Automatic merge of rsync://www.parisc-linux.org/~jejb/git/scsi-for-linus-2.6.git
[linux-2.6] / drivers / acpi / processor_idle.c
1 /*
2  * processor_idle - idle state submodule to the ACPI processor driver
3  *
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  *
10  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
11  *
12  *  This program is free software; you can redistribute it and/or modify
13  *  it under the terms of the GNU General Public License as published by
14  *  the Free Software Foundation; either version 2 of the License, or (at
15  *  your option) any later version.
16  *
17  *  This program is distributed in the hope that it will be useful, but
18  *  WITHOUT ANY WARRANTY; without even the implied warranty of
19  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
20  *  General Public License for more details.
21  *
22  *  You should have received a copy of the GNU General Public License along
23  *  with this program; if not, write to the Free Software Foundation, Inc.,
24  *  59 Temple Place, Suite 330, Boston, MA 02111-1307 USA.
25  *
26  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
27  */
28
29 #include <linux/kernel.h>
30 #include <linux/module.h>
31 #include <linux/init.h>
32 #include <linux/cpufreq.h>
33 #include <linux/proc_fs.h>
34 #include <linux/seq_file.h>
35 #include <linux/acpi.h>
36 #include <linux/dmi.h>
37 #include <linux/moduleparam.h>
38
39 #include <asm/io.h>
40 #include <asm/uaccess.h>
41
42 #include <acpi/acpi_bus.h>
43 #include <acpi/processor.h>
44
45 #define ACPI_PROCESSOR_COMPONENT        0x01000000
46 #define ACPI_PROCESSOR_CLASS            "processor"
47 #define ACPI_PROCESSOR_DRIVER_NAME      "ACPI Processor Driver"
48 #define _COMPONENT              ACPI_PROCESSOR_COMPONENT
49 ACPI_MODULE_NAME                ("acpi_processor")
50
51 #define ACPI_PROCESSOR_FILE_POWER       "power"
52
53 #define US_TO_PM_TIMER_TICKS(t)         ((t * (PM_TIMER_FREQUENCY/1000)) / 1000)
54 #define C2_OVERHEAD                     4       /* 1us (3.579 ticks per us) */
55 #define C3_OVERHEAD                     4       /* 1us (3.579 ticks per us) */
56
57 static void (*pm_idle_save)(void);
58 module_param(max_cstate, uint, 0644);
59
60 static unsigned int nocst = 0;
61 module_param(nocst, uint, 0000);
62
63 /*
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
69  */
70 static unsigned int bm_history = (HZ >= 800 ? 0xFFFFFFFF : ((1U << (HZ / 25)) - 1));
71 module_param(bm_history, uint, 0644);
72 /* --------------------------------------------------------------------------
73                                 Power Management
74    -------------------------------------------------------------------------- */
75
76 /*
77  * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
78  * For now disable this. Probably a bug somewhere else.
79  *
80  * To skip this limit, boot/load with a large max_cstate limit.
81  */
82 static int no_c2c3(struct dmi_system_id *id)
83 {
84         if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
85                 return 0;
86
87         printk(KERN_NOTICE PREFIX "%s detected - C2,C3 disabled."
