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 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/proc_fs.h>
36 #include <linux/seq_file.h>
37 #include <linux/acpi.h>
38 #include <linux/dmi.h>
39 #include <linux/moduleparam.h>
40 #include <linux/sched.h> /* need_resched() */
43 #include <asm/uaccess.h>
45 #include <acpi/acpi_bus.h>
46 #include <acpi/processor.h>
48 #define ACPI_PROCESSOR_COMPONENT 0x01000000
49 #define ACPI_PROCESSOR_CLASS "processor"
50 #define ACPI_PROCESSOR_DRIVER_NAME "ACPI Processor Driver"
51 #define _COMPONENT ACPI_PROCESSOR_COMPONENT
52 ACPI_MODULE_NAME("acpi_processor")
53 #define ACPI_PROCESSOR_FILE_POWER "power"
54 #define US_TO_PM_TIMER_TICKS(t) ((t * (PM_TIMER_FREQUENCY/1000)) / 1000)
55 #define C2_OVERHEAD 4 /* 1us (3.579 ticks per us) */
56 #define C3_OVERHEAD 4 /* 1us (3.579 ticks per us) */
57 static void (*pm_idle_save
) (void);
58 module_param(max_cstate
, uint
, 0644);
60 static unsigned int nocst
= 0;
61 module_param(nocst
, uint
, 0000);
64 * bm_history -- bit-mask with a bit per jiffy of bus-master activity
65 * 1000 HZ: 0xFFFFFFFF: 32 jiffies = 32ms
66 * 800 HZ: 0xFFFFFFFF: 32 jiffies = 40ms
67 * 100 HZ: 0x0000000F: 4 jiffies = 40ms
68 * reduce history for more aggressive entry into C3
70 static unsigned int bm_history
=
71 (HZ
>= 800 ? 0xFFFFFFFF : ((1U << (HZ
/ 25)) - 1));
72 module_param(bm_history
, uint
, 0644);
73 /* --------------------------------------------------------------------------
75 -------------------------------------------------------------------------- */
78 * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
79 * For now disable this. Probably a bug somewhere else.
81 * To skip this limit, boot/load with a large max_cstate limit.
83 static int set_max_cstate(struct dmi_system_id
*id
)
85 if (max_cstate
> ACPI_PROCESSOR_MAX_POWER
)
88 printk(KERN_NOTICE PREFIX
"%s detected - limiting to C%ld max_cstate."
89 " Override with \"processor.max_cstate=%d\"\n", id
->ident
,
90 (long)id
->driver_data
, ACPI_PROCESSOR_MAX_POWER
+ 1);
92 max_cstate
= (long)id
->driver_data
;
97 static struct dmi_system_id __initdata processor_power_dmi_table
[] = {
98 { set_max_cstate
, "IBM ThinkPad R40e", {
99 DMI_MATCH(DMI_BIOS_VENDOR
,"IBM"),
100 DMI_MATCH(DMI_BIOS_VERSION
,"1SET60WW")}, (void *)1},
101 { set_max_cstate
, "IBM ThinkPad R40e", {
102 DMI_MATCH(DMI_BIOS_VENDOR
,"IBM"),
103 DMI_MATCH(DMI_BIOS_VERSION
,"1SET43WW") }, (void*)1},
104 { set_max_cstate
, "IBM ThinkPad R40e", {
105 DMI_MATCH(DMI_BIOS_VENDOR
,"IBM"),
106 DMI_MATCH(DMI_BIOS_VERSION
,"1SET45WW") }, (void*)1},
107 { set_max_cstate
, "IBM ThinkPad R40e", {
108 DMI_MATCH(DMI_BIOS_VENDOR
,"IBM"),
109 DMI_MATCH(DMI_BIOS_VERSION
,"1SET47WW") }, (void*)1},
110 { set_max_cstate
, "IBM ThinkPad R40e", {
111 DMI_MATCH(DMI_BIOS_VENDOR
,"IBM"),
112 DMI_MATCH(DMI_BIOS_VERSION
,"1SET50WW") }, (void*)1},
113 { set_max_cstate
, "IBM ThinkPad R40e", {
114 DMI_MATCH(DMI_BIOS_VENDOR
,"IBM"),
115 DMI_MATCH(DMI_BIOS_VERSION
,"1SET52WW") }, (void*)1},
