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/module.h>
32 #include <linux/acpi.h>
33 #include <linux/dmi.h>
34 #include <linux/sched.h> /* need_resched() */
35 #include <linux/clockchips.h>
36 #include <linux/cpuidle.h>
37 #include <linux/syscore_ops.h>
40 * Include the apic definitions for x86 to have the APIC timer related defines
41 * available also for UP (on SMP it gets magically included via linux/smp.h).
42 * asm/acpi.h is not an option, as it would require more include magic. Also
43 * creating an empty asm-ia64/apic.h would just trade pest vs. cholera.
49 #include <acpi/acpi_bus.h>
50 #include <acpi/processor.h>
52 #define PREFIX "ACPI: "
54 #define ACPI_PROCESSOR_CLASS "processor"
55 #define _COMPONENT ACPI_PROCESSOR_COMPONENT
56 ACPI_MODULE_NAME("processor_idle");
58 static unsigned int max_cstate __read_mostly
= ACPI_PROCESSOR_MAX_POWER
;
59 module_param(max_cstate
, uint
, 0000);
60 static unsigned int nocst __read_mostly
;
61 module_param(nocst
, uint
, 0000);
62 static int bm_check_disable __read_mostly
;
63 module_param(bm_check_disable
, uint
, 0000);
65 static unsigned int latency_factor __read_mostly
= 2;
66 module_param(latency_factor
, uint
, 0644);
68 static DEFINE_PER_CPU(struct cpuidle_device
*, acpi_cpuidle_device
);
70 static DEFINE_PER_CPU(struct acpi_processor_cx
* [CPUIDLE_STATE_MAX
],
73 static int disabled_by_idle_boot_param(void)
75 return boot_option_idle_override
== IDLE_POLL
||
76 boot_option_idle_override
== IDLE_HALT
;
80 * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
81 * For now disable this. Probably a bug somewhere else.
83 * To skip this limit, boot/load with a large max_cstate limit.
85 static int set_max_cstate(const struct dmi_system_id
*id
)
87 if (max_cstate
> ACPI_PROCESSOR_MAX_POWER
)
90 printk(KERN_NOTICE PREFIX
"%s detected - limiting to C%ld max_cstate."
91 " Override with \"processor.max_cstate=%d\"\n", id
->ident
,
92 (long)id
->driver_data
, ACPI_PROCESSOR_MAX_POWER
+ 1);
94 max_cstate
= (long)id
->driver_data
;
99 /* Actually this shouldn't be __cpuinitdata, would be better to fix the
100 callers to only run once -AK */
101 static struct dmi_system_id __cpuinitdata processor_power_dmi_table
[] = {
102 { set_max_cstate
, "Clevo 5600D", {
103 DMI_MATCH(DMI_BIOS_VENDOR
,"Phoenix Technologies LTD"),
104 DMI_MATCH(DMI_BIOS_VERSION
,"SHE845M0.86C.0013.D.0302131307")},
106 { set_max_cstate
, "Pavilion zv5000", {
107 DMI_MATCH(DMI_SYS_VENDOR
, "Hewlett-Packard"),
108 DMI_MATCH(DMI_PRODUCT_NAME
,"Pavilion zv5000 (DS502A#ABA)")},
110 { set_max_cstate
, "Asus L8400B", {
111 DMI_MATCH(DMI_SYS_VENDOR
, "ASUSTeK Computer Inc."),
112 DMI_MATCH(DMI_PRODUCT_NAME
,"L8400B series Notebook PC")},
119 * Callers should disable interrupts before the call and enable
120 * interrupts after return.
122 static void acpi_safe_halt(void)
124 if (!tif_need_resched()) {
130 #ifdef ARCH_APICTIMER_STOPS_ON_C3
133 * Some BIOS implementations switch to C3 in the published C2 state.
134 * This seems to be a common problem on AMD boxen, but other vendors
135 * are affected too. We pick the most conservative approach: we assume
136 * that the local APIC stops in both C2 and C3.
138 static void lapic_timer_check_state(int state
, struct acpi_processor
*pr
,
139 struct acpi_processor_cx
*cx
)
141 struct acpi_processor_power
*pwr
= &pr
->power
;
142 u8 type
= local_apic_timer_c2_ok
? ACPI_STATE_C3
: ACPI_STATE_C2
;
144 if (cpu_has(&cpu_data(pr
->id
), X86_FEATURE_ARAT
))
147 if (amd_e400_c1e_detected
)
148 type
= ACPI_STATE_C1
;
151 * Check, if one of the previous states already marked the lapic
154 if (pwr
->timer_broadcast_on_state
< state
)
157 if (cx
->type
>= type
)
158 pr
->power
.timer_broadcast_on_state
= state
;
161 static void __lapic_timer_propagate_broadcast(void *arg
)
163 struct acpi_processor
*pr
= (struct acpi_processor
*) arg
;
164 unsigned long reason
;
166 reason
= pr
->power
.timer_broadcast_on_state
< INT_MAX
?
