | if (!rt_task(p)) { | |
7302 | /* | |
7303 | * Renice negative nice level userspace | |
7304 | * tasks back to 0: | |
7305 | */ | |
7306 | if (TASK_NICE(p) < 0 && p->mm) | |
7307 | set_user_nice(p, 0); | |
1da177e4 | 7308 | continue; |
dd41f596 | 7309 | } |
1da177e4 | 7310 | |
4cf5d77a | 7311 | spin_lock(&p->pi_lock); |
b29739f9 | 7312 | rq = __task_rq_lock(p); |
1da177e4 | 7313 | |
178be793 | 7314 | normalize_task(rq, p); |
3a5e4dc1 | 7315 | |
b29739f9 | 7316 | __task_rq_unlock(rq); |
4cf5d77a | 7317 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7318 | } while_each_thread(g, p); |
7319 | ||
4cf5d77a | 7320 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7321 | } |
7322 | ||
7323 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
7324 | |
7325 | #ifdef CONFIG_IA64 | |
7326 | /* | |
7327 | * These functions are only useful for the IA64 MCA handling. | |
7328 | * | |
7329 | * They can only be called when the whole system has been | |
7330 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7331 | * activity can take place. Using them for anything else would | |
7332 | * be a serious bug, and as a result, they aren't even visible | |
7333 | * under any other configuration. | |
7334 | */ | |
7335 | ||
7336 | /** | |
7337 | * curr_task - return the current task for a given cpu. | |
7338 | * @cpu: the processor in question. | |
7339 | * | |
7340 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7341 | */ | |
36c8b586 | 7342 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7343 | { |
7344 | return cpu_curr(cpu); | |
7345 | } | |
7346 | ||
7347 | /** | |
7348 | * set_curr_task - set the current task for a given cpu. | |
7349 | * @cpu: the processor in question. | |
7350 | * @p: the task pointer to set. | |
7351 | * | |
7352 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7353 | * are serviced on a separate stack. It allows the architecture to switch the |
7354 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7355 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7356 | * and caller must save the original value of the current task (see | |
7357 | * curr_task() above) and restore that value before reenabling interrupts and | |
7358 | * re-starting the system. | |
7359 | * | |
7360 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7361 | */ | |
36c8b586 | 7362 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7363 | { |
7364 | cpu_curr(cpu) = p; | |
7365 | } | |
7366 | ||
7367 | #endif | |
29f59db3 | 7368 | |
052f1dc7 | 7369 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 7370 | |
bccbe08a PZ |
7371 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7372 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
7373 | { |
7374 | int i; | |
7375 | ||
7376 | for_each_possible_cpu(i) { | |
7377 | if (tg->cfs_rq) | |
7378 | kfree(tg->cfs_rq[i]); | |
7379 | if (tg->se) | |
7380 | kfree(tg->se[i]); | |
6f505b16 PZ |
7381 | } |
7382 | ||
7383 | kfree(tg->cfs_rq); | |
7384 | kfree(tg->se); | |
6f505b16 PZ |
7385 | } |
7386 | ||
bccbe08a | 7387 | static int alloc_fair_sched_group(struct task_group *tg) |
29f59db3 | 7388 | { |
29f59db3 SV |
7389 | struct cfs_rq *cfs_rq; |
7390 | struct sched_entity *se; | |
9b5b7751 | 7391 | struct rq *rq; |
29f59db3 SV |
7392 | int i; |
7393 | ||
9b5b7751 | 7394 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
7395 | if (!tg->cfs_rq) |
7396 | goto err; | |
9b5b7751 | 7397 | tg->se = kzalloc(sizeof(se) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
7398 | if (!tg->se) |
7399 | goto err; | |
052f1dc7 PZ |
7400 | |
7401 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
7402 | |
7403 | for_each_possible_cpu(i) { | |
9b5b7751 | 7404 | rq = cpu_rq(i); |
29f59db3 | 7405 | |
6f505b16 PZ |
7406 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), |
