1 /* 2 * intel_idle.c - native hardware idle loop for modern Intel processors 3 * 4 * Copyright (c) 2010, Intel Corporation. 5 * Len Brown <len.brown@intel.com> 6 * 7 * This program is free software; you can redistribute it and/or modify it 8 * under the terms and conditions of the GNU General Public License, 9 * version 2, as published by the Free Software Foundation. 10 * 11 * This program is distributed in the hope it will be useful, but WITHOUT 12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 14 * more details. 15 * 16 * You should have received a copy of the GNU General Public License along with 17 * this program; if not, write to the Free Software Foundation, Inc., 18 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. 19 */ 20 21 /* 22 * intel_idle is a cpuidle driver that loads on specific Intel processors 23 * in lieu of the legacy ACPI processor_idle driver. The intent is to 24 * make Linux more efficient on these processors, as intel_idle knows 25 * more than ACPI, as well as make Linux more immune to ACPI BIOS bugs. 26 */ 27 28 /* 29 * Design Assumptions 30 * 31 * All CPUs have same idle states as boot CPU 32 * 33 * Chipset BM_STS (bus master status) bit is a NOP 34 * for preventing entry into deep C-stats 35 */ 36 37 /* 38 * Known limitations 39 * 40 * The driver currently initializes for_each_online_cpu() upon modprobe. 41 * It it unaware of subsequent processors hot-added to the system. 42 * This means that if you boot with maxcpus=n and later online 43 * processors above n, those processors will use C1 only. 44 * 45 * ACPI has a .suspend hack to turn off deep c-statees during suspend 46 * to avoid complications with the lapic timer workaround. 47 * Have not seen issues with suspend, but may need same workaround here. 48 * 49 * There is currently no kernel-based automatic probing/loading mechanism 50 * if the driver is built as a module. 51 */ 52 53 /* un-comment DEBUG to enable pr_debug() statements */ 54 #define DEBUG 55 56 #include <linux/kernel.h> 57 #include <linux/cpuidle.h> 58 #include <linux/clockchips.h> 59 #include <trace/events/power.h> 60 #include <linux/sched.h> 61 #include <linux/notifier.h> 62 #include <linux/cpu.h> 63 #include <linux/module.h> 64 #include <asm/cpu_device_id.h> 65 #include <asm/mwait.h> 66 #include <asm/msr.h> 67 68 #define INTEL_IDLE_VERSION "0.4" 69 #define PREFIX "intel_idle: " 70 71 static struct cpuidle_driver intel_idle_driver = { 72 .name = "intel_idle", 73 .owner = THIS_MODULE, 74 }; 75 /* intel_idle.max_cstate=0 disables driver */ 76 static int max_cstate = CPUIDLE_STATE_MAX - 1; 77 78 static unsigned int mwait_substates; 79 80 #define LAPIC_TIMER_ALWAYS_RELIABLE 0xFFFFFFFF 81 /* Reliable LAPIC Timer States, bit 1 for C1 etc. */ 82 static unsigned int lapic_timer_reliable_states = (1 << 1); /* Default to only C1 */ 83 84 struct idle_cpu { 85 struct cpuidle_state *state_table; 86 87 /* 88 * Hardware C-state auto-demotion may not always be optimal. 89 * Indicate which enable bits to clear here. 90 */ 91 unsigned long auto_demotion_disable_flags; 92 bool disable_promotion_to_c1e; 93 }; 94 95 static const struct idle_cpu *icpu; 96 static struct cpuidle_device __percpu *intel_idle_cpuidle_devices; 97 static int intel_idle(struct cpuidle_device *dev, 98 struct cpuidle_driver *drv, int index); 99 static int intel_idle_cpu_init(int cpu); 100 101 static struct cpuidle_state *cpuidle_state_table; 102 103 /* 104 * Set this flag for states where the HW flushes the TLB for us 105 * and so we don't need cross-calls to keep it consistent. 