ACPI: processor: move device _HID into driver
[linux-2.6.git] / 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, 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
11  *
12  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
13  *
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.
18  *
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.
23  *
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.
27  *
28  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
29  */
30
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>
45
46 /*
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.
51  */
52 #ifdef CONFIG_X86
53 #include <asm/apic.h>
54 #endif
55
56 #include <asm/io.h>
57 #include <asm/uaccess.h>
58
59 #include <acpi/acpi_bus.h>
60 #include <acpi/processor.h>
61 #include <asm/processor.h>
62
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 PM_TIMER_TICK_NS                (1000000000ULL/PM_TIMER_FREQUENCY)
68 #define C2_OVERHEAD                     1       /* 1us */
69 #define C3_OVERHEAD                     1       /* 1us */
70 #define PM_TIMER_TICKS_TO_US(p)         (((p) * 1000)/(PM_TIMER_FREQUENCY/1000))
71
72 static unsigned int max_cstate __read_mostly = ACPI_PROCESSOR_MAX_POWER;
73 module_param(max_cstate, uint, 0000);
74 static unsigned int nocst __read_mostly;
75 module_param(nocst, uint, 0000);
76
77 static unsigned int latency_factor __read_mostly = 2;
78 module_param(latency_factor, uint, 0644);
79
80 static s64 us_to_pm_timer_ticks(s64 t)
81 {
82         return div64_u64(t * PM_TIMER_FREQUENCY, 1000000);
83 }
84 /*
85  * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
86  * For now disable this. Probably a bug somewhere else.
87  *
88  * To skip this limit, boot/load with a large max_cstate limit.
89  */
90 static int set_max_cstate(const struct dmi_system_id *id)
91 {
92         if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
93                 return 0;
94
95         printk(KERN_NOTICE PREFIX "%s detected - limiting to C%ld max_cstate."
96                " Override with \"processor.max_cstate=%d\"\n", id->ident,
97                (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
98
99         max_cstate = (long)id->driver_data;
100
101         return 0;
102 }
103
104 /* Actually this shouldn't be __cpuinitdata, would be better to fix the
105    callers to only run once -AK */
106 static struct dmi_system_id __cpuinitdata processor_power_dmi_table[] = {
107         { set_max_cstate, "Clevo 5600D", {
108           DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
109           DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")},
110          (void *)2},
111         {},
112 };
113
114
115 /*
116  * Callers should disable interrupts before the call and enable
117  * interrupts after return.
118  */
119 static void acpi_safe_halt(void)
120 {
121         current_thread_info()->status &= ~TS_POLLING;
122         /*
123          * TS_POLLING-cleared state must be visible before we
124          * test NEED_RESCHED:
125          */
126         smp_mb();
127         if (!need_resched()) {
128                 safe_halt();
129                 local_irq_disable();
130         }
131         current_thread_info()->status |= TS_POLLING;
132 }
133
134 #ifdef ARCH_APICTIMER_STOPS_ON_C3
135
136 /*
137  * Some BIOS implementations switch to C3 in the published C2 state.
138  * This seems to be a common problem on AMD boxen, but other vendors
139  * are affected too. We pick the most conservative approach: we assume
140  * that the local APIC stops in both C2 and C3.
141  */
142 static void acpi_timer_check_state(int state, struct acpi_processor *pr,
143                                    struct acpi_processor_cx *cx)
144 {
145         struct acpi_processor_power *pwr = &pr->power;
146         u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2;
147
148         if (cpu_has(&cpu_data(pr->id), X86_FEATURE_ARAT))
149                 return;
150
151         /*
152          * Check, if one of the previous states already marked the lapic
153          * unstable
154          */
155         if (pwr->timer_broadcast_on_state < state)
156                 return;
157
158         if (cx->type >= type)
159                 pr->power.timer_broadcast_on_state = state;
160 }
161
162 static void acpi_propagate_timer_broadcast(struct acpi_processor *pr)
163 {
164         unsigned long reason;
165
166         reason = pr->power.timer_broadcast_on_state < INT_MAX ?
