]>
git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - kernel/cgroup/rstat.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 #include "cgroup-internal.h"
4 #include <linux/sched/cputime.h>
6 static DEFINE_SPINLOCK(cgroup_rstat_lock
);
7 static DEFINE_PER_CPU(raw_spinlock_t
, cgroup_rstat_cpu_lock
);
9 static void cgroup_base_stat_flush(struct cgroup
*cgrp
, int cpu
);
11 static struct cgroup_rstat_cpu
*cgroup_rstat_cpu(struct cgroup
*cgrp
, int cpu
)
13 return per_cpu_ptr(cgrp
->rstat_cpu
, cpu
);
17 * cgroup_rstat_updated - keep track of updated rstat_cpu
18 * @cgrp: target cgroup
19 * @cpu: cpu on which rstat_cpu was updated
21 * @cgrp's rstat_cpu on @cpu was updated. Put it on the parent's matching
22 * rstat_cpu->updated_children list. See the comment on top of
23 * cgroup_rstat_cpu definition for details.
25 void cgroup_rstat_updated(struct cgroup
*cgrp
, int cpu
)
27 raw_spinlock_t
*cpu_lock
= per_cpu_ptr(&cgroup_rstat_cpu_lock
, cpu
);
28 struct cgroup
*parent
;
31 /* nothing to do for root */
32 if (!cgroup_parent(cgrp
))
36 * Speculative already-on-list test. This may race leading to
37 * temporary inaccuracies, which is fine.
39 * Because @parent's updated_children is terminated with @parent
40 * instead of NULL, we can tell whether @cgrp is on the list by
41 * testing the next pointer for NULL.
43 if (cgroup_rstat_cpu(cgrp
, cpu
)->updated_next
)
46 raw_spin_lock_irqsave(cpu_lock
, flags
);
48 /* put @cgrp and all ancestors on the corresponding updated lists */
49 for (parent
= cgroup_parent(cgrp
); parent
;
50 cgrp
= parent
, parent
= cgroup_parent(cgrp
)) {
51 struct cgroup_rstat_cpu
*rstatc
= cgroup_rstat_cpu(cgrp
, cpu
);
52 struct cgroup_rstat_cpu
*prstatc
= cgroup_rstat_cpu(parent
, cpu
);
55 * Both additions and removals are bottom-up. If a cgroup
56 * is already in the tree, all ancestors are.
58 if (rstatc
->updated_next
)
61 rstatc
->updated_next
= prstatc
->updated_children
;
62 prstatc
->updated_children
= cgrp
;
65 raw_spin_unlock_irqrestore(cpu_lock
, flags
);
69 * cgroup_rstat_cpu_pop_updated - iterate and dismantle rstat_cpu updated tree
70 * @pos: current position
71 * @root: root of the tree to traversal
74 * Walks the udpated rstat_cpu tree on @cpu from @root. %NULL @pos starts
75 * the traversal and %NULL return indicates the end. During traversal,
76 * each returned cgroup is unlinked from the tree. Must be called with the
77 * matching cgroup_rstat_cpu_lock held.
79 * The only ordering guarantee is that, for a parent and a child pair
80 * covered by a given traversal, if a child is visited, its parent is
81 * guaranteed to be visited afterwards.
83 static struct cgroup
*cgroup_rstat_cpu_pop_updated(struct cgroup
*pos
,
84 struct cgroup
*root
, int cpu
)
86 struct cgroup_rstat_cpu
*rstatc
;
92 * We're gonna walk down to the first leaf and visit/remove it. We
93 * can pick whatever unvisited node as the starting point.
98 pos
= cgroup_parent(pos
);
100 /* walk down to the first leaf */
102 rstatc
= cgroup_rstat_cpu(pos
, cpu
);
103 if (rstatc
->updated_children
== pos
)
105 pos
= rstatc
->updated_children
;
109 * Unlink @pos from the tree. As the updated_children list is
110 * singly linked, we have to walk it to find the removal point.
111 * However, due to the way we traverse, @pos will be the first
112 * child in most cases. The only exception is @root.
