time: Kill off CONFIG_GENERIC_TIME
[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/slab.h>
36 #include <linux/proc_fs.h>
37 #include <linux/seq_file.h>
38 #include <linux/acpi.h>
39 #include <linux/dmi.h>
40 #include <linux/moduleparam.h>
41 #include <linux/sched.h>        /* need_resched() */
42 #include <linux/pm_qos_params.h>
43 #include <linux/clockchips.h>
44 #include <linux/cpuidle.h>
45 #include <linux/irqflags.h>
46
47 /*
48  * Include the apic definitions for x86 to have the APIC timer related defines
49  * available also for UP (on SMP it gets magically included via linux/smp.h).
50  * asm/acpi.h is not an option, as it would require more include magic. Also
51  * creating an empty asm-ia64/apic.h would just trade pest vs. cholera.
52  */
53 #ifdef CONFIG_X86
54 #include <asm/apic.h>
55 #endif
56
57 #include <asm/io.h>
58 #include <asm/uaccess.h>
59
60 #include <acpi/acpi_bus.h>
61 #include <acpi/processor.h>
62 #include <asm/processor.h>
63
64 #define PREFIX "ACPI: "
65
66 #define ACPI_PROCESSOR_CLASS            "processor"
67 #define _COMPONENT              ACPI_PROCESSOR_COMPONENT
68 ACPI_MODULE_NAME("processor_idle");
69 #define ACPI_PROCESSOR_FILE_POWER       "power"
70 #define PM_TIMER_TICK_NS                (1000000000ULL/PM_TIMER_FREQUENCY)
71 #define C2_OVERHEAD                     1       /* 1us */
72 #define C3_OVERHEAD                     1       /* 1us */
73 #define PM_TIMER_TICKS_TO_US(p)         (((p) * 1000)/(PM_TIMER_FREQUENCY/1000))
74
75 static unsigned int max_cstate __read_mostly = ACPI_PROCESSOR_MAX_POWER;
76 module_param(max_cstate, uint, 0000);
77 static unsigned int nocst __read_mostly;
78 module_param(nocst, uint, 0000);
79 static int bm_check_disable __read_mostly;
80 module_param(bm_check_disable, uint, 0000);
81
82 static unsigned int latency_factor __read_mostly = 2;
83 module_param(latency_factor, uint, 0644);
84
85 #ifdef CONFIG_ACPI_PROCFS
86 static u64 us_to_pm_timer_ticks(s64 t)
87 {
88         return div64_u64(t * PM_TIMER_FREQUENCY, 1000000);
89 }
90 #endif
91
92 /*
93  * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
94  * For now disable this. Probably a bug somewhere else.
95  *
96  * To skip this limit, boot/load with a large max_cstate limit.
97  */
98 static int set_max_cstate(const struct dmi_system_id *id)
99 {
100         if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
101                 return 0;
102
103         printk(KERN_NOTICE PREFIX "%s detected - limiting to C%ld max_cstate."
104                " Override with \"processor.max_cstate=%d\"\n", id->ident,
105                (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
106
107         max_cstate = (long)id->driver_data;
108
109         return 0;
110 }
111
112 /* Actually this shouldn't be __cpuinitdata, would be better to fix the
113    callers to only run once -AK */
114 static struct dmi_system_id __cpuinitdata processor_power_dmi_table[] = {
115         { set_max_cstate, "Clevo 5600D", {
116           DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
117           DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")},
118          (void *)2},
119         { set_max_cstate, "Pavilion zv5000", {
120           DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
121           DMI_MATCH(DMI_PRODUCT_NAME,"Pavilion zv5000 (DS502A#ABA)")},
122          (void *)1},
123         { set_max_cstate, "Asus L8400B", {
124           DMI_MATCH(DMI_SYS_VENDOR, "ASUSTeK Computer Inc."),
125           DMI_MATCH(DMI_PRODUCT_NAME,"L8400B series Notebook PC")},
126          (void *)1},
127         {},
128 };
129
130
131 /*
132  * Callers should disable interrupts before the call and enable
133  * interrupts after return.
134  */
135 static void acpi_safe_halt(void)
136 {
137         current_thread_info()->status &= ~TS_POLLING;
138         /*
139          * TS_POLLING-cleared state must be visible before we
140          * test NEED_RESCHED:
141          */
142         smp_mb();
143         if (!need_resched()) {
144                 safe_halt();
145                 local_irq_disable();
146         }
147         current_thread_info()->status |= TS_POLLING;
148 }
149
150 #ifdef ARCH_APICTIMER_STOPS_ON_C3
151
152 /*
153  * Some BIOS implementations switch to C3 in the published C2 state.
154  * This seems to be a common problem on AMD boxen, but other vendors
155  * are affected too. We pick the most conservative approach: we assume
156  * that the local APIC stops in both C2 and C3.
