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/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() */
41 #include <linux/pm_qos_params.h>
42 #include <linux/clockchips.h>
43 #include <linux/cpuidle.h>
44 #include <linux/irqflags.h>
47 * Include the apic definitions for x86 to have the APIC timer related defines
48 * available also for UP (on SMP it gets magically included via linux/smp.h).
49 * asm/acpi.h is not an option, as it would require more include magic. Also
50 * creating an empty asm-ia64/apic.h would just trade pest vs. cholera.
57 #include <asm/uaccess.h>
59 #include <acpi/acpi_bus.h>
60 #include <acpi/processor.h>
61 #include <asm/processor.h>
63 #define ACPI_PROCESSOR_CLASS "processor"
64 #define _COMPONENT ACPI_PROCESSOR_COMPONENT
65 ACPI_MODULE_NAME("processor_idle");
66 #define ACPI_PROCESSOR_FILE_POWER "power"
67 #define US_TO_PM_TIMER_TICKS(t) ((t * (PM_TIMER_FREQUENCY/1000)) / 1000)
68 #define PM_TIMER_TICK_NS (1000000000ULL/PM_TIMER_FREQUENCY)
69 #define C2_OVERHEAD 1 /* 1us */
70 #define C3_OVERHEAD 1 /* 1us */
71 #define PM_TIMER_TICKS_TO_US(p) (((p) * 1000)/(PM_TIMER_FREQUENCY/1000))
73 static unsigned int max_cstate __read_mostly
= ACPI_PROCESSOR_MAX_POWER
;
74 module_param(max_cstate
, uint
, 0000);
75 static unsigned int nocst __read_mostly
;
76 module_param(nocst
, uint
, 0000);
78 static unsigned int latency_factor __read_mostly
= 2;
79 module_param(latency_factor
, uint
, 0644);
82 * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
83 * For now disable this. Probably a bug somewhere else.
85 * To skip this limit, boot/load with a large max_cstate limit.
87 static int set_max_cstate(const struct dmi_system_id
*id
)
89 if (max_cstate
> ACPI_PROCESSOR_MAX_POWER
)
92 printk(KERN_NOTICE PREFIX
"%s detected - limiting to C%ld max_cstate."
93 " Override with \"processor.max_cstate=%d\"\n", id
->ident
,
94 (long)id
->driver_data
, ACPI_PROCESSOR_MAX_POWER
+ 1);
96 max_cstate
= (long)id
->driver_data
;
101 /* Actually this shouldn't be __cpuinitdata, would be better to fix the
102 callers to only run once -AK */
103 static struct dmi_system_id __cpuinitdata processor_power_dmi_table
[] = {
104 { set_max_cstate
, "Clevo 5600D", {
105 DMI_MATCH(DMI_BIOS_VENDOR
,"Phoenix Technologies LTD"),
106 DMI_MATCH(DMI_BIOS_VERSION
,"SHE845M0.86C.0013.D.0302131307")},
111 static inline u32
ticks_elapsed(u32 t1
, u32 t2
)
115 else if (!(acpi_gbl_FADT
.flags
& ACPI_FADT_32BIT_TIMER
))
116 return (((0x00FFFFFF - t1
) + t2
) & 0x00FFFFFF);
118 return ((0xFFFFFFFF - t1
) + t2
);
121 static inline u32
ticks_elapsed_in_us(u32 t1
, u32 t2
)
124 return PM_TIMER_TICKS_TO_US(t2
- t1
);
125 else if (!(acpi_gbl_FADT
.flags
& ACPI_FADT_32BIT_TIMER
))
126 return PM_TIMER_TICKS_TO_US(((0x00FFFFFF - t1
) + t2
) & 0x00FFFFFF);
128 return PM_TIMER_TICKS_TO_US((0xFFFFFFFF - t1
) + t2
);
132 * Callers should disable interrupts before the call and enable
133 * interrupts after return.
135 static void acpi_safe_halt(void)
137 current_thread_info()->status
&= ~TS_POLLING
;
139 * TS_POLLING-cleared state must be visible before we
143 if (!need_resched()) {
147 current_thread_info()->status
|= TS_POLLING
;
150 #ifdef ARCH_APICTIMER_STOPS_ON_C3
153 * Some BIOS implementations switch to C3 in the published C2 state.
