2 * processor_idle - idle state submodule to the ACPI processor driver
4 * Copyright (C) 2001, 2002 Andy Grover <andrew.grover@intel.com>
5 * Copyright (C) 2001, 2002 Paul Diefenbaugh <paul.s.diefenbaugh@intel.com>
6 * Copyright (C) 2004, 2005 Dominik Brodowski <linux@brodo.de>
7 * Copyright (C) 2004 Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
8 * - Added processor hotplug support
9 * Copyright (C) 2005 Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
10 * - Added support for C3 on SMP
12 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
14 * This program is free software; you can redistribute it and/or modify
15 * it under the terms of the GNU General Public License as published by
16 * the Free Software Foundation; either version 2 of the License, or (at
17 * your option) any later version.
19 * This program is distributed in the hope that it will be useful, but
20 * WITHOUT ANY WARRANTY; without even the implied warranty of
21 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
22 * General Public License for more details.
24 * You should have received a copy of the GNU General Public License along
25 * with this program; if not, write to the Free Software Foundation, Inc.,
26 * 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA.
28 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
31 #include <linux/kernel.h>
32 #include <linux/module.h>
33 #include <linux/init.h>
34 #include <linux/cpufreq.h>
35 #include <linux/slab.h>
36 #include <linux/acpi.h>
37 #include <linux/dmi.h>
38 #include <linux/moduleparam.h>
39 #include <linux/sched.h> /* need_resched() */
40 #include <linux/pm_qos.h>
41 #include <linux/clockchips.h>
42 #include <linux/cpuidle.h>
43 #include <linux/irqflags.h>
46 * Include the apic definitions for x86 to have the APIC timer related defines
47 * available also for UP (on SMP it gets magically included via linux/smp.h).
48 * asm/acpi.h is not an option, as it would require more include magic. Also
49 * creating an empty asm-ia64/apic.h would just trade pest vs. cholera.
56 #include <asm/uaccess.h>
58 #include <acpi/acpi_bus.h>
59 #include <acpi/processor.h>
60 #include <asm/processor.h>
62 #define PREFIX "ACPI: "
64 #define ACPI_PROCESSOR_CLASS "processor"
65 #define _COMPONENT ACPI_PROCESSOR_COMPONENT
66 ACPI_MODULE_NAME("processor_idle");
67 #define PM_TIMER_TICK_NS (1000000000ULL/PM_TIMER_FREQUENCY)
68 #define C2_OVERHEAD 1 /* 1us */
69 #define C3_OVERHEAD 1 /* 1us */
70 #define PM_TIMER_TICKS_TO_US(p) (((p) * 1000)/(PM_TIMER_FREQUENCY/1000))
72 static unsigned int max_cstate __read_mostly = ACPI_PROCESSOR_MAX_POWER;
73 module_param(max_cstate, uint, 0000);
74 static unsigned int nocst __read_mostly;
75 module_param(nocst, uint, 0000);
76 static int bm_check_disable __read_mostly;
77 module_param(bm_check_disable, uint, 0000);
79 static unsigned int latency_factor __read_mostly = 2;
80 module_param(latency_factor, uint, 0644);
82 static int disabled_by_idle_boot_param(void)
84 return boot_option_idle_override == IDLE_POLL ||
85 boot_option_idle_override == IDLE_FORCE_MWAIT ||
86 boot_option_idle_override == IDLE_HALT;
90 * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
91 * For now disable this. Probably a bug somewhere else.
93 * To skip this limit, boot/load with a large max_cstate limit.
95 static int set_max_cstate(const struct dmi_system_id *id)
97 if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
100 printk(KERN_NOTICE PREFIX "%s detected - limiting to C%ld max_cstate."
