For multi-cluster platforms like i.MX8QM, the best cpufreq governor is schedutil, as common cpufreq framework decides default cpufreq governor in static compile, so this patch adds dynamic switch of cpufreq governor according to cluster number, changing it via sysfs interface, although it is ugly, but it realizes dynamic cpufreq governor select for users. Signed-off-by: Anson Huang <Anson.Huang@nxp.com>
294 lines
7.8 KiB
C
294 lines
7.8 KiB
C
/*
|
|
* Copyright 2017 NXP
|
|
*
|
|
* This program is free software; you can redistribute it and/or modify
|
|
* it under the terms of the GNU General Public License version 2 as
|
|
* published by the Free Software Foundation.
|
|
*/
|
|
|
|
#include <linux/arm-smccc.h>
|
|
#include <linux/clk.h>
|
|
#include <linux/cpu.h>
|
|
#include <linux/cpufreq.h>
|
|
#include <linux/cpu_cooling.h>
|
|
#include <linux/err.h>
|
|
#include <linux/module.h>
|
|
#include <linux/of.h>
|
|
#include <linux/pm_opp.h>
|
|
#include <linux/platform_device.h>
|
|
#include <linux/regulator/consumer.h>
|
|
#include <linux/syscalls.h>
|
|
#include <soc/imx/fsl_sip.h>
|
|
|
|
#define MAX_CLUSTER_NUM 2
|
|
|
|
static struct delayed_work cpufreq_governor_daemon;
|
|
|
|
static DEFINE_SPINLOCK(cpufreq_psci_lock);
|
|
|
|
struct imx8_cpufreq {
|
|
struct clk *cpu_clk;
|
|
};
|
|
|
|
struct imx8_cpufreq cluster_freq[MAX_CLUSTER_NUM];
|
|
static struct cpufreq_frequency_table *freq_table[MAX_CLUSTER_NUM];
|
|
static unsigned int transition_latency[MAX_CLUSTER_NUM];
|
|
struct device *cpu_dev;
|
|
static struct thermal_cooling_device *cdev[2];
|
|
|
|
static void cpufreq_governor_daemon_handler(struct work_struct *work)
|
|
{
|
|
int fd, i;
|
|
unsigned char cluster_governor[MAX_CLUSTER_NUM][54] = {
|
|
"/sys/devices/system/cpu/cpu0/cpufreq/scaling_governor",
|
|
"",
|
|
};
|
|
|
|
/* generate second cluster's cpufreq governor path */
|
|
sprintf(cluster_governor[MAX_CLUSTER_NUM - 1],
|
|
"%s%d%s", "/sys/devices/system/cpu/cpu", num_online_cpus() - 1,
|
|
"/cpufreq/scaling_governor");
|
|
|
|
for (i = 0; i < MAX_CLUSTER_NUM; i++) {
|
|
fd = sys_open((const char __user __force *)cluster_governor[i],
|
|
O_RDWR, 0700);
|
|
if (fd >= 0) {
|
|
sys_write(fd, "schedutil", strlen("schedutil"));
|
|
sys_close(fd);
|
|
pr_info("switch cluster %d cpu-freq governor to schedutil\n",
|
|
i);
|
|
} else {
|
|
/* re-schedule if sys write is NOT ready */
|
|
schedule_delayed_work(&cpufreq_governor_daemon,
|
|
msecs_to_jiffies(3000));
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
static int imx8_set_target(struct cpufreq_policy *policy, unsigned int index)
|
|
{
|
|
struct arm_smccc_res res;
|
|
unsigned int old_freq, new_freq;
|
|
unsigned int cluster_id = topology_physical_package_id(policy->cpu);
|
|
|
|
new_freq = freq_table[cluster_id][index].frequency;
|
|
old_freq = policy->cur;
|
|
|
|
dev_dbg(cpu_dev, "%u MHz --> %u MHz\n",
|
|
old_freq / 1000, new_freq / 1000);
|
|
|
|
spin_lock(&cpufreq_psci_lock);
|
|
arm_smccc_smc(FSL_SIP_CPUFREQ, FSL_SIP_SET_CPUFREQ,
|
|
cluster_id, new_freq * 1000, 0, 0, 0, 0, &res);
|
|
spin_unlock(&cpufreq_psci_lock);
|
|
|
|
/*
|
|
* As we can only set CPU clock rate in ATF, clock
|
|
* framework does NOT know CPU clock rate is changed,
|
|
* so here do clk_get_rate once to update CPU clock
|
|
* rate, otherwise cat /sys/kernel/debug/clk/xxx/clk_rate
|
|
* will return incorrect rate as it does NOT do a
|
|
* recalculation.
