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                 unsigned long expire_time;
 160
 161                 try_to_freeze();
 162
 163                 /* round robin to cpus */
 164                 expire_time = last_jiffies + round_robin_time * HZ;
 165                 if (time_before(expire_time, jiffies)) {
 166                         last_jiffies = jiffies;
 167                         round_robin_cpu(tsk_index);
 168                 }
 169
 170                 do_sleep = 0;
 171
 172                 expire_time = jiffies + HZ * (100 - idle_pct) / 100;
 173
 174                 while (!need_resched()) {
 175                         if (tsc_detected_unstable && !tsc_marked_unstable) {
 176                                 /* TSC could halt in idle, so notify users */
 177                                 mark_tsc_unstable("TSC halts in idle");
 178                                 tsc_marked_unstable = 1;
 179                         }
 180                         if (lapic_detected_unstable && !lapic_marked_unstable) {
 181                                 int i;
 182                                 /* LAPIC could halt in idle, so notify users */
 183                                 for_each_online_cpu(i)
 184                                         clockevents_notify(
 185                                                 CLOCK_EVT_NOTIFY_BROADCAST_ON,
 186                                                 &i);
 187                                 lapic_marked_unstable = 1;
 188                         }
 189                         local_irq_disable();
 190                         cpu = smp_processor_id();
 191                         if (lapic_marked_unstable)
 192                                 clockevents_notify(
 193                                         CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
 194                         stop_critical_timings();
 195
 196                         mwait_idle_with_hints(power_saving_mwait_eax, 1);
 197
 198                         start_critical_timings();
 199                         if (lapic_marked_unstable)
 200                                 clockevents_notify(
 201                                         CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
 202                         local_irq_enable();
 203
 204                         if (time_before(expire_time, jiffies)) {
 205                                 do_sleep = 1;
 206                                 break;
 207                         }
 208                 }
 209
 210                 /*
 211                  * current sched_rt has threshold for rt task running time.
 212                  * When a rt task uses 95% CPU time, the rt thread will be
 213                  * scheduled out for 5% CPU time to not starve other tasks. But
 214                  * the mechanism only works when all CPUs have RT task running,
 215                  * as if one CPU hasn't RT task, RT task from other CPUs will
 216                  * borrow CPU time from this CPU and cause RT task use > 95%
 217                  * CPU time. To make 'avoid starvation' work, takes a nap here.
 218                  */
 219                 if (unlikely(do_sleep))
 220                         schedule_timeout_killable(HZ * idle_pct / 100);
 221
 222                 /* If an external event has set the need_resched flag, then
 223                  * we need to deal with it, or this loop will continue to
 224                  * spin without calling __mwait().
 225                  */
 226                 if (unlikely(need_resched()))
 227                         schedule();
 228         }
 229
 230         exit_round_robin(tsk_index);
 231         return 0;
 232 }
 233
 234 static struct task_struct *ps_tsks[NR_CPUS];
 235 static unsigned int ps_tsk_num;
 236 static int create_power_saving_task(void)
 237 {
 238         int rc;
 239
 240         ps_tsks[ps_tsk_num] = kthread_run(power_saving_thread,
 241                 (void *)(unsigned long)ps_tsk_num,
 242                 "acpi_pad/%d", ps_tsk_num);
 243
 244         if (IS_ERR(ps_tsks[ps_tsk_num])) {
 245                 rc = PTR_ERR(ps_tsks[ps_tsk_num]);
 246                 ps_tsks[ps_tsk_num] = NULL;
 247         } else {
 248                 rc = 0;
 249                 ps_tsk_num++;
 250         }
 251
 252         return rc;
 253 }
 254
 255 static void destroy_power_saving_task(void)
 256 {
 257         if (ps_tsk_num > 0) {
 258                 ps_tsk_num--;
