[deliverable/linux.git] / drivers / cpuidle / governors / menu.c

/*
 * menu.c - the menu idle governor
 *
 * Copyright (C) 2006-2007 Adam Belay <abelay@novell.com>
 * Copyright (C) 2009 Intel Corporation
 * Author:
 *        Arjan van de Ven <arjan@linux.intel.com>
 *
 * This code is licenced under the GPL version 2 as described
 * in the COPYING file that acompanies the Linux Kernel.
 */

#include <linux/kernel.h>
#include <linux/cpuidle.h>
#include <linux/pm_qos.h>
#include <linux/time.h>
#include <linux/ktime.h>
#include <linux/hrtimer.h>
#include <linux/tick.h>
#include <linux/sched.h>
#include <linux/math64.h>
#include <linux/module.h>

#define BUCKETS 12
#define INTERVALS 8
#define RESOLUTION 1024
#define DECAY 8
#define MAX_INTERESTING 50000
#define STDDEV_THRESH 400


/*
 * Concepts and ideas behind the menu governor
 *
 * For the menu governor, there are 3 decision factors for picking a C
 * state:
 * 1) Energy break even point
 * 2) Performance impact
 * 3) Latency tolerance (from pmqos infrastructure)
 * These these three factors are treated independently.
 *
 * Energy break even point
 * -----------------------
 * C state entry and exit have an energy cost, and a certain amount of time in
 * the  C state is required to actually break even on this cost. CPUIDLE
 * provides us this duration in the "target_residency" field. So all that we
 * need is a good prediction of how long we'll be idle. Like the traditional
 * menu governor, we start with the actual known "next timer event" time.
 *
 * Since there are other source of wakeups (interrupts for example) than
 * the next timer event, this estimation is rather optimistic. To get a
 * more realistic estimate, a correction factor is applied to the estimate,
 * that is based on historic behavior. For example, if in the past the actual
 * duration always was 50% of the next timer tick, the correction factor will
 * be 0.5.
 *
 * menu uses a running average for this correction factor, however it uses a
 * set of factors, not just a single factor. This stems from the realization
 * that the ratio is dependent on the order of magnitude of the expected
 * duration; if we expect 500 milliseconds of idle time the likelihood of
 * getting an interrupt very early is much higher than if we expect 50 micro
 * seconds of idle time. A second independent factor that has big impact on
 * the actual factor is if there is (disk) IO outstanding or not.
 * (as a special twist, we consider every sleep longer than 50 milliseconds
 * as perfect; there are no power gains for sleeping longer than this)
 *
 * For these two reasons we keep an array of 12 independent factors, that gets
 * indexed based on the magnitude of the expected duration as well as the
 * "is IO outstanding" property.
 *
 * Repeatable-interval-detector
 * ----------------------------
 * There are some cases where "next timer" is a completely unusable predictor:
 * Those cases where the interval is fixed, for example due to hardware
 * interrupt mitigation, but also due to fixed transfer rate devices such as
 * mice.
 * For this, we use a different predictor: We track the duration of the last 8
 * intervals and if the stand deviation of these 8 intervals is below a
 * threshold value, we use the average of these intervals as prediction.
 *
 * Limiting Performance Impact
 * ---------------------------
 * C states, especially those with large exit latencies, can have a real
 * noticeable impact on workloads, which is not acceptable for most sysadmins,
 * and in addition, less performance has a power price of its own.
 *
 * As a general rule of thumb, menu assumes that the following heuristic
 * holds:
 *     The busier the system, the less impact of C states is acceptable
 *
 * This rule-of-thumb is implemented using a performance-multiplier:
 * If the exit latency times the performance multiplier is longer than
 * the predicted duration, the C state is not considered a candidate
 * for selection due to a too high performance impact. So the higher
 * this multiplier is, the longer we need to be idle to pick a deep C
 * state, and thus the less likely a busy CPU will hit such a deep
 * C state.
 *
 * Two factors are used in determing this multiplier:
 * a value of 10 is added for each point of "per cpu load average" we have.
 * a value of 5 points is added for each process that is waiting for
 * IO on this CPU.
 * (these values are experimentally determined)
 *
 * The load average factor gives a longer term (few seconds) input to the
 * decision, while the iowait value gives a cpu local instantanious input.
 * The iowait factor may look low, but realize that this is also already
 * represented in the system load average.
 *
 */

