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 acpi_processor_performance *acpi_data;
69 	struct cpufreq_frequency_table *freq_table;
70 	unsigned int resume;
71 	unsigned int cpu_feature;
72 	cpumask_var_t freqdomain_cpus;
73 };
74 
75 static DEFINE_PER_CPU(struct acpi_cpufreq_data *, acfreq_data);
76 
77 /* acpi_perf_data is a pointer to percpu data. */
78 static struct acpi_processor_performance __percpu *acpi_perf_data;
79 
80 static struct cpufreq_driver acpi_cpufreq_driver;
81 
82 static unsigned int acpi_pstate_strict;
83 static bool boost_enabled, boost_supported;
84 static struct msr __percpu *msrs;
85 
86 static bool boost_state(unsigned int cpu)
87 {
88 	u32 lo, hi;
89 	u64 msr;
90 
91 	switch (boot_cpu_data.x86_vendor) {
92 	case X86_VENDOR_INTEL:
93 		rdmsr_on_cpu(cpu, MSR_IA32_MISC_ENABLE, &lo, &hi);
94 		msr = lo | ((u64)hi << 32);
95 		return !(msr & MSR_IA32_MISC_ENABLE_TURBO_DISABLE);
96 	case X86_VENDOR_AMD:
97 		rdmsr_on_cpu(cpu, MSR_K7_HWCR, &lo, &hi);
98 		msr = lo | ((u64)hi << 32);
99 		return !(msr & MSR_K7_HWCR_CPB_DIS);
100 	}
101 	return false;
102 }
103 
104 static void boost_set_msrs(bool enable, const struct cpumask *cpumask)
105 {
106 	u32 cpu;
107 	u32 msr_addr;
108 	u64 msr_mask;
109 
110 	switch (boot_cpu_data.x86_vendor) {
111 	case X86_VENDOR_INTEL:
112 		msr_addr = MSR_IA32_MISC_ENABLE;
113 		msr_mask = MSR_IA32_MISC_ENABLE_TURBO_DISABLE;
114 		break;
115 	case X86_VENDOR_AMD:
116 		msr_addr = MSR_K7_HWCR;
117 		msr_mask = MSR_K7_HWCR_CPB_DIS;
118 		break;
119 	default:
120 		return;
121 	}
122 
123 	rdmsr_on_cpus(cpumask, msr_addr, msrs);
124 
125 	for_each_cpu(cpu, cpumask) {
126 		struct msr *reg = per_cpu_ptr(msrs, cpu);
127 		if (enable)
128 			reg->q &= ~msr_mask;
129 		else
130 			reg->q |= msr_mask;
131 	}
132 
133 	wrmsr_on_cpus(cpumask, msr_addr, msrs);
134 }
135 
136 static ssize_t _store_boost(const char *buf, size_t count)
137 {
138 	int ret;
139 	unsigned long val = 0;
140 
141 	if (!boost_supported)
142 		return -EINVAL;
143 
144 	ret = kstrtoul(buf, 10, &val);
145 	if (ret || (val > 1))
146 		return -EINVAL;
147 
148 	if ((val && boost_enabled) || (!val && !boost_enabled))
149 		return count;
150 
151 	get_online_cpus();
152 
153 	boost_set_msrs(val, cpu_online_mask);
154 
155 	put_online_cpus();
156 
157 	boost_enabled = val;
158 	pr_debug("Core Boosting %sabled.\n", val ? "en" : "dis");
159 
160 	return count;
161 }
162 
163 static ssize_t store_global_boost(struct kobject *kobj, struct attribute *attr,
164 				  const char *buf, size_t count)
165 {
166 	return _store_boost(buf, count);
167 }
168 
169 static ssize_t show_global_boost(struct kobject *kobj,
170 				 struct attribute *attr, char *buf)
171 {
172 	return sprintf(buf, "%u\n", boost_enabled);
173 }
174 
175 static struct global_attr global_boost = __ATTR(boost, 0644,
176 						show_global_boost,
177 						store_global_boost);
178 
179 static ssize_t show_freqdomain_cpus(struct cpufreq_policy *policy, char *buf)
180 {
181 	struct acpi_cpufreq_data *data = per_cpu(acfreq_data, policy->cpu);
182 
183 	return cpufreq_show_cpus(data->freqdomain_cpus, buf);
184 }
185 
186 cpufreq_freq_attr_ro(freqdomain_cpus);
187 
188 #ifdef CONFIG_X86_ACPI_CPUFREQ_CPB
189 static ssize_t store_cpb(struct cpufreq_policy *policy, const char *buf,
190 			 size_t count)
191 {
192 	return _store_boost(buf, count);
193 }
194 
195 static ssize_t show_cpb(struct cpufreq_policy *policy, char *buf)
196 {
197 	return sprintf(buf, "%u\n", boost_enabled);
