Merge tag 'for-linus-v3.6-rc1' of git://oss.sgi.com/xfs/xfs
[platform/adaptation/renesas_rcar/renesas_kernel.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/slab.h>
36 #include <linux/acpi.h>
37 #include <linux/dmi.h>
38 #include <linux/moduleparam.h>
39 #include <linux/sched.h>        /* need_resched() */
40 #include <linux/pm_qos.h>
41 #include <linux/clockchips.h>
42 #include <linux/cpuidle.h>
43 #include <linux/irqflags.h>
44
45 /*
46  * Include the apic definitions for x86 to have the APIC timer related defines
47  * available also for UP (on SMP it gets magically included via linux/smp.h).
48  * asm/acpi.h is not an option, as it would require more include magic. Also
49  * creating an empty asm-ia64/apic.h would just trade pest vs. cholera.
50  */
51 #ifdef CONFIG_X86
52 #include <asm/apic.h>
53 #endif
54
55 #include <asm/io.h>
56 #include <asm/uaccess.h>
57
58 #include <acpi/acpi_bus.h>
59 #include <acpi/processor.h>
60 #include <asm/processor.h>
61
62 #define PREFIX "ACPI: "
63
64 #define ACPI_PROCESSOR_CLASS            "processor"
65 #define _COMPONENT              ACPI_PROCESSOR_COMPONENT
66 ACPI_MODULE_NAME("processor_idle");
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 static int bm_check_disable __read_mostly;
77 module_param(bm_check_disable, uint, 0000);
78
79 static unsigned int latency_factor __read_mostly = 2;
80 module_param(latency_factor, uint, 0644);
81
82 static int disabled_by_idle_boot_param(void)
83 {
84         return boot_option_idle_override == IDLE_POLL ||
85                 boot_option_idle_override == IDLE_FORCE_MWAIT ||
86                 boot_option_idle_override == IDLE_HALT;
87 }
88
89 /*
90  * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
91  * For now disable this. Probably a bug somewhere else.
92  *
93  * To skip this limit, boot/load with a large max_cstate limit.
94  */
95 static int set_max_cstate(const struct dmi_system_id *id)
96 {
97         if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
98                 return 0;
99
100         printk(KERN_NOTICE PREFIX "%s detected - limiting to C%ld max_cstate."
101                " Override with \"processor.max_cstate=%d\"\n", id->ident,
102                (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
103
104         max_cstate = (long)id->driver_data;
105
106         return 0;
107 }
108
109 /* Actually this shouldn't be __cpuinitdata, would be better to fix the
110    callers to only run once -AK */
111 static struct dmi_system_id __cpuinitdata processor_power_dmi_table[] = {
112         { set_max_cstate, "Clevo 5600D", {
113           DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
114           DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")},
115          (void *)2},
116         { set_max_cstate, "Pavilion zv5000", {
117           DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
118           DMI_MATCH(DMI_PRODUCT_NAME,"Pavilion zv5000 (DS502A#ABA)")},
119          (void *)1},
120         { set_max_cstate, "Asus L8400B", {
121           DMI_MATCH(DMI_SYS_VENDOR, "ASUSTeK Computer Inc."),
122           DMI_MATCH(DMI_PRODUCT_NAME,"L8400B series Notebook PC")},
123          (void *)1},
124         {},
125 };
126
127
128 /*
129  * Callers should disable interrupts before the call and enable
130  * interrupts after return.
131  */
132 static void acpi_safe_halt(void)
133 {
134         current_thread_info()->status &= ~TS_POLLING;
135         /*
136          * TS_POLLING-cleared state must be visible before we
137          * test NEED_RESCHED:
138          */
139         smp_mb();
140         if (!need_resched()) {
141                 safe_halt();
142                 local_irq_disable();
143         }
144         current_thread_info()->status |= TS_POLLING;
145 }
146
147 #ifdef ARCH_APICTIMER_STOPS_ON_C3
148
149 /*
150  * Some BIOS implementations switch to C3 in the published C2 state.
151  * This seems to be a common problem on AMD boxen, but other vendors
152  * are affected too. We pick the most conservative approach: we assume
153  * that the local APIC stops in both C2 and C3.
154  */
155 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
156                                    struct acpi_processor_cx *cx)
157 {
158         struct acpi_processor_power *pwr = &pr->power;
159         u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2;
160
161         if (cpu_has(&cpu_data(pr->id), X86_FEATURE_ARAT))
162                 return;
163
164         if (amd_e400_c1e_detected)
165                 type = ACPI_STATE_C1;
166
167         /*
168          * Check, if one of the previous states already marked the lapic
169          * unstable
170          */
171         if (pwr->timer_broadcast_on_state < state)
172                 return;
173
174         if (cx->type >= type)
175                 pr->power.timer_broadcast_on_state = state;
176 }
177
178 static void __lapic_timer_propagate_broadcast(void *arg)
179 {
180         struct acpi_processor *pr = (struct acpi_processor *) arg;
181         unsigned long reason;
182
183         reason = pr->power.timer_broadcast_on_state < INT_MAX ?
