xref: /linux/drivers/cpufreq/acpi-cpufreq.c (revision c0c914eca7f251c70facc37dfebeaf176601918d)
1 /*
2  * acpi-cpufreq.c - ACPI Processor P-States 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) 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 #include <linux/slab.h>
37 
38 #include <linux/acpi.h>
39 #include <linux/io.h>
40 #include <linux/delay.h>
41 #include <linux/uaccess.h>
42 
43 #include <acpi/processor.h>
44 
45 #include <asm/msr.h>
46 #include <asm/processor.h>
47 #include <asm/cpufeature.h>
48 
49 MODULE_AUTHOR("Paul Diefenbaugh, Dominik Brodowski");
50 MODULE_DESCRIPTION("ACPI Processor P-States Driver");
51 MODULE_LICENSE("GPL");
52 
53 #define PFX "acpi-cpufreq: "
54 
55 enum {
56 	UNDEFINED_CAPABLE = 0,
57 	SYSTEM_INTEL_MSR_CAPABLE,
58 	SYSTEM_AMD_MSR_CAPABLE,
59 	SYSTEM_IO_CAPABLE,
60 };
61 
62 #define INTEL_MSR_RANGE		(0xffff)
63 #define AMD_MSR_RANGE		(0x7)
64 
65 #define MSR_K7_HWCR_CPB_DIS	(1ULL << 25)
66 
67 struct acpi_cpufreq_data {
68 	struct cpufreq_frequency_table *freq_table;
69 	unsigned int resume;
70 	unsigned int cpu_feature;
71 	unsigned int acpi_perf_cpu;
72 	cpumask_var_t freqdomain_cpus;
73 };
74 
75 /* acpi_perf_data is a pointer to percpu data. */
76 static struct acpi_processor_performance __percpu *acpi_perf_data;
77 
78 static inline struct acpi_processor_performance *to_perf_data(struct acpi_cpufreq_data *data)
79 {
80 	return per_cpu_ptr(acpi_perf_data, data->acpi_perf_cpu);
81 }
82 
83 static struct cpufreq_driver acpi_cpufreq_driver;
84 
85 static unsigned int acpi_pstate_strict;
86 static struct msr __percpu *msrs;
87 
88 static bool boost_state(unsigned int cpu)
89 {
90 	u32 lo, hi;
91 	u64 msr;
92 
93 	switch (boot_cpu_data.x86_vendor) {
94 	case X86_VENDOR_INTEL:
95 		rdmsr_on_cpu(cpu, MSR_IA32_MISC_ENABLE, &lo, &hi);
96 		msr = lo | ((u64)hi << 32);
97 		return !(msr & MSR_IA32_MISC_ENABLE_TURBO_DISABLE);
98 	case X86_VENDOR_AMD:
99 		rdmsr_on_cpu(cpu, MSR_K7_HWCR, &lo, &hi);
100 		msr = lo | ((u64)hi << 32);
101 		return !(msr & MSR_K7_HWCR_CPB_DIS);
102 	}
103 	return false;
104 }
105 
106 static void boost_set_msrs(bool enable, const struct cpumask *cpumask)
107 {
108 	u32 cpu;
109 	u32 msr_addr;
110 	u64 msr_mask;
111 
112 	switch (boot_cpu_data.x86_vendor) {
113 	case X86_VENDOR_INTEL:
114 		msr_addr = MSR_IA32_MISC_ENABLE;
115 		msr_mask = MSR_IA32_MISC_ENABLE_TURBO_DISABLE;
116 		break;
117 	case X86_VENDOR_AMD:
118 		msr_addr = MSR_K7_HWCR;
119 		msr_mask = MSR_K7_HWCR_CPB_DIS;
120 		break;
121 	default:
122 		return;
123 	}
124 
125 	rdmsr_on_cpus(cpumask, msr_addr, msrs);
126 
127 	for_each_cpu(cpu, cpumask) {
128 		struct msr *reg = per_cpu_ptr(msrs, cpu);
129 		if (enable)
130 			reg->q &= ~msr_mask;
131 		else
132 			reg->q |= msr_mask;
133 	}
134 
135 	wrmsr_on_cpus(cpumask, msr_addr, msrs);
136 }
137 
138 static int set_boost(int val)
139 {
140 	get_online_cpus();
141 	boost_set_msrs(val, cpu_online_mask);
142 	put_online_cpus();
143 	pr_debug("Core Boosting %sabled.\n", val ? "en" : "dis");
144 
145 	return 0;
146 }
147 