88                 " Override with \"processor.max_cstate=%d\"\n", id->ident,
89                ACPI_PROCESSOR_MAX_POWER + 1);
90
91         max_cstate = 1;
92
93         return 0;
94 }
95
96
97
98
99 static struct dmi_system_id __initdata processor_power_dmi_table[] = {
100         { no_c2c3, "IBM ThinkPad R40e", {
101           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
102           DMI_MATCH(DMI_BIOS_VERSION,"1SET60WW") }},
103         { no_c2c3, "Medion 41700", {
104           DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
105           DMI_MATCH(DMI_BIOS_VERSION,"R01-A1J") }},
106         {},
107 };
108
109
110 static inline u32
111 ticks_elapsed (
112         u32                     t1,
113         u32                     t2)
114 {
115         if (t2 >= t1)
116                 return (t2 - t1);
117         else if (!acpi_fadt.tmr_val_ext)
118                 return (((0x00FFFFFF - t1) + t2) & 0x00FFFFFF);
119         else
120                 return ((0xFFFFFFFF - t1) + t2);
121 }
122
123
124 static void
125 acpi_processor_power_activate (
126         struct acpi_processor   *pr,
127         struct acpi_processor_cx  *new)
128 {
129         struct acpi_processor_cx  *old;
130
131         if (!pr || !new)
132                 return;
133
134         old = pr->power.state;
135
136         if (old)
137                 old->promotion.count = 0;
138         new->demotion.count = 0;
139
140         /* Cleanup from old state. */
141         if (old) {
142                 switch (old->type) {
143                 case ACPI_STATE_C3:
144                         /* Disable bus master reload */
145                         if (new->type != ACPI_STATE_C3)
146                                 acpi_set_register(ACPI_BITREG_BUS_MASTER_RLD, 0, ACPI_MTX_DO_NOT_LOCK);
147                         break;
148                 }
149         }
150
151         /* Prepare to use new state. */
152         switch (new->type) {
153         case ACPI_STATE_C3:
154                 /* Enable bus master reload */
155                 if (old->type != ACPI_STATE_C3)
156                         acpi_set_register(ACPI_BITREG_BUS_MASTER_RLD, 1, ACPI_MTX_DO_NOT_LOCK);
157                 break;
158         }
159
160         pr->power.state = new;
161
162         return;
163 }
164
165
166 static void acpi_processor_idle (void)
167 {
168         struct acpi_processor   *pr = NULL;
169         struct acpi_processor_cx *cx = NULL;
170         struct acpi_processor_cx *next_state = NULL;
171         int                     sleep_ticks = 0;
172         u32                     t1, t2 = 0;
173
174         pr = processors[_smp_processor_id()];
175         if (!pr)
176                 return;
177
178         /*
179          * Interrupts must be disabled during bus mastering calculations and
180          * for C2/C3 transitions.
181          */
182         local_irq_disable();
183
184         /*
185          * Check whether we truly need to go idle, or should
186          * reschedule:
187          */
188         if (unlikely(need_resched())) {
189                 local_irq_enable();
190                 return;
191         }
192
193         cx = pr->power.state;
194         if (!cx)
195                 goto easy_out;
196
197         /*
198          * Check BM Activity
199          * -----------------
200          * Check for bus mastering activity (if required), record, and check
201          * for demotion.
202          */
203         if (pr->flags.bm_check) {
204                 u32             bm_status = 0;
205                 unsigned long   diff = jiffies - pr->power.bm_check_timestamp;
206
207                 if (diff > 32)
208                         diff = 32;
209
210                 while (diff) {
211                         /* if we didn't get called, assume there was busmaster activity */
212                         diff--;
213                         if (diff)
214                                 pr->power.bm_activity |= 0x1;
215                         pr->power.bm_activity <<= 1;
216                 }
217
218                 acpi_get_register(ACPI_BITREG_BUS_MASTER_STATUS,
219                         &bm_status, ACPI_MTX_DO_NOT_LOCK);
220                 if (bm_status) {
221                         pr->power.bm_activity++;
222                         acpi_set_register(ACPI_BITREG_BUS_MASTER_STATUS,
223                                 1, ACPI_MTX_DO_NOT_LOCK);
224                 }
225                 /*
226                  * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
227                  * the true state of bus mastering activity; forcing us to
228                  * manually check the BMIDEA bit of each IDE channel.
229                  */
230                 else if (errata.piix4.bmisx) {
231                         if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
232                                 || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
233                                 pr->power.bm_activity++;
234                 }
235
236                 pr->power.bm_check_timestamp = jiffies;
237
238                 /*
239                  * Apply bus mastering demotion policy.  Automatically demote
240                  * to avoid a faulty transition.  Note that the processor
241                  * won't enter a low-power state during this call (to this
242                  * funciton) but should upon the next.
243                  *
244                  * TBD: A better policy might be to fallback to the demotion
245                  *      state (use it for this quantum only) istead of
246                  *      demoting -- and rely on duration as our sole demotion
247                  *      qualification.  This may, however, introduce DMA
248                  *      issues (e.g. floppy DMA transfer overrun/underrun).
249                  */
250                 if (pr->power.bm_activity & cx->demotion.threshold.bm) {
251                         local_irq_enable();
252                         next_state = cx->demotion.state;
253                         goto end;
254                 }
255         }
256
257         cx->usage++;
258
259         /*
260          * Sleep:
261          * ------
262          * Invoke the current Cx state to put the processor to sleep.
