2 * acpi_pad.c ACPI Processor Aggregator Driver
4 * Copyright (c) 2009, Intel Corporation.
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.
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
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.
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>
34 #define ACPI_PROCESSOR_AGGREGATOR_CLASS "acpi_pad"
35 #define ACPI_PROCESSOR_AGGREGATOR_DEVICE_NAME "Processor Aggregator"
36 #define ACPI_PROCESSOR_AGGREGATOR_NOTIFY 0x80
37 static DEFINE_MUTEX(isolated_cpus_lock);
39 #define MWAIT_SUBSTATE_MASK (0xf)
40 #define MWAIT_CSTATE_MASK (0xf)
41 #define MWAIT_SUBSTATE_SIZE (4)
42 #define CPUID_MWAIT_LEAF (5)
43 #define CPUID5_ECX_EXTENSIONS_SUPPORTED (0x1)
44 #define CPUID5_ECX_INTERRUPT_BREAK (0x2)
45 static unsigned long power_saving_mwait_eax;
46 static void power_saving_mwait_init(void)
48 unsigned int eax, ebx, ecx, edx;
49 unsigned int highest_cstate = 0;
50 unsigned int highest_subcstate = 0;
53 if (!boot_cpu_has(X86_FEATURE_MWAIT))
55 if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
58 cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &edx);
60 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
61 !(ecx & CPUID5_ECX_INTERRUPT_BREAK))
64 edx >>= MWAIT_SUBSTATE_SIZE;
65 for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
66 if (edx & MWAIT_SUBSTATE_MASK) {
68 highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
71 power_saving_mwait_eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
72 (highest_subcstate - 1);
74 for_each_online_cpu(i)
75 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ON, &i);
77 #if defined(CONFIG_GENERIC_TIME) && defined(CONFIG_X86)
78 switch (boot_cpu_data.x86_vendor) {
80 case X86_VENDOR_INTEL:
82 * AMD Fam10h TSC will tick in all
83 * C/P/S0/S1 states when this bit is set.
85 if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
90 /* TSC could halt in idle, so notify users */
91 mark_tsc_unstable("TSC halts in idle");
96 static unsigned long cpu_weight[NR_CPUS];
97 static int tsk_in_cpu[NR_CPUS] = {[0 ... NR_CPUS-1] = -1};
98 static DECLARE_BITMAP(pad_busy_cpus_bits, NR_CPUS);
99 static void round_robin_cpu(unsigned int tsk_index)
101 struct cpumask *pad_busy_cpus = to_cpumask(pad_busy_cpus_bits);
104 unsigned long min_weight = -1;
105 unsigned long uninitialized_var(preferred_cpu);
107 if (!alloc_cpumask_var(&tmp, GFP_KERNEL))
110 mutex_lock(&isolated_cpus_lock);
112 for_each_cpu(cpu, pad_busy_cpus)
113 cpumask_or(tmp, tmp, topology_thread_cpumask(cpu));
114 cpumask_andnot(tmp, cpu_online_mask, tmp);
115 /* avoid HT sibilings if possible */
116 if (cpumask_empty(tmp))
117 cpumask_andnot(tmp, cpu_online_mask, pad_busy_cpus);
118 if (cpumask_empty(tmp)) {
119 mutex_unlock(&isolated_cpus_lock);
122 for_each_cpu(cpu, tmp) {
123 if (cpu_weight[cpu] < min_weight) {
124 min_weight = cpu_weight[cpu];
129 if (tsk_in_cpu[tsk_index] != -1)
130 cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus);
131 tsk_in_cpu[tsk_index] = preferred_cpu;
132 cpumask_set_cpu(preferred_cpu, pad_busy_cpus);
133 cpu_weight[preferred_cpu]++;
134 mutex_unlock(&isolated_cpus_lock);
136 set_cpus_allowed_ptr(current, cpumask_of(preferred_cpu));
139 static void exit_round_robin(unsigned int tsk_index)
141 struct cpumask *pad_busy_cpus = to_cpumask(pad_busy_cpus_bits);
142 cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus);
143 tsk_in_cpu[tsk_index] = -1;
146 static unsigned int idle_pct = 5; /* percentage */
147 static unsigned int round_robin_time = 10; /* second */
148 static int power_saving_thread(void *data)
150 struct sched_param param = {.sched_priority = 1};
152 unsigned int tsk_index = (unsigned long)data;
153 u64 last_jiffies = 0;
155 sched_setscheduler(current, SCHED_RR, ¶m);
157 while (!kthread_should_stop()) {
163 /* round robin to cpus */
164 if (last_jiffies + round_robin_time * HZ < jiffies) {
165 last_jiffies = jiffies;
166 round_robin_cpu(tsk_index);
171 expire_time = jiffies + HZ * (100 - idle_pct) / 100;
173 while (!need_resched()) {
175 cpu = smp_processor_id();
176 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER,
178 stop_critical_timings();
180 __monitor((void *)¤t_thread_info()->flags, 0, 0);
183 __mwait(power_saving_mwait_eax, 1);
185 start_critical_timings();
186 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT,
190 if (jiffies > expire_time) {
197 * current sched_rt has threshold for rt task running time.