148 static ssize_t show_freqdomain_cpus(struct cpufreq_policy *policy, char *buf)
149 {
150 	struct acpi_cpufreq_data *data = policy->driver_data;
151 
152 	if (unlikely(!data))
153 		return -ENODEV;
154 
155 	return cpufreq_show_cpus(data->freqdomain_cpus, buf);
156 }
157 
158 cpufreq_freq_attr_ro(freqdomain_cpus);
159 
160 #ifdef CONFIG_X86_ACPI_CPUFREQ_CPB
161 static ssize_t store_cpb(struct cpufreq_policy *policy, const char *buf,
162 			 size_t count)
163 {
164 	int ret;
165 	unsigned int val = 0;
166 
167 	if (!acpi_cpufreq_driver.set_boost)
168 		return -EINVAL;
169 
170 	ret = kstrtouint(buf, 10, &val);
171 	if (ret || val > 1)
172 		return -EINVAL;
173 
174 	set_boost(val);
175 
176 	return count;
177 }
178 
179 static ssize_t show_cpb(struct cpufreq_policy *policy, char *buf)
180 {
181 	return sprintf(buf, "%u\n", acpi_cpufreq_driver.boost_enabled);
182 }
183 
184 cpufreq_freq_attr_rw(cpb);
185 #endif
186 
187 static int check_est_cpu(unsigned int cpuid)
188 {
189 	struct cpuinfo_x86 *cpu = &cpu_data(cpuid);
190 
191 	return cpu_has(cpu, X86_FEATURE_EST);
192 }
193 
194 static int check_amd_hwpstate_cpu(unsigned int cpuid)
195 {
196 	struct cpuinfo_x86 *cpu = &cpu_data(cpuid);
197 
198 	return cpu_has(cpu, X86_FEATURE_HW_PSTATE);
199 }
200 
201 static unsigned extract_io(u32 value, struct acpi_cpufreq_data *data)
202 {
203 	struct acpi_processor_performance *perf;
204 	int i;
205 
206 	perf = to_perf_data(data);
207 
208 	for (i = 0; i < perf->state_count; i++) {
209 		if (value == perf->states[i].status)
210 			return data->freq_table[i].frequency;
211 	}
212 	return 0;
213 }
214 
215 static unsigned extract_msr(u32 msr, struct acpi_cpufreq_data *data)
216 {
217 	struct cpufreq_frequency_table *pos;
218 	struct acpi_processor_performance *perf;
219 
220 	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
221 		msr &= AMD_MSR_RANGE;
222 	else
223 		msr &= INTEL_MSR_RANGE;
224 
225 	perf = to_perf_data(data);
226 
227 	cpufreq_for_each_entry(pos, data->freq_table)
228 		if (msr == perf->states[pos->driver_data].status)
229 			return pos->frequency;
230 	return data->freq_table[0].frequency;
231 }
232 
233 static unsigned extract_freq(u32 val, struct acpi_cpufreq_data *data)
234 {
235 	switch (data->cpu_feature) {
236 	case SYSTEM_INTEL_MSR_CAPABLE:
237 	case SYSTEM_AMD_MSR_CAPABLE:
238 		return extract_msr(val, data);
239 	case SYSTEM_IO_CAPABLE:
240 		return extract_io(val, data);
241 	default:
242 		return 0;
243 	}
244 }
245 
246 struct msr_addr {
247 	u32 reg;
248 };
249 
250 struct io_addr {
251 	u16 port;
252 	u8 bit_width;
253 };
254 
255 struct drv_cmd {
256 	unsigned int type;
257 	const struct cpumask *mask;
258 	union {
259 		struct msr_addr msr;
260 		struct io_addr io;
261 	} addr;
262 	u32 val;
263 };
264 
265 /* Called via smp_call_function_single(), on the target CPU */
266 static void do_drv_read(void *_cmd)
267 {
268 	struct drv_cmd *cmd = _cmd;
269 	u32 h;
270 
271 	switch (cmd->type) {
272 	case SYSTEM_INTEL_MSR_CAPABLE:
273 	case SYSTEM_AMD_MSR_CAPABLE:
274 		rdmsr(cmd->addr.msr.reg, cmd->val, h);
275 		break;
