Re: [PATCH 3/4] cpufreq: New 'interactive' governor

From: Vincent Guittot
Date: Wed Feb 08 2012 - 18:01:00 EST


Hi Anton,

Have you got some figures which shows the improvement of the
responsivness compared to other governor like the ondemand one ?
That could be interesting to test interactive governor with
cpufreq-bench and compare the results with ondemand ?

Regards,
Vincent

On 7 February 2012 17:44, Anton Vorontsov <anton.vorontsov@xxxxxxxxxx> wrote:
> From: Mike Chan <mike@xxxxxxxxxxx>
>
> This governor is designed for latency-sensitive workloads, such as
> interactive user interfaces.  The interactive governor aims to be
> significantly more responsive to ramp CPU quickly up when CPU-intensive
> activity begins.
>
> Existing governors sample CPU load at a particular rate, typically
> every X ms.  This can lead to under-powering UI threads for the period of
> time during which the user begins interacting with a previously-idle system
> until the next sample period happens.
>
> The 'interactive' governor uses a different approach. Instead of sampling
> the CPU at a specified rate, the governor will check whether to scale the
> CPU frequency up soon after coming out of idle.  When the CPU comes out of
> idle, a timer is configured to fire within 1-2 ticks.  If the CPU is very
> busy from exiting idle to when the timer fires then we assume the CPU is
> underpowered and ramp to MAX speed.
>
> If the CPU was not sufficiently busy to immediately ramp to MAX speed, then
> the governor evaluates the CPU load since the last speed adjustment,
> choosing the highest value between that longer-term load or the short-term
> load since idle exit to determine the CPU speed to ramp to.
>
> A realtime thread is used for scaling up, giving the remaining tasks the
> CPU performance benefit, unlike existing governors which are more likely to
> schedule rampup work to occur after your performance starved tasks have
> completed.
>
> The tuneables for this governor are:
> /sys/devices/system/cpu/cpufreq/interactive/min_sample_time:
>        The minimum amount of time to spend at the current frequency before
>        ramping down. This is to ensure that the governor has seen enough
>        historic CPU load data to determine the appropriate workload.
>        Default is 20000 uS.
> /sys/devices/system/cpu/cpufreq/interactive/go_hispeed_load
>        The CPU load at which to ramp to max speed.  Default is 95.
>
> Signed-off-by: Mike Chan <mike@xxxxxxxxxxx>
> Signed-off-by: Todd Poynor <toddpoynor@xxxxxxxxxx>
> Signed-off-by: Allen Martin <amartin@xxxxxxxxxx> (submitted improvements)
> Signed-off-by: Axel Haslam <axelhaslam@xxxxxx> (submitted improvements)
> Signed-off-by: Anton Vorontsov <anton.vorontsov@xxxxxxxxxx>
> ---
>  drivers/cpufreq/Kconfig               |   26 ++
>  drivers/cpufreq/Makefile              |    1 +
>  drivers/cpufreq/cpufreq_interactive.c |  700 +++++++++++++++++++++++++++++++++
>  include/linux/cpufreq.h               |    3 +
>  4 files changed, 730 insertions(+), 0 deletions(-)
>  create mode 100644 drivers/cpufreq/cpufreq_interactive.c
>
> diff --git a/drivers/cpufreq/Kconfig b/drivers/cpufreq/Kconfig
> index e24a2a1..c47cc46 100644
> --- a/drivers/cpufreq/Kconfig
> +++ b/drivers/cpufreq/Kconfig
> @@ -99,6 +99,15 @@ config CPU_FREQ_DEFAULT_GOV_CONSERVATIVE
>          Be aware that not all cpufreq drivers support the conservative
>          governor. If unsure have a look at the help section of the
>          driver. Fallback governor will be the performance governor.
> +
> +config CPU_FREQ_DEFAULT_GOV_INTERACTIVE
> +       bool "interactive"
> +       select CPU_FREQ_GOV_INTERACTIVE
> +       help
> +         Use the CPUFreq governor 'interactive' as default. This allows
> +         you to get a full dynamic cpu frequency capable system by simply
> +         loading your cpufreq low-level hardware driver, using the
> +         'interactive' governor for latency-sensitive workloads.
