[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 unsigned char tsc_detected_unstable;
static unsigned char tsc_marked_unstable;
static unsigned char lapic_detected_unstable;
static unsigned char lapic_marked_unstable;

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);

#if 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))
			tsc_detected_unstable = 1;
		if (!boot_cpu_has(X86_FEATURE_ARAT))
			lapic_detected_unstable = 1;
		break;
	default:
		/* TSC & LAPIC could halt in idle */
		tsc_detected_unstable = 1;
		lapic_detected_unstable = 1;
	}
#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()) {
			if (tsc_detected_unstable && !tsc_marked_unstable) {
				/* TSC could halt in idle, so notify users */
				mark_tsc_unstable("TSC halts in idle");
				tsc_marked_unstable = 1;
			}
			if (lapic_detected_unstable && !lapic_marked_unstable) {
				int i;
				/* LAPIC could halt in idle, so notify users */
				for_each_online_cpu(i)
					clockevents_notify(
						CLOCK_EVT_NOTIFY_BROADCAST_ON,
						&i);
				lapic_marked_unstable = 1;
			}
			local_irq_disable();
			cpu = smp_processor_id();
			if (lapic_marked_unstable)
				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();
			if (lapic_marked_unstable)
				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
 * return -1 on failure
 */
static int acpi_pad_pur(acpi_handle handle)
{
	struct acpi_buffer buffer = {ACPI_ALLOCATE_BUFFER, NULL};
	union acpi_object *package;
	int num = -1;

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

	if (!buffer.length || !buffer.pointer)
		return num;

	package = buffer.pointer;

	if (package->type == ACPI_TYPE_PACKAGE &&
		package->package.count == 2 &&
		package->package.elements[0].integer.value == 1) /* rev 1 */

		num = package->package.elements[1].integer.value;