116 { set_max_cstate
, "IBM ThinkPad R40e", {
117 DMI_MATCH(DMI_BIOS_VENDOR
,"IBM"),
118 DMI_MATCH(DMI_BIOS_VERSION
,"1SET55WW") }, (void*)1},
119 { set_max_cstate
, "IBM ThinkPad R40e", {
120 DMI_MATCH(DMI_BIOS_VENDOR
,"IBM"),
121 DMI_MATCH(DMI_BIOS_VERSION
,"1SET56WW") }, (void*)1},
122 { set_max_cstate
, "IBM ThinkPad R40e", {
123 DMI_MATCH(DMI_BIOS_VENDOR
,"IBM"),
124 DMI_MATCH(DMI_BIOS_VERSION
,"1SET59WW") }, (void*)1},
125 { set_max_cstate
, "IBM ThinkPad R40e", {
126 DMI_MATCH(DMI_BIOS_VENDOR
,"IBM"),
127 DMI_MATCH(DMI_BIOS_VERSION
,"1SET60WW") }, (void*)1},
128 { set_max_cstate
, "IBM ThinkPad R40e", {
129 DMI_MATCH(DMI_BIOS_VENDOR
,"IBM"),
130 DMI_MATCH(DMI_BIOS_VERSION
,"1SET61WW") }, (void*)1},
131 { set_max_cstate
, "IBM ThinkPad R40e", {
132 DMI_MATCH(DMI_BIOS_VENDOR
,"IBM"),
133 DMI_MATCH(DMI_BIOS_VERSION
,"1SET62WW") }, (void*)1},
134 { set_max_cstate
, "IBM ThinkPad R40e", {
135 DMI_MATCH(DMI_BIOS_VENDOR
,"IBM"),
136 DMI_MATCH(DMI_BIOS_VERSION
,"1SET64WW") }, (void*)1},
137 { set_max_cstate
, "IBM ThinkPad R40e", {
138 DMI_MATCH(DMI_BIOS_VENDOR
,"IBM"),
139 DMI_MATCH(DMI_BIOS_VERSION
,"1SET65WW") }, (void*)1},
140 { set_max_cstate
, "IBM ThinkPad R40e", {
141 DMI_MATCH(DMI_BIOS_VENDOR
,"IBM"),
142 DMI_MATCH(DMI_BIOS_VERSION
,"1SET68WW") }, (void*)1},
143 { set_max_cstate
, "Medion 41700", {
144 DMI_MATCH(DMI_BIOS_VENDOR
,"Phoenix Technologies LTD"),
145 DMI_MATCH(DMI_BIOS_VERSION
,"R01-A1J")}, (void *)1},
146 { set_max_cstate
, "Clevo 5600D", {
147 DMI_MATCH(DMI_BIOS_VENDOR
,"Phoenix Technologies LTD"),
148 DMI_MATCH(DMI_BIOS_VERSION
,"SHE845M0.86C.0013.D.0302131307")},
153 static inline u32
ticks_elapsed(u32 t1
, u32 t2
)
157 else if (!acpi_fadt
.tmr_val_ext
)
158 return (((0x00FFFFFF - t1
) + t2
) & 0x00FFFFFF);
160 return ((0xFFFFFFFF - t1
) + t2
);
164 acpi_processor_power_activate(struct acpi_processor
*pr
,
165 struct acpi_processor_cx
*new)
167 struct acpi_processor_cx
*old
;
172 old
= pr
->power
.state
;
175 old
->promotion
.count
= 0;
176 new->demotion
.count
= 0;
178 /* Cleanup from old state. */
182 /* Disable bus master reload */
183 if (new->type
!= ACPI_STATE_C3
&& pr
->flags
.bm_check
)
184 acpi_set_register(ACPI_BITREG_BUS_MASTER_RLD
, 0,
185 ACPI_MTX_DO_NOT_LOCK
);
190 /* Prepare to use new state. */
193 /* Enable bus master reload */
194 if (old
->type
!= ACPI_STATE_C3
&& pr
->flags
.bm_check
)
195 acpi_set_register(ACPI_BITREG_BUS_MASTER_RLD
, 1,
196 ACPI_MTX_DO_NOT_LOCK
);
200 pr
->power
.state
= new;
205 static void acpi_safe_halt(void)
207 clear_thread_flag(TIF_POLLING_NRFLAG
);
208 smp_mb__after_clear_bit();
211 set_thread_flag(TIF_POLLING_NRFLAG
);
214 static atomic_t c3_cpu_count
;
216 static void acpi_processor_idle(void)
218 struct acpi_processor
*pr
= NULL
;
219 struct acpi_processor_cx
*cx
= NULL
;
220 struct acpi_processor_cx
*next_state
= NULL
;
224 pr
= processors
[smp_processor_id()];
229 * Interrupts must be disabled during bus mastering calculations and
230 * for C2/C3 transitions.
