[linux-block.git] / drivers / acpi / acpi_pad.c

// SPDX-License-Identifier: GPL-2.0-only
/*
 * acpi_pad.c ACPI Processor Aggregator Driver
 *
 * Copyright (c) 2009, Intel Corporation.
 */

#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 <uapi/linux/sched/types.h>
#include <linux/freezer.h>
#include <linux/cpu.h>
#include <linux/tick.h>
#include <linux/slab.h>
#include <linux/acpi.h>
#include <linux/perf_event.h>
#include <asm/mwait.h>
#include <xen/xen.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);
static DEFINE_MUTEX(round_robin_lock);

static unsigned long power_saving_mwait_eax;

static unsigned char tsc_detected_unstable;
static unsigned char tsc_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_HYGON:
	case X86_VENDOR_AMD:
	case X86_VENDOR_INTEL:
	case X86_VENDOR_ZHAOXIN:
		/*
		 * 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;
		break;
	default:
		/* TSC could halt in idle */
		tsc_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 preferred_cpu;

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

	mutex_lock(&round_robin_lock);
	cpumask_clear(tmp);
	for_each_cpu(cpu, pad_busy_cpus)
		cpumask_or(tmp, tmp, topology_sibling_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(&round_robin_lock);
		free_cpumask_var(tmp);
		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(&round_robin_lock);

	set_cpus_allowed_ptr(current, cpumask_of(preferred_cpu));

	free_cpumask_var(tmp);
}

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 = 1; /* second */
static int power_saving_thread(void *data)
{
	int do_sleep;
	unsigned int tsk_index = (unsigned long)data;
	u64 last_jiffies = 0;

	sched_set_fifo_low(current);

	while (!kthread_should_stop()) {
		unsigned long expire_time;

		/* round robin to cpus */
		expire_time = last_jiffies + round_robin_time * HZ;
		if (time_before(expire_time, 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;
			}
			local_irq_disable();

			perf_lopwr_cb(true);

			tick_broadcast_enable();
			tick_broadcast_enter();
			stop_critical_timings();

			mwait_idle_with_hints(power_saving_mwait_eax, 1);

			start_critical_timings();
			tick_broadcast_exit();

			perf_lopwr_cb(false);

			local_irq_enable();

			if (time_before(expire_time, jiffies)) {
				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 (unlikely(do_sleep))
			schedule_timeout_killable(HZ * idle_pct / 100);

		/* If an external event has set the need_resched flag, then
		 * we need to deal with it, or this loop will continue to
		 * spin without calling __mwait().
		 */
		if (unlikely(need_resched()))
			schedule();
	}

	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;

	ps_tsks[ps_tsk_num] = kthread_run(power_saving_thread,
		(void *)(unsigned long)ps_tsk_num,
		"acpi_pad/%d", ps_tsk_num);

	if (IS_ERR(ps_tsks[ps_tsk_num])) {
		rc = PTR_ERR(ps_tsks[ps_tsk_num]);
		ps_tsks[ps_tsk_num] = NULL;
	} else {
		rc = 0;
		ps_tsk_num++;
	}

	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)
{
	cpus_read_lock();

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

	cpus_read_unlock();
}

static uint32_t acpi_pad_idle_cpus_num(void)
{
	return ps_tsk_num;
}

static ssize_t rrtime_store(struct device *dev,
	struct device_attribute *attr, const char *buf, size_t count)
{
	unsigned long num;

	if (kstrtoul(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 rrtime_show(struct device *dev,
	struct device_attribute *attr, char *buf)
{
	return sysfs_emit(buf, "%d\n", round_robin_time);
}
static DEVICE_ATTR_RW(rrtime);

static ssize_t idlepct_store(struct device *dev,
	struct device_attribute *attr, const char *buf, size_t count)
{
	unsigned long num;