167 CLOCK_EVT_NOTIFY_BROADCAST_ON
: CLOCK_EVT_NOTIFY_BROADCAST_OFF
;
169 clockevents_notify(reason
, &pr
->id
);
172 static void lapic_timer_propagate_broadcast(struct acpi_processor
*pr
)
174 smp_call_function_single(pr
->id
, __lapic_timer_propagate_broadcast
,
178 /* Power(C) State timer broadcast control */
179 static void lapic_timer_state_broadcast(struct acpi_processor
*pr
,
180 struct acpi_processor_cx
*cx
,
183 int state
= cx
- pr
->power
.states
;
185 if (state
>= pr
->power
.timer_broadcast_on_state
) {
186 unsigned long reason
;
188 reason
= broadcast
? CLOCK_EVT_NOTIFY_BROADCAST_ENTER
:
189 CLOCK_EVT_NOTIFY_BROADCAST_EXIT
;
190 clockevents_notify(reason
, &pr
->id
);
196 static void lapic_timer_check_state(int state
, struct acpi_processor
*pr
,
197 struct acpi_processor_cx
*cstate
) { }
198 static void lapic_timer_propagate_broadcast(struct acpi_processor
*pr
) { }
199 static void lapic_timer_state_broadcast(struct acpi_processor
*pr
,
200 struct acpi_processor_cx
*cx
,
207 #ifdef CONFIG_PM_SLEEP
208 static u32 saved_bm_rld
;
210 int acpi_processor_suspend(void)
212 acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD
, &saved_bm_rld
);
216 void acpi_processor_resume(void)
220 acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD
, &resumed_bm_rld
);
221 if (resumed_bm_rld
== saved_bm_rld
)
224 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD
, saved_bm_rld
);
227 static struct syscore_ops acpi_processor_syscore_ops
= {
228 .suspend
= acpi_processor_suspend
,
229 .resume
= acpi_processor_resume
,
232 void acpi_processor_syscore_init(void)
234 register_syscore_ops(&acpi_processor_syscore_ops
);
237 void acpi_processor_syscore_exit(void)
239 unregister_syscore_ops(&acpi_processor_syscore_ops
);
241 #endif /* CONFIG_PM_SLEEP */
243 #if defined(CONFIG_X86)
244 static void tsc_check_state(int state
)
246 switch (boot_cpu_data
.x86_vendor
) {
248 case X86_VENDOR_INTEL
:
250 * AMD Fam10h TSC will tick in all
251 * C/P/S0/S1 states when this bit is set.
253 if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC
))
258 /* TSC could halt in idle, so notify users */
259 if (state
> ACPI_STATE_C1
)
260 mark_tsc_unstable("TSC halts in idle");
264 static void tsc_check_state(int state
) { return; }
267 static int acpi_processor_get_power_info_fadt(struct acpi_processor
*pr
)
276 /* if info is obtained from pblk/fadt, type equals state */
277 pr
->power
.states
[ACPI_STATE_C2
].type
= ACPI_STATE_C2
;
278 pr
->power
.states
[ACPI_STATE_C3
].type
= ACPI_STATE_C3
;
280 #ifndef CONFIG_HOTPLUG_CPU
282 * Check for P_LVL2_UP flag before entering C2 and above on
285 if ((num_online_cpus() > 1) &&
286 !(acpi_gbl_FADT
.flags
& ACPI_FADT_C2_MP_SUPPORTED
))
290 /* determine C2 and C3 address from pblk */
291 pr
->power
.states
[ACPI_STATE_C2
].address
= pr
->pblk
+ 4;
292 pr
->power
.states
[ACPI_STATE_C3
].address
= pr
->pblk
+ 5;
294 /* determine latencies from FADT */
295 pr
->power
.states
[ACPI_STATE_C2
].latency
= acpi_gbl_FADT
.c2_latency
;
296 pr
->power
.states
[ACPI_STATE_C3
].latency
= acpi_gbl_FADT
.c3_latency
;
299 * FADT specified C2 latency must be less than or equal to
302 if (acpi_gbl_FADT
.c2_latency
> ACPI_PROCESSOR_MAX_C2_LATENCY
) {
303 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
304 "C2 latency too large [%d]\n", acpi_gbl_FADT
.c2_latency
));
