Shaohua Li | 8e0af51 | 2009-07-27 18:11:02 -0400 | [diff] [blame] | 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; |
| 103 | unsigned long min_weight = -1, preferred_cpu; |
| 104 | |
| 105 | if (!alloc_cpumask_var(&tmp, GFP_KERNEL)) |
| 106 | return; |
| 107 | |
| 108 | mutex_lock(&isolated_cpus_lock); |
| 109 | cpumask_clear(tmp); |
| 110 | for_each_cpu(cpu, pad_busy_cpus) |
| 111 | cpumask_or(tmp, tmp, topology_thread_cpumask(cpu)); |
| 112 | cpumask_andnot(tmp, cpu_online_mask, tmp); |
| 113 | /* avoid HT sibilings if possible */ |
| 114 | if (cpumask_empty(tmp)) |
| 115 | cpumask_andnot(tmp, cpu_online_mask, pad_busy_cpus); |
| 116 | if (cpumask_empty(tmp)) { |
| 117 | mutex_unlock(&isolated_cpus_lock); |
| 118 | return; |
| 119 | } |
| 120 | for_each_cpu(cpu, tmp) { |
| 121 | if (cpu_weight[cpu] < min_weight) { |
| 122 | min_weight = cpu_weight[cpu]; |
| 123 | preferred_cpu = cpu; |
| 124 | } |
| 125 | } |
| 126 | |
| 127 | if (tsk_in_cpu[tsk_index] != -1) |
| 128 | cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus); |
| 129 | tsk_in_cpu[tsk_index] = preferred_cpu; |
| 130 | cpumask_set_cpu(preferred_cpu, pad_busy_cpus); |
| 131 | cpu_weight[preferred_cpu]++; |
| 132 | mutex_unlock(&isolated_cpus_lock); |
| 133 | |
| 134 | set_cpus_allowed_ptr(current, cpumask_of(preferred_cpu)); |
| 135 | } |
| 136 | |
| 137 | static void exit_round_robin(unsigned int tsk_index) |
| 138 | { |
| 139 | struct cpumask *pad_busy_cpus = to_cpumask(pad_busy_cpus_bits); |
| 140 | cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus); |
| 141 | tsk_in_cpu[tsk_index] = -1; |
| 142 | } |
| 143 | |
| 144 | static unsigned int idle_pct = 5; /* percentage */ |
| 145 | static unsigned int round_robin_time = 10; /* second */ |
| 146 | static int power_saving_thread(void *data) |
| 147 | { |
| 148 | struct sched_param param = {.sched_priority = 1}; |
| 149 | int do_sleep; |
| 150 | unsigned int tsk_index = (unsigned long)data; |
| 151 | u64 last_jiffies = 0; |
| 152 | |
| 153 | sched_setscheduler(current, SCHED_RR, ¶m); |
| 154 | |
| 155 | while (!kthread_should_stop()) { |
| 156 | int cpu; |
| 157 | u64 expire_time; |
| 158 | |
| 159 | try_to_freeze(); |
| 160 | |
| 161 | /* round robin to cpus */ |
| 162 | if (last_jiffies + round_robin_time * HZ < jiffies) { |
| 163 | last_jiffies = jiffies; |
| 164 | round_robin_cpu(tsk_index); |
| 165 | } |
| 166 | |
| 167 | do_sleep = 0; |
| 168 | |
| 169 | current_thread_info()->status &= ~TS_POLLING; |
| 170 | /* |
| 171 | * TS_POLLING-cleared state must be visible before we test |
| 172 | * NEED_RESCHED: |
| 173 | */ |
| 174 | smp_mb(); |
| 175 | |
| 176 | expire_time = jiffies + HZ * (100 - idle_pct) / 100; |
| 177 | |
| 178 | while (!need_resched()) { |
| 179 | local_irq_disable(); |
| 180 | cpu = smp_processor_id(); |
| 181 | clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, |
| 182 | &cpu); |
| 183 | stop_critical_timings(); |
| 184 | |
| 185 | __monitor((void *)¤t_thread_info()->flags, 0, 0); |
| 186 | smp_mb(); |
| 187 | if (!need_resched()) |
| 188 | __mwait(power_saving_mwait_eax, 1); |
| 189 | |
| 190 | start_critical_timings(); |
| 191 | clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, |
| 192 | &cpu); |
| 193 | local_irq_enable(); |
| 194 | |
| 195 | if (jiffies > expire_time) { |
| 196 | do_sleep = 1; |
| 197 | break; |
| 198 | } |
| 199 | } |
| 200 | |
| 201 | current_thread_info()->status |= TS_POLLING; |
| 202 | |
| 203 | /* |
| 204 | * current sched_rt has threshold for rt task running time. |
| 205 | * When a rt task uses 95% CPU time, the rt thread will be |
| 206 | * scheduled out for 5% CPU time to not starve other tasks. But |
| 207 | * the mechanism only works when all CPUs have RT task running, |