157  */
158 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
159                                    struct acpi_processor_cx *cx)
160 {
161         struct acpi_processor_power *pwr = &pr->power;
162         u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2;
163
164         if (cpu_has(&cpu_data(pr->id), X86_FEATURE_ARAT))
165                 return;
166
167         if (boot_cpu_has(X86_FEATURE_AMDC1E))
168                 type = ACPI_STATE_C1;
169
170         /*
171          * Check, if one of the previous states already marked the lapic
172          * unstable
173          */
174         if (pwr->timer_broadcast_on_state < state)
175                 return;
176
177         if (cx->type >= type)
178                 pr->power.timer_broadcast_on_state = state;
179 }
180
181 static void __lapic_timer_propagate_broadcast(void *arg)
182 {
183         struct acpi_processor *pr = (struct acpi_processor *) arg;
184         unsigned long reason;
185
186         reason = pr->power.timer_broadcast_on_state < INT_MAX ?
187                 CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
188
189         clockevents_notify(reason, &pr->id);
190 }
191
192 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr)
193 {
194         smp_call_function_single(pr->id, __lapic_timer_propagate_broadcast,
195                                  (void *)pr, 1);
196 }
197
198 /* Power(C) State timer broadcast control */
199 static void lapic_timer_state_broadcast(struct acpi_processor *pr,
200                                        struct acpi_processor_cx *cx,
201                                        int broadcast)
202 {
203         int state = cx - pr->power.states;
204
205         if (state >= pr->power.timer_broadcast_on_state) {
206                 unsigned long reason;
207
208                 reason = broadcast ?  CLOCK_EVT_NOTIFY_BROADCAST_ENTER :
209                         CLOCK_EVT_NOTIFY_BROADCAST_EXIT;
210                 clockevents_notify(reason, &pr->id);
211         }
212 }
213
214 #else
215
216 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
217                                    struct acpi_processor_cx *cstate) { }
218 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) { }
219 static void lapic_timer_state_broadcast(struct acpi_processor *pr,
220                                        struct acpi_processor_cx *cx,
221                                        int broadcast)
222 {
223 }
224
225 #endif
226
227 /*
228  * Suspend / resume control
229  */
230 static int acpi_idle_suspend;
231 static u32 saved_bm_rld;
232
233 static void acpi_idle_bm_rld_save(void)
234 {
235         acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &saved_bm_rld);
236 }
237 static void acpi_idle_bm_rld_restore(void)
238 {
239         u32 resumed_bm_rld;
240
241         acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &resumed_bm_rld);
242
243         if (resumed_bm_rld != saved_bm_rld)
244                 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, saved_bm_rld);
245 }
246
247 int acpi_processor_suspend(struct acpi_device * device, pm_message_t state)
248 {
249         if (acpi_idle_suspend == 1)
250                 return 0;
251
252         acpi_idle_bm_rld_save();
253         acpi_idle_suspend = 1;
254         return 0;
255 }
256
257 int acpi_processor_resume(struct acpi_device * device)
258 {
259         if (acpi_idle_suspend == 0)
260                 return 0;
261
262         acpi_idle_bm_rld_restore();
263         acpi_idle_suspend = 0;
264         return 0;
265 }
266
267 #if defined(CONFIG_X86)
268 static void tsc_check_state(int state)
269 {
270         switch (boot_cpu_data.x86_vendor) {
271         case X86_VENDOR_AMD:
272         case X86_VENDOR_INTEL:
273                 /*
274                  * AMD Fam10h TSC will tick in all
275                  * C/P/S0/S1 states when this bit is set.
276                  */
277                 if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
278                         return;
279
280                 /*FALL THROUGH*/
281         default:
282                 /* TSC could halt in idle, so notify users */
283                 if (state > ACPI_STATE_C1)
284                         mark_tsc_unstable("TSC halts in idle");
285         }
286 }
287 #else
288 static void tsc_check_state(int state) { return; }
289 #endif
290
291 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
292 {
293
294         if (!pr)
295                 return -EINVAL;
296
297         if (!pr->pblk)
298                 return -ENODEV;
299
300         /* if info is obtained from pblk/fadt, type equals state */
301         pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
302         pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
303
304 #ifndef CONFIG_HOTPLUG_CPU
305         /*
306          * Check for P_LVL2_UP flag before entering C2 and above on
307          * an SMP system.
308          */
309         if ((num_online_cpus() > 1) &&
310             !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
311                 return -ENODEV;
312 #endif
313
314         /* determine C2 and C3 address from pblk */
315         pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
316         pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
317
318         /* determine latencies from FADT */
319         pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.C2latency;
320         pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.C3latency;
321
322         /*
323          * FADT specified C2 latency must be less than or equal to
324          * 100 microseconds.
325          */
326         if (acpi_gbl_FADT.C2latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
327                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
328                         "C2 latency too large [%d]\n", acpi_gbl_FADT.C2latency));
329                 /* invalidate C2 */
330                 pr->power.states[ACPI_STATE_C2].address = 0;
331         }
332
333         /*
334          * FADT supplied C3 latency must be less than or equal to
335          * 1000 microseconds.