184                 CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
185
186         clockevents_notify(reason, &pr->id);
187 }
188
189 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr)
190 {
191         smp_call_function_single(pr->id, __lapic_timer_propagate_broadcast,
192                                  (void *)pr, 1);
193 }
194
195 /* Power(C) State timer broadcast control */
196 static void lapic_timer_state_broadcast(struct acpi_processor *pr,
197                                        struct acpi_processor_cx *cx,
198                                        int broadcast)
199 {
200         int state = cx - pr->power.states;
201
202         if (state >= pr->power.timer_broadcast_on_state) {
203                 unsigned long reason;
204
205                 reason = broadcast ?  CLOCK_EVT_NOTIFY_BROADCAST_ENTER :
206                         CLOCK_EVT_NOTIFY_BROADCAST_EXIT;
207                 clockevents_notify(reason, &pr->id);
208         }
209 }
210
211 #else
212
213 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
214                                    struct acpi_processor_cx *cstate) { }
215 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) { }
216 static void lapic_timer_state_broadcast(struct acpi_processor *pr,
217                                        struct acpi_processor_cx *cx,
218                                        int broadcast)
219 {
220 }
221
222 #endif
223
224 static u32 saved_bm_rld;
225
226 static void acpi_idle_bm_rld_save(void)
227 {
228         acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &saved_bm_rld);
229 }
230 static void acpi_idle_bm_rld_restore(void)
231 {
232         u32 resumed_bm_rld;
233
234         acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &resumed_bm_rld);
235
236         if (resumed_bm_rld != saved_bm_rld)
237                 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, saved_bm_rld);
238 }
239
240 int acpi_processor_suspend(struct device *dev)
241 {
242         acpi_idle_bm_rld_save();
243         return 0;
244 }
245
246 int acpi_processor_resume(struct device *dev)
247 {
248         acpi_idle_bm_rld_restore();
249         return 0;
250 }
251
252 #if defined(CONFIG_X86)
253 static void tsc_check_state(int state)
254 {
255         switch (boot_cpu_data.x86_vendor) {
256         case X86_VENDOR_AMD:
257         case X86_VENDOR_INTEL:
258                 /*
259                  * AMD Fam10h TSC will tick in all
260                  * C/P/S0/S1 states when this bit is set.
261                  */
262                 if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
263                         return;
264
265                 /*FALL THROUGH*/
266         default:
267                 /* TSC could halt in idle, so notify users */
268                 if (state > ACPI_STATE_C1)
269                         mark_tsc_unstable("TSC halts in idle");
270         }
271 }
272 #else
273 static void tsc_check_state(int state) { return; }
274 #endif
275
276 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
277 {
278
279         if (!pr)
280                 return -EINVAL;
281
282         if (!pr->pblk)
283                 return -ENODEV;
284
285         /* if info is obtained from pblk/fadt, type equals state */
286         pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
287         pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
288
289 #ifndef CONFIG_HOTPLUG_CPU
290         /*
291          * Check for P_LVL2_UP flag before entering C2 and above on
292          * an SMP system.
293          */
294         if ((num_online_cpus() > 1) &&
295             !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
296                 return -ENODEV;
297 #endif
298
299         /* determine C2 and C3 address from pblk */
300         pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
301         pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
302
303         /* determine latencies from FADT */
304         pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.c2_latency;
305         pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.c3_latency;
306
307         /*
308          * FADT specified C2 latency must be less than or equal to
309          * 100 microseconds.
310          */
311         if (acpi_gbl_FADT.c2_latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
312                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
313                         "C2 latency too large [%d]\n", acpi_gbl_FADT.c2_latency));
314                 /* invalidate C2 */
315                 pr->power.states[ACPI_STATE_C2].address = 0;
316         }
317
318         /*
319          * FADT supplied C3 latency must be less than or equal to
320          * 1000 microseconds.