>  endchoice
>
>  config CPU_FREQ_GOV_PERFORMANCE
> @@ -179,6 +188,23 @@ config CPU_FREQ_GOV_CONSERVATIVE
>
>          If in doubt, say N.
>
> +config CPU_FREQ_GOV_INTERACTIVE
> +       tristate "'interactive' cpufreq policy governor"
> +       help
> +         'interactive' - This driver adds a dynamic cpufreq policy governor
> +         designed for latency-sensitive workloads.
> +
> +         This governor attempts to reduce the latency of clock
> +         increases so that the system is more responsive to
> +         interactive workloads.
> +
> +         To compile this driver as a module, choose M here: the
> +         module will be called cpufreq_interactive.
> +
> +         For details, take a look at linux/Documentation/cpu-freq.
> +
> +         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 ac000fa..f84c99b 100644
> --- a/drivers/cpufreq/Makefile
> +++ b/drivers/cpufreq/Makefile
> @@ -9,6 +9,7 @@ obj-$(CONFIG_CPU_FREQ_GOV_POWERSAVE)    += cpufreq_powersave.o
>  obj-$(CONFIG_CPU_FREQ_GOV_USERSPACE)   += cpufreq_userspace.o
>  obj-$(CONFIG_CPU_FREQ_GOV_ONDEMAND)    += cpufreq_ondemand.o
>  obj-$(CONFIG_CPU_FREQ_GOV_CONSERVATIVE)        += cpufreq_conservative.o
> +obj-$(CONFIG_CPU_FREQ_GOV_INTERACTIVE) += cpufreq_interactive.o
>
>  # CPUfreq cross-arch helpers
>  obj-$(CONFIG_CPU_FREQ_TABLE)           += freq_table.o
> diff --git a/drivers/cpufreq/cpufreq_interactive.c b/drivers/cpufreq/cpufreq_interactive.c
> new file mode 100644
> index 0000000..188096a
> --- /dev/null
> +++ b/drivers/cpufreq/cpufreq_interactive.c
> @@ -0,0 +1,700 @@
> +/*
> + * Copyright (C) 2010 Google, Inc.
> + *
> + * This software is licensed under the terms of the GNU General Public
> + * License version 2, as published by the Free Software Foundation, and
> + * may be copied, distributed, and modified under those terms.
> + *
> + * This program is distributed in the hope that it will be useful,
> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> + * GNU General Public License for more details.
> + *
> + * Author: Mike Chan (mike@xxxxxxxxxxx)
> + */
> +
> +#include <linux/module.h>
> +#include <linux/cpu.h>
> +#include <linux/cpumask.h>
> +#include <linux/cpufreq.h>
> +#include <linux/mutex.h>
> +#include <linux/sched.h>
> +#include <linux/tick.h>
> +#include <linux/time.h>
> +#include <linux/timer.h>
> +#include <linux/workqueue.h>
> +#include <linux/kthread.h>
> +#include <linux/mutex.h>
> +
> +static atomic_t active_count = ATOMIC_INIT(0);
> +
> +struct cpufreq_interactive_cpuinfo {
> +       struct timer_list cpu_timer;
> +       int timer_idlecancel;
> +       u64 time_in_idle;
> +       u64 idle_exit_time;
> +       u64 timer_run_time;
> +       int idling;
> +       u64 freq_change_time;
> +       u64 freq_change_time_in_idle;
> +       struct cpufreq_policy *policy;
> +       struct cpufreq_frequency_table *freq_table;
> +       unsigned int target_freq;
> +       int governor_enabled;
> +};
> +
> +static DEFINE_PER_CPU(struct cpufreq_interactive_cpuinfo, cpuinfo);
> +
> +/* Workqueues handle frequency scaling */
> +static struct task_struct *up_task;
> +static struct workqueue_struct *down_wq;
> +static struct work_struct freq_scale_down_work;
> +static cpumask_t up_cpumask;
> +static spinlock_t up_cpumask_lock;
> +static cpumask_t down_cpumask;
> +static spinlock_t down_cpumask_lock;
> +static struct mutex set_speed_lock;
> +
> +/* Hi speed to bump to from lo speed when load burst (default max) */
> +static u64 hispeed_freq;
> +
> +/* Go to hi speed when CPU load at or above this value. */
> +#define DEFAULT_GO_HISPEED_LOAD 95
> +static unsigned long go_hispeed_load;
> +
> +/*
> + * The minimum amount of time to spend at a frequency before we can ramp down.