154 * This seems to be a common problem on AMD boxen, but other vendors
155 * are affected too. We pick the most conservative approach: we assume
156 * that the local APIC stops in both C2 and C3.
158 static void acpi_timer_check_state(int state
, struct acpi_processor
*pr
,
159 struct acpi_processor_cx
*cx
)
161 struct acpi_processor_power
*pwr
= &pr
->power
;
162 u8 type
= local_apic_timer_c2_ok
? ACPI_STATE_C3
: ACPI_STATE_C2
;
165 * Check, if one of the previous states already marked the lapic
168 if (pwr
->timer_broadcast_on_state
< state
)
171 if (cx
->type
>= type
)
172 pr
->power
.timer_broadcast_on_state
= state
;
175 static void acpi_propagate_timer_broadcast(struct acpi_processor
*pr
)
177 unsigned long reason
;
179 reason
= pr
->power
.timer_broadcast_on_state
< INT_MAX
?
180 CLOCK_EVT_NOTIFY_BROADCAST_ON
: CLOCK_EVT_NOTIFY_BROADCAST_OFF
;
182 clockevents_notify(reason
, &pr
->id
);
185 /* Power(C) State timer broadcast control */
186 static void acpi_state_timer_broadcast(struct acpi_processor
*pr
,
187 struct acpi_processor_cx
*cx
,
190 int state
= cx
- pr
->power
.states
;
192 if (state
>= pr
->power
.timer_broadcast_on_state
) {
193 unsigned long reason
;
195 reason
= broadcast
? CLOCK_EVT_NOTIFY_BROADCAST_ENTER
:
196 CLOCK_EVT_NOTIFY_BROADCAST_EXIT
;
197 clockevents_notify(reason
, &pr
->id
);
203 static void acpi_timer_check_state(int state
, struct acpi_processor
*pr
,
204 struct acpi_processor_cx
*cstate
) { }
205 static void acpi_propagate_timer_broadcast(struct acpi_processor
*pr
) { }
206 static void acpi_state_timer_broadcast(struct acpi_processor
*pr
,
207 struct acpi_processor_cx
*cx
,
215 * Suspend / resume control
217 static int acpi_idle_suspend
;
219 int acpi_processor_suspend(struct acpi_device
* device
, pm_message_t state
)
221 acpi_idle_suspend
= 1;
225 int acpi_processor_resume(struct acpi_device
* device
)
227 acpi_idle_suspend
= 0;
231 #if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86)
232 static int tsc_halts_in_c(int state
)
234 switch (boot_cpu_data
.x86_vendor
) {
236 case X86_VENDOR_INTEL
:
238 * AMD Fam10h TSC will tick in all
239 * C/P/S0/S1 states when this bit is set.
241 if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC
))
246 return state
> ACPI_STATE_C1
;
251 static int acpi_processor_get_power_info_fadt(struct acpi_processor
*pr
)
260 /* if info is obtained from pblk/fadt, type equals state */
261 pr
->power
.states
[ACPI_STATE_C2
].type
= ACPI_STATE_C2
;
262 pr
->power
.states
[ACPI_STATE_C3
].type
= ACPI_STATE_C3
;
264 #ifndef CONFIG_HOTPLUG_CPU
266 * Check for P_LVL2_UP flag before entering C2 and above on
269 if ((num_online_cpus() > 1) &&
270 !(acpi_gbl_FADT
.flags
& ACPI_FADT_C2_MP_SUPPORTED
))
274 /* determine C2 and C3 address from pblk */
275 pr
->power
.states
[ACPI_STATE_C2
].address
= pr
->pblk
+ 4;
276 pr
->power
.states
[ACPI_STATE_C3
].address
= pr
->pblk
+ 5;
278 /* determine latencies from FADT */
279 pr
->power
.states
[ACPI_STATE_C2
].latency
= acpi_gbl_FADT
.C2latency
;
280 pr
->power
.states
[ACPI_STATE_C3
].latency
= acpi_gbl_FADT
.C3latency
;
282 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
283 "lvl2[0x%08x] lvl3[0x%08x]\n",
284 pr
->power
.states
[ACPI_STATE_C2
].address
,
285 pr
->power
.states
[ACPI_STATE_C3
].address
));
