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                         __monitor((void *)&current_thread_info()->flags, 0, 0);
 196                         smp_mb();
 197                         if (!need_resched())
 198                                 __mwait(power_saving_mwait_eax, 1);
 199
 200                         start_critical_timings();
 201                         if (lapic_marked_unstable)
 202                                 clockevents_notify(
 203                                         CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
 204                         local_irq_enable();
 205
 206                         if (jiffies > expire_time) {
 207                                 do_sleep = 1;
 208                                 break;
 209                         }
 210                 }
 211
 212                 /*
 213                  * current sched_rt has threshold for rt task running time.
 214                  * When a rt task uses 95% CPU time, the rt thread will be
 215                  * scheduled out for 5% CPU time to not starve other tasks. But
 216                  * the mechanism only works when all CPUs have RT task running,
 217                  * as if one CPU hasn't RT task, RT task from other CPUs will
 218                  * borrow CPU time from this CPU and cause RT task use > 95%
 219                  * CPU time. To make 'avoid starvation' work, takes a nap here.
 220                  */
 221                 if (do_sleep)
 222                         schedule_timeout_killable(HZ * idle_pct / 100);
 223         }
 224
 225         exit_round_robin(tsk_index);
 226         return 0;
 227 }
 228
 229 static struct task_struct *ps_tsks[NR_CPUS];
 230 static unsigned int ps_tsk_num;
 231 static int create_power_saving_task(void)
 232 {
 233         int rc;
 234
 235         ps_tsks[ps_tsk_num] = kthread_run(power_saving_thread,
 236                 (void *)(unsigned long)ps_tsk_num,
 237                 "acpi_pad/%d", ps_tsk_num);
 238
 239         if (IS_ERR(ps_tsks[ps_tsk_num])) {
 240                 rc = PTR_ERR(ps_tsks[ps_tsk_num]);
 241                 ps_tsks[ps_tsk_num] = NULL;
 242         } else {
 243                 rc = 0;
 244                 ps_tsk_num++;
 245         }
 246
 247         return rc;
 248 }
 249
 250 static void destroy_power_saving_task(void)
 251 {
 252         if (ps_tsk_num > 0) {
 253                 ps_tsk_num--;
 254                 kthread_stop(ps_tsks[ps_tsk_num]);
 255                 ps_tsks[ps_tsk_num] = NULL;
 256         }
 257 }
 258
 259 static void set_power_saving_task_num(unsigned int num)
 260 {
 261         if (num > ps_tsk_num) {
 262                 while (ps_tsk_num < num) {
 263                         if (create_power_saving_task())
 264                                 return;
 265                 }
 266         } else if (num < ps_tsk_num) {
 267                 while (ps_tsk_num > num)
 268                         destroy_power_saving_task();
 269         }
 270 }
 271
 272 static void acpi_pad_idle_cpus(unsigned int num_cpus)
 273 {
 274         get_online_cpus();
 275
 276         num_cpus = min_t(unsigned int, num_cpus, num_online_cpus());
 277         set_power_saving_task_num(num_cpus);
 278
 279         put_online_cpus();
 280 }
 281
 282 static uint32_t acpi_pad_idle_cpus_num(void)
 283 {
 284         return ps_tsk_num;
 285 }
 286
 287 static ssize_t acpi_pad_rrtime_store(struct device *dev,
 288         struct device_attribute *attr, const char *buf, size_t count)
 289 {
 290         unsigned long num;
 291         if (kstrtoul(buf, 0, &num))
 292                 return -EINVAL;
 293         if (num < 1 || num >= 100)
 294                 return -EINVAL;
 295         mutex_lock(&isolated_cpus_lock);
 296         round_robin_time = num;
 297         mutex_unlock(&isolated_cpus_lock);
 298         return count;
 299 }
 300
 301 static ssize_t acpi_pad_rrtime_show(struct device *dev,
