1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_SCHED_TOPOLOGY_H
3 #define _LINUX_SCHED_TOPOLOGY_H
4
5 #include <linux/topology.h>
6
7 #include <linux/sched/idle.h>
8
9 /*
10 * sched-domains (multiprocessor balancing) declarations:
11 */
12
13 /* Generate SD flag indexes */
14 #define SD_FLAG(name, mflags) __##name,
15 enum {
16 #include <linux/sched/sd_flags.h>
17 __SD_FLAG_CNT,
18 };
19 #undef SD_FLAG
20 /* Generate SD flag bits */
21 #define SD_FLAG(name, mflags) name = 1 << __##name,
22 enum {
23 #include <linux/sched/sd_flags.h>
24 };
25 #undef SD_FLAG
26
27 struct sd_flag_debug {
28 unsigned int meta_flags;
29 char *name;
30 };
31 extern const struct sd_flag_debug sd_flag_debug[];
32
33 struct sched_domain_topology_level;
34
35 #ifdef CONFIG_SCHED_SMT
36 extern int cpu_smt_flags(void);
37 extern const struct cpumask *tl_smt_mask(struct sched_domain_topology_level *tl, int cpu);
38 #endif
39
40 #ifdef CONFIG_SCHED_CLUSTER
41 extern int cpu_cluster_flags(void);
42 extern const struct cpumask *tl_cls_mask(struct sched_domain_topology_level *tl, int cpu);
43 #endif
44
45 #ifdef CONFIG_SCHED_MC
46 extern int cpu_core_flags(void);
47 extern const struct cpumask *tl_mc_mask(struct sched_domain_topology_level *tl, int cpu);
48 #endif
49
50 extern const struct cpumask *tl_pkg_mask(struct sched_domain_topology_level *tl, int cpu);
51
52 extern int arch_asym_cpu_priority(int cpu);
53
54 struct sched_domain_attr {
55 int relax_domain_level;
56 };
57
58 #define SD_ATTR_INIT (struct sched_domain_attr) { \
59 .relax_domain_level = -1, \
60 }
61
62 extern int sched_domain_level_max;
63
64 struct sched_group;
65
66 struct sched_domain_shared {
67 atomic_t ref;
68 atomic_t nr_busy_cpus;
69 int has_idle_cores;
70 int nr_idle_scan;
71 };
72
73 struct sched_domain {
74 /* These fields must be setup */
75 struct sched_domain __rcu *parent; /* top domain must be null terminated */
76 struct sched_domain __rcu *child; /* bottom domain must be null terminated */
77 struct sched_group *groups; /* the balancing groups of the domain */
78 unsigned long min_interval; /* Minimum balance interval ms */
79 unsigned long max_interval; /* Maximum balance interval ms */
80 unsigned int busy_factor; /* less balancing by factor if busy */
81 unsigned int imbalance_pct; /* No balance until over watermark */
82 unsigned int cache_nice_tries; /* Leave cache hot tasks for # tries */
83 unsigned int imb_numa_nr; /* Nr running tasks that allows a NUMA imbalance */
84
85 int nohz_idle; /* NOHZ IDLE status */
86 int flags; /* See SD_* */
87 int level;
88
89 /* Runtime fields. */
90 unsigned long last_balance; /* init to jiffies. units in jiffies */
91 unsigned int balance_interval; /* initialise to 1. units in ms. */
92 unsigned int nr_balance_failed; /* initialise to 0 */
93
94 /* idle_balance() stats */
95 unsigned int newidle_call;
96 unsigned int newidle_success;
97 unsigned int newidle_ratio;
98 u64 newidle_stamp;
99 u64 max_newidle_lb_cost;
100 unsigned long last_decay_max_lb_cost;
101
102 #ifdef CONFIG_SCHEDSTATS
103 /* sched_balance_rq() stats */
104 unsigned int lb_count[CPU_MAX_IDLE_TYPES];
105 unsigned int lb_failed[CPU_MAX_IDLE_TYPES];
106 unsigned int lb_balanced[CPU_MAX_IDLE_TYPES];
107 unsigned int lb_imbalance_load[CPU_MAX_IDLE_TYPES];
108 unsigned int lb_imbalance_util[CPU_MAX_IDLE_TYPES];
109 unsigned int lb_imbalance_task[CPU_MAX_IDLE_TYPES];
110 unsigned int lb_imbalance_misfit[CPU_MAX_IDLE_TYPES];
111 unsigned int lb_gained[CPU_MAX_IDLE_TYPES];
112 unsigned int lb_hot_gained[CPU_MAX_IDLE_TYPES];
113 unsigned int lb_nobusyg[CPU_MAX_IDLE_TYPES];
114 unsigned int lb_nobusyq[CPU_MAX_IDLE_TYPES];
115
116 /* Active load balancing */
117 unsigned int alb_count;
118 unsigned int alb_failed;
119 unsigned int alb_pushed;
120
121 /* SD_BALANCE_EXEC stats */
122 unsigned int sbe_count;
123 unsigned int sbe_balanced;
124 unsigned int sbe_pushed;
125
126 /* SD_BALANCE_FORK stats */
127 unsigned int sbf_count;
128 unsigned int sbf_balanced;
129 unsigned int sbf_pushed;
130
131 /* try_to_wake_up() stats */
132 unsigned int ttwu_wake_remote;
133 unsigned int ttwu_move_affine;
134 unsigned int ttwu_move_balance;
135 #endif
136 char *name;
137 union {
138 void *private; /* used during construction */
139 struct rcu_head rcu; /* used during destruction */
140 };
141 struct sched_domain_shared *shared;
142
143 unsigned int span_weight;
144 /*
145 * See sched_domain_span(), on why flex arrays are broken.