7407 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
7408 | if (!cfs_rq) |
7409 | goto err; | |
7410 | ||
6f505b16 PZ |
7411 | se = kmalloc_node(sizeof(struct sched_entity), |
7412 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
7413 | if (!se) |
7414 | goto err; | |
7415 | ||
052f1dc7 | 7416 | init_tg_cfs_entry(rq, tg, cfs_rq, se, i, 0); |
bccbe08a PZ |
7417 | } |
7418 | ||
7419 | return 1; | |
7420 | ||
7421 | err: | |
7422 | return 0; | |
7423 | } | |
7424 | ||
7425 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
7426 | { | |
7427 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
7428 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
7429 | } | |
7430 | ||
7431 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
7432 | { | |
7433 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
7434 | } | |
7435 | #else | |
7436 | static inline void free_fair_sched_group(struct task_group *tg) | |
7437 | { | |
7438 | } | |
7439 | ||
7440 | static inline int alloc_fair_sched_group(struct task_group *tg) | |
7441 | { | |
7442 | return 1; | |
7443 | } | |
7444 | ||
7445 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
7446 | { | |
7447 | } | |
7448 | ||
7449 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
7450 | { | |
7451 | } | |
052f1dc7 PZ |
7452 | #endif |
7453 | ||
7454 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
7455 | static void free_rt_sched_group(struct task_group *tg) |
7456 | { | |
7457 | int i; | |
7458 | ||
7459 | for_each_possible_cpu(i) { | |
7460 | if (tg->rt_rq) | |
7461 | kfree(tg->rt_rq[i]); | |
7462 | if (tg->rt_se) | |
7463 | kfree(tg->rt_se[i]); | |
7464 | } | |
7465 | ||
7466 | kfree(tg->rt_rq); | |
7467 | kfree(tg->rt_se); | |
7468 | } | |
7469 | ||
7470 | static int alloc_rt_sched_group(struct task_group *tg) | |
7471 | { | |
7472 | struct rt_rq *rt_rq; | |
7473 | struct sched_rt_entity *rt_se; | |
7474 | struct rq *rq; | |
7475 | int i; | |
7476 | ||
7477 | tg->rt_rq = kzalloc(sizeof(rt_rq) * NR_CPUS, GFP_KERNEL); | |
7478 | if (!tg->rt_rq) | |
7479 | goto err; | |
7480 | tg->rt_se = kzalloc(sizeof(rt_se) * NR_CPUS, GFP_KERNEL); | |
7481 | if (!tg->rt_se) | |
7482 | goto err; | |
7483 | ||
7484 | tg->rt_runtime = 0; | |
7485 | ||
7486 | for_each_possible_cpu(i) { | |
7487 | rq = cpu_rq(i); | |
7488 | ||
6f505b16 PZ |
7489 | rt_rq = kmalloc_node(sizeof(struct rt_rq), |
7490 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
7491 | if (!rt_rq) | |
7492 | goto err; | |
29f59db3 | 7493 | |
6f505b16 PZ |
7494 | rt_se = kmalloc_node(sizeof(struct sched_rt_entity), |
7495 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
7496 | if (!rt_se) | |
7497 | goto err; | |
29f59db3 | 7498 | |
6f505b16 | 7499 | init_tg_rt_entry(rq, tg, rt_rq, rt_se, i, 0); |
29f59db3 SV |
7500 | } |
7501 | ||
bccbe08a PZ |
7502 | return 1; |
7503 | ||
7504 | err: | |
7505 | return 0; | |
7506 | } | |
7507 | ||
7508 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
7509 | { | |
7510 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
7511 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
7512 | } | |
7513 | ||
7514 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
7515 | { | |
7516 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
7517 | } | |
7518 | #else | |
7519 | static inline void free_rt_sched_group(struct task_group *tg) | |
7520 | { | |
7521 | } | |
7522 | ||
7523 | static inline int alloc_rt_sched_group(struct task_group *tg) | |
7524 | { | |
7525 | return 1; | |
7526 | } | |
7527 | ||
7528 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
7529 | { | |
7530 | } | |
7531 | ||
7532 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
7533 | { | |
7534 | } | |
7535 | #endif | |
7536 | ||
7537 | static void free_sched_group(struct task_group *tg) | |
7538 | { | |
7539 | free_fair_sched_group(tg); | |