	if (kstrtoul(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 idlepct_show(struct device *dev,
	struct device_attribute *attr, char *buf)
{
	return sysfs_emit(buf, "%d\n", idle_pct);
}
static DEVICE_ATTR_RW(idlepct);

static ssize_t idlecpus_store(struct device *dev,
	struct device_attribute *attr, const char *buf, size_t count)
{
	unsigned long num;

	if (kstrtoul(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 idlecpus_show(struct device *dev,
	struct device_attribute *attr, char *buf)
{
	return cpumap_print_to_pagebuf(false, buf,
				       to_cpumask(pad_busy_cpus_bits));
}

static DEVICE_ATTR_RW(idlecpus);

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;
}

static void acpi_pad_handle_notify(acpi_handle handle)
{
	int num_cpus;
	uint32_t idle_cpus;
	struct acpi_buffer param = {
		.length = 4,
		.pointer = (void *)&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_evaluate_ost(handle, ACPI_PROCESSOR_AGGREGATOR_NOTIFY, 0, &param);
	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_netlink_event(device->pnp.device_class,
			dev_name(&device->dev), event, 0);
		break;
	default:
		pr_warn("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 void acpi_pad_remove(struct acpi_device *device)
{
	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);
}

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)
{
	/* Xen ACPI PAD is used when running as Xen Dom0. */
	if (xen_initial_domain())
		return -ENODEV;