	kfree(buffer.pointer);
	return num;
}

/* 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);
	num_cpus = acpi_pad_pur(handle);
	if (num_cpus < 0) {
		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;
0dc698b9 VP	46
	47	static unsigned char tsc_detected_unstable;
	48	static unsigned char tsc_marked_unstable;
8aa4b14e CG	49	static unsigned char lapic_detected_unstable;
8aa4b14e CG	50	static unsigned char lapic_marked_unstable;
0dc698b9	51
8e0af514 SL	52	static void power_saving_mwait_init(void)
	53	{
	54	unsigned int eax, ebx, ecx, edx;
	55	unsigned int highest_cstate = 0;
	56	unsigned int highest_subcstate = 0;
	57	int i;
	58
	59	if (!boot_cpu_has(X86_FEATURE_MWAIT))
	60	return;
	61	if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
	62	return;
	63
	64	cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &edx);
	65
	66	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) \|\|
	67	!(ecx & CPUID5_ECX_INTERRUPT_BREAK))
	68	return;
	69
	70	edx >>= MWAIT_SUBSTATE_SIZE;
	71	for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
	72	if (edx & MWAIT_SUBSTATE_MASK) {
	73	highest_cstate = i;
	74	highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
	75	}
	76	}
	77	power_saving_mwait_eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) \|
	78	(highest_subcstate - 1);
	79
592913ec	80	#if defined(CONFIG_X86)
8e0af514 SL	81	switch (boot_cpu_data.x86_vendor) {
	82	case X86_VENDOR_AMD:
	83	case X86_VENDOR_INTEL:
	84	/*
	85	* AMD Fam10h TSC will tick in all
	86	* C/P/S0/S1 states when this bit is set.
	87	*/
8aa4b14e CG	88	if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
	89	tsc_detected_unstable = 1;
	90	if (!boot_cpu_has(X86_FEATURE_ARAT))
	91	lapic_detected_unstable = 1;
	92	break;
8e0af514	93	default:
8aa4b14e	94	/* TSC & LAPIC could halt in idle */
0dc698b9	95	tsc_detected_unstable = 1;
8aa4b14e	96	lapic_detected_unstable = 1;
8e0af514 SL	97	}
	98	#endif
	99	}
	100
	101	static unsigned long cpu_weight[NR_CPUS];
	102	static int tsk_in_cpu[NR_CPUS] = {[0 ... NR_CPUS-1] = -1};
	103	static DECLARE_BITMAP(pad_busy_cpus_bits, NR_CPUS);
	104	static void round_robin_cpu(unsigned int tsk_index)
	105	{
	106	struct cpumask *pad_busy_cpus = to_cpumask(pad_busy_cpus_bits);
	107	cpumask_var_t tmp;
	108	int cpu;
f67538f8 AM	109	unsigned long min_weight = -1;
f67538f8 AM	110	unsigned long uninitialized_var(preferred_cpu);
8e0af514 SL	111
	112	if (!alloc_cpumask_var(&tmp, GFP_KERNEL))
	113	return;
	114
	115	mutex_lock(&isolated_cpus_lock);
	116	cpumask_clear(tmp);
	117	for_each_cpu(cpu, pad_busy_cpus)
	118	cpumask_or(tmp, tmp, topology_thread_cpumask(cpu));
	119	cpumask_andnot(tmp, cpu_online_mask, tmp);
	120	/* avoid HT sibilings if possible */
	121	if (cpumask_empty(tmp))
	122	cpumask_andnot(tmp, cpu_online_mask, pad_busy_cpus);
	123	if (cpumask_empty(tmp)) {
	124	mutex_unlock(&isolated_cpus_lock);
	125	return;
	126	}
	127	for_each_cpu(cpu, tmp) {
	128	if (cpu_weight[cpu] < min_weight) {
	129	min_weight = cpu_weight[cpu];
	130	preferred_cpu = cpu;
	131	}
	132	}
	133
	134	if (tsk_in_cpu[tsk_index] != -1)
	135	cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus);
	136	tsk_in_cpu[tsk_index] = preferred_cpu;
	137	cpumask_set_cpu(preferred_cpu, pad_busy_cpus);
	138	cpu_weight[preferred_cpu]++;
	139	mutex_unlock(&isolated_cpus_lock);
	140
	141	set_cpus_allowed_ptr(current, cpumask_of(preferred_cpu));
	142	}
	143
	144	static void exit_round_robin(unsigned int tsk_index)
	145	{
	146	struct cpumask *pad_busy_cpus = to_cpumask(pad_busy_cpus_bits);
	147	cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus);
	148	tsk_in_cpu[tsk_index] = -1;
	149	}
	150
	151	static unsigned int idle_pct = 5; /* percentage */
	152	static unsigned int round_robin_time = 10; /* second */