336          */
337         if (acpi_gbl_FADT.C3latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
338                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
339                         "C3 latency too large [%d]\n", acpi_gbl_FADT.C3latency));
340                 /* invalidate C3 */
341                 pr->power.states[ACPI_STATE_C3].address = 0;
342         }
343
344         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
345                           "lvl2[0x%08x] lvl3[0x%08x]\n",
346                           pr->power.states[ACPI_STATE_C2].address,
347                           pr->power.states[ACPI_STATE_C3].address));
348
349         return 0;
350 }
351
352 static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
353 {
354         if (!pr->power.states[ACPI_STATE_C1].valid) {
355                 /* set the first C-State to C1 */
356                 /* all processors need to support C1 */
357                 pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
358                 pr->power.states[ACPI_STATE_C1].valid = 1;
359                 pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
360         }
361         /* the C0 state only exists as a filler in our array */
362         pr->power.states[ACPI_STATE_C0].valid = 1;
363         return 0;
364 }
365
366 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
367 {
368         acpi_status status = 0;
369         u64 count;
370         int current_count;
371         int i;
372         struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
373         union acpi_object *cst;
374
375
376         if (nocst)
377                 return -ENODEV;
378
379         current_count = 0;
380
381         status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
382         if (ACPI_FAILURE(status)) {
383                 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
384                 return -ENODEV;
385         }
386
387         cst = buffer.pointer;
388
389         /* There must be at least 2 elements */
390         if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
391                 printk(KERN_ERR PREFIX "not enough elements in _CST\n");
392                 status = -EFAULT;
393                 goto end;
394         }
395
396         count = cst->package.elements[0].integer.value;
397
398         /* Validate number of power states. */
399         if (count < 1 || count != cst->package.count - 1) {
400                 printk(KERN_ERR PREFIX "count given by _CST is not valid\n");
401                 status = -EFAULT;
402                 goto end;
403         }
404
405         /* Tell driver that at least _CST is supported. */
406         pr->flags.has_cst = 1;
407
408         for (i = 1; i <= count; i++) {
409                 union acpi_object *element;
410                 union acpi_object *obj;
411                 struct acpi_power_register *reg;
412                 struct acpi_processor_cx cx;
413
414                 memset(&cx, 0, sizeof(cx));
415
416                 element = &(cst->package.elements[i]);
417                 if (element->type != ACPI_TYPE_PACKAGE)
418                         continue;
419
420                 if (element->package.count != 4)
421                         continue;
422
423                 obj = &(element->package.elements[0]);
424
425                 if (obj->type != ACPI_TYPE_BUFFER)
426                         continue;
427
428                 reg = (struct acpi_power_register *)obj->buffer.pointer;
429
430                 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
431                     (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
432                         continue;
433
434                 /* There should be an easy way to extract an integer... */
435                 obj = &(element->package.elements[1]);
436                 if (obj->type != ACPI_TYPE_INTEGER)
437                         continue;
438
439                 cx.type = obj->integer.value;
440                 /*
441                  * Some buggy BIOSes won't list C1 in _CST -
442                  * Let acpi_processor_get_power_info_default() handle them later
443                  */
444                 if (i == 1 && cx.type != ACPI_STATE_C1)
445                         current_count++;
446
447                 cx.address = reg->address;
448                 cx.index = current_count + 1;
449
450                 cx.entry_method = ACPI_CSTATE_SYSTEMIO;
451                 if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) {
452                         if (acpi_processor_ffh_cstate_probe
453                                         (pr->id, &cx, reg) == 0) {
454                                 cx.entry_method = ACPI_CSTATE_FFH;
455                         } else if (cx.type == ACPI_STATE_C1) {
456                                 /*
457                                  * C1 is a special case where FIXED_HARDWARE
458                                  * can be handled in non-MWAIT way as well.
459                                  * In that case, save this _CST entry info.
460                                  * Otherwise, ignore this info and continue.
461                                  */
462                                 cx.entry_method = ACPI_CSTATE_HALT;
463                                 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
464                         } else {
465                                 continue;
466                         }
467                         if (cx.type == ACPI_STATE_C1 &&
468                                         (idle_halt || idle_nomwait)) {
469                                 /*
470                                  * In most cases the C1 space_id obtained from
471                                  * _CST object is FIXED_HARDWARE access mode.
472                                  * But when the option of idle=halt is added,
473                                  * the entry_method type should be changed from
474                                  * CSTATE_FFH to CSTATE_HALT.
475                                  * When the option of idle=nomwait is added,
476                                  * the C1 entry_method type should be
477                                  * CSTATE_HALT.
