]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
97fb7a0a IM |
2 | /* |
3 | * Scheduler internal types and methods: | |
4 | */ | |
029632fb | 5 | #include <linux/sched.h> |
325ea10c | 6 | |
dfc3401a | 7 | #include <linux/sched/autogroup.h> |
e6017571 | 8 | #include <linux/sched/clock.h> |
325ea10c | 9 | #include <linux/sched/coredump.h> |
55687da1 | 10 | #include <linux/sched/cpufreq.h> |
325ea10c IM |
11 | #include <linux/sched/cputime.h> |
12 | #include <linux/sched/deadline.h> | |
b17b0153 | 13 | #include <linux/sched/debug.h> |
ef8bd77f | 14 | #include <linux/sched/hotplug.h> |
325ea10c IM |
15 | #include <linux/sched/idle.h> |
16 | #include <linux/sched/init.h> | |
17 | #include <linux/sched/isolation.h> | |
18 | #include <linux/sched/jobctl.h> | |
19 | #include <linux/sched/loadavg.h> | |
20 | #include <linux/sched/mm.h> | |
21 | #include <linux/sched/nohz.h> | |
22 | #include <linux/sched/numa_balancing.h> | |
23 | #include <linux/sched/prio.h> | |
24 | #include <linux/sched/rt.h> | |
25 | #include <linux/sched/signal.h> | |
321a874a | 26 | #include <linux/sched/smt.h> |
325ea10c IM |
27 | #include <linux/sched/stat.h> |
28 | #include <linux/sched/sysctl.h> | |
29930025 | 29 | #include <linux/sched/task.h> |
68db0cf1 | 30 | #include <linux/sched/task_stack.h> |
325ea10c IM |
31 | #include <linux/sched/topology.h> |
32 | #include <linux/sched/user.h> | |
33 | #include <linux/sched/wake_q.h> | |
34 | #include <linux/sched/xacct.h> | |
35 | ||
36 | #include <uapi/linux/sched/types.h> | |
ef8bd77f | 37 | |
3866e845 | 38 | #include <linux/binfmts.h> |
325ea10c IM |
39 | #include <linux/blkdev.h> |
40 | #include <linux/compat.h> | |
41 | #include <linux/context_tracking.h> | |
42 | #include <linux/cpufreq.h> | |
43 | #include <linux/cpuidle.h> | |
44 | #include <linux/cpuset.h> | |
45 | #include <linux/ctype.h> | |
46 | #include <linux/debugfs.h> | |
47 | #include <linux/delayacct.h> | |
6aa140fa | 48 | #include <linux/energy_model.h> |
325ea10c IM |
49 | #include <linux/init_task.h> |
50 | #include <linux/kprobes.h> | |
51 | #include <linux/kthread.h> | |
52 | #include <linux/membarrier.h> | |
53 | #include <linux/migrate.h> | |
54 | #include <linux/mmu_context.h> | |
55 | #include <linux/nmi.h> | |
56 | #include <linux/proc_fs.h> | |
57 | #include <linux/prefetch.h> | |
58 | #include <linux/profile.h> | |
eb414681 | 59 | #include <linux/psi.h> |
325ea10c IM |
60 | #include <linux/rcupdate_wait.h> |
61 | #include <linux/security.h> | |
029632fb | 62 | #include <linux/stop_machine.h> |
325ea10c IM |
63 | #include <linux/suspend.h> |
64 | #include <linux/swait.h> | |
65 | #include <linux/syscalls.h> | |
66 | #include <linux/task_work.h> | |
67 | #include <linux/tsacct_kern.h> | |
68 | ||
69 | #include <asm/tlb.h> | |
85c2ce91 | 70 | #include <asm-generic/vmlinux.lds.h> |
029632fb | 71 | |
7fce777c | 72 | #ifdef CONFIG_PARAVIRT |
325ea10c | 73 | # include <asm/paravirt.h> |
7fce777c IM |
74 | #endif |
75 | ||
391e43da | 76 | #include "cpupri.h" |
6bfd6d72 | 77 | #include "cpudeadline.h" |
029632fb | 78 | |
9148a3a1 | 79 | #ifdef CONFIG_SCHED_DEBUG |
6d3aed3d | 80 | # define SCHED_WARN_ON(x) WARN_ONCE(x, #x) |
9148a3a1 | 81 | #else |
6d3aed3d | 82 | # define SCHED_WARN_ON(x) ({ (void)(x), 0; }) |
9148a3a1 PZ |
83 | #endif |
84 | ||
45ceebf7 | 85 | struct rq; |
442bf3aa | 86 | struct cpuidle_state; |
45ceebf7 | 87 | |
da0c1e65 KT |
88 | /* task_struct::on_rq states: */ |
89 | #define TASK_ON_RQ_QUEUED 1 | |
cca26e80 | 90 | #define TASK_ON_RQ_MIGRATING 2 |
da0c1e65 | 91 | |
029632fb PZ |
92 | extern __read_mostly int scheduler_running; |
93 | ||
45ceebf7 PG |
94 | extern unsigned long calc_load_update; |
95 | extern atomic_long_t calc_load_tasks; | |
96 | ||
3289bdb4 | 97 | extern void calc_global_load_tick(struct rq *this_rq); |
d60585c5 | 98 | extern long calc_load_fold_active(struct rq *this_rq, long adjust); |
3289bdb4 | 99 | |
029632fb PZ |
100 | /* |
101 | * Helpers for converting nanosecond timing to jiffy resolution | |
102 | */ | |
103 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) | |
104 | ||
cc1f4b1f LZ |
105 | /* |
106 | * Increase resolution of nice-level calculations for 64-bit architectures. | |
107 | * The extra resolution improves shares distribution and load balancing of | |
108 | * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup | |
109 | * hierarchies, especially on larger systems. This is not a user-visible change | |
110 | * and does not change the user-interface for setting shares/weights. | |
111 | * | |
112 | * We increase resolution only if we have enough bits to allow this increased | |
97fb7a0a IM |
113 | * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit |
114 | * are pretty high and the returns do not justify the increased costs. | |
2159197d | 115 | * |
97fb7a0a IM |
116 | * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to |
117 | * increase coverage and consistency always enable it on 64-bit platforms. | |
cc1f4b1f | 118 | */ |
2159197d | 119 | #ifdef CONFIG_64BIT |
172895e6 | 120 | # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT) |
6ecdd749 | 121 | # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT) |
26cf5222 MW |
122 | # define scale_load_down(w) \ |
123 | ({ \ | |
124 | unsigned long __w = (w); \ | |
125 | if (__w) \ | |
126 | __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \ | |
127 | __w; \ | |
128 | }) | |
cc1f4b1f | 129 | #else |
172895e6 | 130 | # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT) |
cc1f4b1f LZ |
131 | # define scale_load(w) (w) |
132 | # define scale_load_down(w) (w) | |
133 | #endif | |
134 | ||
6ecdd749 | 135 | /* |
172895e6 YD |
136 | * Task weight (visible to users) and its load (invisible to users) have |
137 | * independent resolution, but they should be well calibrated. We use | |
138 | * scale_load() and scale_load_down(w) to convert between them. The | |
139 | * following must be true: | |
140 | * | |
141 | * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD | |
142 | * | |
6ecdd749 | 143 | */ |
172895e6 | 144 | #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT) |
029632fb | 145 | |
332ac17e DF |
146 | /* |
147 | * Single value that decides SCHED_DEADLINE internal math precision. | |
148 | * 10 -> just above 1us | |
149 | * 9 -> just above 0.5us | |
150 | */ | |
97fb7a0a | 151 | #define DL_SCALE 10 |
029632fb PZ |
152 | |
153 | /* | |
97fb7a0a | 154 | * Single value that denotes runtime == period, ie unlimited time. |
029632fb | 155 | */ |
97fb7a0a | 156 | #define RUNTIME_INF ((u64)~0ULL) |
029632fb | 157 | |
20f9cd2a HA |
158 | static inline int idle_policy(int policy) |
159 | { | |
160 | return policy == SCHED_IDLE; | |
161 | } | |
d50dde5a DF |
162 | static inline int fair_policy(int policy) |
163 | { | |
164 | return policy == SCHED_NORMAL || policy == SCHED_BATCH; | |
165 | } | |
166 | ||
029632fb PZ |
167 | static inline int rt_policy(int policy) |
168 | { | |
d50dde5a | 169 | return policy == SCHED_FIFO || policy == SCHED_RR; |
029632fb PZ |
170 | } |
171 | ||
aab03e05 DF |
172 | static inline int dl_policy(int policy) |
173 | { | |
174 | return policy == SCHED_DEADLINE; | |
175 | } | |
20f9cd2a HA |
176 | static inline bool valid_policy(int policy) |
177 | { | |
178 | return idle_policy(policy) || fair_policy(policy) || | |
179 | rt_policy(policy) || dl_policy(policy); | |
180 | } | |
aab03e05 | 181 | |
1da1843f VK |
182 | static inline int task_has_idle_policy(struct task_struct *p) |
183 | { | |
184 | return idle_policy(p->policy); | |
185 | } | |
186 | ||
029632fb PZ |
187 | static inline int task_has_rt_policy(struct task_struct *p) |
188 | { | |
189 | return rt_policy(p->policy); | |
190 | } | |
191 | ||
aab03e05 DF |
192 | static inline int task_has_dl_policy(struct task_struct *p) |
193 | { | |
194 | return dl_policy(p->policy); | |
195 | } | |
196 | ||
07881166 JL |
197 | #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT) |
198 | ||
d76343c6 VS |
199 | static inline void update_avg(u64 *avg, u64 sample) |
200 | { | |
201 | s64 diff = sample - *avg; | |
202 | *avg += diff / 8; | |
203 | } | |
204 | ||
794a56eb JL |
205 | /* |
206 | * !! For sched_setattr_nocheck() (kernel) only !! | |
207 | * | |
208 | * This is actually gross. :( | |
209 | * | |
210 | * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE | |
211 | * tasks, but still be able to sleep. We need this on platforms that cannot | |
212 | * atomically change clock frequency. Remove once fast switching will be | |
213 | * available on such platforms. | |
214 | * | |
215 | * SUGOV stands for SchedUtil GOVernor. | |
216 | */ | |
217 | #define SCHED_FLAG_SUGOV 0x10000000 | |
218 | ||
219 | static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se) | |
220 | { | |
221 | #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL | |
222 | return unlikely(dl_se->flags & SCHED_FLAG_SUGOV); | |
223 | #else | |
224 | return false; | |
225 | #endif | |
226 | } | |
227 | ||
2d3d891d DF |
228 | /* |
229 | * Tells if entity @a should preempt entity @b. | |
230 | */ | |
332ac17e DF |
231 | static inline bool |
232 | dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b) | |
2d3d891d | 233 | { |
794a56eb JL |
234 | return dl_entity_is_special(a) || |
235 | dl_time_before(a->deadline, b->deadline); | |
2d3d891d DF |
236 | } |
237 | ||
029632fb PZ |
238 | /* |
239 | * This is the priority-queue data structure of the RT scheduling class: | |
240 | */ | |
241 | struct rt_prio_array { | |
242 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
243 | struct list_head queue[MAX_RT_PRIO]; | |
244 | }; | |
245 | ||
246 | struct rt_bandwidth { | |
247 | /* nests inside the rq lock: */ | |
248 | raw_spinlock_t rt_runtime_lock; | |
249 | ktime_t rt_period; | |
250 | u64 rt_runtime; | |
251 | struct hrtimer rt_period_timer; | |
4cfafd30 | 252 | unsigned int rt_period_active; |
029632fb | 253 | }; |
a5e7be3b JL |
254 | |
255 | void __dl_clear_params(struct task_struct *p); | |
256 | ||
332ac17e DF |
257 | /* |
258 | * To keep the bandwidth of -deadline tasks and groups under control | |
259 | * we need some place where: | |
260 | * - store the maximum -deadline bandwidth of the system (the group); | |
261 | * - cache the fraction of that bandwidth that is currently allocated. | |
262 | * | |
263 | * This is all done in the data structure below. It is similar to the | |
264 | * one used for RT-throttling (rt_bandwidth), with the main difference | |
265 | * that, since here we are only interested in admission control, we | |
266 | * do not decrease any runtime while the group "executes", neither we | |
267 | * need a timer to replenish it. | |
268 | * | |
269 | * With respect to SMP, the bandwidth is given on a per-CPU basis, | |
270 | * meaning that: | |
271 | * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU; | |
272 | * - dl_total_bw array contains, in the i-eth element, the currently | |
273 | * allocated bandwidth on the i-eth CPU. | |
274 | * Moreover, groups consume bandwidth on each CPU, while tasks only | |
275 | * consume bandwidth on the CPU they're running on. | |
276 | * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw | |
277 | * that will be shown the next time the proc or cgroup controls will | |
278 | * be red. It on its turn can be changed by writing on its own | |
279 | * control. | |
280 | */ | |
281 | struct dl_bandwidth { | |
97fb7a0a IM |
282 | raw_spinlock_t dl_runtime_lock; |
283 | u64 dl_runtime; | |
284 | u64 dl_period; | |
332ac17e DF |
285 | }; |
286 | ||
287 | static inline int dl_bandwidth_enabled(void) | |
288 | { | |
1724813d | 289 | return sysctl_sched_rt_runtime >= 0; |
332ac17e DF |
290 | } |
291 | ||
332ac17e | 292 | struct dl_bw { |
97fb7a0a IM |
293 | raw_spinlock_t lock; |
294 | u64 bw; | |
295 | u64 total_bw; | |
332ac17e DF |
296 | }; |
297 | ||
daec5798 LA |
298 | static inline void __dl_update(struct dl_bw *dl_b, s64 bw); |
299 | ||
7f51412a | 300 | static inline |
8c0944ce | 301 | void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus) |
7f51412a JL |
302 | { |
303 | dl_b->total_bw -= tsk_bw; | |
daec5798 | 304 | __dl_update(dl_b, (s32)tsk_bw / cpus); |
7f51412a JL |
305 | } |
306 | ||
307 | static inline | |
daec5798 | 308 | void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus) |
7f51412a JL |
309 | { |
310 | dl_b->total_bw += tsk_bw; | |
daec5798 | 311 | __dl_update(dl_b, -((s32)tsk_bw / cpus)); |
7f51412a JL |
312 | } |
313 | ||
60ffd5ed LA |
314 | static inline bool __dl_overflow(struct dl_bw *dl_b, unsigned long cap, |