 302         struct device_attribute *attr, char *buf)
 303 {
 304         return scnprintf(buf, PAGE_SIZE, "%d\n", round_robin_time);
 305 }
 306 static DEVICE_ATTR(rrtime, S_IRUGO|S_IWUSR,
 307         acpi_pad_rrtime_show,
 308         acpi_pad_rrtime_store);
 309
 310 static ssize_t acpi_pad_idlepct_store(struct device *dev,
 311         struct device_attribute *attr, const char *buf, size_t count)
 312 {
 313         unsigned long num;
 314         if (kstrtoul(buf, 0, &num))
 315                 return -EINVAL;
 316         if (num < 1 || num >= 100)
 317                 return -EINVAL;
 318         mutex_lock(&isolated_cpus_lock);
 319         idle_pct = num;
 320         mutex_unlock(&isolated_cpus_lock);
 321         return count;
 322 }
 323
 324 static ssize_t acpi_pad_idlepct_show(struct device *dev,
 325         struct device_attribute *attr, char *buf)
 326 {
 327         return scnprintf(buf, PAGE_SIZE, "%d\n", idle_pct);
 328 }
 329 static DEVICE_ATTR(idlepct, S_IRUGO|S_IWUSR,
 330         acpi_pad_idlepct_show,
 331         acpi_pad_idlepct_store);
 332
 333 static ssize_t acpi_pad_idlecpus_store(struct device *dev,
 334         struct device_attribute *attr, const char *buf, size_t count)
 335 {
 336         unsigned long num;
 337         if (kstrtoul(buf, 0, &num))
 338                 return -EINVAL;
 339         mutex_lock(&isolated_cpus_lock);
 340         acpi_pad_idle_cpus(num);
 341         mutex_unlock(&isolated_cpus_lock);
 342         return count;
 343 }
 344
 345 static ssize_t acpi_pad_idlecpus_show(struct device *dev,
 346         struct device_attribute *attr, char *buf)
 347 {
 348         int n = 0;
 349         n = cpumask_scnprintf(buf, PAGE_SIZE-2, to_cpumask(pad_busy_cpus_bits));
 350         buf[n++] = '\n';
 351         buf[n] = '\0';
 352         return n;
 353 }
 354 static DEVICE_ATTR(idlecpus, S_IRUGO|S_IWUSR,
 355         acpi_pad_idlecpus_show,
 356         acpi_pad_idlecpus_store);
 357
 358 static int acpi_pad_add_sysfs(struct acpi_device *device)
 359 {
 360         int result;
 361
 362         result = device_create_file(&device->dev, &dev_attr_idlecpus);
 363         if (result)
 364                 return -ENODEV;
 365         result = device_create_file(&device->dev, &dev_attr_idlepct);
 366         if (result) {
 367                 device_remove_file(&device->dev, &dev_attr_idlecpus);
 368                 return -ENODEV;
 369         }
 370         result = device_create_file(&device->dev, &dev_attr_rrtime);
 371         if (result) {
 372                 device_remove_file(&device->dev, &dev_attr_idlecpus);
 373                 device_remove_file(&device->dev, &dev_attr_idlepct);
 374                 return -ENODEV;
 375         }
 376         return 0;
 377 }
 378
 379 static void acpi_pad_remove_sysfs(struct acpi_device *device)
 380 {
 381         device_remove_file(&device->dev, &dev_attr_idlecpus);
 382         device_remove_file(&device->dev, &dev_attr_idlepct);
 383         device_remove_file(&device->dev, &dev_attr_rrtime);
 384 }
 385
 386 /*
 387  * Query firmware how many CPUs should be idle
 388  * return -1 on failure
 389  */
 390 static int acpi_pad_pur(acpi_handle handle)
 391 {
 392         struct acpi_buffer buffer = {ACPI_ALLOCATE_BUFFER, NULL};
 393         union acpi_object *package;
 394         int num = -1;
 395
 396         if (ACPI_FAILURE(acpi_evaluate_object(handle, "_PUR", NULL, &buffer)))
 397                 return num;
 398
 399         if (!buffer.length || !buffer.pointer)
 400                 return num;
 401
 402         package = buffer.pointer;
 403
 404         if (package->type == ACPI_TYPE_PACKAGE &&
 405                 package->package.count == 2 &&
 406                 package->package.elements[0].integer.value == 1) /* rev 1 */
 407
 408                 num = package->package.elements[1].integer.value;
 409
 410         kfree(buffer.pointer);
 411         return num;
 412 }
 413
 414 /* Notify firmware how many CPUs are idle */
