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