If CONFIG_SMP, cpufreq skips loops_per_jiffy update, because different arch has different per-cpu loops_per_jiffy definition.

Signed-off-by: Richard Zhao richard.zhao@linaro.org --- arch/arm/kernel/smp.c | 54 +++++++++++++++++++++++++++++++++++++++++++++++++ 1 files changed, 54 insertions(+), 0 deletions(-)

diff --git a/arch/arm/kernel/smp.c b/arch/arm/kernel/smp.c index ef5640b..ac9cadc 100644 --- a/arch/arm/kernel/smp.c +++ b/arch/arm/kernel/smp.c @@ -25,6 +25,7 @@ #include <linux/percpu.h> #include <linux/clockchips.h> #include <linux/completion.h> +#include <linux/cpufreq.h>

#include <linux/atomic.h> #include <asm/cacheflush.h> @@ -631,3 +632,56 @@ int setup_profiling_timer(unsigned int multiplier) { return -EINVAL; } + +#ifdef CONFIG_CPU_FREQ + +static DEFINE_PER_CPU(unsigned long, l_p_j_ref); +static DEFINE_PER_CPU(unsigned long, l_p_j_ref_freq); +static unsigned long global_l_p_j_ref; +static unsigned long global_l_p_j_ref_freq; + +static int cpufreq_callback(struct notifier_block *nb, + unsigned long val, void *data) +{ + struct cpufreq_freqs *freq = data; + int cpu = freq->cpu; + + if (freq->flags & CPUFREQ_CONST_LOOPS) + return NOTIFY_OK; + + if (!per_cpu(l_p_j_ref, cpu)) { + per_cpu(l_p_j_ref, cpu) = + per_cpu(cpu_data, cpu).loops_per_jiffy; + per_cpu(l_p_j_ref_freq, cpu) = freq->old; + if (!global_l_p_j_ref) { + global_l_p_j_ref = loops_per_jiffy; + global_l_p_j_ref_freq = freq->old; + } + } + + if ((val == CPUFREQ_PRECHANGE && freq->old < freq->new) || + (val == CPUFREQ_POSTCHANGE && freq->old > freq->new) || + (val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) { + loops_per_jiffy = cpufreq_scale(global_l_p_j_ref, + global_l_p_j_ref_freq, + freq->new); + per_cpu(cpu_data, cpu).loops_per_jiffy = + cpufreq_scale(per_cpu(l_p_j_ref, cpu), + per_cpu(l_p_j_ref_freq, cpu), + freq->new); + } + return NOTIFY_OK; +} + +static struct notifier_block cpufreq_notifier = { + .notifier_call = cpufreq_callback, +}; + +static int __init register_cpufreq_notifier(void) +{ + return cpufreq_register_notifier(&cpufreq_notifier, + CPUFREQ_TRANSITION_NOTIFIER); +} +core_initcall(register_cpufreq_notifier); + +#endif

arm registered cpufreq transition notifier to recalculate it.

Signed-off-by: Richard Zhao richard.zhao@linaro.org --- drivers/cpufreq/omap-cpufreq.c | 36 ------------------------------------ 1 files changed, 0 insertions(+), 36 deletions(-)

diff --git a/drivers/cpufreq/omap-cpufreq.c b/drivers/cpufreq/omap-cpufreq.c index 5d04c57..17da4c4 100644 --- a/drivers/cpufreq/omap-cpufreq.c +++ b/drivers/cpufreq/omap-cpufreq.c @@ -37,16 +37,6 @@

#include <mach/hardware.h>

-#ifdef CONFIG_SMP -struct lpj_info { - unsigned long ref; - unsigned int freq; -}; - -static DEFINE_PER_CPU(struct lpj_info, lpj_ref); -static struct lpj_info global_lpj_ref; -#endif - static struct cpufreq_frequency_table *freq_table; static atomic_t freq_table_users = ATOMIC_INIT(0); static struct clk *mpu_clk; @@ -118,32 +108,6 @@ static int omap_target(struct cpufreq_policy *policy, ret = clk_set_rate(mpu_clk, freqs.new * 1000); freqs.new = omap_getspeed(policy->cpu);