276 	case SYSTEM_IO_CAPABLE:
277 		acpi_os_read_port((acpi_io_address)cmd->addr.io.port,
278 				&cmd->val,
279 				(u32)cmd->addr.io.bit_width);
280 		break;
281 	default:
282 		break;
283 	}
284 }
285 
286 /* Called via smp_call_function_many(), on the target CPUs */
287 static void do_drv_write(void *_cmd)
288 {
289 	struct drv_cmd *cmd = _cmd;
290 	u32 lo, hi;
291 
292 	switch (cmd->type) {
293 	case SYSTEM_INTEL_MSR_CAPABLE:
294 		rdmsr(cmd->addr.msr.reg, lo, hi);
295 		lo = (lo & ~INTEL_MSR_RANGE) | (cmd->val & INTEL_MSR_RANGE);
296 		wrmsr(cmd->addr.msr.reg, lo, hi);
297 		break;
298 	case SYSTEM_AMD_MSR_CAPABLE:
299 		wrmsr(cmd->addr.msr.reg, cmd->val, 0);
300 		break;
301 	case SYSTEM_IO_CAPABLE:
302 		acpi_os_write_port((acpi_io_address)cmd->addr.io.port,
303 				cmd->val,
304 				(u32)cmd->addr.io.bit_width);
305 		break;
306 	default:
307 		break;
308 	}
309 }
310 
311 static void drv_read(struct drv_cmd *cmd)
312 {
313 	int err;
314 	cmd->val = 0;
315 
316 	err = smp_call_function_any(cmd->mask, do_drv_read, cmd, 1);
317 	WARN_ON_ONCE(err);	/* smp_call_function_any() was buggy? */
318 }
319 
320 static void drv_write(struct drv_cmd *cmd)
321 {
322 	int this_cpu;
323 
324 	this_cpu = get_cpu();
325 	if (cpumask_test_cpu(this_cpu, cmd->mask))
326 		do_drv_write(cmd);
327 	smp_call_function_many(cmd->mask, do_drv_write, cmd, 1);
328 	put_cpu();
329 }
330 
331 static u32
332 get_cur_val(const struct cpumask *mask, struct acpi_cpufreq_data *data)
333 {
334 	struct acpi_processor_performance *perf;
335 	struct drv_cmd cmd;
336 
337 	if (unlikely(cpumask_empty(mask)))
338 		return 0;
339 
340 	switch (data->cpu_feature) {
341 	case SYSTEM_INTEL_MSR_CAPABLE:
342 		cmd.type = SYSTEM_INTEL_MSR_CAPABLE;
343 		cmd.addr.msr.reg = MSR_IA32_PERF_CTL;
344 		break;
345 	case SYSTEM_AMD_MSR_CAPABLE:
346 		cmd.type = SYSTEM_AMD_MSR_CAPABLE;
347 		cmd.addr.msr.reg = MSR_AMD_PERF_CTL;
348 		break;
349 	case SYSTEM_IO_CAPABLE:
350 		cmd.type = SYSTEM_IO_CAPABLE;
351 		perf = to_perf_data(data);
352 		cmd.addr.io.port = perf->control_register.address;
353 		cmd.addr.io.bit_width = perf->control_register.bit_width;
354 		break;
355 	default:
356 		return 0;
357 	}
358 
359 	cmd.mask = mask;
360 	drv_read(&cmd);
361 
362 	pr_debug("get_cur_val = %u\n", cmd.val);
363 
364 	return cmd.val;
365 }
366 
367 static unsigned int get_cur_freq_on_cpu(unsigned int cpu)
368 {
369 	struct acpi_cpufreq_data *data;
370 	struct cpufreq_policy *policy;
371 	unsigned int freq;
372 	unsigned int cached_freq;
373 
374 	pr_debug("get_cur_freq_on_cpu (%d)\n", cpu);
375 
376 	policy = cpufreq_cpu_get_raw(cpu);
377 	if (unlikely(!policy))
378 		return 0;
379 
380 	data = policy->driver_data;
381 	if (unlikely(!data || !data->freq_table))
382 		return 0;
383 
384 	cached_freq = data->freq_table[to_perf_data(data)->state].frequency;
385 	freq = extract_freq(get_cur_val(cpumask_of(cpu), data), data);
386 	if (freq != cached_freq) {
387 		/*
388 		 * The dreaded BIOS frequency change behind our back.
389 		 * Force set the frequency on next target call.