	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
2025cf9e	1	// SPDX-License-Identifier: GPL-2.0-only
8e0af514 SL	2	/*
	3	* acpi_pad.c ACPI Processor Aggregator Driver
	4	*
	5	* Copyright (c) 2009, Intel Corporation.
8e0af514 SL	6	*/
	7
	8	#include <linux/kernel.h>
	9	#include <linux/cpumask.h>
	10	#include <linux/module.h>
	11	#include <linux/init.h>
	12	#include <linux/types.h>
	13	#include <linux/kthread.h>
ae7e81c0	14	#include <uapi/linux/sched/types.h>
8e0af514 SL	15	#include <linux/freezer.h>
8e0af514 SL	16	#include <linux/cpu.h>
979081e7	17	#include <linux/tick.h>
5a0e3ad6	18	#include <linux/slab.h>
8b48463f	19	#include <linux/acpi.h>
2a606a18	20	#include <linux/perf_event.h>
bc83cccc	21	#include <asm/mwait.h>
e311404f	22	#include <xen/xen.h>
8e0af514	23
a40770a9	24	#define ACPI_PROCESSOR_AGGREGATOR_CLASS "acpi_pad"
8e0af514 SL	25	#define ACPI_PROCESSOR_AGGREGATOR_DEVICE_NAME "Processor Aggregator"
	26	#define ACPI_PROCESSOR_AGGREGATOR_NOTIFY 0x80
	27	static DEFINE_MUTEX(isolated_cpus_lock);
5f160126	28	static DEFINE_MUTEX(round_robin_lock);
8e0af514	29
8e0af514	30	static unsigned long power_saving_mwait_eax;
0dc698b9 VP	31
	32	static unsigned char tsc_detected_unstable;
	33	static unsigned char tsc_marked_unstable;
	34
8e0af514 SL	35	static void power_saving_mwait_init(void)
	36	{
	37	unsigned int eax, ebx, ecx, edx;
	38	unsigned int highest_cstate = 0;
	39	unsigned int highest_subcstate = 0;
	40	int i;
	41
	42	if (!boot_cpu_has(X86_FEATURE_MWAIT))
	43	return;
	44	if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
	45	return;
	46
	47	cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &edx);
	48
	49	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) \|\|
	50	!(ecx & CPUID5_ECX_INTERRUPT_BREAK))
	51	return;
	52
	53	edx >>= MWAIT_SUBSTATE_SIZE;
	54	for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
	55	if (edx & MWAIT_SUBSTATE_MASK) {
	56	highest_cstate = i;
	57	highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
	58	}
	59	}
	60	power_saving_mwait_eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) \|
	61	(highest_subcstate - 1);
	62
592913ec	63	#if defined(CONFIG_X86)
8e0af514	64	switch (boot_cpu_data.x86_vendor) {
7377ed4b	65	case X86_VENDOR_HYGON:
8e0af514 SL	66	case X86_VENDOR_AMD:
8e0af514 SL	67	case X86_VENDOR_INTEL:
773b2f30	68	case X86_VENDOR_ZHAOXIN:
8e0af514 SL	69	/*
	70	* AMD Fam10h TSC will tick in all
	71	* C/P/S0/S1 states when this bit is set.
	72	*/
8aa4b14e CG	73	if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
8aa4b14e CG	74	tsc_detected_unstable = 1;
8aa4b14e	75	break;
8e0af514	76	default:
3ff70551	77	/* TSC could halt in idle */
0dc698b9	78	tsc_detected_unstable = 1;
8e0af514 SL	79	}
	80	#endif
	81	}
	82
	83	static unsigned long cpu_weight[NR_CPUS];
	84	static int tsk_in_cpu[NR_CPUS] = {[0 ... NR_CPUS-1] = -1};
	85	static DECLARE_BITMAP(pad_busy_cpus_bits, NR_CPUS);
	86	static void round_robin_cpu(unsigned int tsk_index)
	87	{
	88	struct cpumask *pad_busy_cpus = to_cpumask(pad_busy_cpus_bits);
	89	cpumask_var_t tmp;
	90	int cpu;
f67538f8	91	unsigned long min_weight = -1;
134043cd	92	unsigned long preferred_cpu;
8e0af514 SL	93