7540 | free_rt_sched_group(tg); | |
7541 | kfree(tg); | |
7542 | } | |
7543 | ||
7544 | /* allocate runqueue etc for a new task group */ | |
7545 | struct task_group *sched_create_group(void) | |
7546 | { | |
7547 | struct task_group *tg; | |
7548 | unsigned long flags; | |
7549 | int i; | |
7550 | ||
7551 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7552 | if (!tg) | |
7553 | return ERR_PTR(-ENOMEM); | |
7554 | ||
7555 | if (!alloc_fair_sched_group(tg)) | |
7556 | goto err; | |
7557 | ||
7558 | if (!alloc_rt_sched_group(tg)) | |
7559 | goto err; | |
7560 | ||
8ed36996 | 7561 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 7562 | for_each_possible_cpu(i) { |
bccbe08a PZ |
7563 | register_fair_sched_group(tg, i); |
7564 | register_rt_sched_group(tg, i); | |
9b5b7751 | 7565 | } |
6f505b16 | 7566 | list_add_rcu(&tg->list, &task_groups); |
8ed36996 | 7567 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 7568 | |
9b5b7751 | 7569 | return tg; |
29f59db3 SV |
7570 | |
7571 | err: | |
6f505b16 | 7572 | free_sched_group(tg); |
29f59db3 SV |
7573 | return ERR_PTR(-ENOMEM); |
7574 | } | |
7575 | ||
9b5b7751 | 7576 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7577 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7578 | { |
29f59db3 | 7579 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7580 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7581 | } |
7582 | ||
9b5b7751 | 7583 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7584 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 7585 | { |
8ed36996 | 7586 | unsigned long flags; |
9b5b7751 | 7587 | int i; |
29f59db3 | 7588 | |
8ed36996 | 7589 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 7590 | for_each_possible_cpu(i) { |
bccbe08a PZ |
7591 | unregister_fair_sched_group(tg, i); |
7592 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 7593 | } |
6f505b16 | 7594 | list_del_rcu(&tg->list); |
8ed36996 | 7595 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 7596 | |
9b5b7751 | 7597 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 7598 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
7599 | } |
7600 | ||
9b5b7751 | 7601 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7602 | * The caller of this function should have put the task in its new group |
7603 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7604 | * reflect its new group. | |
9b5b7751 SV |
7605 | */ |
7606 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
7607 | { |
7608 | int on_rq, running; | |
7609 | unsigned long flags; | |
7610 | struct rq *rq; | |
7611 | ||
7612 | rq = task_rq_lock(tsk, &flags); | |
7613 | ||
29f59db3 SV |
7614 | update_rq_clock(rq); |
7615 | ||
051a1d1a | 7616 | running = task_current(rq, tsk); |
29f59db3 SV |
7617 | on_rq = tsk->se.on_rq; |
7618 | ||
0e1f3483 | 7619 | if (on_rq) |
29f59db3 | 7620 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
7621 | if (unlikely(running)) |
7622 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 7623 | |
6f505b16 | 7624 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 7625 | |
810b3817 PZ |
7626 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7627 | if (tsk->sched_class->moved_group) | |
7628 | tsk->sched_class->moved_group(tsk); | |
7629 | #endif | |
7630 | ||
0e1f3483 HS |
7631 | if (unlikely(running)) |
7632 | tsk->sched_class->set_curr_task(rq); | |
7633 | if (on_rq) | |
7074badb | 7634 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7635 | |
29f59db3 SV |
7636 | task_rq_unlock(rq, &flags); |
7637 | } | |
7638 | ||
052f1dc7 | 7639 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
7640 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
7641 | { | |
7642 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
7643 | struct rq *rq = cfs_rq->rq; | |
7644 | int on_rq; | |
