xref: /openbmc/linux/include/linux/sched/topology.h (revision ae40e94f)
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  * Increase resolution of cpu_capacity calculations
11  */
12 #define SCHED_CAPACITY_SHIFT	SCHED_FIXEDPOINT_SHIFT
13 #define SCHED_CAPACITY_SCALE	(1L << SCHED_CAPACITY_SHIFT)
14 
15 /*
16  * sched-domains (multiprocessor balancing) declarations:
17  */
18 #ifdef CONFIG_SMP
19 
20 #define SD_LOAD_BALANCE		0x0001	/* Do load balancing on this domain. */
21 #define SD_BALANCE_NEWIDLE	0x0002	/* Balance when about to become idle */
22 #define SD_BALANCE_EXEC		0x0004	/* Balance on exec */
23 #define SD_BALANCE_FORK		0x0008	/* Balance on fork, clone */
24 #define SD_BALANCE_WAKE		0x0010  /* Balance on wakeup */
25 #define SD_WAKE_AFFINE		0x0020	/* Wake task to waking CPU */
26 #define SD_ASYM_CPUCAPACITY	0x0040  /* Domain members have different CPU capacities */
27 #define SD_SHARE_CPUCAPACITY	0x0080	/* Domain members share CPU capacity */
28 #define SD_SHARE_POWERDOMAIN	0x0100	/* Domain members share power domain */
29 #define SD_SHARE_PKG_RESOURCES	0x0200	/* Domain members share CPU pkg resources */
30 #define SD_SERIALIZE		0x0400	/* Only a single load balancing instance */
31 #define SD_ASYM_PACKING		0x0800  /* Place busy groups earlier in the domain */
32 #define SD_PREFER_SIBLING	0x1000	/* Prefer to place tasks in a sibling domain */
33 #define SD_OVERLAP		0x2000	/* sched_domains of this level overlap */
34 #define SD_NUMA			0x4000	/* cross-node balancing */
35 
36 #ifdef CONFIG_SCHED_SMT
37 static inline int cpu_smt_flags(void)
38 {
39 	return SD_SHARE_CPUCAPACITY | SD_SHARE_PKG_RESOURCES;
40 }
41 #endif
42 
43 #ifdef CONFIG_SCHED_MC
44 static inline int cpu_core_flags(void)
45 {
46 	return SD_SHARE_PKG_RESOURCES;
47 }
48 #endif
49 
50 #ifdef CONFIG_NUMA
51 static inline int cpu_numa_flags(void)
52 {
53 	return SD_NUMA;
54 }
55 #endif
56 
57 extern int arch_asym_cpu_priority(int cpu);
58 
59 struct sched_domain_attr {
60 	int relax_domain_level;
61 };
62 
63 #define SD_ATTR_INIT	(struct sched_domain_attr) {	\
64 	.relax_domain_level = -1,			\
65 }
66 
67 extern int sched_domain_level_max;
68 
69 struct sched_group;
70 
71 struct sched_domain_shared {
72 	atomic_t	ref;
73 	atomic_t	nr_busy_cpus;
74 	int		has_idle_cores;
75 };
76 
77 struct sched_domain {
78 	/* These fields must be setup */
79 	struct sched_domain *parent;	/* top domain must be null terminated */
80 	struct sched_domain *child;	/* bottom domain must be null terminated */
81 	struct sched_group *groups;	/* the balancing groups of the domain */
82 	unsigned long min_interval;	/* Minimum balance interval ms */
83 	unsigned long max_interval;	/* Maximum balance interval ms */
84 	unsigned int busy_factor;	/* less balancing by factor if busy */
85 	unsigned int imbalance_pct;	/* No balance until over watermark */
86 	unsigned int cache_nice_tries;	/* Leave cache hot tasks for # tries */
87 	unsigned int busy_idx;
88 	unsigned int idle_idx;
89 	unsigned int newidle_idx;
90 	unsigned int wake_idx;
91 	unsigned int forkexec_idx;
92 
93 	int nohz_idle;			/* NOHZ IDLE status */
94 	int flags;			/* See SD_* */
95 	int level;
96 
97 	/* Runtime fields. */
98 	unsigned long last_balance;	/* init to jiffies. units in jiffies */
99 	unsigned int balance_interval;	/* initialise to 1. units in ms. */
100 	unsigned int nr_balance_failed; /* initialise to 0 */
101 
102 	/* idle_balance() stats */
103 	u64 max_newidle_lb_cost;
104 	unsigned long next_decay_max_lb_cost;
105 
106 	u64 avg_scan_cost;		/* select_idle_sibling */
107 
108 #ifdef CONFIG_SCHEDSTATS
109 	/* load_balance() stats */
110 	unsigned int lb_count[CPU_MAX_IDLE_TYPES];
111 	unsigned int lb_failed[CPU_MAX_IDLE_TYPES];