235 * Check whether we truly need to go idle, or should
238 if (unlikely(need_resched())) {
243 cx
= pr
->power
.state
;
255 * Check for bus mastering activity (if required), record, and check
258 if (pr
->flags
.bm_check
) {
260 unsigned long diff
= jiffies
- pr
->power
.bm_check_timestamp
;
266 /* if we didn't get called, assume there was busmaster activity */
269 pr
->power
.bm_activity
|= 0x1;
270 pr
->power
.bm_activity
<<= 1;
273 acpi_get_register(ACPI_BITREG_BUS_MASTER_STATUS
,
274 &bm_status
, ACPI_MTX_DO_NOT_LOCK
);
276 pr
->power
.bm_activity
++;
277 acpi_set_register(ACPI_BITREG_BUS_MASTER_STATUS
,
278 1, ACPI_MTX_DO_NOT_LOCK
);
281 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
282 * the true state of bus mastering activity; forcing us to
283 * manually check the BMIDEA bit of each IDE channel.
285 else if (errata
.piix4
.bmisx
) {
286 if ((inb_p(errata
.piix4
.bmisx
+ 0x02) & 0x01)
287 || (inb_p(errata
.piix4
.bmisx
+ 0x0A) & 0x01))
288 pr
->power
.bm_activity
++;
291 pr
->power
.bm_check_timestamp
= jiffies
;
294 * Apply bus mastering demotion policy. Automatically demote
295 * to avoid a faulty transition. Note that the processor
296 * won't enter a low-power state during this call (to this
297 * funciton) but should upon the next.
299 * TBD: A better policy might be to fallback to the demotion
300 * state (use it for this quantum only) istead of
301 * demoting -- and rely on duration as our sole demotion
302 * qualification. This may, however, introduce DMA
303 * issues (e.g. floppy DMA transfer overrun/underrun).
305 if (pr
->power
.bm_activity
& cx
->demotion
.threshold
.bm
) {
307 next_state
= cx
->demotion
.state
;
312 #ifdef CONFIG_HOTPLUG_CPU
314 * Check for P_LVL2_UP flag before entering C2 and above on
315 * an SMP system. We do it here instead of doing it at _CST/P_LVL
316 * detection phase, to work cleanly with logical CPU hotplug.
318 if ((cx
->type
!= ACPI_STATE_C1
) && (num_online_cpus() > 1) &&
319 !pr
->flags
.has_cst
&& !acpi_fadt
.plvl2_up
)
320 cx
= &pr
->power
.states
[ACPI_STATE_C1
];
328 * Invoke the current Cx state to put the processor to sleep.
330 if (cx
->type
== ACPI_STATE_C2
|| cx
->type
== ACPI_STATE_C3
) {
331 clear_thread_flag(TIF_POLLING_NRFLAG
);
332 smp_mb__after_clear_bit();
333 if (need_resched()) {
334 set_thread_flag(TIF_POLLING_NRFLAG
);
345 * Use the appropriate idle routine, the one that would
346 * be used without acpi C-states.
354 * TBD: Can't get time duration while in C1, as resumes
355 * go to an ISR rather than here. Need to instrument
356 * base interrupt handler.
358 sleep_ticks
= 0xFFFFFFFF;
362 /* Get start time (ticks) */
363 t1
= inl(acpi_fadt
.xpm_tmr_blk
.address
);
366 /* Dummy op - must do something useless after P_LVL2 read */
367 t2
= inl(acpi_fadt
.xpm_tmr_blk
.address
);
368 /* Get end time (ticks) */
369 t2
= inl(acpi_fadt
.xpm_tmr_blk
.address
);
370 /* Re-enable interrupts */
372 set_thread_flag(TIF_POLLING_NRFLAG
);
373 /* Compute time (ticks) that we were actually asleep */
375 ticks_elapsed(t1
, t2
) - cx
->latency_ticks
- C2_OVERHEAD
;
380 if (pr
->flags
.bm_check
) {
381 if (atomic_inc_return(&c3_cpu_count
) ==
384 * All CPUs are trying to go to C3
385 * Disable bus master arbitration
387 acpi_set_register(ACPI_BITREG_ARB_DISABLE
, 1,
388 ACPI_MTX_DO_NOT_LOCK
);
391 /* SMP with no shared cache... Invalidate cache */
392 ACPI_FLUSH_CPU_CACHE();
395 /* Get start time (ticks) */
396 t1
= inl(acpi_fadt
.xpm_tmr_blk
.address
);
399 /* Dummy op - must do something useless after P_LVL3 read */
400 t2
= inl(acpi_fadt
.xpm_tmr_blk
.address
);
401 /* Get end time (ticks) */
402 t2
= inl(acpi_fadt
.xpm_tmr_blk
.address
);
403 if (pr
->flags
.bm_check
) {
404 /* Enable bus master arbitration */
405 atomic_dec(&c3_cpu_count
);
406 acpi_set_register(ACPI_BITREG_ARB_DISABLE
, 0,
407 ACPI_MTX_DO_NOT_LOCK
);
410 /* Re-enable interrupts */
412 set_thread_flag(TIF_POLLING_NRFLAG
);
413 /* Compute time (ticks) that we were actually asleep */
415 ticks_elapsed(t1
, t2
) - cx
->latency_ticks
- C3_OVERHEAD
;
423 next_state
= pr
->power
.state
;
425 #ifdef CONFIG_HOTPLUG_CPU
426 /* Don't do promotion/demotion */
427 if ((cx
->type
== ACPI_STATE_C1
) && (num_online_cpus() > 1) &&
428 !pr
->flags
.has_cst
&& !acpi_fadt
.plvl2_up
) {
437 * Track the number of longs (time asleep is greater than threshold)
438 * and promote when the count threshold is reached. Note that bus
439 * mastering activity may prevent promotions.