	153	static int power_saving_thread(void *data)
	154	{
	155	struct sched_param param = {.sched_priority = 1};
	156	int do_sleep;
	157	unsigned int tsk_index = (unsigned long)data;
	158	u64 last_jiffies = 0;
	159
	160	sched_setscheduler(current, SCHED_RR, &param);
	161
	162	while (!kthread_should_stop()) {
	163	int cpu;
	164	u64 expire_time;
	165
	166	try_to_freeze();
	167
	168	/* round robin to cpus */
	169	if (last_jiffies + round_robin_time * HZ < jiffies) {
	170	last_jiffies = jiffies;
	171	round_robin_cpu(tsk_index);
	172	}
	173
	174	do_sleep = 0;
175
8e0af514 SL	176	expire_time = jiffies + HZ * (100 - idle_pct) / 100;
	177
	178	while (!need_resched()) {
0dc698b9 VP	179	if (tsc_detected_unstable && !tsc_marked_unstable) {
	180	/* TSC could halt in idle, so notify users */
	181	mark_tsc_unstable("TSC halts in idle");
	182	tsc_marked_unstable = 1;
	183	}
8aa4b14e CG	184	if (lapic_detected_unstable && !lapic_marked_unstable) {
	185	int i;
	186	/* LAPIC could halt in idle, so notify users */
	187	for_each_online_cpu(i)
	188	clockevents_notify(
	189	CLOCK_EVT_NOTIFY_BROADCAST_ON,
	190	&i);
	191	lapic_marked_unstable = 1;
	192	}
8e0af514 SL	193	local_irq_disable();
8e0af514 SL	194	cpu = smp_processor_id();
8aa4b14e CG	195	if (lapic_marked_unstable)
	196	clockevents_notify(
	197	CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
8e0af514 SL	198	stop_critical_timings();
	199
	200	__monitor((void *)&current_thread_info()->flags, 0, 0);
	201	smp_mb();
	202	if (!need_resched())
	203	__mwait(power_saving_mwait_eax, 1);
	204
	205	start_critical_timings();
8aa4b14e CG	206	if (lapic_marked_unstable)
	207	clockevents_notify(
	208	CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
8e0af514 SL	209	local_irq_enable();
	210
	211	if (jiffies > expire_time) {
	212	do_sleep = 1;
	213	break;
	214	}
	215	}
	216
8e0af514 SL	217	/*
	218	* current sched_rt has threshold for rt task running time.
	219	* When a rt task uses 95% CPU time, the rt thread will be
	220	* scheduled out for 5% CPU time to not starve other tasks. But
	221	* the mechanism only works when all CPUs have RT task running,
	222	* as if one CPU hasn't RT task, RT task from other CPUs will
	223	* borrow CPU time from this CPU and cause RT task use > 95%
3b8cb427	224	* CPU time. To make 'avoid starvation' work, takes a nap here.
8e0af514 SL	225	*/
	226	if (do_sleep)
	227	schedule_timeout_killable(HZ * idle_pct / 100);
	228	}
	229
	230	exit_round_robin(tsk_index);
	231	return 0;
	232	}
	233
	234	static struct task_struct *ps_tsks[NR_CPUS];
	235	static unsigned int ps_tsk_num;
	236	static int create_power_saving_task(void)
	237	{
3b8cb427 CG	238	int rc = -ENOMEM;
3b8cb427 CG	239
8e0af514 SL	240	ps_tsks[ps_tsk_num] = kthread_run(power_saving_thread,
	241	(void *)(unsigned long)ps_tsk_num,
	242	"power_saving/%d", ps_tsk_num);
3b8cb427 CG	243	rc = IS_ERR(ps_tsks[ps_tsk_num]) ? PTR_ERR(ps_tsks[ps_tsk_num]) : 0;
3b8cb427 CG	244	if (!rc)
8e0af514	245	ps_tsk_num++;
3b8cb427 CG	246	else
	247	ps_tsks[ps_tsk_num] = NULL;
	248
	249	return rc;
8e0af514 SL	250	}
	251
	252	static void destroy_power_saving_task(void)
	253	{
	254	if (ps_tsk_num > 0) {
	255	ps_tsk_num--;
	256	kthread_stop(ps_tsks[ps_tsk_num]);
3b8cb427	257	ps_tsks[ps_tsk_num] = NULL;
8e0af514 SL	258	}
	259	}
	260
	261	static void set_power_saving_task_num(unsigned int num)
	262	{
	263	if (num > ps_tsk_num) {
	264	while (ps_tsk_num < num) {
	265	if (create_power_saving_task())
	266	return;
	267	}
	268	} else if (num < ps_tsk_num) {
	269	while (ps_tsk_num > num)
	270	destroy_power_saving_task();
	271	}
	272	}
	273
3b8cb427	274	static void acpi_pad_idle_cpus(unsigned int num_cpus)
8e0af514 SL	275	{
	276	get_online_cpus();
	277
	278	num_cpus = min_t(unsigned int, num_cpus, num_online_cpus());