struct menu_device {
	int		last_state_idx;
	int             needs_update;

	unsigned int	expected_us;
	u64		predicted_us;
	unsigned int	exit_us;
	unsigned int	bucket;
	u64		correction_factor[BUCKETS];
	u32		intervals[INTERVALS];
	int		interval_ptr;
};


#define LOAD_INT(x) ((x) >> FSHIFT)
#define LOAD_FRAC(x) LOAD_INT(((x) & (FIXED_1-1)) * 100)

static int get_loadavg(void)
{
	unsigned long this = this_cpu_load();


	return LOAD_INT(this) * 10 + LOAD_FRAC(this) / 10;
}

static inline int which_bucket(unsigned int duration)
{
	int bucket = 0;

	/*
	 * We keep two groups of stats; one with no
	 * IO pending, one without.
	 * This allows us to calculate
	 * E(duration)|iowait
	 */
	if (nr_iowait_cpu(smp_processor_id()))
		bucket = BUCKETS/2;

	if (duration < 10)
		return bucket;
	if (duration < 100)
		return bucket + 1;
	if (duration < 1000)
		return bucket + 2;
	if (duration < 10000)
		return bucket + 3;
	if (duration < 100000)
		return bucket + 4;
	return bucket + 5;
}

/*
 * Return a multiplier for the exit latency that is intended
 * to take performance requirements into account.
 * The more performance critical we estimate the system
 * to be, the higher this multiplier, and thus the higher
 * the barrier to go to an expensive C state.
 */
static inline int performance_multiplier(void)
{
	int mult = 1;

	/* for higher loadavg, we are more reluctant */

	mult += 2 * get_loadavg();

	/* for IO wait tasks (per cpu!) we add 5x each */
	mult += 10 * nr_iowait_cpu(smp_processor_id());

	return mult;
}

static DEFINE_PER_CPU(struct menu_device, menu_devices);

static void menu_update(struct cpuidle_driver *drv, struct cpuidle_device *dev);

/* This implements DIV_ROUND_CLOSEST but avoids 64 bit division */
static u64 div_round64(u64 dividend, u32 divisor)
{
	return div_u64(dividend + (divisor / 2), divisor);
}

/*
 * Try detecting repeating patterns by keeping track of the last 8
 * intervals, and checking if the standard deviation of that set
 * of points is below a threshold. If it is... then use the
 * average of these 8 points as the estimated value.
 */
static void detect_repeating_patterns(struct menu_device *data)
{
	int i;
	uint64_t avg = 0;
	uint64_t stddev = 0; /* contains the square of the std deviation */

	/* first calculate average and standard deviation of the past */
	for (i = 0; i < INTERVALS; i++)
		avg += data->intervals[i];
	avg = avg / INTERVALS;

	/* if the avg is beyond the known next tick, it's worthless */
	if (avg > data->expected_us)
		return;

	for (i = 0; i < INTERVALS; i++)
		stddev += (data->intervals[i] - avg) *
			  (data->intervals[i] - avg);

	stddev = stddev / INTERVALS;

	/*
	 * now.. if stddev is small.. then assume we have a
	 * repeating pattern and predict we keep doing this.
	 */

	if (avg && stddev < STDDEV_THRESH)
		data->predicted_us = avg;
}

/**
 * menu_select - selects the next idle state to enter
 * @drv: cpuidle driver containing state data
 * @dev: the CPU
 */
static int menu_select(struct cpuidle_driver *drv, struct cpuidle_device *dev)
{
	struct menu_device *data = &__get_cpu_var(menu_devices);
	int latency_req = pm_qos_request(PM_QOS_CPU_DMA_LATENCY);
	int power_usage = -1;
	int i;
	int multiplier;
	struct timespec t;

	if (data->needs_update) {
		menu_update(drv, dev);
		data->needs_update = 0;
	}

	data->last_state_idx = 0;
	data->exit_us = 0;

