Commit | Line | Data |
---|---|---|
8e0af514 SL |
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 <acpi/acpi_bus.h> | |
31 | #include <acpi/acpi_drivers.h> | |
32 | ||
33 | #define ACPI_PROCESSOR_AGGREGATOR_CLASS "processor_aggregator" | |
34 | #define ACPI_PROCESSOR_AGGREGATOR_DEVICE_NAME "Processor Aggregator" | |
35 | #define ACPI_PROCESSOR_AGGREGATOR_NOTIFY 0x80 | |
36 | static DEFINE_MUTEX(isolated_cpus_lock); | |
37 | ||
38 | #define MWAIT_SUBSTATE_MASK (0xf) | |
39 | #define MWAIT_CSTATE_MASK (0xf) | |
40 | #define MWAIT_SUBSTATE_SIZE (4) | |
41 | #define CPUID_MWAIT_LEAF (5) | |
42 | #define CPUID5_ECX_EXTENSIONS_SUPPORTED (0x1) | |
43 | #define CPUID5_ECX_INTERRUPT_BREAK (0x2) | |
44 | static unsigned long power_saving_mwait_eax; | |
45 | static void power_saving_mwait_init(void) | |
46 | { | |
47 | unsigned int eax, ebx, ecx, edx; | |
48 | unsigned int highest_cstate = 0; | |
49 | unsigned int highest_subcstate = 0; | |
50 | int i; | |
51 | ||
52 | if (!boot_cpu_has(X86_FEATURE_MWAIT)) | |
53 | return; | |
54 | if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF) | |
55 | return; | |
56 | ||
57 | cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &edx); | |
58 | ||
59 | if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) || | |
60 | !(ecx & CPUID5_ECX_INTERRUPT_BREAK)) | |
61 | return; | |
62 | ||
63 | edx >>= MWAIT_SUBSTATE_SIZE; | |
64 | for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) { | |
65 | if (edx & MWAIT_SUBSTATE_MASK) { | |
66 | highest_cstate = i; | |
67 | highest_subcstate = edx & MWAIT_SUBSTATE_MASK; | |
68 | } | |
69 | } | |
70 | power_saving_mwait_eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) | | |
71 | (highest_subcstate - 1); | |
72 | ||
73 | for_each_online_cpu(i) | |
74 | clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ON, &i); | |
75 | ||
76 | #if defined(CONFIG_GENERIC_TIME) && 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 | return; | |
86 | ||
87 | /*FALL THROUGH*/ | |
88 | default: | |
89 | /* TSC could halt in idle, so notify users */ | |
90 | mark_tsc_unstable("TSC halts in idle"); | |
91 | } | |
92 | #endif | |
93 | } | |
94 | ||
95 | static unsigned long cpu_weight[NR_CPUS]; | |
96 | static int tsk_in_cpu[NR_CPUS] = {[0 ... NR_CPUS-1] = -1}; | |
97 | static DECLARE_BITMAP(pad_busy_cpus_bits, NR_CPUS); | |
98 | static void round_robin_cpu(unsigned int tsk_index) | |
99 | { | |
100 | struct cpumask *pad_busy_cpus = to_cpumask(pad_busy_cpus_bits); | |
101 | cpumask_var_t tmp; | |
102 | int cpu; | |
f67538f8 AM |
103 | unsigned long min_weight = -1; |
104 | unsigned long uninitialized_var(preferred_cpu); | |
8e0af514 SL |
105 | |
106 | if (!alloc_cpumask_var(&tmp, GFP_KERNEL)) | |
107 | return; | |
108 | ||
109 | mutex_lock(&isolated_cpus_lock); | |
110 | cpumask_clear(tmp); | |
111 | for_each_cpu(cpu, pad_busy_cpus) | |
112 | cpumask_or(tmp, tmp, topology_thread_cpumask(cpu)); | |
113 | cpumask_andnot(tmp, cpu_online_mask, tmp); | |
114 | /* avoid HT sibilings if possible */ | |
115 | if (cpumask_empty(tmp)) | |
116 | cpumask_andnot(tmp, cpu_online_mask, pad_busy_cpus); | |
117 | if (cpumask_empty(tmp)) { | |
118 | mutex_unlock(&isolated_cpus_lock); | |
119 | return; | |
120 | } | |
121 | for_each_cpu(cpu, tmp) { | |
122 | if (cpu_weight[cpu] < min_weight) { | |
123 | min_weight = cpu_weight[cpu]; | |
124 | preferred_cpu = cpu; | |