321          */
322         if (acpi_gbl_FADT.c3_latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
323                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
324                         "C3 latency too large [%d]\n", acpi_gbl_FADT.c3_latency));
325                 /* invalidate C3 */
326                 pr->power.states[ACPI_STATE_C3].address = 0;
327         }
328
329         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
330                           "lvl2[0x%08x] lvl3[0x%08x]\n",
331                           pr->power.states[ACPI_STATE_C2].address,
332                           pr->power.states[ACPI_STATE_C3].address));
333
334         return 0;
335 }
336
337 static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
338 {
339         if (!pr->power.states[ACPI_STATE_C1].valid) {
340                 /* set the first C-State to C1 */
341                 /* all processors need to support C1 */
342                 pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
343                 pr->power.states[ACPI_STATE_C1].valid = 1;
344                 pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
345         }
346         /* the C0 state only exists as a filler in our array */
347         pr->power.states[ACPI_STATE_C0].valid = 1;
348         return 0;
349 }
350
351 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
352 {
353         acpi_status status = 0;
354         u64 count;
355         int current_count;
356         int i;
357         struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
358         union acpi_object *cst;
359
360
361         if (nocst)
362                 return -ENODEV;
363
364         current_count = 0;
365
366         status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
367         if (ACPI_FAILURE(status)) {
368                 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
369                 return -ENODEV;
370         }
371
372         cst = buffer.pointer;
373
374         /* There must be at least 2 elements */
375         if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
376                 printk(KERN_ERR PREFIX "not enough elements in _CST\n");
377                 status = -EFAULT;
378                 goto end;
379         }
380
381         count = cst->package.elements[0].integer.value;
382
383         /* Validate number of power states. */
384         if (count < 1 || count != cst->package.count - 1) {
385                 printk(KERN_ERR PREFIX "count given by _CST is not valid\n");
386                 status = -EFAULT;
387                 goto end;
388         }
389
390         /* Tell driver that at least _CST is supported. */
391         pr->flags.has_cst = 1;
392
393         for (i = 1; i <= count; i++) {
394                 union acpi_object *element;
395                 union acpi_object *obj;
396                 struct acpi_power_register *reg;
397                 struct acpi_processor_cx cx;
398
399                 memset(&cx, 0, sizeof(cx));
400
401                 element = &(cst->package.elements[i]);
402                 if (element->type != ACPI_TYPE_PACKAGE)
403                         continue;
404
405                 if (element->package.count != 4)
406                         continue;
407
408                 obj = &(element->package.elements[0]);
409
410                 if (obj->type != ACPI_TYPE_BUFFER)
411                         continue;
412
413                 reg = (struct acpi_power_register *)obj->buffer.pointer;
414
415                 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
416                     (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
417                         continue;
418
419                 /* There should be an easy way to extract an integer... */
420                 obj = &(element->package.elements[1]);
421                 if (obj->type != ACPI_TYPE_INTEGER)
422                         continue;
423
424                 cx.type = obj->integer.value;
425                 /*
426                  * Some buggy BIOSes won't list C1 in _CST -
427                  * Let acpi_processor_get_power_info_default() handle them later
428                  */
429                 if (i == 1 && cx.type != ACPI_STATE_C1)
430                         current_count++;
431
432                 cx.address = reg->address;
433                 cx.index = current_count + 1;
434
435                 cx.entry_method = ACPI_CSTATE_SYSTEMIO;
436                 if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) {
437                         if (acpi_processor_ffh_cstate_probe
438                                         (pr->id, &cx, reg) == 0) {
439                                 cx.entry_method = ACPI_CSTATE_FFH;
440                         } else if (cx.type == ACPI_STATE_C1) {
441                                 /*
442                                  * C1 is a special case where FIXED_HARDWARE
443                                  * can be handled in non-MWAIT way as well.
444                                  * In that case, save this _CST entry info.
445                                  * Otherwise, ignore this info and continue.
446                                  */
447                                 cx.entry_method = ACPI_CSTATE_HALT;
448                                 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
449                         } else {
450                                 continue;
451                         }
452                         if (cx.type == ACPI_STATE_C1 &&
453                             (boot_option_idle_override == IDLE_NOMWAIT)) {
454                                 /*
455                                  * In most cases the C1 space_id obtained from
456                                  * _CST object is FIXED_HARDWARE access mode.
457                                  * But when the option of idle=halt is added,
458                                  * the entry_method type should be changed from
459                                  * CSTATE_FFH to CSTATE_HALT.
460                                  * When the option of idle=nomwait is added,
461                                  * the C1 entry_method type should be
462                                  * CSTATE_HALT.
463                                  */
464                                 cx.entry_method = ACPI_CSTATE_HALT;
465                                 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
466                         }
467                 } else {
468                         snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI IOPORT 0x%x",
469                                  cx.address);
470                 }
471
472                 if (cx.type == ACPI_STATE_C1) {
473                         cx.valid = 1;
474                 }
475
476                 obj = &(element->package.elements[2]);
477                 if (obj->type != ACPI_TYPE_INTEGER)
478                         continue;
479
480                 cx.latency = obj->integer.value;
481
482                 obj = &(element->package.elements[3]);
483                 if (obj->type != ACPI_TYPE_INTEGER)
484                         continue;
485
486                 cx.power = obj->integer.value;
487
488                 current_count++;
489                 memcpy(&(pr->power.states[current_count]), &cx, sizeof(cx));
490
491                 /*
492                  * We support total ACPI_PROCESSOR_MAX_POWER - 1
493                  * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
494                  */
495                 if (current_count >= (ACPI_PROCESSOR_MAX_POWER - 1)) {
496                         printk(KERN_WARNING
497                                "Limiting number of power states to max (%d)\n",
498                                ACPI_PROCESSOR_MAX_POWER);
499                         printk(KERN_WARNING
500                                "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
501                         break;
502                 }
503         }
504
505         ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n",
506                           current_count));
507
508         /* Validate number of power states discovered */
509         if (current_count < 2)
510                 status = -EFAULT;
511
512       end:
513         kfree(buffer.pointer);
514
515         return status;
516 }
517
518 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
519                                            struct acpi_processor_cx *cx)
520 {
521         static int bm_check_flag = -1;
522         static int bm_control_flag = -1;
523
524
525         if (!cx->address)
526                 return;
527
528         /*
529          * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
530          * DMA transfers are used by any ISA device to avoid livelock.