 415 static void acpi_pad_ost(acpi_handle handle, int stat,
 416         uint32_t idle_cpus)
 417 {
 418         union acpi_object params[3] = {
 419                 {.type = ACPI_TYPE_INTEGER,},
 420                 {.type = ACPI_TYPE_INTEGER,},
 421                 {.type = ACPI_TYPE_BUFFER,},
 422         };
 423         struct acpi_object_list arg_list = {3, params};
 424
 425         params[0].integer.value = ACPI_PROCESSOR_AGGREGATOR_NOTIFY;
 426         params[1].integer.value =  stat;
 427         params[2].buffer.length = 4;
 428         params[2].buffer.pointer = (void *)&idle_cpus;
 429         acpi_evaluate_object(handle, "_OST", &arg_list, NULL);
 430 }
 431
 432 static void acpi_pad_handle_notify(acpi_handle handle)
 433 {
 434         int num_cpus;
 435         uint32_t idle_cpus;
 436
 437         mutex_lock(&isolated_cpus_lock);
 438         num_cpus = acpi_pad_pur(handle);
 439         if (num_cpus < 0) {
 440                 mutex_unlock(&isolated_cpus_lock);
 441                 return;
 442         }
 443         acpi_pad_idle_cpus(num_cpus);
 444         idle_cpus = acpi_pad_idle_cpus_num();
 445         acpi_pad_ost(handle, 0, idle_cpus);
 446         mutex_unlock(&isolated_cpus_lock);
 447 }
 448
 449 static void acpi_pad_notify(acpi_handle handle, u32 event,
 450         void *data)
 451 {
 452         struct acpi_device *device = data;
 453
 454         switch (event) {
 455         case ACPI_PROCESSOR_AGGREGATOR_NOTIFY:
 456                 acpi_pad_handle_notify(handle);
 457                 acpi_bus_generate_netlink_event(device->pnp.device_class,
 458                         dev_name(&device->dev), event, 0);
 459                 break;
 460         default:
 461                 pr_warn("Unsupported event [0x%x]\n", event);
 462                 break;
 463         }
 464 }
 465
 466 static int acpi_pad_add(struct acpi_device *device)
 467 {
 468         acpi_status status;
 469
 470         strcpy(acpi_device_name(device), ACPI_PROCESSOR_AGGREGATOR_DEVICE_NAME);
 471         strcpy(acpi_device_class(device), ACPI_PROCESSOR_AGGREGATOR_CLASS);
 472
 473         if (acpi_pad_add_sysfs(device))
 474                 return -ENODEV;
 475
 476         status = acpi_install_notify_handler(device->handle,
 477                 ACPI_DEVICE_NOTIFY, acpi_pad_notify, device);
 478         if (ACPI_FAILURE(status)) {
 479                 acpi_pad_remove_sysfs(device);
 480                 return -ENODEV;
 481         }
 482
 483         return 0;
 484 }
 485
 486 static int acpi_pad_remove(struct acpi_device *device)
 487 {
 488         mutex_lock(&isolated_cpus_lock);
 489         acpi_pad_idle_cpus(0);
 490         mutex_unlock(&isolated_cpus_lock);
 491
 492         acpi_remove_notify_handler(device->handle,
 493                 ACPI_DEVICE_NOTIFY, acpi_pad_notify);
 494         acpi_pad_remove_sysfs(device);
 495         return 0;
 496 }
 497
 498 static const struct acpi_device_id pad_device_ids[] = {
 499         {"ACPI000C", 0},
 500         {"", 0},
 501 };
 502 MODULE_DEVICE_TABLE(acpi, pad_device_ids);
 503
 504 static struct acpi_driver acpi_pad_driver = {
 505         .name = "processor_aggregator",
 506         .class = ACPI_PROCESSOR_AGGREGATOR_CLASS,
 507         .ids = pad_device_ids,
 508         .ops = {
 509                 .add = acpi_pad_add,
 510                 .remove = acpi_pad_remove,
 511         },
 512 };
 513
 514 static int __init acpi_pad_init(void)
 515 {
 516         power_saving_mwait_init();
 517         if (power_saving_mwait_eax == 0)
 518                 return -EINVAL;
 519
 520         return acpi_bus_register_driver(&acpi_pad_driver);
 521 }
 522
 523 static void __exit acpi_pad_exit(void)
 524 {
 525         acpi_bus_unregister_driver(&acpi_pad_driver);
 526 }
 527
 528 module_init(acpi_pad_init);
 529 module_exit(acpi_pad_exit);
 530 MODULE_AUTHOR("Shaohua Li<shaohua.li@intel.com>");
 531 MODULE_DESCRIPTION("ACPI Processor Aggregator Driver");
 532 MODULE_LICENSE("GPL");