	279	set_power_saving_task_num(num_cpus);
	280
	281	put_online_cpus();
8e0af514 SL	282	}
	283
	284	static uint32_t acpi_pad_idle_cpus_num(void)
	285	{
	286	return ps_tsk_num;
	287	}
	288
	289	static ssize_t acpi_pad_rrtime_store(struct device *dev,
	290	struct device_attribute attr, const char buf, size_t count)
	291	{
	292	unsigned long num;
	293	if (strict_strtoul(buf, 0, &num))
	294	return -EINVAL;
	295	if (num < 1 \|\| num >= 100)
	296	return -EINVAL;
	297	mutex_lock(&isolated_cpus_lock);
	298	round_robin_time = num;
	299	mutex_unlock(&isolated_cpus_lock);
	300	return count;
	301	}
	302
	303	static ssize_t acpi_pad_rrtime_show(struct device *dev,
	304	struct device_attribute attr, char buf)
	305	{
	306	return scnprintf(buf, PAGE_SIZE, "%d", round_robin_time);
	307	}
	308	static DEVICE_ATTR(rrtime, S_IRUGO\|S_IWUSR,
	309	acpi_pad_rrtime_show,
	310	acpi_pad_rrtime_store);
	311
	312	static ssize_t acpi_pad_idlepct_store(struct device *dev,
	313	struct device_attribute attr, const char buf, size_t count)
	314	{
	315	unsigned long num;
	316	if (strict_strtoul(buf, 0, &num))
	317	return -EINVAL;
	318	if (num < 1 \|\| num >= 100)
	319	return -EINVAL;
	320	mutex_lock(&isolated_cpus_lock);
	321	idle_pct = num;
	322	mutex_unlock(&isolated_cpus_lock);
	323	return count;
	324	}
	325
	326	static ssize_t acpi_pad_idlepct_show(struct device *dev,
	327	struct device_attribute attr, char buf)
	328	{
	329	return scnprintf(buf, PAGE_SIZE, "%d", idle_pct);
	330	}
	331	static DEVICE_ATTR(idlepct, S_IRUGO\|S_IWUSR,
	332	acpi_pad_idlepct_show,
	333	acpi_pad_idlepct_store);
	334
	335	static ssize_t acpi_pad_idlecpus_store(struct device *dev,
	336	struct device_attribute attr, const char buf, size_t count)
	337	{
	338	unsigned long num;
	339	if (strict_strtoul(buf, 0, &num))
	340	return -EINVAL;
	341	mutex_lock(&isolated_cpus_lock);
	342	acpi_pad_idle_cpus(num);
	343	mutex_unlock(&isolated_cpus_lock);
	344	return count;
	345	}
346
347	static ssize_t acpi_pad_idlecpus_show(struct device *dev,
348	struct device_attribute attr, char buf)
349	{
350	return cpumask_scnprintf(buf, PAGE_SIZE,
351	to_cpumask(pad_busy_cpus_bits));
352	}
353	static DEVICE_ATTR(idlecpus, S_IRUGO\|S_IWUSR,
354	acpi_pad_idlecpus_show,
355	acpi_pad_idlecpus_store);
356
357	static int acpi_pad_add_sysfs(struct acpi_device *device)
358	{
359	int result;
360
361	result = device_create_file(&device->dev, &dev_attr_idlecpus);
362	if (result)
363	return -ENODEV;
364	result = device_create_file(&device->dev, &dev_attr_idlepct);
365	if (result) {
366	device_remove_file(&device->dev, &dev_attr_idlecpus);
367	return -ENODEV;
368	}
369	result = device_create_file(&device->dev, &dev_attr_rrtime);
370	if (result) {
371	device_remove_file(&device->dev, &dev_attr_idlecpus);
372	device_remove_file(&device->dev, &dev_attr_idlepct);
373	return -ENODEV;
374	}
375	return 0;
376	}
377
378	static void acpi_pad_remove_sysfs(struct acpi_device *device)
379	{
380	device_remove_file(&device->dev, &dev_attr_idlecpus);
381	device_remove_file(&device->dev, &dev_attr_idlepct);
382	device_remove_file(&device->dev, &dev_attr_rrtime);
383	}
384
c9ad8e06 LB	385	/*
	386	* Query firmware how many CPUs should be idle
	387	* return -1 on failure
	388	*/
	389	static int acpi_pad_pur(acpi_handle handle)
8e0af514 SL	390	{
8e0af514 SL	391	struct acpi_buffer buffer = {ACPI_ALLOCATE_BUFFER, NULL};
8e0af514	392	union acpi_object *package;
c9ad8e06	393	int num = -1;
8e0af514	394
3b8cb427	395	if (ACPI_FAILURE(acpi_evaluate_object(handle, "_PUR", NULL, &buffer)))
c9ad8e06	396	return num;
3b8cb427 CG	397
3b8cb427 CG	398	if (!buffer.length \|\| !buffer.pointer)
c9ad8e06	399	return num;
3b8cb427	400
8e0af514	401	package = buffer.pointer;
c9ad8e06 LB	402
	403	if (package->type == ACPI_TYPE_PACKAGE &&
