xref: /openbmc/linux/arch/x86/kernel/itmt.c (revision fa0dadde)
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 
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_header *itmt_sysctl_header;
81 
82 /**
83  * sched_set_itmt_support() - Indicate platform supports ITMT
84  *
85  * This function is used by the OS to indicate to scheduler that the platform
86  * is capable of supporting the ITMT feature.
87  *
88  * The current scheme has the pstate driver detects if the system
89  * is ITMT capable and call sched_set_itmt_support.
90  *
91  * This must be done only after sched_set_itmt_core_prio
92  * has been called to set the cpus' priorities.
93  * It must not be called with cpu hot plug lock
94  * held as we need to acquire the lock to rebuild sched domains
95  * later.
96  *
97  * Return: 0 on success
98  */
99 int sched_set_itmt_support(void)
100 {
101 	mutex_lock(&itmt_update_mutex);
102 
103 	if (sched_itmt_capable) {
104 		mutex_unlock(&itmt_update_mutex);
105 		return 0;
106 	}
107 
108 	itmt_sysctl_header = register_sysctl("kernel", itmt_kern_table);
109 	if (!itmt_sysctl_header) {
110 		mutex_unlock(&itmt_update_mutex);
111 		return -ENOMEM;
112 	}
113 
114 	sched_itmt_capable = true;
115 
116 	sysctl_sched_itmt_enabled = 1;
117 
118 	x86_topology_update = true;
119 	rebuild_sched_domains();
120 
121 	mutex_unlock(&itmt_update_mutex);
122 
123 	return 0;
124 }
125 
126 /**
127  * sched_clear_itmt_support() - Revoke platform's support of ITMT
128  *
129  * This function is used by the OS to indicate that it has
130  * revoked the platform's support of ITMT feature.
131  *
132  * It must not be called with cpu hot plug lock
133  * held as we need to acquire the lock to rebuild sched domains
134  * later.
135  */
136 void sched_clear_itmt_support(void)
137 {
138 	mutex_lock(&itmt_update_mutex);
139 
140 	if (!sched_itmt_capable) {
141 		mutex_unlock(&itmt_update_mutex);
142 		return;
143 	}
144 	sched_itmt_capable = false;
145 
146 	if (itmt_sysctl_header) {
147 		unregister_sysctl_table(itmt_sysctl_header);
148 		itmt_sysctl_header = NULL;
149 	}
150 
151 	if (sysctl_sched_itmt_enabled) {
152 		/* disable sched_itmt if we are no longer ITMT capable */
153 		sysctl_sched_itmt_enabled = 0;
154 		x86_topology_update = true;
155 		rebuild_sched_domains();
156 	}
157 
158 	mutex_unlock(&itmt_update_mutex);
159 }
160 
161 int arch_asym_cpu_priority(int cpu)
162 {
163 	return per_cpu(sched_core_priority, cpu);
164 }
165 
166 /**
167  * sched_set_itmt_core_prio() - Set CPU priority based on ITMT
168  * @prio:	Priority of cpu core
169  * @core_cpu:	The cpu number associated with the core
170  *
171  * The pstate driver will find out the max boost frequency
172  * and call this function to set a priority proportional
173  * to the max boost frequency. CPU with higher boost
174  * frequency will receive higher priority.
175  *
176  * No need to rebuild sched domain after updating
177  * the CPU priorities. The sched domains have no
178  * dependency on CPU priorities.
179  */
180 void sched_set_itmt_core_prio(int prio, int core_cpu)
181 {
182 	int cpu, i = 1;
183 
184 	for_each_cpu(cpu, topology_sibling_cpumask(core_cpu)) {
185 		int smt_prio;
186 
187 		/*
188 		 * Ensure that the siblings are moved to the end
189 		 * of the priority chain and only used when
190 		 * all other high priority cpus are out of capacity.
191 		 */
192 		smt_prio = prio * smp_num_siblings / (i * i);
193 		per_cpu(sched_core_priority, cpu) = smt_prio;
194 		i++;
195 	}
196 }
197