> + */
> +#define DEFAULT_MIN_SAMPLE_TIME (20 * USEC_PER_MSEC)
> +static unsigned long min_sample_time;
> +
> +/*
> + * The sample rate of the timer used to increase frequency
> + */
> +#define DEFAULT_TIMER_RATE (20 * USEC_PER_MSEC)
> +static unsigned long timer_rate;
> +
> +static int cpufreq_governor_interactive(struct cpufreq_policy *policy,
> +               unsigned int event);
> +
> +#ifndef CONFIG_CPU_FREQ_DEFAULT_GOV_INTERACTIVE
> +static
> +#endif
> +struct cpufreq_governor cpufreq_gov_interactive = {
> +       .name = "interactive",
> +       .governor = cpufreq_governor_interactive,
> +       .max_transition_latency = 10000000,
> +       .owner = THIS_MODULE,
> +};
> +
> +static void cpufreq_interactive_timer(unsigned long data)
> +{
> +       unsigned int delta_idle;
> +       unsigned int delta_time;
> +       int cpu_load;
> +       int load_since_change;
> +       u64 time_in_idle;
> +       u64 idle_exit_time;
> +       struct cpufreq_interactive_cpuinfo *pcpu =
> +               &per_cpu(cpuinfo, data);
> +       u64 now_idle;
> +       unsigned int new_freq;
> +       unsigned int index;
> +       unsigned long flags;
> +
> +       smp_rmb();
> +
> +       if (!pcpu->governor_enabled)
> +               goto exit;
> +
> +       /*
> +        * Once pcpu->timer_run_time is updated to >= pcpu->idle_exit_time,
> +        * this lets idle exit know the current idle time sample has
> +        * been processed, and idle exit can generate a new sample and
> +        * re-arm the timer.  This prevents a concurrent idle
> +        * exit on that CPU from writing a new set of info at the same time
> +        * the timer function runs (the timer function can't use that info
> +        * until more time passes).
> +        */
> +       time_in_idle = pcpu->time_in_idle;
> +       idle_exit_time = pcpu->idle_exit_time;
> +       now_idle = get_cpu_idle_time_us(data, &pcpu->timer_run_time);
> +       smp_wmb();
> +
> +       /* If we raced with cancelling a timer, skip. */
> +       if (!idle_exit_time)
> +               goto exit;
> +
> +       delta_idle = (unsigned int) (now_idle - time_in_idle);
> +       delta_time = (unsigned int) (pcpu->timer_run_time - idle_exit_time);
> +
> +       /*
> +        * If timer ran less than 1ms after short-term sample started, retry.
> +        */
> +       if (delta_time < 1000)
> +               goto rearm;
> +
> +       if (delta_idle > delta_time)
> +               cpu_load = 0;
> +       else
> +               cpu_load = 100 * (delta_time - delta_idle) / delta_time;
> +
> +       delta_idle = (unsigned int) (now_idle - pcpu->freq_change_time_in_idle);
> +       delta_time = (unsigned int) (pcpu->timer_run_time - pcpu->freq_change_time);
> +
> +       if ((delta_time == 0) || (delta_idle > delta_time))
> +               load_since_change = 0;
> +       else
> +               load_since_change =
> +                       100 * (delta_time - delta_idle) / delta_time;
> +
> +       /*
> +        * Choose greater of short-term load (since last idle timer
> +        * started or timer function re-armed itself) or long-term load
> +        * (since last frequency change).
