1/*
  2 * itmt.c: Support Intel Turbo Boost Max Technology 3.0
  3 *
  4 * (C) Copyright 2016 Intel Corporation
  5 * Author: Tim Chen <tim.c.chen@linux.intel.com>
  6 *
  7 * This program is free software; you can redistribute it and/or
  8 * modify it under the terms of the GNU General Public License
  9 * as published by the Free Software Foundation; version 2
 10 * of the License.
 11 *
 12 * On platforms supporting Intel Turbo Boost Max Technology 3.0, (ITMT),
 13 * the maximum turbo frequencies of some cores in a CPU package may be
 14 * higher than for the other cores in the same package.  In that case,
 15 * better performance can be achieved by making the scheduler prefer
 16 * to run tasks on the CPUs with higher max turbo frequencies.
 17 *
 18 * This file provides functions and data structures for enabling the
 19 * scheduler to favor scheduling on cores can be boosted to a higher
 20 * frequency under ITMT.
 21 */
 22
 23#include <linux/sched.h>
 24#include <linux/cpumask.h>
 25#include <linux/cpuset.h>
 26#include <linux/mutex.h>
 27#include <linux/sysctl.h>
 28#include <linux/nodemask.h>
 29
 30static DEFINE_MUTEX(itmt_update_mutex);
 31DEFINE_PER_CPU_READ_MOSTLY(int, sched_core_priority);
 32
 33/* Boolean to track if system has ITMT capabilities */
 34static bool __read_mostly sched_itmt_capable;
 35
 36/*
 37 * Boolean to control whether we want to move processes to cpu capable
 38 * of higher turbo frequency for cpus supporting Intel Turbo Boost Max
 39 * Technology 3.0.
 40 *
 41 * It can be set via /proc/sys/kernel/sched_itmt_enabled
 42 */
 43unsigned int __read_mostly sysctl_sched_itmt_enabled;
 44
 45static int sched_itmt_update_handler(struct ctl_table *table, int write,
 46				     void __user *buffer, size_t *lenp,
 47				     loff_t *ppos)
 48{
 49	unsigned int old_sysctl;
 50	int ret;
 51
 52	mutex_lock(&itmt_update_mutex);
 53
 54	if (!sched_itmt_capable) {
 55		mutex_unlock(&itmt_update_mutex);
 56		return -EINVAL;
 57	}
 58
 59	old_sysctl = sysctl_sched_itmt_enabled;
 60	ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
 61
 62	if (!ret && write && old_sysctl != sysctl_sched_itmt_enabled) {
 63		x86_topology_update = true;
 64		rebuild_sched_domains();
 65	}
 66
 67	mutex_unlock(&itmt_update_mutex);
 68
 69	return ret;
 70}
 71
 72static unsigned int zero;
 73static unsigned int one = 1;
 74static struct ctl_table itmt_kern_table[] = {
 75	{
 76		.procname	= "sched_itmt_enabled",
 77		.data		= &sysctl_sched_itmt_enabled,
 78		.maxlen		= sizeof(unsigned int),
 79		.mode		= 0644,
 80		.proc_handler	= sched_itmt_update_handler,
 81		.extra1		= &zero,
 82		.extra2		= &one,
 83	},
 84	{}
 85};
 86
 87static struct ctl_table itmt_root_table[] = {
 88	{
 89		.procname	= "kernel",
 90		.mode		= 0555,
 91		.child		= itmt_kern_table,
 92	},
 93	{}
 94};
 95
 96static struct ctl_table_header *itmt_sysctl_header;
 97
 98/**
 99 * sched_set_itmt_support() - Indicate platform supports ITMT
100 *
101 * This function is used by the OS to indicate to scheduler that the platform
102 * is capable of supporting the ITMT feature.
103 *
104 * The current scheme has the pstate driver detects if the system
105 * is ITMT capable and call sched_set_itmt_support.
106 *
107 * This must be done only after sched_set_itmt_core_prio
108 * has been called to set the cpus' priorities.
109 * It must not be called with cpu hot plug lock
110 * held as we need to acquire the lock to rebuild sched domains
111 * later.
112 *
113 * Return: 0 on success
114 */
115int sched_set_itmt_support(void)
116{
117	mutex_lock(&itmt_update_mutex);
118
119	if (sched_itmt_capable) {
120		mutex_unlock(&itmt_update_mutex);
121		return 0;
122	}
123
124	itmt_sysctl_header = register_sysctl_table(itmt_root_table);
125	if (!itmt_sysctl_header) {
126		mutex_unlock(&itmt_update_mutex);
127		return -ENOMEM;
128	}
129
130	sched_itmt_capable = true;
131
132	sysctl_sched_itmt_enabled = 1;
133
134	x86_topology_update = true;
135	rebuild_sched_domains();
136
137	mutex_unlock(&itmt_update_mutex);
138
139	return 0;
140}
141
142/**
143 * sched_clear_itmt_support() - Revoke platform's support of ITMT
144 *
145 * This function is used by the OS to indicate that it has
146 * revoked the platform's support of ITMT feature.
147 *
148 * It must not be called with cpu hot plug lock
149 * held as we need to acquire the lock to rebuild sched domains
150 * later.
151 */
152void sched_clear_itmt_support(void)
153{
154	mutex_lock(&itmt_update_mutex);
155
156	if (!sched_itmt_capable) {
157		mutex_unlock(&itmt_update_mutex);
158		return;
159	}
160	sched_itmt_capable = false;
161
162	if (itmt_sysctl_header) {
163		unregister_sysctl_table(itmt_sysctl_header);
164		itmt_sysctl_header = NULL;
165	}
166
167	if (sysctl_sched_itmt_enabled) {
168		/* disable sched_itmt if we are no longer ITMT capable */
169		sysctl_sched_itmt_enabled = 0;
170		x86_topology_update = true;
171		rebuild_sched_domains();
172	}
173
174	mutex_unlock(&itmt_update_mutex);
175}
176
177int arch_asym_cpu_priority(int cpu)
178{
179	return per_cpu(sched_core_priority, cpu);
180}
181
182/**
183 * sched_set_itmt_core_prio() - Set CPU priority based on ITMT
184 * @prio:	Priority of cpu core
185 * @core_cpu:	The cpu number associated with the core
186 *
187 * The pstate driver will find out the max boost frequency
188 * and call this function to set a priority proportional
189 * to the max boost frequency. CPU with higher boost
190 * frequency will receive higher priority.
191 *
192 * No need to rebuild sched domain after updating
193 * the CPU priorities. The sched domains have no
194 * dependency on CPU priorities.
195 */
196void sched_set_itmt_core_prio(int prio, int core_cpu)
197{
198	int cpu, i = 1;
199
200	for_each_cpu(cpu, topology_sibling_cpumask(core_cpu)) {
201		int smt_prio;
202
203		/*
204		 * Ensure that the siblings are moved to the end
205		 * of the priority chain and only used when
206		 * all other high priority cpus are out of capacity.
207		 */
208		smt_prio = prio * smp_num_siblings / i;
209		per_cpu(sched_core_priority, cpu) = smt_prio;
210		i++;
211	}
212}