478                                  */
479                                 cx.entry_method = ACPI_CSTATE_HALT;
480                                 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
481                         }
482                 } else {
483                         snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI IOPORT 0x%x",
484                                  cx.address);
485                 }
486
487                 if (cx.type == ACPI_STATE_C1) {
488                         cx.valid = 1;
489                 }
490
491                 obj = &(element->package.elements[2]);
492                 if (obj->type != ACPI_TYPE_INTEGER)
493                         continue;
494
495                 cx.latency = obj->integer.value;
496
497                 obj = &(element->package.elements[3]);
498                 if (obj->type != ACPI_TYPE_INTEGER)
499                         continue;
500
501                 cx.power = obj->integer.value;
502
503                 current_count++;
504                 memcpy(&(pr->power.states[current_count]), &cx, sizeof(cx));
505
506                 /*
507                  * We support total ACPI_PROCESSOR_MAX_POWER - 1
508                  * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
509                  */
510                 if (current_count >= (ACPI_PROCESSOR_MAX_POWER - 1)) {
511                         printk(KERN_WARNING
512                                "Limiting number of power states to max (%d)\n",
513                                ACPI_PROCESSOR_MAX_POWER);
514                         printk(KERN_WARNING
515                                "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
516                         break;
517                 }
518         }
519
520         ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n",
521                           current_count));
522
523         /* Validate number of power states discovered */
524         if (current_count < 2)
525                 status = -EFAULT;
526
527       end:
528         kfree(buffer.pointer);
529
530         return status;
531 }
532
533 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
534                                            struct acpi_processor_cx *cx)
535 {
536         static int bm_check_flag = -1;
537         static int bm_control_flag = -1;
538
539
540         if (!cx->address)
541                 return;
542
543         /*
544          * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
545          * DMA transfers are used by any ISA device to avoid livelock.
546          * Note that we could disable Type-F DMA (as recommended by
547          * the erratum), but this is known to disrupt certain ISA
548          * devices thus we take the conservative approach.
549          */
550         else if (errata.piix4.fdma) {
551                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
552                                   "C3 not supported on PIIX4 with Type-F DMA\n"));
553                 return;
554         }
555
556         /* All the logic here assumes flags.bm_check is same across all CPUs */
557         if (bm_check_flag == -1) {
558                 /* Determine whether bm_check is needed based on CPU  */
559                 acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
560                 bm_check_flag = pr->flags.bm_check;
561                 bm_control_flag = pr->flags.bm_control;
562         } else {
563                 pr->flags.bm_check = bm_check_flag;
564                 pr->flags.bm_control = bm_control_flag;
565         }
566
567         if (pr->flags.bm_check) {
568                 if (!pr->flags.bm_control) {
569                         if (pr->flags.has_cst != 1) {
570                                 /* bus mastering control is necessary */
571                                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
572                                         "C3 support requires BM control\n"));
573                                 return;
574                         } else {
575                                 /* Here we enter C3 without bus mastering */
576                                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
577                                         "C3 support without BM control\n"));
578                         }
579                 }
580         } else {
581                 /*
582                  * WBINVD should be set in fadt, for C3 state to be
583                  * supported on when bm_check is not required.
584                  */
585                 if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
586                         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
587                                           "Cache invalidation should work properly"
588                                           " for C3 to be enabled on SMP systems\n"));
589                         return;
590                 }
591         }
592
593         /*
594          * Otherwise we've met all of our C3 requirements.
595          * Normalize the C3 latency to expidite policy.  Enable
596          * checking of bus mastering status (bm_check) so we can
597          * use this in our C3 policy
598          */
599         cx->valid = 1;
600
601         cx->latency_ticks = cx->latency;
602         /*
603          * On older chipsets, BM_RLD needs to be set
604          * in order for Bus Master activity to wake the
605          * system from C3.  Newer chipsets handle DMA
606          * during C3 automatically and BM_RLD is a NOP.
607          * In either case, the proper way to
608          * handle BM_RLD is to set it and leave it set.
