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