projects/linux·blob assoc-array-fixes · refs

771 lines · 21.6 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2#include "cgroup-internal.h"
3
4#include <linux/cpumask.h>
5#include <linux/sched/cputime.h>
6
7#include <linux/bpf.h>
8#include <linux/btf.h>
9#include <linux/btf_ids.h>
10
11#include <trace/events/cgroup.h>
12
13static DEFINE_SPINLOCK(rstat_base_lock);
14static DEFINE_PER_CPU(struct llist_head, rstat_backlog_list);
15
16static void cgroup_base_stat_flush(struct cgroup *cgrp, int cpu);
17
18/*
19 * Determines whether a given css can participate in rstat.
20 * css's that are cgroup::self use rstat for base stats.
21 * Other css's associated with a subsystem use rstat only when
22 * they define the ss->css_rstat_flush callback.
23 */
24static inline bool css_uses_rstat(struct cgroup_subsys_state *css)
25{
26 return css_is_self(css) || css->ss->css_rstat_flush != NULL;
27}
28
29static struct css_rstat_cpu *css_rstat_cpu(
30 struct cgroup_subsys_state *css, int cpu)
31{
32 return per_cpu_ptr(css->rstat_cpu, cpu);
33}
34
35static struct cgroup_rstat_base_cpu *cgroup_rstat_base_cpu(
36 struct cgroup *cgrp, int cpu)
37{
38 return per_cpu_ptr(cgrp->rstat_base_cpu, cpu);
39}
40
41static spinlock_t *ss_rstat_lock(struct cgroup_subsys *ss)
42{
43 if (ss)
44 return &ss->rstat_ss_lock;
45
46 return &rstat_base_lock;
47}
48
49static inline struct llist_head *ss_lhead_cpu(struct cgroup_subsys *ss, int cpu)
50{
51 if (ss)
52 return per_cpu_ptr(ss->lhead, cpu);
53 return per_cpu_ptr(&rstat_backlog_list, cpu);
54}
55
56/**
57 * __css_rstat_updated - keep track of updated rstat_cpu
58 * @css: target cgroup subsystem state
59 * @cpu: cpu on which rstat_cpu was updated
60 *
61 * Atomically inserts the css in the ss's llist for the given cpu. This is
62 * reentrant safe i.e. safe against softirq, hardirq and nmi. The ss's llist
63 * will be processed at the flush time to create the update tree.
64 *
65 * NOTE: if the user needs the guarantee that the updater either add itself in
66 * the lockless list or the concurrent flusher flushes its updated stats, a
67 * memory barrier is needed before the call to __css_rstat_updated() i.e. a
68 * barrier after updating the per-cpu stats and before calling
69 * __css_rstat_updated().
70 */
71void __css_rstat_updated(struct cgroup_subsys_state *css, int cpu)
72{
73 struct llist_head *lhead;
74 struct css_rstat_cpu *rstatc;
75 struct llist_node *self;
76
77 /* Prevent access to uninitialized rstat pointers. */
78 if (!css_uses_rstat(css))
79 return;
80
81 lockdep_assert_preemption_disabled();
82
83 /*
84 * The lockless insertion below relies on NMI-safe cmpxchg;
85 * bail out in NMI on archs that don't provide it.
86 */
87 if (!IS_ENABLED(CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG) && in_nmi())
88 return;
89
90 rstatc = css_rstat_cpu(css, cpu);
91 /*
92 * If already on list return. This check is racy and smp_mb() is needed
93 * to pair it with the smp_mb() in css_process_update_tree() if the
94 * guarantee that the updated stats are visible to concurrent flusher is
95 * needed.
96 */
97 if (llist_on_list(&rstatc->lnode))
98 return;
99
100 /*
101 * This function can be renentered by irqs and nmis for the same cgroup
102 * and may try to insert the same per-cpu lnode into the llist. Note
103 * that llist_add() does not protect against such scenarios. In addition
104 * this same per-cpu lnode can be modified through init_llist_node()
105 * from css_rstat_flush() running on a different CPU.
106 *
107 * To protect against such stacked contexts of irqs/nmis, we use the
108 * fact that lnode points to itself when not on a list and then use
109 * try_cmpxchg() to atomically set to NULL to select the winner
110 * which will call llist_add(). The losers can assume the insertion is
111 * successful and the winner will eventually add the per-cpu lnode to
112 * the llist.