198 }
199 
200 cpufreq_freq_attr_rw(cpb);
201 #endif
202 
203 static int check_est_cpu(unsigned int cpuid)
204 {
205 	struct cpuinfo_x86 *cpu = &cpu_data(cpuid);
206 
207 	return cpu_has(cpu, X86_FEATURE_EST);
208 }
209 
210 static int check_amd_hwpstate_cpu(unsigned int cpuid)
211 {
212 	struct cpuinfo_x86 *cpu = &cpu_data(cpuid);
213 
214 	return cpu_has(cpu, X86_FEATURE_HW_PSTATE);
215 }
216 
217 static unsigned extract_io(u32 value, struct acpi_cpufreq_data *data)
218 {
219 	struct acpi_processor_performance *perf;
220 	int i;
221 
222 	perf = data->acpi_data;
223 
224 	for (i = 0; i < perf->state_count; i++) {
225 		if (value == perf->states[i].status)
226 			return data->freq_table[i].frequency;
227 	}
228 	return 0;
229 }
230 
231 static unsigned extract_msr(u32 msr, struct acpi_cpufreq_data *data)
232 {
233 	int i;
234 	struct acpi_processor_performance *perf;
235 
236 	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
237 		msr &= AMD_MSR_RANGE;
238 	else
239 		msr &= INTEL_MSR_RANGE;
240 
241 	perf = data->acpi_data;
242 
243 	for (i = 0; data->freq_table[i].frequency != CPUFREQ_TABLE_END; i++) {
244 		if (msr == perf->states[data->freq_table[i].driver_data].status)
245 			return data->freq_table[i].frequency;
246 	}
247 	return data->freq_table[0].frequency;
248 }
249 
250 static unsigned extract_freq(u32 val, struct acpi_cpufreq_data *data)
251 {
252 	switch (data->cpu_feature) {
253 	case SYSTEM_INTEL_MSR_CAPABLE:
254 	case SYSTEM_AMD_MSR_CAPABLE:
255 		return extract_msr(val, data);
256 	case SYSTEM_IO_CAPABLE:
257 		return extract_io(val, data);
258 	default:
259 		return 0;
260 	}
261 }
262 
263 struct msr_addr {
264 	u32 reg;
265 };
266 
267 struct io_addr {
268 	u16 port;
269 	u8 bit_width;
270 };
271 
272 struct drv_cmd {
273 	unsigned int type;
274 	const struct cpumask *mask;
275 	union {
276 		struct msr_addr msr;
277 		struct io_addr io;
278 	} addr;
279 	u32 val;
280 };
281 
282 /* Called via smp_call_function_single(), on the target CPU */
283 static void do_drv_read(void *_cmd)
284 {
285 	struct drv_cmd *cmd = _cmd;
286 	u32 h;
287 
288 	switch (cmd->type) {
289 	case SYSTEM_INTEL_MSR_CAPABLE:
290 	case SYSTEM_AMD_MSR_CAPABLE:
291 		rdmsr(cmd->addr.msr.reg, cmd->val, h);
292 		break;
293 	case SYSTEM_IO_CAPABLE:
294 		acpi_os_read_port((acpi_io_address)cmd->addr.io.port,
295 				&cmd->val,
296 				(u32)cmd->addr.io.bit_width);
297 		break;
298 	default:
299 		break;
300 	}
301 }
302 
303 /* Called via smp_call_function_many(), on the target CPUs */
304 static void do_drv_write(void *_cmd)
305 {
306 	struct drv_cmd *cmd = _cmd;
307 	u32 lo, hi;
308 
309 	switch (cmd->type) {
310 	case SYSTEM_INTEL_MSR_CAPABLE:
311 		rdmsr(cmd->addr.msr.reg, lo, hi);
312 		lo = (lo & ~INTEL_MSR_RANGE) | (cmd->val & INTEL_MSR_RANGE);
313 		wrmsr(cmd->addr.msr.reg, lo, hi);
314 		break;
315 	case SYSTEM_AMD_MSR_CAPABLE:
316 		wrmsr(cmd->addr.msr.reg, cmd->val, 0);
317 		break;
318 	case SYSTEM_IO_CAPABLE:
319 		acpi_os_write_port((acpi_io_address)cmd->addr.io.port,
320 				cmd->val,
321 				(u32)cmd->addr.io.bit_width);
322 		break;
323 	default:
324 		break;
325 	}
326 }
327 
328 static void drv_read(struct drv_cmd *cmd)
329 {
330 	int err;
331 	cmd->val = 0;
332 
333 	err = smp_call_function_any(cmd->mask, do_drv_read, cmd, 1);
334 	WARN_ON_ONCE(err);	/* smp_call_function_any() was buggy? */
335 }
336 
337 static void drv_write(struct drv_cmd *cmd)
338 {
339 	int this_cpu;
340 
341 	this_cpu = get_cpu();
342 	if (cpumask_test_cpu(this_cpu, cmd->mask))
