[net-next-2.6.git] / drivers / acpi / acpi_pad.c

/*
 * acpi_pad.c ACPI Processor Aggregator Driver
 *
 * Copyright (c) 2009, Intel Corporation.
 *
 * This program is free software; you can redistribute it and/or modify it
 * under the terms and conditions of the GNU General Public License,
 * version 2, as published by the Free Software Foundation.
 *
 * This program is distributed in the hope 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.
 *
 * You should have received a copy of the GNU General Public License along with
 * this program; if not, write to the Free Software Foundation, Inc.,
 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
 *
 */

#include <linux/kernel.h>
#include <linux/cpumask.h>
#include <linux/module.h>
#include <linux/init.h>
#include <linux/types.h>
#include <linux/kthread.h>
#include <linux/freezer.h>
#include <linux/cpu.h>
#include <linux/clockchips.h>
#include <linux/slab.h>
#include <acpi/acpi_bus.h>
#include <acpi/acpi_drivers.h>

#define ACPI_PROCESSOR_AGGREGATOR_CLASS	"acpi_pad"
#define ACPI_PROCESSOR_AGGREGATOR_DEVICE_NAME "Processor Aggregator"
#define ACPI_PROCESSOR_AGGREGATOR_NOTIFY 0x80
static DEFINE_MUTEX(isolated_cpus_lock);

#define MWAIT_SUBSTATE_MASK	(0xf)
#define MWAIT_CSTATE_MASK	(0xf)
#define MWAIT_SUBSTATE_SIZE	(4)
#define CPUID_MWAIT_LEAF (5)
#define CPUID5_ECX_EXTENSIONS_SUPPORTED (0x1)
#define CPUID5_ECX_INTERRUPT_BREAK	(0x2)
static unsigned long power_saving_mwait_eax;
static void power_saving_mwait_init(void)
{
	unsigned int eax, ebx, ecx, edx;
	unsigned int highest_cstate = 0;
	unsigned int highest_subcstate = 0;
	int i;

	if (!boot_cpu_has(X86_FEATURE_MWAIT))
		return;
	if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
		return;

	cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &edx);

	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
	    !(ecx & CPUID5_ECX_INTERRUPT_BREAK))
		return;

	edx >>= MWAIT_SUBSTATE_SIZE;
	for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
		if (edx & MWAIT_SUBSTATE_MASK) {
			highest_cstate = i;
			highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
		}
	}
	power_saving_mwait_eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
		(highest_subcstate - 1);

	for_each_online_cpu(i)
		clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ON, &i);

#if defined(CONFIG_GENERIC_TIME) && defined(CONFIG_X86)
	switch (boot_cpu_data.x86_vendor) {
	case X86_VENDOR_AMD:
	case X86_VENDOR_INTEL:
		/*
		 * AMD Fam10h TSC will tick in all
		 * C/P/S0/S1 states when this bit is set.
		 */
		if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
			return;

		/*FALL THROUGH*/
	default:
		/* TSC could halt in idle, so notify users */
		mark_tsc_unstable("TSC halts in idle");
	}
#endif
}

static unsigned long cpu_weight[NR_CPUS];
static int tsk_in_cpu[NR_CPUS] = {[0 ... NR_CPUS-1] = -1};
static DECLARE_BITMAP(pad_busy_cpus_bits, NR_CPUS);
static void round_robin_cpu(unsigned int tsk_index)
{
	struct cpumask *pad_busy_cpus = to_cpumask(pad_busy_cpus_bits);
	cpumask_var_t tmp;
	int cpu;
	unsigned long min_weight = -1;
	unsigned long uninitialized_var(preferred_cpu);

	if (!alloc_cpumask_var(&tmp, GFP_KERNEL))
		return;

	mutex_lock(&isolated_cpus_lock);
	cpumask_clear(tmp);
	for_each_cpu(cpu, pad_busy_cpus)
		cpumask_or(tmp, tmp, topology_thread_cpumask(cpu));
	cpumask_andnot(tmp, cpu_online_mask, tmp);
	/* avoid HT sibilings if possible */
	if (cpumask_empty(tmp))
		cpumask_andnot(tmp, cpu_online_mask, pad_busy_cpus);
	if (cpumask_empty(tmp)) {
		mutex_unlock(&isolated_cpus_lock);
		return;
	}
	for_each_cpu(cpu, tmp) {
		if (cpu_weight[cpu] < min_weight) {
			min_weight = cpu_weight[cpu];
			preferred_cpu = cpu;
		}
	}

	if (tsk_in_cpu[tsk_index] != -1)
		cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus);
	tsk_in_cpu[tsk_index] = preferred_cpu;
	cpumask_set_cpu(preferred_cpu, pad_busy_cpus);
	cpu_weight[preferred_cpu]++;
	mutex_unlock(&isolated_cpus_lock);