290 static int acpi_processor_get_power_info_default(struct acpi_processor
*pr
)
292 if (!pr
->power
.states
[ACPI_STATE_C1
].valid
) {
293 /* set the first C-State to C1 */
294 /* all processors need to support C1 */
295 pr
->power
.states
[ACPI_STATE_C1
].type
= ACPI_STATE_C1
;
296 pr
->power
.states
[ACPI_STATE_C1
].valid
= 1;
297 pr
->power
.states
[ACPI_STATE_C1
].entry_method
= ACPI_CSTATE_HALT
;
299 /* the C0 state only exists as a filler in our array */
300 pr
->power
.states
[ACPI_STATE_C0
].valid
= 1;
304 static int acpi_processor_get_power_info_cst(struct acpi_processor
*pr
)
306 acpi_status status
= 0;
310 struct acpi_buffer buffer
= { ACPI_ALLOCATE_BUFFER
, NULL
};
311 union acpi_object
*cst
;
319 status
= acpi_evaluate_object(pr
->handle
, "_CST", NULL
, &buffer
);
320 if (ACPI_FAILURE(status
)) {
321 ACPI_DEBUG_PRINT((ACPI_DB_INFO
, "No _CST, giving up\n"));
325 cst
= buffer
.pointer
;
327 /* There must be at least 2 elements */
328 if (!cst
|| (cst
->type
!= ACPI_TYPE_PACKAGE
) || cst
->package
.count
< 2) {
329 printk(KERN_ERR PREFIX
"not enough elements in _CST\n");
334 count
= cst
->package
.elements
[0].integer
.value
;
336 /* Validate number of power states. */
337 if (count
< 1 || count
!= cst
->package
.count
- 1) {
338 printk(KERN_ERR PREFIX
"count given by _CST is not valid\n");
343 /* Tell driver that at least _CST is supported. */
344 pr
->flags
.has_cst
= 1;
346 for (i
= 1; i
<= count
; i
++) {
347 union acpi_object
*element
;
348 union acpi_object
*obj
;
349 struct acpi_power_register
*reg
;
350 struct acpi_processor_cx cx
;
352 memset(&cx
, 0, sizeof(cx
));
354 element
= &(cst
->package
.elements
[i
]);
355 if (element
->type
!= ACPI_TYPE_PACKAGE
)
358 if (element
->package
.count
!= 4)
361 obj
= &(element
->package
.elements
[0]);
363 if (obj
->type
!= ACPI_TYPE_BUFFER
)
366 reg
= (struct acpi_power_register
*)obj
->buffer
.pointer
;
368 if (reg
->space_id
!= ACPI_ADR_SPACE_SYSTEM_IO
&&
369 (reg
->space_id
!= ACPI_ADR_SPACE_FIXED_HARDWARE
))
372 /* There should be an easy way to extract an integer... */
373 obj
= &(element
->package
.elements
[1]);
374 if (obj
->type
!= ACPI_TYPE_INTEGER
)
377 cx
.type
= obj
->integer
.value
;
379 * Some buggy BIOSes won't list C1 in _CST -
380 * Let acpi_processor_get_power_info_default() handle them later
382 if (i
== 1 && cx
.type
!= ACPI_STATE_C1
)
385 cx
.address
= reg
->address
;
386 cx
.index
= current_count
+ 1;
388 cx
.entry_method
= ACPI_CSTATE_SYSTEMIO
;
389 if (reg
->space_id
== ACPI_ADR_SPACE_FIXED_HARDWARE
) {
390 if (acpi_processor_ffh_cstate_probe
391 (pr
->id
, &cx
, reg
) == 0) {
392 cx
.entry_method
= ACPI_CSTATE_FFH
;
393 } else if (cx
.type
== ACPI_STATE_C1
) {
395 * C1 is a special case where FIXED_HARDWARE
396 * can be handled in non-MWAIT way as well.
397 * In that case, save this _CST entry info.
398 * Otherwise, ignore this info and continue.
400 cx
.entry_method
= ACPI_CSTATE_HALT
;
401 snprintf(cx
.desc
, ACPI_CX_DESC_LEN
, "ACPI HLT");
405 if (cx
.type
== ACPI_STATE_C1
&&
406 (idle_halt
|| idle_nomwait
)) {
408 * In most cases the C1 space_id obtained from
409 * _CST object is FIXED_HARDWARE access mode.
410 * But when the option of idle=halt is added,
411 * the entry_method type should be changed from
412 * CSTATE_FFH to CSTATE_HALT.