263          */
264         switch (cx->type) {
265
266         case ACPI_STATE_C1:
267                 /*
268                  * Invoke C1.
269                  * Use the appropriate idle routine, the one that would
270                  * be used without acpi C-states.
271                  */
272                 if (pm_idle_save)
273                         pm_idle_save();
274                 else
275                         safe_halt();
276                 /*
277                  * TBD: Can't get time duration while in C1, as resumes
278                  *      go to an ISR rather than here.  Need to instrument
279                  *      base interrupt handler.
280                  */
281                 sleep_ticks = 0xFFFFFFFF;
282                 break;
283
284         case ACPI_STATE_C2:
285                 /* Get start time (ticks) */
286                 t1 = inl(acpi_fadt.xpm_tmr_blk.address);
287                 /* Invoke C2 */
288                 inb(cx->address);
289                 /* Dummy op - must do something useless after P_LVL2 read */
290                 t2 = inl(acpi_fadt.xpm_tmr_blk.address);
291                 /* Get end time (ticks) */
292                 t2 = inl(acpi_fadt.xpm_tmr_blk.address);
293                 /* Re-enable interrupts */
294                 local_irq_enable();
295                 /* Compute time (ticks) that we were actually asleep */
296                 sleep_ticks = ticks_elapsed(t1, t2) - cx->latency_ticks - C2_OVERHEAD;
297                 break;
298
299         case ACPI_STATE_C3:
300                 /* Disable bus master arbitration */
301                 acpi_set_register(ACPI_BITREG_ARB_DISABLE, 1, ACPI_MTX_DO_NOT_LOCK);
302                 /* Get start time (ticks) */
303                 t1 = inl(acpi_fadt.xpm_tmr_blk.address);
304                 /* Invoke C3 */
305                 inb(cx->address);
306                 /* Dummy op - must do something useless after P_LVL3 read */
307                 t2 = inl(acpi_fadt.xpm_tmr_blk.address);
308                 /* Get end time (ticks) */
309                 t2 = inl(acpi_fadt.xpm_tmr_blk.address);
310                 /* Enable bus master arbitration */
311                 acpi_set_register(ACPI_BITREG_ARB_DISABLE, 0, ACPI_MTX_DO_NOT_LOCK);
312                 /* Re-enable interrupts */
313                 local_irq_enable();
314                 /* Compute time (ticks) that we were actually asleep */
315                 sleep_ticks = ticks_elapsed(t1, t2) - cx->latency_ticks - C3_OVERHEAD;
316                 break;
317
318         default:
319                 local_irq_enable();
320                 return;
321         }
322
323         next_state = pr->power.state;
324
325         /*
326          * Promotion?
327          * ----------
328          * Track the number of longs (time asleep is greater than threshold)
329          * and promote when the count threshold is reached.  Note that bus
330          * mastering activity may prevent promotions.
331          * Do not promote above max_cstate.
332          */
333         if (cx->promotion.state &&
334             ((cx->promotion.state - pr->power.states) <= max_cstate)) {
335                 if (sleep_ticks > cx->promotion.threshold.ticks) {
336                         cx->promotion.count++;
337                         cx->demotion.count = 0;
338                         if (cx->promotion.count >= cx->promotion.threshold.count) {
339                                 if (pr->flags.bm_check) {
340                                         if (!(pr->power.bm_activity & cx->promotion.threshold.bm)) {
341                                                 next_state = cx->promotion.state;
342                                                 goto end;
343                                         }
344                                 }
345                                 else {
346                                         next_state = cx->promotion.state;
347                                         goto end;
348                                 }
349                         }
350                 }
351         }
352
353         /*
354          * Demotion?
355          * ---------
356          * Track the number of shorts (time asleep is less than time threshold)
357          * and demote when the usage threshold is reached.