	404	package->package.count == 2 &&
	405	package->package.elements[0].integer.value == 1) /* rev 1 */
	406
	407	num = package->package.elements[1].integer.value;
	408
8e0af514	409	kfree(buffer.pointer);
c9ad8e06	410	return num;
8e0af514 SL	411	}
	412
	413	/* Notify firmware how many CPUs are idle */
	414	static void acpi_pad_ost(acpi_handle handle, int stat,
	415	uint32_t idle_cpus)
	416	{
	417	union acpi_object params[3] = {
	418	{.type = ACPI_TYPE_INTEGER,},
	419	{.type = ACPI_TYPE_INTEGER,},
	420	{.type = ACPI_TYPE_BUFFER,},
	421	};
	422	struct acpi_object_list arg_list = {3, params};
	423
	424	params[0].integer.value = ACPI_PROCESSOR_AGGREGATOR_NOTIFY;
	425	params[1].integer.value = stat;
	426	params[2].buffer.length = 4;
	427	params[2].buffer.pointer = (void *)&idle_cpus;
	428	acpi_evaluate_object(handle, "_OST", &arg_list, NULL);
	429	}
	430
	431	static void acpi_pad_handle_notify(acpi_handle handle)
	432	{
3b8cb427	433	int num_cpus;
8e0af514 SL	434	uint32_t idle_cpus;
	435
	436	mutex_lock(&isolated_cpus_lock);
c9ad8e06 LB	437	num_cpus = acpi_pad_pur(handle);
c9ad8e06 LB	438	if (num_cpus < 0) {
8e0af514 SL	439	mutex_unlock(&isolated_cpus_lock);
	440	return;
	441	}
3b8cb427	442	acpi_pad_idle_cpus(num_cpus);
8e0af514	443	idle_cpus = acpi_pad_idle_cpus_num();
3b8cb427	444	acpi_pad_ost(handle, 0, idle_cpus);
8e0af514 SL	445	mutex_unlock(&isolated_cpus_lock);
	446	}
	447
	448	static void acpi_pad_notify(acpi_handle handle, u32 event,
	449	void *data)
	450	{
	451	struct acpi_device *device = data;
	452
	453	switch (event) {
	454	case ACPI_PROCESSOR_AGGREGATOR_NOTIFY:
	455	acpi_pad_handle_notify(handle);
	456	acpi_bus_generate_proc_event(device, event, 0);
	457	acpi_bus_generate_netlink_event(device->pnp.device_class,
	458	dev_name(&device->dev), event, 0);
	459	break;
	460	default:
	461	printk(KERN_WARNING"Unsupported event [0x%x]\n", event);
	462	break;
	463	}
	464	}
	465
	466	static int acpi_pad_add(struct acpi_device *device)
	467	{
	468	acpi_status status;
	469
	470	strcpy(acpi_device_name(device), ACPI_PROCESSOR_AGGREGATOR_DEVICE_NAME);
	471	strcpy(acpi_device_class(device), ACPI_PROCESSOR_AGGREGATOR_CLASS);
	472
	473	if (acpi_pad_add_sysfs(device))
	474	return -ENODEV;
	475
	476	status = acpi_install_notify_handler(device->handle,
	477	ACPI_DEVICE_NOTIFY, acpi_pad_notify, device);
	478	if (ACPI_FAILURE(status)) {
	479	acpi_pad_remove_sysfs(device);
	480	return -ENODEV;
	481	}
	482
	483	return 0;
	484	}
	485
	486	static int acpi_pad_remove(struct acpi_device *device,
	487	int type)
	488	{
	489	mutex_lock(&isolated_cpus_lock);
	490	acpi_pad_idle_cpus(0);
	491	mutex_unlock(&isolated_cpus_lock);
	492
	493	acpi_remove_notify_handler(device->handle,
	494	ACPI_DEVICE_NOTIFY, acpi_pad_notify);
	495	acpi_pad_remove_sysfs(device);
	496	return 0;
	497	}
	498
	499	static const struct acpi_device_id pad_device_ids[] = {
	500	{"ACPI000C", 0},
	501	{"", 0},
	502	};
	503	MODULE_DEVICE_TABLE(acpi, pad_device_ids);
	504
	505	static struct acpi_driver acpi_pad_driver = {
	506	.name = "processor_aggregator",
	507	.class = ACPI_PROCESSOR_AGGREGATOR_CLASS,
	508	.ids = pad_device_ids,
509	.ops = {
510	.add = acpi_pad_add,
511	.remove = acpi_pad_remove,
512	},
513	};
514
515	static int __init acpi_pad_init(void)
516	{
517	power_saving_mwait_init();
518	if (power_saving_mwait_eax == 0)
519	return -EINVAL;
520
521	return acpi_bus_register_driver(&acpi_pad_driver);
522	}
523
524	static void __exit acpi_pad_exit(void)
525	{
526	acpi_bus_unregister_driver(&acpi_pad_driver);
527	}
528
529	module_init(acpi_pad_init);
530	module_exit(acpi_pad_exit);
531	MODULE_AUTHOR("Shaohua Li<shaohua.li@intel.com>");
532	MODULE_DESCRIPTION("ACPI Processor Aggregator Driver");
533	MODULE_LICENSE("GPL");