| 208 | * as if one CPU hasn't RT task, RT task from other CPUs will |
| 209 | * borrow CPU time from this CPU and cause RT task use > 95% |
| 210 | * CPU time. To make 'avoid staration' work, takes a nap here. |
| 211 | */ |
| 212 | if (do_sleep) |
| 213 | schedule_timeout_killable(HZ * idle_pct / 100); |
| 214 | } |
| 215 | |
| 216 | exit_round_robin(tsk_index); |
| 217 | return 0; |
| 218 | } |
| 219 | |
| 220 | static struct task_struct *ps_tsks[NR_CPUS]; |
| 221 | static unsigned int ps_tsk_num; |
| 222 | static int create_power_saving_task(void) |
| 223 | { |
| 224 | ps_tsks[ps_tsk_num] = kthread_run(power_saving_thread, |
| 225 | (void *)(unsigned long)ps_tsk_num, |
| 226 | "power_saving/%d", ps_tsk_num); |
| 227 | if (ps_tsks[ps_tsk_num]) { |
| 228 | ps_tsk_num++; |
| 229 | return 0; |
| 230 | } |
| 231 | return -EINVAL; |
| 232 | } |
| 233 | |
| 234 | static void destroy_power_saving_task(void) |
| 235 | { |
| 236 | if (ps_tsk_num > 0) { |
| 237 | ps_tsk_num--; |
| 238 | kthread_stop(ps_tsks[ps_tsk_num]); |
| 239 | } |
| 240 | } |
| 241 | |
| 242 | static void set_power_saving_task_num(unsigned int num) |
| 243 | { |
| 244 | if (num > ps_tsk_num) { |
| 245 | while (ps_tsk_num < num) { |
| 246 | if (create_power_saving_task()) |
| 247 | return; |
| 248 | } |
| 249 | } else if (num < ps_tsk_num) { |
| 250 | while (ps_tsk_num > num) |
| 251 | destroy_power_saving_task(); |
| 252 | } |
| 253 | } |
| 254 | |
| 255 | static int acpi_pad_idle_cpus(unsigned int num_cpus) |
| 256 | { |
| 257 | get_online_cpus(); |
| 258 | |
| 259 | num_cpus = min_t(unsigned int, num_cpus, num_online_cpus()); |
| 260 | set_power_saving_task_num(num_cpus); |
| 261 | |
| 262 | put_online_cpus(); |
| 263 | return 0; |
| 264 | } |
| 265 | |
| 266 | static uint32_t acpi_pad_idle_cpus_num(void) |
| 267 | { |
| 268 | return ps_tsk_num; |
| 269 | } |
| 270 | |
| 271 | static ssize_t acpi_pad_rrtime_store(struct device *dev, |
| 272 | struct device_attribute *attr, const char *buf, size_t count) |
| 273 | { |
| 274 | unsigned long num; |
| 275 | if (strict_strtoul(buf, 0, &num)) |
| 276 | return -EINVAL; |
| 277 | if (num < 1 || num >= 100) |
| 278 | return -EINVAL; |
| 279 | mutex_lock(&isolated_cpus_lock); |
| 280 | round_robin_time = num; |
| 281 | mutex_unlock(&isolated_cpus_lock); |
| 282 | return count; |
| 283 | } |
| 284 | |
| 285 | static ssize_t acpi_pad_rrtime_show(struct device *dev, |
| 286 | struct device_attribute *attr, char *buf) |
| 287 | { |
| 288 | return scnprintf(buf, PAGE_SIZE, "%d", round_robin_time); |
| 289 | } |
| 290 | static DEVICE_ATTR(rrtime, S_IRUGO|S_IWUSR, |
| 291 | acpi_pad_rrtime_show, |
| 292 | acpi_pad_rrtime_store); |
| 293 | |
| 294 | static ssize_t acpi_pad_idlepct_store(struct device *dev, |
| 295 | struct device_attribute *attr, const char *buf, size_t count) |
| 296 | { |
| 297 | unsigned long num; |
| 298 | if (strict_strtoul(buf, 0, &num)) |
| 299 | return -EINVAL; |
| 300 | if (num < 1 || num >= 100) |
| 301 | return -EINVAL; |
| 302 | mutex_lock(&isolated_cpus_lock); |
| 303 | idle_pct = num; |
| 304 | mutex_unlock(&isolated_cpus_lock); |
| 305 | return count; |
| 306 | } |
| 307 | |
| 308 | static ssize_t acpi_pad_idlepct_show(struct device *dev, |
| 309 | struct device_attribute *attr, char *buf) |
| 310 | { |
| 311 | return scnprintf(buf, PAGE_SIZE, "%d", idle_pct); |
| 312 | } |
| 313 | static DEVICE_ATTR(idlepct, S_IRUGO|S_IWUSR, |
| 314 | acpi_pad_idlepct_show, |
| 315 | acpi_pad_idlepct_store); |
| 316 | |
| 317 | static ssize_t acpi_pad_idlecpus_store(struct device *dev, |
| 318 | struct device_attribute *attr, const char *buf, size_t count) |
| 319 | { |
| 320 | unsigned long num; |
| 321 | if (strict_strtoul(buf, 0, &num)) |
| 322 | return -EINVAL; |