358          */
359         if (cx->demotion.state) {
360                 if (sleep_ticks < cx->demotion.threshold.ticks) {
361                         cx->demotion.count++;
362                         cx->promotion.count = 0;
363                         if (cx->demotion.count >= cx->demotion.threshold.count) {
364                                 next_state = cx->demotion.state;
365                                 goto end;
366                         }
367                 }
368         }
369
370 end:
371         /*
372          * Demote if current state exceeds max_cstate
373          */
374         if ((pr->power.state - pr->power.states) > max_cstate) {
375                 if (cx->demotion.state)
376                         next_state = cx->demotion.state;
377         }
378
379         /*
380          * New Cx State?
381          * -------------
382          * If we're going to start using a new Cx state we must clean up
383          * from the previous and prepare to use the new.
384          */
385         if (next_state != pr->power.state)
386                 acpi_processor_power_activate(pr, next_state);
387
388         return;
389
390  easy_out:
391         /* do C1 instead of busy loop */
392         if (pm_idle_save)
393                 pm_idle_save();
394         else
395                 safe_halt();
396         return;
397 }
398
399
400 static int
401 acpi_processor_set_power_policy (
402         struct acpi_processor   *pr)
403 {
404         unsigned int i;
405         unsigned int state_is_set = 0;
406         struct acpi_processor_cx *lower = NULL;
407         struct acpi_processor_cx *higher = NULL;
408         struct acpi_processor_cx *cx;
409
410         ACPI_FUNCTION_TRACE("acpi_processor_set_power_policy");
411
412         if (!pr)
413                 return_VALUE(-EINVAL);
414
415         /*
416          * This function sets the default Cx state policy (OS idle handler).
417          * Our scheme is to promote quickly to C2 but more conservatively
418          * to C3.  We're favoring C2  for its characteristics of low latency
419          * (quick response), good power savings, and ability to allow bus
420          * mastering activity.  Note that the Cx state policy is completely
421          * customizable and can be altered dynamically.
422          */
423
424         /* startup state */
425         for (i=1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
426                 cx = &pr->power.states[i];
427                 if (!cx->valid)
428                         continue;
429
430                 if (!state_is_set)
431                         pr->power.state = cx;
432                 state_is_set++;
433                 break;
434         }
435
436         if (!state_is_set)
437                 return_VALUE(-ENODEV);
438
439         /* demotion */
440         for (i=1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
441                 cx = &pr->power.states[i];
442                 if (!cx->valid)
443                         continue;
444
445                 if (lower) {
446                         cx->demotion.state = lower;
447                         cx->demotion.threshold.ticks = cx->latency_ticks;
448                         cx->demotion.threshold.count = 1;
449                         if (cx->type == ACPI_STATE_C3)
450                                 cx->demotion.threshold.bm = bm_history;
451                 }
452
453                 lower = cx;
454         }
455
456         /* promotion */
457         for (i = (ACPI_PROCESSOR_MAX_POWER - 1); i > 0; i--) {
458                 cx = &pr->power.states[i];
459                 if (!cx->valid)
460                         continue;
461
462                 if (higher) {
463                         cx->promotion.state  = higher;
464                         cx->promotion.threshold.ticks = cx->latency_ticks;
465                         if (cx->type >= ACPI_STATE_C2)
466                                 cx->promotion.threshold.count = 4;
467                         else
468                                 cx->promotion.threshold.count = 10;
469                         if (higher->type == ACPI_STATE_C3)
470                                 cx->promotion.threshold.bm = bm_history;
471                 }
472
473                 higher = cx;
474         }
475
476         return_VALUE(0);
477 }
478
479
480 static int acpi_processor_get_power_info_fadt (struct acpi_processor *pr)
481 {
482         int i;
483
484         ACPI_FUNCTION_TRACE("acpi_processor_get_power_info_fadt");
485
486         if (!pr)
487                 return_VALUE(-EINVAL);
488
489         if (!pr->pblk)
490                 return_VALUE(-ENODEV);
491