114 if (rstatc
->updated_next
) {
115 struct cgroup
*parent
= cgroup_parent(pos
);
116 struct cgroup_rstat_cpu
*prstatc
= cgroup_rstat_cpu(parent
, cpu
);
117 struct cgroup_rstat_cpu
*nrstatc
;
118 struct cgroup
**nextp
;
120 nextp
= &prstatc
->updated_children
;
122 nrstatc
= cgroup_rstat_cpu(*nextp
, cpu
);
126 WARN_ON_ONCE(*nextp
== parent
);
127 nextp
= &nrstatc
->updated_next
;
130 *nextp
= rstatc
->updated_next
;
131 rstatc
->updated_next
= NULL
;
136 /* only happens for @root */
140 /* see cgroup_rstat_flush() */
141 static void cgroup_rstat_flush_locked(struct cgroup
*cgrp
, bool may_sleep
)
142 __releases(&cgroup_rstat_lock
) __acquires(&cgroup_rstat_lock
)
146 lockdep_assert_held(&cgroup_rstat_lock
);
148 for_each_possible_cpu(cpu
) {
149 raw_spinlock_t
*cpu_lock
= per_cpu_ptr(&cgroup_rstat_cpu_lock
,
151 struct cgroup
*pos
= NULL
;
153 raw_spin_lock(cpu_lock
);
154 while ((pos
= cgroup_rstat_cpu_pop_updated(pos
, cgrp
, cpu
))) {
155 struct cgroup_subsys_state
*css
;
157 cgroup_base_stat_flush(pos
, cpu
);
160 list_for_each_entry_rcu(css
, &pos
->rstat_css_list
,
162 css
->ss
->css_rstat_flush(css
, cpu
);
165 raw_spin_unlock(cpu_lock
);
167 /* if @may_sleep, play nice and yield if necessary */
168 if (may_sleep
&& (need_resched() ||
169 spin_needbreak(&cgroup_rstat_lock
))) {
170 spin_unlock_irq(&cgroup_rstat_lock
);
173 spin_lock_irq(&cgroup_rstat_lock
);
179 * cgroup_rstat_flush - flush stats in @cgrp's subtree
180 * @cgrp: target cgroup
182 * Collect all per-cpu stats in @cgrp's subtree into the global counters
183 * and propagate them upwards. After this function returns, all cgroups in
184 * the subtree have up-to-date ->stat.
186 * This also gets all cgroups in the subtree including @cgrp off the
187 * ->updated_children lists.
189 * This function may block.
191 void cgroup_rstat_flush(struct cgroup
*cgrp
)
195 spin_lock_irq(&cgroup_rstat_lock
);
196 cgroup_rstat_flush_locked(cgrp
, true);
197 spin_unlock_irq(&cgroup_rstat_lock
);
201 * cgroup_rstat_flush_irqsafe - irqsafe version of cgroup_rstat_flush()
202 * @cgrp: target cgroup
204 * This function can be called from any context.
206 void cgroup_rstat_flush_irqsafe(struct cgroup
*cgrp
)
210 spin_lock_irqsave(&cgroup_rstat_lock
, flags
);
211 cgroup_rstat_flush_locked(cgrp
, false);
212 spin_unlock_irqrestore(&cgroup_rstat_lock
, flags
);
216 * cgroup_rstat_flush_begin - flush stats in @cgrp's subtree and hold
217 * @cgrp: target cgroup
219 * Flush stats in @cgrp's subtree and prevent further flushes. Must be
220 * paired with cgroup_rstat_flush_release().
222 * This function may block.