106 * If this flag is set, SW flushes the TLB, so even if the 107 * HW doesn't do the flushing, this flag is safe to use. 108 */ 109 #define CPUIDLE_FLAG_TLB_FLUSHED 0x10000 110 111 /* 112 * MWAIT takes an 8-bit "hint" in EAX "suggesting" 113 * the C-state (top nibble) and sub-state (bottom nibble) 114 * 0x00 means "MWAIT(C1)", 0x10 means "MWAIT(C2)" etc. 115 * 116 * We store the hint at the top of our "flags" for each state. 117 */ 118 #define flg2MWAIT(flags) (((flags) >> 24) & 0xFF) 119 #define MWAIT2flg(eax) ((eax & 0xFF) << 24) 120 121 /* 122 * States are indexed by the cstate number, 123 * which is also the index into the MWAIT hint array. 124 * Thus C0 is a dummy. 125 */ 126 static struct cpuidle_state nehalem_cstates[CPUIDLE_STATE_MAX] = { 127 { 128 .name = "C1-NHM", 129 .desc = "MWAIT 0x00", 130 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID, 131 .exit_latency = 3, 132 .target_residency = 6, 133 .enter = &intel_idle }, 134 { 135 .name = "C1E-NHM", 136 .desc = "MWAIT 0x01", 137 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_TIME_VALID, 138 .exit_latency = 10, 139 .target_residency = 20, 140 .enter = &intel_idle }, 141 { 142 .name = "C3-NHM", 143 .desc = "MWAIT 0x10", 144 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED, 145 .exit_latency = 20, 146 .target_residency = 80, 147 .enter = &intel_idle }, 148 { 149 .name = "C6-NHM", 150 .desc = "MWAIT 0x20", 151 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED, 152 .exit_latency = 200, 153 .target_residency = 800, 154 .enter = &intel_idle }, 155 { 156 .enter = NULL } 157 }; 158 159 static struct cpuidle_state snb_cstates[CPUIDLE_STATE_MAX] = { 160 { 161 .name = "C1-SNB", 162 .desc = "MWAIT 0x00", 163 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID, 164 .exit_latency = 2, 165 .target_residency = 2, 166 .enter = &intel_idle }, 167 { 168 .name = "C1E-SNB", 169 .desc = "MWAIT 0x01", 170 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_TIME_VALID, 171 .exit_latency = 10, 172 .target_residency = 20, 173 .enter = &intel_idle }, 174 { 175 .name = "C3-SNB", 176 .desc = "MWAIT 0x10", 177 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED, 178 .exit_latency = 80, 179 .target_residency = 211, 180 .enter = &intel_idle }, 181 { 182 .name = "C6-SNB", 183 .desc = "MWAIT 0x20", 184 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED, 185 .exit_latency = 104, 186 .target_residency = 345, 187 .enter = &intel_idle }, 188 { 189 .name = "C7-SNB", 190 .desc = "MWAIT 0x30", 191 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED, 192 .exit_latency = 109, 193 .target_residency = 345, 194 .enter = &intel_idle }, 195 { 196 .enter = NULL } 197 }; 198 199 static struct cpuidle_state ivb_cstates[CPUIDLE_STATE_MAX] = { 200 { 201 .name = "C1-IVB", 202 .desc = "MWAIT 0x00", 203 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID, 204 .exit_latency = 1, 205 .target_residency = 1, 206 .enter = &intel_idle }, 207 { 208 .name = "C1E-IVB", 209 .desc = "MWAIT 0x01", 210 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_TIME_VALID, 211 .exit_latency = 10, 212 .target_residency = 20, 213 .enter = &intel_idle }, 214 { 215 .name = "C3-IVB", 216 .desc = "MWAIT 0x10", 217 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED, 218 .exit_latency = 59, 219 .target_residency = 156, 220 .enter = &intel_idle }, 221 { 222 .name = "C6-IVB", 223 .desc = "MWAIT 0x20", 224 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED, 225 .exit_latency = 80, 226 .target_residency = 300, 227 .enter = &intel_idle }, 228 { 229 .name = "C7-IVB", 230 .desc = "MWAIT 0x30", 231 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED, 232 .exit_latency = 87, 233 .target_residency = 300, 234 .enter = &intel_idle }, 235 { 236 .enter = NULL } 237 }; 238 239 static struct cpuidle_state hsw_cstates[CPUIDLE_STATE_MAX] = { 240 { 241 .name = "C1-HSW", 242 .desc = "MWAIT 0x00", 243 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID, 244 .exit_latency = 2, 245 .target_residency = 2, 246 .enter = &intel_idle }, 247 { 248 .name = "C1E-HSW", 249 .desc = "MWAIT 0x01", 250 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_TIME_VALID, 251 .exit_latency = 10, 252 .target_residency = 20, 253 .enter = &intel_idle }, 254 { 255 .name = "C3-HSW", 256 .desc = "MWAIT 0x10", 257 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED, 258 .exit_latency = 33, 259 .target_residency = 100, 260 .enter = &intel_idle }, 261 { 262 .name = "C6-HSW", 263 .desc = "MWAIT 0x20", 264 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED, 265 .exit_latency = 133, 266 .target_residency = 400, 267 .enter = &intel_idle }, 268 { 269 .name = "C7s-HSW", 270 .desc = "MWAIT 0x32", 271 .flags = MWAIT2flg(0x32) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED, 272 .exit_latency = 166, 273 .target_residency = 500, 274 .enter = &intel_idle }, 275 { 276 .enter = NULL } 277 }; 278 279 static struct cpuidle_state atom_cstates[CPUIDLE_STATE_MAX] = { 280 { 281 .name = "C1E-ATM", 282 .desc = "MWAIT 0x00", 283 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID, 284 .exit_latency = 10, 285 .target_residency = 20, 286 .enter = &intel_idle }, 287 { 288 .name = "C2-ATM", 289 .desc = "MWAIT 0x10", 290 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID, 291 .exit_latency = 20, 292 .target_residency = 80, 293 .enter = &intel_idle }, 294 { 295 .name = "C4-ATM", 296 .desc = "MWAIT 0x30", 297 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED, 298 .exit_latency = 100, 299 .target_residency = 400, 300 .enter = &intel_idle }, 301 { 302 .name = "C6-ATM", 303 .desc = "MWAIT 0x52", 304 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED, 305 .exit_latency = 140, 306 .target_residency = 560, 307 .enter = &intel_idle }, 308 { 309 .enter = NULL } 310 }; 311 312 /** 313 * intel_idle 314 * @dev: cpuidle_device 315 * @drv: cpuidle driver 316 * @index: index of cpuidle state 317 * 318 * Must be called under local_irq_disable(). 319 */ 320 static int intel_idle(struct cpuidle_device *dev, 321 struct cpuidle_driver *drv, int index) 322 { 323 unsigned long ecx = 1; /* break on interrupt flag */ 324 struct cpuidle_state *state = &drv->states[index]; 325 unsigned long eax = flg2MWAIT(state->flags); 326 unsigned int cstate; 327 int cpu = smp_processor_id(); 328 329 cstate = (((eax) >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK) + 1; 330 331 /* 332 * leave_mm() to avoid costly and often unnecessary wakeups 333 * for flushing the user TLB's associated with the active mm. 334 */ 335 if (state->flags & CPUIDLE_FLAG_TLB_FLUSHED) 336 leave_mm(cpu); 337 338 if (!