167                 CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
168
169         clockevents_notify(reason, &pr->id);
170 }
171
172 /* Power(C) State timer broadcast control */
173 static void acpi_state_timer_broadcast(struct acpi_processor *pr,
174                                        struct acpi_processor_cx *cx,
175                                        int broadcast)
176 {
177         int state = cx - pr->power.states;
178
179         if (state >= pr->power.timer_broadcast_on_state) {
180                 unsigned long reason;
181
182                 reason = broadcast ?  CLOCK_EVT_NOTIFY_BROADCAST_ENTER :
183                         CLOCK_EVT_NOTIFY_BROADCAST_EXIT;
184                 clockevents_notify(reason, &pr->id);
185         }
186 }
187
188 #else
189
190 static void acpi_timer_check_state(int state, struct acpi_processor *pr,
191                                    struct acpi_processor_cx *cstate) { }
192 static void acpi_propagate_timer_broadcast(struct acpi_processor *pr) { }
193 static void acpi_state_timer_broadcast(struct acpi_processor *pr,
194                                        struct acpi_processor_cx *cx,
195                                        int broadcast)
196 {
197 }
198
199 #endif
200
201 /*
202  * Suspend / resume control
203  */
204 static int acpi_idle_suspend;
205 static u32 saved_bm_rld;
206
207 static void acpi_idle_bm_rld_save(void)
208 {
209         acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &saved_bm_rld);
210 }
211 static void acpi_idle_bm_rld_restore(void)
212 {
213         u32 resumed_bm_rld;
214
215         acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &resumed_bm_rld);
216
217         if (resumed_bm_rld != saved_bm_rld)
218                 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, saved_bm_rld);
219 }
220
221 int acpi_processor_suspend(struct acpi_device * device, pm_message_t state)
222 {
223         if (acpi_idle_suspend == 1)
224                 return 0;
225
226         acpi_idle_bm_rld_save();
227         acpi_idle_suspend = 1;
228         return 0;
229 }
230
231 int acpi_processor_resume(struct acpi_device * device)
232 {
233         if (acpi_idle_suspend == 0)
234                 return 0;
235
236         acpi_idle_bm_rld_restore();
237         acpi_idle_suspend = 0;
238         return 0;
239 }
240
241 #if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86)
242 static void tsc_check_state(int state)
243 {
244         switch (boot_cpu_data.x86_vendor) {
245         case X86_VENDOR_AMD:
246         case X86_VENDOR_INTEL:
247                 /*
248                  * AMD Fam10h TSC will tick in all
249                  * C/P/S0/S1 states when this bit is set.
250                  */
251                 if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
252                         return;
253
254                 /*FALL THROUGH*/
255         default:
256                 /* TSC could halt in idle, so notify users */
257                 if (state > ACPI_STATE_C1)
258                         mark_tsc_unstable("TSC halts in idle");
259         }
260 }
261 #else
262 static void tsc_check_state(int state) { return; }
263 #endif
264
265 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
266 {
267
268         if (!pr)
269                 return -EINVAL;
270
271         if (!pr->pblk)
272                 return -ENODEV;
273
274         /* if info is obtained from pblk/fadt, type equals state */
275         pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
276         pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
277
278 #ifndef CONFIG_HOTPLUG_CPU
279         /*
280          * Check for P_LVL2_UP flag before entering C2 and above on
281          * an SMP system.
282          */
283         if ((num_online_cpus() > 1) &&
284             !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
285                 return -ENODEV;
286 #endif
287
288         /* determine C2 and C3 address from pblk */
289         pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
290         pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
291
292         /* determine latencies from FADT */
293         pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.C2latency;
294         pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.C3latency;
295
296         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
297                           "lvl2[0x%08x] lvl3[0x%08x]\n",
298                           pr->power.states[ACPI_STATE_C2].address,
299                           pr->power.states[ACPI_STATE_C3].address));
300
301         return 0;
302 }
303
304 static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
305 {
306         if (!pr->power.states[ACPI_STATE_C1].valid) {
307                 /* set the first C-State to C1 */
308                 /* all processors need to support C1 */
309                 pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
310                 pr->power.states[ACPI_STATE_C1].valid = 1;
311                 pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
312         }
313         /* the C0 state only exists as a filler in our array */
314         pr->power.states[ACPI_STATE_C0].valid = 1;
315         return 0;
316 }
317
318 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
319 {
320         acpi_status status = 0;
321         acpi_integer count;
322         int current_count;
323         int i;
324         struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
325         union acpi_object *cst;
326
327
328         if (nocst)
329                 return -ENODEV;
330
331         current_count = 0;
332
333         status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
334         if (ACPI_FAILURE(status)) {
335                 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
336                 return -ENODEV;
337         }
338
339         cst = buffer.pointer;
340
341         /* There must be at least 2 elements */
342         if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
343                 printk(KERN_ERR PREFIX "not enough elements in _CST\n");
344                 status = -EFAULT;
345                 goto end;
346         }
347
348         count = cst->package.elements[0].integer.value;
349
350         /* Validate number of power states. */
351         if (count < 1 || count != cst->package.count - 1) {
352                 printk(KERN_ERR PREFIX "count given by _CST is not valid\n");
353                 status = -EFAULT;
354                 goto end;
355         }
356
357         /* Tell driver that at least _CST is supported. */
358         pr->flags.has_cst = 1;
359
360         for (i = 1; i <= count; i++) {
361                 union acpi_object *element;
