xref: /openbmc/linux/include/linux/sched/topology.h (revision dbf563ee)
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 #ifdef CONFIG_SMP
13 
14 #define SD_BALANCE_NEWIDLE	0x0001	/* Balance when about to become idle */
15 #define SD_BALANCE_EXEC		0x0002	/* Balance on exec */
16 #define SD_BALANCE_FORK		0x0004	/* Balance on fork, clone */
17 #define SD_BALANCE_WAKE		0x0008  /* Balance on wakeup */
18 #define SD_WAKE_AFFINE		0x0010	/* Wake task to waking CPU */
19 #define SD_ASYM_CPUCAPACITY	0x0020  /* Domain members have different CPU capacities */
20 #define SD_SHARE_CPUCAPACITY	0x0040	/* Domain members share CPU capacity */
21 #define SD_SHARE_POWERDOMAIN	0x0080	/* Domain members share power domain */
22 #define SD_SHARE_PKG_RESOURCES	0x0100	/* Domain members share CPU pkg resources */
23 #define SD_SERIALIZE		0x0200	/* Only a single load balancing instance */
24 #define SD_ASYM_PACKING		0x0400  /* Place busy groups earlier in the domain */
25 #define SD_PREFER_SIBLING	0x0800	/* Prefer to place tasks in a sibling domain */
26 #define SD_OVERLAP		0x1000	/* sched_domains of this level overlap */
27 #define SD_NUMA			0x2000	/* cross-node balancing */
28 
29 #ifdef CONFIG_SCHED_SMT
30 static inline int cpu_smt_flags(void)
31 {
32 	return SD_SHARE_CPUCAPACITY | SD_SHARE_PKG_RESOURCES;
33 }
34 #endif
35 
36 #ifdef CONFIG_SCHED_MC
37 static inline int cpu_core_flags(void)
38 {
39 	return SD_SHARE_PKG_RESOURCES;
40 }
41 #endif
42 
43 #ifdef CONFIG_NUMA
44 static inline int cpu_numa_flags(void)
45 {
46 	return SD_NUMA;
47 }
48 #endif
49 
50 extern int arch_asym_cpu_priority(int cpu);
51 
52 struct sched_domain_attr {
53 	int relax_domain_level;
54 };
55 
56 #define SD_ATTR_INIT	(struct sched_domain_attr) {	\
57 	.relax_domain_level = -1,			\
58 }
59 
60 extern int sched_domain_level_max;
61 
62 struct sched_group;
63 
64 struct sched_domain_shared {
65 	atomic_t	ref;
66 	atomic_t	nr_busy_cpus;
67 	int		has_idle_cores;
68 };
69 
70 struct sched_domain {
71 	/* These fields must be setup */
72 	struct sched_domain __rcu *parent;	/* top domain must be null terminated */
73 	struct sched_domain __rcu *child;	/* bottom domain must be null terminated */
74 	struct sched_group *groups;	/* the balancing groups of the domain */
75 	unsigned long min_interval;	/* Minimum balance interval ms */
76 	unsigned long max_interval;	/* Maximum balance interval ms */
77 	unsigned int busy_factor;	/* less balancing by factor if busy */
78 	unsigned int imbalance_pct;	/* No balance until over watermark */
79 	unsigned int cache_nice_tries;	/* Leave cache hot tasks for # tries */
80 
81 	int nohz_idle;			/* NOHZ IDLE status */
82 	int flags;			/* See SD_* */
83 	int level;
84 
85 	/* Runtime fields. */
86 	unsigned long last_balance;	/* init to jiffies. units in jiffies */
87 	unsigned int balance_interval;	/* initialise to 1. units in ms. */
88 	unsigned int nr_balance_failed; /* initialise to 0 */
89 
90 	/* idle_balance() stats */
91 	u64 max_newidle_lb_cost;
92 	unsigned long next_decay_max_lb_cost;
93 
94 	u64 avg_scan_cost;		/* select_idle_sibling */
95 
96 #ifdef CONFIG_SCHEDSTATS
97 	/* load_balance() stats */
98 	unsigned int lb_count[CPU_MAX_IDLE_TYPES];
99 	unsigned int lb_failed[CPU_MAX_IDLE_TYPES];
100 	unsigned int lb_balanced[CPU_MAX_IDLE_TYPES];
101 	unsigned int lb_imbalance[CPU_MAX_IDLE_TYPES];
102 	unsigned int lb_gained[CPU_MAX_IDLE_TYPES];
103 	unsigned int lb_hot_gained[CPU_MAX_IDLE_TYPES];
104 	unsigned int lb_nobusyg[CPU_MAX_IDLE_TYPES];
105 	unsigned int lb_nobusyq[CPU_MAX_IDLE_TYPES];
106 
107 	/* Active load balancing */
108 	unsigned int alb_count;
109 	unsigned int alb_failed;
110 	unsigned int alb_pushed;
111 
112 	/* SD_BALANCE_EXEC stats */
113 	unsigned int sbe_count;
114 	unsigned int sbe_balanced;
115 	unsigned int sbe_pushed;
116 
117 	/* SD_BALANCE_FORK stats */
118 	unsigned int sbf_count;
119 	unsigned int sbf_balanced;
120 	unsigned int sbf_pushed;
121 
122 	/* try_to_wake_up() stats */
123 	unsigned int ttwu_wake_remote;
124 	unsigned int ttwu_move_affine;
125 	unsigned int ttwu_move_balance;
126 #endif
127 #ifdef CONFIG_SCHED_DEBUG
128 	char *name;
129 #endif
130 	union {
131 		void *private;		/* used during construction */
132 		struct rcu_head rcu;	/* used during destruction */
133 	};
134 	struct sched_domain_shared *shared;
135 
136 	unsigned int span_weight;
137 	/*
138 	 * Span of all CPUs in this domain.
