1/* SPDX-License-Identifier: GPL-2.0 */ 2#ifndef _LINUX_ENERGY_MODEL_H 3#define _LINUX_ENERGY_MODEL_H 4#include <linux/cpumask.h> 5#include <linux/device.h> 6#include <linux/jump_label.h> 7#include <linux/kobject.h> 8#include <linux/rcupdate.h> 9#include <linux/sched/cpufreq.h> 10#include <linux/sched/topology.h> 11#include <linux/types.h> 12 13/** 14 * em_perf_state - Performance state of a performance domain 15 * @frequency: The frequency in KHz, for consistency with CPUFreq 16 * @power: The power consumed at this level, in milli-watts (by 1 CPU or 17 by a registered device). It can be a total power: static and 18 dynamic. 19 * @cost: The cost coefficient associated with this level, used during 20 * energy calculation. Equal to: power * max_frequency / frequency 21 */ 22struct em_perf_state { 23 unsigned long frequency; 24 unsigned long power; 25 unsigned long cost; 26}; 27 28/** 29 * em_perf_domain - Performance domain 30 * @table: List of performance states, in ascending order 31 * @nr_perf_states: Number of performance states 32 * @cpus: Cpumask covering the CPUs of the domain. It's here 33 * for performance reasons to avoid potential cache 34 * misses during energy calculations in the scheduler 35 * and simplifies allocating/freeing that memory region. 36 * 37 * In case of CPU device, a "performance domain" represents a group of CPUs 38 * whose performance is scaled together. All CPUs of a performance domain 39 * must have the same micro-architecture. Performance domains often have 40 * a 1-to-1 mapping with CPUFreq policies. In case of other devices the @cpus 41 * field is unused. 42 */ 43struct em_perf_domain { 44 struct em_perf_state *table; 45 int nr_perf_states; 46 unsigned long cpus[]; 47}; 48 49#define em_span_cpus(em) (to_cpumask((em)->cpus)) 50 51#ifdef CONFIG_ENERGY_MODEL 52#define EM_MAX_POWER 0xFFFF 53 54struct em_data_callback { 55 /** 56 * active_power() - Provide power at the next performance state of 57 * a device 58 * @power : Active power at the performance state in mW 59 * (modified) 60 * @freq : Frequency at the performance state in kHz 61 * (modified) 62 * @dev : Device for which we do this operation (can be a CPU) 63 * 64 * active_power() must find the lowest performance state of 'dev' above 65 * 'freq' and update 'power' and 'freq' to the matching active power 66 * and frequency. 67 * 68 * In case of CPUs, the power is the one of a single CPU in the domain, 69 * expressed in milli-watts. It is expected to fit in the 70 * [0, EM_MAX_POWER] range. 71 * 72 * Return 0 on success. 73 */ 74 int (*active_power)(unsigned long *power, unsigned long *freq, 75 struct device *dev); 76}; 77#define EM_DATA_CB(_active_power_cb) { .active_power = &_active_power_cb } 78 79struct em_perf_domain *em_cpu_get(int cpu); 80struct em_perf_domain *em_pd_get(struct device *dev); 81int em_dev_register_perf_domain(struct device *dev, unsigned int nr_states, 82 struct em_data_callback *cb, cpumask_t *span); 83void em_dev_unregister_perf_domain(struct device *dev); 84 85/** 86 * em_cpu_energy() - Estimates the energy consumed by the CPUs of a 87 performance domain 88 * @pd : performance domain for which energy has to be estimated 89 * @max_util : highest utilization among CPUs of the domain 90 * @sum_util : sum of the utilization of all CPUs in the domain 91 * 92 * This function must be used only for CPU devices. There is no validation, 93 * i.e. if the EM is a CPU type and has cpumask allocated. It is called from 94 * the scheduler code quite frequently and that is why there is not checks. 