Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright 2020 Linaro Limited
4 *
5 * Author: Daniel Lezcano <daniel.lezcano@linaro.org>
6 *
7 * The DTPM CPU is based on the energy model. It hooks the CPU in the
8 * DTPM tree which in turns update the power number by propagating the
9 * power number from the CPU energy model information to the parents.
10 *
11 * The association between the power and the performance state, allows
12 * to set the power of the CPU at the OPP granularity.
13 *
14 * The CPU hotplug is supported and the power numbers will be updated
15 * if a CPU is hot plugged / unplugged.
16 */
17#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
18
19#include <linux/cpumask.h>
20#include <linux/cpufreq.h>
21#include <linux/cpuhotplug.h>
22#include <linux/dtpm.h>
23#include <linux/energy_model.h>
24#include <linux/of.h>
25#include <linux/pm_qos.h>
26#include <linux/slab.h>
27
28struct dtpm_cpu {
29 struct dtpm dtpm;
30 struct freq_qos_request qos_req;
31 int cpu;
32};
33
34static DEFINE_PER_CPU(struct dtpm_cpu *, dtpm_per_cpu);
35
36static struct dtpm_cpu *to_dtpm_cpu(struct dtpm *dtpm)
37{
38 return container_of(dtpm, struct dtpm_cpu, dtpm);
39}
40
41static u64 set_pd_power_limit(struct dtpm *dtpm, u64 power_limit)
42{
43 struct dtpm_cpu *dtpm_cpu = to_dtpm_cpu(dtpm);
44 struct em_perf_domain *pd = em_cpu_get(dtpm_cpu->cpu);
45 struct cpumask cpus;
46 unsigned long freq;
47 u64 power;
48 int i, nr_cpus;
49
50 cpumask_and(&cpus, cpu_online_mask, to_cpumask(pd->cpus));
51 nr_cpus = cpumask_weight(&cpus);
52
53 for (i = 0; i < pd->nr_perf_states; i++) {
54
55 power = pd->table[i].power * nr_cpus;
56
57 if (power > power_limit)
58 break;
59 }
60
61 freq = pd->table[i - 1].frequency;
62
63 freq_qos_update_request(&dtpm_cpu->qos_req, freq);
64
65 power_limit = pd->table[i - 1].power * nr_cpus;
66
67 return power_limit;
68}
69
70static u64 scale_pd_power_uw(struct cpumask *pd_mask, u64 power)
71{
72 unsigned long max, sum_util = 0;
73 int cpu;
74
75 /*
76 * The capacity is the same for all CPUs belonging to
77 * the same perf domain.
78 */
79 max = arch_scale_cpu_capacity(cpumask_first(pd_mask));
80
81 for_each_cpu_and(cpu, pd_mask, cpu_online_mask)
82 sum_util += sched_cpu_util(cpu);
83
84 return (power * ((sum_util << 10) / max)) >> 10;
85}
86
87static u64 get_pd_power_uw(struct dtpm *dtpm)
88{
89 struct dtpm_cpu *dtpm_cpu = to_dtpm_cpu(dtpm);
90 struct em_perf_domain *pd;
91 struct cpumask *pd_mask;
92 unsigned long freq;
93 int i;
94
95 pd = em_cpu_get(dtpm_cpu->cpu);
96
97 pd_mask = em_span_cpus(pd);
98
99 freq = cpufreq_quick_get(dtpm_cpu->cpu);
100
101 for (i = 0; i < pd->nr_perf_states; i++) {
102
103 if (pd->table[i].frequency < freq)
104 continue;
105
106 return scale_pd_power_uw(pd_mask, pd->table[i].power);
107 }
108
109 return 0;
110}
111
112static int update_pd_power_uw(struct dtpm *dtpm)
113{
114 struct dtpm_cpu *dtpm_cpu = to_dtpm_cpu(dtpm);
115 struct em_perf_domain *em = em_cpu_get(dtpm_cpu->cpu);
116 struct cpumask cpus;
117 int nr_cpus;
118
119 cpumask_and(&cpus, cpu_online_mask, to_cpumask(em->cpus));
120 nr_cpus = cpumask_weight(&cpus);
121
122 dtpm->power_min = em->table[0].power;