	/* Special case when user has set very strict latency requirement */
	if (unlikely(latency_req == 0))
		return 0;

	/* determine the expected residency time, round up */
	t = ktime_to_timespec(tick_nohz_get_sleep_length());
	data->expected_us =
		t.tv_sec * USEC_PER_SEC + t.tv_nsec / NSEC_PER_USEC;


	data->bucket = which_bucket(data->expected_us);

	multiplier = performance_multiplier();

	/*
	 * if the correction factor is 0 (eg first time init or cpu hotplug
	 * etc), we actually want to start out with a unity factor.
	 */
	if (data->correction_factor[data->bucket] == 0)
		data->correction_factor[data->bucket] = RESOLUTION * DECAY;

	/* Make sure to round up for half microseconds */
	data->predicted_us = div_round64(data->expected_us * data->correction_factor[data->bucket],
					 RESOLUTION * DECAY);

	detect_repeating_patterns(data);

	/*
	 * We want to default to C1 (hlt), not to busy polling
	 * unless the timer is happening really really soon.
	 */
	if (data->expected_us > 5 &&
	    !drv->states[CPUIDLE_DRIVER_STATE_START].disabled &&
		dev->states_usage[CPUIDLE_DRIVER_STATE_START].disable == 0)
		data->last_state_idx = CPUIDLE_DRIVER_STATE_START;

	/*
	 * Find the idle state with the lowest power while satisfying
	 * our constraints.
	 */
	for (i = CPUIDLE_DRIVER_STATE_START; i < drv->state_count; i++) {
		struct cpuidle_state *s = &drv->states[i];
		struct cpuidle_state_usage *su = &dev->states_usage[i];

		if (s->disabled || su->disable)
			continue;
		if (s->target_residency > data->predicted_us)
			continue;
		if (s->exit_latency > latency_req)
			continue;
		if (s->exit_latency * multiplier > data->predicted_us)
			continue;

		if (s->power_usage < power_usage) {
			power_usage = s->power_usage;
			data->last_state_idx = i;
			data->exit_us = s->exit_latency;
		}
	}

	return data->last_state_idx;
}

/**
 * menu_reflect - records that data structures need update
 * @dev: the CPU
 * @index: the index of actual entered state
 *
 * NOTE: it's important to be fast here because this operation will add to
 *       the overall exit latency.
 */
static void menu_reflect(struct cpuidle_device *dev, int index)
{
	struct menu_device *data = &__get_cpu_var(menu_devices);
	data->last_state_idx = index;
	if (index >= 0)
		data->needs_update = 1;
}

/**
 * menu_update - attempts to guess what happened after entry
 * @drv: cpuidle driver containing state data
 * @dev: the CPU
 */
static void menu_update(struct cpuidle_driver *drv, struct cpuidle_device *dev)
{
	struct menu_device *data = &__get_cpu_var(menu_devices);
	int last_idx = data->last_state_idx;
	unsigned int last_idle_us = cpuidle_get_last_residency(dev);
	struct cpuidle_state *target = &drv->states[last_idx];
	unsigned int measured_us;
	u64 new_factor;

	/*
	 * Ugh, this idle state doesn't support residency measurements, so we
	 * are basically lost in the dark.  As a compromise, assume we slept
	 * for the whole expected time.
	 */
	if (unlikely(!(target->flags & CPUIDLE_FLAG_TIME_VALID)))
		last_idle_us = data->expected_us;


	measured_us = last_idle_us;

	/*
	 * We correct for the exit latency; we are assuming here that the
	 * exit latency happens after the event that we're interested in.
	 */
	if (measured_us > data->exit_us)
		measured_us -= data->exit_us;


	/* update our correction ratio */

	new_factor = data->correction_factor[data->bucket]
			* (DECAY - 1) / DECAY;

	if (data->expected_us > 0 && measured_us < MAX_INTERESTING)
		new_factor += RESOLUTION * measured_us / data->expected_us;
	else
		/*
		 * we were idle so long that we count it as a perfect
		 * prediction
		 */
		new_factor += RESOLUTION;