306 pr
->power
.states
[ACPI_STATE_C2
].address
= 0;
310 * FADT supplied C3 latency must be less than or equal to
313 if (acpi_gbl_FADT
.c3_latency
> ACPI_PROCESSOR_MAX_C3_LATENCY
) {
314 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
315 "C3 latency too large [%d]\n", acpi_gbl_FADT
.c3_latency
));
317 pr
->power
.states
[ACPI_STATE_C3
].address
= 0;
320 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
321 "lvl2[0x%08x] lvl3[0x%08x]\n",
322 pr
->power
.states
[ACPI_STATE_C2
].address
,
323 pr
->power
.states
[ACPI_STATE_C3
].address
));
328 static int acpi_processor_get_power_info_default(struct acpi_processor
*pr
)
330 if (!pr
->power
.states
[ACPI_STATE_C1
].valid
) {
331 /* set the first C-State to C1 */
332 /* all processors need to support C1 */
333 pr
->power
.states
[ACPI_STATE_C1
].type
= ACPI_STATE_C1
;
334 pr
->power
.states
[ACPI_STATE_C1
].valid
= 1;
335 pr
->power
.states
[ACPI_STATE_C1
].entry_method
= ACPI_CSTATE_HALT
;
337 /* the C0 state only exists as a filler in our array */
338 pr
->power
.states
[ACPI_STATE_C0
].valid
= 1;
342 static int acpi_processor_get_power_info_cst(struct acpi_processor
*pr
)
344 acpi_status status
= 0;
348 struct acpi_buffer buffer
= { ACPI_ALLOCATE_BUFFER
, NULL
};
349 union acpi_object
*cst
;
357 status
= acpi_evaluate_object(pr
->handle
, "_CST", NULL
, &buffer
);
358 if (ACPI_FAILURE(status
)) {
359 ACPI_DEBUG_PRINT((ACPI_DB_INFO
, "No _CST, giving up\n"));
363 cst
= buffer
.pointer
;
365 /* There must be at least 2 elements */
366 if (!cst
|| (cst
->type
!= ACPI_TYPE_PACKAGE
) || cst
->package
.count
< 2) {
367 printk(KERN_ERR PREFIX
"not enough elements in _CST\n");
372 count
= cst
->package
.elements
[0].integer
.value
;
374 /* Validate number of power states. */
375 if (count
< 1 || count
!= cst
->package
.count
- 1) {
376 printk(KERN_ERR PREFIX
"count given by _CST is not valid\n");
381 /* Tell driver that at least _CST is supported. */
382 pr
->flags
.has_cst
= 1;
384 for (i
= 1; i
<= count
; i
++) {
385 union acpi_object
*element
;
386 union acpi_object
*obj
;
387 struct acpi_power_register
*reg
;
388 struct acpi_processor_cx cx
;
390 memset(&cx
, 0, sizeof(cx
));
392 element
= &(cst
->package
.elements
[i
]);
393 if (element
->type
!= ACPI_TYPE_PACKAGE
)
396 if (element
->package
.count
!= 4)
399 obj
= &(element
->package
.elements
[0]);
401 if (obj
->type
!= ACPI_TYPE_BUFFER
)
404 reg
= (struct acpi_power_register
*)obj
->buffer
.pointer
;
406 if (reg
->space_id
!= ACPI_ADR_SPACE_SYSTEM_IO
&&
407 (reg
->space_id
!= ACPI_ADR_SPACE_FIXED_HARDWARE
))
410 /* There should be an easy way to extract an integer... */
411 obj
= &(element
->package
.elements
[1]);
412 if (obj
->type
!= ACPI_TYPE_INTEGER
)
415 cx
.type
= obj
->integer
.value
;
417 * Some buggy BIOSes won't list C1 in _CST -
418 * Let acpi_processor_get_power_info_default() handle them later
420 if (i
== 1 && cx
.type
!= ACPI_STATE_C1
)
423 cx
.address
= reg
->address
;
424 cx
.index
= current_count
+ 1;
426 cx
.entry_method
= ACPI_CSTATE_SYSTEMIO
;
427 if (reg
->space_id
== ACPI_ADR_SPACE_FIXED_HARDWARE
) {
428 if (acpi_processor_ffh_cstate_probe
429 (pr
->id
, &cx
, reg
) == 0) {
430 cx
.entry_method
= ACPI_CSTATE_FFH
;
431 } else if (cx
.type
== ACPI_STATE_C1
) {
433 * C1 is a special case where FIXED_HARDWARE
434 * can be handled in non-MWAIT way as well.