	set_cpus_allowed_ptr(current, cpumask_of(preferred_cpu));
}

static void exit_round_robin(unsigned int tsk_index)
{
	struct cpumask *pad_busy_cpus = to_cpumask(pad_busy_cpus_bits);
	cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus);
	tsk_in_cpu[tsk_index] = -1;
}

static unsigned int idle_pct = 5; /* percentage */
static unsigned int round_robin_time = 10; /* second */
static int power_saving_thread(void *data)
{
	struct sched_param param = {.sched_priority = 1};
	int do_sleep;
	unsigned int tsk_index = (unsigned long)data;
	u64 last_jiffies = 0;

	sched_setscheduler(current, SCHED_RR, &param);

	while (!kthread_should_stop()) {
		int cpu;
		u64 expire_time;

		try_to_freeze();

		/* round robin to cpus */
		if (last_jiffies + round_robin_time * HZ < jiffies) {
			last_jiffies = jiffies;
			round_robin_cpu(tsk_index);
		}

		do_sleep = 0;

		expire_time = jiffies + HZ * (100 - idle_pct) / 100;

		while (!need_resched()) {
			local_irq_disable();
			cpu = smp_processor_id();
			clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER,
				&cpu);
			stop_critical_timings();

			__monitor((void *)&current_thread_info()->flags, 0, 0);
			smp_mb();
			if (!need_resched())
				__mwait(power_saving_mwait_eax, 1);

			start_critical_timings();
			clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT,
				&cpu);
			local_irq_enable();

			if (jiffies > expire_time) {
				do_sleep = 1;
				break;
			}
		}

		/*
		 * current sched_rt has threshold for rt task running time.
		 * When a rt task uses 95% CPU time, the rt thread will be
		 * scheduled out for 5% CPU time to not starve other tasks. But
		 * the mechanism only works when all CPUs have RT task running,
		 * as if one CPU hasn't RT task, RT task from other CPUs will
		 * borrow CPU time from this CPU and cause RT task use > 95%
		 * CPU time. To make 'avoid starvation' work, takes a nap here.
		 */
		if (do_sleep)
			schedule_timeout_killable(HZ * idle_pct / 100);
	}

	exit_round_robin(tsk_index);
	return 0;
}

static struct task_struct *ps_tsks[NR_CPUS];
static unsigned int ps_tsk_num;
static int create_power_saving_task(void)
{
	int rc = -ENOMEM;

	ps_tsks[ps_tsk_num] = kthread_run(power_saving_thread,
		(void *)(unsigned long)ps_tsk_num,
		"power_saving/%d", ps_tsk_num);
	rc = IS_ERR(ps_tsks[ps_tsk_num]) ? PTR_ERR(ps_tsks[ps_tsk_num]) : 0;
	if (!rc)
		ps_tsk_num++;
	else
		ps_tsks[ps_tsk_num] = NULL;

	return rc;
}

static void destroy_power_saving_task(void)
{
	if (ps_tsk_num > 0) {
		ps_tsk_num--;
		kthread_stop(ps_tsks[ps_tsk_num]);
		ps_tsks[ps_tsk_num] = NULL;
	}
}

static void set_power_saving_task_num(unsigned int num)
{
	if (num > ps_tsk_num) {
		while (ps_tsk_num < num) {
			if (create_power_saving_task())
				return;
		}
	} else if (num < ps_tsk_num) {
		while (ps_tsk_num > num)
			destroy_power_saving_task();
	}
}

static void acpi_pad_idle_cpus(unsigned int num_cpus)
{
	get_online_cpus();

	num_cpus = min_t(unsigned int, num_cpus, num_online_cpus());
	set_power_saving_task_num(num_cpus);

	put_online_cpus();
}

static uint32_t acpi_pad_idle_cpus_num(void)
{
	return ps_tsk_num;
}

static ssize_t acpi_pad_rrtime_store(struct device *dev,
	struct device_attribute *attr, const char *buf, size_t count)
{
	unsigned long num;
	if (strict_strtoul(buf, 0, &num))
		return -EINVAL;
	if (num < 1 || num >= 100)
		return -EINVAL;
	mutex_lock(&isolated_cpus_lock);
	round_robin_time = num;
	mutex_unlock(&isolated_cpus_lock);
	return count;
}

static ssize_t acpi_pad_rrtime_show(struct device *dev,
	struct device_attribute *attr, char *buf)
{
	return scnprintf(buf, PAGE_SIZE, "%d", round_robin_time);
}
static DEVICE_ATTR(rrtime, S_IRUGO|S_IWUSR,
	acpi_pad_rrtime_show,
	acpi_pad_rrtime_store);