492         for (i = 0; i < ACPI_PROCESSOR_MAX_POWER; i++)
493                 memset(pr->power.states, 0, sizeof(struct acpi_processor_cx));
494
495         /* if info is obtained from pblk/fadt, type equals state */
496         pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
497         pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
498         pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
499
500         /* the C0 state only exists as a filler in our array,
501          * and all processors need to support C1 */
502         pr->power.states[ACPI_STATE_C0].valid = 1;
503         pr->power.states[ACPI_STATE_C1].valid = 1;
504
505         /* determine C2 and C3 address from pblk */
506         pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
507         pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
508
509         /* determine latencies from FADT */
510         pr->power.states[ACPI_STATE_C2].latency = acpi_fadt.plvl2_lat;
511         pr->power.states[ACPI_STATE_C3].latency = acpi_fadt.plvl3_lat;
512
513         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
514                           "lvl2[0x%08x] lvl3[0x%08x]\n",
515                           pr->power.states[ACPI_STATE_C2].address,
516                           pr->power.states[ACPI_STATE_C3].address));
517
518         return_VALUE(0);
519 }
520
521
522 static int acpi_processor_get_power_info_cst (struct acpi_processor *pr)
523 {
524         acpi_status             status = 0;
525         acpi_integer            count;
526         int                     i;
527         struct acpi_buffer      buffer = {ACPI_ALLOCATE_BUFFER, NULL};
528         union acpi_object       *cst;
529
530         ACPI_FUNCTION_TRACE("acpi_processor_get_power_info_cst");
531
532         if (errata.smp)
533                 return_VALUE(-ENODEV);
534
535         if (nocst)
536                 return_VALUE(-ENODEV);
537
538         pr->power.count = 0;
539         for (i = 0; i < ACPI_PROCESSOR_MAX_POWER; i++)
540                 memset(pr->power.states, 0, sizeof(struct acpi_processor_cx));
541
542         status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
543         if (ACPI_FAILURE(status)) {
544                 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
545                 return_VALUE(-ENODEV);
546         }
547
548         cst = (union acpi_object *) buffer.pointer;
549
550         /* There must be at least 2 elements */
551         if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
552                 ACPI_DEBUG_PRINT((ACPI_DB_ERROR, "not enough elements in _CST\n"));
553                 status = -EFAULT;
554                 goto end;
555         }
556
557         count = cst->package.elements[0].integer.value;
558
559         /* Validate number of power states. */
560         if (count < 1 || count != cst->package.count - 1) {
561                 ACPI_DEBUG_PRINT((ACPI_DB_ERROR, "count given by _CST is not valid\n"));
562                 status = -EFAULT;
563                 goto end;
564         }
565
566         /* We support up to ACPI_PROCESSOR_MAX_POWER. */
567         if (count > ACPI_PROCESSOR_MAX_POWER) {
568                 printk(KERN_WARNING "Limiting number of power states to max (%d)\n", ACPI_PROCESSOR_MAX_POWER);
569                 printk(KERN_WARNING "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
570                 count = ACPI_PROCESSOR_MAX_POWER;
571         }
572
573         /* Tell driver that at least _CST is supported. */
574         pr->flags.has_cst = 1;
575
576         for (i = 1; i <= count; i++) {
577                 union acpi_object *element;
578                 union acpi_object *obj;
579                 struct acpi_power_register *reg;
580                 struct acpi_processor_cx cx;
581
582                 memset(&cx, 0, sizeof(cx));
583
584                 element = (union acpi_object *) &(cst->package.elements[i]);
585                 if (element->type != ACPI_TYPE_PACKAGE)
586                         continue;
587
588                 if (element->package.count != 4)
589                         continue;
590
591                 obj = (union acpi_object *) &(element->package.elements[0]);
592
593                 if (obj->type != ACPI_TYPE_BUFFER)
594                         continue;
595
596                 reg = (struct acpi_power_register *) obj->buffer.pointer;
597
598                 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
599                         (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
600                         continue;
601
602                 cx.address = (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) ?
