linux/drivers/cpufreq/mediatek-cpufreq-hw.c

344 lines
8.2 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0
/*
* Copyright (c) 2020 MediaTek Inc.
*/
#include <linux/bitfield.h>
#include <linux/cpufreq.h>
#include <linux/energy_model.h>
#include <linux/init.h>
#include <linux/iopoll.h>
#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/of_address.h>
#include <linux/of_platform.h>
#include <linux/slab.h>
#define LUT_MAX_ENTRIES 32U
#define LUT_FREQ GENMASK(11, 0)
#define LUT_ROW_SIZE 0x4
#define CPUFREQ_HW_STATUS BIT(0)
#define SVS_HW_STATUS BIT(1)
#define POLL_USEC 1000
#define TIMEOUT_USEC 300000
enum {
REG_FREQ_LUT_TABLE,
REG_FREQ_ENABLE,
REG_FREQ_PERF_STATE,
REG_FREQ_HW_STATE,
REG_EM_POWER_TBL,
REG_FREQ_LATENCY,
REG_ARRAY_SIZE,
};
struct mtk_cpufreq_data {
struct cpufreq_frequency_table *table;
void __iomem *reg_bases[REG_ARRAY_SIZE];
struct resource *res;
void __iomem *base;
int nr_opp;
};
static const u16 cpufreq_mtk_offsets[REG_ARRAY_SIZE] = {
[REG_FREQ_LUT_TABLE] = 0x0,
[REG_FREQ_ENABLE] = 0x84,
[REG_FREQ_PERF_STATE] = 0x88,
[REG_FREQ_HW_STATE] = 0x8c,
[REG_EM_POWER_TBL] = 0x90,
[REG_FREQ_LATENCY] = 0x110,
};
static int __maybe_unused
PM: EM: convert power field to micro-Watts precision and align drivers The milli-Watts precision causes rounding errors while calculating efficiency cost for each OPP. This is especially visible in the 'simple' Energy Model (EM), where the power for each OPP is provided from OPP framework. This can cause some OPPs to be marked inefficient, while using micro-Watts precision that might not happen. Update all EM users which access 'power' field and assume the value is in milli-Watts. Solve also an issue with potential overflow in calculation of energy estimation on 32bit machine. It's needed now since the power value (thus the 'cost' as well) are higher. Example calculation which shows the rounding error and impact: power = 'dyn-power-coeff' * volt_mV * volt_mV * freq_MHz power_a_uW = (100 * 600mW * 600mW * 500MHz) / 10^6 = 18000 power_a_mW = (100 * 600mW * 600mW * 500MHz) / 10^9 = 18 power_b_uW = (100 * 605mW * 605mW * 600MHz) / 10^6 = 21961 power_b_mW = (100 * 605mW * 605mW * 600MHz) / 10^9 = 21 max_freq = 2000MHz cost_a_mW = 18 * 2000MHz/500MHz = 72 cost_a_uW = 18000 * 2000MHz/500MHz = 72000 cost_b_mW = 21 * 2000MHz/600MHz = 70 // <- artificially better cost_b_uW = 21961 * 2000MHz/600MHz = 73203 The 'cost_b_mW' (which is based on old milli-Watts) is misleadingly better that the 'cost_b_uW' (this patch uses micro-Watts) and such would have impact on the 'inefficient OPPs' information in the Cpufreq framework. This patch set removes the rounding issue. Signed-off-by: Lukasz Luba <lukasz.luba@arm.com> Acked-by: Daniel Lezcano <daniel.lezcano@linaro.org> Acked-by: Viresh Kumar <viresh.kumar@linaro.org> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2022-07-07 15:15:52 +08:00
mtk_cpufreq_get_cpu_power(struct device *cpu_dev, unsigned long *uW,
unsigned long *KHz)
{
struct mtk_cpufreq_data *data;
struct cpufreq_policy *policy;
int i;
policy = cpufreq_cpu_get_raw(cpu_dev->id);
if (!policy)
return 0;
data = policy->driver_data;
for (i = 0; i < data->nr_opp; i++) {
if (data->table[i].frequency < *KHz)
break;
}
i--;
*KHz = data->table[i].frequency;
PM: EM: convert power field to micro-Watts precision and align drivers The milli-Watts precision causes rounding errors while calculating efficiency cost for each OPP. This is especially visible in the 'simple' Energy Model (EM), where the power for each OPP is provided from OPP framework. This can cause some OPPs to be marked inefficient, while using micro-Watts precision that might not happen. Update all EM users which access 'power' field and assume the value is in milli-Watts. Solve also an issue with potential overflow in calculation of energy estimation on 32bit