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>
30#include <linux/slab.h>
31#include <acpi/acpi_bus.h>
32#include <acpi/acpi_drivers.h>
33#include <asm/mwait.h>
34
35#define ACPI_PROCESSOR_AGGREGATOR_CLASS	"acpi_pad"
36#define ACPI_PROCESSOR_AGGREGATOR_DEVICE_NAME "Processor Aggregator"
37#define ACPI_PROCESSOR_AGGREGATOR_NOTIFY 0x80
38static DEFINE_MUTEX(isolated_cpus_lock);
39static DEFINE_MUTEX(round_robin_lock);
40
41static unsigned long power_saving_mwait_eax;
42
43static unsigned char tsc_detected_unstable;
44static unsigned char tsc_marked_unstable;
45static unsigned char lapic_detected_unstable;
46static unsigned char lapic_marked_unstable;
47
48static void power_saving_mwait_init(void)
49{
50	unsigned int eax, ebx, ecx, edx;
51	unsigned int highest_cstate = 0;
52	unsigned int highest_subcstate = 0;
53	int i;
54
55	if (!boot_cpu_has(X86_FEATURE_MWAIT))
56		return;
57	if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
58		return;
59
60	cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &edx);
61
62	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
63	    !(ecx & CPUID5_ECX_INTERRUPT_BREAK))
64		return;
65
66	edx >>= MWAIT_SUBSTATE_SIZE;
67	for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
68		if (edx & MWAIT_SUBSTATE_MASK) {
69			highest_cstate = i;
70			highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
71		}
72	}
73	power_saving_mwait_eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
74		(highest_subcstate - 1);
75
76#if defined(CONFIG_X86)
77	switch (boot_cpu_data.x86_vendor) {
78	case X86_VENDOR_AMD:
79	case X86_VENDOR_INTEL:
80		/*
81		 * AMD Fam10h TSC will tick in all
82		 * C/P/S0/S1 states when this bit is set.
83		 */
84		if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
85			tsc_detected_unstable = 1;
86		if (!boot_cpu_has(X86_FEATURE_ARAT))
87			lapic_detected_unstable = 1;
88		break;
89	default:
90		/* TSC & LAPIC could halt in idle */
91		tsc_detected_unstable = 1;
92		lapic_detected_unstable = 1;
93	}
94#endif
95}
96
97static unsigned long cpu_weight[NR_CPUS];
98static int tsk_in_cpu[NR_CPUS] = {[0 ... NR_CPUS-1] = -1};
99static DECLARE_BITMAP(pad_busy_cpus_bits, NR_CPUS);
100static void round_robin_cpu(unsigned int tsk_index)
101{
102	struct cpumask *pad_busy_cpus = to_cpumask(pad_busy_cpus_bits);
103	cpumask_var_t tmp;
104	int cpu;
105	unsigned long min_weight = -1;
106	unsigned long uninitialized_var(preferred_cpu);
107
108	if (!alloc_cpumask_var(&tmp, GFP_KERNEL))
109		return;
110
111	mutex_lock(&round_robin_lock);
112	cpumask_clear(tmp);
113	for_each_cpu(cpu, pad_busy_cpus)
114		cpumask_or(tmp, tmp, topology_thread_cpumask(cpu));
115	cpumask_andnot(tmp, cpu_online_mask, tmp);
116	/* avoid HT sibilings if possible */
117	if (cpumask_empty(tmp))
118		cpumask_andnot(tmp, cpu_online_mask, pad_busy_cpus);
119	if (cpumask_empty(tmp)) {
120		mutex_unlock(&round_robin_lock);
121		return;
122	}
123	for_each_cpu(cpu, tmp) {
124		if (cpu_weight[cpu] < min_weight) {
125			min_weight = cpu_weight[cpu];
126			preferred_cpu = cpu;
127		}
128	}
129
130	if (tsk_in_cpu[tsk_index] != -1)
131		cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus);
132	tsk_in_cpu[tsk_index] = preferred_cpu;
133	cpumask_set_cpu(preferred_cpu, pad_busy_cpus);
134	cpu_weight[preferred_cpu]++;
135	mutex_unlock(&round_robin_lock);
136
137	set_cpus_allowed_ptr(current, cpumask_of(preferred_cpu));
138}
139
140static void exit_round_robin(unsigned int tsk_index)
141{
142	struct cpumask *pad_busy_cpus = to_cpumask(pad_busy_cpus_bits);
143	cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus);