static ssize_t acpi_pad_idlepct_store(struct device *dev,
	struct device_attribute *attr, const char *buf, size_t count)
{
	unsigned long num;
	if (strict_strtoul(buf, 0, &num))
		return -EINVAL;
	if (num < 1 || num >= 100)
		return -EINVAL;
	mutex_lock(&isolated_cpus_lock);
	idle_pct = num;
	mutex_unlock(&isolated_cpus_lock);
	return count;
}

static ssize_t acpi_pad_idlepct_show(struct device *dev,
	struct device_attribute *attr, char *buf)
{
	return scnprintf(buf, PAGE_SIZE, "%d", idle_pct);
}
static DEVICE_ATTR(idlepct, S_IRUGO|S_IWUSR,
	acpi_pad_idlepct_show,
	acpi_pad_idlepct_store);

static ssize_t acpi_pad_idlecpus_store(struct device *dev,
	struct device_attribute *attr, const char *buf, size_t count)
{
	unsigned long num;
	if (strict_strtoul(buf, 0, &num))
		return -EINVAL;
	mutex_lock(&isolated_cpus_lock);
	acpi_pad_idle_cpus(num);
	mutex_unlock(&isolated_cpus_lock);
	return count;
}

static ssize_t acpi_pad_idlecpus_show(struct device *dev,
	struct device_attribute *attr, char *buf)
{
	return cpumask_scnprintf(buf, PAGE_SIZE,
		to_cpumask(pad_busy_cpus_bits));
}
static DEVICE_ATTR(idlecpus, S_IRUGO|S_IWUSR,
	acpi_pad_idlecpus_show,
	acpi_pad_idlecpus_store);

static int acpi_pad_add_sysfs(struct acpi_device *device)
{
	int result;

	result = device_create_file(&device->dev, &dev_attr_idlecpus);
	if (result)
		return -ENODEV;
	result = device_create_file(&device->dev, &dev_attr_idlepct);
	if (result) {
		device_remove_file(&device->dev, &dev_attr_idlecpus);
		return -ENODEV;
	}
	result = device_create_file(&device->dev, &dev_attr_rrtime);
	if (result) {
		device_remove_file(&device->dev, &dev_attr_idlecpus);
		device_remove_file(&device->dev, &dev_attr_idlepct);
		return -ENODEV;
	}
	return 0;
}

static void acpi_pad_remove_sysfs(struct acpi_device *device)
{
	device_remove_file(&device->dev, &dev_attr_idlecpus);
	device_remove_file(&device->dev, &dev_attr_idlepct);
	device_remove_file(&device->dev, &dev_attr_rrtime);
}

/* Query firmware how many CPUs should be idle */
static int acpi_pad_pur(acpi_handle handle, int *num_cpus)
{
	struct acpi_buffer buffer = {ACPI_ALLOCATE_BUFFER, NULL};
	union acpi_object *package;
	int rev, num, ret = -EINVAL;

	if (ACPI_FAILURE(acpi_evaluate_object(handle, "_PUR", NULL, &buffer)))
		return -EINVAL;

	if (!buffer.length || !buffer.pointer)
		return -EINVAL;

	package = buffer.pointer;
	if (package->type != ACPI_TYPE_PACKAGE || package->package.count != 2)
		goto out;
	rev = package->package.elements[0].integer.value;
	num = package->package.elements[1].integer.value;
	if (rev != 1 || num < 0)
		goto out;
	*num_cpus = num;
	ret = 0;
out:
	kfree(buffer.pointer);
	return ret;
}

/* Notify firmware how many CPUs are idle */
static void acpi_pad_ost(acpi_handle handle, int stat,
	uint32_t idle_cpus)
{
	union acpi_object params[3] = {
		{.type = ACPI_TYPE_INTEGER,},
		{.type = ACPI_TYPE_INTEGER,},
		{.type = ACPI_TYPE_BUFFER,},
	};
	struct acpi_object_list arg_list = {3, params};

	params[0].integer.value = ACPI_PROCESSOR_AGGREGATOR_NOTIFY;
	params[1].integer.value =  stat;
	params[2].buffer.length = 4;
	params[2].buffer.pointer = (void *)&idle_cpus;
	acpi_evaluate_object(handle, "_OST", &arg_list, NULL);
}

static void acpi_pad_handle_notify(acpi_handle handle)
{
	int num_cpus;
	uint32_t idle_cpus;

	mutex_lock(&isolated_cpus_lock);
	if (acpi_pad_pur(handle, &num_cpus)) {
		mutex_unlock(&isolated_cpus_lock);
		return;
	}
	acpi_pad_idle_cpus(num_cpus);
	idle_cpus = acpi_pad_idle_cpus_num();
	acpi_pad_ost(handle, 0, idle_cpus);
	mutex_unlock(&isolated_cpus_lock);
}

