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
,
224 static u32 saved_bm_rld
;
226 static void acpi_idle_bm_rld_save(void)
228 acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD
, &saved_bm_rld
);
230 static void acpi_idle_bm_rld_restore(void)
234 acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD
, &resumed_bm_rld
);
236 if (resumed_bm_rld
!= saved_bm_rld
)
237 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD
, saved_bm_rld
);
240 int acpi_processor_suspend(struct device
*dev
)
242 acpi_idle_bm_rld_save();
246 int acpi_processor_resume(struct device
*dev
)
248 acpi_idle_bm_rld_restore();
252 #if defined(CONFIG_X86)
253 static void tsc_check_state(int state
)
255 switch (boot_cpu_data
.x86_vendor
) {
257 case X86_VENDOR_INTEL
:
259 * AMD Fam10h TSC will tick in all
260 * C/P/S0/S1 states when this bit is set.
262 if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC
))
267 /* TSC could halt in idle, so notify users */
268 if (state
> ACPI_STATE_C1
)
269 mark_tsc_unstable("TSC halts in idle");
273 static void tsc_check_state(int state
) { return; }
276 static int acpi_processor_get_power_info_fadt(struct acpi_processor
*pr
)
285 /* if info is obtained from pblk/fadt, type equals state */
286 pr
->power
.states
[ACPI_STATE_C2
].type
= ACPI_STATE_C2
;
287 pr
->power
.states
[ACPI_STATE_C3
].type
= ACPI_STATE_C3
;
289 #ifndef CONFIG_HOTPLUG_CPU
291 * Check for P_LVL2_UP flag before entering C2 and above on
294 if ((num_online_cpus() > 1) &&
295 !(acpi_gbl_FADT
.flags
& ACPI_FADT_C2_MP_SUPPORTED
))
299 /* determine C2 and C3 address from pblk */
300 pr
->power
.states
[ACPI_STATE_C2
].address
= pr
->pblk
+ 4;
301 pr
->power
.states
[ACPI_STATE_C3
].address
= pr
->pblk
+ 5;
303 /* determine latencies from FADT */
304 pr
->power
.states
[ACPI_STATE_C2
].latency
= acpi_gbl_FADT
.c2_latency
;
305 pr
->power
.states
[ACPI_STATE_C3
].latency
= acpi_gbl_FADT
.c3_latency
;
308 * FADT specified C2 latency must be less than or equal to
311 if (acpi_gbl_FADT
.c2_latency
> ACPI_PROCESSOR_MAX_C2_LATENCY
) {
312 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
313 "C2 latency too large [%d]\n", acpi_gbl_FADT
.c2_latency
));
315 pr
->power
.states
[ACPI_STATE_C2
].address
= 0;
319 * FADT supplied C3 latency must be less than or equal to
322 if (acpi_gbl_FADT
.c3_latency
> ACPI_PROCESSOR_MAX_C3_LATENCY
) {
323 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
324 "C3 latency too large [%d]\n", acpi_gbl_FADT
.c3_latency
));
326 pr
->power
.states
[ACPI_STATE_C3
].address
= 0;
329 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
330 "lvl2[0x%08x] lvl3[0x%08x]\n",
331 pr
->power
.states
[ACPI_STATE_C2
].address
,
332 pr
->power
.states
[ACPI_STATE_C3
].address
));
337 static int acpi_processor_get_power_info_default(struct acpi_processor
*pr
)
339 if (!pr
->power
.states
[ACPI_STATE_C1
].valid
) {
340 /* set the first C-State to C1 */
341 /* all processors need to support C1 */
342 pr
->power
.states
[ACPI_STATE_C1
].type
= ACPI_STATE_C1
;
343 pr