> +        */
> +       if (load_since_change > cpu_load)
> +               cpu_load = load_since_change;
> +
> +       if (cpu_load >= go_hispeed_load) {
> +               if (pcpu->policy->cur == pcpu->policy->min)
> +                       new_freq = hispeed_freq;
> +               else
> +                       new_freq = pcpu->policy->max * cpu_load / 100;
> +       } else {
> +               new_freq = pcpu->policy->cur * cpu_load / 100;
> +       }
> +
> +       if (cpufreq_frequency_table_target(pcpu->policy, pcpu->freq_table,
> +                                          new_freq, CPUFREQ_RELATION_H,
> +                                          &index)) {
> +               pr_warn_once("timer %d: cpufreq_frequency_table_target error\n",
> +                            (int) data);
> +               goto rearm;
> +       }
> +
> +       new_freq = pcpu->freq_table[index].frequency;
> +
> +       if (pcpu->target_freq == new_freq)
> +               goto rearm_if_notmax;
> +
> +       /*
> +        * Do not scale down unless we have been at this frequency for the
> +        * minimum sample time.
> +        */
> +       if (new_freq < pcpu->target_freq) {
> +               if (pcpu->timer_run_time - pcpu->freq_change_time
> +                               < min_sample_time)
> +                       goto rearm;
> +       }
> +
> +       if (new_freq < pcpu->target_freq) {
> +               pcpu->target_freq = new_freq;
> +               spin_lock_irqsave(&down_cpumask_lock, flags);
> +               cpumask_set_cpu(data, &down_cpumask);
> +               spin_unlock_irqrestore(&down_cpumask_lock, flags);
> +               queue_work(down_wq, &freq_scale_down_work);
> +       } else {
> +               pcpu->target_freq = new_freq;
> +               spin_lock_irqsave(&up_cpumask_lock, flags);
> +               cpumask_set_cpu(data, &up_cpumask);
> +               spin_unlock_irqrestore(&up_cpumask_lock, flags);
> +               wake_up_process(up_task);
> +       }
> +
> +rearm_if_notmax:
> +       /*
> +        * Already set max speed and don't see a need to change that,
> +        * wait until next idle to re-evaluate, don't need timer.
> +        */
> +       if (pcpu->target_freq == pcpu->policy->max)
> +               goto exit;
> +
> +rearm:
> +       if (!timer_pending(&pcpu->cpu_timer)) {
> +               /*
> +                * If already at min: if that CPU is idle, don't set timer.
> +                * Else cancel the timer if that CPU goes idle.  We don't
> +                * need to re-evaluate speed until the next idle exit.
> +                */
> +               if (pcpu->target_freq == pcpu->policy->min) {
> +                       smp_rmb();
> +
> +                       if (pcpu->idling)
> +                               goto exit;
> +
> +                       pcpu->timer_idlecancel = 1;
> +               }
> +
> +               pcpu->time_in_idle = get_cpu_idle_time_us(
> +                       data, &pcpu->idle_exit_time);
> +               mod_timer(&pcpu->cpu_timer,
> +                         jiffies + usecs_to_jiffies(timer_rate));
> +       }
> +
> +exit:
> +       return;
> +}
> +
> +static void cpufreq_interactive_idle_start(void)
> +{
> +       struct cpufreq_interactive_cpuinfo *pcpu =
> +               &per_cpu(cpuinfo, smp_processor_id());
> +       int pending;
> +
> +       if (!pcpu->governor_enabled)
> +               return;
> +
> +       pcpu->idling = 1;
> +       smp_wmb();
> +       pending = timer_pending(&pcpu->cpu_timer);
> +
> +       if (pcpu->target_freq != pcpu->policy->min) {
> +#ifdef CONFIG_SMP
> +               /*
> +                * Entering idle while not at lowest speed.  On some
> +                * platforms this can hold the other CPU(s) at that speed
> +                * even though the CPU is idle. Set a timer to re-evaluate
> +                * speed so this idle CPU doesn't hold the other CPUs above
> +                * min indefinitely.  This should probably be a quirk of
> +                * the CPUFreq driver.
> +                */
> +               if (!pending) {
> +                       pcpu->time_in_idle = get_cpu_idle_time_us(
> +                               smp_processor_id(), &pcpu->idle_exit_time);
> +                       pcpu->timer_idlecancel = 0;
> +                       mod_timer(&pcpu->cpu_timer,
> +                                 jiffies + usecs_to_jiffies(timer_rate));
> +               }
> +#endif
> +       } else {
> +               /*
> +                * If at min speed and entering idle after load has
> +                * already been evaluated, and a timer has been set just in
> +                * case the CPU suddenly goes busy, cancel that timer.  The
> +                * CPU didn't go busy; we'll recheck things upon idle exit.