144	tsk_in_cpu[tsk_index] = -1;
145}
146
147static unsigned int idle_pct = 5; /* percentage */
148static unsigned int round_robin_time = 10; /* second */
149static int power_saving_thread(void *data)
150{
151	struct sched_param param = {.sched_priority = 1};
152	int do_sleep;
153	unsigned int tsk_index = (unsigned long)data;
154	u64 last_jiffies = 0;
155
156	sched_setscheduler(current, SCHED_RR, &param);
157
158	while (!kthread_should_stop()) {
159		int cpu;
160		u64 expire_time;
161
162		try_to_freeze();
163
164		/* round robin to cpus */
165		if (last_jiffies + round_robin_time * HZ < jiffies) {
166			last_jiffies = jiffies;
167			round_robin_cpu(tsk_index);
168		}
169
170		do_sleep = 0;
171
172		expire_time = jiffies + HZ * (100 - idle_pct) / 100;
173
174		while (!need_resched()) {
175			if (tsc_detected_unstable && !tsc_marked_unstable) {
176				/* TSC could halt in idle, so notify users */
177				mark_tsc_unstable("TSC halts in idle");
178				tsc_marked_unstable = 1;
179			}
180			if (lapic_detected_unstable && !lapic_marked_unstable) {
181				int i;
182				/* LAPIC could halt in idle, so notify users */
183				for_each_online_cpu(i)
184					clockevents_notify(
185						CLOCK_EVT_NOTIFY_BROADCAST_ON,
186						&i);
187				lapic_marked_unstable = 1;
188			}
189			local_irq_disable();
190			cpu = smp_processor_id();
191			if (lapic_marked_unstable)
192				clockevents_notify(
193					CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
194			stop_critical_timings();
195
196			__monitor((void *)&current_thread_info()->flags, 0, 0);
197			smp_mb();
198			if (!need_resched())
199				__mwait(power_saving_mwait_eax, 1);
200
201			start_critical_timings();
202			if (lapic_marked_unstable)
203				clockevents_notify(
204					CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
205			local_irq_enable();
206
207			if (jiffies > expire_time) {
208				do_sleep = 1;
209				break;
210			}
211		}
212
213		/*
214		 * current sched_rt has threshold for rt task running time.
215		 * When a rt task uses 95% CPU time, the rt thread will be
216		 * scheduled out for 5% CPU time to not starve other tasks. But
217		 * the mechanism only works when all CPUs have RT task running,
218		 * as if one CPU hasn't RT task, RT task from other CPUs will
219		 * borrow CPU time from this CPU and cause RT task use > 95%
220		 * CPU time. To make 'avoid starvation' work, takes a nap here.
221		 */
222		if (do_sleep)
223			schedule_timeout_killable(HZ * idle_pct / 100);
224	}
225
226	exit_round_robin(tsk_index);
227	return 0;
228}
229
230static struct task_struct *ps_tsks[NR_CPUS];
231static unsigned int ps_tsk_num;
232static int create_power_saving_task(void)
233{
234	int rc = -ENOMEM;
235
236	ps_tsks[ps_tsk_num] = kthread_run(power_saving_thread,
237		(void *)(unsigned long)ps_tsk_num,
238		"power_saving/%d", ps_tsk_num);
239	rc = IS_ERR(ps_tsks[ps_tsk_num]) ? PTR_ERR(ps_tsks[ps_tsk_num]) : 0;
240	if (!rc)
241		ps_tsk_num++;
242	else
243		ps_tsks[ps_tsk_num] = NULL;
244
245	return rc;
246}
247
248static void destroy_power_saving_task(void)
249{
250	if (ps_tsk_num > 0) {
251		ps_tsk_num--;
252		kthread_stop(ps_tsks[ps_tsk_num]);
253		ps_tsks[ps_tsk_num] = NULL;
254	}
255}
256
257static void set_power_saving_task_num(unsigned int num)
258{
259	if (num > ps_tsk_num) {
260		while (ps_tsk_num < num) {
261			if (create_power_saving_task())
262				return;
263		}
264	} else if (num < ps_tsk_num) {
265		while (ps_tsk_num > num)
266			destroy_power_saving_task();
267	}
268}
269
270static void acpi_pad_idle_cpus(unsigned int num_cpus)
271{
272	get_online_cpus();