123 dtpm->power_min *= nr_cpus;
124
125 dtpm->power_max = em->table[em->nr_perf_states - 1].power;
126 dtpm->power_max *= nr_cpus;
127
128 return 0;
129}
130
131static void pd_release(struct dtpm *dtpm)
132{
133 struct dtpm_cpu *dtpm_cpu = to_dtpm_cpu(dtpm);
134 struct cpufreq_policy *policy;
135
136 if (freq_qos_request_active(&dtpm_cpu->qos_req))
137 freq_qos_remove_request(&dtpm_cpu->qos_req);
138
139 policy = cpufreq_cpu_get(dtpm_cpu->cpu);
140 if (policy) {
141 for_each_cpu(dtpm_cpu->cpu, policy->related_cpus)
142 per_cpu(dtpm_per_cpu, dtpm_cpu->cpu) = NULL;
143
144 cpufreq_cpu_put(policy);
145 }
146
147 kfree(dtpm_cpu);
148}
149
150static struct dtpm_ops dtpm_ops = {
151 .set_power_uw = set_pd_power_limit,
152 .get_power_uw = get_pd_power_uw,
153 .update_power_uw = update_pd_power_uw,
154 .release = pd_release,
155};
156
157static int cpuhp_dtpm_cpu_offline(unsigned int cpu)
158{
159 struct dtpm_cpu *dtpm_cpu;
160
161 dtpm_cpu = per_cpu(dtpm_per_cpu, cpu);
162 if (dtpm_cpu)
163 dtpm_update_power(&dtpm_cpu->dtpm);
164
165 return 0;
166}
167
168static int cpuhp_dtpm_cpu_online(unsigned int cpu)
169{
170 struct dtpm_cpu *dtpm_cpu;
171
172 dtpm_cpu = per_cpu(dtpm_per_cpu, cpu);
173 if (dtpm_cpu)
174 return dtpm_update_power(&dtpm_cpu->dtpm);
175
176 return 0;
177}
178
179static int __dtpm_cpu_setup(int cpu, struct dtpm *parent)
180{
181 struct dtpm_cpu *dtpm_cpu;
182 struct cpufreq_policy *policy;
183 struct em_perf_domain *pd;
184 char name[CPUFREQ_NAME_LEN];
185 int ret = -ENOMEM;
186
187 dtpm_cpu = per_cpu(dtpm_per_cpu, cpu);
188 if (dtpm_cpu)
189 return 0;
190
191 policy = cpufreq_cpu_get(cpu);
192 if (!policy)
193 return 0;
194
195 pd = em_cpu_get(cpu);
196 if (!pd || em_is_artificial(pd)) {
197 ret = -EINVAL;
198 goto release_policy;
199 }
200
201 dtpm_cpu = kzalloc(sizeof(*dtpm_cpu), GFP_KERNEL);
202 if (!dtpm_cpu) {
203 ret = -ENOMEM;
204 goto release_policy;
205 }
206
207 dtpm_init(&dtpm_cpu->dtpm, &dtpm_ops);
208 dtpm_cpu->cpu = cpu;
209
210 for_each_cpu(cpu, policy->related_cpus)
211 per_cpu(dtpm_per_cpu, cpu) = dtpm_cpu;
212
213 snprintf(name, sizeof(name), "cpu%d-cpufreq", dtpm_cpu->cpu);
214
215 ret = dtpm_register(name, &dtpm_cpu->dtpm, parent);
216 if (ret)
217 goto out_kfree_dtpm_cpu;
218
219 ret = freq_qos_add_request(&policy->constraints,
220 &dtpm_cpu->qos_req, FREQ_QOS_MAX,
221 pd->table[pd->nr_perf_states - 1].frequency);
222 if (ret)
223 goto out_dtpm_unregister;
224
225 cpufreq_cpu_put(policy);
226 return 0;
227
228out_dtpm_unregister:
229 dtpm_unregister(&dtpm_cpu->dtpm);
230 dtpm_cpu = NULL;
231
232out_kfree_dtpm_cpu:
233 for_each_cpu(cpu, policy->related_cpus)
234 per_cpu(dtpm_per_cpu, cpu) = NULL;
235 kfree(dtpm_cpu);
236
237release_policy:
238 cpufreq_cpu_put(policy);
239 return ret;
240}
241
242static int dtpm_cpu_setup(struct dtpm *dtpm, struct device_node *np)
243{
244 int cpu;
245
246 cpu = of_cpu_node_to_id(np);
247 if (cpu < 0)
248 return 0;
249
250 return __dtpm_cpu_setup(cpu, dtpm);
251}
252
253static int dtpm_cpu_init(void)
254{
255 int ret;
256
257 /*
258 * The callbacks at CPU hotplug time are calling
259 * dtpm_update_power() which in turns calls update_pd_power().