413 * When the option of idle=nomwait is added,
414 * the C1 entry_method type should be
417 cx
.entry_method
= ACPI_CSTATE_HALT
;
418 snprintf(cx
.desc
, ACPI_CX_DESC_LEN
, "ACPI HLT");
421 snprintf(cx
.desc
, ACPI_CX_DESC_LEN
, "ACPI IOPORT 0x%x",
425 if (cx
.type
== ACPI_STATE_C1
) {
429 obj
= &(element
->package
.elements
[2]);
430 if (obj
->type
!= ACPI_TYPE_INTEGER
)
433 cx
.latency
= obj
->integer
.value
;
435 obj
= &(element
->package
.elements
[3]);
436 if (obj
->type
!= ACPI_TYPE_INTEGER
)
439 cx
.power
= obj
->integer
.value
;
442 memcpy(&(pr
->power
.states
[current_count
]), &cx
, sizeof(cx
));
445 * We support total ACPI_PROCESSOR_MAX_POWER - 1
446 * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
448 if (current_count
>= (ACPI_PROCESSOR_MAX_POWER
- 1)) {
450 "Limiting number of power states to max (%d)\n",
451 ACPI_PROCESSOR_MAX_POWER
);
453 "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
458 ACPI_DEBUG_PRINT((ACPI_DB_INFO
, "Found %d power states\n",
461 /* Validate number of power states discovered */
462 if (current_count
< 2)
466 kfree(buffer
.pointer
);
471 static void acpi_processor_power_verify_c2(struct acpi_processor_cx
*cx
)
478 * C2 latency must be less than or equal to 100
481 else if (cx
->latency
> ACPI_PROCESSOR_MAX_C2_LATENCY
) {
482 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
483 "latency too large [%d]\n", cx
->latency
));
488 * Otherwise we've met all of our C2 requirements.
489 * Normalize the C2 latency to expidite policy
493 cx
->latency_ticks
= cx
->latency
;
498 static void acpi_processor_power_verify_c3(struct acpi_processor
*pr
,
499 struct acpi_processor_cx
*cx
)
501 static int bm_check_flag
;
508 * C3 latency must be less than or equal to 1000
511 else if (cx
->latency
> ACPI_PROCESSOR_MAX_C3_LATENCY
) {
512 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
513 "latency too large [%d]\n", cx
->latency
));
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
) {
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
;
536 pr
->flags
.bm_check
= bm_check_flag
;
539 if (pr
->flags
.bm_check
) {
540 if (!pr
->flags
.bm_control
) {
541 if (pr
->flags
.has_cst
!= 1) {
542 /* bus mastering control is necessary */
543 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
544 "C3 support requires BM control\n"));
547 /* Here we enter C3 without bus mastering */
548 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
549 "C3 support without BM control\n"));
554 * WBINVD should be set in fadt, for C3 state to be
555 * supported on when bm_check is not required.
557 if (!(acpi_gbl_FADT
.flags
& ACPI_FADT_WBINVD
)) {
558 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
559 "Cache invalidation should work properly"
560 " for C3 to be enabled on SMP systems\n"));
566 * Otherwise we've met all of our C3 requirements.
567 * Normalize the C3 latency to expidite policy. Enable
568 * checking of bus mastering status (bm_check) so we can
569 * use this in our C3 policy
573 cx
->latency_ticks
= cx
->latency
;
575 * On older chipsets, BM_RLD needs to be set
576 * in order for Bus Master activity to wake the
577 * system from C3. Newer chipsets handle DMA
578 * during C3 automatically and BM_RLD is a NOP.
579 * In either case, the proper way to
580 * handle BM_RLD is to set it and leave it set.