101 " Override with \"processor.max_cstate=%d\"\n", id->ident,
102 (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
104 max_cstate = (long)id->driver_data;
109 /* Actually this shouldn't be __cpuinitdata, would be better to fix the
110 callers to only run once -AK */
111 static struct dmi_system_id __cpuinitdata processor_power_dmi_table[] = {
112 { set_max_cstate, "Clevo 5600D", {
113 DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
114 DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")},
116 { set_max_cstate, "Pavilion zv5000", {
117 DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
118 DMI_MATCH(DMI_PRODUCT_NAME,"Pavilion zv5000 (DS502A#ABA)")},
120 { set_max_cstate, "Asus L8400B", {
121 DMI_MATCH(DMI_SYS_VENDOR, "ASUSTeK Computer Inc."),
122 DMI_MATCH(DMI_PRODUCT_NAME,"L8400B series Notebook PC")},
129 * Callers should disable interrupts before the call and enable
130 * interrupts after return.
132 static void acpi_safe_halt(void)
134 current_thread_info()->status &= ~TS_POLLING;
136 * TS_POLLING-cleared state must be visible before we
140 if (!need_resched()) {
144 current_thread_info()->status |= TS_POLLING;
147 #ifdef ARCH_APICTIMER_STOPS_ON_C3
150 * Some BIOS implementations switch to C3 in the published C2 state.
151 * This seems to be a common problem on AMD boxen, but other vendors
152 * are affected too. We pick the most conservative approach: we assume
153 * that the local APIC stops in both C2 and C3.
155 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
156 struct acpi_processor_cx *cx)
158 struct acpi_processor_power *pwr = &pr->power;
159 u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2;
161 if (cpu_has(&cpu_data(pr->id), X86_FEATURE_ARAT))
164 if (amd_e400_c1e_detected)
165 type = ACPI_STATE_C1;
168 * Check, if one of the previous states already marked the lapic
171 if (pwr->timer_broadcast_on_state < state)
174 if (cx->type >= type)
175 pr->power.timer_broadcast_on_state = state;
178 static void __lapic_timer_propagate_broadcast(void *arg)
180 struct acpi_processor *pr = (struct acpi_processor *) arg;
181 unsigned long reason;
183 reason = pr->power.timer_broadcast_on_state < INT_MAX ?
184 CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
186 clockevents_notify(reason, &pr->id);
189 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr)
191 smp_call_function_single(pr->id, __lapic_timer_propagate_broadcast,
195 /* Power(C) State timer broadcast control */
196 static void lapic_timer_state_broadcast(struct acpi_processor *pr,
197 struct acpi_processor_cx *cx,
200 int state = cx - pr->power.states;
202 if (state >= pr->power.timer_broadcast_on_state) {
203 unsigned long reason;
205 reason = broadcast ? CLOCK_EVT_NOTIFY_BROADCAST_ENTER :
206 CLOCK_EVT_NOTIFY_BROADCAST_EXIT;
207 clockevents_notify(reason, &pr->id);
213 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
214 struct acpi_processor_cx *cstate) { }
215 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) { }
216 static void lapic_timer_state_broadcast(struct acpi_processor *pr,
217 struct acpi_processor_cx *cx,
225 * Suspend / resume control
227 static int acpi_idle_suspend;
228 static u32 saved_bm_rld;
230 static void acpi_idle_bm_rld_save(void)
232 acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &saved_bm_rld);
234 static void acpi_idle_bm_rld_restore(void)
238 acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &resumed_bm_rld);
240 if (resumed_bm_rld != saved_bm_rld)
241 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, saved_bm_rld);
244 int acpi_processor_suspend(struct acpi_device * device, pm_message_t state)
246 if (acpi_idle_suspend == 1)
249 acpi_idle_bm_rld_save();
250 acpi_idle_suspend = 1;
254 int acpi_processor_resume(struct acpi_device * device)
256 if (acpi_idle_suspend == 0)
259 acpi_idle_bm_rld_restore();
260 acpi_idle_suspend = 0;
264 #if defined(CONFIG_X86)
265 static void tsc_check_state(int state)
267 switch (boot_cpu_data.x86_vendor) {
269 case X86_VENDOR_INTEL:
271 * AMD Fam10h TSC will tick in all
272 * C/P/S0/S1 states when this bit is set.