260 *
261 * The function update_pd_power() uses the online mask to
262 * figure out the power consumption limits.
263 *
264 * At CPUHP_AP_ONLINE_DYN, the CPU is present in the CPU
265 * online mask when the cpuhp_dtpm_cpu_online function is
266 * called, but the CPU is still in the online mask for the
267 * tear down callback. So the power can not be updated when
268 * the CPU is unplugged.
269 *
270 * At CPUHP_AP_DTPM_CPU_DEAD, the situation is the opposite as
271 * above. The CPU online mask is not up to date when the CPU
272 * is plugged in.
273 *
274 * For this reason, we need to call the online and offline
275 * callbacks at different moments when the CPU online mask is
276 * consistent with the power numbers we want to update.
277 */
278 ret = cpuhp_setup_state(CPUHP_AP_DTPM_CPU_DEAD, "dtpm_cpu:offline",
279 NULL, cpuhp_dtpm_cpu_offline);
280 if (ret < 0)
281 return ret;
282
283 ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "dtpm_cpu:online",
284 cpuhp_dtpm_cpu_online, NULL);
285 if (ret < 0)
286 return ret;
287
288 return 0;
289}
290
291static void dtpm_cpu_exit(void)
292{
293 cpuhp_remove_state_nocalls(CPUHP_AP_ONLINE_DYN);
294 cpuhp_remove_state_nocalls(CPUHP_AP_DTPM_CPU_DEAD);
295}
296
297struct dtpm_subsys_ops dtpm_cpu_ops = {
298 .name = KBUILD_MODNAME,
299 .init = dtpm_cpu_init,
300 .exit = dtpm_cpu_exit,
301 .setup = dtpm_cpu_setup,
302};
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright 2020 Linaro Limited
4 *
5 * Author: Daniel Lezcano <daniel.lezcano@linaro.org>
6 *
7 * The DTPM CPU is based on the energy model. It hooks the CPU in the
8 * DTPM tree which in turns update the power number by propagating the
9 * power number from the CPU energy model information to the parents.
10 *
11 * The association between the power and the performance state, allows
12 * to set the power of the CPU at the OPP granularity.
13 *
14 * The CPU hotplug is supported and the power numbers will be updated
15 * if a CPU is hot plugged / unplugged.
16 */
17#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
18
19#include <linux/cpumask.h>
20#include <linux/cpufreq.h>
21#include <linux/cpuhotplug.h>
22#include <linux/dtpm.h>
23#include <linux/energy_model.h>
24#include <linux/of.h>
25#include <linux/pm_qos.h>
26#include <linux/slab.h>
27#include <linux/units.h>
28
29struct dtpm_cpu {
30 struct dtpm dtpm;
31 struct freq_qos_request qos_req;
32 int cpu;
33};
34
35static DEFINE_PER_CPU(struct dtpm_cpu *, dtpm_per_cpu);
36
37static struct dtpm_cpu *to_dtpm_cpu(struct dtpm *dtpm)
38{
39 return container_of(dtpm, struct dtpm_cpu, dtpm);
40}
41
42static u64 set_pd_power_limit(struct dtpm *dtpm, u64 power_limit)
43{
44 struct dtpm_cpu *dtpm_cpu = to_dtpm_cpu(dtpm);
45 struct em_perf_domain *pd = em_cpu_get(dtpm_cpu->cpu);
46 struct cpumask cpus;
47 unsigned long freq;
48 u64 power;
49 int i, nr_cpus;
50
51 cpumask_and(&cpus, cpu_online_mask, to_cpumask(pd->cpus));
52 nr_cpus = cpumask_weight(&cpus);
53
54 for (i = 0; i < pd->nr_perf_states; i++) {
55
56 power = pd->table[i].power * nr_cpus;
57
58 if (power > power_limit)
59 break;
60 }
61
62 freq = pd->table[i - 1].frequency;
63
64 freq_qos_update_request(&dtpm_cpu->qos_req, freq);
65
66 power_limit = pd->table[i - 1].power * nr_cpus;
67
68 return power_limit;
69}
70
71static u64 scale_pd_power_uw(struct cpumask *pd_mask, u64 power)
72{
73 unsigned long max, sum_util = 0;
74 int cpu;
75
76 /*
77 * The capacity is the same for all CPUs belonging to
78 * the same perf domain.