125 | } | |
126 | } | |
127 | ||
128 | if (tsk_in_cpu[tsk_index] != -1) | |
129 | cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus); | |
130 | tsk_in_cpu[tsk_index] = preferred_cpu; | |
131 | cpumask_set_cpu(preferred_cpu, pad_busy_cpus); | |
132 | cpu_weight[preferred_cpu]++; | |
133 | mutex_unlock(&isolated_cpus_lock); | |
134 | ||
135 | set_cpus_allowed_ptr(current, cpumask_of(preferred_cpu)); | |
136 | } | |
137 | ||
138 | static void exit_round_robin(unsigned int tsk_index) | |
139 | { | |
140 | struct cpumask *pad_busy_cpus = to_cpumask(pad_busy_cpus_bits); | |
141 | cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus); | |
142 | tsk_in_cpu[tsk_index] = -1; | |
143 | } | |
144 | ||
145 | static unsigned int idle_pct = 5; /* percentage */ | |
146 | static unsigned int round_robin_time = 10; /* second */ | |
147 | static int power_saving_thread(void *data) | |
148 | { | |
149 | struct sched_param param = {.sched_priority = 1}; | |
150 | int do_sleep; | |
151 | unsigned int tsk_index = (unsigned long)data; | |
152 | u64 last_jiffies = 0; | |
153 | ||
154 | sched_setscheduler(current, SCHED_RR, ¶m); | |
155 | ||
156 | while (!kthread_should_stop()) { | |
157 | int cpu; | |
158 | u64 expire_time; | |
159 | ||
160 | try_to_freeze(); | |
161 | ||
162 | /* round robin to cpus */ | |
163 | if (last_jiffies + round_robin_time * HZ < jiffies) { | |
164 | last_jiffies = jiffies; | |
165 | round_robin_cpu(tsk_index); | |
166 | } | |
167 | ||
168 | do_sleep = 0; | |
169 | ||
170 | current_thread_info()->status &= ~TS_POLLING; | |
171 | /* | |
172 | * TS_POLLING-cleared state must be visible before we test | |
173 | * NEED_RESCHED: | |
174 | */ | |
175 | smp_mb(); | |
176 | ||
177 | expire_time = jiffies + HZ * (100 - idle_pct) / 100; | |
178 | ||
179 | while (!need_resched()) { | |
180 | local_irq_disable(); | |
181 | cpu = smp_processor_id(); | |
182 | clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, | |
183 | &cpu); | |
184 | stop_critical_timings(); | |
185 | ||
186 | __monitor((void *)¤t_thread_info()->flags, 0, 0); | |
187 | smp_mb(); | |
188 | if (!need_resched()) | |
189 | __mwait(power_saving_mwait_eax, 1); | |
190 | ||
191 | start_critical_timings(); | |
192 | clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, | |
193 | &cpu); | |
194 | local_irq_enable(); | |
195 | ||
196 | if (jiffies > expire_time) { | |
197 | do_sleep = 1; | |
198 | break; | |
199 | } | |
200 | } | |
201 | ||
202 | current_thread_info()->status |= TS_POLLING; | |
203 | ||
204 | /* | |
205 | * current sched_rt has threshold for rt task running time. | |
206 | * When a rt task uses 95% CPU time, the rt thread will be | |
207 | * scheduled out for 5% CPU time to not starve other tasks. But | |
208 | * the mechanism only works when all CPUs have RT task running, | |
209 | * as if one CPU hasn't RT task, RT task from other CPUs will | |
210 | * borrow CPU time from this CPU and cause RT task use > 95% | |
3b8cb427 | 211 | * CPU time. To make 'avoid starvation' work, takes a nap here. |
8e0af514 SL |
212 | */ |
213 | if (do_sleep) | |
214 | schedule_timeout_killable(HZ * idle_pct / 100); | |
215 | } | |
216 | ||
217 | exit_round_robin(tsk_index); | |
218 | return 0; | |
219 | } | |
220 | ||
221 | static struct task_struct *ps_tsks[NR_CPUS]; | |
222 | static unsigned int ps_tsk_num; | |