343 		do_drv_write(cmd);
344 	smp_call_function_many(cmd->mask, do_drv_write, cmd, 1);
345 	put_cpu();
346 }
347 
348 static u32 get_cur_val(const struct cpumask *mask)
349 {
350 	struct acpi_processor_performance *perf;
351 	struct drv_cmd cmd;
352 
353 	if (unlikely(cpumask_empty(mask)))
354 		return 0;
355 
356 	switch (per_cpu(acfreq_data, cpumask_first(mask))->cpu_feature) {
357 	case SYSTEM_INTEL_MSR_CAPABLE:
358 		cmd.type = SYSTEM_INTEL_MSR_CAPABLE;
359 		cmd.addr.msr.reg = MSR_IA32_PERF_CTL;
360 		break;
361 	case SYSTEM_AMD_MSR_CAPABLE:
362 		cmd.type = SYSTEM_AMD_MSR_CAPABLE;
363 		cmd.addr.msr.reg = MSR_AMD_PERF_CTL;
364 		break;
365 	case SYSTEM_IO_CAPABLE:
366 		cmd.type = SYSTEM_IO_CAPABLE;
367 		perf = per_cpu(acfreq_data, cpumask_first(mask))->acpi_data;
368 		cmd.addr.io.port = perf->control_register.address;
369 		cmd.addr.io.bit_width = perf->control_register.bit_width;
370 		break;
371 	default:
372 		return 0;
373 	}
374 
375 	cmd.mask = mask;
376 	drv_read(&cmd);
377 
378 	pr_debug("get_cur_val = %u\n", cmd.val);
379 
380 	return cmd.val;
381 }
382 
383 static unsigned int get_cur_freq_on_cpu(unsigned int cpu)
384 {
385 	struct acpi_cpufreq_data *data = per_cpu(acfreq_data, cpu);
386 	unsigned int freq;
387 	unsigned int cached_freq;
388 
389 	pr_debug("get_cur_freq_on_cpu (%d)\n", cpu);
390 
391 	if (unlikely(data == NULL ||
392 		     data->acpi_data == NULL || data->freq_table == NULL)) {
393 		return 0;
394 	}
395 
396 	cached_freq = data->freq_table[data->acpi_data->state].frequency;
397 	freq = extract_freq(get_cur_val(cpumask_of(cpu)), data);
398 	if (freq != cached_freq) {
399 		/*
400 		 * The dreaded BIOS frequency change behind our back.
401 		 * Force set the frequency on next target call.
402 		 */
403 		data->resume = 1;
404 	}
405 
406 	pr_debug("cur freq = %u\n", freq);
407 
408 	return freq;
409 }
410 
411 static unsigned int check_freqs(const struct cpumask *mask, unsigned int freq,
412 				struct acpi_cpufreq_data *data)
413 {
414 	unsigned int cur_freq;
415 	unsigned int i;
416 
417 	for (i = 0; i < 100; i++) {
418 		cur_freq = extract_freq(get_cur_val(mask), data);
419 		if (cur_freq == freq)
420 			return 1;
421 		udelay(10);
422 	}
423 	return 0;
424 }
425 
426 static int acpi_cpufreq_target(struct cpufreq_policy *policy,
427 			       unsigned int target_freq, unsigned int relation)
428 {
429 	struct acpi_cpufreq_data *data = per_cpu(acfreq_data, policy->cpu);
430 	struct acpi_processor_performance *perf;
431 	struct cpufreq_freqs freqs;
432 	struct drv_cmd cmd;
433 	unsigned int next_state = 0; /* Index into freq_table */
434 	unsigned int next_perf_state = 0; /* Index into perf table */
435 	int result = 0;
436 
437 	pr_debug("acpi_cpufreq_target %d (%d)\n", target_freq, policy->cpu);
438 
439 	if (unlikely(data == NULL ||
440 	     data->acpi_data == NULL || data->freq_table == NULL)) {
441 		return -ENODEV;
442 	}
443 
444 	perf = data->acpi_data;
445 	result = cpufreq_frequency_table_target(policy,
446 						data->freq_table,
447 						target_freq,
448 						relation, &next_state);
449 	if (unlikely(result)) {
450 		result = -ENODEV;
451 		goto out;
452 	}
453 
454 	next_perf_state = data->freq_table[next_state].driver_data;
455 	if (perf->state == next_perf_state) {
456 		if (unlikely(data->resume)) {
457 			pr_debug("Called after resume, resetting to P%d\n",
458 				next_perf_state);
459 			data->resume = 0;
460 		} else {
461 			pr_debug("Already at target state (P%d)\n",