79 */
80 max = arch_scale_cpu_capacity(cpumask_first(pd_mask));
81
82 for_each_cpu_and(cpu, pd_mask, cpu_online_mask)
83 sum_util += sched_cpu_util(cpu);
84
85 return (power * ((sum_util << 10) / max)) >> 10;
86}
87
88static u64 get_pd_power_uw(struct dtpm *dtpm)
89{
90 struct dtpm_cpu *dtpm_cpu = to_dtpm_cpu(dtpm);
91 struct em_perf_domain *pd;
92 struct cpumask *pd_mask;
93 unsigned long freq;
94 int i;
95
96 pd = em_cpu_get(dtpm_cpu->cpu);
97
98 pd_mask = em_span_cpus(pd);
99
100 freq = cpufreq_quick_get(dtpm_cpu->cpu);
101
102 for (i = 0; i < pd->nr_perf_states; i++) {
103
104 if (pd->table[i].frequency < freq)
105 continue;
106
107 return scale_pd_power_uw(pd_mask, pd->table[i].power *
108 MICROWATT_PER_MILLIWATT);
109 }
110
111 return 0;
112}
113
114static int update_pd_power_uw(struct dtpm *dtpm)
115{
116 struct dtpm_cpu *dtpm_cpu = to_dtpm_cpu(dtpm);
117 struct em_perf_domain *em = em_cpu_get(dtpm_cpu->cpu);
118 struct cpumask cpus;
119 int nr_cpus;
120
121 cpumask_and(&cpus, cpu_online_mask, to_cpumask(em->cpus));
122 nr_cpus = cpumask_weight(&cpus);
123
124 dtpm->power_min = em->table[0].power;
125 dtpm->power_min *= MICROWATT_PER_MILLIWATT;
126 dtpm->power_min *= nr_cpus;
127
128 dtpm->power_max = em->table[em->nr_perf_states - 1].power;
129 dtpm->power_max *= MICROWATT_PER_MILLIWATT;
130 dtpm->power_max *= nr_cpus;
131
132 return 0;
133}
134
135static void pd_release(struct dtpm *dtpm)
136{
137 struct dtpm_cpu *dtpm_cpu = to_dtpm_cpu(dtpm);
138 struct cpufreq_policy *policy;
139
140 if (freq_qos_request_active(&dtpm_cpu->qos_req))
141 freq_qos_remove_request(&dtpm_cpu->qos_req);
142
143 policy = cpufreq_cpu_get(dtpm_cpu->cpu);
144 if (policy) {
145 for_each_cpu(dtpm_cpu->cpu, policy->related_cpus)
146 per_cpu(dtpm_per_cpu, dtpm_cpu->cpu) = NULL;
147 }
148
149 kfree(dtpm_cpu);
150}
151
152static struct dtpm_ops dtpm_ops = {
153 .set_power_uw = set_pd_power_limit,
154 .get_power_uw = get_pd_power_uw,
155 .update_power_uw = update_pd_power_uw,
156 .release = pd_release,
157};
158
159static int cpuhp_dtpm_cpu_offline(unsigned int cpu)
160{
161 struct dtpm_cpu *dtpm_cpu;
162
163 dtpm_cpu = per_cpu(dtpm_per_cpu, cpu);
164 if (dtpm_cpu)
165 dtpm_update_power(&dtpm_cpu->dtpm);
166
167 return 0;
168}
169
170static int cpuhp_dtpm_cpu_online(unsigned int cpu)
171{
172 struct dtpm_cpu *dtpm_cpu;
173
174 dtpm_cpu = per_cpu(dtpm_per_cpu, cpu);
175 if (dtpm_cpu)
176 return dtpm_update_power(&dtpm_cpu->dtpm);
177
178 return 0;
179}
180
181static int __dtpm_cpu_setup(int cpu, struct dtpm *parent)
182{
183 struct dtpm_cpu *dtpm_cpu;
184 struct cpufreq_policy *policy;
185 struct em_perf_domain *pd;
186 char name[CPUFREQ_NAME_LEN];
187 int ret = -ENOMEM;
188
189 dtpm_cpu = per_cpu(dtpm_per_cpu, cpu);
190 if (dtpm_cpu)
191 return 0;
192