440 * Do not promote above max_cstate.
442 if (cx
->promotion
.state
&&
443 ((cx
->promotion
.state
- pr
->power
.states
) <= max_cstate
)) {
444 if (sleep_ticks
> cx
->promotion
.threshold
.ticks
) {
445 cx
->promotion
.count
++;
446 cx
->demotion
.count
= 0;
447 if (cx
->promotion
.count
>=
448 cx
->promotion
.threshold
.count
) {
449 if (pr
->flags
.bm_check
) {
451 (pr
->power
.bm_activity
& cx
->
452 promotion
.threshold
.bm
)) {
458 next_state
= cx
->promotion
.state
;
468 * Track the number of shorts (time asleep is less than time threshold)
469 * and demote when the usage threshold is reached.
471 if (cx
->demotion
.state
) {
472 if (sleep_ticks
< cx
->demotion
.threshold
.ticks
) {
473 cx
->demotion
.count
++;
474 cx
->promotion
.count
= 0;
475 if (cx
->demotion
.count
>= cx
->demotion
.threshold
.count
) {
476 next_state
= cx
->demotion
.state
;
484 * Demote if current state exceeds max_cstate
486 if ((pr
->power
.state
- pr
->power
.states
) > max_cstate
) {
487 if (cx
->demotion
.state
)
488 next_state
= cx
->demotion
.state
;
494 * If we're going to start using a new Cx state we must clean up
495 * from the previous and prepare to use the new.
497 if (next_state
!= pr
->power
.state
)
498 acpi_processor_power_activate(pr
, next_state
);
501 static int acpi_processor_set_power_policy(struct acpi_processor
*pr
)
504 unsigned int state_is_set
= 0;
505 struct acpi_processor_cx
*lower
= NULL
;
506 struct acpi_processor_cx
*higher
= NULL
;
507 struct acpi_processor_cx
*cx
;
509 ACPI_FUNCTION_TRACE("acpi_processor_set_power_policy");
512 return_VALUE(-EINVAL
);
515 * This function sets the default Cx state policy (OS idle handler).
516 * Our scheme is to promote quickly to C2 but more conservatively
517 * to C3. We're favoring C2 for its characteristics of low latency
518 * (quick response), good power savings, and ability to allow bus
519 * mastering activity. Note that the Cx state policy is completely
520 * customizable and can be altered dynamically.
