1 // SPDX-License-Identifier: GPL-2.0-or-later
3 #include "cgroup-internal.h"
4 #include "cpuset-internal.h"
15  * Frequency meter - How fast is some event occurring?
19  *   fmeter_init() - initialize a frequency meter.
20  *   fmeter_markevent() - called each time the event happens.
21  *   fmeter_getrate() - returns the recent rate of such events.
22  *   fmeter_update() - internal routine used to update fmeter.
24  * A common data structure is passed to each of these routines,
25  * which is used to keep track of the state required to manage the
28  * The filter works on the number of events marked per unit time.
29  * The filter is single-pole low-pass recursive (IIR).  The time unit
30  * is 1 second.  Arithmetic is done using 32-bit integers scaled to
31  * simulate 3 decimal digits of precision (multiplied by 1000).
33  * With an FM_COEF of 933, and a time base of 1 second, the filter
34  * has a half-life of 10 seconds, meaning that if the events quit
43  * Limit the count of unprocessed events to FM_MAXCNT, so as to avoid
51  * to a value N*1000, where N is the rate of events per second.
59 #define FM_COEF 933		/* coefficient for half-life of 10 secs */
67 	fmp->cnt = 0;  in fmeter_init()
68 	fmp->val = 0;  in fmeter_init()
69 	fmp->time = 0;  in fmeter_init()
70 	spin_lock_init(&fmp->lock);  in fmeter_init()
73 /* Internal meter update - process cnt events and update value */
80 	ticks = now - fmp->time;  in fmeter_update()
86 	while (ticks-- > 0)  in fmeter_update()
87 		fmp->val = (FM_COEF * fmp->val) / FM_SCALE;  in fmeter_update()
88 	fmp->time = now;  in fmeter_update()
90 	fmp->val += ((FM_SCALE - FM_COEF) * fmp->cnt) / FM_SCALE;  in fmeter_update()
91 	fmp->cnt = 0;  in fmeter_update()
97 	spin_lock(&fmp->lock);  in fmeter_markevent()
99 	fmp->cnt = min(FM_MAXCNT, fmp->cnt + FM_SCALE);  in fmeter_markevent()
100 	spin_unlock(&fmp->lock);  in fmeter_markevent()
108 	spin_lock(&fmp->lock);  in fmeter_getrate()
110 	val = fmp->val;  in fmeter_getrate()
111 	spin_unlock(&fmp->lock);  in fmeter_getrate()
116  * Collection of memory_pressure is suppressed unless
124  * __cpuset_memory_pressure_bump - keep stats of per-cpuset reclaims.
126  * Keep a running average of the rate of synchronous (direct)
135  * Display to user space in the per-cpuset read-only file
137  * representing the recent rate of entry into the synchronous
144 	fmeter_markevent(&task_cs(current)->fmeter);  in __cpuset_memory_pressure_bump()
151 	if (val < -1 || val > sched_domain_level_max + 1)  in update_relax_domain_level()
152 		return -EINVAL;  in update_relax_domain_level()
155 	if (val != cs->relax_domain_level) {  in update_relax_domain_level()
156 		cs->relax_domain_level = val;  in update_relax_domain_level()
157 		if (!cpumask_empty(cs->cpus_allowed) &&  in update_relax_domain_level()
169 	cpuset_filetype_t type = cft->private;  in cpuset_write_s64()
170 	int retval = -ENODEV;  in cpuset_write_s64()
179 		pr_info_once("cpuset.%s is deprecated\n", cft->name);  in cpuset_write_s64()
183 		retval = -EINVAL;  in cpuset_write_s64()
195 	cpuset_filetype_t type = cft->private;  in cpuset_read_s64()
199 		return cs->relax_domain_level;  in cpuset_read_s64()
232  * cpuset1_update_tasks_flags - update the spread flags of tasks in the cpuset.
235  * Iterate through each task of @cs updating its spread flags.  As this
244 	css_task_iter_start(&cs->css, 0, &it);  in cpuset1_update_tasks_flags()
255  * cpuset to its next-highest non-empty parent.