274 if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
279 /* TSC could halt in idle, so notify users */
280 if (state > ACPI_STATE_C1)
281 mark_tsc_unstable("TSC halts in idle");
285 static void tsc_check_state(int state) { return; }
288 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
297 /* if info is obtained from pblk/fadt, type equals state */
298 pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
299 pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
301 #ifndef CONFIG_HOTPLUG_CPU
303 * Check for P_LVL2_UP flag before entering C2 and above on
306 if ((num_online_cpus() > 1) &&
307 !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
311 /* determine C2 and C3 address from pblk */
312 pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
313 pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
315 /* determine latencies from FADT */
316 pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.C2latency;
317 pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.C3latency;
320 * FADT specified C2 latency must be less than or equal to
323 if (acpi_gbl_FADT.C2latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
324 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
325 "C2 latency too large [%d]\n", acpi_gbl_FADT.C2latency));
327 pr->power.states[ACPI_STATE_C2].address = 0;
331 * FADT supplied C3 latency must be less than or equal to
334 if (acpi_gbl_FADT.C3latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
335 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
336 "C3 latency too large [%d]\n", acpi_gbl_FADT.C3latency));
338 pr->power.states[ACPI_STATE_C3].address = 0;
341 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
342 "lvl2[0x%08x] lvl3[0x%08x]\n",
343 pr->power.states[ACPI_STATE_C2].address,
344 pr->power.states[ACPI_STATE_C3].address));
349 static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
351 if (!pr->power.states[ACPI_STATE_C1].valid) {
352 /* set the first C-State to C1 */
353 /* all processors need to support C1 */
354 pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
355 pr->power.states[ACPI_STATE_C1].valid = 1;
356 pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
358 /* the C0 state only exists as a filler in our array */
359 pr->power.states[ACPI_STATE_C0].valid = 1;
363 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
365 acpi_status status = 0;
369 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
370 union acpi_object *cst;
378 status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
379 if (ACPI_FAILURE(status)) {
380 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
384 cst = buffer.pointer;
386 /* There must be at least 2 elements */
387 if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
388 printk(KERN_ERR PREFIX "not enough elements in _CST\n");
393 count = cst->package.elements[0].integer.value;
395 /* Validate number of power states. */
396 if (count < 1 || count != cst->package.count - 1) {
397 printk(KERN_ERR PREFIX "count given by _CST is not valid\n");
402 /* Tell driver that at least _CST is supported. */
403 pr->flags.has_cst = 1;
405 for (i = 1; i <= count; i++) {
406 union acpi_object *element;
407 union acpi_object *obj;
408 struct acpi_power_register *reg;
409 struct acpi_processor_cx cx;
411 memset(&cx, 0, sizeof(cx));
413 element = &(cst->package.elements[i]);
414 if (element->type != ACPI_TYPE_PACKAGE)
417 if (element->package.count != 4)
420 obj = &(element->package.elements[0]);
422 if (obj->type != ACPI_TYPE_BUFFER)
425 reg = (struct acpi_power_register *)obj->buffer.pointer;
427 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
428 (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
431 /* There should be an easy way to extract an integer... */
432 obj = &(element->package.elements[1]);
433 if (obj->type != ACPI_TYPE_INTEGER)
436 cx.type = obj->integer.value;
438 * Some buggy BIOSes won't list C1 in _CST -
439 * Let acpi_processor_get_power_info_default() handle them later
441 if (i == 1 && cx.type != ACPI_STATE_C1)
444 cx.address = reg->address;
445 cx.index = current_count + 1;
447 cx.entry_method = ACPI_CSTATE_SYSTEMIO;
448 if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) {
449 if (acpi_processor_ffh_cstate_probe
450 (pr->id, &cx, reg) == 0) {
451 cx.entry_method = ACPI_CSTATE_FFH;
452 } else if (cx.type == ACPI_STATE_C1) {
454 * C1 is a special case where FIXED_HARDWARE
455 * can be handled in non-MWAIT way as well.
456 * In that case, save this _CST entry info.
457 * Otherwise, ignore this info and continue.
459 cx.entry_method = ACPI_CSTATE_HALT;
460 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
464 if (cx.type == ACPI_STATE_C1 &&
465 (boot_option_idle_override == IDLE_NOMWAIT)) {
467 * In most cases the C1 space_id obtained from
468 * _CST object is FIXED_HARDWARE access mode.