	/*
	 * We don't want 0 as factor; we always want at least
	 * a tiny bit of estimated time.
	 */
	if (new_factor == 0)
		new_factor = 1;

	data->correction_factor[data->bucket] = new_factor;

	/* update the repeating-pattern data */
	data->intervals[data->interval_ptr++] = last_idle_us;
	if (data->interval_ptr >= INTERVALS)
		data->interval_ptr = 0;
}

/**
 * menu_enable_device - scans a CPU's states and does setup
 * @drv: cpuidle driver
 * @dev: the CPU
 */
static int menu_enable_device(struct cpuidle_driver *drv,
				struct cpuidle_device *dev)
{
	struct menu_device *data = &per_cpu(menu_devices, dev->cpu);

	memset(data, 0, sizeof(struct menu_device));

	return 0;
}

static struct cpuidle_governor menu_governor = {
	.name =		"menu",
	.rating =	20,
	.enable =	menu_enable_device,
	.select =	menu_select,
	.reflect =	menu_reflect,
	.owner =	THIS_MODULE,
};

/**
 * init_menu - initializes the governor
 */
static int __init init_menu(void)
{
	return cpuidle_register_governor(&menu_governor);
}

/**
 * exit_menu - exits the governor
 */
static void __exit exit_menu(void)
{
	cpuidle_unregister_governor(&menu_governor);
}