7645 | ||
62fb1851 | 7646 | spin_lock_irq(&rq->lock); |
29f59db3 SV |
7647 | |
7648 | on_rq = se->on_rq; | |
62fb1851 | 7649 | if (on_rq) |
29f59db3 SV |
7650 | dequeue_entity(cfs_rq, se, 0); |
7651 | ||
7652 | se->load.weight = shares; | |
7653 | se->load.inv_weight = div64_64((1ULL<<32), shares); | |
7654 | ||
62fb1851 | 7655 | if (on_rq) |
29f59db3 | 7656 | enqueue_entity(cfs_rq, se, 0); |
62fb1851 PZ |
7657 | |
7658 | spin_unlock_irq(&rq->lock); | |
29f59db3 SV |
7659 | } |
7660 | ||
8ed36996 PZ |
7661 | static DEFINE_MUTEX(shares_mutex); |
7662 | ||
4cf86d77 | 7663 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
7664 | { |
7665 | int i; | |
8ed36996 | 7666 | unsigned long flags; |
c61935fd | 7667 | |
62fb1851 PZ |
7668 | /* |
7669 | * A weight of 0 or 1 can cause arithmetics problems. | |
7670 | * (The default weight is 1024 - so there's no practical | |
7671 | * limitation from this.) | |
7672 | */ | |
7673 | if (shares < 2) | |
7674 | shares = 2; | |
7675 | ||
8ed36996 | 7676 | mutex_lock(&shares_mutex); |
9b5b7751 | 7677 | if (tg->shares == shares) |
5cb350ba | 7678 | goto done; |
29f59db3 | 7679 | |
8ed36996 | 7680 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
7681 | for_each_possible_cpu(i) |
7682 | unregister_fair_sched_group(tg, i); | |
8ed36996 | 7683 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
7684 | |
7685 | /* wait for any ongoing reference to this group to finish */ | |
7686 | synchronize_sched(); | |
7687 | ||
7688 | /* | |
7689 | * Now we are free to modify the group's share on each cpu | |
7690 | * w/o tripping rebalance_share or load_balance_fair. | |
7691 | */ | |
9b5b7751 | 7692 | tg->shares = shares; |
62fb1851 | 7693 | for_each_possible_cpu(i) |
9b5b7751 | 7694 | set_se_shares(tg->se[i], shares); |
29f59db3 | 7695 | |
6b2d7700 SV |
7696 | /* |
7697 | * Enable load balance activity on this group, by inserting it back on | |
7698 | * each cpu's rq->leaf_cfs_rq_list. | |
7699 | */ | |
8ed36996 | 7700 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
7701 | for_each_possible_cpu(i) |
7702 | register_fair_sched_group(tg, i); | |
8ed36996 | 7703 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 7704 | done: |
8ed36996 | 7705 | mutex_unlock(&shares_mutex); |
9b5b7751 | 7706 | return 0; |
29f59db3 SV |
7707 | } |
7708 | ||
5cb350ba DG |
7709 | unsigned long sched_group_shares(struct task_group *tg) |
7710 | { | |
7711 | return tg->shares; | |
7712 | } | |
052f1dc7 | 7713 | #endif |
5cb350ba | 7714 | |
052f1dc7 | 7715 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7716 | /* |
9f0c1e56 | 7717 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 7718 | */ |
9f0c1e56 PZ |
7719 | static DEFINE_MUTEX(rt_constraints_mutex); |
7720 | ||
7721 | static unsigned long to_ratio(u64 period, u64 runtime) | |
7722 | { | |
7723 | if (runtime == RUNTIME_INF) | |
7724 | return 1ULL << 16; | |
7725 | ||
2692a240 | 7726 | return div64_64(runtime << 16, period); |
9f0c1e56 PZ |
7727 | } |
7728 | ||
7729 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) | |
6f505b16 PZ |
7730 | { |
7731 | struct task_group *tgi; | |
7732 | unsigned long total = 0; | |
9f0c1e56 PZ |
7733 | unsigned long global_ratio = |
7734 | to_ratio(sysctl_sched_rt_period, | |
7735 | sysctl_sched_rt_runtime < 0 ? | |
7736 | RUNTIME_INF : sysctl_sched_rt_runtime); | |
6f505b16 PZ |
7737 | |
7738 | rcu_read_lock(); | |
9f0c1e56 PZ |
7739 | list_for_each_entry_rcu(tgi, &task_groups, list) { |
7740 | if (tgi == tg) | |
7741 | continue; | |
6f505b16 | 7742 | |
9f0c1e56 PZ |
7743 | total += to_ratio(period, tgi->rt_runtime); |
7744 | } | |
7745 | rcu_read_unlock(); | |
6f505b16 | 7746 | |
9f0c1e56 | 7747 | return total + to_ratio(period, runtime) < global_ratio; |
6f505b16 PZ |
7748 | } |
7749 | ||
521f1a24 DG |