469 * But when the option of idle=halt is added,
470 * the entry_method type should be changed from
471 * CSTATE_FFH to CSTATE_HALT.
472 * When the option of idle=nomwait is added,
473 * the C1 entry_method type should be
476 cx.entry_method = ACPI_CSTATE_HALT;
477 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
480 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI IOPORT 0x%x",
484 if (cx.type == ACPI_STATE_C1) {
488 obj = &(element->package.elements[2]);
489 if (obj->type != ACPI_TYPE_INTEGER)
492 cx.latency = obj->integer.value;
494 obj = &(element->package.elements[3]);
495 if (obj->type != ACPI_TYPE_INTEGER)
498 cx.power = obj->integer.value;
501 memcpy(&(pr->power.states[current_count]), &cx, sizeof(cx));
504 * We support total ACPI_PROCESSOR_MAX_POWER - 1
505 * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
507 if (current_count >= (ACPI_PROCESSOR_MAX_POWER - 1)) {
509 "Limiting number of power states to max (%d)\n",
510 ACPI_PROCESSOR_MAX_POWER);
512 "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
517 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n",
520 /* Validate number of power states discovered */
521 if (current_count < 2)
525 kfree(buffer.pointer);
530 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
531 struct acpi_processor_cx *cx)
533 static int bm_check_flag = -1;
534 static int bm_control_flag = -1;
541 * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
542 * DMA transfers are used by any ISA device to avoid livelock.
543 * Note that we could disable Type-F DMA (as recommended by
544 * the erratum), but this is known to disrupt certain ISA
545 * devices thus we take the conservative approach.
547 else if (errata.piix4.fdma) {
548 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
549 "C3 not supported on PIIX4 with Type-F DMA\n"));
553 /* All the logic here assumes flags.bm_check is same across all CPUs */
554 if (bm_check_flag == -1) {
555 /* Determine whether bm_check is needed based on CPU */
556 acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
557 bm_check_flag = pr->flags.bm_check;
558 bm_control_flag = pr->flags.bm_control;
560 pr->flags.bm_check = bm_check_flag;
561 pr->flags.bm_control = bm_control_flag;
564 if (pr->flags.bm_check) {
565 if (!pr->flags.bm_control) {
566 if (pr->flags.has_cst != 1) {
567 /* bus mastering control is necessary */
568 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
569 "C3 support requires BM control\n"));
572 /* Here we enter C3 without bus mastering */
573 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
574 "C3 support without BM control\n"));
579 * WBINVD should be set in fadt, for C3 state to be
580 * supported on when bm_check is not required.
582 if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
583 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
584 "Cache invalidation should work properly"
585 " for C3 to be enabled on SMP systems\n"));
591 * Otherwise we've met all of our C3 requirements.
592 * Normalize the C3 latency to expidite policy. Enable
593 * checking of bus mastering status (bm_check) so we can
594 * use this in our C3 policy
598 cx->latency_ticks = cx->latency;
600 * On older chipsets, BM_RLD needs to be set
601 * in order for Bus Master activity to wake the
602 * system from C3. Newer chipsets handle DMA
603 * during C3 automatically and BM_RLD is a NOP.
604 * In either case, the proper way to
605 * handle BM_RLD is to set it and leave it set.
607 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
612 static int acpi_processor_power_verify(struct acpi_processor *pr)
615 unsigned int working = 0;
617 pr->power.timer_broadcast_on_state = INT_MAX;
619 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
620 struct acpi_processor_cx *cx = &pr->power.states[i];
631 cx->latency_ticks = cx->latency; /* Normalize latency */
635 acpi_processor_power_verify_c3(pr, cx);
641 lapic_timer_check_state(i, pr, cx);
642 tsc_check_state(cx->type);
646 lapic_timer_propagate_broadcast(pr);
651 static int acpi_processor_get_power_info(struct acpi_processor *pr)
657 /* NOTE: the idle thread may not be running while calling
660 /* Zero initialize all the C-states info. */
661 memset(pr->power.states, 0, sizeof(pr->power.states));
663 result = acpi_processor_get_power_info_cst(pr);
664 if (result == -ENODEV)
665 result = acpi_processor_get_power_info_fadt(pr);
670 acpi_processor_get_power_info_default(pr);
672 pr->power.count = acpi_processor_power_verify(pr);
675 * if one state of type C2 or C3 is available, mark this
676 * CPU as being "idle manageable"
678 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
679 if (pr->power.states[i].valid) {
681 if (pr->power.states[i].type >= ACPI_STATE_C2)
690 * acpi_idle_bm_check - checks if bus master activity was detected
692 static int acpi_idle_bm_check(void)
696 if (bm_check_disable)
699 acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
701 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
703 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
704 * the true state of bus mastering activity; forcing us to
705 * manually check the BMIDEA bit of each IDE channel.