603                         0 : reg->address;
604
605                 /* There should be an easy way to extract an integer... */
606                 obj = (union acpi_object *) &(element->package.elements[1]);
607                 if (obj->type != ACPI_TYPE_INTEGER)
608                         continue;
609
610                 cx.type = obj->integer.value;
611
612                 if ((cx.type != ACPI_STATE_C1) &&
613                     (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO))
614                         continue;
615
616                 if ((cx.type < ACPI_STATE_C1) ||
617                     (cx.type > ACPI_STATE_C3))
618                         continue;
619
620                 obj = (union acpi_object *) &(element->package.elements[2]);
621                 if (obj->type != ACPI_TYPE_INTEGER)
622                         continue;
623
624                 cx.latency = obj->integer.value;
625
626                 obj = (union acpi_object *) &(element->package.elements[3]);
627                 if (obj->type != ACPI_TYPE_INTEGER)
628                         continue;
629
630                 cx.power = obj->integer.value;
631
632                 (pr->power.count)++;
633                 memcpy(&(pr->power.states[pr->power.count]), &cx, sizeof(cx));
634         }
635
636         ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n", pr->power.count));
637
638         /* Validate number of power states discovered */
639         if (pr->power.count < 2)
640                 status = -ENODEV;
641
642 end:
643         acpi_os_free(buffer.pointer);
644
645         return_VALUE(status);
646 }
647
648
649 static void acpi_processor_power_verify_c2(struct acpi_processor_cx *cx)
650 {
651         ACPI_FUNCTION_TRACE("acpi_processor_get_power_verify_c2");
652
653         if (!cx->address)
654                 return_VOID;
655
656         /*
657          * C2 latency must be less than or equal to 100
658          * microseconds.
659          */
660         else if (cx->latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
661                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
662                                   "latency too large [%d]\n",
663                                   cx->latency));
664                 return_VOID;
665         }
666
667         /* We're (currently) only supporting C2 on UP */
668         else if (errata.smp) {
669                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
670                                   "C2 not supported in SMP mode\n"));
671                 return_VOID;
672         }
673
674         /*
675          * Otherwise we've met all of our C2 requirements.
676          * Normalize the C2 latency to expidite policy
677          */
678         cx->valid = 1;
679         cx->latency_ticks = US_TO_PM_TIMER_TICKS(cx->latency);
680
681         return_VOID;
682 }
683
684
685 static void acpi_processor_power_verify_c3(
686         struct acpi_processor *pr,
687         struct acpi_processor_cx *cx)
688 {
689         ACPI_FUNCTION_TRACE("acpi_processor_get_power_verify_c3");
690
691         if (!cx->address)
692                 return_VOID;
693
694         /*
695          * C3 latency must be less than or equal to 1000
696          * microseconds.
697          */
698         else if (cx->latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
699                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
700                                   "latency too large [%d]\n",
701                                   cx->latency));
702                 return_VOID;
703         }
704
705         /* bus mastering control is necessary */
706         else if (!pr->flags.bm_control) {
707                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
708                                   "C3 support requires bus mastering control\n"));
709                 return_VOID;
710         }
711
712         /* We're (currently) only supporting C2 on UP */
713         else if (errata.smp) {
714                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
715                                   "C3 not supported in SMP mode\n"));
716                 return_VOID;
717         }
718
719         /*
720          * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
721          * DMA transfers are used by any ISA device to avoid livelock.
722          * Note that we could disable Type-F DMA (as recommended by
723          * the erratum), but this is known to disrupt certain ISA
724          * devices thus we take the conservative approach.
725          */
726         else if (errata.piix4.fdma) {
727                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
728                         "C3 not supported on PIIX4 with Type-F DMA\n"));
729                 return_VOID;
730         }
731
732         /*
733          * Otherwise we've met all of our C3 requirements.
734          * Normalize the C3 latency to expidite policy.  Enable
735          * checking of bus mastering status (bm_check) so we can
736          * use this in our C3 policy
737          */
738         cx->valid = 1;
739         cx->latency_ticks = US_TO_PM_TIMER_TICKS(cx->latency);
740         pr->flags.bm_check = 1;
741
742         return_VOID;
743 }
744
745
746 static int acpi_processor_power_verify(struct acpi_processor *pr)
747 {
748         unsigned int i;
749         unsigned int working = 0;
750
751         for (i=1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
752                 struct acpi_processor_cx *cx = &pr->power.states[i];
753
754                 switch (cx->type) {
755                 case ACPI_STATE_C1:
756                         cx->valid = 1;
757                         break;
758
759                 case ACPI_STATE_C2:
760                         acpi_processor_power_verify_c2(cx);
761                         break;
762
763                 case ACPI_STATE_C3:
764                         acpi_processor_power_verify_c3(pr, cx);
765                         break;
766                 }
767
768                 if (cx->valid)
769                         working++;
770         }
771
772         return (working);
773 }
774
775 static int acpi_processor_get_power_info (
776         struct acpi_processor   *pr)
777 {
778         unsigned int i;
779         int result;
780
781         ACPI_FUNCTION_TRACE("acpi_processor_get_power_info");
782
783         /* NOTE: the idle thread may not be running while calling
784          * this function */
785
786         result = acpi_processor_get_power_info_cst(pr);
787         if ((result) || (acpi_processor_power_verify(pr) < 2)) {
788                 result = acpi_processor_get_power_info_fadt(pr);
789                 if (result)
790                         return_VALUE(result);
791
792                 if (acpi_processor_power_verify(pr) < 2)
793                         return_VALUE(-ENODEV);
794         }
795
796         /*
797          * Set Default Policy
798          * ------------------
799          * Now that we know which states are supported, set the default
800          * policy.  Note that this policy can be changed dynamically
801          * (e.g. encourage deeper sleeps to conserve battery life when
802          * not on AC).