531          * Note that we could disable Type-F DMA (as recommended by
532          * the erratum), but this is known to disrupt certain ISA
533          * devices thus we take the conservative approach.
534          */
535         else if (errata.piix4.fdma) {
536                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
537                                   "C3 not supported on PIIX4 with Type-F DMA\n"));
538                 return;
539         }
540
541         /* All the logic here assumes flags.bm_check is same across all CPUs */
542         if (bm_check_flag == -1) {
543                 /* Determine whether bm_check is needed based on CPU  */
544                 acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
545                 bm_check_flag = pr->flags.bm_check;
546                 bm_control_flag = pr->flags.bm_control;
547         } else {
548                 pr->flags.bm_check = bm_check_flag;
549                 pr->flags.bm_control = bm_control_flag;
550         }
551
552         if (pr->flags.bm_check) {
553                 if (!pr->flags.bm_control) {
554                         if (pr->flags.has_cst != 1) {
555                                 /* bus mastering control is necessary */
556                                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
557                                         "C3 support requires BM control\n"));
558                                 return;
559                         } else {
560                                 /* Here we enter C3 without bus mastering */
561                                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
562                                         "C3 support without BM control\n"));
563                         }
564                 }
565         } else {
566                 /*
567                  * WBINVD should be set in fadt, for C3 state to be
568                  * supported on when bm_check is not required.
569                  */
570                 if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
571                         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
572                                           "Cache invalidation should work properly"
573                                           " for C3 to be enabled on SMP systems\n"));
574                         return;
575                 }
576         }
577
578         /*
579          * Otherwise we've met all of our C3 requirements.
580          * Normalize the C3 latency to expidite policy.  Enable
581          * checking of bus mastering status (bm_check) so we can
582          * use this in our C3 policy
583          */
584         cx->valid = 1;
585
586         /*
587          * On older chipsets, BM_RLD needs to be set
588          * in order for Bus Master activity to wake the
589          * system from C3.  Newer chipsets handle DMA
590          * during C3 automatically and BM_RLD is a NOP.
591          * In either case, the proper way to
592          * handle BM_RLD is to set it and leave it set.
593          */
594         acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
595
596         return;
597 }
598
599 static int acpi_processor_power_verify(struct acpi_processor *pr)
600 {
601         unsigned int i;
602         unsigned int working = 0;
603
604         pr->power.timer_broadcast_on_state = INT_MAX;
605
606         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
607                 struct acpi_processor_cx *cx = &pr->power.states[i];
608
609                 switch (cx->type) {
610                 case ACPI_STATE_C1:
611                         cx->valid = 1;
612                         break;
613
614                 case ACPI_STATE_C2:
615                         if (!cx->address)
616                                 break;
617                         cx->valid = 1; 
618                         break;
619
620                 case ACPI_STATE_C3:
621                         acpi_processor_power_verify_c3(pr, cx);
622                         break;
623                 }
624                 if (!cx->valid)
625                         continue;
626
627                 lapic_timer_check_state(i, pr, cx);
628                 tsc_check_state(cx->type);
629                 working++;
630         }
631
632         lapic_timer_propagate_broadcast(pr);
633
634         return (working);
635 }
636
637 static int acpi_processor_get_power_info(struct acpi_processor *pr)
638 {
639         unsigned int i;
640         int result;
641
642
643         /* NOTE: the idle thread may not be running while calling
644          * this function */
645
646         /* Zero initialize all the C-states info. */
647         memset(pr->power.states, 0, sizeof(pr->power.states));
648
649         result = acpi_processor_get_power_info_cst(pr);
650         if (result == -ENODEV)
651                 result = acpi_processor_get_power_info_fadt(pr);
652
653         if (result)
654                 return result;
655
656         acpi_processor_get_power_info_default(pr);
657
658         pr->power.count = acpi_processor_power_verify(pr);
659
660         /*
661          * if one state of type C2 or C3 is available, mark this
662          * CPU as being "idle manageable"
663          */
664         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
665                 if (pr->power.states[i].valid) {
666                         pr->power.count = i;
667                         if (pr->power.states[i].type >= ACPI_STATE_C2)
668                                 pr->flags.power = 1;
669                 }
670         }
671
672         return 0;
673 }
674
675 /**
676  * acpi_idle_bm_check - checks if bus master activity was detected
677  */
678 static int acpi_idle_bm_check(void)
679 {
680         u32 bm_status = 0;
681
682         if (bm_check_disable)
683                 return 0;
684
685         acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
686         if (bm_status)
687                 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
688         /*
689          * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
690          * the true state of bus mastering activity; forcing us to
691          * manually check the BMIDEA bit of each IDE channel.
692          */
693         else if (errata.piix4.bmisx) {
694                 if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
695                     || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
696                         bm_status = 1;
697         }
698         return bm_status;
699 }
700
701 /**
702  * acpi_idle_do_entry - a helper function that does C2 and C3 type entry
703  * @cx: cstate data
704  *
705  * Caller disables interrupt before call and enables interrupt after return.