193 policy = cpufreq_cpu_get(cpu);
194 if (!policy)
195 return 0;
196
197 pd = em_cpu_get(cpu);
198 if (!pd || em_is_artificial(pd))
199 return -EINVAL;
200
201 dtpm_cpu = kzalloc(sizeof(*dtpm_cpu), GFP_KERNEL);
202 if (!dtpm_cpu)
203 return -ENOMEM;
204
205 dtpm_init(&dtpm_cpu->dtpm, &dtpm_ops);
206 dtpm_cpu->cpu = cpu;
207
208 for_each_cpu(cpu, policy->related_cpus)
209 per_cpu(dtpm_per_cpu, cpu) = dtpm_cpu;
210
211 snprintf(name, sizeof(name), "cpu%d-cpufreq", dtpm_cpu->cpu);
212
213 ret = dtpm_register(name, &dtpm_cpu->dtpm, parent);
214 if (ret)
215 goto out_kfree_dtpm_cpu;
216
217 ret = freq_qos_add_request(&policy->constraints,
218 &dtpm_cpu->qos_req, FREQ_QOS_MAX,
219 pd->table[pd->nr_perf_states - 1].frequency);
220 if (ret)
221 goto out_dtpm_unregister;
222
223 return 0;
224
225out_dtpm_unregister:
226 dtpm_unregister(&dtpm_cpu->dtpm);
227 dtpm_cpu = NULL;
228
229out_kfree_dtpm_cpu:
230 for_each_cpu(cpu, policy->related_cpus)
231 per_cpu(dtpm_per_cpu, cpu) = NULL;
232 kfree(dtpm_cpu);
233
234 return ret;
235}
236
237static int dtpm_cpu_setup(struct dtpm *dtpm, struct device_node *np)
238{
239 int cpu;
240
241 cpu = of_cpu_node_to_id(np);
242 if (cpu < 0)
243 return 0;
244
245 return __dtpm_cpu_setup(cpu, dtpm);
246}
247
248static int dtpm_cpu_init(void)
249{
250 int ret;
251
252 /*
253 * The callbacks at CPU hotplug time are calling
254 * dtpm_update_power() which in turns calls update_pd_power().
255 *
256 * The function update_pd_power() uses the online mask to
257 * figure out the power consumption limits.
258 *
259 * At CPUHP_AP_ONLINE_DYN, the CPU is present in the CPU
260 * online mask when the cpuhp_dtpm_cpu_online function is
261 * called, but the CPU is still in the online mask for the
262 * tear down callback. So the power can not be updated when
263 * the CPU is unplugged.
264 *
265 * At CPUHP_AP_DTPM_CPU_DEAD, the situation is the opposite as
266 * above. The CPU online mask is not up to date when the CPU
267 * is plugged in.
268 *
269 * For this reason, we need to call the online and offline
270 * callbacks at different moments when the CPU online mask is
271 * consistent with the power numbers we want to update.
272 */
273 ret = cpuhp_setup_state(CPUHP_AP_DTPM_CPU_DEAD, "dtpm_cpu:offline",
274 NULL, cpuhp_dtpm_cpu_offline);
275 if (ret < 0)
276 return ret;
277
278 ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "dtpm_cpu:online",
279 cpuhp_dtpm_cpu_online, NULL);
280 if (ret < 0)
281 return ret;
282
283 return 0;
284}
285
286static void dtpm_cpu_exit(void)
287{
288 cpuhp_remove_state_nocalls(CPUHP_AP_ONLINE_DYN);
289 cpuhp_remove_state_nocalls(CPUHP_AP_DTPM_CPU_DEAD);
290}
291
292struct dtpm_subsys_ops dtpm_cpu_ops = {
293 .name = KBUILD_MODNAME,
294 .init = dtpm_cpu_init,
295 .exit = dtpm_cpu_exit,
296 .setup = dtpm_cpu_setup,
297};