362                 union acpi_object *obj;
363                 struct acpi_power_register *reg;
364                 struct acpi_processor_cx cx;
365
366                 memset(&cx, 0, sizeof(cx));
367
368                 element = &(cst->package.elements[i]);
369                 if (element->type != ACPI_TYPE_PACKAGE)
370                         continue;
371
372                 if (element->package.count != 4)
373                         continue;
374
375                 obj = &(element->package.elements[0]);
376
377                 if (obj->type != ACPI_TYPE_BUFFER)
378                         continue;
379
380                 reg = (struct acpi_power_register *)obj->buffer.pointer;
381
382                 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
383                     (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
384                         continue;
385
386                 /* There should be an easy way to extract an integer... */
387                 obj = &(element->package.elements[1]);
388                 if (obj->type != ACPI_TYPE_INTEGER)
389                         continue;
390
391                 cx.type = obj->integer.value;
392                 /*
393                  * Some buggy BIOSes won't list C1 in _CST -
394                  * Let acpi_processor_get_power_info_default() handle them later
395                  */
396                 if (i == 1 && cx.type != ACPI_STATE_C1)
397                         current_count++;
398
399                 cx.address = reg->address;
400                 cx.index = current_count + 1;
401
402                 cx.entry_method = ACPI_CSTATE_SYSTEMIO;
403                 if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) {
404                         if (acpi_processor_ffh_cstate_probe
405                                         (pr->id, &cx, reg) == 0) {
406                                 cx.entry_method = ACPI_CSTATE_FFH;
407                         } else if (cx.type == ACPI_STATE_C1) {
408                                 /*
409                                  * C1 is a special case where FIXED_HARDWARE
410                                  * can be handled in non-MWAIT way as well.
411                                  * In that case, save this _CST entry info.
412                                  * Otherwise, ignore this info and continue.
413                                  */
414                                 cx.entry_method = ACPI_CSTATE_HALT;
415                                 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
416                         } else {
417                                 continue;
418                         }
419                         if (cx.type == ACPI_STATE_C1 &&
420                                         (idle_halt || idle_nomwait)) {
421                                 /*
422                                  * In most cases the C1 space_id obtained from
423                                  * _CST object is FIXED_HARDWARE access mode.
424                                  * But when the option of idle=halt is added,
425                                  * the entry_method type should be changed from
426                                  * CSTATE_FFH to CSTATE_HALT.
427                                  * When the option of idle=nomwait is added,
428                                  * the C1 entry_method type should be
429                                  * CSTATE_HALT.
430                                  */
431                                 cx.entry_method = ACPI_CSTATE_HALT;
432                                 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
433                         }
434                 } else {
435                         snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI IOPORT 0x%x",
436                                  cx.address);
437                 }
438
439                 if (cx.type == ACPI_STATE_C1) {
440                         cx.valid = 1;
441                 }
442
443                 obj = &(element->package.elements[2]);
444                 if (obj->type != ACPI_TYPE_INTEGER)
445                         continue;
446
447                 cx.latency = obj->integer.value;
448
449                 obj = &(element->package.elements[3]);
450                 if (obj->type != ACPI_TYPE_INTEGER)
451                         continue;
452
453                 cx.power = obj->integer.value;
454
455                 current_count++;
456                 memcpy(&(pr->power.states[current_count]), &cx, sizeof(cx));
457
458                 /*
459                  * We support total ACPI_PROCESSOR_MAX_POWER - 1
460                  * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
461                  */
462                 if (current_count >= (ACPI_PROCESSOR_MAX_POWER - 1)) {
463                         printk(KERN_WARNING
464                                "Limiting number of power states to max (%d)\n",
465                                ACPI_PROCESSOR_MAX_POWER);
466                         printk(KERN_WARNING
467                                "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
468                         break;
469                 }
470         }
471
472         ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n",
473                           current_count));
474
475         /* Validate number of power states discovered */
476         if (current_count < 2)
477                 status = -EFAULT;
478
479       end:
480         kfree(buffer.pointer);
481
482         return status;
483 }
484
485 static void acpi_processor_power_verify_c2(struct acpi_processor_cx *cx)
486 {
487
488         if (!cx->address)
489                 return;
490
491         /*
492          * C2 latency must be less than or equal to 100
493          * microseconds.
494          */
495         else if (cx->latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
496                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
497                                   "latency too large [%d]\n", cx->latency));
498                 return;
499         }
500
501         /*
502          * Otherwise we've met all of our C2 requirements.
503          * Normalize the C2 latency to expidite policy
504          */
505         cx->valid = 1;
506
507         cx->latency_ticks = cx->latency;
508
509         return;
510 }
511
512 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
513                                            struct acpi_processor_cx *cx)
514 {
515         static int bm_check_flag;
516
517
518         if (!cx->address)
519                 return;
520
521         /*
522          * C3 latency must be less than or equal to 1000
523          * microseconds.