MODULE_LICENSE("GPL");
module_init(init_menu);
module_exit(exit_menu);
Commit	Line	Data
4f86d3a8 LB	1	/*
	2	* menu.c - the menu idle governor
	3	*
	4	* Copyright (C) 2006-2007 Adam Belay <abelay@novell.com>
69d25870 AV	5	* Copyright (C) 2009 Intel Corporation
	6	* Author:
	7	* Arjan van de Ven <arjan@linux.intel.com>
4f86d3a8	8	*
69d25870 AV	9	* This code is licenced under the GPL version 2 as described
69d25870 AV	10	* in the COPYING file that acompanies the Linux Kernel.
4f86d3a8 LB	11	*/
	12
	13	#include <linux/kernel.h>
	14	#include <linux/cpuidle.h>
e8db0be1	15	#include <linux/pm_qos.h>
4f86d3a8 LB	16	#include <linux/time.h>
	17	#include <linux/ktime.h>
	18	#include <linux/hrtimer.h>
	19	#include <linux/tick.h>
69d25870	20	#include <linux/sched.h>
5787536e	21	#include <linux/math64.h>
884b17e1	22	#include <linux/module.h>
4f86d3a8	23
69d25870	24	#define BUCKETS 12
1f85f87d	25	#define INTERVALS 8
69d25870	26	#define RESOLUTION 1024
1f85f87d	27	#define DECAY 8
69d25870	28	#define MAX_INTERESTING 50000
1f85f87d AV	29	#define STDDEV_THRESH 400
1f85f87d AV	30
69d25870 AV	31
	32	/*
	33	* Concepts and ideas behind the menu governor
	34	*
	35	* For the menu governor, there are 3 decision factors for picking a C
	36	* state:
	37	* 1) Energy break even point
	38	* 2) Performance impact
	39	* 3) Latency tolerance (from pmqos infrastructure)
	40	* These these three factors are treated independently.
	41	*
	42	* Energy break even point
	43	* -----------------------
	44	* C state entry and exit have an energy cost, and a certain amount of time in
	45	* the C state is required to actually break even on this cost. CPUIDLE
	46	* provides us this duration in the "target_residency" field. So all that we
	47	* need is a good prediction of how long we'll be idle. Like the traditional
	48	* menu governor, we start with the actual known "next timer event" time.
	49	*
	50	* Since there are other source of wakeups (interrupts for example) than
	51	* the next timer event, this estimation is rather optimistic. To get a
	52	* more realistic estimate, a correction factor is applied to the estimate,
	53	* that is based on historic behavior. For example, if in the past the actual
	54	* duration always was 50% of the next timer tick, the correction factor will
	55	* be 0.5.
	56	*
	57	* menu uses a running average for this correction factor, however it uses a
	58	* set of factors, not just a single factor. This stems from the realization
	59	* that the ratio is dependent on the order of magnitude of the expected
	60	* duration; if we expect 500 milliseconds of idle time the likelihood of
	61	* getting an interrupt very early is much higher than if we expect 50 micro
	62	* seconds of idle time. A second independent factor that has big impact on
	63	* the actual factor is if there is (disk) IO outstanding or not.
	64	* (as a special twist, we consider every sleep longer than 50 milliseconds
	65	* as perfect; there are no power gains for sleeping longer than this)
	66	*
	67	* For these two reasons we keep an array of 12 independent factors, that gets
	68	* indexed based on the magnitude of the expected duration as well as the
	69	* "is IO outstanding" property.
	70	*
1f85f87d AV	71	* Repeatable-interval-detector
	72	* ----------------------------
	73	* There are some cases where "next timer" is a completely unusable predictor:
	74	* Those cases where the interval is fixed, for example due to hardware
	75	* interrupt mitigation, but also due to fixed transfer rate devices such as
	76	* mice.
	77	* For this, we use a different predictor: We track the duration of the last 8
	78	* intervals and if the stand deviation of these 8 intervals is below a
	79	* threshold value, we use the average of these intervals as prediction.
	80	*
69d25870 AV	81	* Limiting Performance Impact
	82	* ---------------------------
	83	* C states, especially those with large exit latencies, can have a real
20e3341b	84	* noticeable impact on workloads, which is not acceptable for most sysadmins,