->power
.states
[ACPI_STATE_C1
].valid
= 1;
344 pr
->power
.states
[ACPI_STATE_C1
].entry_method
= ACPI_CSTATE_HALT
;
346 /* the C0 state only exists as a filler in our array */
347 pr
->power
.states
[ACPI_STATE_C0
].valid
= 1;
351 static int acpi_processor_get_power_info_cst(struct acpi_processor
*pr
)
353 acpi_status status
= 0;
357 struct acpi_buffer buffer
= { ACPI_ALLOCATE_BUFFER
, NULL
};
358 union acpi_object
*cst
;
366 status
= acpi_evaluate_object(pr
->handle
, "_CST", NULL
, &buffer
);
367 if (ACPI_FAILURE(status
)) {
368 ACPI_DEBUG_PRINT((ACPI_DB_INFO
, "No _CST, giving up\n"));
372 cst
= buffer
.pointer
;
374 /* There must be at least 2 elements */
375 if (!cst
|| (cst
->type
!= ACPI_TYPE_PACKAGE
) || cst
->package
.count
< 2) {
376 printk(KERN_ERR PREFIX
"not enough elements in _CST\n");
381 count
= cst
->package
.elements
[0].integer
.value
;
383 /* Validate number of power states. */
384 if (count
< 1 || count
!= cst
->package
.count
- 1) {
385 printk(KERN_ERR PREFIX
"count given by _CST is not valid\n");
390 /* Tell driver that at least _CST is supported. */
391 pr
->flags
.has_cst
= 1;
393 for (i
= 1; i
<= count
; i
++) {
394 union acpi_object
*element
;
395 union acpi_object
*obj
;
396 struct acpi_power_register
*reg
;
397 struct acpi_processor_cx cx
;
399 memset(&cx
, 0, sizeof(cx
));
401 element
= &(cst
->package
.elements
[i
]);
402 if (element
->type
!= ACPI_TYPE_PACKAGE
)
405 if (element
->package
.count
!= 4)
408 obj
= &(element
->package
.elements
[0]);
410 if (obj
->type
!= ACPI_TYPE_BUFFER
)
413 reg
= (struct acpi_power_register
*)obj
->buffer
.pointer
;
415 if (reg
->space_id
!= ACPI_ADR_SPACE_SYSTEM_IO
&&
416 (reg
->space_id
!= ACPI_ADR_SPACE_FIXED_HARDWARE
))
419 /* There should be an easy way to extract an integer... */
420 obj
= &(element
->package
.elements
[1]);
421 if (obj
->type
!= ACPI_TYPE_INTEGER
)
424 cx
.type
= obj
->integer
.value
;
426 * Some buggy BIOSes won't list C1 in _CST -
427 * Let acpi_processor_get_power_info_default() handle them later
429 if (i
== 1 && cx
.type
!= ACPI_STATE_C1
)
432 cx
.address
= reg
->address
;
433 cx
.index
= current_count
+ 1;
435 cx
.entry_method
= ACPI_CSTATE_SYSTEMIO
;
436 if (reg
->space_id
== ACPI_ADR_SPACE_FIXED_HARDWARE
) {
437 if (acpi_processor_ffh_cstate_probe
438 (pr
->id
, &cx
, reg
) == 0) {
439 cx
.entry_method
= ACPI_CSTATE_FFH
;
440 } else if (cx
.type
== ACPI_STATE_C1
) {
442 * C1 is a special case where FIXED_HARDWARE
443 * can be handled in non-MWAIT way as well.
444 * In that case, save this _CST entry info.
445 * Otherwise, ignore this info and continue.
447 cx
.entry_method
= ACPI_CSTATE_HALT
;
448 snprintf(cx
.desc
, ACPI_CX_DESC_LEN
, "ACPI HLT");
452 if (cx
.type
== ACPI_STATE_C1
&&
453 (boot_option_idle_override
== IDLE_NOMWAIT
)) {
455 * In most cases the C1 space_id obtained from
456 * _CST object is FIXED_HARDWARE access mode.
457 * But when the option of idle=halt is added,
458 * the entry_method type should be changed from
459 * CSTATE_FFH to CSTATE_HALT.