113 *
114 * Please note that we can not use this_cpu_cmpxchg() here as on some
115 * archs it is not safe against modifications from multiple CPUs.
116 */
117 self = &rstatc->lnode;
118 if (!try_cmpxchg(&rstatc->lnode.next, &self, NULL))
119 return;
120
121 lhead = ss_lhead_cpu(css->ss, cpu);
122 llist_add(&rstatc->lnode, lhead);
123}
124
125/*
126 * BPF-facing wrapper for __css_rstat_updated(). Validate the caller-provided
127 * CPU before passing it to the internal rstat updater.
128 */
129__bpf_kfunc void css_rstat_updated(struct cgroup_subsys_state *css, int cpu)
130{
131 if (unlikely(cpu < 0 || cpu >= nr_cpu_ids || !cpu_possible(cpu)))
132 return;
133
134 __css_rstat_updated(css, cpu);
135}
136
137static void __css_process_update_tree(struct cgroup_subsys_state *css, int cpu)
138{
139 /* put @css and all ancestors on the corresponding updated lists */
140 while (true) {
141 struct css_rstat_cpu *rstatc = css_rstat_cpu(css, cpu);
142 struct cgroup_subsys_state *parent = css->parent;
143 struct css_rstat_cpu *prstatc;
144
145 /*
146 * Both additions and removals are bottom-up. If a cgroup
147 * is already in the tree, all ancestors are.
148 */
149 if (rstatc->updated_next)
150 break;
151
152 /* Root has no parent to link it to, but mark it busy */
153 if (!parent) {
154 rstatc->updated_next = css;
155 break;
156 }
157
158 prstatc = css_rstat_cpu(parent, cpu);
159 rstatc->updated_next = prstatc->updated_children;
160 prstatc->updated_children = css;
161
162 css = parent;
163 }
164}
165
166static void css_process_update_tree(struct cgroup_subsys *ss, int cpu)
167{
168 struct llist_head *lhead = ss_lhead_cpu(ss, cpu);
169 struct llist_node *lnode;
170
171 while ((lnode = llist_del_first_init(lhead))) {
172 struct css_rstat_cpu *rstatc;
173
174 /*
175 * smp_mb() is needed here (more specifically in between
176 * init_llist_node() and per-cpu stats flushing) if the
177 * guarantee is required by a rstat user where etiher the
178 * updater should add itself on the lockless list or the
179 * flusher flush the stats updated by the updater who have
180 * observed that they are already on the list. The
181 * corresponding barrier pair for this one should be before
182 * __css_rstat_updated() by the user.
183 *
184 * For now, there aren't any such user, so not adding the
185 * barrier here but if such a use-case arise, please add
186 * smp_mb() here.
187 */
188
189 rstatc = container_of(lnode, struct css_rstat_cpu, lnode);
190 __css_process_update_tree(rstatc->owner, cpu);
191 }
192}
193
194/**
195 * css_rstat_push_children - push children css's into the given list
196 * @head: current head of the list (= subtree root)
197 * @child: first child of the root
198 * @cpu: target cpu
199 * Return: A new singly linked list of css's to be flushed
200 *
201 * Iteratively traverse down the css_rstat_cpu updated tree level by
202 * level and push all the parents first before their next level children
203 * into a singly linked list via the rstat_flush_next pointer built from the
204 * tail backward like "pushing" css's into a stack. The root is pushed by
205 * the caller.
206 */
207static struct cgroup_subsys_state *css_rstat_push_children(
208 struct cgroup_subsys_state *head,
209 struct cgroup_subsys_state *child, int cpu)
210{
211 struct cgroup_subsys_state *cnext = child; /* Next head of child css level */
212 struct cgroup_subsys_state *ghead = NULL; /* Head of grandchild css level */
213 struct cgroup_subsys_state *parent, *grandchild;
214 struct css_rstat_cpu *crstatc;
215
216 child->rstat_flush_next = NULL;
217
218 /*
219 * The subsystem rstat lock must be held for the whole duration from
220 * here as the rstat_flush_next list is being constructed to when
221 * it is consumed later in css_rstat_flush().