524 for (i
= 1; i
< ACPI_PROCESSOR_MAX_POWER
; i
++) {
525 cx
= &pr
->power
.states
[i
];
530 pr
->power
.state
= cx
;
536 return_VALUE(-ENODEV
);
539 for (i
= 1; i
< ACPI_PROCESSOR_MAX_POWER
; i
++) {
540 cx
= &pr
->power
.states
[i
];
545 cx
->demotion
.state
= lower
;
546 cx
->demotion
.threshold
.ticks
= cx
->latency_ticks
;
547 cx
->demotion
.threshold
.count
= 1;
548 if (cx
->type
== ACPI_STATE_C3
)
549 cx
->demotion
.threshold
.bm
= bm_history
;
556 for (i
= (ACPI_PROCESSOR_MAX_POWER
- 1); i
> 0; i
--) {
557 cx
= &pr
->power
.states
[i
];
562 cx
->promotion
.state
= higher
;
563 cx
->promotion
.threshold
.ticks
= cx
->latency_ticks
;
564 if (cx
->type
>= ACPI_STATE_C2
)
565 cx
->promotion
.threshold
.count
= 4;
567 cx
->promotion
.threshold
.count
= 10;
568 if (higher
->type
== ACPI_STATE_C3
)
569 cx
->promotion
.threshold
.bm
= bm_history
;
578 static int acpi_processor_get_power_info_fadt(struct acpi_processor
*pr
)
580 ACPI_FUNCTION_TRACE("acpi_processor_get_power_info_fadt");
583 return_VALUE(-EINVAL
);
586 return_VALUE(-ENODEV
);
588 /* if info is obtained from pblk/fadt, type equals state */
589 pr
->power
.states
[ACPI_STATE_C2
].type
= ACPI_STATE_C2
;
590 pr
->power
.states
[ACPI_STATE_C3
].type
= ACPI_STATE_C3
;
592 #ifndef CONFIG_HOTPLUG_CPU
594 * Check for P_LVL2_UP flag before entering C2 and above on
597 if ((num_online_cpus() > 1) && !acpi_fadt
.plvl2_up
)
598 return_VALUE(-ENODEV
);
601 /* determine C2 and C3 address from pblk */
602 pr
->power
.states
[ACPI_STATE_C2
].address
= pr
->pblk
+ 4;
603 pr
->power
.states
[ACPI_STATE_C3
].address
= pr
->pblk
+ 5;
605 /* determine latencies from FADT */
606 pr
->power
.states
[ACPI_STATE_C2
].latency
= acpi_fadt
.plvl2_lat
;
607 pr
->power
.states
[ACPI_STATE_C3
].latency
= acpi_fadt
.plvl3_lat
;
609 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
610 "lvl2[0x%08x] lvl3[0x%08x]\n",
611 pr
->power
.states
[ACPI_STATE_C2
].address
,
612 pr
->power
.states
[ACPI_STATE_C3
].address
));
617 static int acpi_processor_get_power_info_default_c1(struct acpi_processor
*pr
)
619 ACPI_FUNCTION_TRACE("acpi_processor_get_power_info_default_c1");
621 /* Zero initialize all the C-states info. */
622 memset(pr
->power
.states
, 0, sizeof(pr
->power
.states
));
624 /* set the first C-State to C1 */
625 pr
->power
.states
[ACPI_STATE_C1
].type
= ACPI_STATE_C1
;
627 /* the C0 state only exists as a filler in our array,
628 * and all processors need to support C1 */
629 pr
->power
.states
[ACPI_STATE_C0
].valid
= 1;
630 pr
->power
.states
[ACPI_STATE_C1
].valid
= 1;
635 static int acpi_processor_get_power_info_cst(struct acpi_processor
*pr
)
637 acpi_status status
= 0;
641 struct acpi_buffer buffer
= { ACPI_ALLOCATE_BUFFER
, NULL
};
642 union acpi_object
*cst
;
644 ACPI_FUNCTION_TRACE("acpi_processor_get_power_info_cst");
647 return_VALUE(-ENODEV
);
651 /* Zero initialize C2 onwards and prepare for fresh CST lookup */
652 for (i
= 2; i
< ACPI_PROCESSOR_MAX_POWER
; i
++)
653 memset(&(pr
->power
.states
[i
]), 0,
654 sizeof(struct acpi_processor_cx
));
656 status
= acpi_evaluate_object(pr
->handle
, "_CST", NULL
, &buffer
);
657 if (ACPI_FAILURE(status
)) {
658 ACPI_DEBUG_PRINT((ACPI_DB_INFO
, "No _CST, giving up\n"));
659 return_VALUE(-ENODEV
);
662 cst
= (union acpi_object
*)buffer
.pointer
;
664 /* There must be at least 2 elements */
665 if (!cst
|| (cst
->type
!= ACPI_TYPE_PACKAGE
) || cst
->package
.count
< 2) {
666 ACPI_DEBUG_PRINT((ACPI_DB_ERROR
,
667 "not enough elements in _CST\n"));
672 count
= cst
->package
.elements
[0].integer
.value
;
674 /* Validate number of power states. */
675 if (count
< 1 || count
!= cst
->package
.count
- 1) {
676 ACPI_DEBUG_PRINT((ACPI_DB_ERROR
,
677 "count given by _CST is not valid\n"));
682 /* Tell driver that at least _CST is supported. */
683 pr
->flags
.has_cst
= 1;
685 for (i
= 1; i
<= count
; i
++) {
686 union acpi_object
*element
;
687 union acpi_object
*obj
;
688 struct acpi_power_register
*reg
;
689 struct acpi_processor_cx cx
;
691 memset(&cx
, 0, sizeof(cx
));
693 element
= (union acpi_object
*)&(cst
->package
.elements
[i
]);
694 if (element
->type
!= ACPI_TYPE_PACKAGE
)
697 if (element
->package
.count
!= 4)
700 obj
= (union acpi_object
*)&(element
->package
.elements
[0]);
702 if (obj
->type
!= ACPI_TYPE_BUFFER
)
705 reg
= (struct acpi_power_register
*)obj
->buffer
.pointer
;
707 if (reg
->space_id
!= ACPI_ADR_SPACE_SYSTEM_IO
&&
708 (reg
->space_id
!= ACPI_ADR_SPACE_FIXED_HARDWARE
))
711 cx
.address
= (reg
->space_id
== ACPI_ADR_SPACE_FIXED_HARDWARE
) ?