435 * In that case, save this _CST entry info.
436 * Otherwise, ignore this info and continue.
438 cx
.entry_method
= ACPI_CSTATE_HALT
;
439 snprintf(cx
.desc
, ACPI_CX_DESC_LEN
, "ACPI HLT");
443 if (cx
.type
== ACPI_STATE_C1
&&
444 (boot_option_idle_override
== IDLE_NOMWAIT
)) {
446 * In most cases the C1 space_id obtained from
447 * _CST object is FIXED_HARDWARE access mode.
448 * But when the option of idle=halt is added,
449 * the entry_method type should be changed from
450 * CSTATE_FFH to CSTATE_HALT.
451 * When the option of idle=nomwait is added,
452 * the C1 entry_method type should be
455 cx
.entry_method
= ACPI_CSTATE_HALT
;
456 snprintf(cx
.desc
, ACPI_CX_DESC_LEN
, "ACPI HLT");
459 snprintf(cx
.desc
, ACPI_CX_DESC_LEN
, "ACPI IOPORT 0x%x",
463 if (cx
.type
== ACPI_STATE_C1
) {
467 obj
= &(element
->package
.elements
[2]);
468 if (obj
->type
!= ACPI_TYPE_INTEGER
)
471 cx
.latency
= obj
->integer
.value
;
473 obj
= &(element
->package
.elements
[3]);
474 if (obj
->type
!= ACPI_TYPE_INTEGER
)
478 memcpy(&(pr
->power
.states
[current_count
]), &cx
, sizeof(cx
));
481 * We support total ACPI_PROCESSOR_MAX_POWER - 1
482 * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
484 if (current_count
>= (ACPI_PROCESSOR_MAX_POWER
- 1)) {
486 "Limiting number of power states to max (%d)\n",
487 ACPI_PROCESSOR_MAX_POWER
);
489 "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
494 ACPI_DEBUG_PRINT((ACPI_DB_INFO
, "Found %d power states\n",
497 /* Validate number of power states discovered */
498 if (current_count
< 2)
502 kfree(buffer
.pointer
);
507 static void acpi_processor_power_verify_c3(struct acpi_processor
*pr
,
508 struct acpi_processor_cx
*cx
)
510 static int bm_check_flag
= -1;
511 static int bm_control_flag
= -1;
518 * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
519 * DMA transfers are used by any ISA device to avoid livelock.
520 * Note that we could disable Type-F DMA (as recommended by
521 * the erratum), but this is known to disrupt certain ISA
522 * devices thus we take the conservative approach.
524 else if (errata
.piix4
.fdma
) {
525 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
526 "C3 not supported on PIIX4 with Type-F DMA\n"));
530 /* All the logic here assumes flags.bm_check is same across all CPUs */
531 if (bm_check_flag
== -1) {
532 /* Determine whether bm_check is needed based on CPU */
533 acpi_processor_power_init_bm_check(&(pr
->flags
), pr
->id
);
534 bm_check_flag
= pr
->flags
.bm_check
;
535 bm_control_flag
= pr
->flags
.bm_control
;
537 pr
->flags
.bm_check
= bm_check_flag
;
538 pr
->flags
.bm_control
= bm_control_flag
;
541 if (pr
->flags
.bm_check
) {
542 if (!pr
->flags
.bm_control
) {
543 if (pr
->flags
.has_cst
!= 1) {
544 /* bus mastering control is necessary */
545 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
546 "C3 support requires BM control\n"));
549 /* Here we enter C3 without bus mastering */
550 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
551 "C3 support without BM control\n"));
556 * WBINVD should be set in fadt, for C3 state to be
557 * supported on when bm_check is not required.
559 if (!(acpi_gbl_FADT
.flags
& ACPI_FADT_WBINVD
)) {
560 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
561 "Cache invalidation should work properly"
562 " for C3 to be enabled on SMP systems\n"));
568 * Otherwise we've met all of our C3 requirements.
569 * Normalize the C3 latency to expidite policy. Enable
570 * checking of bus mastering status (bm_check) so we can
571 * use this in our C3 policy
576 * On older chipsets, BM_RLD needs to be set
577 * in order for Bus Master activity to wake the
578 * system from C3. Newer chipsets handle DMA
579 * during C3 automatically and BM_RLD is a NOP.