198 * When a rt task uses 95% CPU time, the rt thread will be
199 * scheduled out for 5% CPU time to not starve other tasks. But
200 * the mechanism only works when all CPUs have RT task running,
201 * as if one CPU hasn't RT task, RT task from other CPUs will
202 * borrow CPU time from this CPU and cause RT task use > 95%
203 * CPU time. To make 'avoid starvation' work, takes a nap here.
206 schedule_timeout_killable(HZ * idle_pct / 100);
209 exit_round_robin(tsk_index);
213 static struct task_struct *ps_tsks[NR_CPUS];
214 static unsigned int ps_tsk_num;
215 static int create_power_saving_task(void)
219 ps_tsks[ps_tsk_num] = kthread_run(power_saving_thread,
220 (void *)(unsigned long)ps_tsk_num,
221 "power_saving/%d", ps_tsk_num);
222 rc = IS_ERR(ps_tsks[ps_tsk_num]) ? PTR_ERR(ps_tsks[ps_tsk_num]) : 0;
226 ps_tsks[ps_tsk_num] = NULL;
231 static void destroy_power_saving_task(void)
233 if (ps_tsk_num > 0) {
235 kthread_stop(ps_tsks[ps_tsk_num]);
236 ps_tsks[ps_tsk_num] = NULL;
240 static void set_power_saving_task_num(unsigned int num)
242 if (num > ps_tsk_num) {
243 while (ps_tsk_num < num) {
244 if (create_power_saving_task())
247 } else if (num < ps_tsk_num) {
248 while (ps_tsk_num > num)
249 destroy_power_saving_task();
253 static void acpi_pad_idle_cpus(unsigned int num_cpus)
257 num_cpus = min_t(unsigned int, num_cpus, num_online_cpus());
258 set_power_saving_task_num(num_cpus);
263 static uint32_t acpi_pad_idle_cpus_num(void)
268 static ssize_t acpi_pad_rrtime_store(struct device *dev,
269 struct device_attribute *attr, const char *buf, size_t count)
272 if (strict_strtoul(buf, 0, &num))
274 if (num < 1 || num >= 100)
276 mutex_lock(&isolated_cpus_lock);
277 round_robin_time = num;
278 mutex_unlock(&isolated_cpus_lock);
282 static ssize_t acpi_pad_rrtime_show(struct device *dev,
283 struct device_attribute *attr, char *buf)
285 return scnprintf(buf, PAGE_SIZE, "%d", round_robin_time);
287 static DEVICE_ATTR(rrtime, S_IRUGO|S_IWUSR,
288 acpi_pad_rrtime_show,
289 acpi_pad_rrtime_store);
291 static ssize_t acpi_pad_idlepct_store(struct device *dev,
292 struct device_attribute *attr, const char *buf, size_t count)
295 if (strict_strtoul(buf, 0, &num))
297 if (num < 1 || num >= 100)
299 mutex_lock(&isolated_cpus_lock);
301 mutex_unlock(&isolated_cpus_lock);
305 static ssize_t acpi_pad_idlepct_show(struct device *dev,
306 struct device_attribute *attr, char *buf)
308 return scnprintf(buf, PAGE_SIZE, "%d", idle_pct);
310 static DEVICE_ATTR(idlepct, S_IRUGO|S_IWUSR,
311 acpi_pad_idlepct_show,
312 acpi_pad_idlepct_store);
314 static ssize_t acpi_pad_idlecpus_store(struct device *dev,
315 struct device_attribute *attr, const char *buf, size_t count)
318 if (strict_strtoul(buf, 0, &num))
320 mutex_lock(&isolated_cpus_lock);
321 acpi_pad_idle_cpus(num);
322 mutex_unlock(&isolated_cpus_lock);
326 static ssize_t acpi_pad_idlecpus_show(struct device *dev,
327 struct device_attribute *attr, char *buf)
329 return cpumask_scnprintf(buf, PAGE_SIZE,
330 to_cpumask(pad_busy_cpus_bits));
332 static DEVICE_ATTR(idlecpus, S_IRUGO|S_IWUSR,
333 acpi_pad_idlecpus_show,
334 acpi_pad_idlecpus_store);
336 static int acpi_pad_add_sysfs(struct acpi_device *device)
340 result = device_create_file(&device->dev, &dev_attr_idlecpus);
343 result = device_create_file(&device->dev, &dev_attr_idlepct);
345 device_remove_file(&device->dev, &dev_attr_idlecpus);
348 result = device_create_file(&device->dev, &dev_attr_rrtime);
350 device_remove_file(&device->dev, &dev_attr_idlecpus);
351 device_remove_file(&device->dev, &dev_attr_idlepct);