146 *
147 unsigned long span[];
148 */
149 };
150
sched_domain_span(struct sched_domain * sd)151 static inline struct cpumask *sched_domain_span(struct sched_domain *sd)
152 {
153 /*
154 * Turns out that C flexible arrays are fundamentally broken since it
155 * is allowed for offsetof(*sd, span) < sizeof(*sd), this means that
156 * structure initialzation *sd = { ... }; which writes every byte
157 * inside sizeof(*type), will over-write the start of the flexible
158 * array.
159 *
160 * Luckily, the way we allocate sched_domain is by:
161 *
162 * sizeof(*sd) + cpumask_size()
163 *
164 * this means that we have sufficient space for the whole flex array
165 * *outside* of sizeof(*sd). So use that, and avoid using sd->span.
166 */
167 unsigned long *bitmap = (void *)sd + sizeof(*sd);
168 return to_cpumask(bitmap);
169 }
170
171 extern void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[],
172 struct sched_domain_attr *dattr_new);
173
174 /* Allocate an array of sched domains, for partition_sched_domains(). */
175 cpumask_var_t *alloc_sched_domains(unsigned int ndoms);
176 void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms);
177
178 bool cpus_equal_capacity(int this_cpu, int that_cpu);
179 bool cpus_share_cache(int this_cpu, int that_cpu);
180 bool cpus_share_resources(int this_cpu, int that_cpu);
181
182 typedef const struct cpumask *(*sched_domain_mask_f)(struct sched_domain_topology_level *tl, int cpu);
183 typedef int (*sched_domain_flags_f)(void);
184
185 struct sd_data {
186 struct sched_domain *__percpu *sd;
187 struct sched_group *__percpu *sg;
188 struct sched_group_capacity *__percpu *sgc;
189 };
190
191 struct sched_domain_topology_level {
192 sched_domain_mask_f mask;
193 sched_domain_flags_f sd_flags;
194 int numa_level;
195 struct sd_data data;
196 char *name;
197 };
198
199 extern void __init set_sched_topology(struct sched_domain_topology_level *tl);
200 extern void sched_update_asym_prefer_cpu(int cpu, int old_prio, int new_prio);
201
202 #define SDTL_INIT(maskfn, flagsfn, dname) ((struct sched_domain_topology_level) \
203 { .mask = maskfn, .sd_flags = flagsfn, .name = #dname })
204
205 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
206 extern void rebuild_sched_domains_energy(void);
207 #else
rebuild_sched_domains_energy(void)208 static inline void rebuild_sched_domains_energy(void)
209 {
210 }
211 #endif
212
213 #ifndef arch_scale_cpu_capacity
214 /**
215 * arch_scale_cpu_capacity - get the capacity scale factor of a given CPU.
216 * @cpu: the CPU in question.
217 *
218 * Return: the CPU scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
219 *
220 * max_perf(cpu)
221 * ----------------------------- * SCHED_CAPACITY_SCALE
222 * max(max_perf(c) : c \in CPUs)
223 */
224 static __always_inline
arch_scale_cpu_capacity(int cpu)225 unsigned long arch_scale_cpu_capacity(int cpu)
226 {
227 return SCHED_CAPACITY_SCALE;
228 }
229 #endif
230
231 #ifndef arch_scale_hw_pressure
232 static __always_inline
arch_scale_hw_pressure(int cpu)233 unsigned long arch_scale_hw_pressure(int cpu)
234 {
235 return 0;
236 }
237 #endif
238
239 #ifndef arch_update_hw_pressure
240 static __always_inline
arch_update_hw_pressure(const struct cpumask * cpus,unsigned long capped_frequency)241 void arch_update_hw_pressure(const struct cpumask *cpus,
242 unsigned long capped_frequency)
243 { }
244 #endif
245
246 #ifndef arch_scale_freq_ref
247 static __always_inline
arch_scale_freq_ref(int cpu)248 unsigned int arch_scale_freq_ref(int cpu)
249 {
250 return 0;
251 }
252 #endif
253
task_node(const struct task_struct * p)254 static inline int task_node(const struct task_struct *p)
255 {
256 return cpu_to_node(task_cpu(p));
257 }
258
259 #endif /* _LINUX_SCHED_TOPOLOGY_H */
260