580 * In either case, the proper way to
581 * handle BM_RLD is to set it and leave it set.
583 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD
, 1);
588 static int acpi_processor_power_verify(struct acpi_processor
*pr
)
591 unsigned int working
= 0;
593 pr
->power
.timer_broadcast_on_state
= INT_MAX
;
595 for (i
= 1; i
< ACPI_PROCESSOR_MAX_POWER
&& i
<= max_cstate
; i
++) {
596 struct acpi_processor_cx
*cx
= &pr
->power
.states
[i
];
610 acpi_processor_power_verify_c3(pr
, cx
);
616 lapic_timer_check_state(i
, pr
, cx
);
617 tsc_check_state(cx
->type
);
621 lapic_timer_propagate_broadcast(pr
);
626 static int acpi_processor_get_power_info(struct acpi_processor
*pr
)
632 /* NOTE: the idle thread may not be running while calling
635 /* Zero initialize all the C-states info. */
636 memset(pr
->power
.states
, 0, sizeof(pr
->power
.states
));
638 result
= acpi_processor_get_power_info_cst(pr
);
639 if (result
== -ENODEV
)
640 result
= acpi_processor_get_power_info_fadt(pr
);
645 acpi_processor_get_power_info_default(pr
);
647 pr
->power
.count
= acpi_processor_power_verify(pr
);
650 * if one state of type C2 or C3 is available, mark this
651 * CPU as being "idle manageable"
653 for (i
= 1; i
< ACPI_PROCESSOR_MAX_POWER
; i
++) {
654 if (pr
->power
.states
[i
].valid
) {
656 if (pr
->power
.states
[i
].type
>= ACPI_STATE_C2
)
665 * acpi_idle_bm_check - checks if bus master activity was detected
667 static int acpi_idle_bm_check(void)
671 if (bm_check_disable
)
674 acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS
, &bm_status
);
676 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS
, 1);
678 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
679 * the true state of bus mastering activity; forcing us to
680 * manually check the BMIDEA bit of each IDE channel.
682 else if (errata
.piix4
.bmisx
) {
683 if ((inb_p(errata
.piix4
.bmisx
+ 0x02) & 0x01)
684 || (inb_p(errata
.piix4
.bmisx
+ 0x0A) & 0x01))
691 * acpi_idle_do_entry - a helper function that does C2 and C3 type entry
694 * Caller disables interrupt before call and enables interrupt after return.
696 static inline void acpi_idle_do_entry(struct acpi_processor_cx
*cx
)
698 /* Don't trace irqs off for idle */
699 stop_critical_timings();
700 if (cx
->entry_method
== ACPI_CSTATE_FFH
) {
701 /* Call into architectural FFH based C-state */
702 acpi_processor_ffh_cstate_enter(cx
);
703 } else if (cx
->entry_method
== ACPI_CSTATE_HALT
) {
706 /* IO port based C-state */
708 /* Dummy wait op - must do something useless after P_LVL2 read
709 because chipsets cannot guarantee that STPCLK# signal
710 gets asserted in time to freeze execution properly. */
711 inl(acpi_gbl_FADT
.xpm_timer_block
.address
);
713 start_critical_timings();
717 * acpi_idle_enter_c1 - enters an ACPI C1 state-type
718 * @dev: the target CPU
719 * @drv: cpuidle driver containing cpuidle state info
720 * @index: index of target state
722 * This is equivalent to the HALT instruction.