524          */
525         else if (cx->latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
526                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
527                                   "latency too large [%d]\n", cx->latency));
528                 return;
529         }
530
531         /*
532          * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
533          * DMA transfers are used by any ISA device to avoid livelock.
534          * Note that we could disable Type-F DMA (as recommended by
535          * the erratum), but this is known to disrupt certain ISA
536          * devices thus we take the conservative approach.
537          */
538         else if (errata.piix4.fdma) {
539                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
540                                   "C3 not supported on PIIX4 with Type-F DMA\n"));
541                 return;
542         }
543
544         /* All the logic here assumes flags.bm_check is same across all CPUs */
545         if (!bm_check_flag) {
546                 /* Determine whether bm_check is needed based on CPU  */
547                 acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
548                 bm_check_flag = pr->flags.bm_check;
549         } else {
550                 pr->flags.bm_check = bm_check_flag;
551         }
552
553         if (pr->flags.bm_check) {
554                 if (!pr->flags.bm_control) {
555                         if (pr->flags.has_cst != 1) {
556                                 /* bus mastering control is necessary */
557                                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
558                                         "C3 support requires BM control\n"));
559                                 return;
560                         } else {
561                                 /* Here we enter C3 without bus mastering */
562                                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
563                                         "C3 support without BM control\n"));
564                         }
565                 }
566         } else {
567                 /*
568                  * WBINVD should be set in fadt, for C3 state to be
569                  * supported on when bm_check is not required.
570                  */
571                 if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
572                         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
573                                           "Cache invalidation should work properly"
574                                           " for C3 to be enabled on SMP systems\n"));
575                         return;
576                 }
577         }
578
579         /*
580          * Otherwise we've met all of our C3 requirements.
581          * Normalize the C3 latency to expidite policy.  Enable
582          * checking of bus mastering status (bm_check) so we can
583          * use this in our C3 policy
584          */
585         cx->valid = 1;
586
587         cx->latency_ticks = cx->latency;
588         /*
589          * On older chipsets, BM_RLD needs to be set
590          * in order for Bus Master activity to wake the
591          * system from C3.  Newer chipsets handle DMA
592          * during C3 automatically and BM_RLD is a NOP.
593          * In either case, the proper way to
594          * handle BM_RLD is to set it and leave it set.
595          */
596         acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
597
598         return;
599 }
600
601 static int acpi_processor_power_verify(struct acpi_processor *pr)
602 {
603         unsigned int i;
604         unsigned int working = 0;
605
606         pr->power.timer_broadcast_on_state = INT_MAX;
607
608         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
609                 struct acpi_processor_cx *cx = &pr->power.states[i];
610
611                 switch (cx->type) {
612                 case ACPI_STATE_C1:
613                         cx->valid = 1;
614                         break;
615
616                 case ACPI_STATE_C2:
617                         acpi_processor_power_verify_c2(cx);
618                         if (cx->valid)
619                                 acpi_timer_check_state(i, pr, cx);
620                         break;
621
622                 case ACPI_STATE_C3:
623                         acpi_processor_power_verify_c3(pr, cx);
624                         if (cx->valid)
625                                 acpi_timer_check_state(i, pr, cx);
626                         break;
627                 }
628                 if (cx->valid)
629                         tsc_check_state(cx->type);
630
631                 if (cx->valid)
632                         working++;
633         }
634
635         acpi_propagate_timer_broadcast(pr);
636
637         return (working);
638 }
639
640 static int acpi_processor_get_power_info(struct acpi_processor *pr)
641 {
642         unsigned int i;
643         int result;
644
645
646         /* NOTE: the idle thread may not be running while calling
647          * this function */
648
649         /* Zero initialize all the C-states info. */
650         memset(pr->power.states, 0, sizeof(pr->power.states));
651
652         result = acpi_processor_get_power_info_cst(pr);
653         if (result == -ENODEV)
654                 result = acpi_processor_get_power_info_fadt(pr);
655
656         if (result)
657                 return result;
658
659         acpi_processor_get_power_info_default(pr);
660
661         pr->power.count = acpi_processor_power_verify(pr);
662
663         /*
664          * if one state of type C2 or C3 is available, mark this
665          * CPU as being "idle manageable"
666          */