> +                */
> +               if (pending && pcpu->timer_idlecancel) {
> +                       del_timer(&pcpu->cpu_timer);
> +                       /*
> +                        * Ensure last timer run time is after current idle
> +                        * sample start time, so next idle exit will always
> +                        * start a new idle sampling period.
> +                        */
> +                       pcpu->idle_exit_time = 0;
> +                       pcpu->timer_idlecancel = 0;
> +               }
> +       }
> +
> +}
> +
> +static void cpufreq_interactive_idle_end(void)
> +{
> +       struct cpufreq_interactive_cpuinfo *pcpu =
> +               &per_cpu(cpuinfo, smp_processor_id());
> +
> +       pcpu->idling = 0;
> +       smp_wmb();
> +
> +       /*
> +        * Arm the timer for 1-2 ticks later if not already, and if the timer
> +        * function has already processed the previous load sampling
> +        * interval.  (If the timer is not pending but has not processed
> +        * the previous interval, it is probably racing with us on another
> +        * CPU.  Let it compute load based on the previous sample and then
> +        * re-arm the timer for another interval when it's done, rather
> +        * than updating the interval start time to be "now", which doesn't
> +        * give the timer function enough time to make a decision on this
> +        * run.)
> +        */
> +       if (timer_pending(&pcpu->cpu_timer) == 0 &&
> +           pcpu->timer_run_time >= pcpu->idle_exit_time &&
> +           pcpu->governor_enabled) {
> +               pcpu->time_in_idle =
> +                       get_cpu_idle_time_us(smp_processor_id(),
> +                                            &pcpu->idle_exit_time);
> +               pcpu->timer_idlecancel = 0;
> +               mod_timer(&pcpu->cpu_timer,
> +                         jiffies + usecs_to_jiffies(timer_rate));
> +       }
> +
> +}
> +
> +static int cpufreq_interactive_up_task(void *data)
> +{
> +       unsigned int cpu;
> +       cpumask_t tmp_mask;
> +       unsigned long flags;
> +       struct cpufreq_interactive_cpuinfo *pcpu;
> +
> +       while (1) {
> +               set_current_state(TASK_INTERRUPTIBLE);
> +               spin_lock_irqsave(&up_cpumask_lock, flags);
> +
> +               if (cpumask_empty(&up_cpumask)) {
> +                       spin_unlock_irqrestore(&up_cpumask_lock, flags);
> +                       schedule();
> +
> +                       if (kthread_should_stop())
> +                               break;
> +
> +                       spin_lock_irqsave(&up_cpumask_lock, flags);
> +               }
> +
> +               set_current_state(TASK_RUNNING);
> +               tmp_mask = up_cpumask;
> +               cpumask_clear(&up_cpumask);
> +               spin_unlock_irqrestore(&up_cpumask_lock, flags);
> +
> +               for_each_cpu(cpu, &tmp_mask) {
> +                       unsigned int j;
> +                       unsigned int max_freq = 0;
> +
> +                       pcpu = &per_cpu(cpuinfo, cpu);
> +                       smp_rmb();
> +
> +                       if (!pcpu->governor_enabled)
> +                               continue;
> +
> +                       mutex_lock(&set_speed_lock);
> +
> +                       for_each_cpu(j, pcpu->policy->cpus) {
> +                               struct cpufreq_interactive_cpuinfo *pjcpu =
> +                                       &per_cpu(cpuinfo, j);
> +
> +                               if (pjcpu->target_freq > max_freq)
> +                                       max_freq = pjcpu->target_freq;
> +                       }
> +
> +                       if (max_freq != pcpu->policy->cur)
> +                               __cpufreq_driver_target(pcpu->policy,