(lapic_timer_reliable_states & (1 << (cstate)))) 339 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu); 340 341 if (!need_resched()) { 342 343 __monitor((void *)¤t_thread_info()->flags, 0, 0); 344 smp_mb(); 345 if (!need_resched()) 346 __mwait(eax, ecx); 347 } 348 349 if (!(lapic_timer_reliable_states & (1 << (cstate)))) 350 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu); 351 352 return index; 353 } 354 355 static void __setup_broadcast_timer(void *arg) 356 { 357 unsigned long reason = (unsigned long)arg; 358 int cpu = smp_processor_id(); 359 360 reason = reason ? 361 CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF; 362 363 clockevents_notify(reason, &cpu); 364 } 365 366 static int cpu_hotplug_notify(struct notifier_block *n, 367 unsigned long action, void *hcpu) 368 { 369 int hotcpu = (unsigned long)hcpu; 370 struct cpuidle_device *dev; 371 372 switch (action & 0xf) { 373 case CPU_ONLINE: 374 375 if (lapic_timer_reliable_states != LAPIC_TIMER_ALWAYS_RELIABLE) 376 smp_call_function_single(hotcpu, __setup_broadcast_timer, 377 (void *)true, 1); 378 379 /* 380 * Some systems can hotplug a cpu at runtime after 381 * the kernel has booted, we have to initialize the 382 * driver in this case 383 */ 384 dev = per_cpu_ptr(intel_idle_cpuidle_devices, hotcpu); 385 if (!dev->registered) 386 intel_idle_cpu_init(hotcpu); 387 388 break; 389 } 390 return NOTIFY_OK; 391 } 392 393 static struct notifier_block cpu_hotplug_notifier = { 394 .notifier_call = cpu_hotplug_notify, 395 }; 396 397 static void auto_demotion_disable(void *dummy) 398 { 399 unsigned long long msr_bits; 400 401 rdmsrl(MSR_NHM_SNB_PKG_CST_CFG_CTL, msr_bits); 402 msr_bits &= ~(icpu->auto_demotion_disable_flags); 403 wrmsrl(MSR_NHM_SNB_PKG_CST_CFG_CTL, msr_bits); 404 } 405 static void c1e_promotion_disable(void *dummy) 406 { 407 unsigned long long msr_bits; 408 409 rdmsrl(MSR_IA32_POWER_CTL, msr_bits); 410 msr_bits &= ~0x2; 411 wrmsrl(MSR_IA32_POWER_CTL, msr_bits); 412 } 413 414 static const struct idle_cpu idle_cpu_nehalem = { 415 .state_table = nehalem_cstates, 416 .auto_demotion_disable_flags = NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE, 417 .disable_promotion_to_c1e = true, 418 }; 419 420 static const struct idle_cpu idle_cpu_atom = { 421 .state_table = atom_cstates, 422 }; 423 424 static const struct idle_cpu idle_cpu_lincroft = { 425 .state_table = atom_cstates, 426 .auto_demotion_disable_flags = ATM_LNC_C6_AUTO_DEMOTE, 427 }; 428 429 static const struct idle_cpu idle_cpu_snb = { 430 .state_table = snb_cstates, 431 .disable_promotion_to_c1e = true, 432 }; 433 434 static const struct idle_cpu idle_cpu_ivb = { 435 .state_table = ivb_cstates, 436 .disable_promotion_to_c1e = true, 437 }; 438 439 static const struct idle_cpu idle_cpu_hsw = { 440 .state_table = hsw_cstates, 441 .disable_promotion_to_c1e = true, 442 }; 443 444 #define ICPU(model, cpu) \ 445 { X86_VENDOR_INTEL, 6, model, X86_FEATURE_MWAIT, (unsigned long)&cpu } 446 447 static const struct x86_cpu_id intel_idle_ids[] = { 448 ICPU(0x1a, idle_cpu_nehalem), 449 ICPU(0x1e, idle_cpu_nehalem), 450 ICPU(0x1f, idle_cpu_nehalem), 451 ICPU(0x25, idle_cpu_nehalem), 452 ICPU(0x2c, idle_cpu_nehalem), 453 ICPU(0x2e, idle_cpu_nehalem), 454 ICPU(0x1c, idle_cpu_atom), 455 ICPU(0x26, idle_cpu_lincroft), 456 ICPU(0x2f, idle_cpu_nehalem), 