460 * When the option of idle=nomwait is added,
461 * the C1 entry_method type should be
464 cx
.entry_method
= ACPI_CSTATE_HALT
;
465 snprintf(cx
.desc
, ACPI_CX_DESC_LEN
, "ACPI HLT");
468 snprintf(cx
.desc
, ACPI_CX_DESC_LEN
, "ACPI IOPORT 0x%x",
472 if (cx
.type
== ACPI_STATE_C1
) {
476 obj
= &(element
->package
.elements
[2]);
477 if (obj
->type
!= ACPI_TYPE_INTEGER
)
480 cx
.latency
= obj
->integer
.value
;
482 obj
= &(element
->package
.elements
[3]);
483 if (obj
->type
!= ACPI_TYPE_INTEGER
)
487 memcpy(&(pr
->power
.states
[current_count
]), &cx
, sizeof(cx
));
490 * We support total ACPI_PROCESSOR_MAX_POWER - 1
491 * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
493 if (current_count
>= (ACPI_PROCESSOR_MAX_POWER
- 1)) {
495 "Limiting number of power states to max (%d)\n",
496 ACPI_PROCESSOR_MAX_POWER
);
498 "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
503 ACPI_DEBUG_PRINT((ACPI_DB_INFO
, "Found %d power states\n",
506 /* Validate number of power states discovered */
507 if (current_count
< 2)
511 kfree(buffer
.pointer
);
516 static void acpi_processor_power_verify_c3(struct acpi_processor
*pr
,
517 struct acpi_processor_cx
*cx
)
519 static int bm_check_flag
= -1;
520 static int bm_control_flag
= -1;
527 * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
528 * DMA transfers are used by any ISA device to avoid livelock.
529 * Note that we could disable Type-F DMA (as recommended by
530 * the erratum), but this is known to disrupt certain ISA
531 * devices thus we take the conservative approach.
533 else if (errata
.piix4
.fdma
) {
534 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
535 "C3 not supported on PIIX4 with Type-F DMA\n"));
539 /* All the logic here assumes flags.bm_check is same across all CPUs */
540 if (bm_check_flag
== -1) {
541 /* Determine whether bm_check is needed based on CPU */
542 acpi_processor_power_init_bm_check(&(pr
->flags
), pr
->id
);
543 bm_check_flag
= pr
->flags
.bm_check
;
544 bm_control_flag
= pr
->flags
.bm_control
;
546 pr
->flags
.bm_check
= bm_check_flag
;
547 pr
->flags
.bm_control
= bm_control_flag
;
550 if (pr
->flags
.bm_check
) {
551 if (!pr
->flags
.bm_control
) {
552 if (pr
->flags
.has_cst
!= 1) {
553 /* bus mastering control is necessary */
554 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
555 "C3 support requires BM control\n"));
558 /* Here we enter C3 without bus mastering */
559 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
560 "C3 support without BM control\n"));
565 * WBINVD should be set in fadt, for C3 state to be
566 * supported on when bm_check is not required.
568 if (!(acpi_gbl_FADT
.flags
& ACPI_FADT_WBINVD
)) {
569 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
570 "Cache invalidation should work properly"
571 " for C3 to be enabled on SMP systems\n"));
577 * Otherwise we've met all of our C3 requirements.
578 * Normalize the C3 latency to expidite policy. Enable
579 * checking of bus mastering status (bm_check) so we can
580 * use this in our C3 policy
585 * On older chipsets, BM_RLD needs to be set
586 * in order for Bus Master activity to wake the
587 * system from C3. Newer chipsets handle DMA
588 * during C3 automatically and BM_RLD is a NOP.
589 * In either case, the proper way to
590 * handle BM_RLD is to set it and leave it set.