95 * 96 * Return: the sum of the energy consumed by the CPUs of the domain assuming 97 * a capacity state satisfying the max utilization of the domain. 98 */ 99static inline unsigned long em_cpu_energy(struct em_perf_domain *pd, 100 unsigned long max_util, unsigned long sum_util) 101{ 102 unsigned long freq, scale_cpu; 103 struct em_perf_state *ps; 104 int i, cpu; 105 106 /* 107 * In order to predict the performance state, map the utilization of 108 * the most utilized CPU of the performance domain to a requested 109 * frequency, like schedutil. 110 */ 111 cpu = cpumask_first(to_cpumask(pd->cpus)); 112 scale_cpu = arch_scale_cpu_capacity(cpu); 113 ps = &pd->table[pd->nr_perf_states - 1]; 114 freq = map_util_freq(max_util, ps->frequency, scale_cpu); 115 116 /* 117 * Find the lowest performance state of the Energy Model above the 118 * requested frequency. 119 */ 120 for (i = 0; i < pd->nr_perf_states; i++) { 121 ps = &pd->table[i]; 122 if (ps->frequency >= freq) 123 break; 124 } 125 126 /* 127 * The capacity of a CPU in the domain at the performance state (ps) 128 * can be computed as: 129 * 130 * ps->freq * scale_cpu 131 * ps->cap = -------------------- (1) 132 * cpu_max_freq 133 * 134 * So, ignoring the costs of idle states (which are not available in 135 * the EM), the energy consumed by this CPU at that performance state 136 * is estimated as: 137 * 138 * ps->power * cpu_util 139 * cpu_nrg = -------------------- (2) 140 * ps->cap 141 * 142 * since 'cpu_util / ps->cap' represents its percentage of busy time. 143 * 144 * NOTE: Although the result of this computation actually is in 145 * units of power, it can be manipulated as an energy value 146 * over a scheduling period, since it is assumed to be 147 * constant during that interval. 148 * 149 * By injecting (1) in (2), 'cpu_nrg' can be re-expressed as a product 150 * of two terms: 151 * 152 * ps->power * cpu_max_freq cpu_util 153 * cpu_nrg = ------------------------ * --------- (3) 154 * ps->freq scale_cpu 155 * 156 * The first term is static, and is stored in the em_perf_state struct 157 * as 'ps->cost'. 158 * 159 * Since all CPUs of the domain have the same micro-architecture, they 160 * share the same 'ps->cost', and the same CPU capacity. Hence, the 161 * total energy of the domain (which is the simple sum of the energy of 162 * all of its CPUs) can be factorized as: 163 * 164 * ps->cost * \Sum cpu_util 165 * pd_nrg = ------------------------ (4) 166 * scale_cpu 167 */ 168 return ps->cost * sum_util / scale_cpu; 169} 170 171/** 172 * em_pd_nr_perf_states() - Get the number of performance states of a perf. 173 * domain 174 * @pd : performance domain for which this must be done 175 * 176 * Return: the number of performance states in the performance domain table 177 */ 178static inline int em_pd_nr_perf_states(struct em_perf_domain *pd) 179{ 180 return pd->nr_perf_states; 181} 182 183#else 184struct em_data_callback {}; 185#define EM_DATA_CB(_active_power_cb) { } 186 187static inline 188int em_dev_register_perf_domain(struct device *dev, unsigned int nr_states, 189 struct em_data_callback *cb, cpumask_t *span) 190{ 191 return -EINVAL; 192} 193static inline void em_dev_unregister_perf_domain(struct device *dev) 194{ 195} 196static inline struct em_perf_domain *em_cpu_get(int cpu) 197{ 198 return NULL; 199} 200static inline struct em_perf_domain *em_pd_get(struct device *dev) 201{ 202 return NULL; 203} 204static inline unsigned long em_cpu_energy(struct em_perf_domain *pd, 205 unsigned long max_util, unsigned long sum_util) 206{ 207 return 0; 208} 209static inline int em_pd_nr_perf_states(struct em_perf_domain *pd) 210{ 211 return 0; 212} 213#endif 214 215#endif 216