462 				next_perf_state);
463 			goto out;
464 		}
465 	}
466 
467 	switch (data->cpu_feature) {
468 	case SYSTEM_INTEL_MSR_CAPABLE:
469 		cmd.type = SYSTEM_INTEL_MSR_CAPABLE;
470 		cmd.addr.msr.reg = MSR_IA32_PERF_CTL;
471 		cmd.val = (u32) perf->states[next_perf_state].control;
472 		break;
473 	case SYSTEM_AMD_MSR_CAPABLE:
474 		cmd.type = SYSTEM_AMD_MSR_CAPABLE;
475 		cmd.addr.msr.reg = MSR_AMD_PERF_CTL;
476 		cmd.val = (u32) perf->states[next_perf_state].control;
477 		break;
478 	case SYSTEM_IO_CAPABLE:
479 		cmd.type = SYSTEM_IO_CAPABLE;
480 		cmd.addr.io.port = perf->control_register.address;
481 		cmd.addr.io.bit_width = perf->control_register.bit_width;
482 		cmd.val = (u32) perf->states[next_perf_state].control;
483 		break;
484 	default:
485 		result = -ENODEV;
486 		goto out;
487 	}
488 
489 	/* cpufreq holds the hotplug lock, so we are safe from here on */
490 	if (policy->shared_type != CPUFREQ_SHARED_TYPE_ANY)
491 		cmd.mask = policy->cpus;
492 	else
493 		cmd.mask = cpumask_of(policy->cpu);
494 
495 	freqs.old = perf->states[perf->state].core_frequency * 1000;
496 	freqs.new = data->freq_table[next_state].frequency;
497 	cpufreq_notify_transition(policy, &freqs, CPUFREQ_PRECHANGE);
498 
499 	drv_write(&cmd);
500 
501 	if (acpi_pstate_strict) {
502 		if (!check_freqs(cmd.mask, freqs.new, data)) {
503 			pr_debug("acpi_cpufreq_target failed (%d)\n",
504 				policy->cpu);
505 			result = -EAGAIN;
506 			freqs.new = freqs.old;
507 		}
508 	}
509 
510 	cpufreq_notify_transition(policy, &freqs, CPUFREQ_POSTCHANGE);
511 
512 	if (!result)
513 		perf->state = next_perf_state;
514 
515 out:
516 	return result;
517 }
518 
519 static int acpi_cpufreq_verify(struct cpufreq_policy *policy)
520 {
521 	struct acpi_cpufreq_data *data = per_cpu(acfreq_data, policy->cpu);
522 
523 	pr_debug("acpi_cpufreq_verify\n");
524 
525 	return cpufreq_frequency_table_verify(policy, data->freq_table);
526 }
527 
528 static unsigned long
529 acpi_cpufreq_guess_freq(struct acpi_cpufreq_data *data, unsigned int cpu)
530 {
531 	struct acpi_processor_performance *perf = data->acpi_data;
532 
533 	if (cpu_khz) {
534 		/* search the closest match to cpu_khz */
535 		unsigned int i;
536 		unsigned long freq;
537 		unsigned long freqn = perf->states[0].core_frequency * 1000;
538 
539 		for (i = 0; i < (perf->state_count-1); i++) {
540 			freq = freqn;
541 			freqn = perf->states[i+1].core_frequency * 1000;
542 			if ((2 * cpu_khz) > (freqn + freq)) {
543 				perf->state = i;
544 				return freq;
545 			}
546 		}
547 		perf->state = perf->state_count-1;
548 		return freqn;
549 	} else {
550 		/* assume CPU is at P0... */
551 		perf->state = 0;
552 		return perf->states[0].core_frequency * 1000;
553 	}
554 }
555 
556 static void free_acpi_perf_data(void)
557 {
558 	unsigned int i;
559 
560 	/* Freeing a NULL pointer is OK, and alloc_percpu zeroes. */
561 	for_each_possible_cpu(i)
562 		free_cpumask_var(per_cpu_ptr(acpi_perf_data, i)
563 				 ->shared_cpu_map);
564 	free_percpu(acpi_perf_data);
565 }
566 
567 static int boost_notify(struct notifier_block *nb, unsigned long action,
568 		      void *hcpu)
569 {
570 	unsigned cpu = (long)hcpu;
571 	const struct cpumask *cpumask;
572 
573 	cpumask = get_cpu_mask(cpu);
574 
575 	/*
576 	 * Clear the boost-disable bit on the CPU_DOWN path so that
577 	 * this cpu cannot block the remaining ones from boosting. On
578 	 * the CPU_UP path we simply keep the boost-disable flag in
579 	 * sync with the current global state.