-#ifdef CONFIG_SMP - /* - * Note that loops_per_jiffy is not updated on SMP systems in - * cpufreq driver. So, update the per-CPU loops_per_jiffy value - * on frequency transition. We need to update all dependent CPUs. - */ - for_each_cpu(i, policy->cpus) { - struct lpj_info *lpj = &per_cpu(lpj_ref, i); - if (!lpj->freq) { - lpj->ref = per_cpu(cpu_data, i).loops_per_jiffy; - lpj->freq = freqs.old; - } - - per_cpu(cpu_data, i).loops_per_jiffy = - cpufreq_scale(lpj->ref, lpj->freq, freqs.new); - } - - /* And don't forget to adjust the global one */ - if (!global_lpj_ref.freq) { - global_lpj_ref.ref = loops_per_jiffy; - global_lpj_ref.freq = freqs.old; - } - loops_per_jiffy = cpufreq_scale(global_lpj_ref.ref, global_lpj_ref.freq, - freqs.new); -#endif - /* notifiers */ for_each_cpu(i, policy->cpus) { freqs.cpu = i;

The driver get cpu operation point table from device tree cpu0 node, and adjusts operating points using clk and regulator APIs.

It support single core and multi-core ARM SoCs. But currently it assume all cores share the same frequency and voltage.

Signed-off-by: Richard Zhao richard.zhao@linaro.org Reviewed-by: Jamie Iles jamie@jamieiles.com Reviewed-by: Mark Brown broonie@opensource.wolfsonmicro.com Acked-by: Shawn Guo shawn.guo@linaro.org --- .../devicetree/bindings/cpufreq/clk-reg-cpufreq | 21 ++ drivers/cpufreq/Kconfig | 10 + drivers/cpufreq/Makefile | 2 + drivers/cpufreq/clk-reg-cpufreq.c | 308 ++++++++++++++++++++ 4 files changed, 341 insertions(+), 0 deletions(-) create mode 100644 Documentation/devicetree/bindings/cpufreq/clk-reg-cpufreq create mode 100644 drivers/cpufreq/clk-reg-cpufreq.c

diff --git a/Documentation/devicetree/bindings/cpufreq/clk-reg-cpufreq b/Documentation/devicetree/bindings/cpufreq/clk-reg-cpufreq new file mode 100644 index 0000000..e8dc763 --- /dev/null +++ b/Documentation/devicetree/bindings/cpufreq/clk-reg-cpufreq @@ -0,0 +1,21 @@ +Generic cpufreq driver based on clk and regulator APIs + +Required properties in /cpus/cpu@0: +- cpu-freqs : cpu frequency points it supports, in unit of Hz. + Each point takes on cell. +- cpu-volts : cpu voltage ranges required by the frequency points + at the same index, in unit of uV. + Each range takes two cells, one for min, the other for max. +- clk-trans-latency : cpu clk transition latency, in unit of ns. + It takes one cell. + +An example: +cpu@0 { + cpu-freqs = <996000000 792000000 396000000 198000000>; + cpu-volts = < /* min max */ + 1225000 1450000 /* 996M */ + 1100000 1450000 /* 792M */ + 950000 1450000 /* 396M */ + 850000 1450000>; /* 198M */ + clk-trans-latency = <61036>; /* two CLK32 periods */ +}; diff --git a/drivers/cpufreq/Kconfig b/drivers/cpufreq/Kconfig index e24a2a1..95470f1 100644 --- a/drivers/cpufreq/Kconfig +++ b/drivers/cpufreq/Kconfig @@ -179,6 +179,16 @@ config CPU_FREQ_GOV_CONSERVATIVE

If in doubt, say N.