667         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
668                 if (pr->power.states[i].valid) {
669                         pr->power.count = i;
670                         if (pr->power.states[i].type >= ACPI_STATE_C2)
671                                 pr->flags.power = 1;
672                 }
673         }
674
675         return 0;
676 }
677
678 static int acpi_processor_power_seq_show(struct seq_file *seq, void *offset)
679 {
680         struct acpi_processor *pr = seq->private;
681         unsigned int i;
682
683
684         if (!pr)
685                 goto end;
686
687         seq_printf(seq, "active state:            C%zd\n"
688                    "max_cstate:              C%d\n"
689                    "maximum allowed latency: %d usec\n",
690                    pr->power.state ? pr->power.state - pr->power.states : 0,
691                    max_cstate, pm_qos_requirement(PM_QOS_CPU_DMA_LATENCY));
692
693         seq_puts(seq, "states:\n");
694
695         for (i = 1; i <= pr->power.count; i++) {
696                 seq_printf(seq, "   %cC%d:                  ",
697                            (&pr->power.states[i] ==
698                             pr->power.state ? '*' : ' '), i);
699
700                 if (!pr->power.states[i].valid) {
701                         seq_puts(seq, "<not supported>\n");
702                         continue;
703                 }
704
705                 switch (pr->power.states[i].type) {
706                 case ACPI_STATE_C1:
707                         seq_printf(seq, "type[C1] ");
708                         break;
709                 case ACPI_STATE_C2:
710                         seq_printf(seq, "type[C2] ");
711                         break;
712                 case ACPI_STATE_C3:
713                         seq_printf(seq, "type[C3] ");
714                         break;
715                 default:
716                         seq_printf(seq, "type[--] ");
717                         break;
718                 }
719
720                 if (pr->power.states[i].promotion.state)
721                         seq_printf(seq, "promotion[C%zd] ",
722                                    (pr->power.states[i].promotion.state -
723                                     pr->power.states));
724                 else
725                         seq_puts(seq, "promotion[--] ");
726
727                 if (pr->power.states[i].demotion.state)
728                         seq_printf(seq, "demotion[C%zd] ",
729                                    (pr->power.states[i].demotion.state -
730                                     pr->power.states));
731                 else
732                         seq_puts(seq, "demotion[--] ");
733
734                 seq_printf(seq, "latency[%03d] usage[%08d] duration[%020llu]\n",
735                            pr->power.states[i].latency,
736                            pr->power.states[i].usage,
737                            (unsigned long long)pr->power.states[i].time);
738         }
739
740       end:
741         return 0;
742 }
743
744 static int acpi_processor_power_open_fs(struct inode *inode, struct file *file)
745 {
746         return single_open(file, acpi_processor_power_seq_show,
747                            PDE(inode)->data);
748 }
749
750 static const struct file_operations acpi_processor_power_fops = {
751         .owner = THIS_MODULE,
752         .open = acpi_processor_power_open_fs,
753         .read = seq_read,
754         .llseek = seq_lseek,
755         .release = single_release,
756 };
757
758
759 /**
760  * acpi_idle_bm_check - checks if bus master activity was detected
761  */
762 static int acpi_idle_bm_check(void)
763 {
764         u32 bm_status = 0;
765
766         acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
767         if (bm_status)
768                 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
769         /*
770          * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
771          * the true state of bus mastering activity; forcing us to
772          * manually check the BMIDEA bit of each IDE channel.
773          */
774         else if (errata.piix4.bmisx) {
775                 if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
776                     || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
777                         bm_status = 1;
778         }
779         return bm_status;
780 }
781
782 /**
783  * acpi_idle_do_entry - a helper function that does C2 and C3 type entry
784  * @cx: cstate data
785  *
786  * Caller disables interrupt before call and enables interrupt after return.
787  */
788 static inline void acpi_idle_do_entry(struct acpi_processor_cx *cx)
789 {
790         /* Don't trace irqs off for idle */
791         stop_critical_timings();
792         if (cx->entry_method == ACPI_CSTATE_FFH) {
793                 /* Call into architectural FFH based C-state */
794                 acpi_processor_ffh_cstate_enter(cx);
795         } else if (cx->entry_method == ACPI_CSTATE_HALT) {
796                 acpi_safe_halt();
797         } else {
798                 int unused;
799                 /* IO port based C-state */
800                 inb(cx->address);
801                 /* Dummy wait op - must do something useless after P_LVL2 read
802                    because chipsets cannot guarantee that STPCLK# signal
803                    gets asserted in time to freeze execution properly. */
804                 unused = inl(acpi_gbl_FADT.xpm_timer_block.address);
805         }
806         start_critical_timings();
807 }
808
809 /**
810  * acpi_idle_enter_c1 - enters an ACPI C1 state-type
811  * @dev: the target CPU
812  * @state: the state data
813  *
814  * This is equivalent to the HALT instruction.