592 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD
, 1);
597 static int acpi_processor_power_verify(struct acpi_processor
*pr
)
600 unsigned int working
= 0;
602 pr
->power
.timer_broadcast_on_state
= INT_MAX
;
604 for (i
= 1; i
< ACPI_PROCESSOR_MAX_POWER
&& i
<= max_cstate
; i
++) {
605 struct acpi_processor_cx
*cx
= &pr
->power
.states
[i
];
619 acpi_processor_power_verify_c3(pr
, cx
);
625 lapic_timer_check_state(i
, pr
, cx
);
626 tsc_check_state(cx
->type
);
630 lapic_timer_propagate_broadcast(pr
);
635 static int acpi_processor_get_power_info(struct acpi_processor
*pr
)
641 /* NOTE: the idle thread may not be running while calling
644 /* Zero initialize all the C-states info. */
645 memset(pr
->power
.states
, 0, sizeof(pr
->power
.states
));
647 result
= acpi_processor_get_power_info_cst(pr
);
648 if (result
== -ENODEV
)
649 result
= acpi_processor_get_power_info_fadt(pr
);
654 acpi_processor_get_power_info_default(pr
);
656 pr
->power
.count
= acpi_processor_power_verify(pr
);
659 * if one state of type C2 or C3 is available, mark this
660 * CPU as being "idle manageable"
662 for (i
= 1; i
< ACPI_PROCESSOR_MAX_POWER
; i
++) {
663 if (pr
->power
.states
[i
].valid
) {
665 if (pr
->power
.states
[i
].type
>= ACPI_STATE_C2
)
674 * acpi_idle_bm_check - checks if bus master activity was detected
676 static int acpi_idle_bm_check(void)
680 if (bm_check_disable
)
683 acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS
, &bm_status
);
685 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS
, 1);
687 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
688 * the true state of bus mastering activity; forcing us to
689 * manually check the BMIDEA bit of each IDE channel.
691 else if (errata
.piix4
.bmisx
) {
692 if ((inb_p(errata
.piix4
.bmisx
+ 0x02) & 0x01)
693 || (inb_p(errata
.piix4
.bmisx
+ 0x0A) & 0x01))
700 * acpi_idle_do_entry - a helper function that does C2 and C3 type entry
703 * Caller disables interrupt before call and enables interrupt after return.
705 static inline void acpi_idle_do_entry(struct acpi_processor_cx
*cx
)
707 /* Don't trace irqs off for idle */
708 stop_critical_timings();
709 if (cx
->entry_method
== ACPI_CSTATE_FFH
) {
710 /* Call into architectural FFH based C-state */
711 acpi_processor_ffh_cstate_enter(cx
);
712 } else if (cx
->entry_method
== ACPI_CSTATE_HALT
) {
715 /* IO port based C-state */
717 /* Dummy wait op - must do something useless after P_LVL2 read
718 because chipsets cannot guarantee that STPCLK# signal
719 gets asserted in time to freeze execution properly. */
720 inl(acpi_gbl_FADT
.xpm_timer_block
.address
);
722 start_critical_timings();
726 * acpi_idle_enter_c1 - enters an ACPI C1 state-type
727 * @dev: the target CPU
728 * @drv: cpuidle driver containing cpuidle state info
729 * @index: index of target state
731 * This is equivalent to the HALT instruction.
