[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 <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:
		/*
		 * 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 uninitialized_var(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)
{
	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()) {
		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();
			tick_broadcast_enable();
			tick_broadcast_enter();
			stop_critical_timings();

			mwait_idle_with_hints(power_saving_mwait_eax, 1);

			start_critical_timings();
			tick_broadcast_exit();
			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)
{
	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 (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 acpi_pad_rrtime_show(struct device *dev,
	struct device_attribute *attr, char *buf)
{
	return scnprintf(buf, PAGE_SIZE, "%d\n", 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 (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 acpi_pad_idlepct_show(struct device *dev,
	struct device_attribute *attr, char *buf)
{
	return scnprintf(buf, PAGE_SIZE, "%d\n", 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 (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 acpi_pad_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(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;
}

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 int 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);
	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)
{
	/* 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>
bc83cccc	20	#include <asm/mwait.h>
e311404f	21	#include <xen/xen.h>
8e0af514	22
a40770a9	23	#define ACPI_PROCESSOR_AGGREGATOR_CLASS "acpi_pad"
8e0af514 SL	24	#define ACPI_PROCESSOR_AGGREGATOR_DEVICE_NAME "Processor Aggregator"
	25	#define ACPI_PROCESSOR_AGGREGATOR_NOTIFY 0x80
	26	static DEFINE_MUTEX(isolated_cpus_lock);
5f160126	27	static DEFINE_MUTEX(round_robin_lock);
8e0af514	28
8e0af514	29	static unsigned long power_saving_mwait_eax;
0dc698b9 VP	30
	31	static unsigned char tsc_detected_unstable;
	32	static unsigned char tsc_marked_unstable;
	33
8e0af514 SL	34	static void power_saving_mwait_init(void)
	35	{
	36	unsigned int eax, ebx, ecx, edx;
	37	unsigned int highest_cstate = 0;
	38	unsigned int highest_subcstate = 0;
	39	int i;
	40
	41	if (!boot_cpu_has(X86_FEATURE_MWAIT))
	42	return;
	43	if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
	44	return;
	45
	46	cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &edx);
	47
	48	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) \|\|
	49	!(ecx & CPUID5_ECX_INTERRUPT_BREAK))
	50	return;
	51
	52	edx >>= MWAIT_SUBSTATE_SIZE;
	53	for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
	54	if (edx & MWAIT_SUBSTATE_MASK) {
	55	highest_cstate = i;
	56	highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
	57	}
	58	}
	59	power_saving_mwait_eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) \|
	60	(highest_subcstate - 1);
	61
592913ec	62	#if defined(CONFIG_X86)
8e0af514	63	switch (boot_cpu_data.x86_vendor) {
7377ed4b	64	case X86_VENDOR_HYGON:
8e0af514 SL	65	case X86_VENDOR_AMD:
	66	case X86_VENDOR_INTEL:
	67	/*
	68	* AMD Fam10h TSC will tick in all
	69	* C/P/S0/S1 states when this bit is set.
	70	*/
8aa4b14e CG	71	if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
8aa4b14e CG	72	tsc_detected_unstable = 1;
8aa4b14e	73	break;
8e0af514	74	default:
3ff70551	75	/* TSC could halt in idle */
0dc698b9	76	tsc_detected_unstable = 1;
8e0af514 SL	77	}
	78	#endif
	79	}
	80
	81	static unsigned long cpu_weight[NR_CPUS];
	82	static int tsk_in_cpu[NR_CPUS] = {[0 ... NR_CPUS-1] = -1};
	83	static DECLARE_BITMAP(pad_busy_cpus_bits, NR_CPUS);
	84	static void round_robin_cpu(unsigned int tsk_index)