706  */
707 static inline void acpi_idle_do_entry(struct acpi_processor_cx *cx)
708 {
709         /* Don't trace irqs off for idle */
710         stop_critical_timings();
711         if (cx->entry_method == ACPI_CSTATE_FFH) {
712                 /* Call into architectural FFH based C-state */
713                 acpi_processor_ffh_cstate_enter(cx);
714         } else if (cx->entry_method == ACPI_CSTATE_HALT) {
715                 acpi_safe_halt();
716         } else {
717                 /* IO port based C-state */
718                 inb(cx->address);
719                 /* Dummy wait op - must do something useless after P_LVL2 read
720                    because chipsets cannot guarantee that STPCLK# signal
721                    gets asserted in time to freeze execution properly. */
722                 inl(acpi_gbl_FADT.xpm_timer_block.address);
723         }
724         start_critical_timings();
725 }
726
727 /**
728  * acpi_idle_enter_c1 - enters an ACPI C1 state-type
729  * @dev: the target CPU
730  * @drv: cpuidle driver containing cpuidle state info
731  * @index: index of target state
732  *
733  * This is equivalent to the HALT instruction.
734  */
735 static int acpi_idle_enter_c1(struct cpuidle_device *dev,
736                 struct cpuidle_driver *drv, int index)
737 {
738         ktime_t  kt1, kt2;
739         s64 idle_time;
740         struct acpi_processor *pr;
741         struct cpuidle_state_usage *state_usage = &dev->states_usage[index];
742         struct acpi_processor_cx *cx = cpuidle_get_statedata(state_usage);
743
744         pr = __this_cpu_read(processors);
745         dev->last_residency = 0;
746
747         if (unlikely(!pr))
748                 return -EINVAL;
749
750         local_irq_disable();
751
752
753         lapic_timer_state_broadcast(pr, cx, 1);
754         kt1 = ktime_get_real();
755         acpi_idle_do_entry(cx);
756         kt2 = ktime_get_real();
757         idle_time =  ktime_to_us(ktime_sub(kt2, kt1));
758
759         /* Update device last_residency*/
760         dev->last_residency = (int)idle_time;
761
762         local_irq_enable();
763         lapic_timer_state_broadcast(pr, cx, 0);
764
765         return index;
766 }
767
768
769 /**
770  * acpi_idle_play_dead - enters an ACPI state for long-term idle (i.e. off-lining)
771  * @dev: the target CPU
772  * @index: the index of suggested state
773  */
774 static int acpi_idle_play_dead(struct cpuidle_device *dev, int index)
775 {
776         struct cpuidle_state_usage *state_usage = &dev->states_usage[index];
777         struct acpi_processor_cx *cx = cpuidle_get_statedata(state_usage);
778
779         ACPI_FLUSH_CPU_CACHE();
780
781         while (1) {
782
783                 if (cx->entry_method == ACPI_CSTATE_HALT)
784                         safe_halt();
785                 else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) {
786                         inb(cx->address);
787                         /* See comment in acpi_idle_do_entry() */
788                         inl(acpi_gbl_FADT.xpm_timer_block.address);
789                 } else
790                         return -ENODEV;
791         }
792
793         /* Never reached */
794         return 0;
795 }
796
797 /**
798  * acpi_idle_enter_simple - enters an ACPI state without BM handling
799  * @dev: the target CPU
800  * @drv: cpuidle driver with cpuidle state information
801  * @index: the index of suggested state
802  */
803 static int acpi_idle_enter_simple(struct cpuidle_device *dev,
804                 struct cpuidle_driver *drv, int index)
805 {
806         struct acpi_processor *pr;
807         struct cpuidle_state_usage *state_usage = &dev->states_usage[index];
808         struct acpi_processor_cx *cx = cpuidle_get_statedata(state_usage);
809         ktime_t  kt1, kt2;
810         s64 idle_time_ns;
811         s64 idle_time;
812
813         pr = __this_cpu_read(processors);
814         dev->last_residency = 0;
815
816         if (unlikely(!pr))
817                 return -EINVAL;
818
819         local_irq_disable();
820
821
822         if (cx->entry_method != ACPI_CSTATE_FFH) {
823                 current_thread_info()->status &= ~TS_POLLING;
824                 /*
825                  * TS_POLLING-cleared state must be visible before we test
826                  * NEED_RESCHED:
827                  */
828                 smp_mb();
829
830                 if (unlikely(need_resched())) {
831                         current_thread_info()->status |= TS_POLLING;
832                         local_irq_enable();
833                         return -EINVAL;
834                 }
835         }
836
837         /*
838          * Must be done before busmaster disable as we might need to
839          * access HPET !