223 | static int create_power_saving_task(void) | |
224 | { | |
3b8cb427 CG |
225 | int rc = -ENOMEM; |
226 | ||
8e0af514 SL |
227 | ps_tsks[ps_tsk_num] = kthread_run(power_saving_thread, |
228 | (void *)(unsigned long)ps_tsk_num, | |
229 | "power_saving/%d", ps_tsk_num); | |
3b8cb427 CG |
230 | rc = IS_ERR(ps_tsks[ps_tsk_num]) ? PTR_ERR(ps_tsks[ps_tsk_num]) : 0; |
231 | if (!rc) | |
8e0af514 | 232 | ps_tsk_num++; |
3b8cb427 CG |
233 | else |
234 | ps_tsks[ps_tsk_num] = NULL; | |
235 | ||
236 | return rc; | |
8e0af514 SL |
237 | } |
238 | ||
239 | static void destroy_power_saving_task(void) | |
240 | { | |
241 | if (ps_tsk_num > 0) { | |
242 | ps_tsk_num--; | |
243 | kthread_stop(ps_tsks[ps_tsk_num]); | |
3b8cb427 | 244 | ps_tsks[ps_tsk_num] = NULL; |
8e0af514 SL |
245 | } |
246 | } | |
247 | ||
248 | static void set_power_saving_task_num(unsigned int num) | |
249 | { | |
250 | if (num > ps_tsk_num) { | |
251 | while (ps_tsk_num < num) { | |
252 | if (create_power_saving_task()) | |
253 | return; | |
254 | } | |
255 | } else if (num < ps_tsk_num) { | |
256 | while (ps_tsk_num > num) | |
257 | destroy_power_saving_task(); | |
258 | } | |
259 | } | |
260 | ||
3b8cb427 | 261 | static void acpi_pad_idle_cpus(unsigned int num_cpus) |
8e0af514 SL |
262 | { |
263 | get_online_cpus(); | |
264 | ||
265 | num_cpus = min_t(unsigned int, num_cpus, num_online_cpus()); | |
266 | set_power_saving_task_num(num_cpus); | |
267 | ||
268 | put_online_cpus(); | |
8e0af514 SL |
269 | } |
270 | ||
271 | static uint32_t acpi_pad_idle_cpus_num(void) | |
272 | { | |
273 | return ps_tsk_num; | |
274 | } | |
275 | ||
276 | static ssize_t acpi_pad_rrtime_store(struct device *dev, | |
277 | struct device_attribute *attr, const char *buf, size_t count) | |
278 | { | |
279 | unsigned long num; | |
280 | if (strict_strtoul(buf, 0, &num)) | |
281 | return -EINVAL; | |
282 | if (num < 1 || num >= 100) | |
283 | return -EINVAL; | |
284 | mutex_lock(&isolated_cpus_lock); | |
285 | round_robin_time = num; | |
286 | mutex_unlock(&isolated_cpus_lock); | |
287 | return count; | |
288 | } | |
289 | ||
290 | static ssize_t acpi_pad_rrtime_show(struct device *dev, | |
291 | struct device_attribute *attr, char *buf) | |
292 | { | |
293 | return scnprintf(buf, PAGE_SIZE, "%d", round_robin_time); | |
294 | } | |
295 | static DEVICE_ATTR(rrtime, S_IRUGO|S_IWUSR, | |
296 | acpi_pad_rrtime_show, | |
297 | acpi_pad_rrtime_store); | |
298 | ||
299 | static ssize_t acpi_pad_idlepct_store(struct device *dev, | |
300 | struct device_attribute *attr, const char *buf, size_t count) | |
301 | { | |
302 | unsigned long num; | |
303 | if (strict_strtoul(buf, 0, &num)) | |
304 | return -EINVAL; | |
305 | if (num < 1 || num >= 100) | |
306 | return -EINVAL; | |
307 | mutex_lock(&isolated_cpus_lock); | |
308 | idle_pct = num; | |
309 | mutex_unlock(&isolated_cpus_lock); | |
310 | return count; | |
311 | } | |
312 | ||
313 | static ssize_t acpi_pad_idlepct_show(struct device *dev, | |
314 | struct device_attribute *attr, char *buf) | |
315 | { | |
316 | return scnprintf(buf, PAGE_SIZE, "%d", idle_pct); | |
317 | } | |
318 | static DEVICE_ATTR(idlepct, S_IRUGO|S_IWUSR, | |
319 | acpi_pad_idlepct_show, | |
320 | acpi_pad_idlepct_store); | |
321 | ||
322 | static ssize_t acpi_pad_idlecpus_store(struct device *dev, | |
323 | struct device_attribute *attr, const char *buf, size_t count) | |