 259                 kthread_stop(ps_tsks[ps_tsk_num]);
 260                 ps_tsks[ps_tsk_num] = NULL;
 261         }
 262 }
 263
 264 static void set_power_saving_task_num(unsigned int num)
 265 {
 266         if (num > ps_tsk_num) {
 267                 while (ps_tsk_num < num) {
 268                         if (create_power_saving_task())
 269                                 return;
 270                 }
 271         } else if (num < ps_tsk_num) {
 272                 while (ps_tsk_num > num)
 273                         destroy_power_saving_task();
 274         }
 275 }
 276
 277 static void acpi_pad_idle_cpus(unsigned int num_cpus)
 278 {
 279         get_online_cpus();
 280
 281         num_cpus = min_t(unsigned int, num_cpus, num_online_cpus());
 282         set_power_saving_task_num(num_cpus);
 283
 284         put_online_cpus();
 285 }
 286
 287 static uint32_t acpi_pad_idle_cpus_num(void)
 288 {
 289         return ps_tsk_num;
 290 }
 291
 292 static ssize_t acpi_pad_rrtime_store(struct device *dev,
 293         struct device_attribute *attr, const char *buf, size_t count)
 294 {
 295         unsigned long num;
 296         if (kstrtoul(buf, 0, &num))
 297                 return -EINVAL;
 298         if (num < 1 || num >= 100)
 299                 return -EINVAL;
 300         mutex_lock(&isolated_cpus_lock);
 301         round_robin_time = num;
 302         mutex_unlock(&isolated_cpus_lock);
 303         return count;
 304 }
 305
 306 static ssize_t acpi_pad_rrtime_show(struct device *dev,
 307         struct device_attribute *attr, char *buf)
 308 {
 309         return scnprintf(buf, PAGE_SIZE, "%d\n", round_robin_time);
 310 }
 311 static DEVICE_ATTR(rrtime, S_IRUGO|S_IWUSR,
 312         acpi_pad_rrtime_show,
 313         acpi_pad_rrtime_store);
 314
 315 static ssize_t acpi_pad_idlepct_store(struct device *dev,
 316         struct device_attribute *attr, const char *buf, size_t count)
 317 {
 318         unsigned long num;
 319         if (kstrtoul(buf, 0, &num))
 320                 return -EINVAL;
 321         if (num < 1 || num >= 100)
 322                 return -EINVAL;
 323         mutex_lock(&isolated_cpus_lock);
 324         idle_pct = num;
 325         mutex_unlock(&isolated_cpus_lock);
 326         return count;
 327 }
 328
 329 static ssize_t acpi_pad_idlepct_show(struct device *dev,
 330         struct device_attribute *attr, char *buf)
 331 {
 332         return scnprintf(buf, PAGE_SIZE, "%d\n", idle_pct);
 333 }
 334 static DEVICE_ATTR(idlepct, S_IRUGO|S_IWUSR,
 335         acpi_pad_idlepct_show,
 336         acpi_pad_idlepct_store);
 337
 338 static ssize_t acpi_pad_idlecpus_store(struct device *dev,
 339         struct device_attribute *attr, const char *buf, size_t count)
 340 {
 341         unsigned long num;
 342         if (kstrtoul(buf, 0, &num))
 343                 return -EINVAL;
 344         mutex_lock(&isolated_cpus_lock);
 345         acpi_pad_idle_cpus(num);
 346         mutex_unlock(&isolated_cpus_lock);
 347         return count;
 348 }
 349
 350 static ssize_t acpi_pad_idlecpus_show(struct device *dev,
 351         struct device_attribute *attr, char *buf)
 352 {
 353         int n = 0;
 354         n = cpumask_scnprintf(buf, PAGE_SIZE-2, to_cpumask(pad_busy_cpus_bits));
 355         buf[n++] = '\n';
 356         buf[n] = '\0';
 357         return n;
 358 }
 359 static DEVICE_ATTR(idlecpus, S_IRUGO|S_IWUSR,
 360         acpi_pad_idlecpus_show,
 361         acpi_pad_idlecpus_store);
 362
 363 static int acpi_pad_add_sysfs(struct acpi_device *device)
 364 {
 365         int result;
 366
 367         result = device_create_file(&device->dev, &dev_attr_idlecpus);
 368         if (result)
 369                 return -ENODEV;
 370         result = device_create_file(&device->dev, &dev_attr_idlepct);
 371         if (result) {
 372                 device_remove_file(&device->dev, &dev_attr_idlecpus);
 373                 return -ENODEV;
 374         }
 375         result = device_create_file(&device->dev, &dev_attr_rrtime);
 376         if (result) {
 377                 device_remove_file(&device->dev, &dev_attr_idlecpus);
 378                 device_remove_file(&device->dev, &dev_attr_idlepct);
 379                 return -ENODEV;
 380         }
 381         return 0;
 382 }
 383
 384 static void acpi_pad_remove_sysfs(struct acpi_device *device)
 385 {
 386         device_remove_file(&device->dev, &dev_attr_idlecpus);
 387         device_remove_file(&device->dev, &dev_attr_idlepct);