222 */
223 lockdep_assert_held(ss_rstat_lock(head->ss));
224
225 /*
226 * Notation: -> updated_next pointer
227 * => rstat_flush_next pointer
228 *
229 * Assuming the following sample updated_children lists:
230 * P: C1 -> C2 -> P
231 * C1: G11 -> G12 -> C1
232 * C2: G21 -> G22 -> C2
233 *
234 * After 1st iteration:
235 * head => C2 => C1 => NULL
236 * ghead => G21 => G11 => NULL
237 *
238 * After 2nd iteration:
239 * head => G12 => G11 => G22 => G21 => C2 => C1 => NULL
240 */
241next_level:
242 while (cnext) {
243 child = cnext;
244 cnext = child->rstat_flush_next;
245 parent = child->parent;
246
247 /* updated_next is parent cgroup terminated if !NULL */
248 while (child != parent) {
249 child->rstat_flush_next = head;
250 head = child;
251 crstatc = css_rstat_cpu(child, cpu);
252 grandchild = crstatc->updated_children;
253 if (grandchild != child) {
254 /* Push the grand child to the next level */
255 crstatc->updated_children = child;
256 grandchild->rstat_flush_next = ghead;
257 ghead = grandchild;
258 }
259 child = crstatc->updated_next;
260 crstatc->updated_next = NULL;
261 }
262 }
263
264 if (ghead) {
265 cnext = ghead;
266 ghead = NULL;
267 goto next_level;
268 }
269 return head;
270}
271
272/**
273 * css_rstat_updated_list - build a list of updated css's to be flushed
274 * @root: root of the css subtree to traverse
275 * @cpu: target cpu
276 * Return: A singly linked list of css's to be flushed
277 *
278 * Walks the updated rstat_cpu tree on @cpu from @root. During traversal,
279 * each returned css is unlinked from the updated tree.
280 *
281 * The only ordering guarantee is that, for a parent and a child pair
282 * covered by a given traversal, the child is before its parent in
283 * the list.
284 *
285 * Note that updated_children is self terminated and points to a list of
286 * child css's if not empty. Whereas updated_next is like a sibling link
287 * within the children list and terminated by the parent css. An exception
288 * here is the css root whose updated_next can be self terminated.
289 */
290static struct cgroup_subsys_state *css_rstat_updated_list(
291 struct cgroup_subsys_state *root, int cpu)
292{
293 struct css_rstat_cpu *rstatc = css_rstat_cpu(root, cpu);
294 struct cgroup_subsys_state *head = NULL, *parent, *child;
295
296 css_process_update_tree(root->ss, cpu);
297
298 /* Return NULL if this subtree is not on-list */
299 if (!rstatc->updated_next)
300 return NULL;
301
302 /*
303 * Unlink @root from its parent. As the updated_children list is
304 * singly linked, we have to walk it to find the removal point.
305 */
306 parent = root->parent;
307 if (parent) {
308 struct css_rstat_cpu *prstatc;
309 struct cgroup_subsys_state **nextp;
310
311 prstatc = css_rstat_cpu(parent, cpu);
312 nextp = &prstatc->updated_children;
313 while (*nextp != root) {
314 struct css_rstat_cpu *nrstatc;
315
316 nrstatc = css_rstat_cpu(*nextp, cpu);
317 WARN_ON_ONCE(*nextp == parent);
318 nextp = &nrstatc->updated_next;
319 }
320 *nextp = rstatc->updated_next;
321 }
322
323 rstatc->updated_next = NULL;
324
325 /* Push @root to the list first before pushing the children */
326 head = root;
327 root->rstat_flush_next = NULL;
328 child = rstatc->updated_children;
329 rstatc->updated_children = root;
330 if (child != root)
331 head = css_rstat_push_children(head, child, cpu);
332
333 return head;
334}
335
336/*
337 * A hook for bpf stat collectors to attach to and flush their stats.
338 * Together with providing bpf kfuncs for css_rstat_updated() and
339 * css_rstat_flush(), this enables a complete workflow where bpf progs that
340 * collect cgroup stats can integrate with rstat for efficient flushing.
