cpuset.h

  1/* SPDX-License-Identifier: GPL-2.0 */
  2#ifndef _LINUX_CPUSET_H
  3#define _LINUX_CPUSET_H
  4/*
  5 *  cpuset interface
  6 *
  7 *  Copyright (C) 2003 BULL SA
  8 *  Copyright (C) 2004-2006 Silicon Graphics, Inc.
  9 *
 10 */
 11
 12#include <linux/sched.h>
 13#include <linux/sched/topology.h>
 14#include <linux/sched/task.h>
 15#include <linux/cpumask.h>
 16#include <linux/nodemask.h>
 17#include <linux/mm.h>
 18#include <linux/mmu_context.h>
 19#include <linux/jump_label.h>
 20
 21#ifdef CONFIG_CPUSETS
 22
 23/*
 24 * Static branch rewrites can happen in an arbitrary order for a given
 25 * key. In code paths where we need to loop with read_mems_allowed_begin() and
 26 * read_mems_allowed_retry() to get a consistent view of mems_allowed, we need
 27 * to ensure that begin() always gets rewritten before retry() in the
 28 * disabled -> enabled transition. If not, then if local irqs are disabled
 29 * around the loop, we can deadlock since retry() would always be
 30 * comparing the latest value of the mems_allowed seqcount against 0 as
 31 * begin() still would see cpusets_enabled() as false. The enabled -> disabled
 32 * transition should happen in reverse order for the same reasons (want to stop
 33 * looking at real value of mems_allowed.sequence in retry() first).
 34 */
 35extern struct static_key_false cpusets_pre_enable_key;
 36extern struct static_key_false cpusets_enabled_key;
 37extern struct static_key_false cpusets_insane_config_key;
 38
 39static inline bool cpusets_enabled(void)
 40{
 41	return static_branch_unlikely(&cpusets_enabled_key);
 42}
 43
 44static inline void cpuset_inc(void)
 45{
 46	static_branch_inc_cpuslocked(&cpusets_pre_enable_key);
 47	static_branch_inc_cpuslocked(&cpusets_enabled_key);
 48}
 49
 50static inline void cpuset_dec(void)
 51{
 52	static_branch_dec_cpuslocked(&cpusets_enabled_key);
 53	static_branch_dec_cpuslocked(&cpusets_pre_enable_key);
 54}
 55
 56/*
 57 * This will get enabled whenever a cpuset configuration is considered
 58 * unsupportable in general. E.g. movable only node which cannot satisfy
 59 * any non movable allocations (see update_nodemask). Page allocator
 60 * needs to make additional checks for those configurations and this
 61 * check is meant to guard those checks without any overhead for sane
 62 * configurations.
 63 */
 64static inline bool cpusets_insane_config(void)
 65{
 66	return static_branch_unlikely(&cpusets_insane_config_key);
 67}
 68
 69extern int cpuset_init(void);
 70extern void cpuset_init_smp(void);
 71extern void cpuset_force_rebuild(void);
 72extern void cpuset_update_active_cpus(void);
 73extern void inc_dl_tasks_cs(struct task_struct *task);
 74extern void dec_dl_tasks_cs(struct task_struct *task);
 75extern void cpuset_lock(void);
 76extern void cpuset_unlock(void);
 77extern void cpuset_cpus_allowed_locked(struct task_struct *p, struct cpumask *mask);
 78extern void cpuset_cpus_allowed(struct task_struct *p, struct cpumask *mask);
 79extern bool cpuset_cpus_allowed_fallback(struct task_struct *p);
 80extern bool cpuset_cpu_is_isolated(int cpu);
 81extern nodemask_t cpuset_mems_allowed(struct task_struct *p);
 82#define cpuset_current_mems_allowed (current->mems_allowed)
 83void cpuset_init_current_mems_allowed(void);
 84int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask);
 85
 86extern bool cpuset_current_node_allowed(int node, gfp_t gfp_mask);
 87
 88static inline bool __cpuset_zone_allowed(struct zone *z, gfp_t gfp_mask)
 89{
 90	return cpuset_current_node_allowed(zone_to_nid(z), gfp_mask);
 91}
 92
 93static inline bool cpuset_zone_allowed(struct zone *z, gfp_t gfp_mask)
 94{
 95	if (cpusets_enabled())
 96		return __cpuset_zone_allowed(z, gfp_mask);
 97	return true;
 98}
 99
100extern int cpuset_mems_allowed_intersects(const struct task_struct *tsk1,
101					  const struct task_struct *tsk2);
102
103#ifdef CONFIG_CPUSETS_V1
104#define cpuset_memory_pressure_bump() 				\
105	do {							\
106		if (cpuset_memory_pressure_enabled)		\
107			__cpuset_memory_pressure_bump();	\
108	} while (0)
109extern int cpuset_memory_pressure_enabled;
110extern void __cpuset_memory_pressure_bump(void);
111#else
112static inline void cpuset_memory_pressure_bump(void) { }
113#endif
114
115extern void cpuset_task_status_allowed(struct seq_file *m,
116					struct task_struct *task);
117extern int proc_cpuset_show(struct seq_file *m, struct pid_namespace *ns,
118			    struct pid *pid, struct task_struct *tsk);
119
120extern int cpuset_mem_spread_node(void);
121
122static inline int cpuset_do_page_mem_spread(void)
123{
124	return task_spread_page(current);
125}
126
127extern bool current_cpuset_is_being_rebound(void);
128
129extern void dl_rebuild_rd_accounting(void);
130extern void rebuild_sched_domains(void);
131
132extern void cpuset_print_current_mems_allowed(void);
133extern void cpuset_reset_sched_domains(void);
134
135/*
136 * read_mems_allowed_begin is required when making decisions involving
137 * mems_allowed such as during page allocation. mems_allowed can be updated in
138 * parallel and depending on the new value an operation can fail potentially
139 * causing process failure. A retry loop with read_mems_allowed_begin and
140 * read_mems_allowed_retry prevents these artificial failures.