machine. It's needed now since the power value (thus the 'cost' as well) are higher. Example calculation which shows the rounding error and impact: power = 'dyn-power-coeff' * volt_mV * volt_mV * freq_MHz power_a_uW = (100 * 600mW * 600mW * 500MHz) / 10^6 = 18000 power_a_mW = (100 * 600mW * 600mW * 500MHz) / 10^9 = 18 power_b_uW = (100 * 605mW * 605mW * 600MHz) / 10^6 = 21961 power_b_mW = (100 * 605mW * 605mW * 600MHz) / 10^9 = 21 max_freq = 2000MHz cost_a_mW = 18 * 2000MHz/500MHz = 72 cost_a_uW = 18000 * 2000MHz/500MHz = 72000 cost_b_mW = 21 * 2000MHz/600MHz = 70 // <- artificially better cost_b_uW = 21961 * 2000MHz/600MHz = 73203 The 'cost_b_mW' (which is based on old milli-Watts) is misleadingly better that the 'cost_b_uW' (this patch uses micro-Watts) and such would have impact on the 'inefficient OPPs' information in the Cpufreq framework. This patch set removes the rounding issue. Signed-off-by: Lukasz Luba <lukasz.luba@arm.com> Acked-by: Daniel Lezcano <daniel.lezcano@linaro.org> Acked-by: Viresh Kumar <viresh.kumar@linaro.org> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2022-07-07 15:15:52 +08:00
/* Provide micro-Watts value to the Energy Model */
*uW = readl_relaxed(data->reg_bases[REG_EM_POWER_TBL] +
i * LUT_ROW_SIZE);
return 0;
}
static int mtk_cpufreq_hw_target_index(struct cpufreq_policy *policy,
unsigned int index)
{
struct mtk_cpufreq_data *data = policy->driver_data;
writel_relaxed(index, data->reg_bases[REG_FREQ_PERF_STATE]);
return 0;
}
static unsigned int mtk_cpufreq_hw_get(unsigned int cpu)
{
struct mtk_cpufreq_data *data;
struct cpufreq_policy *policy;
unsigned int index;
policy = cpufreq_cpu_get_raw(cpu);
if (!policy)
return 0;
data = policy->driver_data;
index = readl_relaxed(data->reg_bases[REG_FREQ_PERF_STATE]);
index = min(index, LUT_MAX_ENTRIES - 1);
return data->table[index].frequency;
}
static unsigned int mtk_cpufreq_hw_fast_switch(struct cpufreq_policy *policy,
unsigned int target_freq)
{
struct mtk_cpufreq_data *data = policy->driver_data;
unsigned int index;
index = cpufreq_table_find_index_dl(policy, target_freq, false);
writel_relaxed(index, data->reg_bases[REG_FREQ_PERF_STATE]);
return policy->freq_table[index].frequency;
}
static int mtk_cpu_create_freq_table(struct platform_device *pdev,
struct mtk_cpufreq_data *data)
{
struct device *dev = &pdev->dev;
u32 temp, i, freq, prev_freq = 0;
void __iomem *base_table;
data->table = devm_kcalloc(dev, LUT_MAX_ENTRIES + 1,
sizeof(*data->table), GFP_KERNEL);
if (!data->table)
return -ENOMEM;
base_table = data->reg_bases[REG_FREQ_LUT_TABLE];
for (i = 0; i < LUT_MAX_ENTRIES; i++) {
temp = readl_relaxed(base_table + (i * LUT_ROW_SIZE));
freq = FIELD_GET(LUT_FREQ, temp) * 1000;
if (freq == prev_freq)
break;
data->table[i].frequency = freq;
dev_dbg(dev, "index=%d freq=%d\n", i, data->table[i].frequency);
prev_freq = freq;
}
data->table[i].frequency = CPUFREQ_TABLE_END;
data->nr_opp = i;
return 0;
}
static int mtk_cpu_resources_init(struct platform_device *pdev,
struct cpufreq_policy *policy,
const u16 *offsets)
{
struct mtk_cpufreq_data *data;
struct device *dev = &pdev->dev;
struct resource *res;
struct of_phandle_args args;
void __iomem *base;
int ret, i;
int index;
data = devm_kzalloc(dev, sizeof(*data), GFP_KERNEL);
if (!data)
return -ENOMEM;
ret = of_perf_domain_get_sharing_cpumask(policy->cpu, "performance-domains",
"#performance-domain-cells",
policy->cpus, &args);
if (ret < 0)
return ret;
index = args.args[0];
of_node_put(args.np);
res = platform_get_resource(pdev, IORESOURCE_MEM, index);
if (!res) {
dev_err(dev, "failed to get mem resource %d\n", index);
return -ENODEV;
}
if (!request_mem_region(res->start, resource_size(res), res->name)) {
dev_err(dev, "failed to request resource %pR\n", res);
return -EBUSY;
}
base = ioremap(res->start, resource_size(res));
if (!base) {