390 		 */
391 		data->resume = 1;
392 	}
393 
394 	pr_debug("cur freq = %u\n", freq);
395 
396 	return freq;
397 }
398 
399 static unsigned int check_freqs(const struct cpumask *mask, unsigned int freq,
400 				struct acpi_cpufreq_data *data)
401 {
402 	unsigned int cur_freq;
403 	unsigned int i;
404 
405 	for (i = 0; i < 100; i++) {
406 		cur_freq = extract_freq(get_cur_val(mask, data), data);
407 		if (cur_freq == freq)
408 			return 1;
409 		udelay(10);
410 	}
411 	return 0;
412 }
413 
414 static int acpi_cpufreq_target(struct cpufreq_policy *policy,
415 			       unsigned int index)
416 {
417 	struct acpi_cpufreq_data *data = policy->driver_data;
418 	struct acpi_processor_performance *perf;
419 	struct drv_cmd cmd;
420 	unsigned int next_perf_state = 0; /* Index into perf table */
421 	int result = 0;
422 
423 	if (unlikely(data == NULL || data->freq_table == NULL)) {
424 		return -ENODEV;
425 	}
426 
427 	perf = to_perf_data(data);
428 	next_perf_state = data->freq_table[index].driver_data;
429 	if (perf->state == next_perf_state) {
430 		if (unlikely(data->resume)) {
431 			pr_debug("Called after resume, resetting to P%d\n",
432 				next_perf_state);
433 			data->resume = 0;
434 		} else {
435 			pr_debug("Already at target state (P%d)\n",
436 				next_perf_state);
437 			goto out;
438 		}
439 	}
440 
441 	switch (data->cpu_feature) {
442 	case SYSTEM_INTEL_MSR_CAPABLE:
443 		cmd.type = SYSTEM_INTEL_MSR_CAPABLE;
444 		cmd.addr.msr.reg = MSR_IA32_PERF_CTL;
445 		cmd.val = (u32) perf->states[next_perf_state].control;
446 		break;
447 	case SYSTEM_AMD_MSR_CAPABLE:
448 		cmd.type = SYSTEM_AMD_MSR_CAPABLE;
449 		cmd.addr.msr.reg = MSR_AMD_PERF_CTL;
450 		cmd.val = (u32) perf->states[next_perf_state].control;
451 		break;
452 	case SYSTEM_IO_CAPABLE:
453 		cmd.type = SYSTEM_IO_CAPABLE;
454 		cmd.addr.io.port = perf->control_register.address;
455 		cmd.addr.io.bit_width = perf->control_register.bit_width;
456 		cmd.val = (u32) perf->states[next_perf_state].control;
457 		break;
458 	default:
459 		result = -ENODEV;
460 		goto out;
461 	}
462 
463 	/* cpufreq holds the hotplug lock, so we are safe from here on */
464 	if (policy->shared_type != CPUFREQ_SHARED_TYPE_ANY)
465 		cmd.mask = policy->cpus;
466 	else
467 		cmd.mask = cpumask_of(policy->cpu);
468 
469 	drv_write(&cmd);
470 
471 	if (acpi_pstate_strict) {
472 		if (!check_freqs(cmd.mask, data->freq_table[index].frequency,
473 					data)) {
474 			pr_debug("acpi_cpufreq_target failed (%d)\n",
475 				policy->cpu);
476 			result = -EAGAIN;
477 		}
478 	}
479 
480 	if (!result)
481 		perf->state = next_perf_state;
482 
483 out:
484 	return result;
485 }
486 
487 static unsigned long
488 acpi_cpufreq_guess_freq(struct acpi_cpufreq_data *data, unsigned int cpu)
489 {
490 	struct acpi_processor_performance *perf;
491 
492 	perf = to_perf_data(data);
493 	if (cpu_khz) {
494 		/* search the closest match to cpu_khz */
495 		unsigned int i;
496 		unsigned long freq;
497 		unsigned long freqn = perf->states[0].core_frequency * 1000;
498 
499 		for (i = 0; i < (perf->state_count-1); i++) {
500 			freq = freqn;
501 			freqn = perf->states[i+1].core_frequency * 1000;
502 			if ((2 * cpu_khz) > (freqn + freq)) {
503 				perf->state = i;
504 				return freq;
505 			}
506 		}
507 		perf->state = perf->state_count-1;
508 		return freqn;
509 	} else {
510 		/* assume CPU is at P0... */
511 		perf->state = 0;
512 		return perf->states[0].core_frequency * 1000;
513 	}
514 }
515 
516 static void free_acpi_perf_data(void)
517 {
518 	unsigned int i;
519 
520 	/* Freeing a NULL pointer is OK, and alloc_percpu zeroes. */
521 	for_each_possible_cpu(i)
522 		free_cpumask_var(per_cpu_ptr(acpi_perf_data, i)
523 				 ->shared_cpu_map);
524 	free_percpu(acpi_perf_data);
525 }
526 
527 static int boost_notify(struct notifier_block *nb, unsigned long action,
528 		      void *hcpu)
529 {
530 	unsigned cpu = (long)hcpu;
531 	const struct cpumask *cpumask;
532 
533 	cpumask = get_cpu_mask(cpu);
534 
535 	/*
536 	 * Clear the boost-disable bit on the CPU_DOWN path so that
537 	 * this cpu cannot block the remaining ones from boosting. On
538 	 * the CPU_UP path we simply keep the boost-disable flag in
539 	 * sync with the current global state.
