Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6] / arch / x86 / kernel / cpu / cpufreq / acpi-cpufreq.c
1 /*
2  * acpi-cpufreq.c - ACPI Processor P-States Driver ($Revision: 1.4 $)
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) 2002 - 2004 Dominik Brodowski <linux@brodo.de>
7  *  Copyright (C) 2006       Denis Sadykov <denis.m.sadykov@intel.com>
8  *
9  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
10  *
11  *  This program is free software; you can redistribute it and/or modify
12  *  it under the terms of the GNU General Public License as published by
13  *  the Free Software Foundation; either version 2 of the License, or (at
14  *  your option) any later version.
15  *
16  *  This program is distributed in the hope that it will be useful, but
17  *  WITHOUT ANY WARRANTY; without even the implied warranty of
18  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
19  *  General Public License for more details.
20  *
21  *  You should have received a copy of the GNU General Public License along
22  *  with this program; if not, write to the Free Software Foundation, Inc.,
23  *  59 Temple Place, Suite 330, Boston, MA 02111-1307 USA.
24  *
25  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
26  */
27
28 #include <linux/kernel.h>
29 #include <linux/module.h>
30 #include <linux/init.h>
31 #include <linux/smp.h>
32 #include <linux/sched.h>
33 #include <linux/cpufreq.h>
34 #include <linux/compiler.h>
35 #include <linux/dmi.h>
36
37 #include <linux/acpi.h>
38 #include <acpi/processor.h>
39
40 #include <asm/io.h>
41 #include <asm/msr.h>
42 #include <asm/processor.h>
43 #include <asm/cpufeature.h>
44 #include <asm/delay.h>
45 #include <asm/uaccess.h>
46
47 #define dprintk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_DRIVER, "acpi-cpufreq", msg)
48
49 MODULE_AUTHOR("Paul Diefenbaugh, Dominik Brodowski");
50 MODULE_DESCRIPTION("ACPI Processor P-States Driver");
51 MODULE_LICENSE("GPL");
52
53 enum {
54         UNDEFINED_CAPABLE = 0,
55         SYSTEM_INTEL_MSR_CAPABLE,
56         SYSTEM_IO_CAPABLE,
57 };
58
59 #define INTEL_MSR_RANGE         (0xffff)
60 #define CPUID_6_ECX_APERFMPERF_CAPABILITY       (0x1)
61
62 struct acpi_cpufreq_data {
63         struct acpi_processor_performance *acpi_data;
64         struct cpufreq_frequency_table *freq_table;
65         unsigned int max_freq;
66         unsigned int resume;
67         unsigned int cpu_feature;
68 };
69
70 static DEFINE_PER_CPU(struct acpi_cpufreq_data *, drv_data);
71
72 /* acpi_perf_data is a pointer to percpu data. */
73 static struct acpi_processor_performance *acpi_perf_data;
74
75 static struct cpufreq_driver acpi_cpufreq_driver;
76
77 static unsigned int acpi_pstate_strict;
78
79 static int check_est_cpu(unsigned int cpuid)
80 {
81         struct cpuinfo_x86 *cpu = &cpu_data(cpuid);
82
83         if (cpu->x86_vendor != X86_VENDOR_INTEL ||
84             !cpu_has(cpu, X86_FEATURE_EST))
85                 return 0;
86
87         return 1;
88 }
89
90 static unsigned extract_io(u32 value, struct acpi_cpufreq_data *data)
91 {
92         struct acpi_processor_performance *perf;
93         int i;
94
95         perf = data->acpi_data;
96
97         for (i=0; i<perf->state_count; i++) {
98                 if (value == perf->states[i].status)
99                         return data->freq_table[i].frequency;
100         }
101         return 0;
102 }
103
104 static unsigned extract_msr(u32 msr, struct acpi_cpufreq_data *data)
105 {
106         int i;
107         struct acpi_processor_performance *perf;
108
109         msr &= INTEL_MSR_RANGE;
110         perf = data->acpi_data;
111
112         for (i=0; data->freq_table[i].frequency != CPUFREQ_TABLE_END; i++) {
113                 if (msr == perf->states[data->freq_table[i].index].status)
114                         return data->freq_table[i].frequency;
115         }
116         return data->freq_table[0].frequency;
117 }
118