+config CLK_REG_CPUFREQ_DRIVER + tristate "Generic cpufreq driver using clk and regulator APIs" + depends on HAVE_CLK && OF && REGULATOR + select CPU_FREQ_TABLE + help + This adds generic CPUFreq driver based on clk and regulator APIs. + It assumes all cores of the CPU share the same clock and voltage. + + If in doubt, say N. + menu "x86 CPU frequency scaling drivers" depends on X86 source "drivers/cpufreq/Kconfig.x86" diff --git a/drivers/cpufreq/Makefile b/drivers/cpufreq/Makefile index ce75fcb..2c4eb33 100644 --- a/drivers/cpufreq/Makefile +++ b/drivers/cpufreq/Makefile @@ -13,6 +13,8 @@ obj-$(CONFIG_CPU_FREQ_GOV_CONSERVATIVE) += cpufreq_conservative.o # CPUfreq cross-arch helpers obj-$(CONFIG_CPU_FREQ_TABLE) += freq_table.o

+obj-$(CONFIG_CLK_REG_CPUFREQ_DRIVER) += clk-reg-cpufreq.o + ################################################################################## # x86 drivers. # Link order matters. K8 is preferred to ACPI because of firmware bugs in early diff --git a/drivers/cpufreq/clk-reg-cpufreq.c b/drivers/cpufreq/clk-reg-cpufreq.c new file mode 100644 index 0000000..77c4408 --- /dev/null +++ b/drivers/cpufreq/clk-reg-cpufreq.c @@ -0,0 +1,308 @@ +/* + * Copyright (C) 2011 Freescale Semiconductor, Inc. + */ + +/* + * The code contained herein is licensed under the GNU General Public + * License. You may obtain a copy of the GNU General Public License + * Version 2 or later at the following locations: + * + * http://www.opensource.org/licenses/gpl-license.html + * http://www.gnu.org/copyleft/gpl.html + */ + +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt + +#include <linux/module.h> +#include <linux/cpufreq.h> +#include <linux/clk.h> +#include <linux/regulator/consumer.h> +#include <linux/err.h> +#include <linux/slab.h> +#include <linux/of.h> + +static u32 *cpu_freqs; /* Hz */ +static u32 *cpu_volts; /* uV */ +static u32 trans_latency; /* ns */ +static int cpu_op_nr; +static unsigned int cur_index; + +static struct clk *cpu_clk; +static struct regulator *cpu_reg; +static struct cpufreq_frequency_table *freq_table; + +static int set_cpu_freq(unsigned long freq, int index, int higher) +{ + int ret = 0; + + if (higher && cpu_reg) { + ret = regulator_set_voltage(cpu_reg, + cpu_volts[index * 2], cpu_volts[index * 2 + 1]); + if (ret) { + pr_err("set cpu voltage failed!\n"); + return ret; + } + } + + ret = clk_set_rate(cpu_clk, freq); + if (ret) { + if (cpu_reg) + regulator_set_voltage(cpu_reg, cpu_volts[cur_index * 2], + cpu_volts[cur_index * 2 + 1]); + pr_err("cannot set CPU clock rate\n"); + return ret; + } + + if (!higher && cpu_reg) { + ret = regulator_set_voltage(cpu_reg, + cpu_volts[index * 2], cpu_volts[index * 2 + 1]); + if (ret) + pr_warn("set cpu voltage failed, might run on" + " higher voltage!