224 void cgroup_rstat_flush_hold(struct cgroup
*cgrp
)
225 __acquires(&cgroup_rstat_lock
)
228 spin_lock_irq(&cgroup_rstat_lock
);
229 cgroup_rstat_flush_locked(cgrp
, true);
233 * cgroup_rstat_flush_release - release cgroup_rstat_flush_hold()
235 void cgroup_rstat_flush_release(void)
236 __releases(&cgroup_rstat_lock
)
238 spin_unlock_irq(&cgroup_rstat_lock
);
241 int cgroup_rstat_init(struct cgroup
*cgrp
)
245 /* the root cgrp has rstat_cpu preallocated */
246 if (!cgrp
->rstat_cpu
) {
247 cgrp
->rstat_cpu
= alloc_percpu(struct cgroup_rstat_cpu
);
248 if (!cgrp
->rstat_cpu
)
252 /* ->updated_children list is self terminated */
253 for_each_possible_cpu(cpu
) {
254 struct cgroup_rstat_cpu
*rstatc
= cgroup_rstat_cpu(cgrp
, cpu
);
256 rstatc
->updated_children
= cgrp
;
257 u64_stats_init(&rstatc
->bsync
);
263 void cgroup_rstat_exit(struct cgroup
*cgrp
)
267 cgroup_rstat_flush(cgrp
);
270 for_each_possible_cpu(cpu
) {
271 struct cgroup_rstat_cpu
*rstatc
= cgroup_rstat_cpu(cgrp
, cpu
);
273 if (WARN_ON_ONCE(rstatc
->updated_children
!= cgrp
) ||
274 WARN_ON_ONCE(rstatc
->updated_next
))
278 free_percpu(cgrp
->rstat_cpu
);
279 cgrp
->rstat_cpu
= NULL
;
282 void __init
cgroup_rstat_boot(void)
286 for_each_possible_cpu(cpu
)
287 raw_spin_lock_init(per_cpu_ptr(&cgroup_rstat_cpu_lock
, cpu
));
289 BUG_ON(cgroup_rstat_init(&cgrp_dfl_root
.cgrp
));
293 * Functions for cgroup basic resource statistics implemented on top of
296 static void cgroup_base_stat_add(struct cgroup_base_stat
*dst_bstat
,
297 struct cgroup_base_stat
*src_bstat
)
299 dst_bstat
->cputime
.utime
+= src_bstat
->cputime
.utime
;
300 dst_bstat
->cputime
.stime
+= src_bstat
->cputime
.stime
;
301 dst_bstat
->cputime
.sum_exec_runtime
+= src_bstat
->cputime
.sum_exec_runtime
;
304 static void cgroup_base_stat_sub(struct cgroup_base_stat
*dst_bstat
,
305 struct cgroup_base_stat
*src_bstat
)
307 dst_bstat
->cputime
.utime
-= src_bstat
->cputime
.utime
;
308 dst_bstat
->cputime
.stime
-= src_bstat
->cputime
.stime
;
309 dst_bstat
->cputime
.sum_exec_runtime
-= src_bstat
->cputime
.sum_exec_runtime
;
312 static void cgroup_base_stat_flush(struct cgroup
*cgrp
, int cpu
)
314 struct cgroup
*parent
= cgroup_parent(cgrp
);
315 struct cgroup_rstat_cpu
*rstatc
= cgroup_rstat_cpu(cgrp
, cpu
);
316 struct cgroup_base_stat cur
, delta
;
319 /* fetch the current per-cpu values */
321 seq
= __u64_stats_fetch_begin(&rstatc
->bsync
);
322 cur
.cputime
= rstatc
->bstat
.cputime
;
323 } while (__u64_stats_fetch_retry(&rstatc
->bsync
, seq
));
325 /* propagate percpu delta to global */
327 cgroup_base_stat_sub(&delta
, &rstatc
->last_bstat
);
328 cgroup_base_stat_add(&cgrp
->bstat
, &delta
);
329 cgroup_base_stat_add(&rstatc
->last_bstat
, &delta
);
331 /* propagate global delta to parent */
334 cgroup_base_stat_sub(&delta
, &cgrp
->last_bstat
);
335 cgroup_base_stat_add(&parent
->bstat
, &delta
);
336 cgroup_base_stat_add(&cgrp