582 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD
, 1);
587 static int acpi_processor_power_verify(struct acpi_processor
*pr
)
590 unsigned int working
= 0;
592 pr
->power
.timer_broadcast_on_state
= INT_MAX
;
594 for (i
= 1; i
< ACPI_PROCESSOR_MAX_POWER
; i
++) {
595 struct acpi_processor_cx
*cx
= &pr
->power
.states
[i
];
603 acpi_processor_power_verify_c2(cx
);
605 acpi_timer_check_state(i
, pr
, cx
);
609 acpi_processor_power_verify_c3(pr
, cx
);
611 acpi_timer_check_state(i
, pr
, cx
);
619 acpi_propagate_timer_broadcast(pr
);
624 static int acpi_processor_get_power_info(struct acpi_processor
*pr
)
630 /* NOTE: the idle thread may not be running while calling
633 /* Zero initialize all the C-states info. */
634 memset(pr
->power
.states
, 0, sizeof(pr
->power
.states
));
636 result
= acpi_processor_get_power_info_cst(pr
);
637 if (result
== -ENODEV
)
638 result
= acpi_processor_get_power_info_fadt(pr
);
643 acpi_processor_get_power_info_default(pr
);
645 pr
->power
.count
= acpi_processor_power_verify(pr
);
648 * if one state of type C2 or C3 is available, mark this
649 * CPU as being "idle manageable"
651 for (i
= 1; i
< ACPI_PROCESSOR_MAX_POWER
; i
++) {
652 if (pr
->power
.states
[i
].valid
) {
654 if (pr
->power
.states
[i
].type
>= ACPI_STATE_C2
)
662 static int acpi_processor_power_seq_show(struct seq_file
*seq
, void *offset
)
664 struct acpi_processor
*pr
= seq
->private;
671 seq_printf(seq
, "active state: C%zd\n"
673 "bus master activity: %08x\n"
674 "maximum allowed latency: %d usec\n",
675 pr
->power
.state
? pr
->power
.state
- pr
->power
.states
: 0,
676 max_cstate
, (unsigned)pr
->power
.bm_activity
,
677 pm_qos_requirement(PM_QOS_CPU_DMA_LATENCY
));
679 seq_puts(seq
, "states:\n");
681 for (i
= 1; i
<= pr
->power
.count
; i
++) {
682 seq_printf(seq
, " %cC%d: ",
683 (&pr
->power
.states
[i
] ==
684 pr
->power
.state
? '*' : ' '), i
);
686 if (!pr
->power
.states
[i
].valid
) {
687 seq_puts(seq
, "<not supported>\n");
691 switch (pr
->power
.states
[i
].type
) {
693 seq_printf(seq
, "type[C1] ");
696 seq_printf(seq
, "type[C2] ");
699 seq_printf(seq
, "type[C3] ");
702 seq_printf(seq
, "type[--] ");
706 if (pr
->power
.states
[i
].promotion
.state
)
707 seq_printf(seq
, "promotion[C%zd] ",
708 (pr
->power
.states
[i
].promotion
.state
-
711 seq_puts(seq
, "promotion[--] ");
713 if (pr
->power
.states
[i
].demotion
.state
)
714 seq_printf(seq
, "demotion[C%zd] ",
715 (pr
->power
.states
[i
].demotion
.state
-
718 seq_puts(seq
, "demotion[--] ");
720 seq_printf(seq
, "latency[%03d] usage[%08d] duration[%020llu]\n",
721 pr
->power
.states
[i
].latency
,
722 pr
->power
.states
[i
].usage
,
723 (unsigned long long)pr
->power
.states
[i
].time
);
730 static int acpi_processor_power_open_fs(struct inode
*inode
, struct file
*file
)
732 return single_open(file
, acpi_processor_power_seq_show
,
736 static const struct file_operations acpi_processor_power_fops
= {
737 .owner
= THIS_MODULE
,
738 .open
= acpi_processor_power_open_fs
,
741 .release
= single_release
,
746 * acpi_idle_bm_check - checks if bus master activity was detected
748 static int acpi_idle_bm_check(void)
752 acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS
, &bm_status
);
754 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS
, 1);
756 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
757 * the true state of bus mastering activity; forcing us to
758 * manually check the BMIDEA bit of each IDE channel.
760 else if (errata
.piix4
.bmisx
) {
761 if ((inb_p(errata
.piix4
.bmisx
+ 0x02) & 0x01)
762 || (inb_p(errata
.piix4
.bmisx
+ 0x0A) & 0x01))
769 * acpi_idle_do_entry - a helper function that does C2 and C3 type entry
772 * Caller disables interrupt before call and enables interrupt after return.