707 else if (errata.piix4.bmisx) {
708 if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
709 || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
716 * acpi_idle_do_entry - a helper function that does C2 and C3 type entry
719 * Caller disables interrupt before call and enables interrupt after return.
721 static inline void acpi_idle_do_entry(struct acpi_processor_cx *cx)
723 /* Don't trace irqs off for idle */
724 stop_critical_timings();
725 if (cx->entry_method == ACPI_CSTATE_FFH) {
726 /* Call into architectural FFH based C-state */
727 acpi_processor_ffh_cstate_enter(cx);
728 } else if (cx->entry_method == ACPI_CSTATE_HALT) {
731 /* IO port based C-state */
733 /* Dummy wait op - must do something useless after P_LVL2 read
734 because chipsets cannot guarantee that STPCLK# signal
735 gets asserted in time to freeze execution properly. */
736 inl(acpi_gbl_FADT.xpm_timer_block.address);
738 start_critical_timings();
742 * acpi_idle_enter_c1 - enters an ACPI C1 state-type
743 * @dev: the target CPU
744 * @drv: cpuidle driver containing cpuidle state info
745 * @index: index of target state
747 * This is equivalent to the HALT instruction.
749 static int acpi_idle_enter_c1(struct cpuidle_device *dev,
750 struct cpuidle_driver *drv, int index)
754 struct acpi_processor *pr;
755 struct cpuidle_state_usage *state_usage = &dev->states_usage[index];
756 struct acpi_processor_cx *cx = cpuidle_get_statedata(state_usage);
758 pr = __this_cpu_read(processors);
759 dev->last_residency = 0;
766 if (acpi_idle_suspend) {
772 lapic_timer_state_broadcast(pr, cx, 1);
773 kt1 = ktime_get_real();
774 acpi_idle_do_entry(cx);
775 kt2 = ktime_get_real();
776 idle_time = ktime_to_us(ktime_sub(kt2, kt1));
778 /* Update device last_residency*/
779 dev->last_residency = (int)idle_time;
783 lapic_timer_state_broadcast(pr, cx, 0);
790 * acpi_idle_play_dead - enters an ACPI state for long-term idle (i.e. off-lining)
791 * @dev: the target CPU
792 * @index: the index of suggested state
794 static int acpi_idle_play_dead(struct cpuidle_device *dev, int index)
796 struct cpuidle_state_usage *state_usage = &dev->states_usage[index];
797 struct acpi_processor_cx *cx = cpuidle_get_statedata(state_usage);
799 ACPI_FLUSH_CPU_CACHE();
803 if (cx->entry_method == ACPI_CSTATE_HALT)
805 else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) {
807 /* See comment in acpi_idle_do_entry() */
808 inl(acpi_gbl_FADT.xpm_timer_block.address);
818 * acpi_idle_enter_simple - enters an ACPI state without BM handling
819 * @dev: the target CPU
820 * @drv: cpuidle driver with cpuidle state information
821 * @index: the index of suggested state
823 static int acpi_idle_enter_simple(struct cpuidle_device *dev,
824 struct cpuidle_driver *drv, int index)
826 struct acpi_processor *pr;
827 struct cpuidle_state_usage *state_usage = &dev->states_usage[index];