733 static int acpi_idle_enter_c1(struct cpuidle_device
*dev
,
734 struct cpuidle_driver
*drv
, int index
)
738 struct acpi_processor
*pr
;
739 struct cpuidle_state_usage
*state_usage
= &dev
->states_usage
[index
];
740 struct acpi_processor_cx
*cx
= cpuidle_get_statedata(state_usage
);
742 pr
= __this_cpu_read(processors
);
743 dev
->last_residency
= 0;
751 lapic_timer_state_broadcast(pr
, cx
, 1);
752 kt1
= ktime_get_real();
753 acpi_idle_do_entry(cx
);
754 kt2
= ktime_get_real();
755 idle_time
= ktime_to_us(ktime_sub(kt2
, kt1
));
757 /* Update device last_residency*/
758 dev
->last_residency
= (int)idle_time
;
761 lapic_timer_state_broadcast(pr
, cx
, 0);
768 * acpi_idle_play_dead - enters an ACPI state for long-term idle (i.e. off-lining)
769 * @dev: the target CPU
770 * @index: the index of suggested state
772 static int acpi_idle_play_dead(struct cpuidle_device
*dev
, int index
)
774 struct cpuidle_state_usage
*state_usage
= &dev
->states_usage
[index
];
775 struct acpi_processor_cx
*cx
= cpuidle_get_statedata(state_usage
);
777 ACPI_FLUSH_CPU_CACHE();
781 if (cx
->entry_method
== ACPI_CSTATE_HALT
)
783 else if (cx
->entry_method
== ACPI_CSTATE_SYSTEMIO
) {
785 /* See comment in acpi_idle_do_entry() */
786 inl(acpi_gbl_FADT
.xpm_timer_block
.address
);
796 * acpi_idle_enter_simple - enters an ACPI state without BM handling
797 * @dev: the target CPU
798 * @drv: cpuidle driver with cpuidle state information
799 * @index: the index of suggested state
801 static int acpi_idle_enter_simple(struct cpuidle_device
*dev
,
802 struct cpuidle_driver
*drv
, int index
)
804 struct acpi_processor
*pr
;
805 struct cpuidle_state_usage
*state_usage
= &dev
->states_usage
[index
];
806 struct acpi_processor_cx
*cx
= cpuidle_get_statedata(state_usage
);
811 pr
= __this_cpu_read(processors
);
812 dev
->last_residency
= 0;
820 if (cx
->entry_method
!= ACPI_CSTATE_FFH
) {
821 current_thread_info()->status
&= ~TS_POLLING
;
823 * TS_POLLING-cleared state must be visible before we test
828 if (unlikely(need_resched())) {
829 current_thread_info()->status
|= TS_POLLING
;
836 * Must be done before busmaster disable as we might need to
839 lapic_timer_state_broadcast(pr
, cx
, 1);
841 if (cx
->type
== ACPI_STATE_C3
)
842 ACPI_FLUSH_CPU_CACHE();
844 kt1
= ktime_get_real();
845 /* Tell the scheduler that we are going deep-idle: */
846 sched_clock_idle_sleep_event();
847 acpi_idle_do_entry(cx
);
848 kt2
= ktime_get_real();
849 idle_time_ns
= ktime_to_ns(ktime_sub(kt2
, kt1
));
850 idle_time
= idle_time_ns
;
851 do_div(idle_time
, NSEC_PER_USEC
);
853 /* Update device last_residency*/
854 dev
->last_residency
= (int)idle_time
;
856 /* Tell the scheduler how much we idled: */
857 sched_clock_idle_wakeup_event(idle_time_ns
);
860 if (cx
->entry_method
!= ACPI_CSTATE_FFH
)
861 current_thread_info()->status
|= TS_POLLING
;
863 lapic_timer_state_broadcast(pr
, cx
, 0);
867 static int c3_cpu_count
;
868 static DEFINE_RAW_SPINLOCK(c3_lock
);
871 * acpi_idle_enter_bm - enters C3 with proper BM handling
872 * @dev: the target CPU
873 * @drv: cpuidle driver containing state data
874 * @index: the index of suggested state
876 * If BM is detected, the deepest non-C3 idle state is entered instead.