119 static unsigned extract_freq(u32 val, struct acpi_cpufreq_data *data)
120 {
121         switch (data->cpu_feature) {
122         case SYSTEM_INTEL_MSR_CAPABLE:
123                 return extract_msr(val, data);
124         case SYSTEM_IO_CAPABLE:
125                 return extract_io(val, data);
126         default:
127                 return 0;
128         }
129 }
130
131 struct msr_addr {
132         u32 reg;
133 };
134
135 struct io_addr {
136         u16 port;
137         u8 bit_width;
138 };
139
140 typedef union {
141         struct msr_addr msr;
142         struct io_addr io;
143 } drv_addr_union;
144
145 struct drv_cmd {
146         unsigned int type;
147         cpumask_t mask;
148         drv_addr_union addr;
149         u32 val;
150 };
151
152 static void do_drv_read(struct drv_cmd *cmd)
153 {
154         u32 h;
155
156         switch (cmd->type) {
157         case SYSTEM_INTEL_MSR_CAPABLE:
158                 rdmsr(cmd->addr.msr.reg, cmd->val, h);
159                 break;
160         case SYSTEM_IO_CAPABLE:
161                 acpi_os_read_port((acpi_io_address)cmd->addr.io.port,
162                                 &cmd->val,
163                                 (u32)cmd->addr.io.bit_width);
164                 break;
165         default:
166                 break;
167         }
168 }
169
170 static void do_drv_write(struct drv_cmd *cmd)
171 {
172         u32 lo, hi;
173
174         switch (cmd->type) {
175         case SYSTEM_INTEL_MSR_CAPABLE:
176                 rdmsr(cmd->addr.msr.reg, lo, hi);
177                 lo = (lo & ~INTEL_MSR_RANGE) | (cmd->val & INTEL_MSR_RANGE);
178                 wrmsr(cmd->addr.msr.reg, lo, hi);
179                 break;
180         case SYSTEM_IO_CAPABLE:
181                 acpi_os_write_port((acpi_io_address)cmd->addr.io.port,
182                                 cmd->val,
183                                 (u32)cmd->addr.io.bit_width);
184                 break;
185         default:
186                 break;
187         }
188 }
189
190 static void drv_read(struct drv_cmd *cmd)
191 {
192         cpumask_t saved_mask = current->cpus_allowed;
193         cmd->val = 0;
194
195         set_cpus_allowed_ptr(current, &cmd->mask);
196         do_drv_read(cmd);
197         set_cpus_allowed_ptr(current, &saved_mask);
198 }
199
200 static void drv_write(struct drv_cmd *cmd)
201 {
202         cpumask_t saved_mask = current->cpus_allowed;
203         cpumask_of_cpu_ptr_declare(cpu_mask);
204         unsigned int i;
205
206         for_each_cpu_mask_nr(i, cmd->mask) {
207                 cpumask_of_cpu_ptr_next(cpu_mask, i);
208                 set_cpus_allowed_ptr(current, cpu_mask);
209                 do_drv_write(cmd);
210         }
211
212         set_cpus_allowed_ptr(current, &saved_mask);
213         return;
214 }
215
216 static u32 get_cur_val(const cpumask_t *mask)
217 {
218         struct acpi_processor_performance *perf;
219         struct drv_cmd cmd;
220
221         if (unlikely(cpus_empty(*mask)))
222                 return 0;
223
224         switch (per_cpu(drv_data, first_cpu(*mask))->cpu_feature) {
225         case SYSTEM_INTEL_MSR_CAPABLE:
226                 cmd.type = SYSTEM_INTEL_MSR_CAPABLE;
227                 cmd.addr.msr.reg = MSR_IA32_PERF_STATUS;
228                 break;
229         case SYSTEM_IO_CAPABLE:
230                 cmd.type = SYSTEM_IO_CAPABLE;
231                 perf = per_cpu(drv_data, first_cpu(*mask))->acpi_data;
232                 cmd.addr.io.port = perf->control_register.address;
233                 cmd.addr.io.bit_width = perf->control_register.bit_width;
234                 break;
235         default:
236                 return 0;
237         }
238
239         cmd.mask = *mask;
240
241         drv_read(&cmd);
242
243         dprintk("get_cur_val = %u\n", cmd.val);
244
245         return cmd.val;
246 }
247
248 /*
249  * Return the measured active (C0) frequency on this CPU since last call
250  * to this function.