609          */
610         acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
611
612         return;
613 }
614
615 static int acpi_processor_power_verify(struct acpi_processor *pr)
616 {
617         unsigned int i;
618         unsigned int working = 0;
619
620         pr->power.timer_broadcast_on_state = INT_MAX;
621
622         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
623                 struct acpi_processor_cx *cx = &pr->power.states[i];
624
625                 switch (cx->type) {
626                 case ACPI_STATE_C1:
627                         cx->valid = 1;
628                         break;
629
630                 case ACPI_STATE_C2:
631                         if (!cx->address)
632                                 break;
633                         cx->valid = 1; 
634                         cx->latency_ticks = cx->latency; /* Normalize latency */
635                         break;
636
637                 case ACPI_STATE_C3:
638                         acpi_processor_power_verify_c3(pr, cx);
639                         break;
640                 }
641                 if (!cx->valid)
642                         continue;
643
644                 lapic_timer_check_state(i, pr, cx);
645                 tsc_check_state(cx->type);
646                 working++;
647         }
648
649         lapic_timer_propagate_broadcast(pr);
650
651         return (working);
652 }
653
654 static int acpi_processor_get_power_info(struct acpi_processor *pr)
655 {
656         unsigned int i;
657         int result;
658
659
660         /* NOTE: the idle thread may not be running while calling
661          * this function */
662
663         /* Zero initialize all the C-states info. */
664         memset(pr->power.states, 0, sizeof(pr->power.states));
665
666         result = acpi_processor_get_power_info_cst(pr);
667         if (result == -ENODEV)
668                 result = acpi_processor_get_power_info_fadt(pr);
669
670         if (result)
671                 return result;
672
673         acpi_processor_get_power_info_default(pr);
674
675         pr->power.count = acpi_processor_power_verify(pr);
676
677         /*
678          * if one state of type C2 or C3 is available, mark this
679          * CPU as being "idle manageable"
680          */
681         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
682                 if (pr->power.states[i].valid) {
683                         pr->power.count = i;
684                         if (pr->power.states[i].type >= ACPI_STATE_C2)
685                                 pr->flags.power = 1;
686                 }
687         }
688
689         return 0;
690 }
691
692 #ifdef CONFIG_ACPI_PROCFS
693 static int acpi_processor_power_seq_show(struct seq_file *seq, void *offset)
694 {
695         struct acpi_processor *pr = seq->private;
696         unsigned int i;
697
698
699         if (!pr)
700                 goto end;
701
702         seq_printf(seq, "active state:            C%zd\n"
703                    "max_cstate:              C%d\n"
704                    "maximum allowed latency: %d usec\n",
705                    pr->power.state ? pr->power.state - pr->power.states : 0,
706                    max_cstate, pm_qos_request(PM_QOS_CPU_DMA_LATENCY));
707
708         seq_puts(seq, "states:\n");
709
710         for (i = 1; i <= pr->power.count; i++) {
711                 seq_printf(seq, "   %cC%d:                  ",
712                            (&pr->power.states[i] ==
713                             pr->power.state ? '*' : ' '), i);
714
715                 if (!pr->power.states[i].valid) {
716                         seq_puts(seq, "<not supported>\n");
717                         continue;
718                 }
719
720                 switch (pr->power.states[i].type) {
721                 case ACPI_STATE_C1:
722                         seq_printf(seq, "type[C1] ");
723                         break;
724                 case ACPI_STATE_C2:
725                         seq_printf(seq, "type[C2] ");
726                         break;
727                 case ACPI_STATE_C3:
728                         seq_printf(seq, "type[C3] ");
729                         break;
730                 default:
731                         seq_printf(seq, "type[--] ");
732                         break;
733                 }
734
735                 seq_puts(seq, "promotion[--] ");
736
737                 seq_puts(seq, "demotion[--] ");
738
739                 seq_printf(seq, "latency[%03d] usage[%08d] duration[%020Lu]\n",
740                            pr->power.states[i].latency,
741                            pr->power.states[i].usage,
742                            us_to_pm_timer_ticks(pr->power.states[i].time));
743         }
744
745       end:
746         return 0;
747 }
748
749 static int acpi_processor_power_open_fs(struct inode *inode, struct file *file)
750 {
751         return single_open(file, acpi_processor_power_seq_show,
752                            PDE(inode)->data);
753 }
754
755 static const struct file_operations acpi_processor_power_fops = {
756         .owner = THIS_MODULE,
757         .open = acpi_processor_power_open_fs,
758         .read = seq_read,
759         .llseek = seq_lseek,
760         .release = single_release,
761 };
762 #endif
763
764 /**
765  * acpi_idle_bm_check - checks if bus master activity was detected
766  */
767 static int acpi_idle_bm_check(void)
768 {
769         u32 bm_status = 0;
770
771         if (bm_check_disable)
772                 return 0;
773
774         acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
775         if (bm_status)
776                 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
777         /*
778          * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
779          * the true state of bus mastering activity; forcing us to
780          * manually check the BMIDEA bit of each IDE channel.
781          */
782         else if (errata.piix4.bmisx) {
783                 if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
784                     || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
785                         bm_status = 1;
786         }
787         return bm_status;
788 }
789
790 /**
791  * acpi_idle_do_entry - a helper function that does C2 and C3 type entry
792  * @cx: cstate data
793  *
794  * Caller disables interrupt before call and enables interrupt after return.
795  */
796 static inline void acpi_idle_do_entry(struct acpi_processor_cx *cx)
797 {
798         /* Don't trace irqs off for idle */
799         stop_critical_timings();
800         if (cx->entry_method == ACPI_CSTATE_FFH) {
801                 /* Call into architectural FFH based C-state */
802                 acpi_processor_ffh_cstate_enter(cx);
803         } else if (cx->entry_method == ACPI_CSTATE_HALT) {
804                 acpi_safe_halt();
805         } else {
806                 int unused;
807                 /* IO port based C-state */
808                 inb(cx->address);
809                 /* Dummy wait op - must do something useless after P_LVL2 read
810                    because chipsets cannot guarantee that STPCLK# signal
811                    gets asserted in time to freeze execution properly. */
812                 unused = inl(acpi_gbl_FADT.xpm_timer_block.address);
813         }
814         start_critical_timings();
815 }
816
817 /**
818  * acpi_idle_enter_c1 - enters an ACPI C1 state-type
819  * @dev: the target CPU
820  * @state: the state data
821  *
822  * This is equivalent to the HALT instruction.