324 | { | |
325 | unsigned long num; | |
326 | if (strict_strtoul(buf, 0, &num)) | |
327 | return -EINVAL; | |
328 | mutex_lock(&isolated_cpus_lock); | |
329 | acpi_pad_idle_cpus(num); | |
330 | mutex_unlock(&isolated_cpus_lock); | |
331 | return count; | |
332 | } | |
333 | ||
334 | static ssize_t acpi_pad_idlecpus_show(struct device *dev, | |
335 | struct device_attribute *attr, char *buf) | |
336 | { | |
337 | return cpumask_scnprintf(buf, PAGE_SIZE, | |
338 | to_cpumask(pad_busy_cpus_bits)); | |
339 | } | |
340 | static DEVICE_ATTR(idlecpus, S_IRUGO|S_IWUSR, | |
341 | acpi_pad_idlecpus_show, | |
342 | acpi_pad_idlecpus_store); | |
343 | ||
344 | static int acpi_pad_add_sysfs(struct acpi_device *device) | |
345 | { | |
346 | int result; | |
347 | ||
348 | result = device_create_file(&device->dev, &dev_attr_idlecpus); | |
349 | if (result) | |
350 | return -ENODEV; | |
351 | result = device_create_file(&device->dev, &dev_attr_idlepct); | |
352 | if (result) { | |
353 | device_remove_file(&device->dev, &dev_attr_idlecpus); | |
354 | return -ENODEV; | |
355 | } | |
356 | result = device_create_file(&device->dev, &dev_attr_rrtime); | |
357 | if (result) { | |
358 | device_remove_file(&device->dev, &dev_attr_idlecpus); | |
359 | device_remove_file(&device->dev, &dev_attr_idlepct); | |
360 | return -ENODEV; | |
361 | } | |
362 | return 0; | |
363 | } | |
364 | ||
365 | static void acpi_pad_remove_sysfs(struct acpi_device *device) | |
366 | { | |
367 | device_remove_file(&device->dev, &dev_attr_idlecpus); | |
368 | device_remove_file(&device->dev, &dev_attr_idlepct); | |
369 | device_remove_file(&device->dev, &dev_attr_rrtime); | |
370 | } | |
371 | ||
372 | /* Query firmware how many CPUs should be idle */ | |
373 | static int acpi_pad_pur(acpi_handle handle, int *num_cpus) | |
374 | { | |
375 | struct acpi_buffer buffer = {ACPI_ALLOCATE_BUFFER, NULL}; | |
8e0af514 SL |
376 | union acpi_object *package; |
377 | int rev, num, ret = -EINVAL; | |
378 | ||
3b8cb427 | 379 | if (ACPI_FAILURE(acpi_evaluate_object(handle, "_PUR", NULL, &buffer))) |
8e0af514 | 380 | return -EINVAL; |
3b8cb427 CG |
381 | |
382 | if (!buffer.length || !buffer.pointer) | |
383 | return -EINVAL; | |
384 | ||
8e0af514 SL |
385 | package = buffer.pointer; |
386 | if (package->type != ACPI_TYPE_PACKAGE || package->package.count != 2) | |
387 | goto out; | |
388 | rev = package->package.elements[0].integer.value; | |
389 | num = package->package.elements[1].integer.value; | |
3b8cb427 | 390 | if (rev != 1 || num < 0) |
8e0af514 SL |
391 | goto out; |
392 | *num_cpus = num; | |
393 | ret = 0; | |
394 | out: | |
395 | kfree(buffer.pointer); | |
396 | return ret; | |
397 | } | |
398 | ||
399 | /* Notify firmware how many CPUs are idle */ | |
400 | static void acpi_pad_ost(acpi_handle handle, int stat, | |
401 | uint32_t idle_cpus) | |
402 | { | |
403 | union acpi_object params[3] = { | |
404 | {.type = ACPI_TYPE_INTEGER,}, | |
405 | {.type = ACPI_TYPE_INTEGER,}, | |
406 | {.type = ACPI_TYPE_BUFFER,}, | |
407 | }; | |
408 | struct acpi_object_list arg_list = {3, params}; | |
409 | ||
410 | params[0].integer.value = ACPI_PROCESSOR_AGGREGATOR_NOTIFY; | |
411 | params[1].integer.value = stat; | |
412 | params[2].buffer.length = 4; | |
413 | params[2].buffer.pointer = (void *)&idle_cpus; | |
414 | acpi_evaluate_object(handle, "_OST", &arg_list, NULL); | |