	94	if (!alloc_cpumask_var(&tmp, GFP_KERNEL))
	95	return;
	96
5f160126	97	mutex_lock(&round_robin_lock);
8e0af514 SL	98	cpumask_clear(tmp);
8e0af514 SL	99	for_each_cpu(cpu, pad_busy_cpus)
06931e62	100	cpumask_or(tmp, tmp, topology_sibling_cpumask(cpu));
8e0af514 SL	101	cpumask_andnot(tmp, cpu_online_mask, tmp);
	102	/* avoid HT sibilings if possible */
	103	if (cpumask_empty(tmp))
	104	cpumask_andnot(tmp, cpu_online_mask, pad_busy_cpus);
	105	if (cpumask_empty(tmp)) {
5f160126	106	mutex_unlock(&round_robin_lock);
8b29d29a	107	free_cpumask_var(tmp);
8e0af514 SL	108	return;
	109	}
	110	for_each_cpu(cpu, tmp) {
	111	if (cpu_weight[cpu] < min_weight) {
	112	min_weight = cpu_weight[cpu];
	113	preferred_cpu = cpu;
	114	}
	115	}
	116
	117	if (tsk_in_cpu[tsk_index] != -1)
	118	cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus);
	119	tsk_in_cpu[tsk_index] = preferred_cpu;
	120	cpumask_set_cpu(preferred_cpu, pad_busy_cpus);
	121	cpu_weight[preferred_cpu]++;
5f160126	122	mutex_unlock(&round_robin_lock);
8e0af514 SL	123
8e0af514 SL	124	set_cpus_allowed_ptr(current, cpumask_of(preferred_cpu));
8b29d29a LS	125
8b29d29a LS	126	free_cpumask_var(tmp);
8e0af514 SL	127	}
	128
	129	static void exit_round_robin(unsigned int tsk_index)
	130	{
	131	struct cpumask *pad_busy_cpus = to_cpumask(pad_busy_cpus_bits);
c8eb628c	132
8e0af514 SL	133	cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus);
	134	tsk_in_cpu[tsk_index] = -1;
	135	}
	136
	137	static unsigned int idle_pct = 5; /* percentage */
fa7584e1	138	static unsigned int round_robin_time = 1; /* second */
8e0af514 SL	139	static int power_saving_thread(void *data)
8e0af514 SL	140	{
8e0af514 SL	141	int do_sleep;
	142	unsigned int tsk_index = (unsigned long)data;
	143	u64 last_jiffies = 0;
	144
4ca6c1a0	145	sched_set_fifo_low(current);
8e0af514 SL	146
8e0af514 SL	147	while (!kthread_should_stop()) {
4ff248f3	148	unsigned long expire_time;
8e0af514	149
8e0af514	150	/* round robin to cpus */
4ff248f3 MS	151	expire_time = last_jiffies + round_robin_time * HZ;
4ff248f3 MS	152	if (time_before(expire_time, jiffies)) {
8e0af514 SL	153	last_jiffies = jiffies;
	154	round_robin_cpu(tsk_index);
	155	}
	156
	157	do_sleep = 0;
	158
8e0af514 SL	159	expire_time = jiffies + HZ * (100 - idle_pct) / 100;
	160
	161	while (!need_resched()) {
0dc698b9 VP	162	if (tsc_detected_unstable && !tsc_marked_unstable) {
	163	/* TSC could halt in idle, so notify users */
	164	mark_tsc_unstable("TSC halts in idle");
	165	tsc_marked_unstable = 1;
	166	}
8e0af514	167	local_irq_disable();
2a606a18 SE	168
	169	perf_lopwr_cb(true);
	170
979081e7	171	tick_broadcast_enable();
c7952135	172	tick_broadcast_enter();
8e0af514 SL	173	stop_critical_timings();
8e0af514 SL	174
16824255	175	mwait_idle_with_hints(power_saving_mwait_eax, 1);
8e0af514 SL	176
8e0af514 SL	177	start_critical_timings();
c7952135	178	tick_broadcast_exit();
2a606a18 SE	179
	180	perf_lopwr_cb(false);
	181
8e0af514 SL	182	local_irq_enable();
8e0af514 SL	183
4ff248f3	184	if (time_before(expire_time, jiffies)) {
8e0af514 SL	185	do_sleep = 1;
	186	break;
	187	}
	188	}
	189
8e0af514 SL	190	/*
	191	* current sched_rt has threshold for rt task running time.