823  */
824 static int acpi_idle_enter_c1(struct cpuidle_device *dev,
825                               struct cpuidle_state *state)
826 {
827         ktime_t  kt1, kt2;
828         s64 idle_time;
829         struct acpi_processor *pr;
830         struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
831
832         pr = __get_cpu_var(processors);
833
834         if (unlikely(!pr))
835                 return 0;
836
837         local_irq_disable();
838
839         /* Do not access any ACPI IO ports in suspend path */
840         if (acpi_idle_suspend) {
841                 local_irq_enable();
842                 cpu_relax();
843                 return 0;
844         }
845
846         lapic_timer_state_broadcast(pr, cx, 1);
847         kt1 = ktime_get_real();
848         acpi_idle_do_entry(cx);
849         kt2 = ktime_get_real();
850         idle_time =  ktime_to_us(ktime_sub(kt2, kt1));
851
852         local_irq_enable();
853         cx->usage++;
854         lapic_timer_state_broadcast(pr, cx, 0);
855
856         return idle_time;
857 }
858
859 /**
860  * acpi_idle_enter_simple - enters an ACPI state without BM handling
861  * @dev: the target CPU
862  * @state: the state data
863  */
864 static int acpi_idle_enter_simple(struct cpuidle_device *dev,
865                                   struct cpuidle_state *state)
866 {
867         struct acpi_processor *pr;
868         struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
869         ktime_t  kt1, kt2;
870         s64 idle_time_ns;
871         s64 idle_time;
872
873         pr = __get_cpu_var(processors);
874
875         if (unlikely(!pr))
876                 return 0;
877
878         if (acpi_idle_suspend)
879                 return(acpi_idle_enter_c1(dev, state));
880
881         local_irq_disable();
882
883         if (cx->entry_method != ACPI_CSTATE_FFH) {
884                 current_thread_info()->status &= ~TS_POLLING;
885                 /*
886                  * TS_POLLING-cleared state must be visible before we test
887                  * NEED_RESCHED:
888                  */
889                 smp_mb();
890
891                 if (unlikely(need_resched())) {
892                         current_thread_info()->status |= TS_POLLING;
893                         local_irq_enable();
894                         return 0;
895                 }
896         }
897
898         /*
899          * Must be done before busmaster disable as we might need to
900          * access HPET !
901          */
902         lapic_timer_state_broadcast(pr, cx, 1);
903
904         if (cx->type == ACPI_STATE_C3)
905                 ACPI_FLUSH_CPU_CACHE();
906
907         kt1 = ktime_get_real();
908         /* Tell the scheduler that we are going deep-idle: */
909         sched_clock_idle_sleep_event();
910         acpi_idle_do_entry(cx);
911         kt2 = ktime_get_real();
912         idle_time_ns = ktime_to_ns(ktime_sub(kt2, kt1));
913         idle_time = idle_time_ns;
914         do_div(idle_time, NSEC_PER_USEC);
915
916         /* Tell the scheduler how much we idled: */
917         sched_clock_idle_wakeup_event(idle_time_ns);
918
919         local_irq_enable();
920         if (cx->entry_method != ACPI_CSTATE_FFH)
921                 current_thread_info()->status |= TS_POLLING;
922
923         cx->usage++;
924
925         lapic_timer_state_broadcast(pr, cx, 0);
926         cx->time += idle_time;
927         return idle_time;
928 }
929
930 static int c3_cpu_count;
931 static DEFINE_SPINLOCK(c3_lock);
932
933 /**
934  * acpi_idle_enter_bm - enters C3 with proper BM handling
935  * @dev: the target CPU
936  * @state: the state data
937  *
938  * If BM is detected, the deepest non-C3 idle state is entered instead.
939  */
940 static int acpi_idle_enter_bm(struct cpuidle_device *dev,
941                               struct cpuidle_state *state)
942 {
943         struct acpi_processor *pr;
944         struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
945         ktime_t  kt1, kt2;
946         s64 idle_time_ns;
947         s64 idle_time;
948
949
950         pr = __get_cpu_var(processors);
951
952         if (unlikely(!pr))
953                 return 0;
954
955         if (acpi_idle_suspend)
956                 return(acpi_idle_enter_c1(dev, state));
957
958         if (!cx->bm_sts_skip && acpi_idle_bm_check()) {
959                 if (dev->safe_state) {
960                         dev->last_state = dev->safe_state;
961                         return dev->safe_state->enter(dev, dev->safe_state);
962                 } else {
963                         local_irq_disable();
964                         acpi_safe_halt();
965                         local_irq_enable();
966                         return 0;
967                 }
968         }
969
970         local_irq_disable();
971
972         if (cx->entry_method != ACPI_CSTATE_FFH) {
973                 current_thread_info()->status &= ~TS_POLLING;
974                 /*
975                  * TS_POLLING-cleared state must be visible before we test
976                  * NEED_RESCHED:
977                  */
978                 smp_mb();
979
980                 if (unlikely(need_resched())) {
981                         current_thread_info()->status |= TS_POLLING;
982                         local_irq_enable();
983                         return 0;
984                 }
985         }
986
987         acpi_unlazy_tlb(smp_processor_id());
988
989         /* Tell the scheduler that we are going deep-idle: */
990         sched_clock_idle_sleep_event();
991         /*
992          * Must be done before busmaster disable as we might need to
993          * access HPET !