580 	 */
581 
582 	switch (action) {
583 	case CPU_UP_PREPARE:
584 	case CPU_UP_PREPARE_FROZEN:
585 		boost_set_msrs(boost_enabled, cpumask);
586 		break;
587 
588 	case CPU_DOWN_PREPARE:
589 	case CPU_DOWN_PREPARE_FROZEN:
590 		boost_set_msrs(1, cpumask);
591 		break;
592 
593 	default:
594 		break;
595 	}
596 
597 	return NOTIFY_OK;
598 }
599 
600 
601 static struct notifier_block boost_nb = {
602 	.notifier_call          = boost_notify,
603 };
604 
605 /*
606  * acpi_cpufreq_early_init - initialize ACPI P-States library
607  *
608  * Initialize the ACPI P-States library (drivers/acpi/processor_perflib.c)
609  * in order to determine correct frequency and voltage pairings. We can
610  * do _PDC and _PSD and find out the processor dependency for the
611  * actual init that will happen later...
612  */
613 static int __init acpi_cpufreq_early_init(void)
614 {
615 	unsigned int i;
616 	pr_debug("acpi_cpufreq_early_init\n");
617 
618 	acpi_perf_data = alloc_percpu(struct acpi_processor_performance);
619 	if (!acpi_perf_data) {
620 		pr_debug("Memory allocation error for acpi_perf_data.\n");
621 		return -ENOMEM;
622 	}
623 	for_each_possible_cpu(i) {
624 		if (!zalloc_cpumask_var_node(
625 			&per_cpu_ptr(acpi_perf_data, i)->shared_cpu_map,
626 			GFP_KERNEL, cpu_to_node(i))) {
627 
628 			/* Freeing a NULL pointer is OK: alloc_percpu zeroes. */
629 			free_acpi_perf_data();
630 			return -ENOMEM;
631 		}
632 	}
633 
634 	/* Do initialization in ACPI core */
635 	acpi_processor_preregister_performance(acpi_perf_data);
636 	return 0;
637 }
638 
639 #ifdef CONFIG_SMP
640 /*
641  * Some BIOSes do SW_ANY coordination internally, either set it up in hw
642  * or do it in BIOS firmware and won't inform about it to OS. If not
643  * detected, this has a side effect of making CPU run at a different speed
644  * than OS intended it to run at. Detect it and handle it cleanly.
645  */
646 static int bios_with_sw_any_bug;
647 
648 static int sw_any_bug_found(const struct dmi_system_id *d)
649 {
650 	bios_with_sw_any_bug = 1;
651 	return 0;
652 }
653 
654 static const struct dmi_system_id sw_any_bug_dmi_table[] = {
655 	{
656 		.callback = sw_any_bug_found,
657 		.ident = "Supermicro Server X6DLP",
658 		.matches = {
659 			DMI_MATCH(DMI_SYS_VENDOR, "Supermicro"),
660 			DMI_MATCH(DMI_BIOS_VERSION, "080010"),
661 			DMI_MATCH(DMI_PRODUCT_NAME, "X6DLP"),
662 		},
663 	},
664 	{ }
665 };
666 
667 static int acpi_cpufreq_blacklist(struct cpuinfo_x86 *c)
668 {
669 	/* Intel Xeon Processor 7100 Series Specification Update
670 	 * http://www.intel.com/Assets/PDF/specupdate/314554.pdf
671 	 * AL30: A Machine Check Exception (MCE) Occurring during an
672 	 * Enhanced Intel SpeedStep Technology Ratio Change May Cause
673 	 * Both Processor Cores to Lock Up. */
674 	if (c->x86_vendor == X86_VENDOR_INTEL) {
675 		if ((c->x86 == 15) &&
676 		    (c->x86_model == 6) &&
677 		    (c->x86_mask == 8)) {
678 			printk(KERN_INFO "acpi-cpufreq: Intel(R) "
679 			    "Xeon(R) 7100 Errata AL30, processors may "
680 			    "lock up on frequency changes: disabling "
681 			    "acpi-cpufreq.\n");
682 			return -ENODEV;
683 		    }
684 		}
685 	return 0;
686 }
687 #endif
688 
689 static int acpi_cpufreq_cpu_init(struct cpufreq_policy *policy)
690 {
691 	unsigned int i;
692 	unsigned int valid_states = 0;
693 	unsigned int cpu = policy->cpu;
694 	struct acpi_cpufreq_data *data;
695 	unsigned int result = 0;
696 	struct cpuinfo_x86 *c = &cpu_data(policy->cpu);
697 	struct acpi_processor_performance *perf;
698 #ifdef CONFIG_SMP
699 	static int blacklisted;
700 #endif
701 
702 	pr_debug("acpi_cpufreq_cpu_init\n");
703 
704 #ifdef CONFIG_SMP
705 	if (blacklisted)
706 		return blacklisted;
707 	blacklisted = acpi_cpufreq_blacklist(c);
708 	if (blacklisted)
709 		return blacklisted;
710 #endif
711 
712 	data = kzalloc(sizeof(*data), GFP_KERNEL);
713 	if (!data)
714 		return -ENOMEM;
715 
716 	if (!zalloc_cpumask_var(&data->freqdomain_cpus, GFP_KERNEL)) {
717 		result = -ENOMEM;
718 		goto err_free;
719 	}
720 
721 	data->acpi_data = per_cpu_ptr(acpi_perf_data, cpu);
722 	per_cpu(acfreq_data, cpu) = data;
723 
724 	if (cpu_has(c, X86_FEATURE_CONSTANT_TSC))
725 		acpi_cpufreq_driver.flags |= CPUFREQ_CONST_LOOPS;
726 
727 	result = acpi_processor_register_performance(data->acpi_data, cpu);
728 	if (result)
729 		goto err_free_mask;
730 
731 	perf = data->acpi_data;
732 	policy->shared_type = perf->shared_type;
733 
734 	/*
735 	 * Will let policy->cpus know about dependency only when software
736 	 * coordination is required.
