1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * itmt.c: Support Intel Turbo Boost Max Technology 3.0
4 *
5 * (C) Copyright 2016 Intel Corporation
6 * Author: Tim Chen <tim.c.chen@linux.intel.com>
7 *
8 * On platforms supporting Intel Turbo Boost Max Technology 3.0, (ITMT),
9 * the maximum turbo frequencies of some cores in a CPU package may be
10 * higher than for the other cores in the same package. In that case,
11 * better performance can be achieved by making the scheduler prefer
12 * to run tasks on the CPUs with higher max turbo frequencies.
13 *
14 * This file provides functions and data structures for enabling the
15 * scheduler to favor scheduling on cores can be boosted to a higher
16 * frequency under ITMT.
17 */
18
19 #include <linux/sched.h>
20 #include <linux/cpumask.h>
21 #include <linux/cpuset.h>
22 #include <linux/mutex.h>
23 #include <linux/sysctl.h>
24 #include <linux/nodemask.h>
25
26 static DEFINE_MUTEX(itmt_update_mutex);
27 DEFINE_PER_CPU_READ_MOSTLY(int, sched_core_priority);
28
29 /* Boolean to track if system has ITMT capabilities */
30 static bool __read_mostly sched_itmt_capable;
31
32 /*
33 * Boolean to control whether we want to move processes to cpu capable
34 * of higher turbo frequency for cpus supporting Intel Turbo Boost Max
35 * Technology 3.0.
36 *
37 * It can be set via /proc/sys/kernel/sched_itmt_enabled
38 */
39 unsigned int __read_mostly sysctl_sched_itmt_enabled;
40
sched_itmt_update_handler(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)41 static int sched_itmt_update_handler(struct ctl_table *table, int write,
42 void *buffer, size_t *lenp, loff_t *ppos)
43 {
44 unsigned int old_sysctl;
45 int ret;
46
47 mutex_lock(&itmt_update_mutex);
48
49 if (!sched_itmt_capable) {
50 mutex_unlock(&itmt_update_mutex);
51 return -EINVAL;
52 }
53
54 old_sysctl = sysctl_sched_itmt_enabled;
55 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
56
57 if (!ret && write && old_sysctl != sysctl_sched_itmt_enabled) {
58 x86_topology_update = true;
59 rebuild_sched_domains();
60 }
61
62 mutex_unlock(&itmt_update_mutex);
63
64 return ret;
65 }
66
67 static struct ctl_table itmt_kern_table[] = {
68 {
69 .procname = "sched_itmt_enabled",
70 .data = &sysctl_sched_itmt_enabled,
71 .maxlen = sizeof(unsigned int),
72 .mode = 0644,
73 .proc_handler = sched_itmt_update_handler,
74 .extra1 = SYSCTL_ZERO,
75 .extra2 = SYSCTL_ONE,
76 },
77 {}
78 };
79
80 static struct ctl_table itmt_root_table[] = {
81 {
82 .procname = "kernel",
83 .mode = 0555,
84 .child = itmt_kern_table,
85 },
86 {}
87 };
88
89 static struct ctl_table_header *itmt_sysctl_header;
90
91 /**
92 * sched_set_itmt_support() - Indicate platform supports ITMT
93 *
94 * This function is used by the OS to indicate to scheduler that the platform
95 * is capable of supporting the ITMT feature.
96 *
97 * The current scheme has the pstate driver detects if the system
98 * is ITMT capable and call sched_set_itmt_support.
99 *
100 * This must be done only after sched_set_itmt_core_prio
101 * has been called to set the cpus' priorities.
102 * It must not be called with cpu hot plug lock
103 * held as we need to acquire the lock to rebuild sched domains
104 * later.
105 *
106 * Return: 0 on success
107 */
sched_set_itmt_support(void)108 int sched_set_itmt_support(void)
109 {
110 mutex_lock(&itmt_update_mutex);
111
112 if (sched_itmt_capable) {
113 mutex_unlock(&itmt_update_mutex);
114 return 0;
115 }
116
117 itmt_sysctl_header = register_sysctl_table(itmt_root_table);
118 if (!itmt_sysctl_header) {
119 mutex_unlock(&itmt_update_mutex);
120 return -ENOMEM;
121 }
122
123 sched_itmt_capable = true;
124
125 sysctl_sched_itmt_enabled = 1;
126
127 x86_topology_update = true;
128 rebuild_sched_domains();
129
130 mutex_unlock(&itmt_update_mutex);
131
132 return 0;
133 }
134
135 /**
136 * sched_clear_itmt_support() - Revoke platform's support of ITMT
137 *
138 * This function is used by the OS to indicate that it has
139 * revoked the platform's support of ITMT feature.
140 *
141 * It must not be called with cpu hot plug lock
142 * held as we need to acquire the lock to rebuild sched domains
143 * later.
144 */
sched_clear_itmt_support(void)145 void sched_clear_itmt_support(void)
146 {
147 mutex_lock(&itmt_update_mutex);
148
149 if (!sched_itmt_capable) {
150 mutex_unlock(&itmt_update_mutex);
151 return;
152 }
153 sched_itmt_capable = false;
154
155 if (itmt_sysctl_header) {
156 unregister_sysctl_table(itmt_sysctl_header);
157 itmt_sysctl_header = NULL;
158 }
159
160 if (sysctl_sched_itmt_enabled) {
161 /* disable sched_itmt if we are no longer ITMT capable */
162 sysctl_sched_itmt_enabled = 0;
163 x86_topology_update = true;
164 rebuild_sched_domains();
165 }
166
167 mutex_unlock(&itmt_update_mutex);
168 }
169
arch_asym_cpu_priority(int cpu)170 int arch_asym_cpu_priority(int cpu)
171 {
172 return per_cpu(sched_core_priority, cpu);
173 }
174
175 /**
176 * sched_set_itmt_core_prio() - Set CPU priority based on ITMT
177 * @prio: Priority of cpu core
178 * @core_cpu: The cpu number associated with the core
179 *
180 * The pstate driver will find out the max boost frequency
181 * and call this function to set a priority proportional
182 * to the max boost frequency. CPU with higher boost
183 * frequency will receive higher priority.
184 *
185 * No need to rebuild sched domain after updating
186 * the CPU priorities. The sched domains have no
187 * dependency on CPU priorities.
188 */
sched_set_itmt_core_prio(int prio,int core_cpu)189 void sched_set_itmt_core_prio(int prio, int core_cpu)
190 {
191 int cpu, i = 1;
192
193 for_each_cpu(cpu, topology_sibling_cpumask(core_cpu)) {
194 int smt_prio;
195
196 /*
197 * Ensure that the siblings are moved to the end
198 * of the priority chain and only used when
199 * all other high priority cpus are out of capacity.
200 */
201 smt_prio = prio * smp_num_siblings / i;
202 per_cpu(sched_core_priority, cpu) = smt_prio;
203 i++;
204 }
205 }
206