273
274	num_cpus = min_t(unsigned int, num_cpus, num_online_cpus());
275	set_power_saving_task_num(num_cpus);
276
277	put_online_cpus();
278}
279
280static uint32_t acpi_pad_idle_cpus_num(void)
281{
282	return ps_tsk_num;
283}
284
285static ssize_t acpi_pad_rrtime_store(struct device *dev,
286	struct device_attribute *attr, const char *buf, size_t count)
287{
288	unsigned long num;
289	if (strict_strtoul(buf, 0, &num))
290		return -EINVAL;
291	if (num < 1 || num >= 100)
292		return -EINVAL;
293	mutex_lock(&isolated_cpus_lock);
294	round_robin_time = num;
295	mutex_unlock(&isolated_cpus_lock);
296	return count;
297}
298
299static ssize_t acpi_pad_rrtime_show(struct device *dev,
300	struct device_attribute *attr, char *buf)
301{
302	return scnprintf(buf, PAGE_SIZE, "%d\n", round_robin_time);
303}
304static DEVICE_ATTR(rrtime, S_IRUGO|S_IWUSR,
305	acpi_pad_rrtime_show,
306	acpi_pad_rrtime_store);
307
308static ssize_t acpi_pad_idlepct_store(struct device *dev,
309	struct device_attribute *attr, const char *buf, size_t count)
310{
311	unsigned long num;
312	if (strict_strtoul(buf, 0, &num))
313		return -EINVAL;
314	if (num < 1 || num >= 100)
315		return -EINVAL;
316	mutex_lock(&isolated_cpus_lock);
317	idle_pct = num;
318	mutex_unlock(&isolated_cpus_lock);
319	return count;
320}
321
322static ssize_t acpi_pad_idlepct_show(struct device *dev,
323	struct device_attribute *attr, char *buf)
324{
325	return scnprintf(buf, PAGE_SIZE, "%d\n", idle_pct);
326}
327static DEVICE_ATTR(idlepct, S_IRUGO|S_IWUSR,
328	acpi_pad_idlepct_show,
329	acpi_pad_idlepct_store);
330
331static ssize_t acpi_pad_idlecpus_store(struct device *dev,
332	struct device_attribute *attr, const char *buf, size_t count)
333{
334	unsigned long num;
335	if (strict_strtoul(buf, 0, &num))
336		return -EINVAL;
337	mutex_lock(&isolated_cpus_lock);
338	acpi_pad_idle_cpus(num);
339	mutex_unlock(&isolated_cpus_lock);
340	return count;
341}
342
343static ssize_t acpi_pad_idlecpus_show(struct device *dev,
344	struct device_attribute *attr, char *buf)
345{
346	int n = 0;
347	n = cpumask_scnprintf(buf, PAGE_SIZE-2, to_cpumask(pad_busy_cpus_bits));
348	buf[n++] = '\n';
349	buf[n] = '\0';
350	return n;
351}
352static DEVICE_ATTR(idlecpus, S_IRUGO|S_IWUSR,
353	acpi_pad_idlecpus_show,
354	acpi_pad_idlecpus_store);
355
356static int acpi_pad_add_sysfs(struct acpi_device *device)
357{
358	int result;
359
360	result = device_create_file(&device->dev, &dev_attr_idlecpus);
361	if (result)
362		return -ENODEV;
363	result = device_create_file(&device->dev, &dev_attr_idlepct);
364	if (result) {
365		device_remove_file(&device->dev, &dev_attr_idlecpus);
366		return -ENODEV;
367	}
368	result = device_create_file(&device->dev, &dev_attr_rrtime);
369	if (result) {
370		device_remove_file(&device->dev, &dev_attr_idlecpus);
371		device_remove_file(&device->dev, &dev_attr_idlepct);
372		return -ENODEV;
373	}
374	return 0;
375}
376
377static void acpi_pad_remove_sysfs(struct acpi_device *device)
378{
379	device_remove_file(&device->dev, &dev_attr_idlecpus);
380	device_remove_file(&device->dev, &dev_attr_idlepct);
381	device_remove_file(&device->dev, &dev_attr_rrtime);
382}
383
384/*
385 * Query firmware how many CPUs should be idle
386 * return -1 on failure
387 */
388static int acpi_pad_pur(acpi_handle handle)
389{
390	struct acpi_buffer buffer = {ACPI_ALLOCATE_BUFFER, NULL};
391	union acpi_object *package;
392	int num = -1;
393
394	if (ACPI_FAILURE(acpi_evaluate_object(handle, "_PUR", NULL, &buffer)))
395		return num;
396
397	if (!buffer.length || !buffer.pointer)
398		return num;
399