840          */
841         lapic_timer_state_broadcast(pr, cx, 1);
842
843         if (cx->type == ACPI_STATE_C3)
844                 ACPI_FLUSH_CPU_CACHE();
845
846         kt1 = ktime_get_real();
847         /* Tell the scheduler that we are going deep-idle: */
848         sched_clock_idle_sleep_event();
849         acpi_idle_do_entry(cx);
850         kt2 = ktime_get_real();
851         idle_time_ns = ktime_to_ns(ktime_sub(kt2, kt1));
852         idle_time = idle_time_ns;
853         do_div(idle_time, NSEC_PER_USEC);
854
855         /* Update device last_residency*/
856         dev->last_residency = (int)idle_time;
857
858         /* Tell the scheduler how much we idled: */
859         sched_clock_idle_wakeup_event(idle_time_ns);
860
861         local_irq_enable();
862         if (cx->entry_method != ACPI_CSTATE_FFH)
863                 current_thread_info()->status |= TS_POLLING;
864
865         lapic_timer_state_broadcast(pr, cx, 0);
866         return index;
867 }
868
869 static int c3_cpu_count;
870 static DEFINE_RAW_SPINLOCK(c3_lock);
871
872 /**
873  * acpi_idle_enter_bm - enters C3 with proper BM handling
874  * @dev: the target CPU
875  * @drv: cpuidle driver containing state data
876  * @index: the index of suggested state
877  *
878  * If BM is detected, the deepest non-C3 idle state is entered instead.
879  */
880 static int acpi_idle_enter_bm(struct cpuidle_device *dev,
881                 struct cpuidle_driver *drv, int index)
882 {
883         struct acpi_processor *pr;
884         struct cpuidle_state_usage *state_usage = &dev->states_usage[index];
885         struct acpi_processor_cx *cx = cpuidle_get_statedata(state_usage);
886         ktime_t  kt1, kt2;
887         s64 idle_time_ns;
888         s64 idle_time;
889
890
891         pr = __this_cpu_read(processors);
892         dev->last_residency = 0;
893
894         if (unlikely(!pr))
895                 return -EINVAL;
896
897         if (!cx->bm_sts_skip && acpi_idle_bm_check()) {
898                 if (drv->safe_state_index >= 0) {
899                         return drv->states[drv->safe_state_index].enter(dev,
900                                                 drv, drv->safe_state_index);
901                 } else {
902                         local_irq_disable();
903                         acpi_safe_halt();
904                         local_irq_enable();
905                         return -EBUSY;
906                 }
907         }
908
909         local_irq_disable();
910
911
912         if (cx->entry_method != ACPI_CSTATE_FFH) {
913                 current_thread_info()->status &= ~TS_POLLING;
914                 /*
915                  * TS_POLLING-cleared state must be visible before we test
916                  * NEED_RESCHED:
917                  */
918                 smp_mb();
919
920                 if (unlikely(need_resched())) {
921                         current_thread_info()->status |= TS_POLLING;
922                         local_irq_enable();
923                         return -EINVAL;
924                 }
925         }
926
927         acpi_unlazy_tlb(smp_processor_id());
928
929         /* Tell the scheduler that we are going deep-idle: */
930         sched_clock_idle_sleep_event();
931         /*
932          * Must be done before busmaster disable as we might need to
933          * access HPET !
934          */
935         lapic_timer_state_broadcast(pr, cx, 1);
936
937         kt1 = ktime_get_real();
938         /*
939          * disable bus master
940          * bm_check implies we need ARB_DIS
941          * !bm_check implies we need cache flush
942          * bm_control implies whether we can do ARB_DIS
943          *
944          * That leaves a case where bm_check is set and bm_control is
945          * not set. In that case we cannot do much, we enter C3
946          * without doing anything.
947          */
948         if (pr->flags.bm_check && pr->flags.bm_control) {
949                 raw_spin_lock(&c3_lock);
950                 c3_cpu_count++;
951                 /* Disable bus master arbitration when all CPUs are in C3 */
952                 if (c3_cpu_count == num_online_cpus())
953                         acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
954                 raw_spin_unlock(&c3_lock);
955         } else if (!pr->flags.bm_check) {
956                 ACPI_FLUSH_CPU_CACHE();
957         }
958
959         acpi_idle_do_entry(cx);
960
961         /* Re-enable bus master arbitration */
962         if (pr->flags.bm_check && pr->flags.bm_control) {
963                 raw_spin_lock(&c3_lock);
964                 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
965                 c3_cpu_count--;
966                 raw_spin_unlock(&c3_lock);
967         }
968         kt2 = ktime_get_real();
969         idle_time_ns = ktime_to_ns(ktime_sub(kt2, kt1));
970         idle_time = idle_time_ns;
971         do_div(idle_time, NSEC_PER_USEC);
972
973         /* Update device last_residency*/
974         dev->last_residency = (int)idle_time;
975
976         /* Tell the scheduler how much we idled: */
977         sched_clock_idle_wakeup_event(idle_time_ns);
978
979         local_irq_enable();
980         if (cx->entry_method != ACPI_CSTATE_FFH)
981                 current_thread_info()->status |= TS_POLLING;
982
983         lapic_timer_state_broadcast(pr, cx, 0);
984         return index;
985 }
986
987 struct cpuidle_driver acpi_idle_driver = {
988         .name =         "acpi_idle",
989         .owner =        THIS_MODULE,
990 };
991
992 /**
993  * acpi_processor_setup_cpuidle_cx - prepares and configures CPUIDLE
994  * device i.e. per-cpu data
995  *
996  * @pr: the ACPI processor
997  */
998 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr)
999 {
1000         int i, count = CPUIDLE_DRIVER_STATE_START;
1001         struct acpi_processor_cx *cx;
1002         struct cpuidle_state_usage *state_usage;
1003         struct cpuidle_device *dev = &pr->power.dev;
1004