774 static inline void acpi_idle_do_entry(struct acpi_processor_cx
*cx
)
776 /* Don't trace irqs off for idle */
777 stop_critical_timings();
778 if (cx
->entry_method
== ACPI_CSTATE_FFH
) {
779 /* Call into architectural FFH based C-state */
780 acpi_processor_ffh_cstate_enter(cx
);
781 } else if (cx
->entry_method
== ACPI_CSTATE_HALT
) {
785 /* IO port based C-state */
787 /* Dummy wait op - must do something useless after P_LVL2 read
788 because chipsets cannot guarantee that STPCLK# signal
789 gets asserted in time to freeze execution properly. */
790 unused
= inl(acpi_gbl_FADT
.xpm_timer_block
.address
);
792 start_critical_timings();
796 * acpi_idle_enter_c1 - enters an ACPI C1 state-type
797 * @dev: the target CPU
798 * @state: the state data
800 * This is equivalent to the HALT instruction.
802 static int acpi_idle_enter_c1(struct cpuidle_device
*dev
,
803 struct cpuidle_state
*state
)
806 struct acpi_processor
*pr
;
807 struct acpi_processor_cx
*cx
= cpuidle_get_statedata(state
);
809 pr
= __get_cpu_var(processors
);
816 /* Do not access any ACPI IO ports in suspend path */
817 if (acpi_idle_suspend
) {
823 t1
= inl(acpi_gbl_FADT
.xpm_timer_block
.address
);
824 acpi_idle_do_entry(cx
);
825 t2
= inl(acpi_gbl_FADT
.xpm_timer_block
.address
);
830 return ticks_elapsed_in_us(t1
, t2
);
834 * acpi_idle_enter_simple - enters an ACPI state without BM handling
835 * @dev: the target CPU
836 * @state: the state data
838 static int acpi_idle_enter_simple(struct cpuidle_device
*dev
,
839 struct cpuidle_state
*state
)
841 struct acpi_processor
*pr
;
842 struct acpi_processor_cx
*cx
= cpuidle_get_statedata(state
);
846 pr
= __get_cpu_var(processors
);
851 if (acpi_idle_suspend
)
852 return(acpi_idle_enter_c1(dev
, state
));
855 current_thread_info()->status
&= ~TS_POLLING
;
857 * TS_POLLING-cleared state must be visible before we test
862 if (unlikely(need_resched())) {
863 current_thread_info()->status
|= TS_POLLING
;
869 * Must be done before busmaster disable as we might need to
872 acpi_state_timer_broadcast(pr
, cx
, 1);
874 if (cx
->type
== ACPI_STATE_C3
)
875 ACPI_FLUSH_CPU_CACHE();
877 t1
= inl(acpi_gbl_FADT
.xpm_timer_block
.address
);
878 /* Tell the scheduler that we are going deep-idle: */
879 sched_clock_idle_sleep_event();
880 acpi_idle_do_entry(cx
);
881 t2
= inl(acpi_gbl_FADT
.xpm_timer_block
.address
);
883 #if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86)
884 /* TSC could halt in idle, so notify users */
885 if (tsc_halts_in_c(cx
->type
))
886 mark_tsc_unstable("TSC halts in idle");;
888 sleep_ticks
= ticks_elapsed(t1
, t2
);
890 /* Tell the scheduler how much we idled: */
891 sched_clock_idle_wakeup_event(sleep_ticks
*PM_TIMER_TICK_NS
);
894 current_thread_info()->status
|= TS_POLLING
;
898 acpi_state_timer_broadcast(pr
, cx
, 0);
899 cx
->time
+= sleep_ticks
;
900 return ticks_elapsed_in_us(t1
, t2
);
903 static int c3_cpu_count
;
904 static DEFINE_SPINLOCK(c3_lock
);
907 * acpi_idle_enter_bm - enters C3 with proper BM handling
908 * @dev: the target CPU
909 * @state: the state data
911 * If BM is detected, the deepest non-C3 idle state is entered instead.