	85	{
	86	struct cpumask *pad_busy_cpus = to_cpumask(pad_busy_cpus_bits);
	87	cpumask_var_t tmp;
	88	int cpu;
f67538f8 AM	89	unsigned long min_weight = -1;
f67538f8 AM	90	unsigned long uninitialized_var(preferred_cpu);
8e0af514 SL	91
	92	if (!alloc_cpumask_var(&tmp, GFP_KERNEL))
	93	return;
	94
5f160126	95	mutex_lock(&round_robin_lock);
8e0af514 SL	96	cpumask_clear(tmp);
8e0af514 SL	97	for_each_cpu(cpu, pad_busy_cpus)
06931e62	98	cpumask_or(tmp, tmp, topology_sibling_cpumask(cpu));
8e0af514 SL	99	cpumask_andnot(tmp, cpu_online_mask, tmp);
	100	/* avoid HT sibilings if possible */
	101	if (cpumask_empty(tmp))
	102	cpumask_andnot(tmp, cpu_online_mask, pad_busy_cpus);
	103	if (cpumask_empty(tmp)) {
5f160126	104	mutex_unlock(&round_robin_lock);
8b29d29a	105	free_cpumask_var(tmp);
8e0af514 SL	106	return;
	107	}
	108	for_each_cpu(cpu, tmp) {
	109	if (cpu_weight[cpu] < min_weight) {
	110	min_weight = cpu_weight[cpu];
	111	preferred_cpu = cpu;
	112	}
	113	}
	114
	115	if (tsk_in_cpu[tsk_index] != -1)
	116	cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus);
	117	tsk_in_cpu[tsk_index] = preferred_cpu;
	118	cpumask_set_cpu(preferred_cpu, pad_busy_cpus);
	119	cpu_weight[preferred_cpu]++;
5f160126	120	mutex_unlock(&round_robin_lock);
8e0af514 SL	121
8e0af514 SL	122	set_cpus_allowed_ptr(current, cpumask_of(preferred_cpu));
8b29d29a LS	123
8b29d29a LS	124	free_cpumask_var(tmp);
8e0af514 SL	125	}
	126
	127	static void exit_round_robin(unsigned int tsk_index)
	128	{
	129	struct cpumask *pad_busy_cpus = to_cpumask(pad_busy_cpus_bits);
	130	cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus);
	131	tsk_in_cpu[tsk_index] = -1;
	132	}
	133
	134	static unsigned int idle_pct = 5; /* percentage */
fa7584e1	135	static unsigned int round_robin_time = 1; /* second */
8e0af514 SL	136	static int power_saving_thread(void *data)
	137	{
	138	struct sched_param param = {.sched_priority = 1};
	139	int do_sleep;
	140	unsigned int tsk_index = (unsigned long)data;
	141	u64 last_jiffies = 0;
	142
	143	sched_setscheduler(current, SCHED_RR, &param);
	144
	145	while (!kthread_should_stop()) {
4ff248f3	146	unsigned long expire_time;
8e0af514	147
8e0af514	148	/* round robin to cpus */
4ff248f3 MS	149	expire_time = last_jiffies + round_robin_time * HZ;
4ff248f3 MS	150	if (time_before(expire_time, jiffies)) {
8e0af514 SL	151	last_jiffies = jiffies;
	152	round_robin_cpu(tsk_index);
	153	}
	154
	155	do_sleep = 0;
	156
8e0af514 SL	157	expire_time = jiffies + HZ * (100 - idle_pct) / 100;
	158
	159	while (!need_resched()) {
0dc698b9 VP	160	if (tsc_detected_unstable && !tsc_marked_unstable) {
	161	/* TSC could halt in idle, so notify users */
	162	mark_tsc_unstable("TSC halts in idle");
	163	tsc_marked_unstable = 1;
	164	}
8e0af514	165	local_irq_disable();
979081e7	166	tick_broadcast_enable();
c7952135	167	tick_broadcast_enter();
8e0af514 SL	168	stop_critical_timings();
8e0af514 SL	169
16824255	170	mwait_idle_with_hints(power_saving_mwait_eax, 1);
8e0af514 SL	171
8e0af514 SL	172	start_critical_timings();
c7952135	173	tick_broadcast_exit();
8e0af514 SL	174	local_irq_enable();
8e0af514 SL	175
4ff248f3	176	if (time_before(expire_time, jiffies)) {
8e0af514 SL	177	do_sleep = 1;
	178	break;
	179	}
	180	}
	181
8e0af514 SL	182	/*
	183	* current sched_rt has threshold for rt task running time.
	184	* When a rt task uses 95% CPU time, the rt thread will be
	185	* scheduled out for 5% CPU time to not starve other tasks. But
	186	* the mechanism only works when all CPUs have RT task running,
	187	* as if one CPU hasn't RT task, RT task from other CPUs will
	188	* borrow CPU time from this CPU and cause RT task use > 95%
3b8cb427	189	* CPU time. To make 'avoid starvation' work, takes a nap here.