341 *
342 * A static noinline declaration here could cause the compiler to optimize away
343 * the function. A global noinline declaration will keep the definition, but may
344 * optimize away the callsite. Therefore, __weak is needed to ensure that the
345 * call is still emitted, by telling the compiler that we don't know what the
346 * function might eventually be.
347 */
348
349__bpf_hook_start();
350
351__weak noinline void bpf_rstat_flush(struct cgroup *cgrp,
352 struct cgroup *parent, int cpu)
353{
354}
355
356__bpf_hook_end();
357
358/*
359 * Helper functions for locking.
360 *
361 * This makes it easier to diagnose locking issues and contention in
362 * production environments. The parameter @cpu_in_loop indicate lock
363 * was released and re-taken when collection data from the CPUs. The
364 * value -1 is used when obtaining the main lock else this is the CPU
365 * number processed last.
366 */
367static inline void __css_rstat_lock(struct cgroup_subsys_state *css,
368 int cpu_in_loop)
369 __acquires(ss_rstat_lock(css->ss))
370{
371 struct cgroup *cgrp = css->cgroup;
372 spinlock_t *lock;
373 bool contended;
374
375 lock = ss_rstat_lock(css->ss);
376 contended = !spin_trylock_irq(lock);
377 if (contended) {
378 trace_cgroup_rstat_lock_contended(cgrp, cpu_in_loop, contended);
379 spin_lock_irq(lock);
380 }
381 trace_cgroup_rstat_locked(cgrp, cpu_in_loop, contended);
382}
383
384static inline void __css_rstat_unlock(struct cgroup_subsys_state *css,
385 int cpu_in_loop)
386 __releases(ss_rstat_lock(css->ss))
387{
388 struct cgroup *cgrp = css->cgroup;
389 spinlock_t *lock;
390
391 lock = ss_rstat_lock(css->ss);
392 trace_cgroup_rstat_unlock(cgrp, cpu_in_loop, false);
393 spin_unlock_irq(lock);
394}
395
396/**
397 * css_rstat_flush - flush stats in @css's rstat subtree
398 * @css: target cgroup subsystem state
399 *
400 * Collect all per-cpu stats in @css's subtree into the global counters
401 * and propagate them upwards. After this function returns, all rstat
402 * nodes in the subtree have up-to-date ->stat.
403 *
404 * This also gets all rstat nodes in the subtree including @css off the
405 * ->updated_children lists.
406 *
407 * This function may block.
408 */
409__bpf_kfunc void css_rstat_flush(struct cgroup_subsys_state *css)
410{
411 int cpu;
412 bool is_self = css_is_self(css);
413
414 /*
415 * Since bpf programs can call this function, prevent access to
416 * uninitialized rstat pointers.
417 */
418 if (!css_uses_rstat(css))
419 return;
420
421 might_sleep();
422 for_each_possible_cpu(cpu) {
423 struct cgroup_subsys_state *pos;
424
425 /* Reacquire for each CPU to avoid disabling IRQs too long */
426 __css_rstat_lock(css, cpu);
427 pos = css_rstat_updated_list(css, cpu);
428 for (; pos; pos = pos->rstat_flush_next) {
429 if (is_self) {
430 cgroup_base_stat_flush(pos->cgroup, cpu);
431 bpf_rstat_flush(pos->cgroup,
432 cgroup_parent(pos->cgroup), cpu);
433 } else
434 pos->ss->css_rstat_flush(pos, cpu);
435 }
436 __css_rstat_unlock(css, cpu);
437 if (!cond_resched())
438 cpu_relax();
439 }
440}
441
442int css_rstat_init(struct cgroup_subsys_state *css)
443{
444 struct cgroup *cgrp = css->cgroup;
445 int cpu;
446 bool is_self = css_is_self(css);
447
448 if (is_self) {
449 /* the root cgrp has rstat_base_cpu preallocated */
450 if (!cgrp->rstat_base_cpu) {