540 	 */
541 
542 	switch (action) {
543 	case CPU_UP_PREPARE:
544 	case CPU_UP_PREPARE_FROZEN:
545 		boost_set_msrs(acpi_cpufreq_driver.boost_enabled, cpumask);
546 		break;
547 
548 	case CPU_DOWN_PREPARE:
549 	case CPU_DOWN_PREPARE_FROZEN:
550 		boost_set_msrs(1, cpumask);
551 		break;
552 
553 	default:
554 		break;
555 	}
556 
557 	return NOTIFY_OK;
558 }
559 
560 
561 static struct notifier_block boost_nb = {
562 	.notifier_call          = boost_notify,
563 };
564 
565 /*
566  * acpi_cpufreq_early_init - initialize ACPI P-States library
567  *
568  * Initialize the ACPI P-States library (drivers/acpi/processor_perflib.c)
569  * in order to determine correct frequency and voltage pairings. We can
570  * do _PDC and _PSD and find out the processor dependency for the
571  * actual init that will happen later...
572  */
573 static int __init acpi_cpufreq_early_init(void)
574 {
575 	unsigned int i;
576 	pr_debug("acpi_cpufreq_early_init\n");
577 
578 	acpi_perf_data = alloc_percpu(struct acpi_processor_performance);
579 	if (!acpi_perf_data) {
580 		pr_debug("Memory allocation error for acpi_perf_data.\n");
581 		return -ENOMEM;
582 	}
583 	for_each_possible_cpu(i) {
584 		if (!zalloc_cpumask_var_node(
585 			&per_cpu_ptr(acpi_perf_data, i)->shared_cpu_map,
586 			GFP_KERNEL, cpu_to_node(i))) {
587 
588 			/* Freeing a NULL pointer is OK: alloc_percpu zeroes. */
589 			free_acpi_perf_data();
590 			return -ENOMEM;
591 		}
592 	}
593 
594 	/* Do initialization in ACPI core */
595 	acpi_processor_preregister_performance(acpi_perf_data);
596 	return 0;
597 }
598 
599 #ifdef CONFIG_SMP
600 /*
601  * Some BIOSes do SW_ANY coordination internally, either set it up in hw
602  * or do it in BIOS firmware and won't inform about it to OS. If not
603  * detected, this has a side effect of making CPU run at a different speed
604  * than OS intended it to run at. Detect it and handle it cleanly.
605  */
606 static int bios_with_sw_any_bug;
607 
608 static int sw_any_bug_found(const struct dmi_system_id *d)
609 {
610 	bios_with_sw_any_bug = 1;
611 	return 0;
612 }
613 
614 static const struct dmi_system_id sw_any_bug_dmi_table[] = {
615 	{
616 		.callback = sw_any_bug_found,
617 		.ident = "Supermicro Server X6DLP",
618 		.matches = {
619 			DMI_MATCH(DMI_SYS_VENDOR, "Supermicro"),
620 			DMI_MATCH(DMI_BIOS_VERSION, "080010"),
621 			DMI_MATCH(DMI_PRODUCT_NAME, "X6DLP"),
622 		},
623 	},
624 	{ }
625 };
626 
627 static int acpi_cpufreq_blacklist(struct cpuinfo_x86 *c)
628 {
629 	/* Intel Xeon Processor 7100 Series Specification Update
630 	 * http://www.intel.com/Assets/PDF/specupdate/314554.pdf
631 	 * AL30: A Machine Check Exception (MCE) Occurring during an
632 	 * Enhanced Intel SpeedStep Technology Ratio Change May Cause
633 	 * Both Processor Cores to Lock Up. */
634 	if (c->x86_vendor == X86_VENDOR_INTEL) {
635 		if ((c->x86 == 15) &&
636 		    (c->x86_model == 6) &&
637 		    (c->x86_mask == 8)) {
638 			printk(KERN_INFO "acpi-cpufreq: Intel(R) "
639 			    "Xeon(R) 7100 Errata AL30, processors may "
640 			    "lock up on frequency changes: disabling "
641 			    "acpi-cpufreq.\n");
642 			return -ENODEV;
643 		    }
644 		}
645 	return 0;
646 }
647 #endif
648 
649 static int acpi_cpufreq_cpu_init(struct cpufreq_policy *policy)
650 {
651 	unsigned int i;
652 	unsigned int valid_states = 0;
653 	unsigned int cpu = policy->cpu;
654 	struct acpi_cpufreq_data *data;
655 	unsigned int result = 0;
656 	struct cpuinfo_x86 *c = &cpu_data(policy->cpu);
657 	struct acpi_processor_performance *perf;
658 #ifdef CONFIG_SMP
659 	static int blacklisted;
660 #endif
661 
662 	pr_debug("acpi_cpufreq_cpu_init\n");
663 
664 #ifdef CONFIG_SMP
665 	if (blacklisted)
666 		return blacklisted;
667 	blacklisted = acpi_cpufreq_blacklist(c);
668 	if (blacklisted)
669 		return blacklisted;
670 #endif
671 
672 	data = kzalloc(sizeof(*data), GFP_KERNEL);
673 	if (!data)
674 		return -ENOMEM;
675 
676 	if (!zalloc_cpumask_var(&data->freqdomain_cpus, GFP_KERNEL)) {
677 		result = -ENOMEM;
678 		goto err_free;
679 	}
680 
681 	perf = per_cpu_ptr(acpi_perf_data, cpu);
682 	data->acpi_perf_cpu = cpu;
683 	policy->driver_data = data;
684 
685 	if (cpu_has(c, X86_FEATURE_CONSTANT_TSC))
686 		acpi_cpufreq_driver.flags |= CPUFREQ_CONST_LOOPS;
687 
688 	result = acpi_processor_register_performance(perf, cpu);
689 	if (result)
690 		goto err_free_mask;
691 
692 	policy->shared_type = perf->shared_type;
693 
694 	/*
695 	 * Will let policy->cpus know about dependency only when software
696 	 * coordination is required.