815  */
816 static int acpi_idle_enter_c1(struct cpuidle_device *dev,
817                               struct cpuidle_state *state)
818 {
819         ktime_t  kt1, kt2;
820         s64 idle_time;
821         struct acpi_processor *pr;
822         struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
823
824         pr = __get_cpu_var(processors);
825
826         if (unlikely(!pr))
827                 return 0;
828
829         local_irq_disable();
830
831         /* Do not access any ACPI IO ports in suspend path */
832         if (acpi_idle_suspend) {
833                 acpi_safe_halt();
834                 local_irq_enable();
835                 return 0;
836         }
837
838         kt1 = ktime_get_real();
839         acpi_idle_do_entry(cx);
840         kt2 = ktime_get_real();
841         idle_time =  ktime_to_us(ktime_sub(kt2, kt1));
842
843         local_irq_enable();
844         cx->usage++;
845
846         return idle_time;
847 }
848
849 /**
850  * acpi_idle_enter_simple - enters an ACPI state without BM handling
851  * @dev: the target CPU
852  * @state: the state data
853  */
854 static int acpi_idle_enter_simple(struct cpuidle_device *dev,
855                                   struct cpuidle_state *state)
856 {
857         struct acpi_processor *pr;
858         struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
859         ktime_t  kt1, kt2;
860         s64 idle_time;
861         s64 sleep_ticks = 0;
862
863         pr = __get_cpu_var(processors);
864
865         if (unlikely(!pr))
866                 return 0;
867
868         if (acpi_idle_suspend)
869                 return(acpi_idle_enter_c1(dev, state));
870
871         local_irq_disable();
872         current_thread_info()->status &= ~TS_POLLING;
873         /*
874          * TS_POLLING-cleared state must be visible before we test
875          * NEED_RESCHED:
876          */
877         smp_mb();
878
879         if (unlikely(need_resched())) {
880                 current_thread_info()->status |= TS_POLLING;
881                 local_irq_enable();
882                 return 0;
883         }
884
885         /*
886          * Must be done before busmaster disable as we might need to
887          * access HPET !
888          */
889         acpi_state_timer_broadcast(pr, cx, 1);
890
891         if (cx->type == ACPI_STATE_C3)
892                 ACPI_FLUSH_CPU_CACHE();
893
894         kt1 = ktime_get_real();
895         /* Tell the scheduler that we are going deep-idle: */
896         sched_clock_idle_sleep_event();
897         acpi_idle_do_entry(cx);
898         kt2 = ktime_get_real();
899         idle_time =  ktime_to_us(ktime_sub(kt2, kt1));
900
901         sleep_ticks = us_to_pm_timer_ticks(idle_time);
902
903         /* Tell the scheduler how much we idled: */
904         sched_clock_idle_wakeup_event(sleep_ticks*PM_TIMER_TICK_NS);
905
906         local_irq_enable();
907         current_thread_info()->status |= TS_POLLING;
908
909         cx->usage++;
910
911         acpi_state_timer_broadcast(pr, cx, 0);
912         cx->time += sleep_ticks;
913         return idle_time;
914 }
915
916 static int c3_cpu_count;
917 static DEFINE_SPINLOCK(c3_lock);
918
919 /**
920  * acpi_idle_enter_bm - enters C3 with proper BM handling
921  * @dev: the target CPU
922  * @state: the state data
923  *
924  * If BM is detected, the deepest non-C3 idle state is entered instead.
925  */
926 static int acpi_idle_enter_bm(struct cpuidle_device *dev,
927                               struct cpuidle_state *state)
928 {
929         struct acpi_processor *pr;
930         struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
931         ktime_t  kt1, kt2;
932         s64 idle_time;
933         s64 sleep_ticks = 0;
934
935
936         pr = __get_cpu_var(processors);
937
938         if (unlikely(!pr))
939                 return 0;
940
941         if (acpi_idle_suspend)
942                 return(acpi_idle_enter_c1(dev, state));
943
944         if (acpi_idle_bm_check()) {
945                 if (dev->safe_state) {
946                         dev->last_state = dev->safe_state;
947                         return dev->safe_state->enter(dev, dev->safe_state);
948                 } else {
949                         local_irq_disable();
950                         acpi_safe_halt();
951                         local_irq_enable();
952                         return 0;
953                 }
954         }
955
956         local_irq_disable();
957         current_thread_info()->status &= ~TS_POLLING;
958         /*
959          * TS_POLLING-cleared state must be visible before we test
960          * NEED_RESCHED:
961          */
962         smp_mb();
963
964         if (unlikely(need_resched())) {
965                 current_thread_info()->status |= TS_POLLING;
966                 local_irq_enable();
967                 return 0;
968         }
969
970         acpi_unlazy_tlb(smp_processor_id());
971
972         /* Tell the scheduler that we are going deep-idle: */
973         sched_clock_idle_sleep_event();
974         /*
975          * Must be done before busmaster disable as we might need to
976          * access HPET !