static void acpi_pad_notify(acpi_handle handle, u32 event,
	void *data)
{
	struct acpi_device *device = data;

	switch (event) {
	case ACPI_PROCESSOR_AGGREGATOR_NOTIFY:
		acpi_pad_handle_notify(handle);
		acpi_bus_generate_proc_event(device, event, 0);
		acpi_bus_generate_netlink_event(device->pnp.device_class,
			dev_name(&device->dev), event, 0);
		break;
	default:
		printk(KERN_WARNING"Unsupported event [0x%x]\n", event);
		break;
	}
}

static int acpi_pad_add(struct acpi_device *device)
{
	acpi_status status;

	strcpy(acpi_device_name(device), ACPI_PROCESSOR_AGGREGATOR_DEVICE_NAME);
	strcpy(acpi_device_class(device), ACPI_PROCESSOR_AGGREGATOR_CLASS);

	if (acpi_pad_add_sysfs(device))
		return -ENODEV;

	status = acpi_install_notify_handler(device->handle,
		ACPI_DEVICE_NOTIFY, acpi_pad_notify, device);
	if (ACPI_FAILURE(status)) {
		acpi_pad_remove_sysfs(device);
		return -ENODEV;
	}

	return 0;
}

static int acpi_pad_remove(struct acpi_device *device,
	int type)
{
	mutex_lock(&isolated_cpus_lock);
	acpi_pad_idle_cpus(0);
	mutex_unlock(&isolated_cpus_lock);

	acpi_remove_notify_handler(device->handle,
		ACPI_DEVICE_NOTIFY, acpi_pad_notify);
	acpi_pad_remove_sysfs(device);
	return 0;
}

static const struct acpi_device_id pad_device_ids[] = {
	{"ACPI000C", 0},
	{"", 0},
};
MODULE_DEVICE_TABLE(acpi, pad_device_ids);

static struct acpi_driver acpi_pad_driver = {
	.name = "processor_aggregator",
	.class = ACPI_PROCESSOR_AGGREGATOR_CLASS,
	.ids = pad_device_ids,
	.ops = {
		.add = acpi_pad_add,
		.remove = acpi_pad_remove,
	},
};

static int __init acpi_pad_init(void)
{
	power_saving_mwait_init();
	if (power_saving_mwait_eax == 0)
		return -EINVAL;

	return acpi_bus_register_driver(&acpi_pad_driver);
}

static void __exit acpi_pad_exit(void)
{
	acpi_bus_unregister_driver(&acpi_pad_driver);
}