->last_bstat
, &delta
);
340 static struct cgroup_rstat_cpu
*
341 cgroup_base_stat_cputime_account_begin(struct cgroup
*cgrp
)
343 struct cgroup_rstat_cpu
*rstatc
;
345 rstatc
= get_cpu_ptr(cgrp
->rstat_cpu
);
346 u64_stats_update_begin(&rstatc
->bsync
);
350 static void cgroup_base_stat_cputime_account_end(struct cgroup
*cgrp
,
351 struct cgroup_rstat_cpu
*rstatc
)
353 u64_stats_update_end(&rstatc
->bsync
);
354 cgroup_rstat_updated(cgrp
, smp_processor_id());
358 void __cgroup_account_cputime(struct cgroup
*cgrp
, u64 delta_exec
)
360 struct cgroup_rstat_cpu
*rstatc
;
362 rstatc
= cgroup_base_stat_cputime_account_begin(cgrp
);
363 rstatc
->bstat
.cputime
.sum_exec_runtime
+= delta_exec
;
364 cgroup_base_stat_cputime_account_end(cgrp
, rstatc
);
367 void __cgroup_account_cputime_field(struct cgroup
*cgrp
,
368 enum cpu_usage_stat index
, u64 delta_exec
)
370 struct cgroup_rstat_cpu
*rstatc
;
372 rstatc
= cgroup_base_stat_cputime_account_begin(cgrp
);
377 rstatc
->bstat
.cputime
.utime
+= delta_exec
;
381 case CPUTIME_SOFTIRQ
:
382 rstatc
->bstat
.cputime
.stime
+= delta_exec
;
388 cgroup_base_stat_cputime_account_end(cgrp
, rstatc
);
392 * compute the cputime for the root cgroup by getting the per cpu data
393 * at a global level, then categorizing the fields in a manner consistent
394 * with how it is done by __cgroup_account_cputime_field for each bit of
395 * cpu time attributed to a cgroup.
397 static void root_cgroup_cputime(struct task_cputime
*cputime
)
403 cputime
->sum_exec_runtime
= 0;
404 for_each_possible_cpu(i
) {
405 struct kernel_cpustat kcpustat
;
406 u64
*cpustat
= kcpustat
.cpustat
;
410 kcpustat_cpu_fetch(&kcpustat
, i
);
412 user
+= cpustat
[CPUTIME_USER
];
413 user
+= cpustat
[CPUTIME_NICE
];
414 cputime
->utime
+= user
;
416 sys
+= cpustat
[CPUTIME_SYSTEM
];
417 sys
+= cpustat
[CPUTIME_IRQ
];
418 sys
+= cpustat
[CPUTIME_SOFTIRQ
];
419 cputime
->stime
+= sys
;
421 cputime
->sum_exec_runtime
+= user
;
422 cputime
->sum_exec_runtime
+= sys
;
423 cputime
->sum_exec_runtime
+= cpustat
[CPUTIME_STEAL
];
424 cputime
->sum_exec_runtime
+= cpustat
[CPUTIME_GUEST
];
425 cputime
->sum_exec_runtime
+= cpustat
[CPUTIME_GUEST_NICE
];
429 void cgroup_base_stat_cputime_show(struct seq_file
*seq
)
431 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
432 u64 usage
, utime
, stime
;
433 struct task_cputime cputime
;
435 if (cgroup_parent(cgrp
)) {
436 cgroup_rstat_flush_hold(cgrp
);
437 usage
= cgrp
->bstat
.cputime
.sum_exec_runtime
;
438 cputime_adjust(&cgrp
->bstat
.cputime
, &cgrp
->prev_cputime
,
440 cgroup_rstat_flush_release();
442 root_cgroup_cputime(&cputime
);
443 usage
= cputime
.sum_exec_runtime
;
444 utime
= cputime
.utime
;
445 stime
= cputime
.stime
;
448 do_div(usage
, NSEC_PER_USEC
);
449 do_div(utime
, NSEC_PER_USEC
);
450 do_div(stime
, NSEC_PER_USEC
);
452 seq_printf(seq
, "usage_usec %llu\n"
454 "system_usec %llu\n",
455 usage
, utime
, stime
);