457 ICPU(0x2a, idle_cpu_snb), 458 ICPU(0x2d, idle_cpu_snb), 459 ICPU(0x3a, idle_cpu_ivb), 460 ICPU(0x3e, idle_cpu_ivb), 461 ICPU(0x3c, idle_cpu_hsw), 462 ICPU(0x3f, idle_cpu_hsw), 463 ICPU(0x45, idle_cpu_hsw), 464 ICPU(0x46, idle_cpu_hsw), 465 {} 466 }; 467 MODULE_DEVICE_TABLE(x86cpu, intel_idle_ids); 468 469 /* 470 * intel_idle_probe() 471 */ 472 static int intel_idle_probe(void) 473 { 474 unsigned int eax, ebx, ecx; 475 const struct x86_cpu_id *id; 476 477 if (max_cstate == 0) { 478 pr_debug(PREFIX "disabled\n"); 479 return -EPERM; 480 } 481 482 id = x86_match_cpu(intel_idle_ids); 483 if (!id) { 484 if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL && 485 boot_cpu_data.x86 == 6) 486 pr_debug(PREFIX "does not run on family %d model %d\n", 487 boot_cpu_data.x86, boot_cpu_data.x86_model); 488 return -ENODEV; 489 } 490 491 if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF) 492 return -ENODEV; 493 494 cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates); 495 496 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) || 497 !(ecx & CPUID5_ECX_INTERRUPT_BREAK) || 498 !mwait_substates) 499 return -ENODEV; 500 501 pr_debug(PREFIX "MWAIT substates: 0x%x\n", mwait_substates); 502 503 icpu = (const struct idle_cpu *)id->driver_data; 504 cpuidle_state_table = icpu->state_table; 505 506 if (boot_cpu_has(X86_FEATURE_ARAT)) /* Always Reliable APIC Timer */ 507 lapic_timer_reliable_states = LAPIC_TIMER_ALWAYS_RELIABLE; 508 else 509 on_each_cpu(__setup_broadcast_timer, (void *)true, 1); 510 511 pr_debug(PREFIX "v" INTEL_IDLE_VERSION 512 " model 0x%X\n", boot_cpu_data.x86_model); 513 514 pr_debug(PREFIX "lapic_timer_reliable_states 0x%x\n", 515 lapic_timer_reliable_states); 516 return 0; 517 } 518 519 /* 520 * intel_idle_cpuidle_devices_uninit() 521 * unregister, free cpuidle_devices 522 */ 523 static void intel_idle_cpuidle_devices_uninit(void) 524 { 525 int i; 526 struct cpuidle_device *dev; 527 528 for_each_online_cpu(i) { 529 dev = per_cpu_ptr(intel_idle_cpuidle_devices, i); 530 cpuidle_unregister_device(dev); 531 } 532 533 free_percpu(intel_idle_cpuidle_devices); 534 return; 535 } 536 /* 537 * intel_idle_cpuidle_driver_init() 538 * allocate, initialize cpuidle_states 539 */ 540 static int intel_idle_cpuidle_driver_init(void) 541 { 542 int cstate; 543 struct cpuidle_driver *drv = &intel_idle_driver; 544 545 drv->state_count = 1; 546 547 for (cstate = 0; cstate < CPUIDLE_STATE_MAX; ++cstate) { 548 int num_substates, mwait_hint, mwait_cstate, mwait_substate; 549 550 if (cpuidle_state_table[cstate].enter == NULL) 551 break; 552 553 if (cstate + 1 > max_cstate) { 554 printk(PREFIX "max_cstate %d reached\n", 555 max_cstate); 556 break; 557 } 558 559 mwait_hint = flg2MWAIT(cpuidle_state_table[cstate].flags); 560 mwait_cstate = MWAIT_HINT2CSTATE(mwait_hint); 561 mwait_substate = MWAIT_HINT2SUBSTATE(mwait_hint); 562 563 /* does the state exist in CPUID.MWAIT? */ 564 num_substates = (mwait_substates >> ((mwait_cstate + 1) * 4)) 565 & MWAIT_SUBSTATE_MASK; 566 567 /* if sub-state in table is not enumerated by CPUID */ 568 if ((mwait_substate + 1) > num_substates) 569 continue; 570 571 if (((mwait_cstate + 1) > 2) && 572 !boot_cpu_has(X86_FEATURE_NONSTOP_TSC)) 573 mark_tsc_unstable("TSC halts in idle" 574 " states deeper than C2"); 575 576 drv->states[drv->state_count] = /* structure copy */ 