Re: [PATCH v13 2/2] cpufreq: mediatek-hw: Add support for CPUFREQ HW

From: Viresh Kumar
Date: Tue Aug 03 2021 - 03:13:08 EST


On 30-07-21, 00:08, Hector Yuan wrote:
> From: "Hector.Yuan" <hector.yuan@xxxxxxxxxxxx>
>
> Add cpufreq HW support

Please write a proper commit log, what you are adding and which SoCs
it will apply to. Also add a full stop (.) at the end.

> Signed-off-by: Hector.Yuan <hector.yuan@xxxxxxxxxxxx>
> ---
> drivers/cpufreq/Kconfig.arm | 12 ++
> drivers/cpufreq/Makefile | 1 +
> drivers/cpufreq/mediatek-cpufreq-hw.c | 357 +++++++++++++++++++++++++++++++++
> include/linux/cpufreq.h | 39 ++++

The changes to cpufreq.h should be done in a separate patch.

> diff --git a/drivers/cpufreq/mediatek-cpufreq-hw.c b/drivers/cpufreq/mediatek-cpufreq-hw.c
> new file mode 100644
> index 0000000..ca50a3a
> --- /dev/null
> +++ b/drivers/cpufreq/mediatek-cpufreq-hw.c
> @@ -0,0 +1,357 @@
> +// 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/pm_qos.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 cpufreq_mtk {
> + struct cpufreq_frequency_table *table;
> + void __iomem *reg_bases[REG_ARRAY_SIZE];
> + int nr_opp;
> + cpumask_t related_cpus;
> +};
> +
> +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 struct cpufreq_mtk *mtk_freq_domain_map[NR_CPUS];
> +
> +static int __maybe_unused
> +mtk_cpufreq_get_cpu_power(unsigned long *mW,
> + unsigned long *KHz, struct device *cpu_dev)
> +{
> + struct cpufreq_mtk *c;
> + struct cpufreq_policy *policy;
> + int i;
> +
> + policy = cpufreq_cpu_get_raw(cpu_dev->id);
> + if (!policy)
> + return 0;
> +
> + c = mtk_freq_domain_map[policy->cpu];
> +
> + for (i = 0; i < c->nr_opp; i++) {
> + if (c->table[i].frequency < *KHz)
> + break;
> + }
> + i--;
> +
> + *KHz = c->table[i].frequency;
> + *mW = readl_relaxed(c->reg_bases[REG_EM_POWER_TBL] +
> + i * LUT_ROW_SIZE) / 1000;
> +
> + return 0;
> +}
> +
> +static int mtk_cpufreq_hw_target_index(struct cpufreq_policy *policy,
> + unsigned int index)
> +{
> + struct cpufreq_mtk *c = policy->driver_data;
> +
> + writel_relaxed(index, c->reg_bases[REG_FREQ_PERF_STATE]);
> +
> + return 0;
> +}
> +
> +static unsigned int mtk_cpufreq_hw_get(unsigned int cpu)
> +{
> + struct cpufreq_mtk *c;
> + struct cpufreq_policy *policy;
> + unsigned int index;
> +
> + policy = cpufreq_cpu_get_raw(cpu);
> + if (!policy)
> + return 0;
> +
> + c = mtk_freq_domain_map[policy->cpu];
> +
> + index = readl_relaxed(c->reg_bases[REG_FREQ_PERF_STATE]);
> + index = min(index, LUT_MAX_ENTRIES - 1);
> +
> + return c->table[index].frequency;
> +}
> +
> +static unsigned int mtk_cpufreq_hw_fast_switch(struct cpufreq_policy *policy,
> + unsigned int target_freq)
> +{
> + struct cpufreq_mtk *c = policy->driver_data;
> + unsigned int index;
> +
> + index = cpufreq_table_find_index_dl(policy, target_freq);
> +
> + writel_relaxed(index, c->reg_bases[REG_FREQ_PERF_STATE]);
> +
> + return policy->freq_table[index].frequency;
> +}
> +
> +static int mtk_cpu_create_freq_table(struct platform_device *pdev,
> + struct cpufreq_mtk *c)
> +{
> + struct device *dev = &pdev->dev;
> + void __iomem *base_table;
> + u32 data, i, freq, prev_freq = 0;
> +
> + c->table = devm_kcalloc(dev, LUT_MAX_ENTRIES + 1,
> + sizeof(*c->table), GFP_KERNEL);
> + if (!c->table)
> + return -ENOMEM;
> +
> + base_table = c->reg_bases[REG_FREQ_LUT_TABLE];
> +
> + for (i = 0; i < LUT_MAX_ENTRIES; i++) {
> + data = readl_relaxed(base_table + (i * LUT_ROW_SIZE));
> + freq = FIELD_GET(LUT_FREQ, data) * 1000;
> +
> + if (freq == prev_freq)
> + break;
> +
> + c->table[i].frequency = freq;
> +
> + dev_dbg(dev, "index=%d freq=%d\n",
> + i, c->table[i].frequency);

Won't this fit in a single line ?