 388         device_remove_file(&device->dev, &dev_attr_rrtime);
 389 }
 390
 391 /*
 392  * Query firmware how many CPUs should be idle
 393  * return -1 on failure
 394  */
 395 static int acpi_pad_pur(acpi_handle handle)
 396 {
 397         struct acpi_buffer buffer = {ACPI_ALLOCATE_BUFFER, NULL};
 398         union acpi_object *package;
 399         int num = -1;
 400
 401         if (ACPI_FAILURE(acpi_evaluate_object(handle, "_PUR", NULL, &buffer)))
 402                 return num;
 403
 404         if (!buffer.length || !buffer.pointer)
 405                 return num;
 406
 407         package = buffer.pointer;
 408
 409         if (package->type == ACPI_TYPE_PACKAGE &&
 410                 package->package.count == 2 &&
 411                 package->package.elements[0].integer.value == 1) /* rev 1 */
 412
 413                 num = package->package.elements[1].integer.value;
 414
 415         kfree(buffer.pointer);
 416         return num;
 417 }
 418
 419 static void acpi_pad_handle_notify(acpi_handle handle)
 420 {
 421         int num_cpus;
 422         uint32_t idle_cpus;
 423         struct acpi_buffer param = {
 424                 .length = 4,
 425                 .pointer = (void *)&idle_cpus,
 426         };
 427
 428         mutex_lock(&isolated_cpus_lock);
 429         num_cpus = acpi_pad_pur(handle);
 430         if (num_cpus < 0) {
 431                 mutex_unlock(&isolated_cpus_lock);
 432                 return;
 433         }
 434         acpi_pad_idle_cpus(num_cpus);
 435         idle_cpus = acpi_pad_idle_cpus_num();
 436         acpi_evaluate_ost(handle, ACPI_PROCESSOR_AGGREGATOR_NOTIFY, 0, &param);
 437         mutex_unlock(&isolated_cpus_lock);
 438 }
 439
 440 static void acpi_pad_notify(acpi_handle handle, u32 event,
 441         void *data)
 442 {
 443         struct acpi_device *device = data;
 444
 445         switch (event) {
 446         case ACPI_PROCESSOR_AGGREGATOR_NOTIFY:
 447                 acpi_pad_handle_notify(handle);
 448                 acpi_bus_generate_netlink_event(device->pnp.device_class,
 449                         dev_name(&device->dev), event, 0);
 450                 break;
 451         default:
 452                 pr_warn("Unsupported event [0x%x]\n", event);
 453                 break;
 454         }
 455 }
 456
 457 static int acpi_pad_add(struct acpi_device *device)
 458 {
 459         acpi_status status;
 460
 461         strcpy(acpi_device_name(device), ACPI_PROCESSOR_AGGREGATOR_DEVICE_NAME);
 462         strcpy(acpi_device_class(device), ACPI_PROCESSOR_AGGREGATOR_CLASS);
 463
 464         if (acpi_pad_add_sysfs(device))
 465                 return -ENODEV;
 466
 467         status = acpi_install_notify_handler(device->handle,
 468                 ACPI_DEVICE_NOTIFY, acpi_pad_notify, device);
 469         if (ACPI_FAILURE(status)) {
 470                 acpi_pad_remove_sysfs(device);
 471                 return -ENODEV;
 472         }
 473
 474         return 0;
 475 }
 476
 477 static int acpi_pad_remove(struct acpi_device *device)
 478 {
 479         mutex_lock(&isolated_cpus_lock);
 480         acpi_pad_idle_cpus(0);
 481         mutex_unlock(&isolated_cpus_lock);
 482
 483         acpi_remove_notify_handler(device->handle,
 484                 ACPI_DEVICE_NOTIFY, acpi_pad_notify);
 485         acpi_pad_remove_sysfs(device);
 486         return 0;
 487 }
 488
 489 static const struct acpi_device_id pad_device_ids[] = {
 490         {"ACPI000C", 0},
 491         {"", 0},
 492 };
 493 MODULE_DEVICE_TABLE(acpi, pad_device_ids);
 494
 495 static struct acpi_driver acpi_pad_driver = {
 496         .name = "processor_aggregator",
 497         .class = ACPI_PROCESSOR_AGGREGATOR_CLASS,
 498         .ids = pad_device_ids,
 499         .ops = {
 500                 .add = acpi_pad_add,
 501                 .remove = acpi_pad_remove,
 502         },
 503 };
 504
 505 static int __init acpi_pad_init(void)
 506 {
 507         power_saving_mwait_init();
 508         if (power_saving_mwait_eax == 0)
 509                 return -EINVAL;
 510
 511         return acpi_bus_register_driver(&acpi_pad_driver);
 512 }
 513
 514 static void __exit acpi_pad_exit(void)
 515 {
 516         acpi_bus_unregister_driver(&acpi_pad_driver);
 517 }
 518
 519 module_init(acpi_pad_init);
 520 module_exit(acpi_pad_exit);
 521 MODULE_AUTHOR("Shaohua Li<shaohua.li@intel.com>");
 522 MODULE_DESCRIPTION("ACPI Processor Aggregator Driver");
 523 MODULE_LICENSE("GPL");