251  * Input: cpu number
252  * Return: Average CPU frequency in terms of max frequency (zero on error)
253  *
254  * We use IA32_MPERF and IA32_APERF MSRs to get the measured performance
255  * over a period of time, while CPU is in C0 state.
256  * IA32_MPERF counts at the rate of max advertised frequency
257  * IA32_APERF counts at the rate of actual CPU frequency
258  * Only IA32_APERF/IA32_MPERF ratio is architecturally defined and
259  * no meaning should be associated with absolute values of these MSRs.
260  */
261 static unsigned int get_measured_perf(unsigned int cpu)
262 {
263         union {
264                 struct {
265                         u32 lo;
266                         u32 hi;
267                 } split;
268                 u64 whole;
269         } aperf_cur, mperf_cur;
270
271         cpumask_t saved_mask;
272         cpumask_of_cpu_ptr(cpu_mask, cpu);
273         unsigned int perf_percent;
274         unsigned int retval;
275
276         saved_mask = current->cpus_allowed;
277         set_cpus_allowed_ptr(current, cpu_mask);
278         if (get_cpu() != cpu) {
279                 /* We were not able to run on requested processor */
280                 put_cpu();
281                 return 0;
282         }
283
284         rdmsr(MSR_IA32_APERF, aperf_cur.split.lo, aperf_cur.split.hi);
285         rdmsr(MSR_IA32_MPERF, mperf_cur.split.lo, mperf_cur.split.hi);
286
287         wrmsr(MSR_IA32_APERF, 0,0);
288         wrmsr(MSR_IA32_MPERF, 0,0);
289
290 #ifdef __i386__
291         /*
292          * We dont want to do 64 bit divide with 32 bit kernel
293          * Get an approximate value. Return failure in case we cannot get
294          * an approximate value.
295          */
296         if (unlikely(aperf_cur.split.hi || mperf_cur.split.hi)) {
297                 int shift_count;
298                 u32 h;
299
300                 h = max_t(u32, aperf_cur.split.hi, mperf_cur.split.hi);
301                 shift_count = fls(h);
302
303                 aperf_cur.whole >>= shift_count;
304                 mperf_cur.whole >>= shift_count;
305         }
306
307         if (((unsigned long)(-1) / 100) < aperf_cur.split.lo) {
308                 int shift_count = 7;
309                 aperf_cur.split.lo >>= shift_count;
310                 mperf_cur.split.lo >>= shift_count;
311         }
312
313         if (aperf_cur.split.lo && mperf_cur.split.lo)
314                 perf_percent = (aperf_cur.split.lo * 100) / mperf_cur.split.lo;
315         else
316                 perf_percent = 0;
317
318 #else
319         if (unlikely(((unsigned long)(-1) / 100) < aperf_cur.whole)) {
320                 int shift_count = 7;
321                 aperf_cur.whole >>= shift_count;
322                 mperf_cur.whole >>= shift_count;
323         }
324
325         if (aperf_cur.whole && mperf_cur.whole)
326                 perf_percent = (aperf_cur.whole * 100) / mperf_cur.whole;
327         else
328                 perf_percent = 0;
329
330 #endif
331
332         retval = per_cpu(drv_data, cpu)->max_freq * perf_percent / 100;
333
334         put_cpu();
335         set_cpus_allowed_ptr(current, &saved_mask);
336
337         dprintk("cpu %d: performance percent %d\n", cpu, perf_percent);
338         return retval;
339 }
340
341 static unsigned int get_cur_freq_on_cpu(unsigned int cpu)
342 {
343         cpumask_of_cpu_ptr(cpu_mask, cpu);
344         struct acpi_cpufreq_data *data = per_cpu(drv_data, cpu);
345         unsigned int freq;
346         unsigned int cached_freq;
347
348         dprintk("get_cur_freq_on_cpu (%d)\n", cpu);
349
350         if (unlikely(data == NULL ||
351                      data->acpi_data == NULL || data->freq_table == NULL)) {
352                 return 0;
353         }
354
355         cached_freq = data->freq_table[data->acpi_data->state].frequency;
356         freq = extract_freq(get_cur_val(cpu_mask), data);
357         if (freq != cached_freq) {
358                 /*
359                  * The dreaded BIOS frequency change behind our back.