803          */
804         result = acpi_processor_set_power_policy(pr);
805         if (result)
806                 return_VALUE(result);
807
808         /*
809          * if one state of type C2 or C3 is available, mark this
810          * CPU as being "idle manageable"
811          */
812         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
813                 if (pr->power.states[i].valid)
814                         pr->power.count = i;
815                 if ((pr->power.states[i].valid) &&
816                     (pr->power.states[i].type >= ACPI_STATE_C2))
817                         pr->flags.power = 1;
818         }
819
820         return_VALUE(0);
821 }
822
823 int acpi_processor_cst_has_changed (struct acpi_processor *pr)
824 {
825         int                     result = 0;
826
827         ACPI_FUNCTION_TRACE("acpi_processor_cst_has_changed");
828
829         if (!pr)
830                 return_VALUE(-EINVAL);
831
832         if (errata.smp || nocst) {
833                 return_VALUE(-ENODEV);
834         }
835
836         if (!pr->flags.power_setup_done)
837                 return_VALUE(-ENODEV);
838
839         /* Fall back to the default idle loop */
840         pm_idle = pm_idle_save;
841         synchronize_sched();  /* Relies on interrupts forcing exit from idle. */
842
843         pr->flags.power = 0;
844         result = acpi_processor_get_power_info(pr);
845         if ((pr->flags.power == 1) && (pr->flags.power_setup_done))
846                 pm_idle = acpi_processor_idle;
847
848         return_VALUE(result);
849 }
850
851 /* proc interface */
852
853 static int acpi_processor_power_seq_show(struct seq_file *seq, void *offset)
854 {
855         struct acpi_processor   *pr = (struct acpi_processor *)seq->private;
856         unsigned int            i;
857
858         ACPI_FUNCTION_TRACE("acpi_processor_power_seq_show");
859
860         if (!pr)
861                 goto end;
862
863         seq_printf(seq, "active state:            C%zd\n"
864                         "max_cstate:              C%d\n"
865                         "bus master activity:     %08x\n",
866                         pr->power.state ? pr->power.state - pr->power.states : 0,
867                         max_cstate,
868                         (unsigned)pr->power.bm_activity);
869
870         seq_puts(seq, "states:\n");
871
872         for (i = 1; i <= pr->power.count; i++) {
873                 seq_printf(seq, "   %cC%d:                  ",
874                         (&pr->power.states[i] == pr->power.state?'*':' '), i);
875
876                 if (!pr->power.states[i].valid) {
877                         seq_puts(seq, "<not supported>\n");
878                         continue;
879                 }
880
881                 switch (pr->power.states[i].type) {
882                 case ACPI_STATE_C1:
883                         seq_printf(seq, "type[C1] ");
884                         break;
885                 case ACPI_STATE_C2:
886                         seq_printf(seq, "type[C2] ");
887                         break;
888                 case ACPI_STATE_C3:
889                         seq_printf(seq, "type[C3] ");
890                         break;
891                 default:
892                         seq_printf(seq, "type[--] ");
893                         break;
894                 }
895
896                 if (pr->power.states[i].promotion.state)
897                         seq_printf(seq, "promotion[C%zd] ",
898                                 (pr->power.states[i].promotion.state -
899                                  pr->power.states));
900                 else
901                         seq_puts(seq, "promotion[--] ");
902
903                 if (pr->power.states[i].demotion.state)
904                         seq_printf(seq, "demotion[C%zd] ",
905                                 (pr->power.states[i].demotion.state -
906                                  pr->power.states));
907                 else
908                         seq_puts(seq, "demotion[--] ");
909
910                 seq_printf(seq, "latency[%03d] usage[%08d]\n",