> +                                                       max_freq,
> +                                                       CPUFREQ_RELATION_H);
> +                       mutex_unlock(&set_speed_lock);
> +
> +                       pcpu->freq_change_time_in_idle =
> +                               get_cpu_idle_time_us(cpu,
> +                                                    &pcpu->freq_change_time);
> +               }
> +       }
> +
> +       return 0;
> +}
> +
> +static void cpufreq_interactive_freq_down(struct work_struct *work)
> +{
> +       unsigned int cpu;
> +       cpumask_t tmp_mask;
> +       unsigned long flags;
> +       struct cpufreq_interactive_cpuinfo *pcpu;
> +
> +       spin_lock_irqsave(&down_cpumask_lock, flags);
> +       tmp_mask = down_cpumask;
> +       cpumask_clear(&down_cpumask);
> +       spin_unlock_irqrestore(&down_cpumask_lock, flags);
> +
> +       for_each_cpu(cpu, &tmp_mask) {
> +               unsigned int j;
> +               unsigned int max_freq = 0;
> +
> +               pcpu = &per_cpu(cpuinfo, cpu);
> +               smp_rmb();
> +
> +               if (!pcpu->governor_enabled)
> +                       continue;
> +
> +               mutex_lock(&set_speed_lock);
> +
> +               for_each_cpu(j, pcpu->policy->cpus) {
> +                       struct cpufreq_interactive_cpuinfo *pjcpu =
> +                               &per_cpu(cpuinfo, j);
> +
> +                       if (pjcpu->target_freq > max_freq)
> +                               max_freq = pjcpu->target_freq;
> +               }
> +
> +               if (max_freq != pcpu->policy->cur)
> +                       __cpufreq_driver_target(pcpu->policy, max_freq,
> +                                               CPUFREQ_RELATION_H);
> +
> +               mutex_unlock(&set_speed_lock);
> +               pcpu->freq_change_time_in_idle =
> +                       get_cpu_idle_time_us(cpu,
> +                                            &pcpu->freq_change_time);
> +       }
> +}
> +
> +static ssize_t show_hispeed_freq(struct kobject *kobj,
> +                                struct attribute *attr, char *buf)
> +{
> +       return sprintf(buf, "%llu\n", hispeed_freq);
> +}
> +
> +static ssize_t store_hispeed_freq(struct kobject *kobj,
> +                                 struct attribute *attr, const char *buf,
> +                                 size_t count)
> +{
> +       int ret;
> +       u64 val;
> +
> +       ret = strict_strtoull(buf, 0, &val);
> +       if (ret < 0)
> +               return ret;
> +       hispeed_freq = val;
> +       return count;
> +}
> +
> +static struct global_attr hispeed_freq_attr = __ATTR(hispeed_freq, 0644,
> +               show_hispeed_freq, store_hispeed_freq);
> +
> +
> +static ssize_t show_go_hispeed_load(struct kobject *kobj,
> +                                    struct attribute *attr, char *buf)
> +{
> +       return sprintf(buf, "%lu\n", go_hispeed_load);
> +}
> +
> +static ssize_t store_go_hispeed_load(struct kobject *kobj,
> +                       struct attribute *attr, const char *buf, size_t count)
> +{
> +       int ret;
> +       unsigned long val;
> +
> +       ret = strict_strtoul(buf, 0, &val);
> +       if (ret < 0)
> +               return ret;
> +       go_hispeed_load = val;
> +       return count;
> +}
> +
> +static struct global_attr go_hispeed_load_attr = __ATTR(go_hispeed_load, 0644,
> +               show_go_hispeed_load, store_go_hispeed_load);
> +
> +static ssize_t show_min_sample_time(struct kobject *kobj,
> +                               struct attribute *attr, char *buf)
> +{
> +       return sprintf(buf, "%lu\n", min_sample_time);
> +}
> +
> +static ssize_t store_min_sample_time(struct kobject *kobj,