828 struct acpi_processor_cx *cx = cpuidle_get_statedata(state_usage);
833 pr = __this_cpu_read(processors);
834 dev->last_residency = 0;
841 if (acpi_idle_suspend) {
847 if (cx->entry_method != ACPI_CSTATE_FFH) {
848 current_thread_info()->status &= ~TS_POLLING;
850 * TS_POLLING-cleared state must be visible before we test
855 if (unlikely(need_resched())) {
856 current_thread_info()->status |= TS_POLLING;
863 * Must be done before busmaster disable as we might need to
866 lapic_timer_state_broadcast(pr, cx, 1);
868 if (cx->type == ACPI_STATE_C3)
869 ACPI_FLUSH_CPU_CACHE();
871 kt1 = ktime_get_real();
872 /* Tell the scheduler that we are going deep-idle: */
873 sched_clock_idle_sleep_event();
874 acpi_idle_do_entry(cx);
875 kt2 = ktime_get_real();
876 idle_time_ns = ktime_to_ns(ktime_sub(kt2, kt1));
877 idle_time = idle_time_ns;
878 do_div(idle_time, NSEC_PER_USEC);
880 /* Update device last_residency*/
881 dev->last_residency = (int)idle_time;
883 /* Tell the scheduler how much we idled: */
884 sched_clock_idle_wakeup_event(idle_time_ns);
887 if (cx->entry_method != ACPI_CSTATE_FFH)
888 current_thread_info()->status |= TS_POLLING;
892 lapic_timer_state_broadcast(pr, cx, 0);
893 cx->time += idle_time;
897 static int c3_cpu_count;
898 static DEFINE_RAW_SPINLOCK(c3_lock);
901 * acpi_idle_enter_bm - enters C3 with proper BM handling
902 * @dev: the target CPU
903 * @drv: cpuidle driver containing state data
904 * @index: the index of suggested state
906 * If BM is detected, the deepest non-C3 idle state is entered instead.
908 static int acpi_idle_enter_bm(struct cpuidle_device *dev,
909 struct cpuidle_driver *drv, int index)
911 struct acpi_processor *pr;
912 struct cpuidle_state_usage *state_usage = &dev->states_usage[index];
913 struct acpi_processor_cx *cx = cpuidle_get_statedata(state_usage);
919 pr = __this_cpu_read(processors);
920 dev->last_residency = 0;
925 if (!cx->bm_sts_skip && acpi_idle_bm_check()) {
926 if (drv->safe_state_index >= 0) {
927 return drv->states[drv->safe_state_index].enter(dev,
928 drv, drv->safe_state_index);
931 if (!acpi_idle_suspend)
940 if (acpi_idle_suspend) {
946 if (cx->entry_method != ACPI_CSTATE_FFH) {
947 current_thread_info()->status &= ~TS_POLLING;
949 * TS_POLLING-cleared state must be visible before we test
954 if (unlikely(need_resched())) {
955 current_thread_info()->status |= TS_POLLING;
961 acpi_unlazy_tlb(smp_processor_id());
963 /* Tell the scheduler that we are going deep-idle: */
964 sched_clock_idle_sleep_event();
966 * Must be done before busmaster disable as we might need to
969 lapic_timer_state_broadcast(pr, cx, 1);
971 kt1 = ktime_get_real();
974 * bm_check implies we need ARB_DIS
975 * !bm_check implies we need cache flush
976 * bm_control implies whether we can do ARB_DIS
978 * That leaves a case where bm_check is set and bm_control is
979 * not set. In that case we cannot do much, we enter C3
980 * without doing anything.