994          */
995         lapic_timer_state_broadcast(pr, cx, 1);
996
997         kt1 = ktime_get_real();
998         /*
999          * disable bus master
1000          * bm_check implies we need ARB_DIS
1001          * !bm_check implies we need cache flush
1002          * bm_control implies whether we can do ARB_DIS
1003          *
1004          * That leaves a case where bm_check is set and bm_control is
1005          * not set. In that case we cannot do much, we enter C3
1006          * without doing anything.
1007          */
1008         if (pr->flags.bm_check && pr->flags.bm_control) {
1009                 spin_lock(&c3_lock);
1010                 c3_cpu_count++;
1011                 /* Disable bus master arbitration when all CPUs are in C3 */
1012                 if (c3_cpu_count == num_online_cpus())
1013                         acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
1014                 spin_unlock(&c3_lock);
1015         } else if (!pr->flags.bm_check) {
1016                 ACPI_FLUSH_CPU_CACHE();
1017         }
1018
1019         acpi_idle_do_entry(cx);
1020
1021         /* Re-enable bus master arbitration */
1022         if (pr->flags.bm_check && pr->flags.bm_control) {
1023                 spin_lock(&c3_lock);
1024                 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
1025                 c3_cpu_count--;
1026                 spin_unlock(&c3_lock);
1027         }
1028         kt2 = ktime_get_real();
1029         idle_time_ns = ktime_to_ns(ktime_sub(kt2, kt1));
1030         idle_time = idle_time_ns;
1031         do_div(idle_time, NSEC_PER_USEC);
1032
1033         /* Tell the scheduler how much we idled: */
1034         sched_clock_idle_wakeup_event(idle_time_ns);
1035
1036         local_irq_enable();
1037         if (cx->entry_method != ACPI_CSTATE_FFH)
1038                 current_thread_info()->status |= TS_POLLING;
1039
1040         cx->usage++;
1041
1042         lapic_timer_state_broadcast(pr, cx, 0);
1043         cx->time += idle_time;
1044         return idle_time;
1045 }
1046
1047 struct cpuidle_driver acpi_idle_driver = {
1048         .name =         "acpi_idle",
1049         .owner =        THIS_MODULE,
1050 };
1051
1052 /**
1053  * acpi_processor_setup_cpuidle - prepares and configures CPUIDLE
1054  * @pr: the ACPI processor
1055  */
1056 static int acpi_processor_setup_cpuidle(struct acpi_processor *pr)
1057 {
1058         int i, count = CPUIDLE_DRIVER_STATE_START;
1059         struct acpi_processor_cx *cx;
1060         struct cpuidle_state *state;
1061         struct cpuidle_device *dev = &pr->power.dev;
1062
1063         if (!pr->flags.power_setup_done)
1064                 return -EINVAL;
1065
1066         if (pr->flags.power == 0) {
1067                 return -EINVAL;
1068         }
1069
1070         dev->cpu = pr->id;
1071         for (i = 0; i < CPUIDLE_STATE_MAX; i++) {
1072                 dev->states[i].name[0] = '\0';
1073                 dev->states[i].desc[0] = '\0';
1074         }
1075
1076         if (max_cstate == 0)
1077                 max_cstate = 1;
1078
1079         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
1080                 cx = &pr->power.states[i];
1081                 state = &dev->states[count];
1082
1083                 if (!cx->valid)
1084                         continue;
1085
1086 #ifdef CONFIG_HOTPLUG_CPU
1087                 if ((cx->type != ACPI_STATE_C1) && (num_online_cpus() > 1) &&
1088                     !pr->flags.has_cst &&
1089                     !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
1090                         continue;
1091 #endif
1092                 cpuidle_set_statedata(state, cx);
1093
1094                 snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
1095                 strncpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
1096                 state->exit_latency = cx->latency;
1097                 state->target_residency = cx->latency * latency_factor;
1098                 state->power_usage = cx->power;
1099
1100                 state->flags = 0;
1101                 switch (cx->type) {
1102                         case ACPI_STATE_C1:
1103                         state->flags |= CPUIDLE_FLAG_SHALLOW;
1104                         if (cx->entry_method == ACPI_CSTATE_FFH)
1105                                 state->flags |= CPUIDLE_FLAG_TIME_VALID;
1106
1107                         state->enter = acpi_idle_enter_c1;
1108                         dev->safe_state = state;
1109                         break;
1110
1111                         case ACPI_STATE_C2:
1112                         state->flags |= CPUIDLE_FLAG_BALANCED;
1113                         state->flags |= CPUIDLE_FLAG_TIME_VALID;
1114                         state->enter = acpi_idle_enter_simple;
1115                         dev->safe_state = state;
1116                         break;
1117
1118                         case ACPI_STATE_C3:
1119                         state->flags |= CPUIDLE_FLAG_DEEP;
1120                         state->flags |= CPUIDLE_FLAG_TIME_VALID;
1121                         state->flags |= CPUIDLE_FLAG_CHECK_BM;
1122                         state->enter = pr->flags.bm_check ?