69d25870 AV	85	* and in addition, less performance has a power price of its own.
	86	*
	87	* As a general rule of thumb, menu assumes that the following heuristic
	88	* holds:
	89	* The busier the system, the less impact of C states is acceptable
	90	*
	91	* This rule-of-thumb is implemented using a performance-multiplier:
	92	* If the exit latency times the performance multiplier is longer than
	93	* the predicted duration, the C state is not considered a candidate
	94	* for selection due to a too high performance impact. So the higher
	95	* this multiplier is, the longer we need to be idle to pick a deep C
	96	* state, and thus the less likely a busy CPU will hit such a deep
	97	* C state.
	98	*
	99	* Two factors are used in determing this multiplier:
	100	* a value of 10 is added for each point of "per cpu load average" we have.
	101	* a value of 5 points is added for each process that is waiting for
	102	* IO on this CPU.
	103	* (these values are experimentally determined)
	104	*
	105	* The load average factor gives a longer term (few seconds) input to the
	106	* decision, while the iowait value gives a cpu local instantanious input.
	107	* The iowait factor may look low, but realize that this is also already
	108	* represented in the system load average.
	109	*
	110	*/
4f86d3a8 LB	111
	112	struct menu_device {
	113	int last_state_idx;
672917dc	114	int needs_update;
4f86d3a8 LB	115
4f86d3a8 LB	116	unsigned int expected_us;
56e6943b	117	u64 predicted_us;
69d25870 AV	118	unsigned int exit_us;
	119	unsigned int bucket;
	120	u64 correction_factor[BUCKETS];
1f85f87d AV	121	u32 intervals[INTERVALS];
1f85f87d AV	122	int interval_ptr;
4f86d3a8 LB	123	};
4f86d3a8 LB	124
69d25870 AV	125
	126	#define LOAD_INT(x) ((x) >> FSHIFT)
	127	#define LOAD_FRAC(x) LOAD_INT(((x) & (FIXED_1-1)) * 100)
	128
	129	static int get_loadavg(void)
	130	{
	131	unsigned long this = this_cpu_load();
	132
	133
	134	return LOAD_INT(this) * 10 + LOAD_FRAC(this) / 10;
	135	}
	136
	137	static inline int which_bucket(unsigned int duration)
	138	{
	139	int bucket = 0;
	140
	141	/*
	142	* We keep two groups of stats; one with no
	143	* IO pending, one without.
	144	* This allows us to calculate
	145	* E(duration)\|iowait
	146	*/
8c215bd3	147	if (nr_iowait_cpu(smp_processor_id()))
69d25870 AV	148	bucket = BUCKETS/2;
	149
	150	if (duration < 10)
	151	return bucket;
	152	if (duration < 100)
	153	return bucket + 1;
	154	if (duration < 1000)
	155	return bucket + 2;
	156	if (duration < 10000)
	157	return bucket + 3;
	158	if (duration < 100000)
	159	return bucket + 4;
	160	return bucket + 5;
	161	}
	162
	163	/*
	164	* Return a multiplier for the exit latency that is intended
	165	* to take performance requirements into account.
	166	* The more performance critical we estimate the system
	167	* to be, the higher this multiplier, and thus the higher
	168	* the barrier to go to an expensive C state.
	169	*/
	170	static inline int performance_multiplier(void)
	171	{
	172	int mult = 1;
	173
	174	/* for higher loadavg, we are more reluctant */
	175
	176	mult += 2 * get_loadavg();
	177
	178	/* for IO wait tasks (per cpu!) we add 5x each */
8c215bd3	179	mult += 10 * nr_iowait_cpu(smp_processor_id());
69d25870 AV	180
	181	return mult;
	182	}
	183
4f86d3a8 LB	184	static DEFINE_PER_CPU(struct menu_device, menu_devices);
4f86d3a8 LB	185
46bcfad7	186	static void menu_update(struct cpuidle_driver drv, struct cpuidle_device dev);
672917dc	187
5787536e SH	188	/* This implements DIV_ROUND_CLOSEST but avoids 64 bit division */
	189	static u64 div_round64(u64 dividend, u32 divisor)
	190	{
	191	return div_u64(dividend + (divisor / 2), divisor);
	192	}
	193
1f85f87d AV	194	/*
	195	* Try detecting repeating patterns by keeping track of the last 8
	196	* intervals, and checking if the standard deviation of that set
	197	* of points is below a threshold. If it is... then use the
	198	* average of these 8 points as the estimated value.
