Version:  2.6.34 2.6.35 2.6.36 2.6.37 2.6.38 2.6.39 3.0 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 3.10 3.11 3.12 3.13 3.14

Linux/drivers/acpi/acpi_pad.c

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

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us