913 static int acpi_idle_enter_bm(struct cpuidle_device
*dev
,
914 struct cpuidle_state
*state
)
916 struct acpi_processor
*pr
;
917 struct acpi_processor_cx
*cx
= cpuidle_get_statedata(state
);
921 pr
= __get_cpu_var(processors
);
926 if (acpi_idle_suspend
)
927 return(acpi_idle_enter_c1(dev
, state
));
929 if (acpi_idle_bm_check()) {
930 if (dev
->safe_state
) {
931 dev
->last_state
= dev
->safe_state
;
932 return dev
->safe_state
->enter(dev
, dev
->safe_state
);
942 current_thread_info()->status
&= ~TS_POLLING
;
944 * TS_POLLING-cleared state must be visible before we test
949 if (unlikely(need_resched())) {
950 current_thread_info()->status
|= TS_POLLING
;
955 acpi_unlazy_tlb(smp_processor_id());
957 /* Tell the scheduler that we are going deep-idle: */
958 sched_clock_idle_sleep_event();
960 * Must be done before busmaster disable as we might need to
963 acpi_state_timer_broadcast(pr
, cx
, 1);
967 * bm_check implies we need ARB_DIS
968 * !bm_check implies we need cache flush
969 * bm_control implies whether we can do ARB_DIS
971 * That leaves a case where bm_check is set and bm_control is
972 * not set. In that case we cannot do much, we enter C3
973 * without doing anything.
975 if (pr
->flags
.bm_check
&& pr
->flags
.bm_control
) {
978 /* Disable bus master arbitration when all CPUs are in C3 */
979 if (c3_cpu_count
== num_online_cpus())
980 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE
, 1);
981 spin_unlock(&c3_lock
);
982 } else if (!pr
->flags
.bm_check
) {
983 ACPI_FLUSH_CPU_CACHE();
986 t1
= inl(acpi_gbl_FADT
.xpm_timer_block
.address
);
987 acpi_idle_do_entry(cx
);
988 t2
= inl(acpi_gbl_FADT
.xpm_timer_block
.address
);
990 /* Re-enable bus master arbitration */
991 if (pr
->flags
.bm_check
&& pr
->flags
.bm_control
) {
993 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE
, 0);
995 spin_unlock(&c3_lock
);
998 #if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86)
999 /* TSC could halt in idle, so notify users */
1000 if (tsc_halts_in_c(ACPI_STATE_C3
))
1001 mark_tsc_unstable("TSC halts in idle");
1003 sleep_ticks
= ticks_elapsed(t1
, t2
);
1004 /* Tell the scheduler how much we idled: */
1005 sched_clock_idle_wakeup_event(sleep_ticks
*PM_TIMER_TICK_NS
);
1008 current_thread_info()->status
|= TS_POLLING
;
1012 acpi_state_timer_broadcast(pr
, cx
, 0);
1013 cx
->time
+= sleep_ticks
;
1014 return ticks_elapsed_in_us(t1
, t2
);
1017 struct cpuidle_driver acpi_idle_driver
= {
1018 .name
= "acpi_idle",
1019 .owner
= THIS_MODULE
,
1023 * acpi_processor_setup_cpuidle - prepares and configures CPUIDLE
1024 * @pr: the ACPI processor
1026 static int acpi_processor_setup_cpuidle(struct acpi_processor
*pr
)
1028 int i
, count
= CPUIDLE_DRIVER_STATE_START
;
1029 struct acpi_processor_cx
*cx
;
1030 struct cpuidle_state
*state
;
1031 struct cpuidle_device
*dev
= &pr
->power
.dev
;
1033 if (!pr
->flags
.power_setup_done
)
1036 if (pr
->flags
.power
== 0) {
1041 for (i
= 0; i
< CPUIDLE_STATE_MAX
; i
++) {
1042 dev
->states
[i
].name
[0] = '\0';
1043 dev
->states
[i
].desc
[0] = '\0';
1046 for (i
= 1; i
< ACPI_PROCESSOR_MAX_POWER
&& i
<= max_cstate
; i
++) {
1047 cx
= &pr
->power
.states
[i
];
1048 state
= &dev
->states
[count
];
1053 #ifdef CONFIG_HOTPLUG_CPU
1054 if ((cx
->type
!= ACPI_STATE_C1
) && (num_online_cpus() > 1) &&
1055 !pr
->flags
.has_cst
&&
1056 !(acpi_gbl_FADT
.flags
& ACPI_FADT_C2_MP_SUPPORTED
))
1059 cpuidle_set_statedata(state
, cx
);
1061 snprintf(state
->name
, CPUIDLE_NAME_LEN
, "C%d", i
);
1062 strncpy(state
->desc
, cx
->desc
, CPUIDLE_DESC_LEN
);
1063 state
->exit_latency
= cx
->latency
;
1064 state
->target_residency
= cx
->latency
* latency_factor
;
1065 state
->power_usage
= cx
->power
;
1070 state
->flags
|= CPUIDLE_FLAG_SHALLOW
;
1071 if (cx
->entry_method
== ACPI_CSTATE_FFH
)
1072 state
->flags
|= CPUIDLE_FLAG_TIME_VALID
;
1074 state
->enter
= acpi_idle_enter_c1
;
1075 dev
->safe_state
= state
;
1079 state
->flags
|= CPUIDLE_FLAG_BALANCED
;
1080 state
->flags
|= CPUIDLE_FLAG_TIME_VALID
;
1081 state
->enter
= acpi_idle_enter_simple
;
1082 dev
->safe_state
= state
;
1086 state
->flags
|= CPUIDLE_FLAG_DEEP
;
1087 state
->flags
|= CPUIDLE_FLAG_TIME_VALID
;
1088 state
->flags
|= CPUIDLE_FLAG_CHECK_BM
;
1089 state
->enter
= pr
->flags
.bm_check
?