	199	*/
	200	static void detect_repeating_patterns(struct menu_device *data)
	201	{
	202	int i;
	203	uint64_t avg = 0;
	204	uint64_t stddev = 0; /* contains the square of the std deviation */
	205
	206	/* first calculate average and standard deviation of the past */
	207	for (i = 0; i < INTERVALS; i++)
	208	avg += data->intervals[i];
	209	avg = avg / INTERVALS;
	210
	211	/* if the avg is beyond the known next tick, it's worthless */
	212	if (avg > data->expected_us)
	213	return;
	214
	215	for (i = 0; i < INTERVALS; i++)
	216	stddev += (data->intervals[i] - avg) *
	217	(data->intervals[i] - avg);
	218
	219	stddev = stddev / INTERVALS;
	220
	221	/*
	222	* now.. if stddev is small.. then assume we have a
	223	* repeating pattern and predict we keep doing this.
	224	*/
	225
	226	if (avg && stddev < STDDEV_THRESH)
	227	data->predicted_us = avg;
	228	}
	229
4f86d3a8 LB	230	/**
4f86d3a8 LB	231	* menu_select - selects the next idle state to enter
46bcfad7	232	* @drv: cpuidle driver containing state data
4f86d3a8 LB	233	* @dev: the CPU
4f86d3a8 LB	234	*/
46bcfad7	235	static int menu_select(struct cpuidle_driver drv, struct cpuidle_device dev)
4f86d3a8 LB	236	{
4f86d3a8 LB	237	struct menu_device *data = &__get_cpu_var(menu_devices);
ed77134b	238	int latency_req = pm_qos_request(PM_QOS_CPU_DMA_LATENCY);
02401c06	239	int power_usage = -1;
4f86d3a8	240	int i;
69d25870	241	int multiplier;
7467571f	242	struct timespec t;
69d25870	243
672917dc	244	if (data->needs_update) {
46bcfad7	245	menu_update(drv, dev);
672917dc CZ	246	data->needs_update = 0;
	247	}
	248
1c6fe036 AV	249	data->last_state_idx = 0;
	250	data->exit_us = 0;
	251
a2bd9202	252	/* Special case when user has set very strict latency requirement */
69d25870	253	if (unlikely(latency_req == 0))
a2bd9202	254	return 0;
a2bd9202	255
69d25870	256	/* determine the expected residency time, round up */
7467571f	257	t = ktime_to_timespec(tick_nohz_get_sleep_length());
4f86d3a8	258	data->expected_us =
7467571f	259	t.tv_sec * USEC_PER_SEC + t.tv_nsec / NSEC_PER_USEC;
69d25870 AV	260
	261
	262	data->bucket = which_bucket(data->expected_us);
	263
	264	multiplier = performance_multiplier();
	265
	266	/*
	267	* if the correction factor is 0 (eg first time init or cpu hotplug
	268	* etc), we actually want to start out with a unity factor.
	269	*/
	270	if (data->correction_factor[data->bucket] == 0)
	271	data->correction_factor[data->bucket] = RESOLUTION * DECAY;
	272
	273	/* Make sure to round up for half microseconds */
5787536e SH	274	data->predicted_us = div_round64(data->expected_us * data->correction_factor[data->bucket],
5787536e SH	275	RESOLUTION * DECAY);
69d25870	276
1f85f87d AV	277	detect_repeating_patterns(data);
1f85f87d AV	278
69d25870 AV	279	/*
	280	* We want to default to C1 (hlt), not to busy polling
	281	* unless the timer is happening really really soon.
	282	*/
3a53396b	283	if (data->expected_us > 5 &&
cbc9ef02	284	!drv->states[CPUIDLE_DRIVER_STATE_START].disabled &&
dc7fd275	285	dev->states_usage[CPUIDLE_DRIVER_STATE_START].disable == 0)
69d25870	286	data->last_state_idx = CPUIDLE_DRIVER_STATE_START;
4f86d3a8	287
71abbbf8 AL	288	/*
	289	* Find the idle state with the lowest power while satisfying
	290	* our constraints.
	291	*/
46bcfad7 DD	292	for (i = CPUIDLE_DRIVER_STATE_START; i < drv->state_count; i++) {
46bcfad7 DD	293	struct cpuidle_state *s = &drv->states[i];
dc7fd275	294	struct cpuidle_state_usage *su = &dev->states_usage[i];
4f86d3a8	295
cbc9ef02	296	if (s->disabled \|\| su->disable)
3a53396b	297	continue;
4f86d3a8	298	if (s->target_residency > data->predicted_us)
71abbbf8	299	continue;
a2bd9202	300	if (s->exit_latency > latency_req)
71abbbf8	301	continue;
69d25870	302	if (s->exit_latency * multiplier > data->predicted_us)
71abbbf8 AL	303	continue;
	304
	305	if (s->power_usage < power_usage) {
	306	power_usage = s->power_usage;
	307	data->last_state_idx = i;
	308	data->exit_us = s->exit_latency;
	309	}
4f86d3a8 LB	310	}
4f86d3a8 LB	311