714 /* There should be an easy way to extract an integer... */
715 obj
= (union acpi_object
*)&(element
->package
.elements
[1]);
716 if (obj
->type
!= ACPI_TYPE_INTEGER
)
719 cx
.type
= obj
->integer
.value
;
721 if ((cx
.type
!= ACPI_STATE_C1
) &&
722 (reg
->space_id
!= ACPI_ADR_SPACE_SYSTEM_IO
))
725 if ((cx
.type
< ACPI_STATE_C2
) || (cx
.type
> ACPI_STATE_C3
))
728 obj
= (union acpi_object
*)&(element
->package
.elements
[2]);
729 if (obj
->type
!= ACPI_TYPE_INTEGER
)
732 cx
.latency
= obj
->integer
.value
;
734 obj
= (union acpi_object
*)&(element
->package
.elements
[3]);
735 if (obj
->type
!= ACPI_TYPE_INTEGER
)
738 cx
.power
= obj
->integer
.value
;
741 memcpy(&(pr
->power
.states
[current_count
]), &cx
, sizeof(cx
));
744 * We support total ACPI_PROCESSOR_MAX_POWER - 1
745 * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
747 if (current_count
>= (ACPI_PROCESSOR_MAX_POWER
- 1)) {
749 "Limiting number of power states to max (%d)\n",
750 ACPI_PROCESSOR_MAX_POWER
);
752 "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
757 ACPI_DEBUG_PRINT((ACPI_DB_INFO
, "Found %d power states\n",
760 /* Validate number of power states discovered */
761 if (current_count
< 2)
765 acpi_os_free(buffer
.pointer
);
767 return_VALUE(status
);
770 static void acpi_processor_power_verify_c2(struct acpi_processor_cx
*cx
)
772 ACPI_FUNCTION_TRACE("acpi_processor_get_power_verify_c2");
778 * C2 latency must be less than or equal to 100
781 else if (cx
->latency
> ACPI_PROCESSOR_MAX_C2_LATENCY
) {
782 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
783 "latency too large [%d]\n", cx
->latency
));
788 * Otherwise we've met all of our C2 requirements.
789 * Normalize the C2 latency to expidite policy
792 cx
->latency_ticks
= US_TO_PM_TIMER_TICKS(cx
->latency
);
797 static void acpi_processor_power_verify_c3(struct acpi_processor
*pr
,
798 struct acpi_processor_cx
*cx
)
800 static int bm_check_flag
;
802 ACPI_FUNCTION_TRACE("acpi_processor_get_power_verify_c3");
808 * C3 latency must be less than or equal to 1000
811 else if (cx
->latency
> ACPI_PROCESSOR_MAX_C3_LATENCY
) {
812 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
813 "latency too large [%d]\n", cx
->latency
));
818 * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
819 * DMA transfers are used by any ISA device to avoid livelock.
820 * Note that we could disable Type-F DMA (as recommended by
821 * the erratum), but this is known to disrupt certain ISA
822 * devices thus we take the conservative approach.
824 else if (errata
.piix4
.fdma
) {
825 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
826 "C3 not supported on PIIX4 with Type-F DMA\n"));
830 /* All the logic here assumes flags.bm_check is same across all CPUs */
831 if (!bm_check_flag
) {
832 /* Determine whether bm_check is needed based on CPU */
833 acpi_processor_power_init_bm_check(&(pr
->flags
), pr
->id
);
834 bm_check_flag
= pr
->flags
.bm_check
;
836 pr
->flags
.bm_check
= bm_check_flag
;
839 if (pr
->flags
.bm_check
) {
840 /* bus mastering control is necessary */
841 if (!pr
->flags
.bm_control
) {
842 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
843 "C3 support requires bus mastering control\n"));
848 * WBINVD should be set in fadt, for C3 state to be
849 * supported on when bm_check is not required.