878 static int acpi_idle_enter_bm(struct cpuidle_device
*dev
,
879 struct cpuidle_driver
*drv
, int index
)
881 struct acpi_processor
*pr
;
882 struct cpuidle_state_usage
*state_usage
= &dev
->states_usage
[index
];
883 struct acpi_processor_cx
*cx
= cpuidle_get_statedata(state_usage
);
889 pr
= __this_cpu_read(processors
);
890 dev
->last_residency
= 0;
895 if (!cx
->bm_sts_skip
&& acpi_idle_bm_check()) {
896 if (drv
->safe_state_index
>= 0) {
897 return drv
->states
[drv
->safe_state_index
].enter(dev
,
898 drv
, drv
->safe_state_index
);
910 if (cx
->entry_method
!= ACPI_CSTATE_FFH
) {
911 current_thread_info()->status
&= ~TS_POLLING
;
913 * TS_POLLING-cleared state must be visible before we test
918 if (unlikely(need_resched())) {
919 current_thread_info()->status
|= TS_POLLING
;
925 acpi_unlazy_tlb(smp_processor_id());
927 /* Tell the scheduler that we are going deep-idle: */
928 sched_clock_idle_sleep_event();
930 * Must be done before busmaster disable as we might need to
933 lapic_timer_state_broadcast(pr
, cx
, 1);
935 kt1
= ktime_get_real();
938 * bm_check implies we need ARB_DIS
939 * !bm_check implies we need cache flush
940 * bm_control implies whether we can do ARB_DIS
942 * That leaves a case where bm_check is set and bm_control is
943 * not set. In that case we cannot do much, we enter C3
944 * without doing anything.
946 if (pr
->flags
.bm_check
&& pr
->flags
.bm_control
) {
947 raw_spin_lock(&c3_lock
);
949 /* Disable bus master arbitration when all CPUs are in C3 */
950 if (c3_cpu_count
== num_online_cpus())
951 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE
, 1);
952 raw_spin_unlock(&c3_lock
);
953 } else if (!pr
->flags
.bm_check
) {
954 ACPI_FLUSH_CPU_CACHE();
957 acpi_idle_do_entry(cx
);
959 /* Re-enable bus master arbitration */
960 if (pr
->flags
.bm_check
&& pr
->flags
.bm_control
) {
961 raw_spin_lock(&c3_lock
);
962 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE
, 0);
964 raw_spin_unlock(&c3_lock
);
966 kt2
= ktime_get_real();
967 idle_time_ns
= ktime_to_ns(ktime_sub(kt2
, kt1
));
968 idle_time
= idle_time_ns
;
969 do_div(idle_time
, NSEC_PER_USEC
);
971 /* Update device last_residency*/
972 dev
->last_residency
= (int)idle_time
;
974 /* Tell the scheduler how much we idled: */
975 sched_clock_idle_wakeup_event(idle_time_ns
);
978 if (cx
->entry_method
!= ACPI_CSTATE_FFH
)
979 current_thread_info()->status
|= TS_POLLING
;
981 lapic_timer_state_broadcast(pr
, cx
, 0);
985 struct cpuidle_driver acpi_idle_driver
= {
987 .owner
= THIS_MODULE
,
991 * acpi_processor_setup_cpuidle_cx - prepares and configures CPUIDLE
992 * device i.e. per-cpu data
994 * @pr: the ACPI processor
996 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor
*pr
)
998 int i
, count
= CPUIDLE_DRIVER_STATE_START
;
999 struct acpi_processor_cx
*cx
;
1000 struct cpuidle_state_usage
*state_usage
;
1001 struct cpuidle_device
*dev
= &pr
->power
.dev
;
1003 if (!pr
->flags
.power_setup_done
)
1006 if (pr
->flags
.power
== 0) {
1012 if (max_cstate
== 0)
1015 for (i
= 1; i
< ACPI_PROCESSOR_MAX_POWER
&& i