697 	 */
698 	if (policy->shared_type == CPUFREQ_SHARED_TYPE_ALL ||
699 	    policy->shared_type == CPUFREQ_SHARED_TYPE_ANY) {
700 		cpumask_copy(policy->cpus, perf->shared_cpu_map);
701 	}
702 	cpumask_copy(data->freqdomain_cpus, perf->shared_cpu_map);
703 
704 #ifdef CONFIG_SMP
705 	dmi_check_system(sw_any_bug_dmi_table);
706 	if (bios_with_sw_any_bug && !policy_is_shared(policy)) {
707 		policy->shared_type = CPUFREQ_SHARED_TYPE_ALL;
708 		cpumask_copy(policy->cpus, topology_core_cpumask(cpu));
709 	}
710 
711 	if (check_amd_hwpstate_cpu(cpu) && !acpi_pstate_strict) {
712 		cpumask_clear(policy->cpus);
713 		cpumask_set_cpu(cpu, policy->cpus);
714 		cpumask_copy(data->freqdomain_cpus,
715 			     topology_sibling_cpumask(cpu));
716 		policy->shared_type = CPUFREQ_SHARED_TYPE_HW;
717 		pr_info_once(PFX "overriding BIOS provided _PSD data\n");
718 	}
719 #endif
720 
721 	/* capability check */
722 	if (perf->state_count <= 1) {
723 		pr_debug("No P-States\n");
724 		result = -ENODEV;
725 		goto err_unreg;
726 	}
727 
728 	if (perf->control_register.space_id != perf->status_register.space_id) {
729 		result = -ENODEV;
730 		goto err_unreg;
731 	}
732 
733 	switch (perf->control_register.space_id) {
734 	case ACPI_ADR_SPACE_SYSTEM_IO:
735 		if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
736 		    boot_cpu_data.x86 == 0xf) {
737 			pr_debug("AMD K8 systems must use native drivers.\n");
738 			result = -ENODEV;
739 			goto err_unreg;
740 		}
741 		pr_debug("SYSTEM IO addr space\n");
742 		data->cpu_feature = SYSTEM_IO_CAPABLE;
743 		break;
744 	case ACPI_ADR_SPACE_FIXED_HARDWARE:
745 		pr_debug("HARDWARE addr space\n");
746 		if (check_est_cpu(cpu)) {
747 			data->cpu_feature = SYSTEM_INTEL_MSR_CAPABLE;
748 			break;
749 		}
750 		if (check_amd_hwpstate_cpu(cpu)) {
751 			data->cpu_feature = SYSTEM_AMD_MSR_CAPABLE;
752 			break;
753 		}
754 		result = -ENODEV;
755 		goto err_unreg;
756 	default:
757 		pr_debug("Unknown addr space %d\n",
758 			(u32) (perf->control_register.space_id));
759 		result = -ENODEV;
760 		goto err_unreg;
761 	}
762 
763 	data->freq_table = kzalloc(sizeof(*data->freq_table) *
764 		    (perf->state_count+1), GFP_KERNEL);
765 	if (!data->freq_table) {
766 		result = -ENOMEM;
767 		goto err_unreg;
768 	}
769 
770 	/* detect transition latency */
771 	policy->cpuinfo.transition_latency = 0;
772 	for (i = 0; i < perf->state_count; i++) {
773 		if ((perf->states[i].transition_latency * 1000) >
774 		    policy->cpuinfo.transition_latency)
775 			policy->cpuinfo.transition_latency =
776 			    perf->states[i].transition_latency * 1000;
777 	}
778 
779 	/* Check for high latency (>20uS) from buggy BIOSes, like on T42 */
780 	if (perf->control_register.space_id == ACPI_ADR_SPACE_FIXED_HARDWARE &&
781 	    policy->cpuinfo.transition_latency > 20 * 1000) {
782 		policy->cpuinfo.transition_latency = 20 * 1000;
783 		printk_once(KERN_INFO
784 			    "P-state transition latency capped at 20 uS\n");
785 	}
786 
787 	/* table init */
788 	for (i = 0; i < perf->state_count; i++) {
789 		if (i > 0 && perf->states[i].core_frequency >=
790 		    data->freq_table[valid_states-1].frequency / 1000)