315 | u64 old_bw, u64 new_bw) | |
7f51412a JL |
316 | { |
317 | return dl_b->bw != -1 && | |
60ffd5ed | 318 | cap_scale(dl_b->bw, cap) < dl_b->total_bw - old_bw + new_bw; |
7f51412a JL |
319 | } |
320 | ||
b4118988 LA |
321 | /* |
322 | * Verify the fitness of task @p to run on @cpu taking into account the | |
323 | * CPU original capacity and the runtime/deadline ratio of the task. | |
324 | * | |
325 | * The function will return true if the CPU original capacity of the | |
326 | * @cpu scaled by SCHED_CAPACITY_SCALE >= runtime/deadline ratio of the | |
327 | * task and false otherwise. | |
328 | */ | |
329 | static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu) | |
330 | { | |
331 | unsigned long cap = arch_scale_cpu_capacity(cpu); | |
332 | ||
333 | return cap_scale(p->dl.dl_deadline, cap) >= p->dl.dl_runtime; | |
334 | } | |
335 | ||
f2cb1360 | 336 | extern void init_dl_bw(struct dl_bw *dl_b); |
97fb7a0a | 337 | extern int sched_dl_global_validate(void); |
06a76fe0 | 338 | extern void sched_dl_do_global(void); |
97fb7a0a | 339 | extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr); |
06a76fe0 NP |
340 | extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr); |
341 | extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr); | |
342 | extern bool __checkparam_dl(const struct sched_attr *attr); | |
06a76fe0 | 343 | extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr); |
97fb7a0a IM |
344 | extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed); |
345 | extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial); | |
06a76fe0 | 346 | extern bool dl_cpu_busy(unsigned int cpu); |
029632fb PZ |
347 | |
348 | #ifdef CONFIG_CGROUP_SCHED | |
349 | ||
350 | #include <linux/cgroup.h> | |
eb414681 | 351 | #include <linux/psi.h> |
029632fb PZ |
352 | |
353 | struct cfs_rq; | |
354 | struct rt_rq; | |
355 | ||
35cf4e50 | 356 | extern struct list_head task_groups; |
029632fb PZ |
357 | |
358 | struct cfs_bandwidth { | |
359 | #ifdef CONFIG_CFS_BANDWIDTH | |
97fb7a0a IM |
360 | raw_spinlock_t lock; |
361 | ktime_t period; | |
362 | u64 quota; | |
363 | u64 runtime; | |
364 | s64 hierarchical_quota; | |
97fb7a0a | 365 | |
66567fcb | 366 | u8 idle; |
367 | u8 period_active; | |
66567fcb | 368 | u8 slack_started; |
97fb7a0a IM |
369 | struct hrtimer period_timer; |
370 | struct hrtimer slack_timer; | |
371 | struct list_head throttled_cfs_rq; | |
372 | ||
373 | /* Statistics: */ | |
374 | int nr_periods; | |
375 | int nr_throttled; | |
376 | u64 throttled_time; | |
029632fb PZ |
377 | #endif |
378 | }; | |
379 | ||
97fb7a0a | 380 | /* Task group related information */ |
029632fb PZ |
381 | struct task_group { |
382 | struct cgroup_subsys_state css; | |
383 | ||
384 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
97fb7a0a IM |
385 | /* schedulable entities of this group on each CPU */ |
386 | struct sched_entity **se; | |
387 | /* runqueue "owned" by this group on each CPU */ | |
388 | struct cfs_rq **cfs_rq; | |
389 | unsigned long shares; | |
029632fb | 390 | |
fa6bddeb | 391 | #ifdef CONFIG_SMP |
b0367629 WL |
392 | /* |
393 | * load_avg can be heavily contended at clock tick time, so put | |
394 | * it in its own cacheline separated from the fields above which | |
395 | * will also be accessed at each tick. | |
396 | */ | |
97fb7a0a | 397 | atomic_long_t load_avg ____cacheline_aligned; |
029632fb | 398 | #endif |
fa6bddeb | 399 | #endif |
029632fb PZ |
400 | |
401 | #ifdef CONFIG_RT_GROUP_SCHED | |
97fb7a0a IM |
402 | struct sched_rt_entity **rt_se; |
403 | struct rt_rq **rt_rq; | |
029632fb | 404 | |
97fb7a0a | 405 | struct rt_bandwidth rt_bandwidth; |
029632fb PZ |
406 | #endif |
407 | ||
97fb7a0a IM |
408 | struct rcu_head rcu; |
409 | struct list_head list; | |
029632fb | 410 | |
97fb7a0a IM |
411 | struct task_group *parent; |
412 | struct list_head siblings; | |
413 | struct list_head children; | |
029632fb PZ |
414 | |
415 | #ifdef CONFIG_SCHED_AUTOGROUP | |
97fb7a0a | 416 | struct autogroup *autogroup; |
029632fb PZ |
417 | #endif |
418 | ||
97fb7a0a | 419 | struct cfs_bandwidth cfs_bandwidth; |
2480c093 PB |
420 | |
421 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
422 | /* The two decimal precision [%] value requested from user-space */ | |
423 | unsigned int uclamp_pct[UCLAMP_CNT]; | |
424 | /* Clamp values requested for a task group */ | |
425 | struct uclamp_se uclamp_req[UCLAMP_CNT]; | |
0b60ba2d PB |
426 | /* Effective clamp values used for a task group */ |
427 | struct uclamp_se uclamp[UCLAMP_CNT]; | |
2480c093 PB |
428 | #endif |
429 | ||
029632fb PZ |
430 | }; |
431 | ||
432 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
433 | #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD | |
434 | ||
435 | /* | |
436 | * A weight of 0 or 1 can cause arithmetics problems. | |
437 | * A weight of a cfs_rq is the sum of weights of which entities | |
438 | * are queued on this cfs_rq, so a weight of a entity should not be | |
439 | * too large, so as the shares value of a task group. | |
440 | * (The default weight is 1024 - so there's no practical | |
441 | * limitation from this.) | |
442 | */ | |
97fb7a0a IM |
443 | #define MIN_SHARES (1UL << 1) |
444 | #define MAX_SHARES (1UL << 18) | |
029632fb PZ |
445 | #endif |
446 | ||
029632fb PZ |
447 | typedef int (*tg_visitor)(struct task_group *, void *); |
448 | ||
449 | extern int walk_tg_tree_from(struct task_group *from, | |
450 | tg_visitor down, tg_visitor up, void *data); | |
451 | ||
452 | /* | |
453 | * Iterate the full tree, calling @down when first entering a node and @up when | |
454 | * leaving it for the final time. | |
455 | * | |
456 | * Caller must hold rcu_lock or sufficient equivalent. | |
457 | */ | |
458 | static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) | |
459 | { | |
460 | return walk_tg_tree_from(&root_task_group, down, up, data); | |
461 | } | |
462 | ||
463 | extern int tg_nop(struct task_group *tg, void *data); | |
464 | ||
465 | extern void free_fair_sched_group(struct task_group *tg); | |
466 | extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent); | |
8663e24d | 467 | extern void online_fair_sched_group(struct task_group *tg); |
6fe1f348 | 468 | extern void unregister_fair_sched_group(struct task_group *tg); |
029632fb PZ |
469 | extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
470 | struct sched_entity *se, int cpu, | |
471 | struct sched_entity *parent); | |
472 | extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b); | |
029632fb PZ |
473 | |
474 | extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b); | |
77a4d1a1 | 475 | extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b); |
029632fb PZ |
476 | extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq); |
477 | ||
478 | extern void free_rt_sched_group(struct task_group *tg); | |
479 | extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent); | |
480 | extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, | |
481 | struct sched_rt_entity *rt_se, int cpu, | |
482 | struct sched_rt_entity *parent); | |
8887cd99 NP |
483 | extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us); |
484 | extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us); | |
485 | extern long sched_group_rt_runtime(struct task_group *tg); | |
486 | extern long sched_group_rt_period(struct task_group *tg); | |
487 | extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk); | |
029632fb | 488 | |
25cc7da7 LZ |
489 | extern struct task_group *sched_create_group(struct task_group *parent); |
490 | extern void sched_online_group(struct task_group *tg, | |
491 | struct task_group *parent); | |
492 | extern void sched_destroy_group(struct task_group *tg); | |
493 | extern void sched_offline_group(struct task_group *tg); | |
494 | ||
495 | extern void sched_move_task(struct task_struct *tsk); | |
496 | ||
497 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
498 | extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); | |
ad936d86 BP |
499 | |
500 | #ifdef CONFIG_SMP | |
501 | extern void set_task_rq_fair(struct sched_entity *se, | |
502 | struct cfs_rq *prev, struct cfs_rq *next); | |
503 | #else /* !CONFIG_SMP */ | |
504 | static inline void set_task_rq_fair(struct sched_entity *se, | |
505 | struct cfs_rq *prev, struct cfs_rq *next) { } | |
506 | #endif /* CONFIG_SMP */ | |
507 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
25cc7da7 | 508 | |
029632fb PZ |
509 | #else /* CONFIG_CGROUP_SCHED */ |
510 | ||
511 | struct cfs_bandwidth { }; | |
512 | ||
513 | #endif /* CONFIG_CGROUP_SCHED */ | |
514 | ||
515 | /* CFS-related fields in a runqueue */ | |
516 | struct cfs_rq { | |
97fb7a0a | 517 | struct load_weight load; |
97fb7a0a | 518 | unsigned int nr_running; |
43e9f7f2 VK |
519 | unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */ |
520 | unsigned int idle_h_nr_running; /* SCHED_IDLE */ | |
029632fb | 521 | |
97fb7a0a IM |
522 | u64 exec_clock; |
523 | u64 min_vruntime; | |
029632fb | 524 | #ifndef CONFIG_64BIT |
97fb7a0a | 525 | u64 min_vruntime_copy; |
029632fb PZ |
526 | #endif |
527 | ||
97fb7a0a | 528 | struct rb_root_cached tasks_timeline; |
029632fb | 529 | |
029632fb PZ |
530 | /* |
531 | * 'curr' points to currently running entity on this cfs_rq. | |
532 | * It is set to NULL otherwise (i.e when none are currently running). | |
533 | */ | |
97fb7a0a IM |
534 | struct sched_entity *curr; |
535 | struct sched_entity *next; | |
536 | struct sched_entity *last; | |
537 | struct sched_entity *skip; | |
029632fb PZ |
538 | |
539 | #ifdef CONFIG_SCHED_DEBUG | |
97fb7a0a | 540 | unsigned int nr_spread_over; |
029632fb PZ |
541 | #endif |
542 | ||
2dac754e PT |
543 | #ifdef CONFIG_SMP |
544 | /* | |
9d89c257 | 545 | * CFS load tracking |
2dac754e | 546 | */ |
97fb7a0a | 547 | struct sched_avg avg; |
2a2f5d4e | 548 | #ifndef CONFIG_64BIT |
97fb7a0a | 549 | u64 load_last_update_time_copy; |
9d89c257 | 550 | #endif |
2a2f5d4e PZ |
551 | struct { |
552 | raw_spinlock_t lock ____cacheline_aligned; | |
553 | int nr; | |
554 | unsigned long load_avg; | |
555 | unsigned long util_avg; | |
9f683953 | 556 | unsigned long runnable_avg; |
2a2f5d4e | 557 | } removed; |
82958366 | 558 | |
9d89c257 | 559 | #ifdef CONFIG_FAIR_GROUP_SCHED |
97fb7a0a IM |
560 | unsigned long tg_load_avg_contrib; |
561 | long propagate; | |
562 | long prop_runnable_sum; | |
0e2d2aaa | 563 | |
82958366 PT |
564 | /* |
565 | * h_load = weight * f(tg) | |
566 | * | |
567 | * Where f(tg) is the recursive weight fraction assigned to | |
568 | * this group. | |
569 | */ | |
97fb7a0a IM |
570 | unsigned long h_load; |
571 | u64 last_h_load_update; | |
572 | struct sched_entity *h_load_next; | |
68520796 | 573 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
82958366 PT |
574 | #endif /* CONFIG_SMP */ |
575 | ||
029632fb | 576 | #ifdef CONFIG_FAIR_GROUP_SCHED |
97fb7a0a | 577 | struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */ |
029632fb PZ |
578 | |
579 | /* | |
580 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
581 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities | |
582 | * (like users, containers etc.) | |
583 | * | |
97fb7a0a IM |
584 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU. |
585 | * This list is used during load balance. | |
029632fb | 586 | */ |
97fb7a0a IM |
587 | int on_list; |
588 | struct list_head leaf_cfs_rq_list; | |
589 | struct task_group *tg; /* group that "owns" this runqueue */ | |
029632fb | 590 | |
029632fb | 591 | #ifdef CONFIG_CFS_BANDWIDTH |
97fb7a0a | 592 | int runtime_enabled; |
97fb7a0a IM |
593 | s64 runtime_remaining; |
594 | ||
595 | u64 throttled_clock; | |
596 | u64 throttled_clock_task; | |
597 | u64 throttled_clock_task_time; | |
598 | int throttled; | |
599 | int throttle_count; | |
600 | struct list_head throttled_list; | |
029632fb PZ |
601 | #endif /* CONFIG_CFS_BANDWIDTH */ |
602 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
603 | }; | |
604 | ||
605 | static inline int rt_bandwidth_enabled(void) | |
606 | { | |
607 | return sysctl_sched_rt_runtime >= 0; | |
608 | } | |
609 | ||
b6366f04 | 610 | /* RT IPI pull logic requires IRQ_WORK */ |
4bdced5c | 611 | #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP) |
b6366f04 SR |
612 | # define HAVE_RT_PUSH_IPI |
613 | #endif | |
614 | ||
029632fb PZ |
615 | /* Real-Time classes' related field in a runqueue: */ |
616 | struct rt_rq { | |
97fb7a0a IM |
617 | struct rt_prio_array active; |
618 | unsigned int rt_nr_running; | |
619 | unsigned int rr_nr_running; | |
029632fb PZ |