\n"); + ret = 0; + } + + return ret; +} + +static int clk_reg_verify_speed(struct cpufreq_policy *policy) +{ + return cpufreq_frequency_table_verify(policy, freq_table); +} + +static unsigned int clk_reg_get_speed(unsigned int cpu) +{ + return clk_get_rate(cpu_clk) / 1000; +} + +static int clk_reg_set_target(struct cpufreq_policy *policy, + unsigned int target_freq, unsigned int relation) +{ + struct cpufreq_freqs freqs; + unsigned long freq_Hz; + int cpu; + int ret = 0; + unsigned int index; + + cpufreq_frequency_table_target(policy, freq_table, + target_freq, relation, &index); + freq_Hz = clk_round_rate(cpu_clk, cpu_freqs[index]); + freq_Hz = freq_Hz ? freq_Hz : cpu_freqs[index]; + freqs.old = clk_get_rate(cpu_clk) / 1000; + freqs.new = freq_Hz / 1000; + freqs.flags = 0; + + if (freqs.old == freqs.new) + return 0; + + for_each_possible_cpu(cpu) { + freqs.cpu = cpu; + cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE); + } + + ret = set_cpu_freq(freq_Hz, index, (freqs.new > freqs.old)); + if (ret) + freqs.new = clk_get_rate(cpu_clk) / 1000; + else + cur_index = index; + + for_each_possible_cpu(cpu) { + freqs.cpu = cpu; + cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE); + } + + return ret; +} + +static int clk_reg_cpufreq_init(struct cpufreq_policy *policy) +{ + int ret; + + if (policy->cpu >= num_possible_cpus()) + return -EINVAL; + + policy->cur = clk_get_rate(cpu_clk) / 1000; + policy->shared_type = CPUFREQ_SHARED_TYPE_ANY; + cpumask_setall(policy->cpus); + policy->cpuinfo.transition_latency = trans_latency; + + ret = cpufreq_frequency_table_cpuinfo(policy, freq_table); + + if (ret < 0) { + pr_err("invalid frequency table for cpu %d\n", + policy->cpu); + return ret; + } + + cpufreq_frequency_table_get_attr(freq_table, policy->cpu); + cpufreq_frequency_table_target(policy, freq_table, policy->cur, + CPUFREQ_RELATION_H, &cur_index); + return 0; +} + +static int clk_reg_cpufreq_exit(struct cpufreq_policy *policy) +{ + cpufreq_frequency_table_put_attr(policy->cpu); + return 0; +} + +static struct freq_attr *clk_reg_cpufreq_attr[] = { + &cpufreq_freq_attr_scaling_available_freqs, + NULL, +}; + +static struct cpufreq_driver clk_reg_cpufreq_driver = { + .flags = CPUFREQ_STICKY, + .verify = clk_reg_verify_speed, + .target = clk_reg_set_target, + .get = clk_reg_get_speed, + .init = clk_reg_cpufreq_init, + .exit = clk_reg_cpufreq_exit, + .name = "clk-reg", + .attr = clk_reg_cpufreq_attr, +}; + +static u32 max_freq = UINT_MAX / 1000; /* kHz */ +module_param(max_freq, uint, 0); +MODULE_PARM_DESC(max_freq, "max cpu frequency in unit of kHz"); + +static int __devinit clk_reg_cpufreq_driver_init(void) +{ + struct device_node *cpu0; + const struct property *pp; + int max_idx = 0, min_idx = 0; + int i, ret; + + cpu0 = of_find_node_by_path("/cpus/cpu@0"); + if (!cpu0) + return -ENODEV; + + pp = of_find_property(cpu0, "cpu-freqs", NULL); + if (!pp) { + ret = -ENODEV; + goto put_node; + } + cpu_op_nr = pp->length / sizeof(u32); + if (!cpu_op_nr) { + ret = -ENODEV; + goto put_node; + } + ret = -ENOMEM; + cpu_freqs = kzalloc(sizeof(*cpu_freqs) * cpu_op_nr, GFP_KERNEL); + if (!cpu_freqs) + goto put_node; + of_property_read_u32_array(cpu0, "cpu-freqs", cpu_freqs, cpu_op_nr); + + pp = of_find_property(cpu0, "cpu-volts", NULL); + if (pp) { + if (cpu_op_nr * 2 == pp->length / sizeof(u32)) { + cpu_volts = kzalloc(sizeof(*cpu_volts) * cpu_op_nr * 2, + GFP_KERNEL); + if (!cpu_volts) + goto free_cpu_freqs; + of_property_read_u32_array(cpu0, "cpu-volts", + cpu_volts, cpu_op_nr * 2); + } else + pr_warn("invalid cpu_volts!