| 323 | mutex_lock(&isolated_cpus_lock); |
| 324 | acpi_pad_idle_cpus(num); |
| 325 | mutex_unlock(&isolated_cpus_lock); |
| 326 | return count; |
| 327 | } |
| 328 | |
| 329 | static ssize_t acpi_pad_idlecpus_show(struct device *dev, |
| 330 | struct device_attribute *attr, char *buf) |
| 331 | { |
| 332 | return cpumask_scnprintf(buf, PAGE_SIZE, |
| 333 | to_cpumask(pad_busy_cpus_bits)); |
| 334 | } |
| 335 | static DEVICE_ATTR(idlecpus, S_IRUGO|S_IWUSR, |
| 336 | acpi_pad_idlecpus_show, |
| 337 | acpi_pad_idlecpus_store); |
| 338 | |
| 339 | static int acpi_pad_add_sysfs(struct acpi_device *device) |
| 340 | { |
| 341 | int result; |
| 342 | |
| 343 | result = device_create_file(&device->dev, &dev_attr_idlecpus); |
| 344 | if (result) |
| 345 | return -ENODEV; |
| 346 | result = device_create_file(&device->dev, &dev_attr_idlepct); |
| 347 | if (result) { |
| 348 | device_remove_file(&device->dev, &dev_attr_idlecpus); |
| 349 | return -ENODEV; |
| 350 | } |
| 351 | result = device_create_file(&device->dev, &dev_attr_rrtime); |
| 352 | if (result) { |
| 353 | device_remove_file(&device->dev, &dev_attr_idlecpus); |
| 354 | device_remove_file(&device->dev, &dev_attr_idlepct); |
| 355 | return -ENODEV; |
| 356 | } |
| 357 | return 0; |
| 358 | } |
| 359 | |
| 360 | static void acpi_pad_remove_sysfs(struct acpi_device *device) |
| 361 | { |
| 362 | device_remove_file(&device->dev, &dev_attr_idlecpus); |
| 363 | device_remove_file(&device->dev, &dev_attr_idlepct); |
| 364 | device_remove_file(&device->dev, &dev_attr_rrtime); |
| 365 | } |
| 366 | |
| 367 | /* Query firmware how many CPUs should be idle */ |
| 368 | static int acpi_pad_pur(acpi_handle handle, int *num_cpus) |
| 369 | { |
| 370 | struct acpi_buffer buffer = {ACPI_ALLOCATE_BUFFER, NULL}; |
| 371 | acpi_status status; |
| 372 | union acpi_object *package; |
| 373 | int rev, num, ret = -EINVAL; |
| 374 | |
| 375 | status = acpi_evaluate_object(handle, "_PUR", NULL, &buffer); |
| 376 | if (ACPI_FAILURE(status)) |
| 377 | return -EINVAL; |
| 378 | package = buffer.pointer; |
| 379 | if (package->type != ACPI_TYPE_PACKAGE || package->package.count != 2) |
| 380 | goto out; |
| 381 | rev = package->package.elements[0].integer.value; |
| 382 | num = package->package.elements[1].integer.value; |
| 383 | if (rev != 1) |
| 384 | goto out; |
| 385 | *num_cpus = num; |
| 386 | ret = 0; |
| 387 | out: |
| 388 | kfree(buffer.pointer); |
| 389 | return ret; |
| 390 | } |
| 391 | |
| 392 | /* Notify firmware how many CPUs are idle */ |
| 393 | static void acpi_pad_ost(acpi_handle handle, int stat, |
| 394 | uint32_t idle_cpus) |
| 395 | { |
| 396 | union acpi_object params[3] = { |
| 397 | {.type = ACPI_TYPE_INTEGER,}, |
| 398 | {.type = ACPI_TYPE_INTEGER,}, |
| 399 | {.type = ACPI_TYPE_BUFFER,}, |
| 400 | }; |
| 401 | struct acpi_object_list arg_list = {3, params}; |
| 402 | |
| 403 | params[0].integer.value = ACPI_PROCESSOR_AGGREGATOR_NOTIFY; |
| 404 | params[1].integer.value = stat; |
| 405 | params[2].buffer.length = 4; |
| 406 | params[2].buffer.pointer = (void *)&idle_cpus; |
| 407 | acpi_evaluate_object(handle, "_OST", &arg_list, NULL); |
| 408 | } |
| 409 | |
| 410 | static void acpi_pad_handle_notify(acpi_handle handle) |
| 411 | { |
| 412 | int num_cpus, ret; |
| 413 | uint32_t idle_cpus; |
| 414 | |
| 415 | mutex_lock(&isolated_cpus_lock); |
| 416 | if (acpi_pad_pur(handle, &num_cpus)) { |
| 417 | mutex_unlock(&isolated_cpus_lock); |
| 418 | return; |
| 419 | } |
| 420 | ret = acpi_pad_idle_cpus(num_cpus); |
| 421 | idle_cpus = acpi_pad_idle_cpus_num(); |
| 422 | if (!ret) |
| 423 | acpi_pad_ost(handle, 0, idle_cpus); |
| 424 | else |
| 425 | acpi_pad_ost(handle, 1, 0); |
| 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"); |