620 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
621 | struct { | |
97fb7a0a | 622 | int curr; /* highest queued rt task prio */ |
029632fb | 623 | #ifdef CONFIG_SMP |
97fb7a0a | 624 | int next; /* next highest */ |
029632fb PZ |
625 | #endif |
626 | } highest_prio; | |
627 | #endif | |
628 | #ifdef CONFIG_SMP | |
97fb7a0a IM |
629 | unsigned long rt_nr_migratory; |
630 | unsigned long rt_nr_total; | |
631 | int overloaded; | |
632 | struct plist_head pushable_tasks; | |
371bf427 | 633 | |
b6366f04 | 634 | #endif /* CONFIG_SMP */ |
97fb7a0a | 635 | int rt_queued; |
f4ebcbc0 | 636 | |
97fb7a0a IM |
637 | int rt_throttled; |
638 | u64 rt_time; | |
639 | u64 rt_runtime; | |
029632fb | 640 | /* Nests inside the rq lock: */ |
97fb7a0a | 641 | raw_spinlock_t rt_runtime_lock; |
029632fb PZ |
642 | |
643 | #ifdef CONFIG_RT_GROUP_SCHED | |
97fb7a0a | 644 | unsigned long rt_nr_boosted; |
029632fb | 645 | |
97fb7a0a IM |
646 | struct rq *rq; |
647 | struct task_group *tg; | |
029632fb PZ |
648 | #endif |
649 | }; | |
650 | ||
296b2ffe VG |
651 | static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq) |
652 | { | |
653 | return rt_rq->rt_queued && rt_rq->rt_nr_running; | |
654 | } | |
655 | ||
aab03e05 DF |
656 | /* Deadline class' related fields in a runqueue */ |
657 | struct dl_rq { | |
658 | /* runqueue is an rbtree, ordered by deadline */ | |
97fb7a0a | 659 | struct rb_root_cached root; |
aab03e05 | 660 | |
97fb7a0a | 661 | unsigned long dl_nr_running; |
1baca4ce JL |
662 | |
663 | #ifdef CONFIG_SMP | |
664 | /* | |
665 | * Deadline values of the currently executing and the | |
666 | * earliest ready task on this rq. Caching these facilitates | |
dfcb245e | 667 | * the decision whether or not a ready but not running task |
1baca4ce JL |
668 | * should migrate somewhere else. |
669 | */ | |
670 | struct { | |
97fb7a0a IM |
671 | u64 curr; |
672 | u64 next; | |
1baca4ce JL |
673 | } earliest_dl; |
674 | ||
97fb7a0a IM |
675 | unsigned long dl_nr_migratory; |
676 | int overloaded; | |
1baca4ce JL |
677 | |
678 | /* | |
679 | * Tasks on this rq that can be pushed away. They are kept in | |
680 | * an rb-tree, ordered by tasks' deadlines, with caching | |
681 | * of the leftmost (earliest deadline) element. | |
682 | */ | |
97fb7a0a | 683 | struct rb_root_cached pushable_dl_tasks_root; |
332ac17e | 684 | #else |
97fb7a0a | 685 | struct dl_bw dl_bw; |
1baca4ce | 686 | #endif |
e36d8677 LA |
687 | /* |
688 | * "Active utilization" for this runqueue: increased when a | |
689 | * task wakes up (becomes TASK_RUNNING) and decreased when a | |
690 | * task blocks | |
691 | */ | |
97fb7a0a | 692 | u64 running_bw; |
4da3abce | 693 | |
8fd27231 LA |
694 | /* |
695 | * Utilization of the tasks "assigned" to this runqueue (including | |
696 | * the tasks that are in runqueue and the tasks that executed on this | |
697 | * CPU and blocked). Increased when a task moves to this runqueue, and | |
698 | * decreased when the task moves away (migrates, changes scheduling | |
699 | * policy, or terminates). | |
700 | * This is needed to compute the "inactive utilization" for the | |
701 | * runqueue (inactive utilization = this_bw - running_bw). | |
702 | */ | |
97fb7a0a IM |
703 | u64 this_bw; |
704 | u64 extra_bw; | |
8fd27231 | 705 | |
4da3abce LA |
706 | /* |
707 | * Inverse of the fraction of CPU utilization that can be reclaimed | |
708 | * by the GRUB algorithm. | |
709 | */ | |
97fb7a0a | 710 | u64 bw_ratio; |
aab03e05 DF |
711 | }; |
712 | ||
c0796298 VG |
713 | #ifdef CONFIG_FAIR_GROUP_SCHED |
714 | /* An entity is a task if it doesn't "own" a runqueue */ | |
715 | #define entity_is_task(se) (!se->my_q) | |
0dacee1b | 716 | |
9f683953 VG |
717 | static inline void se_update_runnable(struct sched_entity *se) |
718 | { | |
719 | if (!entity_is_task(se)) | |
720 | se->runnable_weight = se->my_q->h_nr_running; | |
721 | } | |
722 | ||
723 | static inline long se_runnable(struct sched_entity *se) | |
724 | { | |
725 | if (entity_is_task(se)) | |
726 | return !!se->on_rq; | |
727 | else | |
728 | return se->runnable_weight; | |
729 | } | |
730 | ||
c0796298 VG |
731 | #else |
732 | #define entity_is_task(se) 1 | |
0dacee1b | 733 | |
9f683953 VG |
734 | static inline void se_update_runnable(struct sched_entity *se) {} |
735 | ||
736 | static inline long se_runnable(struct sched_entity *se) | |
737 | { | |
738 | return !!se->on_rq; | |
739 | } | |
c0796298 VG |
740 | #endif |
741 | ||
029632fb | 742 | #ifdef CONFIG_SMP |
c0796298 VG |
743 | /* |
744 | * XXX we want to get rid of these helpers and use the full load resolution. | |
745 | */ | |
746 | static inline long se_weight(struct sched_entity *se) | |
747 | { | |
748 | return scale_load_down(se->load.weight); | |
749 | } | |
750 | ||
029632fb | 751 | |
afe06efd TC |
752 | static inline bool sched_asym_prefer(int a, int b) |
753 | { | |
754 | return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b); | |
755 | } | |
756 | ||
6aa140fa QP |
757 | struct perf_domain { |
758 | struct em_perf_domain *em_pd; | |
759 | struct perf_domain *next; | |
760 | struct rcu_head rcu; | |
761 | }; | |
762 | ||
630246a0 QP |
763 | /* Scheduling group status flags */ |
764 | #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */ | |
2802bf3c | 765 | #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */ |
630246a0 | 766 | |
029632fb PZ |
767 | /* |
768 | * We add the notion of a root-domain which will be used to define per-domain | |
769 | * variables. Each exclusive cpuset essentially defines an island domain by | |
97fb7a0a | 770 | * fully partitioning the member CPUs from any other cpuset. Whenever a new |
029632fb PZ |
771 | * exclusive cpuset is created, we also create and attach a new root-domain |
772 | * object. | |
773 | * | |
774 | */ | |
775 | struct root_domain { | |
97fb7a0a IM |
776 | atomic_t refcount; |
777 | atomic_t rto_count; | |
778 | struct rcu_head rcu; | |
779 | cpumask_var_t span; | |
780 | cpumask_var_t online; | |
029632fb | 781 | |
757ffdd7 VS |
782 | /* |
783 | * Indicate pullable load on at least one CPU, e.g: | |
784 | * - More than one runnable task | |
785 | * - Running task is misfit | |
786 | */ | |
575638d1 | 787 | int overload; |
4486edd1 | 788 | |
2802bf3c MR |
789 | /* Indicate one or more cpus over-utilized (tipping point) */ |
790 | int overutilized; | |
791 | ||
1baca4ce JL |
792 | /* |
793 | * The bit corresponding to a CPU gets set here if such CPU has more | |
794 | * than one runnable -deadline task (as it is below for RT tasks). | |
795 | */ | |
97fb7a0a IM |
796 | cpumask_var_t dlo_mask; |
797 | atomic_t dlo_count; | |
798 | struct dl_bw dl_bw; | |
799 | struct cpudl cpudl; | |
1baca4ce | 800 | |
4bdced5c SRRH |
801 | #ifdef HAVE_RT_PUSH_IPI |
802 | /* | |
803 | * For IPI pull requests, loop across the rto_mask. | |
804 | */ | |
97fb7a0a IM |
805 | struct irq_work rto_push_work; |
806 | raw_spinlock_t rto_lock; | |
4bdced5c | 807 | /* These are only updated and read within rto_lock */ |
97fb7a0a IM |
808 | int rto_loop; |
809 | int rto_cpu; | |
4bdced5c | 810 | /* These atomics are updated outside of a lock */ |
97fb7a0a IM |
811 | atomic_t rto_loop_next; |
812 | atomic_t rto_loop_start; | |
4bdced5c | 813 | #endif |
029632fb PZ |
814 | /* |
815 | * The "RT overload" flag: it gets set if a CPU has more than | |
816 | * one runnable RT task. | |
817 | */ | |
97fb7a0a IM |
818 | cpumask_var_t rto_mask; |
819 | struct cpupri cpupri; | |
cd92bfd3 | 820 | |
97fb7a0a | 821 | unsigned long max_cpu_capacity; |
6aa140fa QP |
822 | |
823 | /* | |
824 | * NULL-terminated list of performance domains intersecting with the | |
825 | * CPUs of the rd. Protected by RCU. | |
826 | */ | |
7ba7319f | 827 | struct perf_domain __rcu *pd; |
029632fb PZ |
828 | }; |
829 | ||
f2cb1360 | 830 | extern void init_defrootdomain(void); |
8d5dc512 | 831 | extern int sched_init_domains(const struct cpumask *cpu_map); |
f2cb1360 | 832 | extern void rq_attach_root(struct rq *rq, struct root_domain *rd); |
364f5665 SRV |
833 | extern void sched_get_rd(struct root_domain *rd); |
834 | extern void sched_put_rd(struct root_domain *rd); | |
029632fb | 835 | |
4bdced5c SRRH |
836 | #ifdef HAVE_RT_PUSH_IPI |
837 | extern void rto_push_irq_work_func(struct irq_work *work); | |
838 | #endif | |
029632fb PZ |
839 | #endif /* CONFIG_SMP */ |
840 | ||
69842cba PB |
841 | #ifdef CONFIG_UCLAMP_TASK |
842 | /* | |
843 | * struct uclamp_bucket - Utilization clamp bucket | |
844 | * @value: utilization clamp value for tasks on this clamp bucket | |
845 | * @tasks: number of RUNNABLE tasks on this clamp bucket | |
846 | * | |
847 | * Keep track of how many tasks are RUNNABLE for a given utilization | |
848 | * clamp value. | |
849 | */ | |
850 | struct uclamp_bucket { | |
851 | unsigned long value : bits_per(SCHED_CAPACITY_SCALE); | |
852 | unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE); | |
853 | }; | |
854 | ||
855 | /* | |
856 | * struct uclamp_rq - rq's utilization clamp | |
857 | * @value: currently active clamp values for a rq | |
858 | * @bucket: utilization clamp buckets affecting a rq | |
859 | * | |
860 | * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values. | |
861 | * A clamp value is affecting a rq when there is at least one task RUNNABLE | |
862 | * (or actually running) with that value. | |
863 | * | |
864 | * There are up to UCLAMP_CNT possible different clamp values, currently there | |
865 | * are only two: minimum utilization and maximum utilization. | |
866 | * | |
867 | * All utilization clamping values are MAX aggregated, since: | |
868 | * - for util_min: we want to run the CPU at least at the max of the minimum | |
869 | * utilization required by its currently RUNNABLE tasks. | |
870 | * - for util_max: we want to allow the CPU to run up to the max of the | |
871 | * maximum utilization allowed by its currently RUNNABLE tasks. | |
872 | * | |
873 | * Since on each system we expect only a limited number of different | |
874 | * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track | |
875 | * the metrics required to compute all the per-rq utilization clamp values. | |
876 | */ | |
877 | struct uclamp_rq { | |
878 | unsigned int value; | |
879 | struct uclamp_bucket bucket[UCLAMP_BUCKETS]; | |
880 | }; | |
881 | #endif /* CONFIG_UCLAMP_TASK */ | |
882 | ||
029632fb PZ |
883 | /* |
884 | * This is the main, per-CPU runqueue data structure. | |
885 | * | |
886 | * Locking rule: those places that want to lock multiple runqueues | |
887 | * (such as the load balancing or the thread migration code), lock | |
888 | * acquire operations must be ordered by ascending &runqueue. | |
889 | */ | |
890 | struct rq { | |
891 | /* runqueue lock: */ | |
97fb7a0a | 892 | raw_spinlock_t lock; |
029632fb PZ |
893 | |
894 | /* | |
895 | * nr_running and cpu_load should be in the same cacheline because | |
896 | * remote CPUs use both these fields when doing load calculation. | |
897 | */ | |
97fb7a0a | 898 | unsigned int nr_running; |
0ec8aa00 | 899 | #ifdef CONFIG_NUMA_BALANCING |
97fb7a0a IM |
900 | unsigned int nr_numa_running; |
901 | unsigned int nr_preferred_running; | |
a4739eca | 902 | unsigned int numa_migrate_on; |
0ec8aa00 | 903 | #endif |
3451d024 | 904 | #ifdef CONFIG_NO_HZ_COMMON |
9fd81dd5 | 905 | #ifdef CONFIG_SMP |
e022e0d3 | 906 | unsigned long last_blocked_load_update_tick; |
f643ea22 | 907 | unsigned int has_blocked_load; |
90b5363a | 908 | call_single_data_t nohz_csd; |
9fd81dd5 | 909 | #endif /* CONFIG_SMP */ |
00357f5e | 910 | unsigned int nohz_tick_stopped; |
90b5363a | 911 | atomic_t nohz_flags; |
9fd81dd5 | 912 | #endif /* CONFIG_NO_HZ_COMMON */ |
dcdedb24 | 913 | |
126c2092 PZ |
914 | #ifdef CONFIG_SMP |
915 | unsigned int ttwu_pending; | |
916 | #endif | |
97fb7a0a | 917 | u64 nr_switches; |
029632fb | 918 | |
69842cba PB |
919 | #ifdef CONFIG_UCLAMP_TASK |
920 | /* Utilization clamp values based on CPU's RUNNABLE tasks */ | |
921 | struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned; | |
e496187d PB |
922 | unsigned int uclamp_flags; |
923 | #define UCLAMP_FLAG_IDLE 0x01 | |
69842cba PB |
924 | #endif |
925 | ||
97fb7a0a IM |
926 | struct cfs_rq cfs; |
927 | struct rt_rq rt; | |
928 | struct dl_rq dl; | |
029632fb PZ |
929 | |
930 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
97fb7a0a IM |
931 | /* list of leaf cfs_rq on this CPU: */ |
932 | struct list_head leaf_cfs_rq_list; | |
933 | struct list_head *tmp_alone_branch; | |
a35b6466 PZ |
934 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