911                         pr->power.states[i].latency,
912                         pr->power.states[i].usage);
913         }
914
915 end:
916         return_VALUE(0);
917 }
918
919 static int acpi_processor_power_open_fs(struct inode *inode, struct file *file)
920 {
921         return single_open(file, acpi_processor_power_seq_show,
922                                                 PDE(inode)->data);
923 }
924
925 static struct file_operations acpi_processor_power_fops = {
926         .open           = acpi_processor_power_open_fs,
927         .read           = seq_read,
928         .llseek         = seq_lseek,
929         .release        = single_release,
930 };
931
932
933 int acpi_processor_power_init(struct acpi_processor *pr, struct acpi_device *device)
934 {
935         acpi_status             status = 0;
936         static int              first_run = 0;
937         struct proc_dir_entry   *entry = NULL;
938         unsigned int i;
939
940         ACPI_FUNCTION_TRACE("acpi_processor_power_init");
941
942         if (!first_run) {
943                 dmi_check_system(processor_power_dmi_table);
944                 if (max_cstate < ACPI_C_STATES_MAX)
945                         printk(KERN_NOTICE "ACPI: processor limited to max C-state %d\n", max_cstate);
946                 first_run++;
947         }
948
949         if (!errata.smp && (pr->id == 0) && acpi_fadt.cst_cnt && !nocst) {
950                 status = acpi_os_write_port(acpi_fadt.smi_cmd, acpi_fadt.cst_cnt, 8);
951                 if (ACPI_FAILURE(status)) {
952                         ACPI_DEBUG_PRINT((ACPI_DB_ERROR,
953                                           "Notifying BIOS of _CST ability failed\n"));
954                 }
955         }
956
957         acpi_processor_get_power_info(pr);
958
959         /*
960          * Install the idle handler if processor power management is supported.
961          * Note that we use previously set idle handler will be used on
962          * platforms that only support C1.
963          */
964         if ((pr->flags.power) && (!boot_option_idle_override)) {
965                 printk(KERN_INFO PREFIX "CPU%d (power states:", pr->id);
966                 for (i = 1; i <= pr->power.count; i++)
967                         if (pr->power.states[i].valid)
968                                 printk(" C%d[C%d]", i, pr->power.states[i].type);
969                 printk(")\n");
970
971                 if (pr->id == 0) {
972                         pm_idle_save = pm_idle;
973                         pm_idle = acpi_processor_idle;
974                 }
975         }
976
977         /* 'power' [R] */
978         entry = create_proc_entry(ACPI_PROCESSOR_FILE_POWER,
979                 S_IRUGO, acpi_device_dir(device));
980         if (!entry)
981                 ACPI_DEBUG_PRINT((ACPI_DB_ERROR,
982                         "Unable to create '%s' fs entry\n",
983                         ACPI_PROCESSOR_FILE_POWER));
984         else {
985                 entry->proc_fops = &acpi_processor_power_fops;
986                 entry->data = acpi_driver_data(device);
987                 entry->owner = THIS_MODULE;
988         }
989
990         pr->flags.power_setup_done = 1;
991
992         return_VALUE(0);
993 }
994
995 int acpi_processor_power_exit(struct acpi_processor *pr, struct acpi_device *device)
996 {
997         ACPI_FUNCTION_TRACE("acpi_processor_power_exit");
998
999         pr->flags.power_setup_done = 0;
1000
1001         if (acpi_device_dir(device))
1002                 remove_proc_entry(ACPI_PROCESSOR_FILE_POWER,acpi_device_dir(device));
1003
1004         /* Unregister the idle handler when processor #0 is removed. */
1005         if (pr->id == 0) {
1006                 pm_idle = pm_idle_save;
1007
1008                 /*
1009                  * We are about to unload the current idle thread pm callback
1010                  * (pm_idle), Wait for all processors to update cached/local
1011                  * copies of pm_idle before proceeding.
1012                  */
1013                 cpu_idle_wait();
1014         }
1015
1016         return_VALUE(0);
1017 }