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