360                  * Force set the frequency on next target call.
361                  */
362                 data->resume = 1;
363         }
364
365         dprintk("cur freq = %u\n", freq);
366
367         return freq;
368 }
369
370 static unsigned int check_freqs(const cpumask_t *mask, unsigned int freq,
371                                 struct acpi_cpufreq_data *data)
372 {
373         unsigned int cur_freq;
374         unsigned int i;
375
376         for (i=0; i<100; i++) {
377                 cur_freq = extract_freq(get_cur_val(mask), data);
378                 if (cur_freq == freq)
379                         return 1;
380                 udelay(10);
381         }
382         return 0;
383 }
384
385 static int acpi_cpufreq_target(struct cpufreq_policy *policy,
386                                unsigned int target_freq, unsigned int relation)
387 {
388         struct acpi_cpufreq_data *data = per_cpu(drv_data, policy->cpu);
389         struct acpi_processor_performance *perf;
390         struct cpufreq_freqs freqs;
391         cpumask_t online_policy_cpus;
392         struct drv_cmd cmd;
393         unsigned int next_state = 0; /* Index into freq_table */
394         unsigned int next_perf_state = 0; /* Index into perf table */
395         unsigned int i;
396         int result = 0;
397
398         dprintk("acpi_cpufreq_target %d (%d)\n", target_freq, policy->cpu);
399
400         if (unlikely(data == NULL ||
401              data->acpi_data == NULL || data->freq_table == NULL)) {
402                 return -ENODEV;
403         }
404
405         perf = data->acpi_data;
406         result = cpufreq_frequency_table_target(policy,
407                                                 data->freq_table,
408                                                 target_freq,
409                                                 relation, &next_state);
410         if (unlikely(result))
411                 return -ENODEV;
412
413 #ifdef CONFIG_HOTPLUG_CPU
414         /* cpufreq holds the hotplug lock, so we are safe from here on */
415         cpus_and(online_policy_cpus, cpu_online_map, policy->cpus);
416 #else
417         online_policy_cpus = policy->cpus;
418 #endif
419
420         next_perf_state = data->freq_table[next_state].index;
421         if (perf->state == next_perf_state) {
422                 if (unlikely(data->resume)) {
423                         dprintk("Called after resume, resetting to P%d\n",
424                                 next_perf_state);
425                         data->resume = 0;
426                 } else {
427                         dprintk("Already at target state (P%d)\n",
428                                 next_perf_state);
429                         return 0;
430                 }
431         }
432
433         switch (data->cpu_feature) {
434         case SYSTEM_INTEL_MSR_CAPABLE:
435                 cmd.type = SYSTEM_INTEL_MSR_CAPABLE;
436                 cmd.addr.msr.reg = MSR_IA32_PERF_CTL;
437                 cmd.val = (u32) perf->states[next_perf_state].control;
438                 break;
439         case SYSTEM_IO_CAPABLE:
440                 cmd.type = SYSTEM_IO_CAPABLE;
441                 cmd.addr.io.port = perf->control_register.address;
442                 cmd.addr.io.bit_width = perf->control_register.bit_width;
443                 cmd.val = (u32) perf->states[next_perf_state].control;
444                 break;
445         default:
446                 return -ENODEV;
447         }
448
449         cpus_clear(cmd.mask);
450
451         if (policy->shared_type != CPUFREQ_SHARED_TYPE_ANY)
452                 cmd.mask = online_policy_cpus;
453         else
454                 cpu_set(policy->cpu, cmd.mask);
455
456         freqs.old = perf->states[perf->state].core_frequency * 1000;
457         freqs.new = data->freq_table[next_state].frequency;
458         for_each_cpu_mask_nr(i, cmd.mask) {
459                 freqs.cpu = i;
460                 cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
461         }
462
463         drv_write(&cmd);
464
465         if (acpi_pstate_strict) {
466                 if (!check_freqs(&cmd.mask, freqs.new, data)) {
467                         dprintk("acpi_cpufreq_target failed (%d)\n",
468                                 policy->cpu);
469                         return -EAGAIN;
470                 }
471         }
472
473         for_each_cpu_mask_nr(i, cmd.mask) {
474                 freqs.cpu = i;
475                 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
476         }