|
|
*/
|
|
clk_get_rate(cluster_freq[cluster_id].cpu_clk);
|
|
|
|
return 0;
|
|
}
|
|
|
|
static int imx8_cpufreq_init(struct cpufreq_policy *policy)
|
|
{
|
|
int cluster_id = topology_physical_package_id(policy->cpu);
|
|
int ret = 0;
|
|
|
|
policy->clk = cluster_freq[cluster_id].cpu_clk;
|
|
policy->cur = clk_get_rate(cluster_freq[cluster_id].cpu_clk) / 1000;
|
|
/*
|
|
* The driver only supports the SMP configuartion where all processors
|
|
* share the clock and voltage and clock.
|
|
*/
|
|
cpumask_copy(policy->cpus, topology_core_cpumask(policy->cpu));
|
|
|
|
ret = cpufreq_table_validate_and_show(policy, freq_table[cluster_id]);
|
|
if (ret) {
|
|
pr_err("%s: invalid frequency table: %d\n", __func__, ret);
|
|
return ret;
|
|
}
|
|
|
|
policy->cpuinfo.transition_latency = transition_latency[cluster_id];
|
|
policy->suspend_freq = policy->max;
|
|
|
|
pr_info("%s: cluster %d running at freq %d MHz, suspend freq %d MHz\n",
|
|
__func__, cluster_id, policy->cur / 1000,
|
|
policy->suspend_freq / 1000);
|
|
|
|
return ret;
|
|
}
|
|
|
|
static void imx8_cpufreq_ready(struct cpufreq_policy *policy)
|
|
{
|
|
struct device_node *np = of_get_cpu_node(policy->cpu, NULL);
|
|
unsigned int cluster_id = topology_physical_package_id(policy->cpu);
|
|
|
|
if (of_find_property(np, "#cooling-cells", NULL)) {
|
|
cdev[cluster_id] = of_cpufreq_cooling_register(np, policy);
|
|
|
|
if (IS_ERR(cdev[cluster_id]) && PTR_ERR(cdev[cluster_id]) != -ENOSYS) {
|
|
pr_err("cpu%d is not running as cooling device: %ld\n",
|
|
policy->cpu, PTR_ERR(cdev[cluster_id]));
|
|
|
|
cdev[cluster_id] = NULL;
|
|
}
|
|
}
|
|
|
|
of_node_put(np);
|
|
}
|
|
|
|
static struct cpufreq_driver imx8_cpufreq_driver = {
|
|
.flags = CPUFREQ_NEED_INITIAL_FREQ_CHECK,
|
|
.verify = cpufreq_generic_frequency_table_verify,
|
|
.target_index = imx8_set_target,
|
|
.get = cpufreq_generic_get,
|
|
.init = imx8_cpufreq_init,
|
|
.name = "imx8-cpufreq",
|
|
.attr = cpufreq_generic_attr,
|
|
.ready = imx8_cpufreq_ready,
|
|
#ifdef CONFIG_PM
|
|
.suspend = cpufreq_generic_suspend,
|
|
#endif
|
|
};
|
|
|
|
static int imx8_cpufreq_probe(struct platform_device *pdev)
|
|
{
|
|
struct device_node *np;
|
|
int ret = 0;
|
|
int i, cluster_id;
|
|
struct device *first_cpu_dev = NULL;
|
|
|
|
cpu_dev = get_cpu_device(0);
|
|
|
|
if (!cpu_dev) {
|
|
pr_err("failed to get cpu device 0\n");
|
|
return -ENODEV;
|
|
}
|
|
|
|
np = of_node_get(cpu_dev->of_node);
|
|
if (!np) {
|
|
pr_warn("failed to find cpu 0 node\n");
|
|
return -ENODEV;
|
|
}
|
|
|
|
ret = dev_pm_opp_of_add_table(cpu_dev);
|
|
if (ret < 0) {
|
|
dev_err(cpu_dev, "failed to init OPP table: %d\n", ret);
|
|
goto put_node;
|
|
}
|
|
|
|
cluster_id = topology_physical_package_id(0);
|
|
cluster_freq[cluster_id].cpu_clk = devm_clk_get(cpu_dev, NULL);
|
|
if (IS_ERR(cluster_freq[cluster_id].cpu_clk)) {
|
|
dev_err(cpu_dev, "failed to get cluster %d clock\n", cluster_id);
|
|