791 			continue;
792 
793 		data->freq_table[valid_states].driver_data = i;
794 		data->freq_table[valid_states].frequency =
795 		    perf->states[i].core_frequency * 1000;
796 		valid_states++;
797 	}
798 	data->freq_table[valid_states].frequency = CPUFREQ_TABLE_END;
799 	perf->state = 0;
800 
801 	result = cpufreq_table_validate_and_show(policy, data->freq_table);
802 	if (result)
803 		goto err_freqfree;
804 
805 	if (perf->states[0].core_frequency * 1000 != policy->cpuinfo.max_freq)
806 		printk(KERN_WARNING FW_WARN "P-state 0 is not max freq\n");
807 
808 	switch (perf->control_register.space_id) {
809 	case ACPI_ADR_SPACE_SYSTEM_IO:
810 		/*
811 		 * The core will not set policy->cur, because
812 		 * cpufreq_driver->get is NULL, so we need to set it here.
813 		 * However, we have to guess it, because the current speed is
814 		 * unknown and not detectable via IO ports.
815 		 */
816 		policy->cur = acpi_cpufreq_guess_freq(data, policy->cpu);
817 		break;
818 	case ACPI_ADR_SPACE_FIXED_HARDWARE:
819 		acpi_cpufreq_driver.get = get_cur_freq_on_cpu;
820 		break;
821 	default:
822 		break;
823 	}
824 
825 	/* notify BIOS that we exist */
826 	acpi_processor_notify_smm(THIS_MODULE);
827 
828 	pr_debug("CPU%u - ACPI performance management activated.\n", cpu);
829 	for (i = 0; i < perf->state_count; i++)
830 		pr_debug("     %cP%d: %d MHz, %d mW, %d uS\n",
831 			(i == perf->state ? '*' : ' '), i,
832 			(u32) perf->states[i].core_frequency,
833 			(u32) perf->states[i].power,
834 			(u32) perf->states[i].transition_latency);
835 
836 	/*
837 	 * the first call to ->target() should result in us actually
838 	 * writing something to the appropriate registers.
839 	 */
840 	data->resume = 1;
841 
842 	return result;
843 
844 err_freqfree:
845 	kfree(data->freq_table);
846 err_unreg:
847 	acpi_processor_unregister_performance(cpu);
848 err_free_mask:
849 	free_cpumask_var(data->freqdomain_cpus);
850 err_free:
851 	kfree(data);
852 	policy->driver_data = NULL;
853 
854 	return result;
855 }
856 
857 static int acpi_cpufreq_cpu_exit(struct cpufreq_policy *policy)
858 {
859 	struct acpi_cpufreq_data *data = policy->driver_data;
860 
861 	pr_debug("acpi_cpufreq_cpu_exit\n");
862 
863 	if (data) {
864 		policy->driver_data = NULL;
865 		acpi_processor_unregister_performance(data->acpi_perf_cpu);
866 		free_cpumask_var(data->freqdomain_cpus);
867 		kfree(data->freq_table);
868 		kfree(data);
869 	}
870 
871 	return 0;
872 }
873 
874 static int acpi_cpufreq_resume(struct cpufreq_policy *policy)
875 {
876 	struct acpi_cpufreq_data *data = policy->driver_data;
877 
878 	pr_debug("acpi_cpufreq_resume\n");
879 
880 	data->resume = 1;
881 
882 	return 0;
883 }
884 
885 static struct freq_attr *acpi_cpufreq_attr[] = {
886 	&cpufreq_freq_attr_scaling_available_freqs,
887 	&freqdomain_cpus,
888 #ifdef CONFIG_X86_ACPI_CPUFREQ_CPB
889 	&cpb,
890 #endif
891 	NULL,
892 };
893 
894 static struct cpufreq_driver acpi_cpufreq_driver = {
895 	.verify		= cpufreq_generic_frequency_table_verify,
896 	.target_index	= acpi_cpufreq_target,
897 	.bios_limit	= acpi_processor_get_bios_limit,
898 	.init		= acpi_cpufreq_cpu_init,