> +                       struct attribute *attr, const char *buf, size_t count)
> +{
> +       int ret;
> +       unsigned long val;
> +
> +       ret = strict_strtoul(buf, 0, &val);
> +       if (ret < 0)
> +               return ret;
> +       min_sample_time = val;
> +       return count;
> +}
> +
> +static struct global_attr min_sample_time_attr = __ATTR(min_sample_time, 0644,
> +               show_min_sample_time, store_min_sample_time);
> +
> +static ssize_t show_timer_rate(struct kobject *kobj,
> +                       struct attribute *attr, char *buf)
> +{
> +       return sprintf(buf, "%lu\n", timer_rate);
> +}
> +
> +static ssize_t store_timer_rate(struct kobject *kobj,
> +                       struct attribute *attr, const char *buf, size_t count)
> +{
> +       int ret;
> +       unsigned long val;
> +
> +       ret = strict_strtoul(buf, 0, &val);
> +       if (ret < 0)
> +               return ret;
> +       timer_rate = val;
> +       return count;
> +}
> +
> +static struct global_attr timer_rate_attr = __ATTR(timer_rate, 0644,
> +               show_timer_rate, store_timer_rate);
> +
> +static struct attribute *interactive_attributes[] = {
> +       &hispeed_freq_attr.attr,
> +       &go_hispeed_load_attr.attr,
> +       &min_sample_time_attr.attr,
> +       &timer_rate_attr.attr,
> +       NULL,
> +};
> +
> +static struct attribute_group interactive_attr_group = {
> +       .attrs = interactive_attributes,
> +       .name = "interactive",
> +};
> +
> +static int cpufreq_governor_interactive(struct cpufreq_policy *policy,
> +               unsigned int event)
> +{
> +       int rc;
> +       unsigned int j;
> +       struct cpufreq_interactive_cpuinfo *pcpu;
> +       struct cpufreq_frequency_table *freq_table;
> +
> +       switch (event) {
> +       case CPUFREQ_GOV_START:
> +               if (!cpu_online(policy->cpu))
> +                       return -EINVAL;
> +
> +               freq_table =
> +                       cpufreq_frequency_get_table(policy->cpu);
> +
> +               for_each_cpu(j, policy->cpus) {
> +                       pcpu = &per_cpu(cpuinfo, j);
> +                       pcpu->policy = policy;
> +                       pcpu->target_freq = policy->cur;
> +                       pcpu->freq_table = freq_table;
> +                       pcpu->freq_change_time_in_idle =
> +                               get_cpu_idle_time_us(j,
> +                                            &pcpu->freq_change_time);
> +                       pcpu->governor_enabled = 1;
> +                       smp_wmb();
> +               }
> +
> +               if (!hispeed_freq)
> +                       hispeed_freq = policy->max;
> +
> +               /*
> +                * Do not register the idle hook and create sysfs
> +                * entries if we have already done so.
> +                */
> +               if (atomic_inc_return(&active_count) > 1)
> +                       return 0;
> +
> +               rc = sysfs_create_group(cpufreq_global_kobject,
> +                               &interactive_attr_group);
> +               if (rc)
> +                       return rc;
> +
> +               break;
> +
> +       case CPUFREQ_GOV_STOP:
> +               for_each_cpu(j, policy->cpus) {
> +                       pcpu = &per_cpu(cpuinfo, j);
> +                       pcpu->governor_enabled = 0;
> +                       smp_wmb();
> +                       del_timer_sync(&pcpu->cpu_timer);
> +
> +                       /*
> +                        * Reset idle exit time since we may cancel the timer
> +                        * before it can run after the last idle exit time,
> +                        * to avoid tripping the check in idle exit for a timer
> +                        * that is trying to run.