577 cpuidle_state_table[cstate]; 578 579 drv->state_count += 1; 580 } 581 582 if (icpu->auto_demotion_disable_flags) 583 on_each_cpu(auto_demotion_disable, NULL, 1); 584 585 if (icpu->disable_promotion_to_c1e) /* each-cpu is redundant */ 586 on_each_cpu(c1e_promotion_disable, NULL, 1); 587 588 return 0; 589 } 590 591 592 /* 593 * intel_idle_cpu_init() 594 * allocate, initialize, register cpuidle_devices 595 * @cpu: cpu/core to initialize 596 */ 597 static int intel_idle_cpu_init(int cpu) 598 { 599 int cstate; 600 struct cpuidle_device *dev; 601 602 dev = per_cpu_ptr(intel_idle_cpuidle_devices, cpu); 603 604 dev->state_count = 1; 605 606 for (cstate = 0; cstate < CPUIDLE_STATE_MAX; ++cstate) { 607 int num_substates, mwait_hint, mwait_cstate, mwait_substate; 608 609 if (cpuidle_state_table[cstate].enter == NULL) 610 continue; 611 612 if (cstate + 1 > max_cstate) { 613 printk(PREFIX "max_cstate %d reached\n", max_cstate); 614 break; 615 } 616 617 mwait_hint = flg2MWAIT(cpuidle_state_table[cstate].flags); 618 mwait_cstate = MWAIT_HINT2CSTATE(mwait_hint); 619 mwait_substate = MWAIT_HINT2SUBSTATE(mwait_hint); 620 621 /* does the state exist in CPUID.MWAIT? */ 622 num_substates = (mwait_substates >> ((mwait_cstate + 1) * 4)) 623 & MWAIT_SUBSTATE_MASK; 624 625 /* if sub-state in table is not enumerated by CPUID */ 626 if ((mwait_substate + 1) > num_substates) 627 continue; 628 629 dev->state_count += 1; 630 } 631 632 dev->cpu = cpu; 633 634 if (cpuidle_register_device(dev)) { 635 pr_debug(PREFIX "cpuidle_register_device %d failed!\n", cpu); 636 intel_idle_cpuidle_devices_uninit(); 637 return -EIO; 638 } 639 640 if (icpu->auto_demotion_disable_flags) 641 smp_call_function_single(cpu, auto_demotion_disable, NULL, 1); 642 643 return 0; 644 } 645 646 static int __init intel_idle_init(void) 647 { 648 int retval, i; 649 650 /* Do not load intel_idle at all for now if idle= is passed */ 651 if (boot_option_idle_override != IDLE_NO_OVERRIDE) 652 return -ENODEV; 653 654 retval = intel_idle_probe(); 655 if (retval) 656 return retval; 657 658 intel_idle_cpuidle_driver_init(); 659 retval = cpuidle_register_driver(&intel_idle_driver); 660 if (retval) { 661 struct cpuidle_driver *drv = cpuidle_get_driver(); 662 printk(KERN_DEBUG PREFIX "intel_idle yielding to %s", 663 drv ? drv->name : "none"); 664 return retval; 665 } 666 667 intel_idle_cpuidle_devices = alloc_percpu(struct cpuidle_device); 668 if (intel_idle_cpuidle_devices == NULL) 669 return -ENOMEM; 670 671 for_each_online_cpu(i) { 672 retval = intel_idle_cpu_init(i); 673 if (retval) { 674 cpuidle_unregister_driver(&intel_idle_driver); 675 return retval; 676 } 677 } 678 register_cpu_notifier(&cpu_hotplug_notifier); 679 680 return 0; 681 } 682 683 static void __exit intel_idle_exit(void) 684 { 685 intel_idle_cpuidle_devices_uninit(); 686 cpuidle_unregister_driver(&intel_idle_driver); 687 688 689 if (lapic_timer_reliable_states != LAPIC_TIMER_ALWAYS_RELIABLE) 690 on_each_cpu(__setup_broadcast_timer, (void *)false, 1); 691 unregister_cpu_notifier(&cpu_hotplug_notifier); 692 693 return; 694 } 695 696 module_init(intel_idle_init); 697 module_exit(intel_idle_exit); 698 699 module_param(max_cstate, int, 0444); 700 701 MODULE_AUTHOR("Len Brown <len.brown@intel.com>"); 702 MODULE_DESCRIPTION("Cpuidle driver for Intel Hardware v" INTEL_IDLE_VERSION); 703 MODULE_LICENSE("GPL"); 704