477         perf->state = next_perf_state;
478
479         return result;
480 }
481
482 static int acpi_cpufreq_verify(struct cpufreq_policy *policy)
483 {
484         struct acpi_cpufreq_data *data = per_cpu(drv_data, policy->cpu);
485
486         dprintk("acpi_cpufreq_verify\n");
487
488         return cpufreq_frequency_table_verify(policy, data->freq_table);
489 }
490
491 static unsigned long
492 acpi_cpufreq_guess_freq(struct acpi_cpufreq_data *data, unsigned int cpu)
493 {
494         struct acpi_processor_performance *perf = data->acpi_data;
495
496         if (cpu_khz) {
497                 /* search the closest match to cpu_khz */
498                 unsigned int i;
499                 unsigned long freq;
500                 unsigned long freqn = perf->states[0].core_frequency * 1000;
501
502                 for (i=0; i<(perf->state_count-1); i++) {
503                         freq = freqn;
504                         freqn = perf->states[i+1].core_frequency * 1000;
505                         if ((2 * cpu_khz) > (freqn + freq)) {
506                                 perf->state = i;
507                                 return freq;
508                         }
509                 }
510                 perf->state = perf->state_count-1;
511                 return freqn;
512         } else {
513                 /* assume CPU is at P0... */
514                 perf->state = 0;
515                 return perf->states[0].core_frequency * 1000;
516         }
517 }
518
519 /*
520  * acpi_cpufreq_early_init - initialize ACPI P-States library
521  *
522  * Initialize the ACPI P-States library (drivers/acpi/processor_perflib.c)
523  * in order to determine correct frequency and voltage pairings. We can
524  * do _PDC and _PSD and find out the processor dependency for the
525  * actual init that will happen later...
526  */
527 static int __init acpi_cpufreq_early_init(void)
528 {
529         dprintk("acpi_cpufreq_early_init\n");
530
531         acpi_perf_data = alloc_percpu(struct acpi_processor_performance);
532         if (!acpi_perf_data) {
533                 dprintk("Memory allocation error for acpi_perf_data.\n");
534                 return -ENOMEM;
535         }
536
537         /* Do initialization in ACPI core */
538         acpi_processor_preregister_performance(acpi_perf_data);
539         return 0;
540 }
541
542 #ifdef CONFIG_SMP
543 /*
544  * Some BIOSes do SW_ANY coordination internally, either set it up in hw
545  * or do it in BIOS firmware and won't inform about it to OS. If not
546  * detected, this has a side effect of making CPU run at a different speed
547  * than OS intended it to run at. Detect it and handle it cleanly.
548  */
549 static int bios_with_sw_any_bug;
550
551 static int sw_any_bug_found(const struct dmi_system_id *d)
552 {
553         bios_with_sw_any_bug = 1;
554         return 0;
555 }
556
557 static const struct dmi_system_id sw_any_bug_dmi_table[] = {
558         {
559                 .callback = sw_any_bug_found,
560                 .ident = "Supermicro Server X6DLP",
561                 .matches = {
562                         DMI_MATCH(DMI_SYS_VENDOR, "Supermicro"),
563                         DMI_MATCH(DMI_BIOS_VERSION, "080010"),
564                         DMI_MATCH(DMI_PRODUCT_NAME, "X6DLP"),
565                 },
566         },
567         { }
568 };
569 #endif
570
571 static int acpi_cpufreq_cpu_init(struct cpufreq_policy *policy)
572 {
573         unsigned int i;
574         unsigned int valid_states = 0;
575         unsigned int cpu = policy->cpu;
576         struct acpi_cpufreq_data *data;
577         unsigned int result = 0;
578         struct cpuinfo_x86 *c = &cpu_data(policy->cpu);
579         struct acpi_processor_performance *perf;
580
581         dprintk("acpi_cpufreq_cpu_init\n");
582
583         data = kzalloc(sizeof(struct acpi_cpufreq_data), GFP_KERNEL);
584         if (!data)
585                 return -ENOMEM;
586
587         data->acpi_data = percpu_ptr(acpi_perf_data, cpu);
588         per_cpu(drv_data, cpu) = data;
589
590         if (cpu_has(c, X86_FEATURE_CONSTANT_TSC))
591                 acpi_cpufreq_driver.flags |= CPUFREQ_CONST_LOOPS;
592
593         result = acpi_processor_register_performance(data->acpi_data, cpu);
594         if (result)
595                 goto err_free;
596
597         perf = data->acpi_data;
598         policy->shared_type = perf->shared_type;
599
600         /*
601          * Will let policy->cpus know about dependency only when software
602          * coordination is required.