451 cgrp->rstat_base_cpu = alloc_percpu(struct cgroup_rstat_base_cpu);
452 if (!cgrp->rstat_base_cpu)
453 return -ENOMEM;
454 }
455 } else if (css->ss->css_rstat_flush == NULL)
456 return 0;
457
458 /* the root cgrp's self css has rstat_cpu preallocated */
459 if (!css->rstat_cpu) {
460 css->rstat_cpu = alloc_percpu(struct css_rstat_cpu);
461 if (!css->rstat_cpu) {
462 if (is_self)
463 free_percpu(cgrp->rstat_base_cpu);
464
465 return -ENOMEM;
466 }
467 }
468
469 /* ->updated_children list is self terminated */
470 for_each_possible_cpu(cpu) {
471 struct css_rstat_cpu *rstatc = css_rstat_cpu(css, cpu);
472
473 rstatc->owner = rstatc->updated_children = css;
474 init_llist_node(&rstatc->lnode);
475
476 if (is_self) {
477 struct cgroup_rstat_base_cpu *rstatbc;
478
479 rstatbc = cgroup_rstat_base_cpu(cgrp, cpu);
480 u64_stats_init(&rstatbc->bsync);
481 }
482 }
483
484 return 0;
485}
486
487void css_rstat_exit(struct cgroup_subsys_state *css)
488{
489 int cpu;
490
491 if (!css_uses_rstat(css))
492 return;
493
494 if (!css->rstat_cpu)
495 return;
496
497 css_rstat_flush(css);
498
499 /* sanity check */
500 for_each_possible_cpu(cpu) {
501 struct css_rstat_cpu *rstatc = css_rstat_cpu(css, cpu);
502
503 if (WARN_ON_ONCE(rstatc->updated_children != css) ||
504 WARN_ON_ONCE(rstatc->updated_next))
505 return;
506 }
507
508 if (css_is_self(css)) {
509 struct cgroup *cgrp = css->cgroup;
510
511 free_percpu(cgrp->rstat_base_cpu);
512 cgrp->rstat_base_cpu = NULL;
513 }
514
515 free_percpu(css->rstat_cpu);
516 css->rstat_cpu = NULL;
517}
518
519/**
520 * ss_rstat_init - subsystem-specific rstat initialization
521 * @ss: target subsystem
522 *
523 * If @ss is NULL, the static locks associated with the base stats
524 * are initialized. If @ss is non-NULL, the subsystem-specific locks
525 * are initialized.
526 */
527int __init ss_rstat_init(struct cgroup_subsys *ss)
528{
529 int cpu;
530
531 if (ss) {
532 ss->lhead = alloc_percpu(struct llist_head);
533 if (!ss->lhead)
534 return -ENOMEM;
535 }
536
537 spin_lock_init(ss_rstat_lock(ss));
538 for_each_possible_cpu(cpu)
539 init_llist_head(ss_lhead_cpu(ss, cpu));
540
541 return 0;
542}
543
544/*
545 * Functions for cgroup basic resource statistics implemented on top of
546 * rstat.
547 */
548static void cgroup_base_stat_add(struct cgroup_base_stat *dst_bstat,
549 struct cgroup_base_stat *src_bstat)
550{
551 dst_bstat->cputime.utime += src_bstat->cputime.utime;
552 dst_bstat->cputime.stime += src_bstat->cputime.stime;
553 dst_bstat->cputime.sum_exec_runtime += src_bstat->cputime.sum_exec_runtime;
554#ifdef CONFIG_SCHED_CORE
555 dst_bstat->forceidle_sum += src_bstat->forceidle_sum;
556#endif
557 dst_bstat->ntime += src_bstat->ntime;
558}
559
560static void cgroup_base_stat_sub(struct cgroup_base_stat *dst_bstat,
561 struct cgroup_base_stat *src_bstat)
562{
563 dst_bstat->cputime.utime -= src_bstat->cputime.utime;
564 dst_bstat->cputime.stime -= src_bstat->cputime.stime;
565 dst_bstat->cputime.sum_exec_runtime -= src_bstat->cputime.sum_exec_runtime;
566#ifdef CONFIG_SCHED_CORE
567 dst_bstat->forceidle_sum -= src_bstat->forceidle_sum;
568#endif
569 dst_bstat->ntime -= src_bstat->ntime;
570}
571
572static void cgroup_base_stat_flush(struct cgroup *cgrp, int cpu)
573{
574 struct cgroup_rstat_base_cpu *rstatbc = cgroup_rstat_base_cpu(cgrp, cpu);
575 struct cgroup *parent = cgroup_parent(cgrp);