724 static int acpi_idle_enter_c1(struct cpuidle_device
*dev
,
725 struct cpuidle_driver
*drv
, int index
)
727 struct acpi_processor
*pr
;
728 struct acpi_processor_cx
*cx
= per_cpu(acpi_cstate
[index
], dev
->cpu
);
730 pr
= __this_cpu_read(processors
);
735 if (cx
->entry_method
== ACPI_CSTATE_FFH
) {
736 if (current_set_polling_and_test())
740 lapic_timer_state_broadcast(pr
, cx
, 1);
741 acpi_idle_do_entry(cx
);
743 lapic_timer_state_broadcast(pr
, cx
, 0);
750 * acpi_idle_play_dead - enters an ACPI state for long-term idle (i.e. off-lining)
751 * @dev: the target CPU
752 * @index: the index of suggested state
754 static int acpi_idle_play_dead(struct cpuidle_device
*dev
, int index
)
756 struct acpi_processor_cx
*cx
= per_cpu(acpi_cstate
[index
], dev
->cpu
);
758 ACPI_FLUSH_CPU_CACHE();
762 if (cx
->entry_method
== ACPI_CSTATE_HALT
)
764 else if (cx
->entry_method
== ACPI_CSTATE_SYSTEMIO
) {
766 /* See comment in acpi_idle_do_entry() */
767 inl(acpi_gbl_FADT
.xpm_timer_block
.address
);
777 * acpi_idle_enter_simple - enters an ACPI state without BM handling
778 * @dev: the target CPU
779 * @drv: cpuidle driver with cpuidle state information
780 * @index: the index of suggested state
782 static int acpi_idle_enter_simple(struct cpuidle_device
*dev
,
783 struct cpuidle_driver
*drv
, int index
)
785 struct acpi_processor
*pr
;
786 struct acpi_processor_cx
*cx
= per_cpu(acpi_cstate
[index
], dev
->cpu
);
788 pr
= __this_cpu_read(processors
);
793 if (cx
->entry_method
== ACPI_CSTATE_FFH
) {
794 if (current_set_polling_and_test())
799 * Must be done before busmaster disable as we might need to
802 lapic_timer_state_broadcast(pr
, cx
, 1);
804 if (cx
->type
== ACPI_STATE_C3
)
805 ACPI_FLUSH_CPU_CACHE();
807 /* Tell the scheduler that we are going deep-idle: */
808 sched_clock_idle_sleep_event();
809 acpi_idle_do_entry(cx
);
811 sched_clock_idle_wakeup_event(0);
813 lapic_timer_state_broadcast(pr
, cx
, 0);
817 static int c3_cpu_count
;
818 static DEFINE_RAW_SPINLOCK(c3_lock
);
821 * acpi_idle_enter_bm - enters C3 with proper BM handling
822 * @dev: the target CPU
823 * @drv: cpuidle driver containing state data
824 * @index: the index of suggested state
826 * If BM is detected, the deepest non-C3 idle state is entered instead.
828 static int acpi_idle_enter_bm(struct cpuidle_device
*dev
,
829 struct cpuidle_driver
*drv
, int index
)
831 struct acpi_processor
*pr
;
832 struct acpi_processor_cx
*cx
= per_cpu(acpi_cstate
[index
], dev
->cpu
);
834 pr
= __this_cpu_read(processors
);
839 if (!cx
->bm_sts_skip
&& acpi_idle_bm_check()) {
840 if (drv
->safe_state_index
>= 0) {
841 return drv
->states
[drv
->safe_state_index
].enter(dev
,
842 drv
, drv
->safe_state_index
);
849 if (cx
->entry_method
== ACPI_CSTATE_FFH
) {
850 if (current_set_polling_and_test())
854 acpi_unlazy_tlb(smp_processor_id());
856 /* Tell the scheduler that we are going deep-idle: */
857 sched_clock_idle_sleep_event();
859 * Must be done before busmaster disable as we might need to
862 lapic_timer_state_broadcast(pr
, cx
, 1);
866 * bm_check implies we need ARB_DIS
867 * !bm_check implies we need cache flush
868 * bm_control implies whether we can do ARB_DIS
870 * That leaves a case where bm_check is set and bm_control is
871 * not set. In that case we cannot do much, we enter C3
872 * without doing anything.