935 | ||
029632fb PZ |
936 | /* |
937 | * This is part of a global counter where only the total sum | |
938 | * over all CPUs matters. A task can increase this counter on | |
939 | * one CPU and if it got migrated afterwards it may decrease | |
940 | * it on another CPU. Always updated under the runqueue lock: | |
941 | */ | |
97fb7a0a | 942 | unsigned long nr_uninterruptible; |
029632fb | 943 | |
4104a562 | 944 | struct task_struct __rcu *curr; |
97fb7a0a IM |
945 | struct task_struct *idle; |
946 | struct task_struct *stop; | |
947 | unsigned long next_balance; | |
948 | struct mm_struct *prev_mm; | |
029632fb | 949 | |
97fb7a0a IM |
950 | unsigned int clock_update_flags; |
951 | u64 clock; | |
23127296 VG |
952 | /* Ensure that all clocks are in the same cache line */ |
953 | u64 clock_task ____cacheline_aligned; | |
954 | u64 clock_pelt; | |
955 | unsigned long lost_idle_time; | |
029632fb | 956 | |
97fb7a0a | 957 | atomic_t nr_iowait; |
029632fb | 958 | |
227a4aad MD |
959 | #ifdef CONFIG_MEMBARRIER |
960 | int membarrier_state; | |
961 | #endif | |
962 | ||
029632fb | 963 | #ifdef CONFIG_SMP |
994aeb7a JFG |
964 | struct root_domain *rd; |
965 | struct sched_domain __rcu *sd; | |
97fb7a0a IM |
966 | |
967 | unsigned long cpu_capacity; | |
968 | unsigned long cpu_capacity_orig; | |
029632fb | 969 | |
97fb7a0a | 970 | struct callback_head *balance_callback; |
029632fb | 971 | |
19a1f5ec | 972 | unsigned char nohz_idle_balance; |
97fb7a0a | 973 | unsigned char idle_balance; |
e3fca9e7 | 974 | |
3b1baa64 MR |
975 | unsigned long misfit_task_load; |
976 | ||
029632fb | 977 | /* For active balancing */ |
97fb7a0a IM |
978 | int active_balance; |
979 | int push_cpu; | |
980 | struct cpu_stop_work active_balance_work; | |
981 | ||
982 | /* CPU of this runqueue: */ | |
983 | int cpu; | |
984 | int online; | |
029632fb | 985 | |
367456c7 PZ |
986 | struct list_head cfs_tasks; |
987 | ||
371bf427 | 988 | struct sched_avg avg_rt; |
3727e0e1 | 989 | struct sched_avg avg_dl; |
11d4afd4 | 990 | #ifdef CONFIG_HAVE_SCHED_AVG_IRQ |
91c27493 | 991 | struct sched_avg avg_irq; |
76504793 TG |
992 | #endif |
993 | #ifdef CONFIG_SCHED_THERMAL_PRESSURE | |
994 | struct sched_avg avg_thermal; | |
91c27493 | 995 | #endif |
97fb7a0a IM |
996 | u64 idle_stamp; |
997 | u64 avg_idle; | |
9bd721c5 JL |
998 | |
999 | /* This is used to determine avg_idle's max value */ | |
97fb7a0a | 1000 | u64 max_idle_balance_cost; |
90b5363a | 1001 | #endif /* CONFIG_SMP */ |
029632fb PZ |
1002 | |
1003 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
97fb7a0a | 1004 | u64 prev_irq_time; |
029632fb PZ |
1005 | #endif |
1006 | #ifdef CONFIG_PARAVIRT | |
97fb7a0a | 1007 | u64 prev_steal_time; |
029632fb PZ |
1008 | #endif |
1009 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
97fb7a0a | 1010 | u64 prev_steal_time_rq; |
029632fb PZ |
1011 | #endif |
1012 | ||
1013 | /* calc_load related fields */ | |
97fb7a0a IM |
1014 | unsigned long calc_load_update; |
1015 | long calc_load_active; | |
029632fb PZ |
1016 | |
1017 | #ifdef CONFIG_SCHED_HRTICK | |
1018 | #ifdef CONFIG_SMP | |
97fb7a0a | 1019 | call_single_data_t hrtick_csd; |
029632fb | 1020 | #endif |
97fb7a0a | 1021 | struct hrtimer hrtick_timer; |
029632fb PZ |
1022 | #endif |
1023 | ||
1024 | #ifdef CONFIG_SCHEDSTATS | |
1025 | /* latency stats */ | |
97fb7a0a IM |
1026 | struct sched_info rq_sched_info; |
1027 | unsigned long long rq_cpu_time; | |
029632fb PZ |
1028 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ |
1029 | ||
1030 | /* sys_sched_yield() stats */ | |
97fb7a0a | 1031 | unsigned int yld_count; |
029632fb PZ |
1032 | |
1033 | /* schedule() stats */ | |
97fb7a0a IM |
1034 | unsigned int sched_count; |
1035 | unsigned int sched_goidle; | |
029632fb PZ |
1036 | |
1037 | /* try_to_wake_up() stats */ | |
97fb7a0a IM |
1038 | unsigned int ttwu_count; |
1039 | unsigned int ttwu_local; | |
029632fb PZ |
1040 | #endif |
1041 | ||
442bf3aa DL |
1042 | #ifdef CONFIG_CPU_IDLE |
1043 | /* Must be inspected within a rcu lock section */ | |
97fb7a0a | 1044 | struct cpuidle_state *idle_state; |
442bf3aa | 1045 | #endif |
029632fb PZ |
1046 | }; |
1047 | ||
62478d99 VG |
1048 | #ifdef CONFIG_FAIR_GROUP_SCHED |
1049 | ||
1050 | /* CPU runqueue to which this cfs_rq is attached */ | |
1051 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) | |
1052 | { | |
1053 | return cfs_rq->rq; | |
1054 | } | |
1055 | ||
1056 | #else | |
1057 | ||
1058 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) | |
1059 | { | |
1060 | return container_of(cfs_rq, struct rq, cfs); | |
1061 | } | |
1062 | #endif | |
1063 | ||
029632fb PZ |
1064 | static inline int cpu_of(struct rq *rq) |
1065 | { | |
1066 | #ifdef CONFIG_SMP | |
1067 | return rq->cpu; | |
1068 | #else | |
1069 | return 0; | |
1070 | #endif | |
1071 | } | |
1072 | ||
1b568f0a PZ |
1073 | |
1074 | #ifdef CONFIG_SCHED_SMT | |
1b568f0a PZ |
1075 | extern void __update_idle_core(struct rq *rq); |
1076 | ||
1077 | static inline void update_idle_core(struct rq *rq) | |
1078 | { | |
1079 | if (static_branch_unlikely(&sched_smt_present)) | |
1080 | __update_idle_core(rq); | |
1081 | } | |
1082 | ||
1083 | #else | |
1084 | static inline void update_idle_core(struct rq *rq) { } | |
1085 | #endif | |
1086 | ||
8b06c55b | 1087 | DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
029632fb | 1088 | |
518cd623 | 1089 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) |
4a32fea9 | 1090 | #define this_rq() this_cpu_ptr(&runqueues) |
518cd623 PZ |
1091 | #define task_rq(p) cpu_rq(task_cpu(p)) |
1092 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
4a32fea9 | 1093 | #define raw_rq() raw_cpu_ptr(&runqueues) |
518cd623 | 1094 | |
1f351d7f JW |
1095 | extern void update_rq_clock(struct rq *rq); |
1096 | ||
cebde6d6 PZ |
1097 | static inline u64 __rq_clock_broken(struct rq *rq) |
1098 | { | |
316c1608 | 1099 | return READ_ONCE(rq->clock); |
cebde6d6 PZ |
1100 | } |
1101 | ||
cb42c9a3 MF |
1102 | /* |
1103 | * rq::clock_update_flags bits | |
1104 | * | |
1105 | * %RQCF_REQ_SKIP - will request skipping of clock update on the next | |
1106 | * call to __schedule(). This is an optimisation to avoid | |
1107 | * neighbouring rq clock updates. | |
1108 | * | |
1109 | * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is | |
1110 | * in effect and calls to update_rq_clock() are being ignored. | |
1111 | * | |
1112 | * %RQCF_UPDATED - is a debug flag that indicates whether a call has been | |
1113 | * made to update_rq_clock() since the last time rq::lock was pinned. | |
1114 | * | |
1115 | * If inside of __schedule(), clock_update_flags will have been | |
1116 | * shifted left (a left shift is a cheap operation for the fast path | |
1117 | * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use, | |
1118 | * | |
1119 | * if (rq-clock_update_flags >= RQCF_UPDATED) | |
1120 | * | |
1121 | * to check if %RQCF_UPADTED is set. It'll never be shifted more than | |
1122 | * one position though, because the next rq_unpin_lock() will shift it | |
1123 | * back. | |
1124 | */ | |
97fb7a0a IM |
1125 | #define RQCF_REQ_SKIP 0x01 |
1126 | #define RQCF_ACT_SKIP 0x02 | |
1127 | #define RQCF_UPDATED 0x04 | |
cb42c9a3 MF |
1128 | |
1129 | static inline void assert_clock_updated(struct rq *rq) | |
1130 | { | |
1131 | /* | |
1132 | * The only reason for not seeing a clock update since the | |
1133 | * last rq_pin_lock() is if we're currently skipping updates. | |
1134 | */ | |
1135 | SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP); | |
1136 | } | |
1137 | ||
78becc27 FW |
1138 | static inline u64 rq_clock(struct rq *rq) |
1139 | { | |
cebde6d6 | 1140 | lockdep_assert_held(&rq->lock); |
cb42c9a3 MF |
1141 | assert_clock_updated(rq); |
1142 | ||
78becc27 FW |
1143 | return rq->clock; |
1144 | } | |
1145 | ||
1146 | static inline u64 rq_clock_task(struct rq *rq) | |
1147 | { | |
cebde6d6 | 1148 | lockdep_assert_held(&rq->lock); |
cb42c9a3 MF |
1149 | assert_clock_updated(rq); |
1150 | ||
78becc27 FW |
1151 | return rq->clock_task; |
1152 | } | |
1153 | ||
05289b90 TG |
1154 | /** |
1155 | * By default the decay is the default pelt decay period. | |
1156 | * The decay shift can change the decay period in | |
1157 | * multiples of 32. | |
1158 | * Decay shift Decay period(ms) | |
1159 | * 0 32 | |
1160 | * 1 64 | |
1161 | * 2 128 | |
1162 | * 3 256 | |
1163 | * 4 512 | |
1164 | */ | |
1165 | extern int sched_thermal_decay_shift; | |
1166 | ||
1167 | static inline u64 rq_clock_thermal(struct rq *rq) | |
1168 | { | |
1169 | return rq_clock_task(rq) >> sched_thermal_decay_shift; | |
1170 | } | |
1171 | ||
adcc8da8 | 1172 | static inline void rq_clock_skip_update(struct rq *rq) |
9edfbfed PZ |
1173 | { |
1174 | lockdep_assert_held(&rq->lock); | |
adcc8da8 DB |
1175 | rq->clock_update_flags |= RQCF_REQ_SKIP; |
1176 | } | |
1177 | ||
1178 | /* | |
595058b6 | 1179 | * See rt task throttling, which is the only time a skip |
adcc8da8 DB |
1180 | * request is cancelled. |
1181 | */ | |
1182 | static inline void rq_clock_cancel_skipupdate(struct rq *rq) | |
1183 | { | |
1184 | lockdep_assert_held(&rq->lock); | |
1185 | rq->clock_update_flags &= ~RQCF_REQ_SKIP; | |
9edfbfed PZ |
1186 | } |
1187 | ||
d8ac8971 MF |
1188 | struct rq_flags { |
1189 | unsigned long flags; | |
1190 | struct pin_cookie cookie; | |
cb42c9a3 MF |
1191 | #ifdef CONFIG_SCHED_DEBUG |
1192 | /* | |
1193 | * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the | |
1194 | * current pin context is stashed here in case it needs to be | |
1195 | * restored in rq_repin_lock(). | |
1196 | */ | |
1197 | unsigned int clock_update_flags; | |
1198 | #endif | |
d8ac8971 MF |
1199 | }; |
1200 | ||
1201 | static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf) | |
1202 | { | |
1203 | rf->cookie = lockdep_pin_lock(&rq->lock); | |
cb42c9a3 MF |
1204 | |
1205 | #ifdef CONFIG_SCHED_DEBUG | |
1206 | rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP); | |
1207 | rf->clock_update_flags = 0; | |
1208 | #endif | |
d8ac8971 MF |
1209 | } |
1210 | ||
1211 | static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf) | |
1212 | { | |
cb42c9a3 MF |
1213 | #ifdef CONFIG_SCHED_DEBUG |
1214 | if (rq->clock_update_flags > RQCF_ACT_SKIP) | |
1215 | rf->clock_update_flags = RQCF_UPDATED; | |
1216 | #endif | |
1217 | ||
d8ac8971 MF |
1218 | lockdep_unpin_lock(&rq->lock, rf->cookie); |
1219 | } | |
1220 | ||
1221 | static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf) | |
1222 | { | |
1223 | lockdep_repin_lock(&rq->lock, rf->cookie); | |
cb42c9a3 MF |
1224 | |
1225 | #ifdef CONFIG_SCHED_DEBUG | |
1226 | /* | |
1227 | * Restore the value we stashed in @rf for this pin context. | |
1228 | */ | |
1229 | rq->clock_update_flags |= rf->clock_update_flags; | |
1230 | #endif | |
d8ac8971 MF |
1231 | } |
1232 | ||
1f351d7f JW |
1233 | struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
1234 | __acquires(rq->lock); | |
1235 | ||
1236 | struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) | |
1237 | __acquires(p->pi_lock) | |
1238 | __acquires(rq->lock); | |
1239 | ||
1240 | static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf) | |
1241 | __releases(rq->lock) | |
1242 | { | |
1243 | rq_unpin_lock(rq, rf); | |
1244 | raw_spin_unlock(&rq->lock); | |
1245 | } | |
1246 | ||
1247 | static inline void | |
1248 | task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf) | |
1249 | __releases(rq->lock) | |
1250 | __releases(p->pi_lock) | |
1251 | { | |
1252 | rq_unpin_lock(rq, rf); | |
1253 | raw_spin_unlock(&rq->lock); | |
1254 | raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); | |
1255 | } | |
1256 | ||
1257 | static inline void | |
1258 | rq_lock_irqsave(struct rq *rq, struct rq_flags *rf) | |
1259 | __acquires(rq->lock) | |
1260 | { | |
1261 | raw_spin_lock_irqsave(&rq->lock, rf->flags); | |
1262 | rq_pin_lock(rq, rf); | |
1263 | } | |
1264 | ||
1265 | static inline void | |
1266 | rq_lock_irq(struct rq *rq, struct rq_flags *rf) | |
1267 | __acquires(rq->lock) | |
1268 | { | |
1269 | raw_spin_lock_irq(&rq->lock); | |
1270 | rq_pin_lock(rq, rf); | |
1271 | } | |
1272 | ||
1273 | static inline void | |
1274 | rq_lock(struct rq *rq, struct rq_flags *rf) | |
1275 | __acquires(rq->lock) | |
1276 | { | |
1277 | raw_spin_lock(&rq->lock); | |
1278 | rq_pin_lock(rq, rf); | |
1279 | } | |
1280 | ||
1281 | static inline void | |
1282 | rq_relock(struct rq *rq, struct rq_flags *rf) | |
1283 | __acquires(rq->lock) | |
1284 | { | |
1285 | raw_spin_lock(&rq->lock); | |
1286 | rq_repin_lock(rq, rf); | |