ret = -ENOENT;
|
|
goto put_node;
|
|
}
|
|
|
|
ret = dev_pm_opp_init_cpufreq_table(cpu_dev, &freq_table[cluster_id]);
|
|
if (ret) {
|
|
dev_err(cpu_dev, "failed to init cpufreq table: %d\n", ret);
|
|
goto out_free_opp;
|
|
}
|
|
|
|
if (of_property_read_u32(np, "clock-latency", &transition_latency[cluster_id]))
|
|
transition_latency[cluster_id] = CPUFREQ_ETERNAL;
|
|
|
|
/* init next cluster if there is */
|
|
for (i = 1; i < num_online_cpus(); i++) {
|
|
if (topology_physical_package_id(i) == topology_physical_package_id(0))
|
|
continue;
|
|
|
|
INIT_DELAYED_WORK(&cpufreq_governor_daemon,
|
|
cpufreq_governor_daemon_handler);
|
|
schedule_delayed_work(&cpufreq_governor_daemon,
|
|
msecs_to_jiffies(3000));
|
|
first_cpu_dev = cpu_dev;
|
|
cpu_dev = get_cpu_device(i);
|
|
if (!cpu_dev) {
|
|
pr_err("failed to get cpu device %d\n", i);
|
|
return -ENODEV;
|
|
}
|
|
|
|
np = of_node_get(cpu_dev->of_node);
|
|
if (!np) {
|
|
pr_warn("failed to find cpu %d node\n", i);
|
|
ret = -ENODEV;
|
|
goto put_node;
|
|
}
|
|
|
|
cluster_id = topology_physical_package_id(i);
|
|
cluster_freq[cluster_id].cpu_clk = devm_clk_get(cpu_dev, NULL);
|
|
if (IS_ERR(cluster_freq[cluster_id].cpu_clk)) {
|
|
dev_err(cpu_dev, "failed to get cluster %d clock\n", cluster_id);
|
|
ret = -ENOENT;
|
|
goto put_node;
|
|
}
|
|
|
|
ret = dev_pm_opp_of_add_table(cpu_dev);
|
|
if (ret < 0) {
|
|
dev_err(cpu_dev, "failed to init OPP table: %d\n", ret);
|
|
goto put_node;
|
|
}
|
|
|
|
ret = dev_pm_opp_init_cpufreq_table(cpu_dev, &freq_table[cluster_id]);
|
|
if (ret) {
|
|
dev_err(cpu_dev, "failed to init cpufreq table: %d\n", ret);
|
|
goto out_free_opp;
|
|
}
|
|
|
|
if (of_property_read_u32(np, "clock-latency", &transition_latency[cluster_id]))
|
|
transition_latency[cluster_id] = CPUFREQ_ETERNAL;
|
|
break;
|
|
}
|
|
|
|
ret = cpufreq_register_driver(&imx8_cpufreq_driver);
|
|
if (ret) {
|
|
dev_err(cpu_dev, "failed register driver: %d\n", ret);
|
|
if (cluster_id > 0 && first_cpu_dev != NULL) {
|
|
dev_pm_opp_free_cpufreq_table(first_cpu_dev, &freq_table[0]);
|
|
dev_pm_opp_of_remove_table(first_cpu_dev);
|
|
}
|
|
goto free_freq_table;
|
|
}
|
|
|
|
of_node_put(np);
|
|
|
|
return 0;
|
|
|
|
free_freq_table:
|
|
dev_pm_opp_free_cpufreq_table(cpu_dev, &freq_table[cluster_id]);
|
|
out_free_opp:
|
|
dev_pm_opp_of_remove_table(cpu_dev);
|
|
put_node:
|
|
of_node_put(np);
|
|
return ret;
|
|
}
|
|
|
|
static int imx8_cpufreq_remove(struct platform_device *pdev)
|
|
{
|
|
cpufreq_unregister_driver(&imx8_cpufreq_driver);
|
|
|
|
return 0;
|
|
}
|
|
|
|
static struct platform_driver imx8_cpufreq_platdrv = {
|
|
.driver = {
|
|
.name = "imx8-cpufreq",
|
|
},
|
|
.probe = imx8_cpufreq_probe,
|
|
.remove = imx8_cpufreq_remove,
|
|
};
|
|
module_platform_driver(imx8_cpufreq_platdrv);
|
|
|
|
MODULE_AUTHOR("Anson Huang <Anson.Huang@nxp.com>");
|
|
MODULE_DESCRIPTION("NXP i.MX8 cpufreq driver");
|
|
MODULE_LICENSE("GPL");
|