982 if (pr->flags.bm_check && pr->flags.bm_control) {
983 raw_spin_lock(&c3_lock);
985 /* Disable bus master arbitration when all CPUs are in C3 */
986 if (c3_cpu_count == num_online_cpus())
987 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
988 raw_spin_unlock(&c3_lock);
989 } else if (!pr->flags.bm_check) {
990 ACPI_FLUSH_CPU_CACHE();
993 acpi_idle_do_entry(cx);
995 /* Re-enable bus master arbitration */
996 if (pr->flags.bm_check && pr->flags.bm_control) {
997 raw_spin_lock(&c3_lock);
998 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
1000 raw_spin_unlock(&c3_lock);
1002 kt2 = ktime_get_real();
1003 idle_time_ns = ktime_to_ns(ktime_sub(kt2, kt1));
1004 idle_time = idle_time_ns;
1005 do_div(idle_time, NSEC_PER_USEC);
1007 /* Update device last_residency*/
1008 dev->last_residency = (int)idle_time;
1010 /* Tell the scheduler how much we idled: */
1011 sched_clock_idle_wakeup_event(idle_time_ns);
1014 if (cx->entry_method != ACPI_CSTATE_FFH)
1015 current_thread_info()->status |= TS_POLLING;
1019 lapic_timer_state_broadcast(pr, cx, 0);
1020 cx->time += idle_time;
1024 struct cpuidle_driver acpi_idle_driver = {
1025 .name = "acpi_idle",
1026 .owner = THIS_MODULE,
1030 * acpi_processor_setup_cpuidle_cx - prepares and configures CPUIDLE
1031 * device i.e. per-cpu data
1033 * @pr: the ACPI processor
1035 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr)
1037 int i, count = CPUIDLE_DRIVER_STATE_START;
1038 struct acpi_processor_cx *cx;
1039 struct cpuidle_state_usage *state_usage;
1040 struct cpuidle_device *dev = &pr->power.dev;
1042 if (!pr->flags.power_setup_done)
1045 if (pr->flags.power == 0) {
1051 if (max_cstate == 0)
1054 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
1055 cx = &pr->power.states[i];
1056 state_usage = &dev->states_usage[count];
1061 #ifdef CONFIG_HOTPLUG_CPU
1062 if ((cx->type != ACPI_STATE_C1) && (num_online_cpus() > 1) &&
1063 !pr->flags.has_cst &&
1064 !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
1068 cpuidle_set_statedata(state_usage, cx);
1071 if (count == CPUIDLE_STATE_MAX)
1075 dev->state_count = count;
1084 * acpi_processor_setup_cpuidle states- prepares and configures cpuidle
1085 * global state data i.e. idle routines
1087 * @pr: the ACPI processor
1089 static int acpi_processor_setup_cpuidle_states(struct acpi_processor *pr)
1091 int i, count = CPUIDLE_DRIVER_STATE_START;
1092 struct acpi_processor_cx *cx;
1093 struct cpuidle_state *state;
1094 struct cpuidle_driver *drv = &acpi_idle_driver;
1096 if (!pr->flags.power_setup_done)
1099 if (pr->flags.power == 0)
1102 drv->safe_state_index = -1;
1103 for (i = 0; i < CPUIDLE_STATE_MAX; i++) {
1104 drv->states[i].name[0] = '\0';
1105 drv->states[i].desc[0] = '\0';
1108 if (max_cstate == 0)
1111 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
1112 cx = &pr->power.states[i];
1117 #ifdef CONFIG_HOTPLUG_CPU
1118 if ((cx->type != ACPI_STATE_C1) && (num_online_cpus() > 1) &&
1119 !pr->flags.has_cst &&
1120 !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
1124 state = &drv->states[count];
1125 snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
1126 strncpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
1127 state->exit_latency = cx->latency;
1128 state->target_residency = cx->latency * latency_factor;
1133 if (cx->entry_method == ACPI_CSTATE_FFH)
1134 state->flags |= CPUIDLE_FLAG_TIME_VALID;
1136 state->enter = acpi_idle_enter_c1;
1137 state->enter_dead = acpi_idle_play_dead;
1138 drv->safe_state_index = count;
1142 state->flags |= CPUIDLE_FLAG_TIME_VALID;
1143 state->enter = acpi_idle_enter_simple;
1144 state->enter_dead = acpi_idle_play_dead;
1145 drv->safe_state_index = count;
1149 state->flags |= CPUIDLE_FLAG_TIME_VALID;
1150 state->enter = pr->flags.bm_check ?