737 	 */
738 	if (policy->shared_type == CPUFREQ_SHARED_TYPE_ALL ||
739 	    policy->shared_type == CPUFREQ_SHARED_TYPE_ANY) {
740 		cpumask_copy(policy->cpus, perf->shared_cpu_map);
741 	}
742 	cpumask_copy(data->freqdomain_cpus, perf->shared_cpu_map);
743 
744 #ifdef CONFIG_SMP
745 	dmi_check_system(sw_any_bug_dmi_table);
746 	if (bios_with_sw_any_bug && !policy_is_shared(policy)) {
747 		policy->shared_type = CPUFREQ_SHARED_TYPE_ALL;
748 		cpumask_copy(policy->cpus, cpu_core_mask(cpu));
749 	}
750 
751 	if (check_amd_hwpstate_cpu(cpu) && !acpi_pstate_strict) {
752 		cpumask_clear(policy->cpus);
753 		cpumask_set_cpu(cpu, policy->cpus);
754 		cpumask_copy(data->freqdomain_cpus, cpu_sibling_mask(cpu));
755 		policy->shared_type = CPUFREQ_SHARED_TYPE_HW;
756 		pr_info_once(PFX "overriding BIOS provided _PSD data\n");
757 	}
758 #endif
759 
760 	/* capability check */
761 	if (perf->state_count <= 1) {
762 		pr_debug("No P-States\n");
763 		result = -ENODEV;
764 		goto err_unreg;
765 	}
766 
767 	if (perf->control_register.space_id != perf->status_register.space_id) {
768 		result = -ENODEV;
769 		goto err_unreg;
770 	}
771 
772 	switch (perf->control_register.space_id) {
773 	case ACPI_ADR_SPACE_SYSTEM_IO:
774 		if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
775 		    boot_cpu_data.x86 == 0xf) {
776 			pr_debug("AMD K8 systems must use native drivers.\n");
777 			result = -ENODEV;
778 			goto err_unreg;
779 		}
780 		pr_debug("SYSTEM IO addr space\n");
781 		data->cpu_feature = SYSTEM_IO_CAPABLE;
782 		break;
783 	case ACPI_ADR_SPACE_FIXED_HARDWARE:
784 		pr_debug("HARDWARE addr space\n");
785 		if (check_est_cpu(cpu)) {
786 			data->cpu_feature = SYSTEM_INTEL_MSR_CAPABLE;
787 			break;
788 		}
789 		if (check_amd_hwpstate_cpu(cpu)) {
790 			data->cpu_feature = SYSTEM_AMD_MSR_CAPABLE;
791 			break;
792 		}
793 		result = -ENODEV;
794 		goto err_unreg;
795 	default:
796 		pr_debug("Unknown addr space %d\n",
797 			(u32) (perf->control_register.space_id));
798 		result = -ENODEV;
799 		goto err_unreg;
800 	}
801 
802 	data->freq_table = kmalloc(sizeof(*data->freq_table) *
803 		    (perf->state_count+1), GFP_KERNEL);
804 	if (!data->freq_table) {
805 		result = -ENOMEM;
806 		goto err_unreg;
807 	}
808 
809 	/* detect transition latency */
810 	policy->cpuinfo.transition_latency = 0;
811 	for (i = 0; i < perf->state_count; i++) {
812 		if ((perf->states[i].transition_latency * 1000) >
813 		    policy->cpuinfo.transition_latency)
814 			policy->cpuinfo.transition_latency =
815 			    perf->states[i].transition_latency * 1000;
816 	}
817 
818 	/* Check for high latency (>20uS) from buggy BIOSes, like on T42 */
819 	if (perf->control_register.space_id == ACPI_ADR_SPACE_FIXED_HARDWARE &&
820 	    policy->cpuinfo.transition_latency > 20 * 1000) {
821 		policy->cpuinfo.transition_latency = 20 * 1000;
822 		printk_once(KERN_INFO
823 			    "P-state transition latency capped at 20 uS\n");
824 	}
825 
826 	/* table init */
827 	for (i = 0; i < perf->state_count; i++) {
828 		if (i > 0 && perf->states[i].core_frequency >=
829 		    data->freq_table[valid_states-1].frequency / 1000)
830 			continue;
831 
832 		data->freq_table[valid_states].driver_data = i;
833 		data->freq_table[valid_states].frequency =
834 		    perf->states[i].core_frequency * 1000;
835 		valid_states++;
836 	}
837 	data->freq_table[valid_states].frequency = CPUFREQ_TABLE_END;
838 	perf->state = 0;
839 
840 	result = cpufreq_frequency_table_cpuinfo(policy, data->freq_table);
841 	if (result)
842 		goto err_freqfree;
843 