874 if (pr
->flags
.bm_check
&& pr
->flags
.bm_control
) {
875 raw_spin_lock(&c3_lock
);
877 /* Disable bus master arbitration when all CPUs are in C3 */
878 if (c3_cpu_count
== num_online_cpus())
879 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE
, 1);
880 raw_spin_unlock(&c3_lock
);
881 } else if (!pr
->flags
.bm_check
) {
882 ACPI_FLUSH_CPU_CACHE();
885 acpi_idle_do_entry(cx
);
887 /* Re-enable bus master arbitration */
888 if (pr
->flags
.bm_check
&& pr
->flags
.bm_control
) {
889 raw_spin_lock(&c3_lock
);
890 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE
, 0);
892 raw_spin_unlock(&c3_lock
);
895 sched_clock_idle_wakeup_event(0);
897 lapic_timer_state_broadcast(pr
, cx
, 0);
901 struct cpuidle_driver acpi_idle_driver
= {
903 .owner
= THIS_MODULE
,
907 * acpi_processor_setup_cpuidle_cx - prepares and configures CPUIDLE
908 * device i.e. per-cpu data
910 * @pr: the ACPI processor
911 * @dev : the cpuidle device
913 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor
*pr
,
914 struct cpuidle_device
*dev
)
916 int i
, count
= CPUIDLE_DRIVER_STATE_START
;
917 struct acpi_processor_cx
*cx
;
919 if (!pr
->flags
.power_setup_done
)
922 if (pr
->flags
.power
== 0) {
934 for (i
= 1; i
< ACPI_PROCESSOR_MAX_POWER
&& i
<= max_cstate
; i
++) {
935 cx
= &pr
->power
.states
[i
];
940 #ifdef CONFIG_HOTPLUG_CPU
941 if ((cx
->type
!= ACPI_STATE_C1
) && (num_online_cpus() > 1) &&
942 !pr
->flags
.has_cst
&&
943 !(acpi_gbl_FADT
.flags
& ACPI_FADT_C2_MP_SUPPORTED
))
946 per_cpu(acpi_cstate
[count
], dev
->cpu
) = cx
;
949 if (count
== CPUIDLE_STATE_MAX
)
953 dev
->state_count
= count
;
962 * acpi_processor_setup_cpuidle states- prepares and configures cpuidle
963 * global state data i.e. idle routines
965 * @pr: the ACPI processor
967 static int acpi_processor_setup_cpuidle_states(struct acpi_processor
*pr
)
969 int i
, count
= CPUIDLE_DRIVER_STATE_START
;
970 struct acpi_processor_cx
*cx
;
971 struct cpuidle_state
*state
;
972 struct cpuidle_driver
*drv
= &acpi_idle_driver
;
974 if (!pr
->flags
.power_setup_done
)
977 if (pr
->flags
.power
== 0)
980 drv
->safe_state_index
= -1;
981 for (i
= CPUIDLE_DRIVER_STATE_START
; i
< CPUIDLE_STATE_MAX
; i
++) {
982 drv
->states
[i
].name
[0] = '\0';
983 drv
->states
[i
].desc
[0] = '\0';
989 for (i
= 1; i
< ACPI_PROCESSOR_MAX_POWER
&& i
<= max_cstate
; i
++) {
990 cx
= &pr
->power
.states
[i
];
995 #ifdef CONFIG_HOTPLUG_CPU
996 if ((cx
->type
!= ACPI_STATE_C1
) && (num_online_cpus() > 1) &&
997 !pr
->flags
.has_cst
&&
998 !(acpi_gbl_FADT
.flags
& ACPI_FADT_C2_MP_SUPPORTED
))
1002 state
= &drv
->states
[count
];
1003 snprintf(state
->name
, CPUIDLE_NAME_LEN
, "C%d", i
);
1004 strncpy(state
->desc
, cx
->desc
, CPUIDLE_DESC_LEN
);
1005 state
->exit_latency
= cx
->latency
;
1006 state
->target_residency
= cx
->latency
* latency_factor
;
1011 if (cx
->entry_method
== ACPI_CSTATE_FFH
)
1012 state
->flags
|= CPUIDLE_FLAG_TIME_VALID
;
1014 state
->enter
= acpi_idle_enter_c1
;
1015 state
->enter_dead
= acpi_idle_play_dead
;
1016 drv
->safe_state_index
= count
;
1020 state
->flags
|= CPUIDLE_FLAG_TIME_VALID
;
1021 state
->enter
= acpi_idle_enter_simple
;
1022 state
->enter_dead
= acpi_idle_play_dead
;
1023 drv
->safe_state_index
= count
;
1027 state
->flags
|= CPUIDLE_FLAG_TIME_VALID
;
1028 state
->enter
= pr
->flags
.bm_check
?
1029 acpi_idle_enter_bm
:
1030 acpi_idle_enter_simple
;
1035 if (count
== CPUIDLE_STATE_MAX
)
1039 drv
->state_count
= count
;
1047 int acpi_processor_hotplug(struct acpi_processor
*pr
)
1050 struct cpuidle_device
*dev
;
1052 if (disabled_by_idle_boot_param())
1062 if (!pr
->flags
.power_setup_done
)
1065 dev
= per_cpu(acpi_cpuidle_device
, pr
->id
);
1066 cpuidle_pause_and_lock();
1067 cpuidle_disable_device(dev
);
1068 acpi_processor_get_power_info(pr
);
1069 if (pr
->flags
.power
) {
1070 acpi_processor_setup_cpuidle_cx(pr
, dev
);
1071 ret
= cpuidle_enable_device(dev
);
1073 cpuidle_resume_and_unlock();
1078 int acpi_processor_cst_has_changed(struct acpi_processor
*pr
)
1081 struct acpi_processor
*_pr
;
1082 struct cpuidle_device
*dev
;
1084 if (disabled_by_idle_boot_param())
1093 if (!pr
->flags
.power_setup_done
)
1097 * FIXME: Design the ACPI notification to make it once per
1098 * system instead of once per-cpu. This condition is a hack
1099 * to make the code that updates C-States be called once.