576 struct cgroup_rstat_base_cpu *prstatbc;
577 struct cgroup_base_stat delta;
578 unsigned seq;
579
580 /* Root-level stats are sourced from system-wide CPU stats */
581 if (!parent)
582 return;
583
584 /* fetch the current per-cpu values */
585 do {
586 seq = __u64_stats_fetch_begin(&rstatbc->bsync);
587 delta = rstatbc->bstat;
588 } while (__u64_stats_fetch_retry(&rstatbc->bsync, seq));
589
590 /* propagate per-cpu delta to cgroup and per-cpu global statistics */
591 cgroup_base_stat_sub(&delta, &rstatbc->last_bstat);
592 cgroup_base_stat_add(&cgrp->bstat, &delta);
593 cgroup_base_stat_add(&rstatbc->last_bstat, &delta);
594 cgroup_base_stat_add(&rstatbc->subtree_bstat, &delta);
595
596 /* propagate cgroup and per-cpu global delta to parent (unless that's root) */
597 if (cgroup_parent(parent)) {
598 delta = cgrp->bstat;
599 cgroup_base_stat_sub(&delta, &cgrp->last_bstat);
600 cgroup_base_stat_add(&parent->bstat, &delta);
601 cgroup_base_stat_add(&cgrp->last_bstat, &delta);
602
603 delta = rstatbc->subtree_bstat;
604 prstatbc = cgroup_rstat_base_cpu(parent, cpu);
605 cgroup_base_stat_sub(&delta, &rstatbc->last_subtree_bstat);
606 cgroup_base_stat_add(&prstatbc->subtree_bstat, &delta);
607 cgroup_base_stat_add(&rstatbc->last_subtree_bstat, &delta);
608 }
609}
610
611static struct cgroup_rstat_base_cpu *
612cgroup_base_stat_cputime_account_begin(struct cgroup *cgrp, unsigned long *flags)
613{
614 struct cgroup_rstat_base_cpu *rstatbc;
615
616 rstatbc = get_cpu_ptr(cgrp->rstat_base_cpu);
617 *flags = u64_stats_update_begin_irqsave(&rstatbc->bsync);
618 return rstatbc;
619}
620
621static void cgroup_base_stat_cputime_account_end(struct cgroup *cgrp,
622 struct cgroup_rstat_base_cpu *rstatbc,
623 unsigned long flags)
624{
625 u64_stats_update_end_irqrestore(&rstatbc->bsync, flags);
626 __css_rstat_updated(&cgrp->self, smp_processor_id());
627 put_cpu_ptr(rstatbc);
628}
629
630void __cgroup_account_cputime(struct cgroup *cgrp, u64 delta_exec)
631{
632 struct cgroup_rstat_base_cpu *rstatbc;
633 unsigned long flags;
634
635 rstatbc = cgroup_base_stat_cputime_account_begin(cgrp, &flags);
636 rstatbc->bstat.cputime.sum_exec_runtime += delta_exec;
637 cgroup_base_stat_cputime_account_end(cgrp, rstatbc, flags);
638}
639
640void __cgroup_account_cputime_field(struct cgroup *cgrp,
641 enum cpu_usage_stat index, u64 delta_exec)
642{
643 struct cgroup_rstat_base_cpu *rstatbc;
644 unsigned long flags;
645
646 rstatbc = cgroup_base_stat_cputime_account_begin(cgrp, &flags);
647
648 switch (index) {
649 case CPUTIME_NICE:
650 rstatbc->bstat.ntime += delta_exec;
651 fallthrough;
652 case CPUTIME_USER:
653 rstatbc->bstat.cputime.utime += delta_exec;
654 break;
655 case CPUTIME_SYSTEM:
656 case CPUTIME_IRQ:
657 case CPUTIME_SOFTIRQ:
658 rstatbc->bstat.cputime.stime += delta_exec;
659 break;
660#ifdef CONFIG_SCHED_CORE
661 case CPUTIME_FORCEIDLE:
662 rstatbc->bstat.forceidle_sum += delta_exec;
663 break;
664#endif
665 default:
666 break;
667 }
668
669 cgroup_base_stat_cputime_account_end(cgrp, rstatbc, flags);
670}
671
672/*
673 * compute the cputime for the root cgroup by getting the per cpu data
674 * at a global level, then categorizing the fields in a manner consistent
675 * with how it is done by __cgroup_account_cputime_field for each bit of
676 * cpu time attributed to a cgroup.