1287 | } | |
1288 | ||
1289 | static inline void | |
1290 | rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf) | |
1291 | __releases(rq->lock) | |
1292 | { | |
1293 | rq_unpin_lock(rq, rf); | |
1294 | raw_spin_unlock_irqrestore(&rq->lock, rf->flags); | |
1295 | } | |
1296 | ||
1297 | static inline void | |
1298 | rq_unlock_irq(struct rq *rq, struct rq_flags *rf) | |
1299 | __releases(rq->lock) | |
1300 | { | |
1301 | rq_unpin_lock(rq, rf); | |
1302 | raw_spin_unlock_irq(&rq->lock); | |
1303 | } | |
1304 | ||
1305 | static inline void | |
1306 | rq_unlock(struct rq *rq, struct rq_flags *rf) | |
1307 | __releases(rq->lock) | |
1308 | { | |
1309 | rq_unpin_lock(rq, rf); | |
1310 | raw_spin_unlock(&rq->lock); | |
1311 | } | |
1312 | ||
246b3b33 JW |
1313 | static inline struct rq * |
1314 | this_rq_lock_irq(struct rq_flags *rf) | |
1315 | __acquires(rq->lock) | |
1316 | { | |
1317 | struct rq *rq; | |
1318 | ||
1319 | local_irq_disable(); | |
1320 | rq = this_rq(); | |
1321 | rq_lock(rq, rf); | |
1322 | return rq; | |
1323 | } | |
1324 | ||
9942f79b | 1325 | #ifdef CONFIG_NUMA |
e3fe70b1 RR |
1326 | enum numa_topology_type { |
1327 | NUMA_DIRECT, | |
1328 | NUMA_GLUELESS_MESH, | |
1329 | NUMA_BACKPLANE, | |
1330 | }; | |
1331 | extern enum numa_topology_type sched_numa_topology_type; | |
9942f79b RR |
1332 | extern int sched_max_numa_distance; |
1333 | extern bool find_numa_distance(int distance); | |
f2cb1360 IM |
1334 | extern void sched_init_numa(void); |
1335 | extern void sched_domains_numa_masks_set(unsigned int cpu); | |
1336 | extern void sched_domains_numa_masks_clear(unsigned int cpu); | |
e0e8d491 | 1337 | extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu); |
f2cb1360 IM |
1338 | #else |
1339 | static inline void sched_init_numa(void) { } | |
1340 | static inline void sched_domains_numa_masks_set(unsigned int cpu) { } | |
1341 | static inline void sched_domains_numa_masks_clear(unsigned int cpu) { } | |
e0e8d491 WL |
1342 | static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu) |
1343 | { | |
1344 | return nr_cpu_ids; | |
1345 | } | |
f2cb1360 IM |
1346 | #endif |
1347 | ||
f809ca9a | 1348 | #ifdef CONFIG_NUMA_BALANCING |
44dba3d5 IM |
1349 | /* The regions in numa_faults array from task_struct */ |
1350 | enum numa_faults_stats { | |
1351 | NUMA_MEM = 0, | |
1352 | NUMA_CPU, | |
1353 | NUMA_MEMBUF, | |
1354 | NUMA_CPUBUF | |
1355 | }; | |
0ec8aa00 | 1356 | extern void sched_setnuma(struct task_struct *p, int node); |
e6628d5b | 1357 | extern int migrate_task_to(struct task_struct *p, int cpu); |
0ad4e3df SD |
1358 | extern int migrate_swap(struct task_struct *p, struct task_struct *t, |
1359 | int cpu, int scpu); | |
13784475 MG |
1360 | extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p); |
1361 | #else | |
1362 | static inline void | |
1363 | init_numa_balancing(unsigned long clone_flags, struct task_struct *p) | |
1364 | { | |
1365 | } | |
f809ca9a MG |
1366 | #endif /* CONFIG_NUMA_BALANCING */ |
1367 | ||
518cd623 PZ |
1368 | #ifdef CONFIG_SMP |
1369 | ||
e3fca9e7 PZ |
1370 | static inline void |
1371 | queue_balance_callback(struct rq *rq, | |
1372 | struct callback_head *head, | |
1373 | void (*func)(struct rq *rq)) | |
1374 | { | |
1375 | lockdep_assert_held(&rq->lock); | |
1376 | ||
1377 | if (unlikely(head->next)) | |
1378 | return; | |
1379 | ||
1380 | head->func = (void (*)(struct callback_head *))func; | |
1381 | head->next = rq->balance_callback; | |
1382 | rq->balance_callback = head; | |
1383 | } | |
1384 | ||
029632fb PZ |
1385 | #define rcu_dereference_check_sched_domain(p) \ |
1386 | rcu_dereference_check((p), \ | |
1387 | lockdep_is_held(&sched_domains_mutex)) | |
1388 | ||
1389 | /* | |
1390 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
337e9b07 | 1391 | * See destroy_sched_domains: call_rcu for details. |
029632fb PZ |
1392 | * |
1393 | * The domain tree of any CPU may only be accessed from within | |
1394 | * preempt-disabled sections. | |
1395 | */ | |
1396 | #define for_each_domain(cpu, __sd) \ | |
518cd623 PZ |
1397 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \ |
1398 | __sd; __sd = __sd->parent) | |
029632fb | 1399 | |
518cd623 PZ |
1400 | /** |
1401 | * highest_flag_domain - Return highest sched_domain containing flag. | |
97fb7a0a | 1402 | * @cpu: The CPU whose highest level of sched domain is to |
518cd623 PZ |
1403 | * be returned. |
1404 | * @flag: The flag to check for the highest sched_domain | |
97fb7a0a | 1405 | * for the given CPU. |
518cd623 | 1406 | * |
97fb7a0a | 1407 | * Returns the highest sched_domain of a CPU which contains the given flag. |
518cd623 PZ |
1408 | */ |
1409 | static inline struct sched_domain *highest_flag_domain(int cpu, int flag) | |
1410 | { | |
1411 | struct sched_domain *sd, *hsd = NULL; | |
1412 | ||
1413 | for_each_domain(cpu, sd) { | |
1414 | if (!(sd->flags & flag)) | |
1415 | break; | |
1416 | hsd = sd; | |
1417 | } | |
1418 | ||
1419 | return hsd; | |
1420 | } | |
1421 | ||
fb13c7ee MG |
1422 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) |
1423 | { | |
1424 | struct sched_domain *sd; | |
1425 | ||
1426 | for_each_domain(cpu, sd) { | |
1427 | if (sd->flags & flag) | |
1428 | break; | |
1429 | } | |
1430 | ||
1431 | return sd; | |
1432 | } | |
1433 | ||
994aeb7a | 1434 | DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc); |
7d9ffa89 | 1435 | DECLARE_PER_CPU(int, sd_llc_size); |
518cd623 | 1436 | DECLARE_PER_CPU(int, sd_llc_id); |
994aeb7a JFG |
1437 | DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared); |
1438 | DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa); | |
1439 | DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing); | |
1440 | DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity); | |
df054e84 | 1441 | extern struct static_key_false sched_asym_cpucapacity; |
518cd623 | 1442 | |
63b2ca30 | 1443 | struct sched_group_capacity { |
97fb7a0a | 1444 | atomic_t ref; |
5e6521ea | 1445 | /* |
172895e6 | 1446 | * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity |
63b2ca30 | 1447 | * for a single CPU. |
5e6521ea | 1448 | */ |
97fb7a0a IM |
1449 | unsigned long capacity; |
1450 | unsigned long min_capacity; /* Min per-CPU capacity in group */ | |
e3d6d0cb | 1451 | unsigned long max_capacity; /* Max per-CPU capacity in group */ |
97fb7a0a IM |
1452 | unsigned long next_update; |
1453 | int imbalance; /* XXX unrelated to capacity but shared group state */ | |
5e6521ea | 1454 | |
005f874d | 1455 | #ifdef CONFIG_SCHED_DEBUG |
97fb7a0a | 1456 | int id; |
005f874d PZ |
1457 | #endif |
1458 | ||
97fb7a0a | 1459 | unsigned long cpumask[0]; /* Balance mask */ |
5e6521ea LZ |
1460 | }; |
1461 | ||
1462 | struct sched_group { | |
97fb7a0a IM |
1463 | struct sched_group *next; /* Must be a circular list */ |
1464 | atomic_t ref; | |
5e6521ea | 1465 | |
97fb7a0a | 1466 | unsigned int group_weight; |
63b2ca30 | 1467 | struct sched_group_capacity *sgc; |
97fb7a0a | 1468 | int asym_prefer_cpu; /* CPU of highest priority in group */ |
5e6521ea LZ |
1469 | |
1470 | /* | |
1471 | * The CPUs this group covers. | |
1472 | * | |
1473 | * NOTE: this field is variable length. (Allocated dynamically | |
1474 | * by attaching extra space to the end of the structure, | |
1475 | * depending on how many CPUs the kernel has booted up with) | |
1476 | */ | |
04f5c362 | 1477 | unsigned long cpumask[]; |
5e6521ea LZ |
1478 | }; |
1479 | ||
ae4df9d6 | 1480 | static inline struct cpumask *sched_group_span(struct sched_group *sg) |
5e6521ea LZ |
1481 | { |
1482 | return to_cpumask(sg->cpumask); | |
1483 | } | |
1484 | ||
1485 | /* | |
e5c14b1f | 1486 | * See build_balance_mask(). |
5e6521ea | 1487 | */ |
e5c14b1f | 1488 | static inline struct cpumask *group_balance_mask(struct sched_group *sg) |
5e6521ea | 1489 | { |
63b2ca30 | 1490 | return to_cpumask(sg->sgc->cpumask); |
5e6521ea LZ |
1491 | } |
1492 | ||
1493 | /** | |
97fb7a0a IM |
1494 | * group_first_cpu - Returns the first CPU in the cpumask of a sched_group. |
1495 | * @group: The group whose first CPU is to be returned. | |
5e6521ea LZ |
1496 | */ |
1497 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
1498 | { | |
ae4df9d6 | 1499 | return cpumask_first(sched_group_span(group)); |
5e6521ea LZ |
1500 | } |
1501 | ||
c1174876 PZ |
1502 | extern int group_balance_cpu(struct sched_group *sg); |
1503 | ||
3866e845 SRRH |
1504 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
1505 | void register_sched_domain_sysctl(void); | |
bbdacdfe | 1506 | void dirty_sched_domain_sysctl(int cpu); |
3866e845 SRRH |
1507 | void unregister_sched_domain_sysctl(void); |
1508 | #else | |
1509 | static inline void register_sched_domain_sysctl(void) | |
1510 | { | |
1511 | } | |
bbdacdfe PZ |
1512 | static inline void dirty_sched_domain_sysctl(int cpu) |
1513 | { | |
1514 | } | |
3866e845 SRRH |
1515 | static inline void unregister_sched_domain_sysctl(void) |
1516 | { | |
1517 | } | |
1518 | #endif | |
1519 | ||
b2a02fc4 | 1520 | extern void flush_smp_call_function_from_idle(void); |
e3baac47 | 1521 | |
b2a02fc4 PZ |
1522 | #else /* !CONFIG_SMP: */ |
1523 | static inline void flush_smp_call_function_from_idle(void) { } | |
b2a02fc4 | 1524 | #endif |
029632fb | 1525 | |
391e43da | 1526 | #include "stats.h" |
1051408f | 1527 | #include "autogroup.h" |
029632fb PZ |
1528 | |
1529 | #ifdef CONFIG_CGROUP_SCHED | |
1530 | ||
1531 | /* | |
1532 | * Return the group to which this tasks belongs. | |
1533 | * | |
8af01f56 TH |
1534 | * We cannot use task_css() and friends because the cgroup subsystem |
1535 | * changes that value before the cgroup_subsys::attach() method is called, | |
1536 | * therefore we cannot pin it and might observe the wrong value. | |
8323f26c PZ |
1537 | * |
1538 | * The same is true for autogroup's p->signal->autogroup->tg, the autogroup | |
1539 | * core changes this before calling sched_move_task(). | |
1540 | * | |
1541 | * Instead we use a 'copy' which is updated from sched_move_task() while | |
1542 | * holding both task_struct::pi_lock and rq::lock. | |
029632fb PZ |
1543 | */ |
1544 | static inline struct task_group *task_group(struct task_struct *p) | |
1545 | { | |
8323f26c | 1546 | return p->sched_task_group; |
029632fb PZ |
1547 | } |
1548 | ||
1549 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
1550 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
1551 | { | |
1552 | #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED) | |
1553 | struct task_group *tg = task_group(p); | |
1554 | #endif | |
1555 | ||
1556 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
ad936d86 | 1557 | set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]); |
029632fb PZ |
1558 | p->se.cfs_rq = tg->cfs_rq[cpu]; |
1559 | p->se.parent = tg->se[cpu]; | |
1560 | #endif | |
1561 | ||
1562 | #ifdef CONFIG_RT_GROUP_SCHED | |
1563 | p->rt.rt_rq = tg->rt_rq[cpu]; | |
1564 | p->rt.parent = tg->rt_se[cpu]; | |
1565 | #endif | |
1566 | } | |
1567 | ||
1568 | #else /* CONFIG_CGROUP_SCHED */ | |
1569 | ||
1570 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
1571 | static inline struct task_group *task_group(struct task_struct *p) | |
1572 | { | |
1573 | return NULL; | |
1574 | } | |
1575 | ||
1576 | #endif /* CONFIG_CGROUP_SCHED */ | |
1577 | ||
1578 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1579 | { | |
1580 | set_task_rq(p, cpu); | |
1581 | #ifdef CONFIG_SMP | |
1582 | /* | |
1583 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
dfcb245e | 1584 | * successfully executed on another CPU. We must ensure that updates of |
029632fb PZ |
1585 | * per-task data have been completed by this moment. |
1586 | */ | |
1587 | smp_wmb(); | |
c65eacbe | 1588 | #ifdef CONFIG_THREAD_INFO_IN_TASK |
c546951d | 1589 | WRITE_ONCE(p->cpu, cpu); |
c65eacbe | 1590 | #else |
c546951d | 1591 | WRITE_ONCE(task_thread_info(p)->cpu, cpu); |
c65eacbe | 1592 | #endif |
ac66f547 | 1593 | p->wake_cpu = cpu; |
029632fb PZ |
1594 | #endif |
1595 | } | |
1596 | ||
1597 | /* | |
1598 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
1599 | */ | |
1600 | #ifdef CONFIG_SCHED_DEBUG | |
c5905afb | 1601 | # include <linux/static_key.h> |
029632fb PZ |
1602 | # define const_debug __read_mostly |
1603 | #else | |
1604 | # define const_debug const | |
1605 | #endif | |
1606 | ||
029632fb PZ |
1607 | #define SCHED_FEAT(name, enabled) \ |
1608 | __SCHED_FEAT_##name , | |
1609 | ||
1610 | enum { | |
391e43da | 1611 | #include "features.h" |
f8b6d1cc | 1612 | __SCHED_FEAT_NR, |
029632fb PZ |
1613 | }; |
1614 | ||