8e0af514	190	*/
5b59c69e	191	if (unlikely(do_sleep))
8e0af514	192	schedule_timeout_killable(HZ * idle_pct / 100);
5b59c69e TC	193
	194	/* If an external event has set the need_resched flag, then
	195	* we need to deal with it, or this loop will continue to
	196	* spin without calling __mwait().
	197	*/
	198	if (unlikely(need_resched()))
	199	schedule();
8e0af514 SL	200	}
	201
	202	exit_round_robin(tsk_index);
	203	return 0;
	204	}
	205
	206	static struct task_struct *ps_tsks[NR_CPUS];
	207	static unsigned int ps_tsk_num;
	208	static int create_power_saving_task(void)
	209	{
5d7e4386	210	int rc;
3b8cb427	211
8e0af514 SL	212	ps_tsks[ps_tsk_num] = kthread_run(power_saving_thread,
8e0af514 SL	213	(void *)(unsigned long)ps_tsk_num,
150ed86f	214	"acpi_pad/%d", ps_tsk_num);
5d7e4386 RR	215
	216	if (IS_ERR(ps_tsks[ps_tsk_num])) {
	217	rc = PTR_ERR(ps_tsks[ps_tsk_num]);
3b8cb427	218	ps_tsks[ps_tsk_num] = NULL;
5d7e4386 RR	219	} else {
	220	rc = 0;
	221	ps_tsk_num++;
	222	}
3b8cb427 CG	223
3b8cb427 CG	224	return rc;
8e0af514 SL	225	}
	226
	227	static void destroy_power_saving_task(void)
	228	{
	229	if (ps_tsk_num > 0) {
	230	ps_tsk_num--;
	231	kthread_stop(ps_tsks[ps_tsk_num]);
3b8cb427	232	ps_tsks[ps_tsk_num] = NULL;
8e0af514 SL	233	}
	234	}
	235
	236	static void set_power_saving_task_num(unsigned int num)
	237	{
	238	if (num > ps_tsk_num) {
	239	while (ps_tsk_num < num) {
	240	if (create_power_saving_task())
	241	return;
	242	}
	243	} else if (num < ps_tsk_num) {
	244	while (ps_tsk_num > num)
	245	destroy_power_saving_task();
	246	}
	247	}
	248
3b8cb427	249	static void acpi_pad_idle_cpus(unsigned int num_cpus)
8e0af514 SL	250	{
	251	get_online_cpus();
	252
	253	num_cpus = min_t(unsigned int, num_cpus, num_online_cpus());
	254	set_power_saving_task_num(num_cpus);
	255
	256	put_online_cpus();
8e0af514 SL	257	}
	258
	259	static uint32_t acpi_pad_idle_cpus_num(void)
	260	{
	261	return ps_tsk_num;
	262	}
	263
	264	static ssize_t acpi_pad_rrtime_store(struct device *dev,
	265	struct device_attribute attr, const char buf, size_t count)
	266	{
	267	unsigned long num;
73d4511a	268	if (kstrtoul(buf, 0, &num))
8e0af514 SL	269	return -EINVAL;
	270	if (num < 1 \|\| num >= 100)
	271	return -EINVAL;
	272	mutex_lock(&isolated_cpus_lock);
	273	round_robin_time = num;
	274	mutex_unlock(&isolated_cpus_lock);
	275	return count;
	276	}
	277
	278	static ssize_t acpi_pad_rrtime_show(struct device *dev,
	279	struct device_attribute attr, char buf)
	280	{
32297abd	281	return scnprintf(buf, PAGE_SIZE, "%d\n", round_robin_time);
8e0af514 SL	282	}
	283	static DEVICE_ATTR(rrtime, S_IRUGO\|S_IWUSR,
	284	acpi_pad_rrtime_show,
	285	acpi_pad_rrtime_store);
	286
	287	static ssize_t acpi_pad_idlepct_store(struct device *dev,
	288	struct device_attribute attr, const char buf, size_t count)
	289	{
	290	unsigned long num;
73d4511a	291	if (kstrtoul(buf, 0, &num))
8e0af514 SL	292	return -EINVAL;
	293	if (num < 1 \|\| num >= 100)
	294	return -EINVAL;
	295	mutex_lock(&isolated_cpus_lock);
	296	idle_pct = num;
	297	mutex_unlock(&isolated_cpus_lock);