844 	if (perf->states[0].core_frequency * 1000 != policy->cpuinfo.max_freq)
845 		printk(KERN_WARNING FW_WARN "P-state 0 is not max freq\n");
846 
847 	switch (perf->control_register.space_id) {
848 	case ACPI_ADR_SPACE_SYSTEM_IO:
849 		/* Current speed is unknown and not detectable by IO port */
850 		policy->cur = acpi_cpufreq_guess_freq(data, policy->cpu);
851 		break;
852 	case ACPI_ADR_SPACE_FIXED_HARDWARE:
853 		acpi_cpufreq_driver.get = get_cur_freq_on_cpu;
854 		policy->cur = get_cur_freq_on_cpu(cpu);
855 		break;
856 	default:
857 		break;
858 	}
859 
860 	/* notify BIOS that we exist */
861 	acpi_processor_notify_smm(THIS_MODULE);
862 
863 	pr_debug("CPU%u - ACPI performance management activated.\n", cpu);
864 	for (i = 0; i < perf->state_count; i++)
865 		pr_debug("     %cP%d: %d MHz, %d mW, %d uS\n",
866 			(i == perf->state ? '*' : ' '), i,
867 			(u32) perf->states[i].core_frequency,
868 			(u32) perf->states[i].power,
869 			(u32) perf->states[i].transition_latency);
870 
871 	cpufreq_frequency_table_get_attr(data->freq_table, policy->cpu);
872 
873 	/*
874 	 * the first call to ->target() should result in us actually
875 	 * writing something to the appropriate registers.
876 	 */
877 	data->resume = 1;
878 
879 	return result;
880 
881 err_freqfree:
882 	kfree(data->freq_table);
883 err_unreg:
884 	acpi_processor_unregister_performance(perf, cpu);
885 err_free_mask:
886 	free_cpumask_var(data->freqdomain_cpus);
887 err_free:
888 	kfree(data);
889 	per_cpu(acfreq_data, cpu) = NULL;
890 
891 	return result;
892 }
893 
894 static int acpi_cpufreq_cpu_exit(struct cpufreq_policy *policy)
895 {
896 	struct acpi_cpufreq_data *data = per_cpu(acfreq_data, policy->cpu);
897 
898 	pr_debug("acpi_cpufreq_cpu_exit\n");
899 
900 	if (data) {
901 		cpufreq_frequency_table_put_attr(policy->cpu);
902 		per_cpu(acfreq_data, policy->cpu) = NULL;
903 		acpi_processor_unregister_performance(data->acpi_data,
904 						      policy->cpu);
905 		free_cpumask_var(data->freqdomain_cpus);
906 		kfree(data->freq_table);
907 		kfree(data);
908 	}
909 
910 	return 0;
911 }
912 
913 static int acpi_cpufreq_resume(struct cpufreq_policy *policy)
914 {
915 	struct acpi_cpufreq_data *data = per_cpu(acfreq_data, policy->cpu);
916 
917 	pr_debug("acpi_cpufreq_resume\n");
918 
919 	data->resume = 1;
920 
921 	return 0;
922 }
923 
924 static struct freq_attr *acpi_cpufreq_attr[] = {
925 	&cpufreq_freq_attr_scaling_available_freqs,
926 	&freqdomain_cpus,
927 	NULL,	/* this is a placeholder for cpb, do not remove */
928 	NULL,
929 };
930 
931 static struct cpufreq_driver acpi_cpufreq_driver = {
932 	.verify		= acpi_cpufreq_verify,
933 	.target		= acpi_cpufreq_target,
934 	.bios_limit	= acpi_processor_get_bios_limit,
935 	.init		= acpi_cpufreq_cpu_init,
936 	.exit		= acpi_cpufreq_cpu_exit,
937 	.resume		= acpi_cpufreq_resume,
938 	.name		= "acpi-cpufreq",
939 	.attr		= acpi_cpufreq_attr,
940 };
941 
942 static void __init acpi_cpufreq_boost_init(void)
943 {
944 	if (boot_cpu_has(X86_FEATURE_CPB) || boot_cpu_has(X86_FEATURE_IDA)) {
945 		msrs = msrs_alloc();
946 
947 		if (!msrs)
948 			return;
949 
950 		boost_supported = true;
951 		boost_enabled = boost_state(0);
952 
953 		get_online_cpus();
954 
955 		/* Force all MSRs to the same value */
956 		boost_set_msrs(boost_enabled, cpu_online_mask);
957 
958 		register_cpu_notifier(&boost_nb);
959 
960 		put_online_cpus();
961 	} else
962 		global_boost.attr.mode = 0444;
963 
964 	/* We create the boost file in any case, though for systems without
965 	 * hardware support it will be read-only and hardwired to return 0.