112 	unsigned int lb_balanced[CPU_MAX_IDLE_TYPES];
113 	unsigned int lb_imbalance[CPU_MAX_IDLE_TYPES];
114 	unsigned int lb_gained[CPU_MAX_IDLE_TYPES];
115 	unsigned int lb_hot_gained[CPU_MAX_IDLE_TYPES];
116 	unsigned int lb_nobusyg[CPU_MAX_IDLE_TYPES];
117 	unsigned int lb_nobusyq[CPU_MAX_IDLE_TYPES];
118 
119 	/* Active load balancing */
120 	unsigned int alb_count;
121 	unsigned int alb_failed;
122 	unsigned int alb_pushed;
123 
124 	/* SD_BALANCE_EXEC stats */
125 	unsigned int sbe_count;
126 	unsigned int sbe_balanced;
127 	unsigned int sbe_pushed;
128 
129 	/* SD_BALANCE_FORK stats */
130 	unsigned int sbf_count;
131 	unsigned int sbf_balanced;
132 	unsigned int sbf_pushed;
133 
134 	/* try_to_wake_up() stats */
135 	unsigned int ttwu_wake_remote;
136 	unsigned int ttwu_move_affine;
137 	unsigned int ttwu_move_balance;
138 #endif
139 #ifdef CONFIG_SCHED_DEBUG
140 	char *name;
141 #endif
142 	union {
143 		void *private;		/* used during construction */
144 		struct rcu_head rcu;	/* used during destruction */
145 	};
146 	struct sched_domain_shared *shared;
147 
148 	unsigned int span_weight;
149 	/*
150 	 * Span of all CPUs in this domain.
151 	 *
152 	 * NOTE: this field is variable length. (Allocated dynamically
153 	 * by attaching extra space to the end of the structure,
154 	 * depending on how many CPUs the kernel has booted up with)
155 	 */
156 	unsigned long span[0];
157 };
158 
159 static inline struct cpumask *sched_domain_span(struct sched_domain *sd)
160 {
161 	return to_cpumask(sd->span);
162 }
163 
164 extern void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[],
165 				    struct sched_domain_attr *dattr_new);
166 
167 /* Allocate an array of sched domains, for partition_sched_domains(). */
168 cpumask_var_t *alloc_sched_domains(unsigned int ndoms);
169 void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms);
170 
171 bool cpus_share_cache(int this_cpu, int that_cpu);
172 
173 typedef const struct cpumask *(*sched_domain_mask_f)(int cpu);
174 typedef int (*sched_domain_flags_f)(void);
175 
176 #define SDTL_OVERLAP	0x01
177 
178 struct sd_data {
179 	struct sched_domain *__percpu *sd;
180 	struct sched_domain_shared *__percpu *sds;
181 	struct sched_group *__percpu *sg;
182 	struct sched_group_capacity *__percpu *sgc;
183 };
184 
185 struct sched_domain_topology_level {
186 	sched_domain_mask_f mask;
187 	sched_domain_flags_f sd_flags;
188 	int		    flags;
189 	int		    numa_level;
190 	struct sd_data      data;
191 #ifdef CONFIG_SCHED_DEBUG
192 	char                *name;
193 #endif
194 };
195 
196 extern void set_sched_topology(struct sched_domain_topology_level *tl);
197 
198 #ifdef CONFIG_SCHED_DEBUG
199 # define SD_INIT_NAME(type)		.name = #type
200 #else
201 # define SD_INIT_NAME(type)
202 #endif
203 
204 #ifndef arch_scale_cpu_capacity
205 static __always_inline
206 unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
207 {
208 	return SCHED_CAPACITY_SCALE;
209 }
210 #endif
211 
212 #else /* CONFIG_SMP */
213 
214 struct sched_domain_attr;
215 
216 static inline void
217 partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[],
218 			struct sched_domain_attr *dattr_new)
219 {
220 }
221 
222 static inline bool cpus_share_cache(int this_cpu, int that_cpu)
223 {
224 	return true;
225 }
226 
227 #ifndef arch_scale_cpu_capacity
228 static __always_inline
229 unsigned long arch_scale_cpu_capacity(void __always_unused *sd, int cpu)
230 {
231 	return SCHED_CAPACITY_SCALE;
232 }
233 #endif
234 
235 #endif	/* !CONFIG_SMP */
236 
237 static inline int task_node(const struct task_struct *p)
238 {
239 	return cpu_to_node(task_cpu(p));
240 }
241 
242 #endif /* _LINUX_SCHED_TOPOLOGY_H */
243