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