966 	 */
967 	if (cpufreq_sysfs_create_file(&(global_boost.attr)))
968 		pr_warn(PFX "could not register global boost sysfs file\n");
969 	else
970 		pr_debug("registered global boost sysfs file\n");
971 }
972 
973 static void __exit acpi_cpufreq_boost_exit(void)
974 {
975 	cpufreq_sysfs_remove_file(&(global_boost.attr));
976 
977 	if (msrs) {
978 		unregister_cpu_notifier(&boost_nb);
979 
980 		msrs_free(msrs);
981 		msrs = NULL;
982 	}
983 }
984 
985 static int __init acpi_cpufreq_init(void)
986 {
987 	int ret;
988 
989 	if (acpi_disabled)
990 		return -ENODEV;
991 
992 	/* don't keep reloading if cpufreq_driver exists */
993 	if (cpufreq_get_current_driver())
994 		return -EEXIST;
995 
996 	pr_debug("acpi_cpufreq_init\n");
997 
998 	ret = acpi_cpufreq_early_init();
999 	if (ret)
1000 		return ret;
1001 
1002 #ifdef CONFIG_X86_ACPI_CPUFREQ_CPB
1003 	/* this is a sysfs file with a strange name and an even stranger
1004 	 * semantic - per CPU instantiation, but system global effect.
1005 	 * Lets enable it only on AMD CPUs for compatibility reasons and
1006 	 * only if configured. This is considered legacy code, which
1007 	 * will probably be removed at some point in the future.
1008 	 */
1009 	if (check_amd_hwpstate_cpu(0)) {
1010 		struct freq_attr **iter;
1011 
1012 		pr_debug("adding sysfs entry for cpb\n");
1013 
1014 		for (iter = acpi_cpufreq_attr; *iter != NULL; iter++)
1015 			;
1016 
1017 		/* make sure there is a terminator behind it */
1018 		if (iter[1] == NULL)
1019 			*iter = &cpb;
1020 	}
1021 #endif
1022 
1023 	ret = cpufreq_register_driver(&acpi_cpufreq_driver);
1024 	if (ret)
1025 		free_acpi_perf_data();
1026 	else
1027 		acpi_cpufreq_boost_init();
1028 
1029 	return ret;
1030 }
1031 
1032 static void __exit acpi_cpufreq_exit(void)
1033 {
1034 	pr_debug("acpi_cpufreq_exit\n");
1035 
1036 	acpi_cpufreq_boost_exit();
1037 
1038 	cpufreq_unregister_driver(&acpi_cpufreq_driver);
1039 
1040 	free_acpi_perf_data();
1041 }
1042 
1043 module_param(acpi_pstate_strict, uint, 0644);
1044 MODULE_PARM_DESC(acpi_pstate_strict,
1045 	"value 0 or non-zero. non-zero -> strict ACPI checks are "
1046 	"performed during frequency changes.");
1047 
1048 late_initcall(acpi_cpufreq_init);
1049 module_exit(acpi_cpufreq_exit);
1050 
1051 static const struct x86_cpu_id acpi_cpufreq_ids[] = {
1052 	X86_FEATURE_MATCH(X86_FEATURE_ACPI),
1053 	X86_FEATURE_MATCH(X86_FEATURE_HW_PSTATE),
1054 	{}
1055 };
1056 MODULE_DEVICE_TABLE(x86cpu, acpi_cpufreq_ids);
1057 
1058 static const struct acpi_device_id processor_device_ids[] = {
1059 	{ACPI_PROCESSOR_OBJECT_HID, },
1060 	{ACPI_PROCESSOR_DEVICE_HID, },
1061 	{},
1062 };
1063 MODULE_DEVICE_TABLE(acpi, processor_device_ids);
1064 
1065 MODULE_ALIAS("acpi");
1066