677 */
678static void root_cgroup_cputime(struct cgroup_base_stat *bstat)
679{
680 struct task_cputime *cputime = &bstat->cputime;
681 int i;
682
683 memset(bstat, 0, sizeof(*bstat));
684 for_each_possible_cpu(i) {
685 struct kernel_cpustat kcpustat;
686 u64 *cpustat = kcpustat.cpustat;
687 u64 user = 0;
688 u64 sys = 0;
689
690 kcpustat_cpu_fetch(&kcpustat, i);
691
692 user += cpustat[CPUTIME_USER];
693 user += cpustat[CPUTIME_NICE];
694 cputime->utime += user;
695
696 sys += cpustat[CPUTIME_SYSTEM];
697 sys += cpustat[CPUTIME_IRQ];
698 sys += cpustat[CPUTIME_SOFTIRQ];
699 cputime->stime += sys;
700
701 cputime->sum_exec_runtime += user;
702 cputime->sum_exec_runtime += sys;
703
704#ifdef CONFIG_SCHED_CORE
705 bstat->forceidle_sum += cpustat[CPUTIME_FORCEIDLE];
706#endif
707 bstat->ntime += cpustat[CPUTIME_NICE];
708 }
709}
710
711
712static void cgroup_force_idle_show(struct seq_file *seq, struct cgroup_base_stat *bstat)
713{
714#ifdef CONFIG_SCHED_CORE
715 u64 forceidle_time = bstat->forceidle_sum;
716
717 do_div(forceidle_time, NSEC_PER_USEC);
718 seq_printf(seq, "core_sched.force_idle_usec %llu\n", forceidle_time);
719#endif
720}
721
722void cgroup_base_stat_cputime_show(struct seq_file *seq)
723{
724 struct cgroup *cgrp = seq_css(seq)->cgroup;
725 struct cgroup_base_stat bstat;
726
727 if (cgroup_parent(cgrp)) {
728 css_rstat_flush(&cgrp->self);
729 __css_rstat_lock(&cgrp->self, -1);
730 bstat = cgrp->bstat;
731 cputime_adjust(&cgrp->bstat.cputime, &cgrp->prev_cputime,
732 &bstat.cputime.utime, &bstat.cputime.stime);
733 __css_rstat_unlock(&cgrp->self, -1);
734 } else {
735 root_cgroup_cputime(&bstat);
736 }
737
738 do_div(bstat.cputime.sum_exec_runtime, NSEC_PER_USEC);
739 do_div(bstat.cputime.utime, NSEC_PER_USEC);
740 do_div(bstat.cputime.stime, NSEC_PER_USEC);
741 do_div(bstat.ntime, NSEC_PER_USEC);
742
743 seq_printf(seq, "usage_usec %llu\n"
744 "user_usec %llu\n"
745 "system_usec %llu\n"
746 "nice_usec %llu\n",
747 bstat.cputime.sum_exec_runtime,
748 bstat.cputime.utime,
749 bstat.cputime.stime,
750 bstat.ntime);
751
752 cgroup_force_idle_show(seq, &bstat);
753}
754
755/* Add bpf kfuncs for css_rstat_updated() and css_rstat_flush() */
756BTF_KFUNCS_START(bpf_rstat_kfunc_ids)
757BTF_ID_FLAGS(func, css_rstat_updated)
758BTF_ID_FLAGS(func, css_rstat_flush, KF_SLEEPABLE)
759BTF_KFUNCS_END(bpf_rstat_kfunc_ids)
760
761static const struct btf_kfunc_id_set bpf_rstat_kfunc_set = {
762 .owner = THIS_MODULE,
763 .set = &bpf_rstat_kfunc_ids,
764};
765
766static int __init bpf_rstat_kfunc_init(void)
767{
768 return register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING,
769 &bpf_rstat_kfunc_set);
770}
771late_initcall(bpf_rstat_kfunc_init);