> +
> + prev_freq = freq;
> + }
> +
> + c->table[i].frequency = CPUFREQ_TABLE_END;
> + c->nr_opp = i;
> +
> + return 0;
> +}
> +
> +static int mtk_cpu_resources_init(struct platform_device *pdev,
> + unsigned int cpu, int index,
> + const u16 *offsets)
> +{
> + struct cpufreq_mtk *c;
> + struct device *dev = &pdev->dev;
> + int ret, i;
> + void __iomem *base;
> +
> + if (mtk_freq_domain_map[cpu])

This should not happen anymore, isn't it ?

> + return 0;
> +
> + c = devm_kzalloc(dev, sizeof(*c), GFP_KERNEL);
> + if (!c)
> + return -ENOMEM;
> +
> + base = devm_platform_ioremap_resource(pdev, index);
> + if (IS_ERR(base))
> + return PTR_ERR(base);
> +
> + for (i = REG_FREQ_LUT_TABLE; i < REG_ARRAY_SIZE; i++)
> + c->reg_bases[i] = base + offsets[i];
> +
> + ret = of_perf_domain_get_sharing_cpumask(index, "performance-domains",

Instead of parsing parsing "performance-domains" twice, I would rather
pass a CPU number here instead of index.

> + "#performance-domain-cells",
> + &c->related_cpus);

You could have directly passed policy->cpus here instead.

> + if (ret) {
> + dev_info(dev, "Domain-%d failed to get related CPUs\n", index);
> + return ret;
> + }
> +
> + ret = mtk_cpu_create_freq_table(pdev, c);
> + if (ret) {
> + dev_info(dev, "Domain-%d failed to create freq table\n", index);
> + return ret;
> + }
> +
> + mtk_freq_domain_map[cpu] = c;

I will rather use policy->driver_data to store this now.

> +
> + return 0;
> +}
> +
> +static int mtk_cpufreq_hw_cpu_init(struct cpufreq_policy *policy)
> +{
> + struct platform_device *pdev = cpufreq_get_driver_data();
> + struct cpufreq_mtk *c;
> + struct device *cpu_dev;
> + struct em_data_callback em_cb = EM_DATA_CB(mtk_cpufreq_get_cpu_power);
> + struct pm_qos_request *qos_request;
> + struct device_node *cpu_np;
> + struct of_phandle_args args;
> + const u16 *offsets;
> + unsigned int latency;
> + int sig, pwr_hw = CPUFREQ_HW_STATUS | SVS_HW_STATUS;
> + int ret;

It looks much more organized when the variable definitions are in
decreasing order of their length, instead of the random order as it is
right now.

> +
> + offsets = of_device_get_match_data(&pdev->dev);
> + if (!offsets)
> + return -EINVAL;
> +
> + cpu_np = of_cpu_device_node_get(policy->cpu);
> + if (!cpu_np) {
> + dev_info(&pdev->dev, "Failed to get cpu %d device\n",
> + policy->cpu);
> + return -ENODEV;
> + }
> +
> + ret = of_parse_phandle_with_args(cpu_np, "performance-domains",
> + "#performance-domain-cells", 0,
> + &args);
> + if (ret < 0)

What about dropping cpu_np and same later in the code as well ?

> + return ret;
> +
> + /* Get the bases of cpufreq for domains */
> + ret = mtk_cpu_resources_init(pdev, policy->cpu, args.args[0], offsets);
> + if (ret) {
> + dev_info(&pdev->dev, "CPUFreq resource init failed\n");
> + return ret;
> + }
> +
> + cpu_dev = get_cpu_device(policy->cpu);
> + if (!cpu_dev) {
> + pr_err("failed to get cpu%d device\n", policy->cpu);
> + return -ENODEV;
> + }
> +
> + c = mtk_freq_domain_map[policy->cpu];
> + if (!c) {
> + pr_err("No scaling support for CPU%d\n", policy->cpu);
> + return -ENODEV;
> + }
> +
> + cpumask_copy(policy->cpus, &c->related_cpus);
> +
> + policy->freq_table = c->table;
> + policy->driver_data = c;

Oh you already do this, you can remove mtk_freq_domain_map array now.