400	package = buffer.pointer;
401
402	if (package->type == ACPI_TYPE_PACKAGE &&
403		package->package.count == 2 &&
404		package->package.elements[0].integer.value == 1) /* rev 1 */
405
406		num = package->package.elements[1].integer.value;
407
408	kfree(buffer.pointer);
409	return num;
410}
411
412/* Notify firmware how many CPUs are idle */
413static void acpi_pad_ost(acpi_handle handle, int stat,
414	uint32_t idle_cpus)
415{
416	union acpi_object params[3] = {
417		{.type = ACPI_TYPE_INTEGER,},
418		{.type = ACPI_TYPE_INTEGER,},
419		{.type = ACPI_TYPE_BUFFER,},
420	};
421	struct acpi_object_list arg_list = {3, params};
422
423	params[0].integer.value = ACPI_PROCESSOR_AGGREGATOR_NOTIFY;
424	params[1].integer.value =  stat;
425	params[2].buffer.length = 4;
426	params[2].buffer.pointer = (void *)&idle_cpus;
427	acpi_evaluate_object(handle, "_OST", &arg_list, NULL);
428}
429
430static void acpi_pad_handle_notify(acpi_handle handle)
431{
432	int num_cpus;
433	uint32_t idle_cpus;
434
435	mutex_lock(&isolated_cpus_lock);
436	num_cpus = acpi_pad_pur(handle);
437	if (num_cpus < 0) {
438		mutex_unlock(&isolated_cpus_lock);
439		return;
440	}
441	acpi_pad_idle_cpus(num_cpus);
442	idle_cpus = acpi_pad_idle_cpus_num();
443	acpi_pad_ost(handle, 0, idle_cpus);
444	mutex_unlock(&isolated_cpus_lock);
445}
446
447static void acpi_pad_notify(acpi_handle handle, u32 event,
448	void *data)
449{
450	struct acpi_device *device = data;
451
452	switch (event) {
453	case ACPI_PROCESSOR_AGGREGATOR_NOTIFY:
454		acpi_pad_handle_notify(handle);
455		acpi_bus_generate_proc_event(device, event, 0);
456		acpi_bus_generate_netlink_event(device->pnp.device_class,
457			dev_name(&device->dev), event, 0);
458		break;
459	default:
460		printk(KERN_WARNING "Unsupported event [0x%x]\n", event);
461		break;
462	}
463}
464
465static int acpi_pad_add(struct acpi_device *device)
466{
467	acpi_status status;
468
469	strcpy(acpi_device_name(device), ACPI_PROCESSOR_AGGREGATOR_DEVICE_NAME);
470	strcpy(acpi_device_class(device), ACPI_PROCESSOR_AGGREGATOR_CLASS);
471
472	if (acpi_pad_add_sysfs(device))
473		return -ENODEV;
474
475	status = acpi_install_notify_handler(device->handle,
476		ACPI_DEVICE_NOTIFY, acpi_pad_notify, device);
477	if (ACPI_FAILURE(status)) {
478		acpi_pad_remove_sysfs(device);
479		return -ENODEV;
480	}
481
482	return 0;
483}
484
485static int acpi_pad_remove(struct acpi_device *device,
486	int type)
487{
488	mutex_lock(&isolated_cpus_lock);
489	acpi_pad_idle_cpus(0);
490	mutex_unlock(&isolated_cpus_lock);
491
492	acpi_remove_notify_handler(device->handle,
493		ACPI_DEVICE_NOTIFY, acpi_pad_notify);
494	acpi_pad_remove_sysfs(device);
495	return 0;
496}
497
498static const struct acpi_device_id pad_device_ids[] = {
499	{"ACPI000C", 0},
500	{"", 0},
501};
502MODULE_DEVICE_TABLE(acpi, pad_device_ids);
503
504static struct acpi_driver acpi_pad_driver = {
505	.name = "processor_aggregator",
506	.class = ACPI_PROCESSOR_AGGREGATOR_CLASS,
507	.ids = pad_device_ids,
508	.ops = {
509		.add = acpi_pad_add,
510		.remove = acpi_pad_remove,
511	},
512};
513
514static int __init acpi_pad_init(void)
515{
516	power_saving_mwait_init();
517	if (power_saving_mwait_eax == 0)
518		return -EINVAL;
519
520	return acpi_bus_register_driver(&acpi_pad_driver);
521}
522
523static void __exit acpi_pad_exit(void)
524{
525	acpi_bus_unregister_driver(&acpi_pad_driver);
526}
527
528module_init(acpi_pad_init);
529module_exit(acpi_pad_exit);
530MODULE_AUTHOR("Shaohua Li<shaohua.li@intel.com>");
531MODULE_DESCRIPTION("ACPI Processor Aggregator Driver");
532MODULE_LICENSE("GPL");
533