851 if (acpi_fadt
.wb_invd
!= 1) {
852 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
853 "Cache invalidation should work properly"
854 " for C3 to be enabled on SMP systems\n"));
857 acpi_set_register(ACPI_BITREG_BUS_MASTER_RLD
,
858 0, ACPI_MTX_DO_NOT_LOCK
);
862 * Otherwise we've met all of our C3 requirements.
863 * Normalize the C3 latency to expidite policy. Enable
864 * checking of bus mastering status (bm_check) so we can
865 * use this in our C3 policy
868 cx
->latency_ticks
= US_TO_PM_TIMER_TICKS(cx
->latency
);
873 static int acpi_processor_power_verify(struct acpi_processor
*pr
)
876 unsigned int working
= 0;
878 #ifdef ARCH_APICTIMER_STOPS_ON_C3
879 struct cpuinfo_x86
*c
= cpu_data
+ pr
->id
;
880 cpumask_t mask
= cpumask_of_cpu(pr
->id
);
882 if (c
->x86_vendor
== X86_VENDOR_INTEL
) {
883 on_each_cpu(switch_ipi_to_APIC_timer
, &mask
, 1, 1);
887 for (i
= 1; i
< ACPI_PROCESSOR_MAX_POWER
; i
++) {
888 struct acpi_processor_cx
*cx
= &pr
->power
.states
[i
];
896 acpi_processor_power_verify_c2(cx
);
900 acpi_processor_power_verify_c3(pr
, cx
);
901 #ifdef ARCH_APICTIMER_STOPS_ON_C3
902 if (cx
->valid
&& c
->x86_vendor
== X86_VENDOR_INTEL
) {
903 on_each_cpu(switch_APIC_timer_to_ipi
,
917 static int acpi_processor_get_power_info(struct acpi_processor
*pr
)
922 ACPI_FUNCTION_TRACE("acpi_processor_get_power_info");
924 /* NOTE: the idle thread may not be running while calling
927 /* Adding C1 state */
928 acpi_processor_get_power_info_default_c1(pr
);
929 result
= acpi_processor_get_power_info_cst(pr
);
930 if (result
== -ENODEV
)
931 acpi_processor_get_power_info_fadt(pr
);
933 pr
->power
.count
= acpi_processor_power_verify(pr
);
938 * Now that we know which states are supported, set the default
939 * policy. Note that this policy can be changed dynamically
940 * (e.g. encourage deeper sleeps to conserve battery life when
943 result
= acpi_processor_set_power_policy(pr
);
945 return_VALUE(result
);
948 * if one state of type C2 or C3 is available, mark this
949 * CPU as being "idle manageable"
951 for (i
= 1; i
< ACPI_PROCESSOR_MAX_POWER
; i
++) {
952 if (pr
->power
.states
[i
].valid
) {
954 if (pr
->power
.states
[i
].type
>= ACPI_STATE_C2
)
962 int acpi_processor_cst_has_changed(struct acpi_processor
*pr
)
966 ACPI_FUNCTION_TRACE("acpi_processor_cst_has_changed");
969 return_VALUE(-EINVAL
);
972 return_VALUE(-ENODEV
);
975 if (!pr
->flags
.power_setup_done
)
976 return_VALUE(-ENODEV
);
978 /* Fall back to the default idle loop */
979 pm_idle
= pm_idle_save
;
980 synchronize_sched(); /* Relies on interrupts forcing exit from idle. */
983 result
= acpi_processor_get_power_info(pr
);
984 if ((pr
->flags
.power
== 1) && (pr
->flags
.power_setup_done
))
985 pm_idle
= acpi_processor_idle
;
987 return_VALUE(result
);
992 static int acpi_processor_power_seq_show(struct seq_file
*seq
, void *offset
)
994 struct acpi_processor
*pr
= (struct acpi_processor
*)seq
->private;
997 ACPI_FUNCTION_TRACE("acpi_processor_power_seq_show");
1002 seq_printf(seq
, "active state: C%zd\n"
1004 "bus master activity: %08x\n",
1005 pr
->power
.state
? pr
->power
.state
- pr
->power
.states
: 0,
1006 max_cstate
, (unsigned)pr
->power
.bm_activity
);
1008 seq_puts(seq
, "states:\n");
1010 for (i
= 1; i
<= pr
->power
.count
; i
++) {
1011 seq_printf(seq
, " %cC%d: ",
1012 (&pr
->power
.states
[i