	298	return count;
	299	}
	300
	301	static ssize_t acpi_pad_idlepct_show(struct device *dev,
	302	struct device_attribute attr, char buf)
	303	{
32297abd	304	return scnprintf(buf, PAGE_SIZE, "%d\n", idle_pct);
8e0af514 SL	305	}
	306	static DEVICE_ATTR(idlepct, S_IRUGO\|S_IWUSR,
	307	acpi_pad_idlepct_show,
	308	acpi_pad_idlepct_store);
	309
	310	static ssize_t acpi_pad_idlecpus_store(struct device *dev,
	311	struct device_attribute attr, const char buf, size_t count)
	312	{
	313	unsigned long num;
73d4511a	314	if (kstrtoul(buf, 0, &num))
8e0af514 SL	315	return -EINVAL;
	316	mutex_lock(&isolated_cpus_lock);
	317	acpi_pad_idle_cpus(num);
	318	mutex_unlock(&isolated_cpus_lock);
	319	return count;
	320	}
	321
	322	static ssize_t acpi_pad_idlecpus_show(struct device *dev,
	323	struct device_attribute attr, char buf)
	324	{
5aaba363 SH	325	return cpumap_print_to_pagebuf(false, buf,
5aaba363 SH	326	to_cpumask(pad_busy_cpus_bits));
8e0af514	327	}
5aaba363	328
8e0af514 SL	329	static DEVICE_ATTR(idlecpus, S_IRUGO\|S_IWUSR,
	330	acpi_pad_idlecpus_show,
	331	acpi_pad_idlecpus_store);
	332
	333	static int acpi_pad_add_sysfs(struct acpi_device *device)
	334	{
	335	int result;
	336
	337	result = device_create_file(&device->dev, &dev_attr_idlecpus);
	338	if (result)
	339	return -ENODEV;
	340	result = device_create_file(&device->dev, &dev_attr_idlepct);
	341	if (result) {
	342	device_remove_file(&device->dev, &dev_attr_idlecpus);
	343	return -ENODEV;
	344	}
	345	result = device_create_file(&device->dev, &dev_attr_rrtime);
	346	if (result) {
	347	device_remove_file(&device->dev, &dev_attr_idlecpus);
	348	device_remove_file(&device->dev, &dev_attr_idlepct);
	349	return -ENODEV;
	350	}
	351	return 0;
	352	}
	353
	354	static void acpi_pad_remove_sysfs(struct acpi_device *device)
	355	{
	356	device_remove_file(&device->dev, &dev_attr_idlecpus);
	357	device_remove_file(&device->dev, &dev_attr_idlepct);
	358	device_remove_file(&device->dev, &dev_attr_rrtime);
	359	}
	360
c9ad8e06 LB	361	/*
	362	* Query firmware how many CPUs should be idle
	363	* return -1 on failure
	364	*/
	365	static int acpi_pad_pur(acpi_handle handle)
8e0af514 SL	366	{
8e0af514 SL	367	struct acpi_buffer buffer = {ACPI_ALLOCATE_BUFFER, NULL};
8e0af514	368	union acpi_object *package;
c9ad8e06	369	int num = -1;
8e0af514	370
3b8cb427	371	if (ACPI_FAILURE(acpi_evaluate_object(handle, "_PUR", NULL, &buffer)))
c9ad8e06	372	return num;
3b8cb427 CG	373
3b8cb427 CG	374	if (!buffer.length \|\| !buffer.pointer)
c9ad8e06	375	return num;
3b8cb427	376
8e0af514	377	package = buffer.pointer;
c9ad8e06 LB	378
	379	if (package->type == ACPI_TYPE_PACKAGE &&
	380	package->package.count == 2 &&
	381	package->package.elements[0].integer.value == 1) /* rev 1 */
	382
	383	num = package->package.elements[1].integer.value;
	384
8e0af514	385	kfree(buffer.pointer);
c9ad8e06	386	return num;
8e0af514 SL	387	}
8e0af514 SL	388
8e0af514 SL	389	static void acpi_pad_handle_notify(acpi_handle handle)
8e0af514 SL	390	{
3b8cb427	391	int num_cpus;
8e0af514	392	uint32_t idle_cpus;