7750 | /* Must be called with tasklist_lock held */ |
7751 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
7752 | { | |
7753 | struct task_struct *g, *p; | |
7754 | do_each_thread(g, p) { | |
7755 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
7756 | return 1; | |
7757 | } while_each_thread(g, p); | |
7758 | return 0; | |
7759 | } | |
7760 | ||
9f0c1e56 | 7761 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
6f505b16 | 7762 | { |
9f0c1e56 PZ |
7763 | u64 rt_runtime, rt_period; |
7764 | int err = 0; | |
7765 | ||
2692a240 | 7766 | rt_period = (u64)sysctl_sched_rt_period * NSEC_PER_USEC; |
9f0c1e56 PZ |
7767 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; |
7768 | if (rt_runtime_us == -1) | |
2692a240 | 7769 | rt_runtime = RUNTIME_INF; |
9f0c1e56 PZ |
7770 | |
7771 | mutex_lock(&rt_constraints_mutex); | |
521f1a24 DG |
7772 | read_lock(&tasklist_lock); |
7773 | if (rt_runtime_us == 0 && tg_has_rt_tasks(tg)) { | |
7774 | err = -EBUSY; | |
7775 | goto unlock; | |
7776 | } | |
9f0c1e56 PZ |
7777 | if (!__rt_schedulable(tg, rt_period, rt_runtime)) { |
7778 | err = -EINVAL; | |
7779 | goto unlock; | |
7780 | } | |
9f0c1e56 PZ |
7781 | tg->rt_runtime = rt_runtime; |
7782 | unlock: | |
521f1a24 | 7783 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7784 | mutex_unlock(&rt_constraints_mutex); |
7785 | ||
7786 | return err; | |
6f505b16 PZ |
7787 | } |
7788 | ||
9f0c1e56 PZ |
7789 | long sched_group_rt_runtime(struct task_group *tg) |
7790 | { | |
7791 | u64 rt_runtime_us; | |
7792 | ||
7793 | if (tg->rt_runtime == RUNTIME_INF) | |
7794 | return -1; | |
7795 | ||
7796 | rt_runtime_us = tg->rt_runtime; | |
7797 | do_div(rt_runtime_us, NSEC_PER_USEC); | |
7798 | return rt_runtime_us; | |
7799 | } | |
052f1dc7 PZ |
7800 | #endif |
7801 | #endif /* CONFIG_GROUP_SCHED */ | |
68318b8e | 7802 | |
052f1dc7 | 7803 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
7804 | |
7805 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 7806 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 7807 | { |
2b01dfe3 PM |
7808 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
7809 | struct task_group, css); | |
68318b8e SV |
7810 | } |
7811 | ||
7812 | static struct cgroup_subsys_state * | |
2b01dfe3 | 7813 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e SV |
7814 | { |
7815 | struct task_group *tg; | |
7816 | ||
2b01dfe3 | 7817 | if (!cgrp->parent) { |
68318b8e | 7818 | /* This is early initialization for the top cgroup */ |
2b01dfe3 | 7819 | init_task_group.css.cgroup = cgrp; |
68318b8e SV |
7820 | return &init_task_group.css; |
7821 | } | |
7822 | ||
7823 | /* we support only 1-level deep hierarchical scheduler atm */ | |
2b01dfe3 | 7824 | if (cgrp->parent->parent) |
68318b8e SV |
7825 | return ERR_PTR(-EINVAL); |
7826 | ||
7827 | tg = sched_create_group(); | |
7828 | if (IS_ERR(tg)) | |
7829 | return ERR_PTR(-ENOMEM); | |
7830 | ||
7831 | /* Bind the cgroup to task_group object we just created */ | |
2b01dfe3 | 7832 | tg->css.cgroup = cgrp; |
68318b8e SV |
7833 | |
7834 | return &tg->css; | |
7835 | } | |
7836 | ||
41a2d6cf IM |
7837 | static void |
7838 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 7839 | { |
2b01dfe3 | 7840 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7841 | |
7842 | sched_destroy_group(tg); | |
7843 | } | |
7844 | ||
41a2d6cf IM |
7845 | static int |
7846 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
7847 | struct task_struct *tsk) | |
68318b8e | 7848 | { |
b68aa230 PZ |
7849 | #ifdef CONFIG_RT_GROUP_SCHED |
7850 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7851 | if (rt_task(tsk) && cgroup_tg(cgrp)->rt_runtime == 0) | |
7852 | return -EINVAL; | |
7853 | #else | |
68318b8e SV |
7854 | /* We don't support RT-tasks being in separate groups */ |
7855 | if (tsk->sched_class != &fair_sched_class) | |