<= max_cstate
; i
++) {
1016 cx
= &pr
->power
.states
[i
];
1017 state_usage
= &dev
->states_usage
[count
];
1022 #ifdef CONFIG_HOTPLUG_CPU
1023 if ((cx
->type
!= ACPI_STATE_C1
) && (num_online_cpus() > 1) &&
1024 !pr
->flags
.has_cst
&&
1025 !(acpi_gbl_FADT
.flags
& ACPI_FADT_C2_MP_SUPPORTED
))
1029 cpuidle_set_statedata(state_usage
, cx
);
1032 if (count
== CPUIDLE_STATE_MAX
)
1036 dev
->state_count
= count
;
1045 * acpi_processor_setup_cpuidle states- prepares and configures cpuidle
1046 * global state data i.e. idle routines
1048 * @pr: the ACPI processor
1050 static int acpi_processor_setup_cpuidle_states(struct acpi_processor
*pr
)
1052 int i
, count
= CPUIDLE_DRIVER_STATE_START
;
1053 struct acpi_processor_cx
*cx
;
1054 struct cpuidle_state
*state
;
1055 struct cpuidle_driver
*drv
= &acpi_idle_driver
;
1057 if (!pr
->flags
.power_setup_done
)
1060 if (pr
->flags
.power
== 0)
1063 drv
->safe_state_index
= -1;
1064 for (i
= 0; i
< CPUIDLE_STATE_MAX
; i
++) {
1065 drv
->states
[i
].name
[0] = '\0';
1066 drv
->states
[i
].desc
[0] = '\0';
1069 if (max_cstate
== 0)
1072 for (i
= 1; i
< ACPI_PROCESSOR_MAX_POWER
&& i
<= max_cstate
; i
++) {
1073 cx
= &pr
->power
.states
[i
];
1078 #ifdef CONFIG_HOTPLUG_CPU
1079 if ((cx
->type
!= ACPI_STATE_C1
) && (num_online_cpus() > 1) &&
1080 !pr
->flags
.has_cst
&&
1081 !(acpi_gbl_FADT
.flags
& ACPI_FADT_C2_MP_SUPPORTED
))
1085 state
= &drv
->states
[count
];
1086 snprintf(state
->name
, CPUIDLE_NAME_LEN
, "C%d", i
);
1087 strncpy(state
->desc
, cx
->desc
, CPUIDLE_DESC_LEN
);
1088 state
->exit_latency
= cx
->latency
;
1089 state
->target_residency
= cx
->latency
* latency_factor
;
1094 if (cx
->entry_method
== ACPI_CSTATE_FFH
)
1095 state
->flags
|= CPUIDLE_FLAG_TIME_VALID
;
1097 state
->enter
= acpi_idle_enter_c1
;
1098 state
->enter_dead
= acpi_idle_play_dead
;
1099 drv
->safe_state_index
= count
;
1103 state
->flags
|= CPUIDLE_FLAG_TIME_VALID
;
1104 state
->enter
= acpi_idle_enter_simple
;
1105 state
->enter_dead
= acpi_idle_play_dead
;
1106 drv
->safe_state_index
= count
;
1110 state
->flags
|= CPUIDLE_FLAG_TIME_VALID
;
1111 state
->enter
= pr
->flags
.bm_check
?
1112 acpi_idle_enter_bm
:
1113 acpi_idle_enter_simple
;
1118 if (count
== CPUIDLE_STATE_MAX
)
1122 drv
->state_count
= count
;
1130 int acpi_processor_hotplug(struct acpi_processor
*pr
)
1134 if (disabled_by_idle_boot_param())
1144 if (!pr
->flags
.power_setup_done
)
1147 cpuidle_pause_and_lock();
1148 cpuidle_disable_device(&pr
->power
.dev
);
1149 acpi_processor_get_power_info(pr
);
1150 if (pr
->flags
.power
) {
1151 acpi_processor_setup_cpuidle_cx(pr
);
1152 ret
= cpuidle_enable_device(&pr
->power
.dev
);
1154 cpuidle_resume_and_unlock();
1159 int acpi_processor_cst_has_changed(struct acpi_processor
*pr
)
1162 struct acpi_processor
*_pr
;
1164 if (disabled_by_idle_boot_param())
1173 if (!pr
->flags
.power_setup_done
)
1177 * FIXME: Design the ACPI notification to make it once per
1178 * system instead of once per-cpu. This condition is a hack
1179 * to make the code that updates C-States be called once.