8b296d94 JL	393	struct acpi_buffer param = {
	394	.length = 4,
	395	.pointer = (void *)&idle_cpus,
	396	};
8e0af514 SL	397
8e0af514 SL	398	mutex_lock(&isolated_cpus_lock);
c9ad8e06 LB	399	num_cpus = acpi_pad_pur(handle);
c9ad8e06 LB	400	if (num_cpus < 0) {
8e0af514 SL	401	mutex_unlock(&isolated_cpus_lock);
	402	return;
	403	}
3b8cb427	404	acpi_pad_idle_cpus(num_cpus);
8e0af514	405	idle_cpus = acpi_pad_idle_cpus_num();
8b296d94	406	acpi_evaluate_ost(handle, ACPI_PROCESSOR_AGGREGATOR_NOTIFY, 0, &param);
8e0af514 SL	407	mutex_unlock(&isolated_cpus_lock);
	408	}
	409
	410	static void acpi_pad_notify(acpi_handle handle, u32 event,
	411	void *data)
	412	{
	413	struct acpi_device *device = data;
	414
	415	switch (event) {
	416	case ACPI_PROCESSOR_AGGREGATOR_NOTIFY:
	417	acpi_pad_handle_notify(handle);
8e0af514 SL	418	acpi_bus_generate_netlink_event(device->pnp.device_class,
	419	dev_name(&device->dev), event, 0);
	420	break;
	421	default:
73d4511a	422	pr_warn("Unsupported event [0x%x]\n", event);
8e0af514 SL	423	break;
	424	}
	425	}
	426
	427	static int acpi_pad_add(struct acpi_device *device)
	428	{
	429	acpi_status status;
	430
	431	strcpy(acpi_device_name(device), ACPI_PROCESSOR_AGGREGATOR_DEVICE_NAME);
	432	strcpy(acpi_device_class(device), ACPI_PROCESSOR_AGGREGATOR_CLASS);
	433
	434	if (acpi_pad_add_sysfs(device))
	435	return -ENODEV;
	436
	437	status = acpi_install_notify_handler(device->handle,
	438	ACPI_DEVICE_NOTIFY, acpi_pad_notify, device);
	439	if (ACPI_FAILURE(status)) {
	440	acpi_pad_remove_sysfs(device);
	441	return -ENODEV;
	442	}
	443
	444	return 0;
	445	}
	446
51fac838	447	static int acpi_pad_remove(struct acpi_device *device)
8e0af514 SL	448	{
	449	mutex_lock(&isolated_cpus_lock);
	450	acpi_pad_idle_cpus(0);
	451	mutex_unlock(&isolated_cpus_lock);
	452
	453	acpi_remove_notify_handler(device->handle,
	454	ACPI_DEVICE_NOTIFY, acpi_pad_notify);
	455	acpi_pad_remove_sysfs(device);
	456	return 0;
	457	}
	458
	459	static const struct acpi_device_id pad_device_ids[] = {
	460	{"ACPI000C", 0},
	461	{"", 0},
	462	};
	463	MODULE_DEVICE_TABLE(acpi, pad_device_ids);
	464
	465	static struct acpi_driver acpi_pad_driver = {
	466	.name = "processor_aggregator",
	467	.class = ACPI_PROCESSOR_AGGREGATOR_CLASS,
	468	.ids = pad_device_ids,
	469	.ops = {
	470	.add = acpi_pad_add,
	471	.remove = acpi_pad_remove,
	472	},
	473	};
	474
	475	static int __init acpi_pad_init(void)
	476	{
e311404f JG	477	/* Xen ACPI PAD is used when running as Xen Dom0. */
	478	if (xen_initial_domain())
	479	return -ENODEV;
	480
8e0af514 SL	481	power_saving_mwait_init();
	482	if (power_saving_mwait_eax == 0)
	483	return -EINVAL;
	484
	485	return acpi_bus_register_driver(&acpi_pad_driver);
	486	}
	487
	488	static void __exit acpi_pad_exit(void)
	489	{
	490	acpi_bus_unregister_driver(&acpi_pad_driver);
	491	}
	492
	493	module_init(acpi_pad_init);
	494	module_exit(acpi_pad_exit);
	495	MODULE_AUTHOR("Shaohua Li<shaohua.li@intel.com>");
	496	MODULE_DESCRIPTION("ACPI Processor Aggregator Driver");
	497	MODULE_LICENSE("GPL");