1151 acpi_idle_enter_bm :
1152 acpi_idle_enter_simple;
1157 if (count == CPUIDLE_STATE_MAX)
1161 drv->state_count = count;
1169 int acpi_processor_hotplug(struct acpi_processor *pr)
1173 if (disabled_by_idle_boot_param())
1183 if (!pr->flags.power_setup_done)
1186 cpuidle_pause_and_lock();
1187 cpuidle_disable_device(&pr->power.dev);
1188 acpi_processor_get_power_info(pr);
1189 if (pr->flags.power) {
1190 acpi_processor_setup_cpuidle_cx(pr);
1191 ret = cpuidle_enable_device(&pr->power.dev);
1193 cpuidle_resume_and_unlock();
1198 int acpi_processor_cst_has_changed(struct acpi_processor *pr)
1201 struct acpi_processor *_pr;
1203 if (disabled_by_idle_boot_param())
1212 if (!pr->flags.power_setup_done)
1216 * FIXME: Design the ACPI notification to make it once per
1217 * system instead of once per-cpu. This condition is a hack
1218 * to make the code that updates C-States be called once.
1221 if (pr->id == 0 && cpuidle_get_driver() == &acpi_idle_driver) {
1223 cpuidle_pause_and_lock();
1224 /* Protect against cpu-hotplug */
1227 /* Disable all cpuidle devices */
1228 for_each_online_cpu(cpu) {
1229 _pr = per_cpu(processors, cpu);
1230 if (!_pr || !_pr->flags.power_setup_done)
1232 cpuidle_disable_device(&_pr->power.dev);
1235 /* Populate Updated C-state information */
1236 acpi_processor_setup_cpuidle_states(pr);
1238 /* Enable all cpuidle devices */
1239 for_each_online_cpu(cpu) {
1240 _pr = per_cpu(processors, cpu);
1241 if (!_pr || !_pr->flags.power_setup_done)
1243 acpi_processor_get_power_info(_pr);
1244 if (_pr->flags.power) {
1245 acpi_processor_setup_cpuidle_cx(_pr);
1246 cpuidle_enable_device(&_pr->power.dev);
1250 cpuidle_resume_and_unlock();
1256 static int acpi_processor_registered;
1258 int __cpuinit acpi_processor_power_init(struct acpi_processor *pr,
1259 struct acpi_device *device)
1261 acpi_status status = 0;
1263 static int first_run;
1265 if (disabled_by_idle_boot_param())
1269 dmi_check_system(processor_power_dmi_table);
1270 max_cstate = acpi_processor_cstate_check(max_cstate);
1271 if (max_cstate < ACPI_C_STATES_MAX)
1273 "ACPI: processor limited to max C-state %d\n",
1281 if (acpi_gbl_FADT.cst_control && !nocst) {
1283 acpi_os_write_port(acpi_gbl_FADT.smi_command, acpi_gbl_FADT.cst_control, 8);
1284 if (ACPI_FAILURE(status)) {
1285 ACPI_EXCEPTION((AE_INFO, status,
1286 "Notifying BIOS of _CST ability failed"));
1290 acpi_processor_get_power_info(pr);
1291 pr->flags.power_setup_done = 1;
1294 * Install the idle handler if processor power management is supported.
1295 * Note that we use previously set idle handler will be used on
1296 * platforms that only support C1.
1298 if (pr->flags.power) {
1299 /* Register acpi_idle_driver if not already registered */
1300 if (!acpi_processor_registered) {
1301 acpi_processor_setup_cpuidle_states(pr);
1302 retval = cpuidle_register_driver(&acpi_idle_driver);
1305 printk(KERN_DEBUG "ACPI: %s registered with cpuidle\n",
1306 acpi_idle_driver.name);
1308 /* Register per-cpu cpuidle_device. Cpuidle driver
1309 * must already be registered before registering device
1311 acpi_processor_setup_cpuidle_cx(pr);
1312 retval = cpuidle_register_device(&pr->power.dev);
1314 if (acpi_processor_registered == 0)
1315 cpuidle_unregister_driver(&acpi_idle_driver);
1318 acpi_processor_registered++;
1323 int acpi_processor_power_exit(struct acpi_processor *pr,
1324 struct acpi_device *device)
1326 if (disabled_by_idle_boot_param())
1329 if (pr->flags.power) {
1330 cpuidle_unregister_device(&pr->power.dev);
1331 acpi_processor_registered--;
1332 if (acpi_processor_registered == 0)
1333 cpuidle_unregister_driver(&acpi_idle_driver);
1336 pr->flags.power_setup_done = 0;