1182 if (pr
->id
== 0 && cpuidle_get_driver() == &acpi_idle_driver
) {
1184 cpuidle_pause_and_lock();
1185 /* Protect against cpu-hotplug */
1188 /* Disable all cpuidle devices */
1189 for_each_online_cpu(cpu
) {
1190 _pr
= per_cpu(processors
, cpu
);
1191 if (!_pr
|| !_pr
->flags
.power_setup_done
)
1193 cpuidle_disable_device(&_pr
->power
.dev
);
1196 /* Populate Updated C-state information */
1197 acpi_processor_setup_cpuidle_states(pr
);
1199 /* Enable all cpuidle devices */
1200 for_each_online_cpu(cpu
) {
1201 _pr
= per_cpu(processors
, cpu
);
1202 if (!_pr
|| !_pr
->flags
.power_setup_done
)
1204 acpi_processor_get_power_info(_pr
);
1205 if (_pr
->flags
.power
) {
1206 acpi_processor_setup_cpuidle_cx(_pr
);
1207 cpuidle_enable_device(&_pr
->power
.dev
);
1211 cpuidle_resume_and_unlock();
1217 static int acpi_processor_registered
;
1219 int __cpuinit
acpi_processor_power_init(struct acpi_processor
*pr
,
1220 struct acpi_device
*device
)
1222 acpi_status status
= 0;
1224 static int first_run
;
1226 if (disabled_by_idle_boot_param())
1230 dmi_check_system(processor_power_dmi_table
);
1231 max_cstate
= acpi_processor_cstate_check(max_cstate
);
1232 if (max_cstate
< ACPI_C_STATES_MAX
)
1234 "ACPI: processor limited to max C-state %d\n",
1242 if (acpi_gbl_FADT
.cst_control
&& !nocst
) {
1244 acpi_os_write_port(acpi_gbl_FADT
.smi_command
, acpi_gbl_FADT
.cst_control
, 8);
1245 if (ACPI_FAILURE(status
)) {
1246 ACPI_EXCEPTION((AE_INFO
, status
,
1247 "Notifying BIOS of _CST ability failed"));
1251 acpi_processor_get_power_info(pr
);
1252 pr
->flags
.power_setup_done
= 1;
1255 * Install the idle handler if processor power management is supported.
1256 * Note that we use previously set idle handler will be used on
1257 * platforms that only support C1.
1259 if (pr
->flags
.power
) {
1260 /* Register acpi_idle_driver if not already registered */
1261 if (!acpi_processor_registered
) {
1262 acpi_processor_setup_cpuidle_states(pr
);
1263 retval
= cpuidle_register_driver(&acpi_idle_driver
);
1266 printk(KERN_DEBUG
"ACPI: %s registered with cpuidle\n",
1267 acpi_idle_driver
.name
);
1269 /* Register per-cpu cpuidle_device. Cpuidle driver
1270 * must already be registered before registering device
1272 acpi_processor_setup_cpuidle_cx(pr
);
1273 retval
= cpuidle_register_device(&pr
->power
.dev
);
1275 if (acpi_processor_registered
== 0)
1276 cpuidle_unregister_driver(&acpi_idle_driver
);
1279 acpi_processor_registered
++;
1284 int acpi_processor_power_exit(struct acpi_processor
*pr
,
1285 struct acpi_device
*device
)
1287 if (disabled_by_idle_boot_param())
1290 if (pr
->flags
.power
) {
1291 cpuidle_unregister_device(&pr
->power
.dev
);
1292 acpi_processor_registered
--;
1293 if (acpi_processor_registered
== 0)
1294 cpuidle_unregister_driver(&acpi_idle_driver
);
1297 pr
->flags
.power_setup_done
= 0;