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(struct task_struct *p, struct cpumask *mask);
78extern bool cpuset_cpus_allowed_fallback(struct task_struct *p);
79extern bool cpuset_cpu_is_isolated(int cpu);
80extern nodemask_t cpuset_mems_allowed(struct task_struct *p);
81#define cpuset_current_mems_allowed (current->mems_allowed)
82void cpuset_init_current_mems_allowed(void);
83int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask);
84
85extern bool cpuset_current_node_allowed(int node, gfp_t gfp_mask);
86
87static inline bool __cpuset_zone_allowed(struct zone *z, gfp_t gfp_mask)
88{
89 return cpuset_current_node_allowed(node: zone_to_nid(zone: z), gfp_mask);
90}
91
92static inline bool cpuset_zone_allowed(struct zone *z, gfp_t gfp_mask)
93{
94 if (cpusets_enabled())
95 return __cpuset_zone_allowed(z, gfp_mask);
96 return true;
97}
98
99extern int cpuset_mems_allowed_intersects(const struct task_struct *tsk1,
100 const struct task_struct *tsk2);
101
102#ifdef CONFIG_CPUSETS_V1
103#define cpuset_memory_pressure_bump() \
104 do { \
105 if (cpuset_memory_pressure_enabled) \
106 __cpuset_memory_pressure_bump(); \
107 } while (0)
108extern int cpuset_memory_pressure_enabled;
109extern void __cpuset_memory_pressure_bump(void);
110#else
111static inline void cpuset_memory_pressure_bump(void) { }
112#endif
113
114extern void cpuset_task_status_allowed(struct seq_file *m,
115 struct task_struct *task);
116extern int proc_cpuset_show(struct seq_file *m, struct pid_namespace *ns,
117 struct pid *pid, struct task_struct *tsk);
118
119extern int cpuset_mem_spread_node(void);
120
121static inline int cpuset_do_page_mem_spread(void)
122{
123 return task_spread_page(current);
124}
125
126extern bool current_cpuset_is_being_rebound(void);
127
128extern void dl_rebuild_rd_accounting(void);
129extern void rebuild_sched_domains(void);
130
131extern void cpuset_print_current_mems_allowed(void);
132extern void cpuset_reset_sched_domains(void);
133
134/*
135 * read_mems_allowed_begin is required when making decisions involving
136 * mems_allowed such as during page allocation. mems_allowed can be updated in
137 * parallel and depending on the new value an operation can fail potentially
138 * causing process failure. A retry loop with read_mems_allowed_begin and
139 * read_mems_allowed_retry prevents these artificial failures.
140 */
141static inline unsigned int read_mems_allowed_begin(void)
142{
143 if (!static_branch_unlikely(&cpusets_pre_enable_key))
144 return 0;
145
146 return read_seqcount_begin(&current->mems_allowed_seq);
147}
148
149/*
150 * If this returns true, the operation that took place after
151 * read_mems_allowed_begin may have failed artificially due to a concurrent
152 * update of mems_allowed. It is up to the caller to retry the operation if
153 * appropriate.
154 */
155static inline bool read_mems_allowed_retry(unsigned int seq)
156{
157 if (!static_branch_unlikely(&cpusets_enabled_key))
158 return false;
159
160 return read_seqcount_retry(&current->mems_allowed_seq, seq);
161}
162
163static inline void set_mems_allowed(nodemask_t nodemask)
164{
165 unsigned long flags;
166
167 task_lock(current);
168 local_irq_save(flags);
169 write_seqcount_begin(&current->mems_allowed_seq);
170 current->mems_allowed = nodemask;
171 write_seqcount_end(&current->mems_allowed_seq);
172 local_irq_restore(flags);
173 task_unlock(current);
174}
175
176extern bool cpuset_node_allowed(struct cgroup *cgroup, int nid);
177#else /* !CONFIG_CPUSETS */
178
179static inline bool cpusets_enabled(void) { return false; }
180
181static inline bool cpusets_insane_config(void) { return false; }
182
183static inline int cpuset_init(void) { return 0; }
184static inline void cpuset_init_smp(void) {}
185
186static inline void cpuset_force_rebuild(void) { }
187
188static inline void cpuset_update_active_cpus(void)
189{
190 partition_sched_domains(1, NULL, NULL);
191}
192
193static inline void inc_dl_tasks_cs(struct task_struct *task) { }
194static inline void dec_dl_tasks_cs(struct task_struct *task) { }
195static inline void cpuset_lock(void) { }
196static inline void cpuset_unlock(void) { }
197
198static inline void cpuset_cpus_allowed(struct task_struct *p,
199 struct cpumask *mask)
200{
201 cpumask_copy(mask, task_cpu_possible_mask(p));
202}
203
204static inline bool cpuset_cpus_allowed_fallback(struct task_struct *p)
205{
206 return false;
207}
208
209static inline bool cpuset_cpu_is_isolated(int cpu)
210{
211 return false;
212}
213
214static inline nodemask_t cpuset_mems_allowed(struct task_struct *p)
215{
216 return node_possible_map;
217}
218
219#define cpuset_current_mems_allowed (node_states[N_MEMORY])
220static inline void cpuset_init_current_mems_allowed(void) {}
221
222static inline int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask)
223{
224 return 1;
225}
226
227static inline bool __cpuset_zone_allowed(struct zone *z, gfp_t gfp_mask)
228{
229 return true;
230}
231
232static inline bool cpuset_zone_allowed(struct zone *z, gfp_t gfp_mask)
233{
234 return true;
235}
236
237static inline int cpuset_mems_allowed_intersects(const struct task_struct *tsk1,
238 const struct task_struct *tsk2)
239{
240 return 1;
241}
242
243static inline void cpuset_memory_pressure_bump(void) {}
244
245static inline void cpuset_task_status_allowed(struct seq_file *m,
246 struct task_struct *task)
247{
248}
249
250static inline int cpuset_mem_spread_node(void)
251{
252 return 0;
253}
254
255static inline int cpuset_do_page_mem_spread(void)
256{
257 return 0;
258}
259
260static inline bool current_cpuset_is_being_rebound(void)
261{
262 return false;
263}
264
265static inline void dl_rebuild_rd_accounting(void)
266{
267}
268
269static inline void rebuild_sched_domains(void)
270{
271 partition_sched_domains(1, NULL, NULL);
272}
273
274static inline void cpuset_reset_sched_domains(void)
275{
276 partition_sched_domains(1, NULL, NULL);
277}
278
279static inline void cpuset_print_current_mems_allowed(void)
280{
281}
282
283static inline void set_mems_allowed(nodemask_t nodemask)
284{
285}
286
287static inline unsigned int read_mems_allowed_begin(void)
288{
289 return 0;
290}
291
292static inline bool read_mems_allowed_retry(unsigned int seq)
293{
294 return false;
295}
296
297static inline bool cpuset_node_allowed(struct cgroup *cgroup, int nid)
298{
299 return true;
300}
301#endif /* !CONFIG_CPUSETS */
302
303#endif /* _LINUX_CPUSET_H */
304