dev_err(dev, "failed to map resource %pR\n", res);
ret = -ENOMEM;
goto release_region;
}
data->base = base;
data->res = res;
for (i = REG_FREQ_LUT_TABLE; i < REG_ARRAY_SIZE; i++)
data->reg_bases[i] = base + offsets[i];
ret = mtk_cpu_create_freq_table(pdev, data);
if (ret) {
dev_info(dev, "Domain-%d failed to create freq table\n", index);
return ret;
}
policy->freq_table = data->table;
policy->driver_data = data;
return 0;
release_region:
release_mem_region(res->start, resource_size(res));
return ret;
}
static int mtk_cpufreq_hw_cpu_init(struct cpufreq_policy *policy)
{
struct platform_device *pdev = cpufreq_get_driver_data();
int sig, pwr_hw = CPUFREQ_HW_STATUS | SVS_HW_STATUS;
struct mtk_cpufreq_data *data;
unsigned int latency;
int ret;
/* Get the bases of cpufreq for domains */
ret = mtk_cpu_resources_init(pdev, policy, platform_get_drvdata(pdev));
if (ret) {
dev_info(&pdev->dev, "CPUFreq resource init failed\n");
return ret;
}
data = policy->driver_data;
latency = readl_relaxed(data->reg_bases[REG_FREQ_LATENCY]) * 1000;
if (!latency)
latency = CPUFREQ_ETERNAL;
policy->cpuinfo.transition_latency = latency;
policy->fast_switch_possible = true;
/* HW should be in enabled state to proceed now */
writel_relaxed(0x1, data->reg_bases[REG_FREQ_ENABLE]);
if (readl_poll_timeout(data->reg_bases[REG_FREQ_HW_STATE], sig,
(sig & pwr_hw) == pwr_hw, POLL_USEC,
TIMEOUT_USEC)) {
if (!(sig & CPUFREQ_HW_STATUS)) {
pr_info("cpufreq hardware of CPU%d is not enabled\n",
policy->cpu);
return -ENODEV;
}
pr_info("SVS of CPU%d is not enabled\n", policy->cpu);
}
return 0;
}
static int mtk_cpufreq_hw_cpu_exit(struct cpufreq_policy *policy)
{
struct mtk_cpufreq_data *data = policy->driver_data;
struct resource *res = data->res;
void __iomem *base = data->base;
/* HW should be in paused state now */
writel_relaxed(0x0, data->reg_bases[REG_FREQ_ENABLE]);
iounmap(base);
release_mem_region(res->start, resource_size(res));
return 0;
}
static void mtk_cpufreq_register_em(struct cpufreq_policy *policy)
{
struct em_data_callback em_cb = EM_DATA_CB(mtk_cpufreq_get_cpu_power);
struct mtk_cpufreq_data *data = policy->driver_data;
em_dev_register_perf_domain(get_cpu_device(policy->cpu), data->nr_opp,
&em_cb, policy->cpus, true);
}
static struct cpufreq_driver cpufreq_mtk_hw_driver = {
.flags = CPUFREQ_NEED_INITIAL_FREQ_CHECK |
CPUFREQ_HAVE_GOVERNOR_PER_POLICY |
CPUFREQ_IS_COOLING_DEV,
.verify = cpufreq_generic_frequency_table_verify,
.target_index = mtk_cpufreq_hw_target_index,
.get = mtk_cpufreq_hw_get,
.init = mtk_cpufreq_hw_cpu_init,
.exit = mtk_cpufreq_hw_cpu_exit,
.register_em = mtk_cpufreq_register_em,
.fast_switch = mtk_cpufreq_hw_fast_switch,
.name = "mtk-cpufreq-hw",
.attr = cpufreq_generic_attr,
};
static int mtk_cpufreq_hw_driver_probe(struct platform_device *pdev)
{
const void *data;
int ret;
data = of_device_get_match_data(&pdev->dev);
if (!data)
return -EINVAL;
platform_set_drvdata(pdev, (void *) data);
cpufreq_mtk_hw_driver.driver_data = pdev;
ret = cpufreq_register_driver(&cpufreq_mtk_hw_driver);
if (ret)
dev_err(&pdev->dev, "CPUFreq HW driver failed to register\n");
return ret;
}
static int mtk_cpufreq_hw_driver_remove(struct platform_device *pdev)
{
cpufreq_unregister_driver(&cpufreq_mtk_hw_driver);
return 0;
}
static const struct of_device_id mtk_cpufreq_hw_match[] = {
{ .compatible = "mediatek,cpufreq-hw", .data = &cpufreq_mtk_offsets },
{}
};
MODULE_DEVICE_TABLE(of, mtk_cpufreq_hw_match);
static struct platform_driver mtk_cpufreq_hw_driver = {
.probe = mtk_cpufreq_hw_driver_probe,
.remove = mtk_cpufreq_hw_driver_remove,
.driver = {
.name = "mtk-cpufreq-hw",
.of_match_table = mtk_cpufreq_hw_match,
},
};
module_platform_driver(mtk_cpufreq_hw_driver);
MODULE_AUTHOR("Hector Yuan <hector.yuan@mediatek.com>");
MODULE_DESCRIPTION("Mediatek cpufreq-hw driver");
MODULE_LICENSE("GPL v2");