1123                                         acpi_idle_enter_bm :
1124                                         acpi_idle_enter_simple;
1125                         break;
1126                 }
1127
1128                 count++;
1129                 if (count == CPUIDLE_STATE_MAX)
1130                         break;
1131         }
1132
1133         dev->state_count = count;
1134
1135         if (!count)
1136                 return -EINVAL;
1137
1138         return 0;
1139 }
1140
1141 int acpi_processor_cst_has_changed(struct acpi_processor *pr)
1142 {
1143         int ret = 0;
1144
1145         if (boot_option_idle_override)
1146                 return 0;
1147
1148         if (!pr)
1149                 return -EINVAL;
1150
1151         if (nocst) {
1152                 return -ENODEV;
1153         }
1154
1155         if (!pr->flags.power_setup_done)
1156                 return -ENODEV;
1157
1158         cpuidle_pause_and_lock();
1159         cpuidle_disable_device(&pr->power.dev);
1160         acpi_processor_get_power_info(pr);
1161         if (pr->flags.power) {
1162                 acpi_processor_setup_cpuidle(pr);
1163                 ret = cpuidle_enable_device(&pr->power.dev);
1164         }
1165         cpuidle_resume_and_unlock();
1166
1167         return ret;
1168 }
1169
1170 int __cpuinit acpi_processor_power_init(struct acpi_processor *pr,
1171                               struct acpi_device *device)
1172 {
1173         acpi_status status = 0;
1174         static int first_run;
1175 #ifdef CONFIG_ACPI_PROCFS
1176         struct proc_dir_entry *entry = NULL;
1177 #endif
1178
1179         if (boot_option_idle_override)
1180                 return 0;
1181
1182         if (!first_run) {
1183                 if (idle_halt) {
1184                         /*
1185                          * When the boot option of "idle=halt" is added, halt
1186                          * is used for CPU IDLE.
1187                          * In such case C2/C3 is meaningless. So the max_cstate
1188                          * is set to one.
1189                          */
1190                         max_cstate = 1;
1191                 }
1192                 dmi_check_system(processor_power_dmi_table);
1193                 max_cstate = acpi_processor_cstate_check(max_cstate);
1194                 if (max_cstate < ACPI_C_STATES_MAX)
1195                         printk(KERN_NOTICE
1196                                "ACPI: processor limited to max C-state %d\n",
1197                                max_cstate);
1198                 first_run++;
1199         }
1200
1201         if (!pr)
1202                 return -EINVAL;
1203
1204         if (acpi_gbl_FADT.cst_control && !nocst) {
1205                 status =
1206                     acpi_os_write_port(acpi_gbl_FADT.smi_command, acpi_gbl_FADT.cst_control, 8);
1207                 if (ACPI_FAILURE(status)) {
1208                         ACPI_EXCEPTION((AE_INFO, status,
1209                                         "Notifying BIOS of _CST ability failed"));
1210                 }
1211         }
1212
1213         acpi_processor_get_power_info(pr);
1214         pr->flags.power_setup_done = 1;
1215
1216         /*
1217          * Install the idle handler if processor power management is supported.
1218          * Note that we use previously set idle handler will be used on
1219          * platforms that only support C1.
1220          */
1221         if (pr->flags.power) {
1222                 acpi_processor_setup_cpuidle(pr);
1223                 if (cpuidle_register_device(&pr->power.dev))
1224                         return -EIO;
1225         }
1226 #ifdef CONFIG_ACPI_PROCFS
1227         /* 'power' [R] */
1228         entry = proc_create_data(ACPI_PROCESSOR_FILE_POWER,
1229                                  S_IRUGO, acpi_device_dir(device),
1230                                  &acpi_processor_power_fops,
1231                                  acpi_driver_data(device));
1232         if (!entry)
1233                 return -EIO;
1234 #endif
1235         return 0;
1236 }
1237
1238 int acpi_processor_power_exit(struct acpi_processor *pr,
1239                               struct acpi_device *device)
1240 {
1241         if (boot_option_idle_override)
1242                 return 0;
1243
1244         cpuidle_unregister_device(&pr->power.dev);
1245         pr->flags.power_setup_done = 0;
1246
1247 #ifdef CONFIG_ACPI_PROCFS
1248         if (acpi_device_dir(device))
1249                 remove_proc_entry(ACPI_PROCESSOR_FILE_POWER,
1250                                   acpi_device_dir(device));
1251 #endif
1252
1253         return 0;
1254 }