141 */
142static inline unsigned int read_mems_allowed_begin(void)
143{
144	if (!static_branch_unlikely(&cpusets_pre_enable_key))
145		return 0;
146
147	return read_seqcount_begin(&current->mems_allowed_seq);
148}
149
150/*
151 * If this returns true, the operation that took place after
152 * read_mems_allowed_begin may have failed artificially due to a concurrent
153 * update of mems_allowed. It is up to the caller to retry the operation if
154 * appropriate.
155 */
156static inline bool read_mems_allowed_retry(unsigned int seq)
157{
158	if (!static_branch_unlikely(&cpusets_enabled_key))
159		return false;
160
161	return read_seqcount_retry(&current->mems_allowed_seq, seq);
162}
163
164static inline void set_mems_allowed(nodemask_t nodemask)
165{
166	unsigned long flags;
167
168	task_lock(current);
169	local_irq_save(flags);
170	write_seqcount_begin(&current->mems_allowed_seq);
171	current->mems_allowed = nodemask;
172	write_seqcount_end(&current->mems_allowed_seq);
173	local_irq_restore(flags);
174	task_unlock(current);
175}
176
177extern void cpuset_nodes_allowed(struct cgroup *cgroup, nodemask_t *mask);
178#else /* !CONFIG_CPUSETS */
179
180static inline bool cpusets_enabled(void) { return false; }
181
182static inline bool cpusets_insane_config(void) { return false; }
183
184static inline int cpuset_init(void) { return 0; }
185static inline void cpuset_init_smp(void) {}
186
187static inline void cpuset_force_rebuild(void) { }
188
189static inline void cpuset_update_active_cpus(void)
190{
191	partition_sched_domains(1, NULL, NULL);
192}
193
194static inline void inc_dl_tasks_cs(struct task_struct *task) { }
195static inline void dec_dl_tasks_cs(struct task_struct *task) { }
196static inline void cpuset_lock(void) { }
197static inline void cpuset_unlock(void) { }
198
199static inline void cpuset_cpus_allowed_locked(struct task_struct *p,
200					struct cpumask *mask)
201{
202	cpumask_copy(mask, task_cpu_possible_mask(p));
203}
204
205static inline void cpuset_cpus_allowed(struct task_struct *p,
206				       struct cpumask *mask)
207{
208	cpuset_cpus_allowed_locked(p, mask);
209}
210
211static inline bool cpuset_cpus_allowed_fallback(struct task_struct *p)
212{
213	return false;
214}
215
216static inline bool cpuset_cpu_is_isolated(int cpu)
217{
218	return false;
219}
220
221static inline nodemask_t cpuset_mems_allowed(struct task_struct *p)
222{
223	return node_possible_map;
224}
225
226#define cpuset_current_mems_allowed (node_states[N_MEMORY])
227static inline void cpuset_init_current_mems_allowed(void) {}
228
229static inline int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask)
230{
231	return 1;
232}
233
234static inline bool __cpuset_zone_allowed(struct zone *z, gfp_t gfp_mask)
235{
236	return true;
237}
238
239static inline bool cpuset_zone_allowed(struct zone *z, gfp_t gfp_mask)
240{
241	return true;
242}
243
244static inline int cpuset_mems_allowed_intersects(const struct task_struct *tsk1,
245						 const struct task_struct *tsk2)
246{
247	return 1;
248}
249
250static inline void cpuset_memory_pressure_bump(void) {}
251
252static inline void cpuset_task_status_allowed(struct seq_file *m,
253						struct task_struct *task)
254{
255}
256
257static inline int cpuset_mem_spread_node(void)
258{
259	return 0;
260}
261
262static inline int cpuset_do_page_mem_spread(void)
263{
264	return 0;
265}
266
267static inline bool current_cpuset_is_being_rebound(void)
268{
269	return false;
270}
271
272static inline void dl_rebuild_rd_accounting(void)
273{
274}
275
276static inline void rebuild_sched_domains(void)
277{
278	partition_sched_domains(1, NULL, NULL);
279}
280
281static inline void cpuset_reset_sched_domains(void)
282{
283	partition_sched_domains(1, NULL, NULL);
284}
285
286static inline void cpuset_print_current_mems_allowed(void)
287{
288}
289
290static inline void set_mems_allowed(nodemask_t nodemask)
291{
292}
293
294static inline unsigned int read_mems_allowed_begin(void)
295{
296	return 0;
297}
298
299static inline bool read_mems_allowed_retry(unsigned int seq)
300{
301	return false;
302}
303
304static inline void cpuset_nodes_allowed(struct cgroup *cgroup, nodemask_t *mask)
305{
306	nodes_copy(*mask, node_states[N_MEMORY]);
307}
308#endif /* !CONFIG_CPUSETS */
309
310#endif /* _LINUX_CPUSET_H */