139 	 *
140 	 * NOTE: this field is variable length. (Allocated dynamically
141 	 * by attaching extra space to the end of the structure,
142 	 * depending on how many CPUs the kernel has booted up with)
143 	 */
144 	unsigned long span[];
145 };
146 
147 static inline struct cpumask *sched_domain_span(struct sched_domain *sd)
148 {
149 	return to_cpumask(sd->span);
150 }
151 
152 extern void partition_sched_domains_locked(int ndoms_new,
153 					   cpumask_var_t doms_new[],
154 					   struct sched_domain_attr *dattr_new);
155 
156 extern void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[],
157 				    struct sched_domain_attr *dattr_new);
158 
159 /* Allocate an array of sched domains, for partition_sched_domains(). */
160 cpumask_var_t *alloc_sched_domains(unsigned int ndoms);
161 void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms);
162 
163 bool cpus_share_cache(int this_cpu, int that_cpu);
164 
165 typedef const struct cpumask *(*sched_domain_mask_f)(int cpu);
166 typedef int (*sched_domain_flags_f)(void);
167 
168 #define SDTL_OVERLAP	0x01
169 
170 struct sd_data {
171 	struct sched_domain *__percpu *sd;
172 	struct sched_domain_shared *__percpu *sds;
173 	struct sched_group *__percpu *sg;
174 	struct sched_group_capacity *__percpu *sgc;
175 };
176 
177 struct sched_domain_topology_level {
178 	sched_domain_mask_f mask;
179 	sched_domain_flags_f sd_flags;
180 	int		    flags;
181 	int		    numa_level;
182 	struct sd_data      data;
183 #ifdef CONFIG_SCHED_DEBUG
184 	char                *name;
185 #endif
186 };
187 
188 extern void set_sched_topology(struct sched_domain_topology_level *tl);
189 
190 #ifdef CONFIG_SCHED_DEBUG
191 # define SD_INIT_NAME(type)		.name = #type
192 #else
193 # define SD_INIT_NAME(type)
194 #endif
195 
196 #else /* CONFIG_SMP */
197 
198 struct sched_domain_attr;
199 
200 static inline void
201 partition_sched_domains_locked(int ndoms_new, cpumask_var_t doms_new[],
202 			       struct sched_domain_attr *dattr_new)
203 {
204 }
205 
206 static inline void
207 partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[],
208 			struct sched_domain_attr *dattr_new)
209 {
210 }
211 
212 static inline bool cpus_share_cache(int this_cpu, int that_cpu)
213 {
214 	return true;
215 }
216 
217 #endif	/* !CONFIG_SMP */
218 
219 #ifndef arch_scale_cpu_capacity
220 /**
221  * arch_scale_cpu_capacity - get the capacity scale factor of a given CPU.
222  * @cpu: the CPU in question.
223  *
224  * Return: the CPU scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
225  *
226  *             max_perf(cpu)
227  *      ----------------------------- * SCHED_CAPACITY_SCALE
228  *      max(max_perf(c) : c \in CPUs)
229  */
230 static __always_inline
231 unsigned long arch_scale_cpu_capacity(int cpu)
232 {
233 	return SCHED_CAPACITY_SCALE;
234 }
235 #endif
236 
237 #ifndef arch_scale_thermal_pressure
238 static __always_inline
239 unsigned long arch_scale_thermal_pressure(int cpu)
240 {
241 	return 0;
242 }
243 #endif
244 
245 #ifndef arch_set_thermal_pressure
246 static __always_inline
247 void arch_set_thermal_pressure(const struct cpumask *cpus,
248 			       unsigned long th_pressure)
249 { }
250 #endif
251 
252 static inline int task_node(const struct task_struct *p)
253 {
254 	return cpu_to_node(task_cpu(p));
255 }
256 
257 #endif /* _LINUX_SCHED_TOPOLOGY_H */
258