module_init(acpi_pad_init);
module_exit(acpi_pad_exit);
MODULE_AUTHOR("Shaohua Li<shaohua.li@intel.com>");
MODULE_DESCRIPTION("ACPI Processor Aggregator Driver");
MODULE_LICENSE("GPL");
Commit	Line	Data
8e0af514 SL	1	/*
	2	* acpi_pad.c ACPI Processor Aggregator Driver
	3	*
	4	* Copyright (c) 2009, Intel Corporation.
	5	*
	6	* This program is free software; you can redistribute it and/or modify it
	7	* under the terms and conditions of the GNU General Public License,
	8	* version 2, as published by the Free Software Foundation.
	9	*
	10	* This program is distributed in the hope it will be useful, but WITHOUT
	11	* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
	12	* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
	13	* more details.
	14	*
	15	* You should have received a copy of the GNU General Public License along with
	16	* this program; if not, write to the Free Software Foundation, Inc.,
	17	* 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
	18	*
	19	*/
	20
	21	#include <linux/kernel.h>
	22	#include <linux/cpumask.h>
	23	#include <linux/module.h>
	24	#include <linux/init.h>
	25	#include <linux/types.h>
	26	#include <linux/kthread.h>
	27	#include <linux/freezer.h>
	28	#include <linux/cpu.h>
	29	#include <linux/clockchips.h>
5a0e3ad6	30	#include <linux/slab.h>
8e0af514 SL	31	#include <acpi/acpi_bus.h>
	32	#include <acpi/acpi_drivers.h>
	33
a40770a9	34	#define ACPI_PROCESSOR_AGGREGATOR_CLASS "acpi_pad"
8e0af514 SL	35	#define ACPI_PROCESSOR_AGGREGATOR_DEVICE_NAME "Processor Aggregator"
	36	#define ACPI_PROCESSOR_AGGREGATOR_NOTIFY 0x80
	37	static DEFINE_MUTEX(isolated_cpus_lock);
	38
	39	#define MWAIT_SUBSTATE_MASK (0xf)
	40	#define MWAIT_CSTATE_MASK (0xf)
	41	#define MWAIT_SUBSTATE_SIZE (4)
	42	#define CPUID_MWAIT_LEAF (5)
	43	#define CPUID5_ECX_EXTENSIONS_SUPPORTED (0x1)
	44	#define CPUID5_ECX_INTERRUPT_BREAK (0x2)
	45	static unsigned long power_saving_mwait_eax;
	46	static void power_saving_mwait_init(void)
	47	{
	48	unsigned int eax, ebx, ecx, edx;
	49	unsigned int highest_cstate = 0;
	50	unsigned int highest_subcstate = 0;
	51	int i;
	52
	53	if (!boot_cpu_has(X86_FEATURE_MWAIT))
	54	return;
	55	if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
	56	return;
	57
	58	cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &edx);
	59
	60	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) \|\|
	61	!(ecx & CPUID5_ECX_INTERRUPT_BREAK))
	62	return;
	63
	64	edx >>= MWAIT_SUBSTATE_SIZE;
	65	for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
	66	if (edx & MWAIT_SUBSTATE_MASK) {
	67	highest_cstate = i;
	68	highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
	69	}
	70	}
	71	power_saving_mwait_eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) \|
	72	(highest_subcstate - 1);
	73
	74	for_each_online_cpu(i)
	75	clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ON, &i);
	76
	77	#if defined(CONFIG_GENERIC_TIME) && defined(CONFIG_X86)
	78	switch (boot_cpu_data.x86_vendor) {
	79	case X86_VENDOR_AMD:
	80	case X86_VENDOR_INTEL:
	81	/*
	82	* AMD Fam10h TSC will tick in all
	83	* C/P/S0/S1 states when this bit is set.
	84	*/
	85	if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
	86	return;
	87
	88	/FALL THROUGH/
	89	default:
	90	/* TSC could halt in idle, so notify users */
	91	mark_tsc_unstable("TSC halts in idle");
	92	}
	93	#endif
	94	}
	95
	96	static unsigned long cpu_weight[NR_CPUS];
	97	static int tsk_in_cpu[NR_CPUS] = {[0 ... NR_CPUS-1] = -1};
	98	static DECLARE_BITMAP(pad_busy_cpus_bits, NR_CPUS);
99	static void round_robin_cpu(unsigned int tsk_index)
100	{
101	struct cpumask *pad_busy_cpus = to_cpumask(pad_busy_cpus_bits);
102	cpumask_var_t tmp;
103	int cpu;
f67538f8 AM	104	unsigned long min_weight = -1;
f67538f8 AM	105	unsigned long uninitialized_var(preferred_cpu);
8e0af514 SL	106
	107	if (!alloc_cpumask_var(&tmp, GFP_KERNEL))
	108	return;
	109
	110	mutex_lock(&isolated_cpus_lock);
	111	cpumask_clear(tmp);