\n"); + } + + if (of_property_read_u32(cpu0, "clk-trans-latency", &trans_latency)) + trans_latency = CPUFREQ_ETERNAL; + + cpu_clk = clk_get(NULL, "cpu"); + if (IS_ERR(cpu_clk)) { + pr_err("failed to get cpu clock\n"); + ret = PTR_ERR(cpu_clk); + goto free_cpu_volts; + } + + if (cpu_volts) { + cpu_reg = regulator_get(NULL, "cpu"); + if (IS_ERR(cpu_reg)) { + pr_warn("regulator cpu get failed.\n"); + cpu_reg = NULL; + } + } + + freq_table = kmalloc(sizeof(struct cpufreq_frequency_table) + * (cpu_op_nr + 1), GFP_KERNEL); + if (!freq_table) { + ret = -ENOMEM; + goto reg_put; + } + + for (i = 0; i < cpu_op_nr; i++) { + freq_table[i].index = i; + if (cpu_freqs[i] > max_freq * 1000) { + freq_table[i].frequency = CPUFREQ_ENTRY_INVALID; + continue; + } + + if (cpu_reg) { + ret = regulator_is_supported_voltage(cpu_reg, + cpu_volts[i * 2], cpu_volts[i * 2 + 1]); + if (ret <= 0) { + freq_table[i].frequency = CPUFREQ_ENTRY_INVALID; + continue; + } + } + freq_table[i].frequency = cpu_freqs[i] / 1000; + max_idx = cpu_freqs[i] > cpu_freqs[max_idx] ? i : max_idx; + min_idx = cpu_freqs[i] < cpu_freqs[min_idx] ? i : min_idx; + } + + freq_table[i].index = i; + freq_table[i].frequency = CPUFREQ_TABLE_END; + + if (cpu_reg && trans_latency != CPUFREQ_ETERNAL) { + ret = regulator_set_voltage_time(cpu_reg, cpu_volts[min_idx], + cpu_volts[max_idx]); + if (ret < 0) { + pr_warn("regulator_set_voltage_time failed. " + "Default use 1us\n"); + ret = 1; + } + trans_latency += ret * 1000; + } + + ret = cpufreq_register_driver(&clk_reg_cpufreq_driver); + if (ret) + goto free_freq_table; + + of_node_put(cpu0); + + return 0; + +free_freq_table: + kfree(freq_table); +reg_put: + if (cpu_reg) + regulator_put(cpu_reg); + clk_put(cpu_clk); +free_cpu_volts: + kfree(cpu_volts); +free_cpu_freqs: + kfree(cpu_freqs); +put_node: + of_node_put(cpu0); + + return ret; +} + +static void clk_reg_cpufreq_driver_exit(void) +{ + cpufreq_unregister_driver(&clk_reg_cpufreq_driver); + kfree(cpu_freqs); + kfree(cpu_volts); + clk_put(cpu_clk); + if (cpu_reg) + regulator_put(cpu_reg); + kfree(freq_table); +} + +module_init(clk_reg_cpufreq_driver_init); +module_exit(clk_reg_cpufreq_driver_exit); + +MODULE_AUTHOR("Freescale Semiconductor Inc. Richard Zhao richard.zhao@freescale.com"); +MODULE_DESCRIPTION("Generic CPUFreq driver based on clk and regulator APIs"); +MODULE_LICENSE("GPL");

cpufreq needs cpu clock to change frequency.

Signed-off-by: Richard Zhao richard.zhao@linaro.org --- arch/arm/mach-imx/clock-imx6q.c | 1 + 1 files changed, 1 insertions(+), 0 deletions(-)

diff --git a/arch/arm/mach-imx/clock-imx6q.c b/arch/arm/mach-imx/clock-imx6q.c index 039a7ab..72acbc2 100644 --- a/arch/arm/mach-imx/clock-imx6q.c +++ b/arch/arm/mach-imx/clock-imx6q.c @@ -1911,6 +1911,7 @@ static struct clk_lookup lookups[] = { _REGISTER_CLOCK(NULL, "gpmi_io_clk", gpmi_io_clk), _REGISTER_CLOCK(NULL, "usboh3_clk", usboh3_clk), _REGISTER_CLOCK(NULL, "sata_clk", sata_clk), + _REGISTER_CLOCK(NULL, "cpu", arm_clk), };

int imx6q_set_lpm(enum mxc_cpu_pwr_mode mode)