1615 | #undef SCHED_FEAT | |
1616 | ||
e9666d10 | 1617 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_JUMP_LABEL) |
765cc3a4 PB |
1618 | |
1619 | /* | |
1620 | * To support run-time toggling of sched features, all the translation units | |
1621 | * (but core.c) reference the sysctl_sched_features defined in core.c. | |
1622 | */ | |
1623 | extern const_debug unsigned int sysctl_sched_features; | |
1624 | ||
f8b6d1cc | 1625 | #define SCHED_FEAT(name, enabled) \ |
c5905afb | 1626 | static __always_inline bool static_branch_##name(struct static_key *key) \ |
f8b6d1cc | 1627 | { \ |
6e76ea8a | 1628 | return static_key_##enabled(key); \ |
f8b6d1cc PZ |
1629 | } |
1630 | ||
1631 | #include "features.h" | |
f8b6d1cc PZ |
1632 | #undef SCHED_FEAT |
1633 | ||
c5905afb | 1634 | extern struct static_key sched_feat_keys[__SCHED_FEAT_NR]; |
f8b6d1cc | 1635 | #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x])) |
765cc3a4 | 1636 | |
e9666d10 | 1637 | #else /* !(SCHED_DEBUG && CONFIG_JUMP_LABEL) */ |
765cc3a4 PB |
1638 | |
1639 | /* | |
1640 | * Each translation unit has its own copy of sysctl_sched_features to allow | |
1641 | * constants propagation at compile time and compiler optimization based on | |
1642 | * features default. | |
1643 | */ | |
1644 | #define SCHED_FEAT(name, enabled) \ | |
1645 | (1UL << __SCHED_FEAT_##name) * enabled | | |
1646 | static const_debug __maybe_unused unsigned int sysctl_sched_features = | |
1647 | #include "features.h" | |
1648 | 0; | |
1649 | #undef SCHED_FEAT | |
1650 | ||
7e6f4c5d | 1651 | #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) |
765cc3a4 | 1652 | |
e9666d10 | 1653 | #endif /* SCHED_DEBUG && CONFIG_JUMP_LABEL */ |
029632fb | 1654 | |
2a595721 | 1655 | extern struct static_key_false sched_numa_balancing; |
cb251765 | 1656 | extern struct static_key_false sched_schedstats; |
cbee9f88 | 1657 | |
029632fb PZ |
1658 | static inline u64 global_rt_period(void) |
1659 | { | |
1660 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
1661 | } | |
1662 | ||
1663 | static inline u64 global_rt_runtime(void) | |
1664 | { | |
1665 | if (sysctl_sched_rt_runtime < 0) | |
1666 | return RUNTIME_INF; | |
1667 | ||
1668 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
1669 | } | |
1670 | ||
029632fb PZ |
1671 | static inline int task_current(struct rq *rq, struct task_struct *p) |
1672 | { | |
1673 | return rq->curr == p; | |
1674 | } | |
1675 | ||
1676 | static inline int task_running(struct rq *rq, struct task_struct *p) | |
1677 | { | |
1678 | #ifdef CONFIG_SMP | |
1679 | return p->on_cpu; | |
1680 | #else | |
1681 | return task_current(rq, p); | |
1682 | #endif | |
1683 | } | |
1684 | ||
da0c1e65 KT |
1685 | static inline int task_on_rq_queued(struct task_struct *p) |
1686 | { | |
1687 | return p->on_rq == TASK_ON_RQ_QUEUED; | |
1688 | } | |
029632fb | 1689 | |
cca26e80 KT |
1690 | static inline int task_on_rq_migrating(struct task_struct *p) |
1691 | { | |
c546951d | 1692 | return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING; |
cca26e80 KT |
1693 | } |
1694 | ||
b13095f0 LZ |
1695 | /* |
1696 | * wake flags | |
1697 | */ | |
97fb7a0a IM |
1698 | #define WF_SYNC 0x01 /* Waker goes to sleep after wakeup */ |
1699 | #define WF_FORK 0x02 /* Child wakeup after fork */ | |
2ebb1771 | 1700 | #define WF_MIGRATED 0x04 /* Internal use, task got migrated */ |
739f70b4 | 1701 | #define WF_ON_CPU 0x08 /* Wakee is on_cpu */ |
b13095f0 | 1702 | |
029632fb PZ |
1703 | /* |
1704 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1705 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1706 | * each task makes to its run queue's load is weighted according to its | |
1707 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
1708 | * scaled version of the new time slice allocation that they receive on time | |
1709 | * slice expiry etc. | |
1710 | */ | |
1711 | ||
97fb7a0a IM |
1712 | #define WEIGHT_IDLEPRIO 3 |
1713 | #define WMULT_IDLEPRIO 1431655765 | |
029632fb | 1714 | |
97fb7a0a IM |
1715 | extern const int sched_prio_to_weight[40]; |
1716 | extern const u32 sched_prio_to_wmult[40]; | |
029632fb | 1717 | |
ff77e468 PZ |
1718 | /* |
1719 | * {de,en}queue flags: | |
1720 | * | |
1721 | * DEQUEUE_SLEEP - task is no longer runnable | |
1722 | * ENQUEUE_WAKEUP - task just became runnable | |
1723 | * | |
1724 | * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks | |
1725 | * are in a known state which allows modification. Such pairs | |
1726 | * should preserve as much state as possible. | |
1727 | * | |
1728 | * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location | |
1729 | * in the runqueue. | |
1730 | * | |
1731 | * ENQUEUE_HEAD - place at front of runqueue (tail if not specified) | |
1732 | * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline) | |
59efa0ba | 1733 | * ENQUEUE_MIGRATED - the task was migrated during wakeup |
ff77e468 PZ |
1734 | * |
1735 | */ | |
1736 | ||
1737 | #define DEQUEUE_SLEEP 0x01 | |
97fb7a0a IM |
1738 | #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */ |
1739 | #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */ | |
1740 | #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */ | |
ff77e468 | 1741 | |
1de64443 | 1742 | #define ENQUEUE_WAKEUP 0x01 |
ff77e468 PZ |
1743 | #define ENQUEUE_RESTORE 0x02 |
1744 | #define ENQUEUE_MOVE 0x04 | |
0a67d1ee | 1745 | #define ENQUEUE_NOCLOCK 0x08 |
ff77e468 | 1746 | |
0a67d1ee PZ |
1747 | #define ENQUEUE_HEAD 0x10 |
1748 | #define ENQUEUE_REPLENISH 0x20 | |
c82ba9fa | 1749 | #ifdef CONFIG_SMP |
0a67d1ee | 1750 | #define ENQUEUE_MIGRATED 0x40 |
c82ba9fa | 1751 | #else |
59efa0ba | 1752 | #define ENQUEUE_MIGRATED 0x00 |
c82ba9fa | 1753 | #endif |
c82ba9fa | 1754 | |
37e117c0 PZ |
1755 | #define RETRY_TASK ((void *)-1UL) |
1756 | ||
c82ba9fa | 1757 | struct sched_class { |
c82ba9fa | 1758 | |
69842cba PB |
1759 | #ifdef CONFIG_UCLAMP_TASK |
1760 | int uclamp_enabled; | |
1761 | #endif | |
1762 | ||
c82ba9fa LZ |
1763 | void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags); |
1764 | void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags); | |
97fb7a0a | 1765 | void (*yield_task) (struct rq *rq); |
0900acf2 | 1766 | bool (*yield_to_task)(struct rq *rq, struct task_struct *p); |
c82ba9fa | 1767 | |
97fb7a0a | 1768 | void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags); |
c82ba9fa | 1769 | |
98c2f700 PZ |
1770 | struct task_struct *(*pick_next_task)(struct rq *rq); |
1771 | ||
6e2df058 | 1772 | void (*put_prev_task)(struct rq *rq, struct task_struct *p); |
a0e813f2 | 1773 | void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first); |
c82ba9fa LZ |
1774 | |
1775 | #ifdef CONFIG_SMP | |
6e2df058 | 1776 | int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf); |
ac66f547 | 1777 | int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags); |
1327237a | 1778 | void (*migrate_task_rq)(struct task_struct *p, int new_cpu); |
c82ba9fa | 1779 | |
97fb7a0a | 1780 | void (*task_woken)(struct rq *this_rq, struct task_struct *task); |
c82ba9fa LZ |
1781 | |
1782 | void (*set_cpus_allowed)(struct task_struct *p, | |
1783 | const struct cpumask *newmask); | |
1784 | ||
1785 | void (*rq_online)(struct rq *rq); | |
1786 | void (*rq_offline)(struct rq *rq); | |
1787 | #endif | |
1788 | ||
97fb7a0a IM |
1789 | void (*task_tick)(struct rq *rq, struct task_struct *p, int queued); |
1790 | void (*task_fork)(struct task_struct *p); | |
1791 | void (*task_dead)(struct task_struct *p); | |
c82ba9fa | 1792 | |
67dfa1b7 KT |
1793 | /* |
1794 | * The switched_from() call is allowed to drop rq->lock, therefore we | |
1795 | * cannot assume the switched_from/switched_to pair is serliazed by | |
1796 | * rq->lock. They are however serialized by p->pi_lock. | |
1797 | */ | |
97fb7a0a IM |
1798 | void (*switched_from)(struct rq *this_rq, struct task_struct *task); |
1799 | void (*switched_to) (struct rq *this_rq, struct task_struct *task); | |
c82ba9fa | 1800 | void (*prio_changed) (struct rq *this_rq, struct task_struct *task, |
97fb7a0a | 1801 | int oldprio); |
c82ba9fa | 1802 | |
97fb7a0a IM |
1803 | unsigned int (*get_rr_interval)(struct rq *rq, |
1804 | struct task_struct *task); | |
c82ba9fa | 1805 | |
97fb7a0a | 1806 | void (*update_curr)(struct rq *rq); |
6e998916 | 1807 | |
97fb7a0a IM |
1808 | #define TASK_SET_GROUP 0 |
1809 | #define TASK_MOVE_GROUP 1 | |
ea86cb4b | 1810 | |
c82ba9fa | 1811 | #ifdef CONFIG_FAIR_GROUP_SCHED |
97fb7a0a | 1812 | void (*task_change_group)(struct task_struct *p, int type); |
c82ba9fa | 1813 | #endif |
85c2ce91 | 1814 | } __aligned(STRUCT_ALIGNMENT); /* STRUCT_ALIGN(), vmlinux.lds.h */ |
029632fb | 1815 | |
3f1d2a31 PZ |
1816 | static inline void put_prev_task(struct rq *rq, struct task_struct *prev) |
1817 | { | |
10e7071b | 1818 | WARN_ON_ONCE(rq->curr != prev); |
6e2df058 | 1819 | prev->sched_class->put_prev_task(rq, prev); |
3f1d2a31 PZ |
1820 | } |
1821 | ||
03b7fad1 | 1822 | static inline void set_next_task(struct rq *rq, struct task_struct *next) |
b2bf6c31 | 1823 | { |
03b7fad1 | 1824 | WARN_ON_ONCE(rq->curr != next); |
a0e813f2 | 1825 | next->sched_class->set_next_task(rq, next, false); |
b2bf6c31 PZ |
1826 | } |
1827 | ||
c3a340f7 SRV |
1828 | /* Defined in include/asm-generic/vmlinux.lds.h */ |
1829 | extern struct sched_class __begin_sched_classes[]; | |
1830 | extern struct sched_class __end_sched_classes[]; | |
1831 | ||
1832 | #define sched_class_highest (__end_sched_classes - 1) | |
1833 | #define sched_class_lowest (__begin_sched_classes - 1) | |
6e2df058 PZ |
1834 | |
1835 | #define for_class_range(class, _from, _to) \ | |
c3a340f7 | 1836 | for (class = (_from); class != (_to); class--) |
6e2df058 | 1837 | |
029632fb | 1838 | #define for_each_class(class) \ |
c3a340f7 | 1839 | for_class_range(class, sched_class_highest, sched_class_lowest) |
029632fb PZ |
1840 | |
1841 | extern const struct sched_class stop_sched_class; | |
aab03e05 | 1842 | extern const struct sched_class dl_sched_class; |
029632fb PZ |
1843 | extern const struct sched_class rt_sched_class; |
1844 | extern const struct sched_class fair_sched_class; | |
1845 | extern const struct sched_class idle_sched_class; | |
1846 | ||
6e2df058 PZ |
1847 | static inline bool sched_stop_runnable(struct rq *rq) |
1848 | { | |
1849 | return rq->stop && task_on_rq_queued(rq->stop); | |
1850 | } | |
1851 | ||
1852 | static inline bool sched_dl_runnable(struct rq *rq) | |
1853 | { | |
1854 | return rq->dl.dl_nr_running > 0; | |
1855 | } | |
1856 | ||
1857 | static inline bool sched_rt_runnable(struct rq *rq) | |
1858 | { | |
1859 | return rq->rt.rt_queued > 0; | |
1860 | } | |
1861 | ||
1862 | static inline bool sched_fair_runnable(struct rq *rq) | |
1863 | { | |
1864 | return rq->cfs.nr_running > 0; | |
1865 | } | |
029632fb | 1866 | |
5d7d6056 | 1867 | extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf); |
98c2f700 | 1868 | extern struct task_struct *pick_next_task_idle(struct rq *rq); |
5d7d6056 | 1869 | |
029632fb PZ |
1870 | #ifdef CONFIG_SMP |
1871 | ||
63b2ca30 | 1872 | extern void update_group_capacity(struct sched_domain *sd, int cpu); |
b719203b | 1873 | |
7caff66f | 1874 | extern void trigger_load_balance(struct rq *rq); |
029632fb | 1875 | |
c5b28038 PZ |
1876 | extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask); |
1877 | ||
029632fb PZ |
1878 | #endif |
1879 | ||
442bf3aa DL |
1880 | #ifdef CONFIG_CPU_IDLE |
1881 | static inline void idle_set_state(struct rq *rq, | |
1882 | struct cpuidle_state *idle_state) | |
1883 | { | |
1884 | rq->idle_state = idle_state; | |
1885 | } | |
1886 | ||
1887 | static inline struct cpuidle_state *idle_get_state(struct rq *rq) | |
1888 | { | |
9148a3a1 | 1889 | SCHED_WARN_ON(!rcu_read_lock_held()); |
97fb7a0a | 1890 | |
442bf3aa DL |
1891 | return rq->idle_state; |
1892 | } | |
1893 | #else | |
1894 | static inline void idle_set_state(struct rq *rq, | |
1895 | struct cpuidle_state *idle_state) | |
1896 | { | |
1897 | } | |
1898 | ||
1899 | static inline struct cpuidle_state *idle_get_state(struct rq *rq) | |
1900 | { | |
1901 | return NULL; | |
1902 | } | |
1903 | #endif | |
1904 | ||
8663effb SRV |
1905 | extern void schedule_idle(void); |
1906 | ||
029632fb PZ |
1907 | extern void sysrq_sched_debug_show(void); |
1908 | extern void sched_init_granularity(void); | |
1909 | extern void update_max_interval(void); | |
1baca4ce JL |
1910 | |
1911 | extern void init_sched_dl_class(void); | |
029632fb PZ |
1912 | extern void init_sched_rt_class(void); |
1913 | extern void init_sched_fair_class(void); | |