	112	for_each_cpu(cpu, pad_busy_cpus)
	113	cpumask_or(tmp, tmp, topology_thread_cpumask(cpu));
	114	cpumask_andnot(tmp, cpu_online_mask, tmp);
	115	/* avoid HT sibilings if possible */
	116	if (cpumask_empty(tmp))
	117	cpumask_andnot(tmp, cpu_online_mask, pad_busy_cpus);
	118	if (cpumask_empty(tmp)) {
	119	mutex_unlock(&isolated_cpus_lock);
	120	return;
	121	}
	122	for_each_cpu(cpu, tmp) {
	123	if (cpu_weight[cpu] < min_weight) {
	124	min_weight = cpu_weight[cpu];
	125	preferred_cpu = cpu;
	126	}
	127	}
	128
	129	if (tsk_in_cpu[tsk_index] != -1)
	130	cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus);
	131	tsk_in_cpu[tsk_index] = preferred_cpu;
	132	cpumask_set_cpu(preferred_cpu, pad_busy_cpus);
	133	cpu_weight[preferred_cpu]++;
	134	mutex_unlock(&isolated_cpus_lock);
	135
	136	set_cpus_allowed_ptr(current, cpumask_of(preferred_cpu));
	137	}
	138
	139	static void exit_round_robin(unsigned int tsk_index)
	140	{
	141	struct cpumask *pad_busy_cpus = to_cpumask(pad_busy_cpus_bits);
	142	cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus);
	143	tsk_in_cpu[tsk_index] = -1;
	144	}
	145
	146	static unsigned int idle_pct = 5; /* percentage */
	147	static unsigned int round_robin_time = 10; /* second */
	148	static int power_saving_thread(void *data)
	149	{
	150	struct sched_param param = {.sched_priority = 1};
	151	int do_sleep;
	152	unsigned int tsk_index = (unsigned long)data;
	153	u64 last_jiffies = 0;
	154
	155	sched_setscheduler(current, SCHED_RR, &param);
	156
	157	while (!kthread_should_stop()) {
	158	int cpu;
	159	u64 expire_time;
	160
	161	try_to_freeze();
	162
	163	/* round robin to cpus */
	164	if (last_jiffies + round_robin_time * HZ < jiffies) {
	165	last_jiffies = jiffies;
	166	round_robin_cpu(tsk_index);
	167	}
	168
	169	do_sleep = 0;
170
8e0af514 SL	171	expire_time = jiffies + HZ * (100 - idle_pct) / 100;
	172
	173	while (!need_resched()) {
	174	local_irq_disable();
	175	cpu = smp_processor_id();
	176	clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER,
	177	&cpu);
	178	stop_critical_timings();
	179
	180	__monitor((void *)&current_thread_info()->flags, 0, 0);
	181	smp_mb();
	182	if (!need_resched())
	183	__mwait(power_saving_mwait_eax, 1);
	184
	185	start_critical_timings();
	186	clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT,
	187	&cpu);
	188	local_irq_enable();
	189
	190	if (jiffies > expire_time) {
	191	do_sleep = 1;
	192	break;
	193	}
	194	}
	195
8e0af514 SL	196	/*
	197	* current sched_rt has threshold for rt task running time.
	198	* When a rt task uses 95% CPU time, the rt thread will be
	199	* scheduled out for 5% CPU time to not starve other tasks. But
	200	* the mechanism only works when all CPUs have RT task running,
	201	* as if one CPU hasn't RT task, RT task from other CPUs will
	202	* borrow CPU time from this CPU and cause RT task use > 95%
3b8cb427	203	* CPU time. To make 'avoid starvation' work, takes a nap here.
8e0af514 SL	204	*/
	205	if (do_sleep)
	206	schedule_timeout_killable(HZ * idle_pct / 100);
	207	}
	208
	209	exit_round_robin(tsk_index);
	210	return 0;
	211	}
	212
	213	static struct task_struct *ps_tsks[NR_CPUS];
	214	static unsigned int ps_tsk_num;
	215	static int create_power_saving_task(void)
	216	{
3b8cb427 CG	217	int rc = -ENOMEM;
3b8cb427 CG	218
8e0af514 SL	219	ps_tsks[ps_tsk_num] = kthread_run(power_saving_thread,
	220	(void *)(unsigned long)ps_tsk_num,
	221	"power_saving/%d", ps_tsk_num);
3b8cb427 CG	222	rc = IS_ERR(ps_tsks[ps_tsk_num]) ? PTR_ERR(ps_tsks[ps_tsk_num]) : 0;
3b8cb427 CG	223	if (!rc)
8e0af514	224	ps_tsk_num++;
3b8cb427 CG	225	else
	226	ps_tsks[ps_tsk_num] = NULL;
	227
	228	return rc;
8e0af514 SL	229	}
	230
	231	static void destroy_power_saving_task(void)
	232	{
	233	if (ps_tsk_num > 0) {
	234	ps_tsk_num--;
	235	kthread_stop(ps_tsks[ps_tsk_num]);
3b8cb427	236	ps_tsks[ps_tsk_num] = NULL;
8e0af514 SL	237	}
	238	}
	239
	240	static void set_power_saving_task_num(unsigned int num)