7856 | return -EINVAL; | |
b68aa230 | 7857 | #endif |
68318b8e SV |
7858 | |
7859 | return 0; | |
7860 | } | |
7861 | ||
7862 | static void | |
2b01dfe3 | 7863 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
7864 | struct cgroup *old_cont, struct task_struct *tsk) |
7865 | { | |
7866 | sched_move_task(tsk); | |
7867 | } | |
7868 | ||
052f1dc7 | 7869 | #ifdef CONFIG_FAIR_GROUP_SCHED |
2b01dfe3 PM |
7870 | static int cpu_shares_write_uint(struct cgroup *cgrp, struct cftype *cftype, |
7871 | u64 shareval) | |
68318b8e | 7872 | { |
2b01dfe3 | 7873 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
7874 | } |
7875 | ||
2b01dfe3 | 7876 | static u64 cpu_shares_read_uint(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 7877 | { |
2b01dfe3 | 7878 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7879 | |
7880 | return (u64) tg->shares; | |
7881 | } | |
052f1dc7 | 7882 | #endif |
68318b8e | 7883 | |
052f1dc7 | 7884 | #ifdef CONFIG_RT_GROUP_SCHED |
9f0c1e56 PZ |
7885 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
7886 | struct file *file, | |
7887 | const char __user *userbuf, | |
7888 | size_t nbytes, loff_t *unused_ppos) | |
6f505b16 | 7889 | { |
9f0c1e56 PZ |
7890 | char buffer[64]; |
7891 | int retval = 0; | |
7892 | s64 val; | |
7893 | char *end; | |
7894 | ||
7895 | if (!nbytes) | |
7896 | return -EINVAL; | |
7897 | if (nbytes >= sizeof(buffer)) | |
7898 | return -E2BIG; | |
7899 | if (copy_from_user(buffer, userbuf, nbytes)) | |
7900 | return -EFAULT; | |
7901 | ||
7902 | buffer[nbytes] = 0; /* nul-terminate */ | |
7903 | ||
7904 | /* strip newline if necessary */ | |
7905 | if (nbytes && (buffer[nbytes-1] == '\n')) | |
7906 | buffer[nbytes-1] = 0; | |
7907 | val = simple_strtoll(buffer, &end, 0); | |
7908 | if (*end) | |
7909 | return -EINVAL; | |
7910 | ||
7911 | /* Pass to subsystem */ | |
7912 | retval = sched_group_set_rt_runtime(cgroup_tg(cgrp), val); | |
7913 | if (!retval) | |
7914 | retval = nbytes; | |
7915 | return retval; | |
6f505b16 PZ |
7916 | } |
7917 | ||
9f0c1e56 PZ |
7918 | static ssize_t cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft, |
7919 | struct file *file, | |
7920 | char __user *buf, size_t nbytes, | |
7921 | loff_t *ppos) | |
6f505b16 | 7922 | { |
9f0c1e56 PZ |
7923 | char tmp[64]; |
7924 | long val = sched_group_rt_runtime(cgroup_tg(cgrp)); | |
7925 | int len = sprintf(tmp, "%ld\n", val); | |
6f505b16 | 7926 | |
9f0c1e56 | 7927 | return simple_read_from_buffer(buf, nbytes, ppos, tmp, len); |
6f505b16 | 7928 | } |
052f1dc7 | 7929 | #endif |
6f505b16 | 7930 | |
fe5c7cc2 | 7931 | static struct cftype cpu_files[] = { |
052f1dc7 | 7932 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
7933 | { |
7934 | .name = "shares", | |
7935 | .read_uint = cpu_shares_read_uint, | |
7936 | .write_uint = cpu_shares_write_uint, | |
7937 | }, | |
052f1dc7 PZ |
7938 | #endif |
7939 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 7940 | { |
9f0c1e56 PZ |
7941 | .name = "rt_runtime_us", |
7942 | .read = cpu_rt_runtime_read, | |
7943 | .write = cpu_rt_runtime_write, | |
6f505b16 | 7944 | }, |
052f1dc7 | 7945 | #endif |
68318b8e SV |
7946 | }; |
7947 | ||
7948 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
7949 | { | |
fe5c7cc2 | 7950 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
7951 | } |
7952 | ||
7953 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
7954 | .name = "cpu", |
7955 | .create = cpu_cgroup_create, | |
7956 | .destroy = cpu_cgroup_destroy, | |
7957 | .can_attach = cpu_cgroup_can_attach, | |
7958 | .attach = cpu_cgroup_attach, | |
7959 | .populate = cpu_cgroup_populate, | |
7960 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
7961 | .early_init = 1, |
7962 | }; | |
7963 | ||
052f1dc7 | 7964 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