262 	 * Find its next-highest non-empty parent, (top cpuset  in remove_tasks_in_empty_cpuset()
266 	while (cpumask_empty(parent->cpus_allowed) ||  in remove_tasks_in_empty_cpuset()
267 			nodes_empty(parent->mems_allowed))  in remove_tasks_in_empty_cpuset()
270 	if (cgroup_transfer_tasks(parent->css.cgroup, cs->css.cgroup)) {  in remove_tasks_in_empty_cpuset()
271 		pr_err("cpuset: failed to transfer tasks out of empty cpuset ");  in remove_tasks_in_empty_cpuset()
272 		pr_cont_cgroup_name(cs->css.cgroup);  in remove_tasks_in_empty_cpuset()
282 	remove_tasks_in_empty_cpuset(s->cs);  in cpuset_migrate_tasks_workfn()
283 	css_put(&s->cs->css);  in cpuset_migrate_tasks_workfn()
294 	cpumask_copy(cs->cpus_allowed, new_cpus);  in cpuset1_hotplug_update_tasks()
295 	cpumask_copy(cs->effective_cpus, new_cpus);  in cpuset1_hotplug_update_tasks()
296 	cs->mems_allowed = *new_mems;  in cpuset1_hotplug_update_tasks()
297 	cs->effective_mems = *new_mems;  in cpuset1_hotplug_update_tasks()
304 	if (cpus_updated && !cpumask_empty(cs->cpus_allowed))  in cpuset1_hotplug_update_tasks()
306 	if (mems_updated && !nodes_empty(cs->mems_allowed))  in cpuset1_hotplug_update_tasks()
309 	is_empty = cpumask_empty(cs->cpus_allowed) ||  in cpuset1_hotplug_update_tasks()
310 		   nodes_empty(cs->mems_allowed);  in cpuset1_hotplug_update_tasks()
317 	if (is_empty && cs->css.cgroup->nr_populated_csets &&  in cpuset1_hotplug_update_tasks()
318 	    css_tryget_online(&cs->css)) {  in cpuset1_hotplug_update_tasks()
323 			css_put(&cs->css);  in cpuset1_hotplug_update_tasks()
327 		s->cs = cs;  in cpuset1_hotplug_update_tasks()
328 		INIT_WORK(&s->work, cpuset_migrate_tasks_workfn);  in cpuset1_hotplug_update_tasks()
329 		schedule_work(&s->work);  in cpuset1_hotplug_update_tasks()
334  * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q?
336  * One cpuset is a subset of another if all its allowed CPUs and
337  * Memory Nodes are a subset of the other, and its exclusive flags
343 	return	cpumask_subset(p->cpus_allowed, q->cpus_allowed) &&  in is_cpuset_subset()
344 		nodes_subset(p->mems_allowed, q->mems_allowed) &&  in is_cpuset_subset()
350  * cpuset1_validate_change() - Validate conditions specific to legacy (v1)
361 	/* Each of our child cpusets must be a subset of us */  in cpuset1_validate_change()
362 	ret = -EBUSY;  in cpuset1_validate_change()
367 	/* On legacy hierarchy, we must be a subset of our parent cpuset. */  in cpuset1_validate_change()
368 	ret = -EACCES;  in cpuset1_validate_change()
381  *  - Print tasks cpuset path into seq_file.
382  *  - Used for /proc/<pid>/cpuset.
391 	retval = -ENOMEM;  in proc_cpuset_show()
399 	retval = cgroup_path_ns_locked(css->cgroup, buf, PATH_MAX,  in proc_cpuset_show()
400 				       current->nsproxy->cgroup_ns);  in proc_cpuset_show()
404 	if (retval == -E2BIG)  in proc_cpuset_show()
405 		retval = -ENAMETOOLONG;  in proc_cpuset_show()
421 	cpuset_filetype_t type = cft->private;  in cpuset_read_u64()
437 		return fmeter_getrate(&cs->fmeter);  in cpuset_read_u64()
454 	cpuset_filetype_t type = cft->private;  in cpuset_write_u64()
460 		retval = -ENODEV;  in cpuset_write_u64()
469 		pr_info_once("cpuset.%s is deprecated\n", cft->name);  in cpuset_write_u64()
473 		pr_info_once("cpuset.%s is deprecated\n", cft->name);  in cpuset_write_u64()
477 		pr_info_once("cpuset.%s is deprecated, use cpuset.cpus.partition instead\n", cft->name);  in cpuset_write_u64()
481 		pr_info_once("cpuset.%s is deprecated\n", cft->name);  in cpuset_write_u64()
485 		pr_info_once("cpuset.%s is deprecated, use memory.pressure with CONFIG_PSI instead\n", cft->name);  in cpuset_write_u64()
489 		pr_info_once("cpuset.%s is deprecated\n", cft->name);  in cpuset_write_u64()
493 		pr_warn_once("cpuset.%s is deprecated\n", cft->name);  in cpuset_write_u64()
497 		retval = -EINVAL;  in cpuset_write_u64()
507  * for the common functions, 'private' gives the type of file