	192	* When a rt task uses 95% CPU time, the rt thread will be
	193	* scheduled out for 5% CPU time to not starve other tasks. But
	194	* the mechanism only works when all CPUs have RT task running,
	195	* as if one CPU hasn't RT task, RT task from other CPUs will
	196	* borrow CPU time from this CPU and cause RT task use > 95%
3b8cb427	197	* CPU time. To make 'avoid starvation' work, takes a nap here.
8e0af514	198	*/
5b59c69e	199	if (unlikely(do_sleep))
8e0af514	200	schedule_timeout_killable(HZ * idle_pct / 100);
5b59c69e TC	201
	202	/* If an external event has set the need_resched flag, then
	203	* we need to deal with it, or this loop will continue to
	204	* spin without calling __mwait().
	205	*/
	206	if (unlikely(need_resched()))
	207	schedule();
8e0af514 SL	208	}
	209
	210	exit_round_robin(tsk_index);
	211	return 0;
	212	}
	213
	214	static struct task_struct *ps_tsks[NR_CPUS];
	215	static unsigned int ps_tsk_num;
	216	static int create_power_saving_task(void)
	217	{
5d7e4386	218	int rc;
3b8cb427	219
8e0af514 SL	220	ps_tsks[ps_tsk_num] = kthread_run(power_saving_thread,
8e0af514 SL	221	(void *)(unsigned long)ps_tsk_num,
150ed86f	222	"acpi_pad/%d", ps_tsk_num);
5d7e4386 RR	223
	224	if (IS_ERR(ps_tsks[ps_tsk_num])) {
	225	rc = PTR_ERR(ps_tsks[ps_tsk_num]);
3b8cb427	226	ps_tsks[ps_tsk_num] = NULL;
5d7e4386 RR	227	} else {
	228	rc = 0;
	229	ps_tsk_num++;
	230	}
3b8cb427 CG	231
3b8cb427 CG	232	return rc;
8e0af514 SL	233	}
	234
	235	static void destroy_power_saving_task(void)
	236	{
	237	if (ps_tsk_num > 0) {
	238	ps_tsk_num--;
	239	kthread_stop(ps_tsks[ps_tsk_num]);
3b8cb427	240	ps_tsks[ps_tsk_num] = NULL;
8e0af514 SL	241	}
	242	}
	243
	244	static void set_power_saving_task_num(unsigned int num)
	245	{
	246	if (num > ps_tsk_num) {
	247	while (ps_tsk_num < num) {
	248	if (create_power_saving_task())
	249	return;
	250	}
	251	} else if (num < ps_tsk_num) {
	252	while (ps_tsk_num > num)
	253	destroy_power_saving_task();
	254	}
	255	}
	256
3b8cb427	257	static void acpi_pad_idle_cpus(unsigned int num_cpus)
8e0af514	258	{
95ac7067	259	cpus_read_lock();
8e0af514 SL	260
	261	num_cpus = min_t(unsigned int, num_cpus, num_online_cpus());
	262	set_power_saving_task_num(num_cpus);
	263
95ac7067	264	cpus_read_unlock();
8e0af514 SL	265	}
	266
	267	static uint32_t acpi_pad_idle_cpus_num(void)
	268	{
	269	return ps_tsk_num;
	270	}
	271
0f39ee83	272	static ssize_t rrtime_store(struct device *dev,
8e0af514 SL	273	struct device_attribute attr, const char buf, size_t count)
	274	{
	275	unsigned long num;
c8eb628c	276
73d4511a	277	if (kstrtoul(buf, 0, &num))
8e0af514 SL	278	return -EINVAL;
	279	if (num < 1 \|\| num >= 100)
	280	return -EINVAL;
	281	mutex_lock(&isolated_cpus_lock);
	282	round_robin_time = num;
	283	mutex_unlock(&isolated_cpus_lock);
	284	return count;
	285	}
	286
0f39ee83	287	static ssize_t rrtime_show(struct device *dev,
8e0af514 SL	288	struct device_attribute attr, char buf)
8e0af514 SL	289	{
92266c65	290	return sysfs_emit(buf, "%d\n", round_robin_time);
8e0af514	291	}
0f39ee83	292	static DEVICE_ATTR_RW(rrtime);
8e0af514	293
0f39ee83	294	static ssize_t idlepct_store(struct device *dev,