977          */
978         acpi_state_timer_broadcast(pr, cx, 1);
979
980         kt1 = ktime_get_real();
981         /*
982          * disable bus master
983          * bm_check implies we need ARB_DIS
984          * !bm_check implies we need cache flush
985          * bm_control implies whether we can do ARB_DIS
986          *
987          * That leaves a case where bm_check is set and bm_control is
988          * not set. In that case we cannot do much, we enter C3
989          * without doing anything.
990          */
991         if (pr->flags.bm_check && pr->flags.bm_control) {
992                 spin_lock(&c3_lock);
993                 c3_cpu_count++;
994                 /* Disable bus master arbitration when all CPUs are in C3 */
995                 if (c3_cpu_count == num_online_cpus())
996                         acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
997                 spin_unlock(&c3_lock);
998         } else if (!pr->flags.bm_check) {
999                 ACPI_FLUSH_CPU_CACHE();
1000         }
1001
1002         acpi_idle_do_entry(cx);
1003
1004         /* Re-enable bus master arbitration */
1005         if (pr->flags.bm_check && pr->flags.bm_control) {
1006                 spin_lock(&c3_lock);
1007                 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
1008                 c3_cpu_count--;
1009                 spin_unlock(&c3_lock);
1010         }
1011         kt2 = ktime_get_real();
1012         idle_time =  ktime_to_us(ktime_sub(kt2, kt1));
1013
1014         sleep_ticks = us_to_pm_timer_ticks(idle_time);
1015         /* Tell the scheduler how much we idled: */
1016         sched_clock_idle_wakeup_event(sleep_ticks*PM_TIMER_TICK_NS);
1017
1018         local_irq_enable();
1019         current_thread_info()->status |= TS_POLLING;
1020
1021         cx->usage++;
1022
1023         acpi_state_timer_broadcast(pr, cx, 0);
1024         cx->time += sleep_ticks;
1025         return idle_time;
1026 }
1027
1028 struct cpuidle_driver acpi_idle_driver = {
1029         .name =         "acpi_idle",
1030         .owner =        THIS_MODULE,
1031 };
1032
1033 /**
1034  * acpi_processor_setup_cpuidle - prepares and configures CPUIDLE
1035  * @pr: the ACPI processor
1036  */
1037 static int acpi_processor_setup_cpuidle(struct acpi_processor *pr)
1038 {
1039         int i, count = CPUIDLE_DRIVER_STATE_START;
1040         struct acpi_processor_cx *cx;
1041         struct cpuidle_state *state;
1042         struct cpuidle_device *dev = &pr->power.dev;
1043
1044         if (!pr->flags.power_setup_done)
1045                 return -EINVAL;
1046
1047         if (pr->flags.power == 0) {
1048                 return -EINVAL;
1049         }
1050
1051         dev->cpu = pr->id;
1052         for (i = 0; i < CPUIDLE_STATE_MAX; i++) {
1053                 dev->states[i].name[0] = '\0';
1054                 dev->states[i].desc[0] = '\0';
1055         }
1056
1057         if (max_cstate == 0)
1058                 max_cstate = 1;
1059
1060         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
1061                 cx = &pr->power.states[i];
1062                 state = &dev->states[count];
1063
1064                 if (!cx->valid)
1065                         continue;
1066
1067 #ifdef CONFIG_HOTPLUG_CPU
1068                 if ((cx->type != ACPI_STATE_C1) && (num_online_cpus() > 1) &&
1069                     !pr->flags.has_cst &&
1070                     !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
1071                         continue;
1072 #endif
1073                 cpuidle_set_statedata(state, cx);
1074
1075                 snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
1076                 strncpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
1077                 state->exit_latency = cx->latency;
1078                 state->target_residency = cx->latency * latency_factor;
1079                 state->power_usage = cx->power;
1080
1081                 state->flags = 0;
1082                 switch (cx->type) {
1083                         case ACPI_STATE_C1:
1084                         state->flags |= CPUIDLE_FLAG_SHALLOW;
1085                         if (cx->entry_method == ACPI_CSTATE_FFH)
1086                                 state->flags |= CPUIDLE_FLAG_TIME_VALID;
1087
1088                         state->enter = acpi_idle_enter_c1;
1089                         dev->safe_state = state;
1090                         break;
1091
1092                         case ACPI_STATE_C2:
1093                         state->flags |= CPUIDLE_FLAG_BALANCED;
1094                         state->flags |= CPUIDLE_FLAG_TIME_VALID;
1095                         state->enter = acpi_idle_enter_simple;
1096                         dev->safe_state = state;
1097                         break;
1098
1099                         case ACPI_STATE_C3:
1100                         state->flags |= CPUIDLE_FLAG_DEEP;
1101                         state->flags |= CPUIDLE_FLAG_TIME_VALID;
1102                         state->flags |= CPUIDLE_FLAG_CHECK_BM;
1103                         state->enter = pr->flags.bm_check ?