357 static void acpi_pad_remove_sysfs(struct acpi_device *device)
359 device_remove_file(&device->dev, &dev_attr_idlecpus);
360 device_remove_file(&device->dev, &dev_attr_idlepct);
361 device_remove_file(&device->dev, &dev_attr_rrtime);
364 /* Query firmware how many CPUs should be idle */
365 static int acpi_pad_pur(acpi_handle handle, int *num_cpus)
367 struct acpi_buffer buffer = {ACPI_ALLOCATE_BUFFER, NULL};
368 union acpi_object *package;
369 int rev, num, ret = -EINVAL;
371 if (ACPI_FAILURE(acpi_evaluate_object(handle, "_PUR", NULL, &buffer)))
374 if (!buffer.length || !buffer.pointer)
377 package = buffer.pointer;
378 if (package->type != ACPI_TYPE_PACKAGE || package->package.count != 2)
380 rev = package->package.elements[0].integer.value;
381 num = package->package.elements[1].integer.value;
382 if (rev != 1 || num < 0)
387 kfree(buffer.pointer);
391 /* Notify firmware how many CPUs are idle */
392 static void acpi_pad_ost(acpi_handle handle, int stat,
395 union acpi_object params[3] = {
396 {.type = ACPI_TYPE_INTEGER,},
397 {.type = ACPI_TYPE_INTEGER,},
398 {.type = ACPI_TYPE_BUFFER,},
400 struct acpi_object_list arg_list = {3, params};
402 params[0].integer.value = ACPI_PROCESSOR_AGGREGATOR_NOTIFY;
403 params[1].integer.value = stat;
404 params[2].buffer.length = 4;
405 params[2].buffer.pointer = (void *)&idle_cpus;
406 acpi_evaluate_object(handle, "_OST", &arg_list, NULL);
409 static void acpi_pad_handle_notify(acpi_handle handle)
414 mutex_lock(&isolated_cpus_lock);
415 if (acpi_pad_pur(handle, &num_cpus)) {
416 mutex_unlock(&isolated_cpus_lock);
419 acpi_pad_idle_cpus(num_cpus);
420 idle_cpus = acpi_pad_idle_cpus_num();
421 acpi_pad_ost(handle, 0, idle_cpus);
422 mutex_unlock(&isolated_cpus_lock);
425 static void acpi_pad_notify(acpi_handle handle, u32 event,
428 struct acpi_device *device = data;
431 case ACPI_PROCESSOR_AGGREGATOR_NOTIFY:
432 acpi_pad_handle_notify(handle);
433 acpi_bus_generate_proc_event(device, event, 0);
434 acpi_bus_generate_netlink_event(device->pnp.device_class,
435 dev_name(&device->dev), event, 0);
438 printk(KERN_WARNING"Unsupported event [0x%x]\n", event);
443 static int acpi_pad_add(struct acpi_device *device)
447 strcpy(acpi_device_name(device), ACPI_PROCESSOR_AGGREGATOR_DEVICE_NAME);
448 strcpy(acpi_device_class(device), ACPI_PROCESSOR_AGGREGATOR_CLASS);
450 if (acpi_pad_add_sysfs(device))
453 status = acpi_install_notify_handler(device->handle,
454 ACPI_DEVICE_NOTIFY, acpi_pad_notify, device);
455 if (ACPI_FAILURE(status)) {
456 acpi_pad_remove_sysfs(device);
463 static int acpi_pad_remove(struct acpi_device *device,
466 mutex_lock(&isolated_cpus_lock);
467 acpi_pad_idle_cpus(0);
468 mutex_unlock(&isolated_cpus_lock);
470 acpi_remove_notify_handler(device->handle,
471 ACPI_DEVICE_NOTIFY, acpi_pad_notify);
472 acpi_pad_remove_sysfs(device);
476 static const struct acpi_device_id pad_device_ids[] = {
480 MODULE_DEVICE_TABLE(acpi, pad_device_ids);
482 static struct acpi_driver acpi_pad_driver = {
483 .name = "processor_aggregator",
484 .class = ACPI_PROCESSOR_AGGREGATOR_CLASS,
485 .ids = pad_device_ids,
488 .remove = acpi_pad_remove,
492 static int __init acpi_pad_init(void)
494 power_saving_mwait_init();
495 if (power_saving_mwait_eax == 0)
498 return acpi_bus_register_driver(&acpi_pad_driver);
501 static void __exit acpi_pad_exit(void)
503 acpi_bus_unregister_driver(&acpi_pad_driver);
506 module_init(acpi_pad_init);
507 module_exit(acpi_pad_exit);
508 MODULE_AUTHOR("Shaohua Li<shaohua.li@intel.com>");
509 MODULE_DESCRIPTION("ACPI Processor Aggregator Driver");
510 MODULE_LICENSE("GPL");