603          */
604         if (policy->shared_type == CPUFREQ_SHARED_TYPE_ALL ||
605             policy->shared_type == CPUFREQ_SHARED_TYPE_ANY) {
606                 policy->cpus = perf->shared_cpu_map;
607         }
608         policy->related_cpus = perf->shared_cpu_map;
609
610 #ifdef CONFIG_SMP
611         dmi_check_system(sw_any_bug_dmi_table);
612         if (bios_with_sw_any_bug && cpus_weight(policy->cpus) == 1) {
613                 policy->shared_type = CPUFREQ_SHARED_TYPE_ALL;
614                 policy->cpus = per_cpu(cpu_core_map, cpu);
615         }
616 #endif
617
618         /* capability check */
619         if (perf->state_count <= 1) {
620                 dprintk("No P-States\n");
621                 result = -ENODEV;
622                 goto err_unreg;
623         }
624
625         if (perf->control_register.space_id != perf->status_register.space_id) {
626                 result = -ENODEV;
627                 goto err_unreg;
628         }
629
630         switch (perf->control_register.space_id) {
631         case ACPI_ADR_SPACE_SYSTEM_IO:
632                 dprintk("SYSTEM IO addr space\n");
633                 data->cpu_feature = SYSTEM_IO_CAPABLE;
634                 break;
635         case ACPI_ADR_SPACE_FIXED_HARDWARE:
636                 dprintk("HARDWARE addr space\n");
637                 if (!check_est_cpu(cpu)) {
638                         result = -ENODEV;
639                         goto err_unreg;
640                 }
641                 data->cpu_feature = SYSTEM_INTEL_MSR_CAPABLE;
642                 break;
643         default:
644                 dprintk("Unknown addr space %d\n",
645                         (u32) (perf->control_register.space_id));
646                 result = -ENODEV;
647                 goto err_unreg;
648         }
649
650         data->freq_table = kmalloc(sizeof(struct cpufreq_frequency_table) *
651                     (perf->state_count+1), GFP_KERNEL);
652         if (!data->freq_table) {
653                 result = -ENOMEM;
654                 goto err_unreg;
655         }
656
657         /* detect transition latency */
658         policy->cpuinfo.transition_latency = 0;
659         for (i=0; i<perf->state_count; i++) {
660                 if ((perf->states[i].transition_latency * 1000) >
661                     policy->cpuinfo.transition_latency)
662                         policy->cpuinfo.transition_latency =
663                             perf->states[i].transition_latency * 1000;
664         }
665
666         data->max_freq = perf->states[0].core_frequency * 1000;
667         /* table init */
668         for (i=0; i<perf->state_count; i++) {
669                 if (i>0 && perf->states[i].core_frequency >=
670                     data->freq_table[valid_states-1].frequency / 1000)
671                         continue;
672
673                 data->freq_table[valid_states].index = i;
674                 data->freq_table[valid_states].frequency =
675                     perf->states[i].core_frequency * 1000;
676                 valid_states++;
677         }
678         data->freq_table[valid_states].frequency = CPUFREQ_TABLE_END;
679         perf->state = 0;
680
681         result = cpufreq_frequency_table_cpuinfo(policy, data->freq_table);
682         if (result)
683                 goto err_freqfree;
684
685         switch (perf->control_register.space_id) {
686         case ACPI_ADR_SPACE_SYSTEM_IO:
687                 /* Current speed is unknown and not detectable by IO port */
688                 policy->cur = acpi_cpufreq_guess_freq(data, policy->cpu);
689                 break;
690         case ACPI_ADR_SPACE_FIXED_HARDWARE:
691                 acpi_cpufreq_driver.get = get_cur_freq_on_cpu;
692                 policy->cur = get_cur_freq_on_cpu(cpu);
693                 break;