> +                        */
> +                       pcpu->idle_exit_time = 0;
> +               }
> +
> +               flush_work(&freq_scale_down_work);
> +               if (atomic_dec_return(&active_count) > 0)
> +                       return 0;
> +
> +               sysfs_remove_group(cpufreq_global_kobject,
> +                               &interactive_attr_group);
> +
> +               break;
> +
> +       case CPUFREQ_GOV_LIMITS:
> +               if (policy->max < policy->cur)
> +                       __cpufreq_driver_target(policy,
> +                                       policy->max, CPUFREQ_RELATION_H);
> +               else if (policy->min > policy->cur)
> +                       __cpufreq_driver_target(policy,
> +                                       policy->min, CPUFREQ_RELATION_L);
> +               break;
> +       }
> +       return 0;
> +}
> +
> +static int cpufreq_interactive_idle_notifier(struct notifier_block *nb,
> +                                            unsigned long val,
> +                                            void *data)
> +{
> +       switch (val) {
> +       case SCHED_IDLE_START:
> +               cpufreq_interactive_idle_start();
> +               break;
> +       case SCHED_IDLE_END:
> +               cpufreq_interactive_idle_end();
> +               break;
> +       }
> +
> +       return 0;
> +}
> +
> +static struct notifier_block cpufreq_interactive_idle_nb = {
> +       .notifier_call = cpufreq_interactive_idle_notifier,
> +};
> +
> +static int __init cpufreq_interactive_init(void)
> +{
> +       unsigned int i;
> +       struct cpufreq_interactive_cpuinfo *pcpu;
> +       struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 };
> +
> +       go_hispeed_load = DEFAULT_GO_HISPEED_LOAD;
> +       min_sample_time = DEFAULT_MIN_SAMPLE_TIME;
> +       timer_rate = DEFAULT_TIMER_RATE;
> +
> +       /* Initalize per-cpu timers */
> +       for_each_possible_cpu(i) {
> +               pcpu = &per_cpu(cpuinfo, i);
> +               init_timer(&pcpu->cpu_timer);
> +               pcpu->cpu_timer.function = cpufreq_interactive_timer;
> +               pcpu->cpu_timer.data = i;
> +       }
> +
> +       up_task = kthread_create(cpufreq_interactive_up_task, NULL,
> +                                "kinteractiveup");
> +       if (IS_ERR(up_task))
> +               return PTR_ERR(up_task);
> +
> +       sched_setscheduler_nocheck(up_task, SCHED_FIFO, &param);
> +       get_task_struct(up_task);
> +
> +       /* No rescuer thread, bind to CPU queuing the work for possibly
> +          warm cache (probably doesn't matter much). */
> +       down_wq = alloc_workqueue("knteractive_down", 0, 1);
> +
> +       if (!down_wq)
> +               goto err_freeuptask;
> +
> +       INIT_WORK(&freq_scale_down_work,
> +                 cpufreq_interactive_freq_down);
> +
> +       spin_lock_init(&up_cpumask_lock);
> +       spin_lock_init(&down_cpumask_lock);
> +       mutex_init(&set_speed_lock);
> +
> +       sched_idle_notifier_register(&cpufreq_interactive_idle_nb);
> +
> +       return cpufreq_register_governor(&cpufreq_gov_interactive);
> +
> +err_freeuptask:
> +       put_task_struct(up_task);
> +       return -ENOMEM;
> +}
> +
> +#ifdef CONFIG_CPU_FREQ_DEFAULT_GOV_INTERACTIVE
> +fs_initcall(cpufreq_interactive_init);
> +#else
> +module_init(cpufreq_interactive_init);
> +#endif
> +
> +static void __exit cpufreq_interactive_exit(void)
> +{
> +       cpufreq_unregister_governor(&cpufreq_gov_interactive);
> +       kthread_stop(up_task);
> +       put_task_struct(up_task);
> +       destroy_workqueue(down_wq);
> +}
> +
> +module_exit(cpufreq_interactive_exit);
> +
> +MODULE_AUTHOR("Mike Chan <mike@xxxxxxxxxxx>");
> +MODULE_DESCRIPTION("'cpufreq_interactive' - A cpufreq governor for "
> +       "Latency sensitive workloads");
> +MODULE_LICENSE("GPL");
> diff --git a/include/linux/cpufreq.h b/include/linux/cpufreq.h
> index 6216115..c6126b9 100644
> --- a/include/linux/cpufreq.h
> +++ b/include/linux/cpufreq.h
> @@ -363,6 +363,9 @@ extern struct cpufreq_governor cpufreq_gov_ondemand;
>  #elif defined(CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE)
>  extern struct cpufreq_governor cpufreq_gov_conservative;
>  #define CPUFREQ_DEFAULT_GOVERNOR       (&cpufreq_gov_conservative)
> +#elif defined(CONFIG_CPU_FREQ_DEFAULT_GOV_INTERACTIVE)
> +extern struct cpufreq_governor cpufreq_gov_interactive;
> +#define CPUFREQ_DEFAULT_GOVERNOR       (&cpufreq_gov_interactive)
>  #endif
>
>
> --
> 1.7.7.6
>
> --
> To unsubscribe from this list: send the line "unsubscribe cpufreq" in
> the body of a message to majordomo@xxxxxxxxxxxxxxx
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/