1914 | ||
9059393e VG |
1915 | extern void reweight_task(struct task_struct *p, int prio); |
1916 | ||
8875125e | 1917 | extern void resched_curr(struct rq *rq); |
029632fb PZ |
1918 | extern void resched_cpu(int cpu); |
1919 | ||
1920 | extern struct rt_bandwidth def_rt_bandwidth; | |
1921 | extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime); | |
1922 | ||
332ac17e DF |
1923 | extern struct dl_bandwidth def_dl_bandwidth; |
1924 | extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime); | |
aab03e05 | 1925 | extern void init_dl_task_timer(struct sched_dl_entity *dl_se); |
209a0cbd | 1926 | extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se); |
aab03e05 | 1927 | |
97fb7a0a IM |
1928 | #define BW_SHIFT 20 |
1929 | #define BW_UNIT (1 << BW_SHIFT) | |
1930 | #define RATIO_SHIFT 8 | |
d505b8af HC |
1931 | #define MAX_BW_BITS (64 - BW_SHIFT) |
1932 | #define MAX_BW ((1ULL << MAX_BW_BITS) - 1) | |
332ac17e DF |
1933 | unsigned long to_ratio(u64 period, u64 runtime); |
1934 | ||
540247fb | 1935 | extern void init_entity_runnable_average(struct sched_entity *se); |
d0fe0b9c | 1936 | extern void post_init_entity_util_avg(struct task_struct *p); |
a75cdaa9 | 1937 | |
76d92ac3 FW |
1938 | #ifdef CONFIG_NO_HZ_FULL |
1939 | extern bool sched_can_stop_tick(struct rq *rq); | |
d84b3131 | 1940 | extern int __init sched_tick_offload_init(void); |
76d92ac3 FW |
1941 | |
1942 | /* | |
1943 | * Tick may be needed by tasks in the runqueue depending on their policy and | |
1944 | * requirements. If tick is needed, lets send the target an IPI to kick it out of | |
1945 | * nohz mode if necessary. | |
1946 | */ | |
1947 | static inline void sched_update_tick_dependency(struct rq *rq) | |
1948 | { | |
1949 | int cpu; | |
1950 | ||
1951 | if (!tick_nohz_full_enabled()) | |
1952 | return; | |
1953 | ||
1954 | cpu = cpu_of(rq); | |
1955 | ||
1956 | if (!tick_nohz_full_cpu(cpu)) | |
1957 | return; | |
1958 | ||
1959 | if (sched_can_stop_tick(rq)) | |
1960 | tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED); | |
1961 | else | |
1962 | tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED); | |
1963 | } | |
1964 | #else | |
d84b3131 | 1965 | static inline int sched_tick_offload_init(void) { return 0; } |
76d92ac3 FW |
1966 | static inline void sched_update_tick_dependency(struct rq *rq) { } |
1967 | #endif | |
1968 | ||
72465447 | 1969 | static inline void add_nr_running(struct rq *rq, unsigned count) |
029632fb | 1970 | { |
72465447 KT |
1971 | unsigned prev_nr = rq->nr_running; |
1972 | ||
1973 | rq->nr_running = prev_nr + count; | |
9f3660c2 | 1974 | |
4486edd1 | 1975 | #ifdef CONFIG_SMP |
3e184501 | 1976 | if (prev_nr < 2 && rq->nr_running >= 2) { |
e90c8fe1 VS |
1977 | if (!READ_ONCE(rq->rd->overload)) |
1978 | WRITE_ONCE(rq->rd->overload, 1); | |
4486edd1 | 1979 | } |
3e184501 | 1980 | #endif |
76d92ac3 FW |
1981 | |
1982 | sched_update_tick_dependency(rq); | |
029632fb PZ |
1983 | } |
1984 | ||
72465447 | 1985 | static inline void sub_nr_running(struct rq *rq, unsigned count) |
029632fb | 1986 | { |
72465447 | 1987 | rq->nr_running -= count; |
76d92ac3 FW |
1988 | /* Check if we still need preemption */ |
1989 | sched_update_tick_dependency(rq); | |
029632fb PZ |
1990 | } |
1991 | ||
029632fb PZ |
1992 | extern void activate_task(struct rq *rq, struct task_struct *p, int flags); |
1993 | extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags); | |
1994 | ||
1995 | extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); | |
1996 | ||
029632fb PZ |
1997 | extern const_debug unsigned int sysctl_sched_nr_migrate; |
1998 | extern const_debug unsigned int sysctl_sched_migration_cost; | |
1999 | ||
029632fb PZ |
2000 | #ifdef CONFIG_SCHED_HRTICK |
2001 | ||
2002 | /* | |
2003 | * Use hrtick when: | |
2004 | * - enabled by features | |
2005 | * - hrtimer is actually high res | |
2006 | */ | |
2007 | static inline int hrtick_enabled(struct rq *rq) | |
2008 | { | |
2009 | if (!sched_feat(HRTICK)) | |
2010 | return 0; | |
2011 | if (!cpu_active(cpu_of(rq))) | |
2012 | return 0; | |
2013 | return hrtimer_is_hres_active(&rq->hrtick_timer); | |
2014 | } | |
2015 | ||
2016 | void hrtick_start(struct rq *rq, u64 delay); | |
2017 | ||
b39e66ea MG |
2018 | #else |
2019 | ||
2020 | static inline int hrtick_enabled(struct rq *rq) | |
2021 | { | |
2022 | return 0; | |
2023 | } | |
2024 | ||
029632fb PZ |
2025 | #endif /* CONFIG_SCHED_HRTICK */ |
2026 | ||
1567c3e3 GG |
2027 | #ifndef arch_scale_freq_tick |
2028 | static __always_inline | |
2029 | void arch_scale_freq_tick(void) | |
2030 | { | |
2031 | } | |
2032 | #endif | |
2033 | ||
dfbca41f PZ |
2034 | #ifndef arch_scale_freq_capacity |
2035 | static __always_inline | |
7673c8a4 | 2036 | unsigned long arch_scale_freq_capacity(int cpu) |
dfbca41f PZ |
2037 | { |
2038 | return SCHED_CAPACITY_SCALE; | |
2039 | } | |
2040 | #endif | |
b5b4860d | 2041 | |
029632fb | 2042 | #ifdef CONFIG_SMP |
c1a280b6 | 2043 | #ifdef CONFIG_PREEMPTION |
029632fb PZ |
2044 | |
2045 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2); | |
2046 | ||
2047 | /* | |
2048 | * fair double_lock_balance: Safely acquires both rq->locks in a fair | |
2049 | * way at the expense of forcing extra atomic operations in all | |
2050 | * invocations. This assures that the double_lock is acquired using the | |
2051 | * same underlying policy as the spinlock_t on this architecture, which | |
2052 | * reduces latency compared to the unfair variant below. However, it | |
2053 | * also adds more overhead and therefore may reduce throughput. | |
2054 | */ | |
2055 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
2056 | __releases(this_rq->lock) | |
2057 | __acquires(busiest->lock) | |
2058 | __acquires(this_rq->lock) | |
2059 | { | |
2060 | raw_spin_unlock(&this_rq->lock); | |
2061 | double_rq_lock(this_rq, busiest); | |
2062 | ||
2063 | return 1; | |
2064 | } | |
2065 | ||
2066 | #else | |
2067 | /* | |
2068 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
2069 | * latency by eliminating extra atomic operations when the locks are | |
97fb7a0a IM |
2070 | * already in proper order on entry. This favors lower CPU-ids and will |
2071 | * grant the double lock to lower CPUs over higher ids under contention, | |
029632fb PZ |
2072 | * regardless of entry order into the function. |
2073 | */ | |
2074 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
2075 | __releases(this_rq->lock) | |
2076 | __acquires(busiest->lock) | |
2077 | __acquires(this_rq->lock) | |
2078 | { | |
2079 | int ret = 0; | |
2080 | ||
2081 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { | |
2082 | if (busiest < this_rq) { | |
2083 | raw_spin_unlock(&this_rq->lock); | |
2084 | raw_spin_lock(&busiest->lock); | |
2085 | raw_spin_lock_nested(&this_rq->lock, | |
2086 | SINGLE_DEPTH_NESTING); | |
2087 | ret = 1; | |
2088 | } else | |
2089 | raw_spin_lock_nested(&busiest->lock, | |
2090 | SINGLE_DEPTH_NESTING); | |
2091 | } | |
2092 | return ret; | |
2093 | } | |
2094 | ||
c1a280b6 | 2095 | #endif /* CONFIG_PREEMPTION */ |
029632fb PZ |
2096 | |
2097 | /* | |
2098 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2099 | */ | |
2100 | static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
2101 | { | |
2102 | if (unlikely(!irqs_disabled())) { | |
97fb7a0a | 2103 | /* printk() doesn't work well under rq->lock */ |
029632fb PZ |
2104 | raw_spin_unlock(&this_rq->lock); |
2105 | BUG_ON(1); | |
2106 | } | |
2107 | ||
2108 | return _double_lock_balance(this_rq, busiest); | |
2109 | } | |
2110 | ||
2111 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) | |
2112 | __releases(busiest->lock) | |
2113 | { | |
2114 | raw_spin_unlock(&busiest->lock); | |
2115 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
2116 | } | |
2117 | ||
74602315 PZ |
2118 | static inline void double_lock(spinlock_t *l1, spinlock_t *l2) |
2119 | { | |
2120 | if (l1 > l2) | |
2121 | swap(l1, l2); | |
2122 | ||
2123 | spin_lock(l1); | |
2124 | spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
2125 | } | |
2126 | ||
60e69eed MG |
2127 | static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2) |
2128 | { | |
2129 | if (l1 > l2) | |
2130 | swap(l1, l2); | |
2131 | ||
2132 | spin_lock_irq(l1); | |
2133 | spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
2134 | } | |
2135 | ||
74602315 PZ |
2136 | static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2) |
2137 | { | |
2138 | if (l1 > l2) | |
2139 | swap(l1, l2); | |
2140 | ||
2141 | raw_spin_lock(l1); | |
2142 | raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
2143 | } | |
2144 | ||
029632fb PZ |
2145 | /* |
2146 | * double_rq_lock - safely lock two runqueues | |
2147 | * | |
2148 | * Note this does not disable interrupts like task_rq_lock, | |
2149 | * you need to do so manually before calling. | |
2150 | */ | |
2151 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
2152 | __acquires(rq1->lock) | |
2153 | __acquires(rq2->lock) | |
2154 | { | |
2155 | BUG_ON(!irqs_disabled()); | |
2156 | if (rq1 == rq2) { | |
2157 | raw_spin_lock(&rq1->lock); | |
2158 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2159 | } else { | |
2160 | if (rq1 < rq2) { | |
2161 | raw_spin_lock(&rq1->lock); | |
2162 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
2163 | } else { | |
2164 | raw_spin_lock(&rq2->lock); | |
2165 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
2166 | } | |
2167 | } | |
2168 | } | |
2169 | ||
2170 | /* | |
2171 | * double_rq_unlock - safely unlock two runqueues | |
2172 | * | |
2173 | * Note this does not restore interrupts like task_rq_unlock, | |
2174 | * you need to do so manually after calling. | |
2175 | */ | |
2176 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
2177 | __releases(rq1->lock) | |
2178 | __releases(rq2->lock) | |
2179 | { | |
2180 | raw_spin_unlock(&rq1->lock); | |
2181 | if (rq1 != rq2) | |
2182 | raw_spin_unlock(&rq2->lock); | |
2183 | else | |
2184 | __release(rq2->lock); | |
2185 | } | |
2186 | ||
f2cb1360 IM |
2187 | extern void set_rq_online (struct rq *rq); |
2188 | extern void set_rq_offline(struct rq *rq); | |
2189 | extern bool sched_smp_initialized; | |
2190 | ||
029632fb PZ |
2191 | #else /* CONFIG_SMP */ |
2192 | ||
2193 | /* | |
2194 | * double_rq_lock - safely lock two runqueues | |
2195 | * | |
2196 | * Note this does not disable interrupts like task_rq_lock, | |
2197 | * you need to do so manually before calling. | |
2198 | */ | |
2199 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
2200 | __acquires(rq1->lock) | |
2201 | __acquires(rq2->lock) | |
2202 | { | |
2203 | BUG_ON(!irqs_disabled()); | |
2204 | BUG_ON(rq1 != rq2); | |
2205 | raw_spin_lock(&rq1->lock); | |
2206 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2207 | } | |
2208 | ||
2209 | /* | |
2210 | * double_rq_unlock - safely unlock two runqueues | |
2211 | * | |
2212 | * Note this does not restore interrupts like task_rq_unlock, | |
2213 | * you need to do so manually after calling. | |
2214 | */ | |
2215 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
2216 | __releases(rq1->lock) | |
2217 | __releases(rq2->lock) | |
2218 | { | |
2219 | BUG_ON(rq1 != rq2); | |
2220 | raw_spin_unlock(&rq1->lock); | |
2221 | __release(rq2->lock); | |
2222 | } | |
2223 | ||
2224 | #endif | |
2225 | ||
2226 | extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq); | |
2227 | extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq); | |
6b55c965 SD |
2228 | |
2229 | #ifdef CONFIG_SCHED_DEBUG | |
9469eb01 PZ |
2230 | extern bool sched_debug_enabled; |
2231 | ||
029632fb PZ |
2232 | extern void print_cfs_stats(struct seq_file *m, int cpu); |
2233 | extern void print_rt_stats(struct seq_file *m, int cpu); | |
acb32132 | 2234 | extern void print_dl_stats(struct seq_file *m, int cpu); |
f6a34630 MM |
2235 | extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq); |
2236 | extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq); | |
2237 | extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq); | |
397f2378 SD |
2238 | #ifdef CONFIG_NUMA_BALANCING |
2239 | extern void | |
2240 | show_numa_stats(struct task_struct *p, struct seq_file *m); | |
2241 | extern void | |
2242 | print_numa_stats(struct seq_file *m, int node, unsigned long tsf, | |
2243 | unsigned long tpf, unsigned long gsf, unsigned long gpf); | |
2244 | #endif /* CONFIG_NUMA_BALANCING */ | |
2245 | #endif /* CONFIG_SCHED_DEBUG */ | |
029632fb PZ |
2246 | |
2247 | extern void init_cfs_rq(struct cfs_rq *cfs_rq); | |
07c54f7a AV |
2248 | extern void init_rt_rq(struct rt_rq *rt_rq); |