8e0af514 SL	295	struct device_attribute attr, const char buf, size_t count)
	296	{
	297	unsigned long num;
c8eb628c	298
73d4511a	299	if (kstrtoul(buf, 0, &num))
8e0af514 SL	300	return -EINVAL;
	301	if (num < 1 \|\| num >= 100)
	302	return -EINVAL;
	303	mutex_lock(&isolated_cpus_lock);
	304	idle_pct = num;
	305	mutex_unlock(&isolated_cpus_lock);
	306	return count;
	307	}
	308
0f39ee83	309	static ssize_t idlepct_show(struct device *dev,
8e0af514 SL	310	struct device_attribute attr, char buf)
8e0af514 SL	311	{
92266c65	312	return sysfs_emit(buf, "%d\n", idle_pct);
8e0af514	313	}
0f39ee83	314	static DEVICE_ATTR_RW(idlepct);
8e0af514	315
0f39ee83	316	static ssize_t idlecpus_store(struct device *dev,
8e0af514 SL	317	struct device_attribute attr, const char buf, size_t count)
	318	{
	319	unsigned long num;
c8eb628c	320
73d4511a	321	if (kstrtoul(buf, 0, &num))
8e0af514 SL	322	return -EINVAL;
	323	mutex_lock(&isolated_cpus_lock);
	324	acpi_pad_idle_cpus(num);
	325	mutex_unlock(&isolated_cpus_lock);
	326	return count;
	327	}
	328
0f39ee83	329	static ssize_t idlecpus_show(struct device *dev,
8e0af514 SL	330	struct device_attribute attr, char buf)
8e0af514 SL	331	{
5aaba363 SH	332	return cpumap_print_to_pagebuf(false, buf,
5aaba363 SH	333	to_cpumask(pad_busy_cpus_bits));
8e0af514	334	}
5aaba363	335
0f39ee83	336	static DEVICE_ATTR_RW(idlecpus);
8e0af514 SL	337
	338	static int acpi_pad_add_sysfs(struct acpi_device *device)
	339	{
	340	int result;
	341
	342	result = device_create_file(&device->dev, &dev_attr_idlecpus);
	343	if (result)
	344	return -ENODEV;
	345	result = device_create_file(&device->dev, &dev_attr_idlepct);
	346	if (result) {
	347	device_remove_file(&device->dev, &dev_attr_idlecpus);
	348	return -ENODEV;
	349	}
	350	result = device_create_file(&device->dev, &dev_attr_rrtime);
	351	if (result) {
	352	device_remove_file(&device->dev, &dev_attr_idlecpus);
	353	device_remove_file(&device->dev, &dev_attr_idlepct);
	354	return -ENODEV;
	355	}
	356	return 0;
	357	}
	358
	359	static void acpi_pad_remove_sysfs(struct acpi_device *device)
	360	{
	361	device_remove_file(&device->dev, &dev_attr_idlecpus);
	362	device_remove_file(&device->dev, &dev_attr_idlepct);
	363	device_remove_file(&device->dev, &dev_attr_rrtime);
	364	}
	365
c9ad8e06 LB	366	/*
	367	* Query firmware how many CPUs should be idle
	368	* return -1 on failure
	369	*/
	370	static int acpi_pad_pur(acpi_handle handle)
8e0af514 SL	371	{
8e0af514 SL	372	struct acpi_buffer buffer = {ACPI_ALLOCATE_BUFFER, NULL};
8e0af514	373	union acpi_object *package;
c9ad8e06	374	int num = -1;
8e0af514	375
3b8cb427	376	if (ACPI_FAILURE(acpi_evaluate_object(handle, "_PUR", NULL, &buffer)))
c9ad8e06	377	return num;
3b8cb427 CG	378
3b8cb427 CG	379	if (!buffer.length \|\| !buffer.pointer)
c9ad8e06	380	return num;
3b8cb427	381
8e0af514	382	package = buffer.pointer;
c9ad8e06 LB	383
	384	if (package->type == ACPI_TYPE_PACKAGE &&
	385	package->package.count == 2 &&
	386	package->package.elements[0].integer.value == 1) /* rev 1 */
	387
	388	num = package->package.elements[1].integer.value;
	389
8e0af514	390	kfree(buffer.pointer);
c9ad8e06	391	return num;
8e0af514 SL	392	}
8e0af514 SL	393