69d25870	312	return data->last_state_idx;
4f86d3a8 LB	313	}
	314
	315	/**
672917dc	316	* menu_reflect - records that data structures need update
4f86d3a8	317	* @dev: the CPU
e978aa7d	318	* @index: the index of actual entered state
4f86d3a8 LB	319	*
	320	* NOTE: it's important to be fast here because this operation will add to
	321	* the overall exit latency.
	322	*/
e978aa7d	323	static void menu_reflect(struct cpuidle_device *dev, int index)
672917dc CZ	324	{
672917dc CZ	325	struct menu_device *data = &__get_cpu_var(menu_devices);
e978aa7d DD	326	data->last_state_idx = index;
	327	if (index >= 0)
	328	data->needs_update = 1;
672917dc CZ	329	}
	330
	331	/**
	332	* menu_update - attempts to guess what happened after entry
46bcfad7	333	* @drv: cpuidle driver containing state data
672917dc CZ	334	* @dev: the CPU
672917dc CZ	335	*/
46bcfad7	336	static void menu_update(struct cpuidle_driver drv, struct cpuidle_device dev)
4f86d3a8 LB	337	{
	338	struct menu_device *data = &__get_cpu_var(menu_devices);
	339	int last_idx = data->last_state_idx;
320eee77	340	unsigned int last_idle_us = cpuidle_get_last_residency(dev);
46bcfad7	341	struct cpuidle_state *target = &drv->states[last_idx];
320eee77	342	unsigned int measured_us;
69d25870	343	u64 new_factor;
4f86d3a8 LB	344
	345	/*
	346	* Ugh, this idle state doesn't support residency measurements, so we
	347	* are basically lost in the dark. As a compromise, assume we slept
69d25870	348	* for the whole expected time.
4f86d3a8	349	*/
320eee77	350	if (unlikely(!(target->flags & CPUIDLE_FLAG_TIME_VALID)))
69d25870 AV	351	last_idle_us = data->expected_us;
	352
	353
	354	measured_us = last_idle_us;
4f86d3a8	355
320eee77	356	/*
69d25870 AV	357	* We correct for the exit latency; we are assuming here that the
69d25870 AV	358	* exit latency happens after the event that we're interested in.
320eee77	359	*/
69d25870 AV	360	if (measured_us > data->exit_us)
	361	measured_us -= data->exit_us;
	362
	363
	364	/* update our correction ratio */
	365
	366	new_factor = data->correction_factor[data->bucket]
	367	* (DECAY - 1) / DECAY;
	368
1c6fe036	369	if (data->expected_us > 0 && measured_us < MAX_INTERESTING)
69d25870	370	new_factor += RESOLUTION * measured_us / data->expected_us;
320eee77	371	else
69d25870 AV	372	/*
	373	* we were idle so long that we count it as a perfect
	374	* prediction
	375	*/
	376	new_factor += RESOLUTION;
320eee77	377
69d25870 AV	378	/*
	379	* We don't want 0 as factor; we always want at least
	380	* a tiny bit of estimated time.
	381	*/
	382	if (new_factor == 0)
	383	new_factor = 1;
320eee77	384
69d25870	385	data->correction_factor[data->bucket] = new_factor;
1f85f87d AV	386
	387	/* update the repeating-pattern data */
	388	data->intervals[data->interval_ptr++] = last_idle_us;
	389	if (data->interval_ptr >= INTERVALS)
	390	data->interval_ptr = 0;
4f86d3a8 LB	391	}
	392
	393	/**
	394	* menu_enable_device - scans a CPU's states and does setup
46bcfad7	395	* @drv: cpuidle driver
4f86d3a8 LB	396	* @dev: the CPU
4f86d3a8 LB	397	*/
46bcfad7 DD	398	static int menu_enable_device(struct cpuidle_driver *drv,
46bcfad7 DD	399	struct cpuidle_device *dev)
4f86d3a8 LB	400	{
	401	struct menu_device *data = &per_cpu(menu_devices, dev->cpu);
	402
	403	memset(data, 0, sizeof(struct menu_device));
	404
	405	return 0;
	406	}
	407
	408	static struct cpuidle_governor menu_governor = {
	409	.name = "menu",
	410	.rating = 20,
	411	.enable = menu_enable_device,
	412	.select = menu_select,
	413	.reflect = menu_reflect,
	414	.owner = THIS_MODULE,
	415	};
	416
	417	/**
	418	* init_menu - initializes the governor
	419	*/
	420	static int __init init_menu(void)
	421	{
	422	return cpuidle_register_governor(&menu_governor);
	423	}
	424
	425	/**
	426	* exit_menu - exits the governor
	427	*/
	428	static void __exit exit_menu(void)
	429	{
	430	cpuidle_unregister_governor(&menu_governor);
	431	}
	432
	433	MODULE_LICENSE("GPL");
	434	module_init(init_menu);
	435	module_exit(exit_menu);