	241	{
	242	if (num > ps_tsk_num) {
	243	while (ps_tsk_num < num) {
	244	if (create_power_saving_task())
	245	return;
	246	}
	247	} else if (num < ps_tsk_num) {
	248	while (ps_tsk_num > num)
	249	destroy_power_saving_task();
	250	}
	251	}
	252
3b8cb427	253	static void acpi_pad_idle_cpus(unsigned int num_cpus)
8e0af514 SL	254	{
	255	get_online_cpus();
	256
	257	num_cpus = min_t(unsigned int, num_cpus, num_online_cpus());
	258	set_power_saving_task_num(num_cpus);
	259
	260	put_online_cpus();
8e0af514 SL	261	}
	262
	263	static uint32_t acpi_pad_idle_cpus_num(void)
	264	{
	265	return ps_tsk_num;
	266	}
	267
	268	static ssize_t acpi_pad_rrtime_store(struct device *dev,
	269	struct device_attribute attr, const char buf, size_t count)
	270	{
	271	unsigned long num;
	272	if (strict_strtoul(buf, 0, &num))
	273	return -EINVAL;
	274	if (num < 1 \|\| num >= 100)
	275	return -EINVAL;
	276	mutex_lock(&isolated_cpus_lock);
	277	round_robin_time = num;
	278	mutex_unlock(&isolated_cpus_lock);
	279	return count;
	280	}
	281
	282	static ssize_t acpi_pad_rrtime_show(struct device *dev,
	283	struct device_attribute attr, char buf)
	284	{
	285	return scnprintf(buf, PAGE_SIZE, "%d", round_robin_time);
	286	}
	287	static DEVICE_ATTR(rrtime, S_IRUGO\|S_IWUSR,
	288	acpi_pad_rrtime_show,
	289	acpi_pad_rrtime_store);
	290
	291	static ssize_t acpi_pad_idlepct_store(struct device *dev,
	292	struct device_attribute attr, const char buf, size_t count)
	293	{
	294	unsigned long num;
	295	if (strict_strtoul(buf, 0, &num))
	296	return -EINVAL;
	297	if (num < 1 \|\| num >= 100)
	298	return -EINVAL;
	299	mutex_lock(&isolated_cpus_lock);
	300	idle_pct = num;
	301	mutex_unlock(&isolated_cpus_lock);
	302	return count;
	303	}
	304
	305	static ssize_t acpi_pad_idlepct_show(struct device *dev,
	306	struct device_attribute attr, char buf)
	307	{
	308	return scnprintf(buf, PAGE_SIZE, "%d", idle_pct);
	309	}
	310	static DEVICE_ATTR(idlepct, S_IRUGO\|S_IWUSR,
	311	acpi_pad_idlepct_show,
	312	acpi_pad_idlepct_store);
	313
	314	static ssize_t acpi_pad_idlecpus_store(struct device *dev,
	315	struct device_attribute attr, const char buf, size_t count)
	316	{
	317	unsigned long num;
	318	if (strict_strtoul(buf, 0, &num))
	319	return -EINVAL;
	320	mutex_lock(&isolated_cpus_lock);
	321	acpi_pad_idle_cpus(num);
	322	mutex_unlock(&isolated_cpus_lock);
	323	return count;
	324	}
325
326	static ssize_t acpi_pad_idlecpus_show(struct device *dev,
327	struct device_attribute attr, char buf)
328	{
329	return cpumask_scnprintf(buf, PAGE_SIZE,
330	to_cpumask(pad_busy_cpus_bits));
331	}
332	static DEVICE_ATTR(idlecpus, S_IRUGO\|S_IWUSR,
333	acpi_pad_idlecpus_show,
334	acpi_pad_idlecpus_store);
335
336	static int acpi_pad_add_sysfs(struct acpi_device *device)
337	{
338	int result;
339
340	result = device_create_file(&device->dev, &dev_attr_idlecpus);
341	if (result)
342	return -ENODEV;
343	result = device_create_file(&device->dev, &dev_attr_idlepct);
344	if (result) {
345	device_remove_file(&device->dev, &dev_attr_idlecpus);
346	return -ENODEV;
347	}
348	result = device_create_file(&device->dev, &dev_attr_rrtime);
349	if (result) {
350	device_remove_file(&device->dev, &dev_attr_idlecpus);
351	device_remove_file(&device->dev, &dev_attr_idlepct);
352	return -ENODEV;
353	}
354	return 0;
355	}
356
357	static void acpi_pad_remove_sysfs(struct acpi_device *device)
358	{
359	device_remove_file(&device->dev, &dev_attr_idlecpus);
360	device_remove_file(&device->dev, &dev_attr_idlepct);
361	device_remove_file(&device->dev, &dev_attr_rrtime);
362	}
363
364	/* Query firmware how many CPUs should be idle */
365	static int acpi_pad_pur(acpi_handle handle, int *num_cpus)
366	{
367	struct acpi_buffer buffer = {ACPI_ALLOCATE_BUFFER, NULL};
8e0af514 SL	368	union acpi_object *package;
	369	int rev, num, ret = -EINVAL;
	370
3b8cb427	371	if (ACPI_FAILURE(acpi_evaluate_object(handle, "_PUR", NULL, &buffer)))
8e0af514	372	return -EINVAL;