415 | } | |
416 | ||
417 | static void acpi_pad_handle_notify(acpi_handle handle) | |
418 | { | |
3b8cb427 | 419 | int num_cpus; |
8e0af514 SL |
420 | uint32_t idle_cpus; |
421 | ||
422 | mutex_lock(&isolated_cpus_lock); | |
423 | if (acpi_pad_pur(handle, &num_cpus)) { | |
424 | mutex_unlock(&isolated_cpus_lock); | |
425 | return; | |
426 | } | |
3b8cb427 | 427 | acpi_pad_idle_cpus(num_cpus); |
8e0af514 | 428 | idle_cpus = acpi_pad_idle_cpus_num(); |
3b8cb427 | 429 | acpi_pad_ost(handle, 0, idle_cpus); |
8e0af514 SL |
430 | mutex_unlock(&isolated_cpus_lock); |
431 | } | |
432 | ||
433 | static void acpi_pad_notify(acpi_handle handle, u32 event, | |
434 | void *data) | |
435 | { | |
436 | struct acpi_device *device = data; | |
437 | ||
438 | switch (event) { | |
439 | case ACPI_PROCESSOR_AGGREGATOR_NOTIFY: | |
440 | acpi_pad_handle_notify(handle); | |
441 | acpi_bus_generate_proc_event(device, event, 0); | |
442 | acpi_bus_generate_netlink_event(device->pnp.device_class, | |
443 | dev_name(&device->dev), event, 0); | |
444 | break; | |
445 | default: | |
446 | printk(KERN_WARNING"Unsupported event [0x%x]\n", event); | |
447 | break; | |
448 | } | |
449 | } | |
450 | ||
451 | static int acpi_pad_add(struct acpi_device *device) | |
452 | { | |
453 | acpi_status status; | |
454 | ||
455 | strcpy(acpi_device_name(device), ACPI_PROCESSOR_AGGREGATOR_DEVICE_NAME); | |
456 | strcpy(acpi_device_class(device), ACPI_PROCESSOR_AGGREGATOR_CLASS); | |
457 | ||
458 | if (acpi_pad_add_sysfs(device)) | |
459 | return -ENODEV; | |
460 | ||
461 | status = acpi_install_notify_handler(device->handle, | |
462 | ACPI_DEVICE_NOTIFY, acpi_pad_notify, device); | |
463 | if (ACPI_FAILURE(status)) { | |
464 | acpi_pad_remove_sysfs(device); | |
465 | return -ENODEV; | |
466 | } | |
467 | ||
468 | return 0; | |
469 | } | |
470 | ||
471 | static int acpi_pad_remove(struct acpi_device *device, | |
472 | int type) | |
473 | { | |
474 | mutex_lock(&isolated_cpus_lock); | |
475 | acpi_pad_idle_cpus(0); | |
476 | mutex_unlock(&isolated_cpus_lock); | |
477 | ||
478 | acpi_remove_notify_handler(device->handle, | |
479 | ACPI_DEVICE_NOTIFY, acpi_pad_notify); | |
480 | acpi_pad_remove_sysfs(device); | |
481 | return 0; | |
482 | } | |
483 | ||
484 | static const struct acpi_device_id pad_device_ids[] = { | |
485 | {"ACPI000C", 0}, | |
486 | {"", 0}, | |
487 | }; | |
488 | MODULE_DEVICE_TABLE(acpi, pad_device_ids); | |
489 | ||
490 | static struct acpi_driver acpi_pad_driver = { | |
491 | .name = "processor_aggregator", | |
492 | .class = ACPI_PROCESSOR_AGGREGATOR_CLASS, | |
493 | .ids = pad_device_ids, | |
494 | .ops = { | |
495 | .add = acpi_pad_add, | |
496 | .remove = acpi_pad_remove, | |
497 | }, | |
498 | }; | |
499 | ||
500 | static int __init acpi_pad_init(void) | |
501 | { | |
502 | power_saving_mwait_init(); | |
503 | if (power_saving_mwait_eax == 0) | |
504 | return -EINVAL; | |
505 | ||
506 | return acpi_bus_register_driver(&acpi_pad_driver); | |
507 | } | |
508 | ||
509 | static void __exit acpi_pad_exit(void) | |
510 | { | |
511 | acpi_bus_unregister_driver(&acpi_pad_driver); | |
512 | } | |
513 | ||
514 | module_init(acpi_pad_init); | |
515 | module_exit(acpi_pad_exit); | |
516 | MODULE_AUTHOR("Shaohua Li<shaohua.li@intel.com>"); | |
517 | MODULE_DESCRIPTION("ACPI Processor Aggregator Driver"); | |
518 | MODULE_LICENSE("GPL"); |