1005         if (!pr->flags.power_setup_done)
1006                 return -EINVAL;
1007
1008         if (pr->flags.power == 0) {
1009                 return -EINVAL;
1010         }
1011
1012         dev->cpu = pr->id;
1013
1014         if (max_cstate == 0)
1015                 max_cstate = 1;
1016
1017         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
1018                 cx = &pr->power.states[i];
1019                 state_usage = &dev->states_usage[count];
1020
1021                 if (!cx->valid)
1022                         continue;
1023
1024 #ifdef CONFIG_HOTPLUG_CPU
1025                 if ((cx->type != ACPI_STATE_C1) && (num_online_cpus() > 1) &&
1026                     !pr->flags.has_cst &&
1027                     !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
1028                         continue;
1029 #endif
1030
1031                 cpuidle_set_statedata(state_usage, cx);
1032
1033                 count++;
1034                 if (count == CPUIDLE_STATE_MAX)
1035                         break;
1036         }
1037
1038         dev->state_count = count;
1039
1040         if (!count)
1041                 return -EINVAL;
1042
1043         return 0;
1044 }
1045
1046 /**
1047  * acpi_processor_setup_cpuidle states- prepares and configures cpuidle
1048  * global state data i.e. idle routines
1049  *
1050  * @pr: the ACPI processor
1051  */
1052 static int acpi_processor_setup_cpuidle_states(struct acpi_processor *pr)
1053 {
1054         int i, count = CPUIDLE_DRIVER_STATE_START;
1055         struct acpi_processor_cx *cx;
1056         struct cpuidle_state *state;
1057         struct cpuidle_driver *drv = &acpi_idle_driver;
1058
1059         if (!pr->flags.power_setup_done)
1060                 return -EINVAL;
1061
1062         if (pr->flags.power == 0)
1063                 return -EINVAL;
1064
1065         drv->safe_state_index = -1;
1066         for (i = 0; i < CPUIDLE_STATE_MAX; i++) {
1067                 drv->states[i].name[0] = '\0';
1068                 drv->states[i].desc[0] = '\0';
1069         }
1070
1071         if (max_cstate == 0)
1072                 max_cstate = 1;
1073
1074         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
1075                 cx = &pr->power.states[i];
1076
1077                 if (!cx->valid)
1078                         continue;
1079
1080 #ifdef CONFIG_HOTPLUG_CPU
1081                 if ((cx->type != ACPI_STATE_C1) && (num_online_cpus() > 1) &&
1082                     !pr->flags.has_cst &&
1083                     !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
1084                         continue;
1085 #endif
1086
1087                 state = &drv->states[count];
1088                 snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
1089                 strncpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
1090                 state->exit_latency = cx->latency;
1091                 state->target_residency = cx->latency * latency_factor;
1092
1093                 state->flags = 0;
1094                 switch (cx->type) {
1095                         case ACPI_STATE_C1:
1096                         if (cx->entry_method == ACPI_CSTATE_FFH)
1097                                 state->flags |= CPUIDLE_FLAG_TIME_VALID;
1098
1099                         state->enter = acpi_idle_enter_c1;
1100                         state->enter_dead = acpi_idle_play_dead;
1101                         drv->safe_state_index = count;
1102                         break;
1103
1104                         case ACPI_STATE_C2:
1105                         state->flags |= CPUIDLE_FLAG_TIME_VALID;
1106                         state->enter = acpi_idle_enter_simple;
1107                         state->enter_dead = acpi_idle_play_dead;
1108                         drv->safe_state_index = count;
1109                         break;
1110
1111                         case ACPI_STATE_C3:
1112                         state->flags |= CPUIDLE_FLAG_TIME_VALID;
1113                         state->enter = pr->flags.bm_check ?
1114                                         acpi_idle_enter_bm :
1115                                         acpi_idle_enter_simple;
1116                         break;
1117                 }
1118
1119                 count++;
1120                 if (count == CPUIDLE_STATE_MAX)
1121                         break;
1122         }
1123
1124         drv->state_count = count;
1125
1126         if (!count)
1127                 return -EINVAL;
1128
1129         return 0;
1130 }
1131
1132 int acpi_processor_hotplug(struct acpi_processor *pr)
1133 {
1134         int ret = 0;
1135
1136         if (disabled_by_idle_boot_param())
1137                 return 0;
1138
1139         if (!pr)
1140                 return -EINVAL;
1141
1142         if (nocst) {
1143                 return -ENODEV;
1144         }
1145
1146         if (!pr->flags.power_setup_done)
1147                 return -ENODEV;
1148
1149         cpuidle_pause_and_lock();
1150         cpuidle_disable_device(&pr->power.dev);
1151         acpi_processor_get_power_info(pr);
1152         if (pr->flags.power) {
1153                 acpi_processor_setup_cpuidle_cx(pr);
1154                 ret = cpuidle_enable_device(&pr->power.dev);
1155         }
1156         cpuidle_resume_and_unlock();
1157
1158         return ret;
1159 }
1160
1161 int acpi_processor_cst_has_changed(struct acpi_processor *pr)
1162 {
1163         int cpu;
1164         struct acpi_processor *_pr;
1165
1166         if (disabled_by_idle_boot_param())
1167                 return 0;
1168
1169         if (!pr)
1170                 return -EINVAL;
1171
1172         if (nocst)
1173                 return -ENODEV;
1174
1175         if (!pr->flags.power_setup_done)
1176                 return -ENODEV;
1177
1178         /*
1179          * FIXME:  Design the ACPI notification to make it once per
1180          * system instead of once per-cpu.  This condition is a hack
1181          * to make the code that updates C-States be called once.