694         default:
695                 break;
696         }
697
698         /* notify BIOS that we exist */
699         acpi_processor_notify_smm(THIS_MODULE);
700
701         /* Check for APERF/MPERF support in hardware */
702         if (c->x86_vendor == X86_VENDOR_INTEL && c->cpuid_level >= 6) {
703                 unsigned int ecx;
704                 ecx = cpuid_ecx(6);
705                 if (ecx & CPUID_6_ECX_APERFMPERF_CAPABILITY)
706                         acpi_cpufreq_driver.getavg = get_measured_perf;
707         }
708
709         dprintk("CPU%u - ACPI performance management activated.\n", cpu);
710         for (i = 0; i < perf->state_count; i++)
711                 dprintk("     %cP%d: %d MHz, %d mW, %d uS\n",
712                         (i == perf->state ? '*' : ' '), i,
713                         (u32) perf->states[i].core_frequency,
714                         (u32) perf->states[i].power,
715                         (u32) perf->states[i].transition_latency);
716
717         cpufreq_frequency_table_get_attr(data->freq_table, policy->cpu);
718
719         /*
720          * the first call to ->target() should result in us actually
721          * writing something to the appropriate registers.
722          */
723         data->resume = 1;
724
725         return result;
726
727 err_freqfree:
728         kfree(data->freq_table);
729 err_unreg:
730         acpi_processor_unregister_performance(perf, cpu);
731 err_free:
732         kfree(data);
733         per_cpu(drv_data, cpu) = NULL;
734
735         return result;
736 }
737
738 static int acpi_cpufreq_cpu_exit(struct cpufreq_policy *policy)
739 {
740         struct acpi_cpufreq_data *data = per_cpu(drv_data, policy->cpu);
741
742         dprintk("acpi_cpufreq_cpu_exit\n");
743
744         if (data) {
745                 cpufreq_frequency_table_put_attr(policy->cpu);
746                 per_cpu(drv_data, policy->cpu) = NULL;
747                 acpi_processor_unregister_performance(data->acpi_data,
748                                                       policy->cpu);
749                 kfree(data);
750         }
751
752         return 0;
753 }
754
755 static int acpi_cpufreq_resume(struct cpufreq_policy *policy)
756 {
757         struct acpi_cpufreq_data *data = per_cpu(drv_data, policy->cpu);
758
759         dprintk("acpi_cpufreq_resume\n");
760
761         data->resume = 1;
762
763         return 0;
764 }
765
766 static struct freq_attr *acpi_cpufreq_attr[] = {
767         &cpufreq_freq_attr_scaling_available_freqs,
768         NULL,
769 };
770
771 static struct cpufreq_driver acpi_cpufreq_driver = {
772         .verify = acpi_cpufreq_verify,
773         .target = acpi_cpufreq_target,
774         .init = acpi_cpufreq_cpu_init,
775         .exit = acpi_cpufreq_cpu_exit,
776         .resume = acpi_cpufreq_resume,
777         .name = "acpi-cpufreq",
778         .owner = THIS_MODULE,
779         .attr = acpi_cpufreq_attr,
780 };
781
782 static int __init acpi_cpufreq_init(void)
783 {
784         int ret;
785
786         dprintk("acpi_cpufreq_init\n");
787
788         ret = acpi_cpufreq_early_init();
789         if (ret)
790                 return ret;
791
792         return cpufreq_register_driver(&acpi_cpufreq_driver);
793 }
794
795 static void __exit acpi_cpufreq_exit(void)
796 {
797         dprintk("acpi_cpufreq_exit\n");
798
799         cpufreq_unregister_driver(&acpi_cpufreq_driver);
800
801         free_percpu(acpi_perf_data);
802
803         return;
804 }
805
806 module_param(acpi_pstate_strict, uint, 0644);
807 MODULE_PARM_DESC(acpi_pstate_strict,
808         "value 0 or non-zero. non-zero -> strict ACPI checks are "
809         "performed during frequency changes.");
810
811 late_initcall(acpi_cpufreq_init);
812 module_exit(acpi_cpufreq_exit);
813
814 MODULE_ALIAS("acpi");