3b8cb427 CG	373
	374	if (!buffer.length \|\| !buffer.pointer)
	375	return -EINVAL;
	376
8e0af514 SL	377	package = buffer.pointer;
	378	if (package->type != ACPI_TYPE_PACKAGE \|\| package->package.count != 2)
	379	goto out;
	380	rev = package->package.elements[0].integer.value;
	381	num = package->package.elements[1].integer.value;
3b8cb427	382	if (rev != 1 \|\| num < 0)
8e0af514 SL	383	goto out;
	384	*num_cpus = num;
	385	ret = 0;
	386	out:
	387	kfree(buffer.pointer);
	388	return ret;
	389	}
	390
	391	/* Notify firmware how many CPUs are idle */
	392	static void acpi_pad_ost(acpi_handle handle, int stat,
	393	uint32_t idle_cpus)
	394	{
	395	union acpi_object params[3] = {
	396	{.type = ACPI_TYPE_INTEGER,},
	397	{.type = ACPI_TYPE_INTEGER,},
	398	{.type = ACPI_TYPE_BUFFER,},
	399	};
	400	struct acpi_object_list arg_list = {3, params};
	401
	402	params[0].integer.value = ACPI_PROCESSOR_AGGREGATOR_NOTIFY;
	403	params[1].integer.value = stat;
	404	params[2].buffer.length = 4;
	405	params[2].buffer.pointer = (void *)&idle_cpus;
	406	acpi_evaluate_object(handle, "_OST", &arg_list, NULL);
	407	}
	408
	409	static void acpi_pad_handle_notify(acpi_handle handle)
	410	{
3b8cb427	411	int num_cpus;
8e0af514 SL	412	uint32_t idle_cpus;
	413
	414	mutex_lock(&isolated_cpus_lock);
	415	if (acpi_pad_pur(handle, &num_cpus)) {
	416	mutex_unlock(&isolated_cpus_lock);
	417	return;
	418	}
3b8cb427	419	acpi_pad_idle_cpus(num_cpus);
8e0af514	420	idle_cpus = acpi_pad_idle_cpus_num();
3b8cb427	421	acpi_pad_ost(handle, 0, idle_cpus);
8e0af514 SL	422	mutex_unlock(&isolated_cpus_lock);
	423	}
	424
	425	static void acpi_pad_notify(acpi_handle handle, u32 event,
	426	void *data)
	427	{
	428	struct acpi_device *device = data;
	429
	430	switch (event) {
	431	case ACPI_PROCESSOR_AGGREGATOR_NOTIFY:
	432	acpi_pad_handle_notify(handle);
	433	acpi_bus_generate_proc_event(device, event, 0);
	434	acpi_bus_generate_netlink_event(device->pnp.device_class,
	435	dev_name(&device->dev), event, 0);
	436	break;
	437	default:
	438	printk(KERN_WARNING"Unsupported event [0x%x]\n", event);
	439	break;
	440	}
	441	}
	442
	443	static int acpi_pad_add(struct acpi_device *device)
	444	{
	445	acpi_status status;
	446
	447	strcpy(acpi_device_name(device), ACPI_PROCESSOR_AGGREGATOR_DEVICE_NAME);
	448	strcpy(acpi_device_class(device), ACPI_PROCESSOR_AGGREGATOR_CLASS);
	449
	450	if (acpi_pad_add_sysfs(device))
	451	return -ENODEV;
	452
	453	status = acpi_install_notify_handler(device->handle,
	454	ACPI_DEVICE_NOTIFY, acpi_pad_notify, device);
	455	if (ACPI_FAILURE(status)) {
	456	acpi_pad_remove_sysfs(device);
	457	return -ENODEV;
	458	}
	459
	460	return 0;
	461	}
	462
	463	static int acpi_pad_remove(struct acpi_device *device,
	464	int type)
	465	{
	466	mutex_lock(&isolated_cpus_lock);
	467	acpi_pad_idle_cpus(0);
	468	mutex_unlock(&isolated_cpus_lock);
	469
	470	acpi_remove_notify_handler(device->handle,
	471	ACPI_DEVICE_NOTIFY, acpi_pad_notify);
	472	acpi_pad_remove_sysfs(device);
	473	return 0;
	474	}
	475
	476	static const struct acpi_device_id pad_device_ids[] = {
	477	{"ACPI000C", 0},
	478	{"", 0},
	479	};
	480	MODULE_DEVICE_TABLE(acpi, pad_device_ids);
	481
	482	static struct acpi_driver acpi_pad_driver = {
	483	.name = "processor_aggregator",
	484	.class = ACPI_PROCESSOR_AGGREGATOR_CLASS,
	485	.ids = pad_device_ids,
486	.ops = {
487	.add = acpi_pad_add,
488	.remove = acpi_pad_remove,
489	},
490	};
491
492	static int __init acpi_pad_init(void)
493	{
494	power_saving_mwait_init();
495	if (power_saving_mwait_eax == 0)
496	return -EINVAL;
497
498	return acpi_bus_register_driver(&acpi_pad_driver);
499	}
500
501	static void __exit acpi_pad_exit(void)
502	{
503	acpi_bus_unregister_driver(&acpi_pad_driver);
504	}
505
506	module_init(acpi_pad_init);
507	module_exit(acpi_pad_exit);
508	MODULE_AUTHOR("Shaohua Li<shaohua.li@intel.com>");
509	MODULE_DESCRIPTION("ACPI Processor Aggregator Driver");
510	MODULE_LICENSE("GPL");