899 	.exit		= acpi_cpufreq_cpu_exit,
900 	.resume		= acpi_cpufreq_resume,
901 	.name		= "acpi-cpufreq",
902 	.attr		= acpi_cpufreq_attr,
903 };
904 
905 static void __init acpi_cpufreq_boost_init(void)
906 {
907 	if (boot_cpu_has(X86_FEATURE_CPB) || boot_cpu_has(X86_FEATURE_IDA)) {
908 		msrs = msrs_alloc();
909 
910 		if (!msrs)
911 			return;
912 
913 		acpi_cpufreq_driver.set_boost = set_boost;
914 		acpi_cpufreq_driver.boost_enabled = boost_state(0);
915 
916 		cpu_notifier_register_begin();
917 
918 		/* Force all MSRs to the same value */
919 		boost_set_msrs(acpi_cpufreq_driver.boost_enabled,
920 			       cpu_online_mask);
921 
922 		__register_cpu_notifier(&boost_nb);
923 
924 		cpu_notifier_register_done();
925 	}
926 }
927 
928 static void acpi_cpufreq_boost_exit(void)
929 {
930 	if (msrs) {
931 		unregister_cpu_notifier(&boost_nb);
932 
933 		msrs_free(msrs);
934 		msrs = NULL;
935 	}
936 }
937 
938 static int __init acpi_cpufreq_init(void)
939 {
940 	int ret;
941 
942 	if (acpi_disabled)
943 		return -ENODEV;
944 
945 	/* don't keep reloading if cpufreq_driver exists */
946 	if (cpufreq_get_current_driver())
947 		return -EEXIST;
948 
949 	pr_debug("acpi_cpufreq_init\n");
950 
951 	ret = acpi_cpufreq_early_init();
952 	if (ret)
953 		return ret;
954 
955 #ifdef CONFIG_X86_ACPI_CPUFREQ_CPB
956 	/* this is a sysfs file with a strange name and an even stranger
957 	 * semantic - per CPU instantiation, but system global effect.
958 	 * Lets enable it only on AMD CPUs for compatibility reasons and
959 	 * only if configured. This is considered legacy code, which
960 	 * will probably be removed at some point in the future.
961 	 */
962 	if (!check_amd_hwpstate_cpu(0)) {
963 		struct freq_attr **attr;
964 
965 		pr_debug("CPB unsupported, do not expose it\n");
966 
967 		for (attr = acpi_cpufreq_attr; *attr; attr++)
968 			if (*attr == &cpb) {
969 				*attr = NULL;
970 				break;
971 			}
972 	}
973 #endif
974 	acpi_cpufreq_boost_init();
975 
976 	ret = cpufreq_register_driver(&acpi_cpufreq_driver);
977 	if (ret) {
978 		free_acpi_perf_data();
979 		acpi_cpufreq_boost_exit();
980 	}
981 	return ret;
982 }
983 
984 static void __exit acpi_cpufreq_exit(void)
985 {
986 	pr_debug("acpi_cpufreq_exit\n");
987 
988 	acpi_cpufreq_boost_exit();
989 
990 	cpufreq_unregister_driver(&acpi_cpufreq_driver);
991 
992 	free_acpi_perf_data();
993 }
994 
995 module_param(acpi_pstate_strict, uint, 0644);
996 MODULE_PARM_DESC(acpi_pstate_strict,
997 	"value 0 or non-zero. non-zero -> strict ACPI checks are "
998 	"performed during frequency changes.");
999 
1000 late_initcall(acpi_cpufreq_init);
1001 module_exit(acpi_cpufreq_exit);
1002 
1003 static const struct x86_cpu_id acpi_cpufreq_ids[] = {
1004 	X86_FEATURE_MATCH(X86_FEATURE_ACPI),
1005 	X86_FEATURE_MATCH(X86_FEATURE_HW_PSTATE),
1006 	{}
1007 };
1008 MODULE_DEVICE_TABLE(x86cpu, acpi_cpufreq_ids);
1009 
1010 static const struct acpi_device_id processor_device_ids[] = {
1011 	{ACPI_PROCESSOR_OBJECT_HID, },
1012 	{ACPI_PROCESSOR_DEVICE_HID, },
1013 	{},
1014 };
1015 MODULE_DEVICE_TABLE(acpi, processor_device_ids);
1016 
1017 MODULE_ALIAS("acpi");
1018