1182          */
1183
1184         if (pr->id == 0 && cpuidle_get_driver() == &acpi_idle_driver) {
1185
1186                 cpuidle_pause_and_lock();
1187                 /* Protect against cpu-hotplug */
1188                 get_online_cpus();
1189
1190                 /* Disable all cpuidle devices */
1191                 for_each_online_cpu(cpu) {
1192                         _pr = per_cpu(processors, cpu);
1193                         if (!_pr || !_pr->flags.power_setup_done)
1194                                 continue;
1195                         cpuidle_disable_device(&_pr->power.dev);
1196                 }
1197
1198                 /* Populate Updated C-state information */
1199                 acpi_processor_setup_cpuidle_states(pr);
1200
1201                 /* Enable all cpuidle devices */
1202                 for_each_online_cpu(cpu) {
1203                         _pr = per_cpu(processors, cpu);
1204                         if (!_pr || !_pr->flags.power_setup_done)
1205                                 continue;
1206                         acpi_processor_get_power_info(_pr);
1207                         if (_pr->flags.power) {
1208                                 acpi_processor_setup_cpuidle_cx(_pr);
1209                                 cpuidle_enable_device(&_pr->power.dev);
1210                         }
1211                 }
1212                 put_online_cpus();
1213                 cpuidle_resume_and_unlock();
1214         }
1215
1216         return 0;
1217 }
1218
1219 static int acpi_processor_registered;
1220
1221 int __cpuinit acpi_processor_power_init(struct acpi_processor *pr,
1222                               struct acpi_device *device)
1223 {
1224         acpi_status status = 0;
1225         int retval;
1226         static int first_run;
1227
1228         if (disabled_by_idle_boot_param())
1229                 return 0;
1230
1231         if (!first_run) {
1232                 dmi_check_system(processor_power_dmi_table);
1233                 max_cstate = acpi_processor_cstate_check(max_cstate);
1234                 if (max_cstate < ACPI_C_STATES_MAX)
1235                         printk(KERN_NOTICE
1236                                "ACPI: processor limited to max C-state %d\n",
1237                                max_cstate);
1238                 first_run++;
1239         }
1240
1241         if (!pr)
1242                 return -EINVAL;
1243
1244         if (acpi_gbl_FADT.cst_control && !nocst) {
1245                 status =
1246                     acpi_os_write_port(acpi_gbl_FADT.smi_command, acpi_gbl_FADT.cst_control, 8);
1247                 if (ACPI_FAILURE(status)) {
1248                         ACPI_EXCEPTION((AE_INFO, status,
1249                                         "Notifying BIOS of _CST ability failed"));
1250                 }
1251         }
1252
1253         acpi_processor_get_power_info(pr);
1254         pr->flags.power_setup_done = 1;
1255
1256         /*
1257          * Install the idle handler if processor power management is supported.
1258          * Note that we use previously set idle handler will be used on
1259          * platforms that only support C1.
1260          */
1261         if (pr->flags.power) {
1262                 /* Register acpi_idle_driver if not already registered */
1263                 if (!acpi_processor_registered) {
1264                         acpi_processor_setup_cpuidle_states(pr);
1265                         retval = cpuidle_register_driver(&acpi_idle_driver);
1266                         if (retval)
1267                                 return retval;
1268                         printk(KERN_DEBUG "ACPI: %s registered with cpuidle\n",
1269                                         acpi_idle_driver.name);
1270                 }
1271                 /* Register per-cpu cpuidle_device. Cpuidle driver
1272                  * must already be registered before registering device
1273                  */
1274                 acpi_processor_setup_cpuidle_cx(pr);
1275                 retval = cpuidle_register_device(&pr->power.dev);
1276                 if (retval) {
1277                         if (acpi_processor_registered == 0)
1278                                 cpuidle_unregister_driver(&acpi_idle_driver);
1279                         return retval;
1280                 }
1281                 acpi_processor_registered++;
1282         }
1283         return 0;
1284 }
1285
1286 int acpi_processor_power_exit(struct acpi_processor *pr,
1287                               struct acpi_device *device)
1288 {
1289         if (disabled_by_idle_boot_param())
1290                 return 0;
1291
1292         if (pr->flags.power) {
1293                 cpuidle_unregister_device(&pr->power.dev);
1294                 acpi_processor_registered--;
1295                 if (acpi_processor_registered == 0)
1296                         cpuidle_unregister_driver(&acpi_idle_driver);
1297         }
1298
1299         pr->flags.power_setup_done = 0;
1300         return 0;
1301 }