1102 if (pr
->id
== 0 && cpuidle_get_driver() == &acpi_idle_driver
) {
1104 /* Protect against cpu-hotplug */
1106 cpuidle_pause_and_lock();
1108 /* Disable all cpuidle devices */
1109 for_each_online_cpu(cpu
) {
1110 _pr
= per_cpu(processors
, cpu
);
1111 if (!_pr
|| !_pr
->flags
.power_setup_done
)
1113 dev
= per_cpu(acpi_cpuidle_device
, cpu
);
1114 cpuidle_disable_device(dev
);
1117 /* Populate Updated C-state information */
1118 acpi_processor_get_power_info(pr
);
1119 acpi_processor_setup_cpuidle_states(pr
);
1121 /* Enable all cpuidle devices */
1122 for_each_online_cpu(cpu
) {
1123 _pr
= per_cpu(processors
, cpu
);
1124 if (!_pr
|| !_pr
->flags
.power_setup_done
)
1126 acpi_processor_get_power_info(_pr
);
1127 if (_pr
->flags
.power
) {
1128 dev
= per_cpu(acpi_cpuidle_device
, cpu
);
1129 acpi_processor_setup_cpuidle_cx(_pr
, dev
);
1130 cpuidle_enable_device(dev
);
1133 cpuidle_resume_and_unlock();
1140 static int acpi_processor_registered
;
1142 int __cpuinit
acpi_processor_power_init(struct acpi_processor
*pr
)
1144 acpi_status status
= 0;
1146 struct cpuidle_device
*dev
;
1147 static int first_run
;
1149 if (disabled_by_idle_boot_param())
1153 dmi_check_system(processor_power_dmi_table
);
1154 max_cstate
= acpi_processor_cstate_check(max_cstate
);
1155 if (max_cstate
< ACPI_C_STATES_MAX
)
1157 "ACPI: processor limited to max C-state %d\n",
1165 if (acpi_gbl_FADT
.cst_control
&& !nocst
) {
1167 acpi_os_write_port(acpi_gbl_FADT
.smi_command
, acpi_gbl_FADT
.cst_control
, 8);
1168 if (ACPI_FAILURE(status
)) {
1169 ACPI_EXCEPTION((AE_INFO
, status
,
1170 "Notifying BIOS of _CST ability failed"));
1174 acpi_processor_get_power_info(pr
);
1175 pr
->flags
.power_setup_done
= 1;
1178 * Install the idle handler if processor power management is supported.
1179 * Note that we use previously set idle handler will be used on
1180 * platforms that only support C1.
1182 if (pr
->flags
.power
) {
1183 /* Register acpi_idle_driver if not already registered */
1184 if (!acpi_processor_registered
) {
1185 acpi_processor_setup_cpuidle_states(pr
);
1186 retval
= cpuidle_register_driver(&acpi_idle_driver
);
1189 printk(KERN_DEBUG
"ACPI: %s registered with cpuidle\n",
1190 acpi_idle_driver
.name
);
1193 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
1196 per_cpu(acpi_cpuidle_device
, pr
->id
) = dev
;
1198 acpi_processor_setup_cpuidle_cx(pr
, dev
);
1200 /* Register per-cpu cpuidle_device. Cpuidle driver
1201 * must already be registered before registering device
1203 retval
= cpuidle_register_device(dev
);
1205 if (acpi_processor_registered
== 0)
1206 cpuidle_unregister_driver(&acpi_idle_driver
);
1209 acpi_processor_registered
++;
1214 int acpi_processor_power_exit(struct acpi_processor
*pr
)
1216 struct cpuidle_device
*dev
= per_cpu(acpi_cpuidle_device
, pr
->id
);
1218 if (disabled_by_idle_boot_param())
1221 if (pr
->flags
.power
) {
1222 cpuidle_unregister_device(dev
);
1223 acpi_processor_registered
--;
1224 if (acpi_processor_registered
== 0)
1225 cpuidle_unregister_driver(&acpi_idle_driver
);
1228 pr
->flags
.power_setup_done
= 0;