> +
> + latency = readl_relaxed(c->reg_bases[REG_FREQ_LATENCY]);
> + if (!latency)
> + latency = CPUFREQ_ETERNAL;
> +
> + /* us convert to ns */
> + policy->cpuinfo.transition_latency = latency * 1000;

You want to multiple CPUFREQ_ETERNAL too ?

> +
> + policy->fast_switch_possible = true;
> +
> + qos_request = kzalloc(sizeof(*qos_request), GFP_KERNEL);

This is a small structure, why not allocate it on stack instead ?

> + if (!qos_request)
> + return -ENOMEM;
> +
> + /* Let CPUs leave idle-off state for SVS CPU initializing */
> + cpu_latency_qos_add_request(qos_request, 0);
> +
> + /* HW should be in enabled state to proceed now */
> + writel_relaxed(0x1, c->reg_bases[REG_FREQ_ENABLE]);
> +
> + if (readl_poll_timeout(c->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);
> + }
> +
> + cpu_latency_qos_remove_request(qos_request);
> +
> + em_dev_register_perf_domain(cpu_dev, c->nr_opp, &em_cb, policy->cpus, true);
> +
> + kfree(qos_request);

Why free after registering for em ? And also move the entire qos thing
into a separate routine instead of adding it to ->init().

> +
> + return 0;
> +}
> +
> +static int mtk_cpufreq_hw_cpu_exit(struct cpufreq_policy *policy)
> +{
> + struct cpufreq_mtk *c;
> +
> + c = mtk_freq_domain_map[policy->cpu];
> +
> + /* HW should be in paused state now */
> + writel_relaxed(0x0, c->reg_bases[REG_FREQ_ENABLE]);

Please make sure each and every resource is freed here and in probe on
failures.

> +
> + return 0;
> +}
> +
> +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,
> + .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)
> +{
> + int ret;
> +
> + 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;
> + }
> +
> + return 0;

You can do return ret here and drop the earlier return and its {}.

> +}
> +
> +static int mtk_cpufreq_hw_driver_remove(struct platform_device *pdev)
> +{
> + return cpufreq_unregister_driver(&cpufreq_mtk_hw_driver);
> +}
> +
> +static const struct of_device_id mtk_cpufreq_hw_match[] = {
> + { .compatible = "mediatek,cpufreq-hw", .data = &cpufreq_mtk_offsets },
> + {}
> +};
> +
> +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_DESCRIPTION("Mediatek cpufreq-hw driver");
> +MODULE_LICENSE("GPL v2");

You can add Module-author as well here if you want.

> diff --git a/include/linux/cpufreq.h b/include/linux/cpufreq.h
> index 9fd7194..4916d70 100644
> --- a/include/linux/cpufreq.h
> +++ b/include/linux/cpufreq.h
> @@ -13,6 +13,8 @@
> #include <linux/completion.h>
> #include <linux/kobject.h>
> #include <linux/notifier.h>
> +#include <linux/of.h>
> +#include <linux/of_device.h>
> #include <linux/pm_qos.h>
> #include <linux/spinlock.h>
> #include <linux/sysfs.h>
> @@ -1036,6 +1038,43 @@ void arch_set_freq_scale(const struct cpumask *cpus,
> }
> #endif
>
> +#ifdef CONFIG_CPU_FREQ

There is another CONFIG_CPU_FREQ a few lines above, please use the
same block for this routine as well.

> +static inline int of_perf_domain_get_sharing_cpumask(int index, const char *list_name,
> + const char *cell_name,
> + struct cpumask *cpumask)
> +{
> + struct device_node *cpu_np;
> + struct of_phandle_args args;
> + int cpu, ret;
> +
> + for_each_possible_cpu(cpu) {
> + cpu_np = of_cpu_device_node_get(cpu);
> + if (!cpu_np)
> + continue;
> +
> + ret = of_parse_phandle_with_args(cpu_np, list_name,
> + cell_name, 0,
> + &args);
> +
> + of_node_put(cpu_np);
> + if (ret < 0)
> + continue;
> +
> + if (index == args.args[0])
> + cpumask_set_cpu(cpu, cpumask);
> + }
> +
> + return 0;
> +}
> +#else
> +static inline int of_perf_domain_get_sharing_cpumask(int index, const char *list_name,
> + const char *cell_name,
> + struct cpumask *cpumask)
> +{
> + return 0;

return -EOPNOTSUPP;

> +}
> +#endif

--
viresh