7965 | |
7966 | #ifdef CONFIG_CGROUP_CPUACCT | |
7967 | ||
7968 | /* | |
7969 | * CPU accounting code for task groups. | |
7970 | * | |
7971 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
7972 | * (balbir@in.ibm.com). | |
7973 | */ | |
7974 | ||
7975 | /* track cpu usage of a group of tasks */ | |
7976 | struct cpuacct { | |
7977 | struct cgroup_subsys_state css; | |
7978 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
7979 | u64 *cpuusage; | |
7980 | }; | |
7981 | ||
7982 | struct cgroup_subsys cpuacct_subsys; | |
7983 | ||
7984 | /* return cpu accounting group corresponding to this container */ | |
7985 | static inline struct cpuacct *cgroup_ca(struct cgroup *cont) | |
7986 | { | |
7987 | return container_of(cgroup_subsys_state(cont, cpuacct_subsys_id), | |
7988 | struct cpuacct, css); | |
7989 | } | |
7990 | ||
7991 | /* return cpu accounting group to which this task belongs */ | |
7992 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
7993 | { | |
7994 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
7995 | struct cpuacct, css); | |
7996 | } | |
7997 | ||
7998 | /* create a new cpu accounting group */ | |
7999 | static struct cgroup_subsys_state *cpuacct_create( | |
8000 | struct cgroup_subsys *ss, struct cgroup *cont) | |
8001 | { | |
8002 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
8003 | ||
8004 | if (!ca) | |
8005 | return ERR_PTR(-ENOMEM); | |
8006 | ||
8007 | ca->cpuusage = alloc_percpu(u64); | |
8008 | if (!ca->cpuusage) { | |
8009 | kfree(ca); | |
8010 | return ERR_PTR(-ENOMEM); | |
8011 | } | |
8012 | ||
8013 | return &ca->css; | |
8014 | } | |
8015 | ||
8016 | /* destroy an existing cpu accounting group */ | |
41a2d6cf IM |
8017 | static void |
8018 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cont) | |
d842de87 SV |
8019 | { |
8020 | struct cpuacct *ca = cgroup_ca(cont); | |
8021 | ||
8022 | free_percpu(ca->cpuusage); | |
8023 | kfree(ca); | |
8024 | } | |
8025 | ||
8026 | /* return total cpu usage (in nanoseconds) of a group */ | |
8027 | static u64 cpuusage_read(struct cgroup *cont, struct cftype *cft) | |
8028 | { | |
8029 | struct cpuacct *ca = cgroup_ca(cont); | |
8030 | u64 totalcpuusage = 0; | |
8031 | int i; | |
8032 | ||
8033 | for_each_possible_cpu(i) { | |
8034 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
8035 | ||
8036 | /* | |
8037 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
8038 | * platforms. | |
8039 | */ | |
8040 | spin_lock_irq(&cpu_rq(i)->lock); | |
8041 | totalcpuusage += *cpuusage; | |
8042 | spin_unlock_irq(&cpu_rq(i)->lock); | |
8043 | } | |
8044 | ||
8045 | return totalcpuusage; | |
8046 | } | |
8047 | ||
8048 | static struct cftype files[] = { | |
8049 | { | |
8050 | .name = "usage", | |
8051 | .read_uint = cpuusage_read, | |
8052 | }, | |
8053 | }; | |
8054 | ||
8055 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8056 | { | |
8057 | return cgroup_add_files(cont, ss, files, ARRAY_SIZE(files)); | |
8058 | } | |
8059 | ||
8060 | /* | |
8061 | * charge this task's execution time to its accounting group. | |
8062 | * | |
8063 | * called with rq->lock held. | |
8064 | */ | |
8065 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
8066 | { | |
8067 | struct cpuacct *ca; | |
8068 | ||
8069 | if (!cpuacct_subsys.active) | |
8070 | return; | |
8071 | ||
8072 | ca = task_ca(tsk); | |
8073 | if (ca) { | |
8074 | u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk)); | |
8075 | ||
8076 | *cpuusage += cputime; | |
8077 | } | |
8078 | } | |
8079 | ||
8080 | struct cgroup_subsys cpuacct_subsys = { | |
8081 | .name = "cpuacct", | |
8082 | .create = cpuacct_create, | |
8083 | .destroy = cpuacct_destroy, | |
8084 | .populate = cpuacct_populate, | |
8085 | .subsys_id = cpuacct_subsys_id, | |
8086 | }; | |
8087 | #endif /* CONFIG_CGROUP_CPUACCT */ |