1104                                         acpi_idle_enter_bm :
1105                                         acpi_idle_enter_simple;
1106                         break;
1107                 }
1108
1109                 count++;
1110                 if (count == CPUIDLE_STATE_MAX)
1111                         break;
1112         }
1113
1114         dev->state_count = count;
1115
1116         if (!count)
1117                 return -EINVAL;
1118
1119         return 0;
1120 }
1121
1122 int acpi_processor_cst_has_changed(struct acpi_processor *pr)
1123 {
1124         int ret = 0;
1125
1126         if (boot_option_idle_override)
1127                 return 0;
1128
1129         if (!pr)
1130                 return -EINVAL;
1131
1132         if (nocst) {
1133                 return -ENODEV;
1134         }
1135
1136         if (!pr->flags.power_setup_done)
1137                 return -ENODEV;
1138
1139         cpuidle_pause_and_lock();
1140         cpuidle_disable_device(&pr->power.dev);
1141         acpi_processor_get_power_info(pr);
1142         if (pr->flags.power) {
1143                 acpi_processor_setup_cpuidle(pr);
1144                 ret = cpuidle_enable_device(&pr->power.dev);
1145         }
1146         cpuidle_resume_and_unlock();
1147
1148         return ret;
1149 }
1150
1151 int __cpuinit acpi_processor_power_init(struct acpi_processor *pr,
1152                               struct acpi_device *device)
1153 {
1154         acpi_status status = 0;
1155         static int first_run;
1156         struct proc_dir_entry *entry = NULL;
1157         unsigned int i;
1158
1159         if (boot_option_idle_override)
1160                 return 0;
1161
1162         if (!first_run) {
1163                 if (idle_halt) {
1164                         /*
1165                          * When the boot option of "idle=halt" is added, halt
1166                          * is used for CPU IDLE.
1167                          * In such case C2/C3 is meaningless. So the max_cstate
1168                          * is set to one.
1169                          */
1170                         max_cstate = 1;
1171                 }
1172                 dmi_check_system(processor_power_dmi_table);
1173                 max_cstate = acpi_processor_cstate_check(max_cstate);
1174                 if (max_cstate < ACPI_C_STATES_MAX)
1175                         printk(KERN_NOTICE
1176                                "ACPI: processor limited to max C-state %d\n",
1177                                max_cstate);
1178                 first_run++;
1179         }
1180
1181         if (!pr)
1182                 return -EINVAL;
1183
1184         if (acpi_gbl_FADT.cst_control && !nocst) {
1185                 status =
1186                     acpi_os_write_port(acpi_gbl_FADT.smi_command, acpi_gbl_FADT.cst_control, 8);
1187                 if (ACPI_FAILURE(status)) {
1188                         ACPI_EXCEPTION((AE_INFO, status,
1189                                         "Notifying BIOS of _CST ability failed"));
1190                 }
1191         }
1192
1193         acpi_processor_get_power_info(pr);
1194         pr->flags.power_setup_done = 1;
1195
1196         /*
1197          * Install the idle handler if processor power management is supported.
1198          * Note that we use previously set idle handler will be used on
1199          * platforms that only support C1.
1200          */
1201         if (pr->flags.power) {
1202                 acpi_processor_setup_cpuidle(pr);
1203                 if (cpuidle_register_device(&pr->power.dev))
1204                         return -EIO;
1205
1206                 printk(KERN_INFO PREFIX "CPU%d (power states:", pr->id);
1207                 for (i = 1; i <= pr->power.count; i++)
1208                         if (pr->power.states[i].valid)
1209                                 printk(" C%d[C%d]", i,
1210                                        pr->power.states[i].type);
1211                 printk(")\n");
1212         }
1213
1214         /* 'power' [R] */
1215         entry = proc_create_data(ACPI_PROCESSOR_FILE_POWER,
1216                                  S_IRUGO, acpi_device_dir(device),
1217                                  &acpi_processor_power_fops,
1218                                  acpi_driver_data(device));
1219         if (!entry)
1220                 return -EIO;
1221         return 0;
1222 }
1223
1224 int acpi_processor_power_exit(struct acpi_processor *pr,
1225                               struct acpi_device *device)
1226 {
1227         if (boot_option_idle_override)
1228                 return 0;
1229
1230         cpuidle_unregister_device(&pr->power.dev);
1231         pr->flags.power_setup_done = 0;
1232
1233         if (acpi_device_dir(device))
1234                 remove_proc_entry(ACPI_PROCESSOR_FILE_POWER,
1235                                   acpi_device_dir(device));
1236
1237         return 0;
1238 }