8e0af514 SL	394	static void acpi_pad_handle_notify(acpi_handle handle)
8e0af514 SL	395	{
3b8cb427	396	int num_cpus;
8e0af514	397	uint32_t idle_cpus;
8b296d94 JL	398	struct acpi_buffer param = {
	399	.length = 4,
	400	.pointer = (void *)&idle_cpus,
	401	};
8e0af514 SL	402
8e0af514 SL	403	mutex_lock(&isolated_cpus_lock);
c9ad8e06 LB	404	num_cpus = acpi_pad_pur(handle);
c9ad8e06 LB	405	if (num_cpus < 0) {
8e0af514 SL	406	mutex_unlock(&isolated_cpus_lock);
	407	return;
	408	}
3b8cb427	409	acpi_pad_idle_cpus(num_cpus);
8e0af514	410	idle_cpus = acpi_pad_idle_cpus_num();
8b296d94	411	acpi_evaluate_ost(handle, ACPI_PROCESSOR_AGGREGATOR_NOTIFY, 0, &param);
8e0af514 SL	412	mutex_unlock(&isolated_cpus_lock);
	413	}
	414
	415	static void acpi_pad_notify(acpi_handle handle, u32 event,
	416	void *data)
	417	{
	418	struct acpi_device *device = data;
	419
	420	switch (event) {
	421	case ACPI_PROCESSOR_AGGREGATOR_NOTIFY:
	422	acpi_pad_handle_notify(handle);
8e0af514 SL	423	acpi_bus_generate_netlink_event(device->pnp.device_class,
	424	dev_name(&device->dev), event, 0);
	425	break;
	426	default:
73d4511a	427	pr_warn("Unsupported event [0x%x]\n", event);
8e0af514 SL	428	break;
	429	}
	430	}
	431
	432	static int acpi_pad_add(struct acpi_device *device)
	433	{
	434	acpi_status status;
	435
	436	strcpy(acpi_device_name(device), ACPI_PROCESSOR_AGGREGATOR_DEVICE_NAME);
	437	strcpy(acpi_device_class(device), ACPI_PROCESSOR_AGGREGATOR_CLASS);
	438
	439	if (acpi_pad_add_sysfs(device))
	440	return -ENODEV;
	441
	442	status = acpi_install_notify_handler(device->handle,
	443	ACPI_DEVICE_NOTIFY, acpi_pad_notify, device);
	444	if (ACPI_FAILURE(status)) {
	445	acpi_pad_remove_sysfs(device);
	446	return -ENODEV;
	447	}
	448
	449	return 0;
	450	}
	451
6c0eb5ba	452	static void acpi_pad_remove(struct acpi_device *device)
8e0af514 SL	453	{
	454	mutex_lock(&isolated_cpus_lock);
	455	acpi_pad_idle_cpus(0);
	456	mutex_unlock(&isolated_cpus_lock);
	457
	458	acpi_remove_notify_handler(device->handle,
	459	ACPI_DEVICE_NOTIFY, acpi_pad_notify);
	460	acpi_pad_remove_sysfs(device);
8e0af514 SL	461	}
	462
	463	static const struct acpi_device_id pad_device_ids[] = {
	464	{"ACPI000C", 0},
	465	{"", 0},
	466	};
	467	MODULE_DEVICE_TABLE(acpi, pad_device_ids);
	468
	469	static struct acpi_driver acpi_pad_driver = {
	470	.name = "processor_aggregator",
	471	.class = ACPI_PROCESSOR_AGGREGATOR_CLASS,
	472	.ids = pad_device_ids,
	473	.ops = {
	474	.add = acpi_pad_add,
	475	.remove = acpi_pad_remove,
	476	},
	477	};
	478
	479	static int __init acpi_pad_init(void)
	480	{
e311404f JG	481	/* Xen ACPI PAD is used when running as Xen Dom0. */
	482	if (xen_initial_domain())
	483	return -ENODEV;
	484
8e0af514 SL	485	power_saving_mwait_init();
	486	if (power_saving_mwait_eax == 0)
	487	return -EINVAL;
	488
	489	return acpi_bus_register_driver(&acpi_pad_driver);
	490	}
	491
	492	static void __exit acpi_pad_exit(void)
	493	{
	494	acpi_bus_unregister_driver(&acpi_pad_driver);
	495	}
	496
	497	module_init(acpi_pad_init);
	498	module_exit(acpi_pad_exit);
	499	MODULE_AUTHOR("Shaohua Li<shaohua.li@intel.com>");
	500	MODULE_DESCRIPTION("ACPI Processor Aggregator Driver");
	501	MODULE_LICENSE("GPL");