1090 acpi_idle_enter_bm
:
1091 acpi_idle_enter_simple
;
1096 if (count
== CPUIDLE_STATE_MAX
)
1100 dev
->state_count
= count
;
1108 int acpi_processor_cst_has_changed(struct acpi_processor
*pr
)
1112 if (boot_option_idle_override
)
1122 if (!pr
->flags
.power_setup_done
)
1125 cpuidle_pause_and_lock();
1126 cpuidle_disable_device(&pr
->power
.dev
);
1127 acpi_processor_get_power_info(pr
);
1128 if (pr
->flags
.power
) {
1129 acpi_processor_setup_cpuidle(pr
);
1130 ret
= cpuidle_enable_device(&pr
->power
.dev
);
1132 cpuidle_resume_and_unlock();
1137 int __cpuinit
acpi_processor_power_init(struct acpi_processor
*pr
,
1138 struct acpi_device
*device
)
1140 acpi_status status
= 0;
1141 static int first_run
;
1142 struct proc_dir_entry
*entry
= NULL
;
1145 if (boot_option_idle_override
)
1151 * When the boot option of "idle=halt" is added, halt
1152 * is used for CPU IDLE.
1153 * In such case C2/C3 is meaningless. So the max_cstate
1158 dmi_check_system(processor_power_dmi_table
);
1159 max_cstate
= acpi_processor_cstate_check(max_cstate
);
1160 if (max_cstate
< ACPI_C_STATES_MAX
)
1162 "ACPI: processor limited to max C-state %d\n",
1170 if (acpi_gbl_FADT
.cst_control
&& !nocst
) {
1172 acpi_os_write_port(acpi_gbl_FADT
.smi_command
, acpi_gbl_FADT
.cst_control
, 8);
1173 if (ACPI_FAILURE(status
)) {
1174 ACPI_EXCEPTION((AE_INFO
, status
,
1175 "Notifying BIOS of _CST ability failed"));
1179 acpi_processor_get_power_info(pr
);
1180 pr
->flags
.power_setup_done
= 1;
1183 * Install the idle handler if processor power management is supported.
1184 * Note that we use previously set idle handler will be used on
1185 * platforms that only support C1.
1187 if (pr
->flags
.power
) {
1188 acpi_processor_setup_cpuidle(pr
);
1189 if (cpuidle_register_device(&pr
->power
.dev
))
1192 printk(KERN_INFO PREFIX
"CPU%d (power states:", pr
->id
);
1193 for (i
= 1; i
<= pr
->power
.count
; i
++)
1194 if (pr
->power
.states
[i
].valid
)
1195 printk(" C%d[C%d]", i
,
1196 pr
->power
.states
[i
].type
);
1201 entry
= proc_create_data(ACPI_PROCESSOR_FILE_POWER
,
1202 S_IRUGO
, acpi_device_dir(device
),
1203 &acpi_processor_power_fops
,
1204 acpi_driver_data(device
));
1210 int acpi_processor_power_exit(struct acpi_processor
*pr
,
1211 struct acpi_device
*device
)
1213 if (boot_option_idle_override
)
1216 cpuidle_unregister_device(&pr
->power
.dev
);
1217 pr
->flags
.power_setup_done
= 0;
1219 if (acpi_device_dir(device
))
1220 remove_proc_entry(ACPI_PROCESSOR_FILE_POWER
,
1221 acpi_device_dir(device
));