] ==
1013 pr
->power
.state
? '*' : ' '), i
);
1015 if (!pr
->power
.states
[i
].valid
) {
1016 seq_puts(seq
, "<not supported>\n");
1020 switch (pr
->power
.states
[i
].type
) {
1022 seq_printf(seq
, "type[C1] ");
1025 seq_printf(seq
, "type[C2] ");
1028 seq_printf(seq
, "type[C3] ");
1031 seq_printf(seq
, "type[--] ");
1035 if (pr
->power
.states
[i
].promotion
.state
)
1036 seq_printf(seq
, "promotion[C%zd] ",
1037 (pr
->power
.states
[i
].promotion
.state
-
1040 seq_puts(seq
, "promotion[--] ");
1042 if (pr
->power
.states
[i
].demotion
.state
)
1043 seq_printf(seq
, "demotion[C%zd] ",
1044 (pr
->power
.states
[i
].demotion
.state
-
1047 seq_puts(seq
, "demotion[--] ");
1049 seq_printf(seq
, "latency[%03d] usage[%08d]\n",
1050 pr
->power
.states
[i
].latency
,
1051 pr
->power
.states
[i
].usage
);
1058 static int acpi_processor_power_open_fs(struct inode
*inode
, struct file
*file
)
1060 return single_open(file
, acpi_processor_power_seq_show
,
1064 static struct file_operations acpi_processor_power_fops
= {
1065 .open
= acpi_processor_power_open_fs
,
1067 .llseek
= seq_lseek
,
1068 .release
= single_release
,
1071 int acpi_processor_power_init(struct acpi_processor
*pr
,
1072 struct acpi_device
*device
)
1074 acpi_status status
= 0;
1075 static int first_run
= 0;
1076 struct proc_dir_entry
*entry
= NULL
;
1079 ACPI_FUNCTION_TRACE("acpi_processor_power_init");
1082 dmi_check_system(processor_power_dmi_table
);
1083 if (max_cstate
< ACPI_C_STATES_MAX
)
1085 "ACPI: processor limited to max C-state %d\n",
1091 return_VALUE(-EINVAL
);
1093 if (acpi_fadt
.cst_cnt
&& !nocst
) {
1095 acpi_os_write_port(acpi_fadt
.smi_cmd
, acpi_fadt
.cst_cnt
, 8);
1096 if (ACPI_FAILURE(status
)) {
1097 ACPI_DEBUG_PRINT((ACPI_DB_ERROR
,
1098 "Notifying BIOS of _CST ability failed\n"));
1102 acpi_processor_get_power_info(pr
);
1105 * Install the idle handler if processor power management is supported.
1106 * Note that we use previously set idle handler will be used on
1107 * platforms that only support C1.
1109 if ((pr
->flags
.power
) && (!boot_option_idle_override
)) {
1110 printk(KERN_INFO PREFIX
"CPU%d (power states:", pr
->id
);
1111 for (i
= 1; i
<= pr
->power
.count
; i
++)
1112 if (pr
->power
.states
[i
].valid
)
1113 printk(" C%d[C%d]", i
,
1114 pr
->power
.states
[i
].type
);
1118 pm_idle_save
= pm_idle
;
1119 pm_idle
= acpi_processor_idle
;
1124 entry
= create_proc_entry(ACPI_PROCESSOR_FILE_POWER
,
1125 S_IRUGO
, acpi_device_dir(device
));
1127 ACPI_DEBUG_PRINT((ACPI_DB_ERROR
,
1128 "Unable to create '%s' fs entry\n",
1129 ACPI_PROCESSOR_FILE_POWER
));
1131 entry
->proc_fops
= &acpi_processor_power_fops
;
1132 entry
->data
= acpi_driver_data(device
);
1133 entry
->owner
= THIS_MODULE
;
1136 pr
->flags
.power_setup_done
= 1;
1141 int acpi_processor_power_exit(struct acpi_processor
*pr
,
1142 struct acpi_device
*device
)
1144 ACPI_FUNCTION_TRACE("acpi_processor_power_exit");
1146 pr
->flags
.power_setup_done
= 0;
1148 if (acpi_device_dir(device
))
1149 remove_proc_entry(ACPI_PROCESSOR_FILE_POWER
,
1150 acpi_device_dir(device
));
1152 /* Unregister the idle handler when processor #0 is removed. */
1154 pm_idle
= pm_idle_save
;
1157 * We are about to unload the current idle thread pm callback
1158 * (pm_idle), Wait for all processors to update cached/local
1159 * copies of pm_idle before proceeding.