2249 | extern void init_dl_rq(struct dl_rq *dl_rq); | |
029632fb | 2250 | |
1ee14e6c BS |
2251 | extern void cfs_bandwidth_usage_inc(void); |
2252 | extern void cfs_bandwidth_usage_dec(void); | |
1c792db7 | 2253 | |
3451d024 | 2254 | #ifdef CONFIG_NO_HZ_COMMON |
00357f5e PZ |
2255 | #define NOHZ_BALANCE_KICK_BIT 0 |
2256 | #define NOHZ_STATS_KICK_BIT 1 | |
a22e47a4 | 2257 | |
a22e47a4 | 2258 | #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT) |
b7031a02 PZ |
2259 | #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT) |
2260 | ||
2261 | #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK) | |
1c792db7 SS |
2262 | |
2263 | #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags) | |
20a5c8cc | 2264 | |
00357f5e | 2265 | extern void nohz_balance_exit_idle(struct rq *rq); |
20a5c8cc | 2266 | #else |
00357f5e | 2267 | static inline void nohz_balance_exit_idle(struct rq *rq) { } |
1c792db7 | 2268 | #endif |
73fbec60 | 2269 | |
daec5798 LA |
2270 | |
2271 | #ifdef CONFIG_SMP | |
2272 | static inline | |
2273 | void __dl_update(struct dl_bw *dl_b, s64 bw) | |
2274 | { | |
2275 | struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw); | |
2276 | int i; | |
2277 | ||
2278 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), | |
2279 | "sched RCU must be held"); | |
2280 | for_each_cpu_and(i, rd->span, cpu_active_mask) { | |
2281 | struct rq *rq = cpu_rq(i); | |
2282 | ||
2283 | rq->dl.extra_bw += bw; | |
2284 | } | |
2285 | } | |
2286 | #else | |
2287 | static inline | |
2288 | void __dl_update(struct dl_bw *dl_b, s64 bw) | |
2289 | { | |
2290 | struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw); | |
2291 | ||
2292 | dl->extra_bw += bw; | |
2293 | } | |
2294 | #endif | |
2295 | ||
2296 | ||
73fbec60 | 2297 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
19d23dbf | 2298 | struct irqtime { |
25e2d8c1 | 2299 | u64 total; |
a499a5a1 | 2300 | u64 tick_delta; |
19d23dbf FW |
2301 | u64 irq_start_time; |
2302 | struct u64_stats_sync sync; | |
2303 | }; | |
73fbec60 | 2304 | |
19d23dbf | 2305 | DECLARE_PER_CPU(struct irqtime, cpu_irqtime); |
73fbec60 | 2306 | |
25e2d8c1 FW |
2307 | /* |
2308 | * Returns the irqtime minus the softirq time computed by ksoftirqd. | |
2309 | * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime | |
2310 | * and never move forward. | |
2311 | */ | |
73fbec60 FW |
2312 | static inline u64 irq_time_read(int cpu) |
2313 | { | |
19d23dbf FW |
2314 | struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu); |
2315 | unsigned int seq; | |
2316 | u64 total; | |
73fbec60 FW |
2317 | |
2318 | do { | |
19d23dbf | 2319 | seq = __u64_stats_fetch_begin(&irqtime->sync); |
25e2d8c1 | 2320 | total = irqtime->total; |
19d23dbf | 2321 | } while (__u64_stats_fetch_retry(&irqtime->sync, seq)); |
73fbec60 | 2322 | |
19d23dbf | 2323 | return total; |
73fbec60 | 2324 | } |
73fbec60 | 2325 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
adaf9fcd RW |
2326 | |
2327 | #ifdef CONFIG_CPU_FREQ | |
b10abd0a | 2328 | DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data); |
adaf9fcd RW |
2329 | |
2330 | /** | |
2331 | * cpufreq_update_util - Take a note about CPU utilization changes. | |
12bde33d | 2332 | * @rq: Runqueue to carry out the update for. |
58919e83 | 2333 | * @flags: Update reason flags. |
adaf9fcd | 2334 | * |
58919e83 RW |
2335 | * This function is called by the scheduler on the CPU whose utilization is |
2336 | * being updated. | |
adaf9fcd RW |
2337 | * |
2338 | * It can only be called from RCU-sched read-side critical sections. | |
adaf9fcd RW |
2339 | * |
2340 | * The way cpufreq is currently arranged requires it to evaluate the CPU | |
2341 | * performance state (frequency/voltage) on a regular basis to prevent it from | |
2342 | * being stuck in a completely inadequate performance level for too long. | |
e0367b12 JL |
2343 | * That is not guaranteed to happen if the updates are only triggered from CFS |
2344 | * and DL, though, because they may not be coming in if only RT tasks are | |
2345 | * active all the time (or there are RT tasks only). | |
adaf9fcd | 2346 | * |
e0367b12 JL |
2347 | * As a workaround for that issue, this function is called periodically by the |
2348 | * RT sched class to trigger extra cpufreq updates to prevent it from stalling, | |
adaf9fcd | 2349 | * but that really is a band-aid. Going forward it should be replaced with |
e0367b12 | 2350 | * solutions targeted more specifically at RT tasks. |
adaf9fcd | 2351 | */ |
12bde33d | 2352 | static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) |
adaf9fcd | 2353 | { |
58919e83 RW |
2354 | struct update_util_data *data; |
2355 | ||
674e7541 VK |
2356 | data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data, |
2357 | cpu_of(rq))); | |
58919e83 | 2358 | if (data) |
12bde33d RW |
2359 | data->func(data, rq_clock(rq), flags); |
2360 | } | |
adaf9fcd | 2361 | #else |
12bde33d | 2362 | static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {} |
adaf9fcd | 2363 | #endif /* CONFIG_CPU_FREQ */ |
be53f58f | 2364 | |
982d9cdc | 2365 | #ifdef CONFIG_UCLAMP_TASK |
686516b5 | 2366 | unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id); |
9d20ad7d PB |
2367 | |
2368 | static __always_inline | |
d2b58a28 VS |
2369 | unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util, |
2370 | struct task_struct *p) | |
982d9cdc | 2371 | { |
686516b5 VS |
2372 | unsigned long min_util = READ_ONCE(rq->uclamp[UCLAMP_MIN].value); |
2373 | unsigned long max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value); | |
982d9cdc | 2374 | |
9d20ad7d PB |
2375 | if (p) { |
2376 | min_util = max(min_util, uclamp_eff_value(p, UCLAMP_MIN)); | |
2377 | max_util = max(max_util, uclamp_eff_value(p, UCLAMP_MAX)); | |
2378 | } | |
2379 | ||
982d9cdc PB |
2380 | /* |
2381 | * Since CPU's {min,max}_util clamps are MAX aggregated considering | |
2382 | * RUNNABLE tasks with _different_ clamps, we can end up with an | |
2383 | * inversion. Fix it now when the clamps are applied. | |
2384 | */ | |
2385 | if (unlikely(min_util >= max_util)) | |
2386 | return min_util; | |
2387 | ||
2388 | return clamp(util, min_util, max_util); | |
2389 | } | |
2390 | #else /* CONFIG_UCLAMP_TASK */ | |
d2b58a28 VS |
2391 | static inline |
2392 | unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util, | |
2393 | struct task_struct *p) | |
9d20ad7d PB |
2394 | { |
2395 | return util; | |
2396 | } | |
982d9cdc PB |
2397 | #endif /* CONFIG_UCLAMP_TASK */ |
2398 | ||
9bdcb44e | 2399 | #ifdef arch_scale_freq_capacity |
97fb7a0a IM |
2400 | # ifndef arch_scale_freq_invariant |
2401 | # define arch_scale_freq_invariant() true | |
2402 | # endif | |
2403 | #else | |
2404 | # define arch_scale_freq_invariant() false | |
9bdcb44e | 2405 | #endif |
d4edd662 | 2406 | |
10a35e68 VG |
2407 | #ifdef CONFIG_SMP |
2408 | static inline unsigned long capacity_orig_of(int cpu) | |
2409 | { | |
2410 | return cpu_rq(cpu)->cpu_capacity_orig; | |
2411 | } | |
2412 | #endif | |
2413 | ||
938e5e4b QP |
2414 | /** |
2415 | * enum schedutil_type - CPU utilization type | |
2416 | * @FREQUENCY_UTIL: Utilization used to select frequency | |
2417 | * @ENERGY_UTIL: Utilization used during energy calculation | |
2418 | * | |
2419 | * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time | |
2420 | * need to be aggregated differently depending on the usage made of them. This | |
2421 | * enum is used within schedutil_freq_util() to differentiate the types of | |
2422 | * utilization expected by the callers, and adjust the aggregation accordingly. | |
2423 | */ | |
2424 | enum schedutil_type { | |
2425 | FREQUENCY_UTIL, | |
2426 | ENERGY_UTIL, | |
2427 | }; | |
2428 | ||
af24bde8 | 2429 | #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL |
938e5e4b | 2430 | |
af24bde8 PB |
2431 | unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs, |
2432 | unsigned long max, enum schedutil_type type, | |
2433 | struct task_struct *p); | |
938e5e4b | 2434 | |
8cc90515 | 2435 | static inline unsigned long cpu_bw_dl(struct rq *rq) |
d4edd662 JL |
2436 | { |
2437 | return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT; | |
2438 | } | |
2439 | ||
8cc90515 VG |
2440 | static inline unsigned long cpu_util_dl(struct rq *rq) |
2441 | { | |
2442 | return READ_ONCE(rq->avg_dl.util_avg); | |
2443 | } | |
2444 | ||
d4edd662 JL |
2445 | static inline unsigned long cpu_util_cfs(struct rq *rq) |
2446 | { | |
a07630b8 PB |
2447 | unsigned long util = READ_ONCE(rq->cfs.avg.util_avg); |
2448 | ||
2449 | if (sched_feat(UTIL_EST)) { | |
2450 | util = max_t(unsigned long, util, | |
2451 | READ_ONCE(rq->cfs.avg.util_est.enqueued)); | |
2452 | } | |
2453 | ||
2454 | return util; | |
d4edd662 | 2455 | } |
371bf427 VG |
2456 | |
2457 | static inline unsigned long cpu_util_rt(struct rq *rq) | |
2458 | { | |
dfa444dc | 2459 | return READ_ONCE(rq->avg_rt.util_avg); |
371bf427 | 2460 | } |
938e5e4b | 2461 | #else /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */ |
af24bde8 PB |
2462 | static inline unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs, |
2463 | unsigned long max, enum schedutil_type type, | |
2464 | struct task_struct *p) | |
938e5e4b | 2465 | { |
af24bde8 | 2466 | return 0; |
938e5e4b | 2467 | } |
af24bde8 | 2468 | #endif /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */ |
9033ea11 | 2469 | |
11d4afd4 | 2470 | #ifdef CONFIG_HAVE_SCHED_AVG_IRQ |
9033ea11 VG |
2471 | static inline unsigned long cpu_util_irq(struct rq *rq) |
2472 | { | |
2473 | return rq->avg_irq.util_avg; | |
2474 | } | |
2e62c474 VG |
2475 | |
2476 | static inline | |
2477 | unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max) | |
2478 | { | |
2479 | util *= (max - irq); | |
2480 | util /= max; | |
2481 | ||
2482 | return util; | |
2483 | ||
2484 | } | |
9033ea11 VG |
2485 | #else |
2486 | static inline unsigned long cpu_util_irq(struct rq *rq) | |
2487 | { | |
2488 | return 0; | |
2489 | } | |
2490 | ||
2e62c474 VG |
2491 | static inline |
2492 | unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max) | |
2493 | { | |
2494 | return util; | |
2495 | } | |
794a56eb | 2496 | #endif |
6aa140fa | 2497 | |
531b5c9f | 2498 | #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL) |
f8a696f2 | 2499 | |
6aa140fa | 2500 | #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus))) |
f8a696f2 PZ |
2501 | |
2502 | DECLARE_STATIC_KEY_FALSE(sched_energy_present); | |
2503 | ||
2504 | static inline bool sched_energy_enabled(void) | |
2505 | { | |
2506 | return static_branch_unlikely(&sched_energy_present); | |
2507 | } | |
2508 | ||
2509 | #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */ | |
2510 | ||
6aa140fa | 2511 | #define perf_domain_span(pd) NULL |
f8a696f2 | 2512 | static inline bool sched_energy_enabled(void) { return false; } |
1f74de87 | 2513 | |
f8a696f2 | 2514 | #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */ |
227a4aad MD |
2515 | |
2516 | #ifdef CONFIG_MEMBARRIER | |
2517 | /* | |
2518 | * The scheduler provides memory barriers required by membarrier between: | |
2519 | * - prior user-space memory accesses and store to rq->membarrier_state, | |
2520 | * - store to rq->membarrier_state and following user-space memory accesses. | |
2521 | * In the same way it provides those guarantees around store to rq->curr. | |
2522 | */ | |
2523 | static inline void membarrier_switch_mm(struct rq *rq, | |
2524 | struct mm_struct *prev_mm, | |
2525 | struct mm_struct *next_mm) | |
2526 | { | |
2527 | int membarrier_state; | |
2528 | ||
2529 | if (prev_mm == next_mm) | |
2530 | return; | |
2531 | ||
2532 | membarrier_state = atomic_read(&next_mm->membarrier_state); | |
2533 | if (READ_ONCE(rq->membarrier_state) == membarrier_state) | |
2534 | return; | |
2535 | ||
2536 | WRITE_ONCE(rq->membarrier_state, membarrier_state); | |
2537 | } | |
2538 | #else | |
2539 | static inline void membarrier_switch_mm(struct rq *rq, | |
2540 | struct mm_struct *prev_mm, | |
2541 | struct mm_struct *next_mm) | |
2542 | { | |
2543 | } | |
2544 | #endif | |
52262ee5 MG |
2545 | |
2546 | #ifdef CONFIG_SMP | |
2547 | static inline bool is_per_cpu_kthread(struct task_struct *p) | |
2548 | { | |
2549 | if (!(p->flags & PF_KTHREAD)) | |
2550 | return false; | |
2551 | ||
2552 | if (p->nr_cpus_allowed != 1) | |
2553 | return false; | |
2554 | ||
2555 | return true; | |
2556 | } | |
2557 | #endif | |
b3212fe2 TG |
2558 | |
2559 | void swake_up_all_locked(struct swait_queue_head *q); | |
2560 | void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait); |