On Tue, Nov 06, 2007 at 02:33:53AM -0800, akpm@linux-foundation.org wrote:
[mmotm.git] / drivers / acpi / processor_perflib.c
blob8ba0ed0b9ddbc912684d398e2eda14357920aed4
1 /*
2 * processor_perflib.c - ACPI Processor P-States Library ($Revision: 71 $)
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 Dominik Brodowski <linux@brodo.de>
7 * Copyright (C) 2004 Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
8 * - Added processor hotplug support
11 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
13 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License as published by
15 * the Free Software Foundation; either version 2 of the License, or (at
16 * your option) any later version.
18 * This program is distributed in the hope that it will be useful, but
19 * WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
21 * General Public License for more details.
23 * You should have received a copy of the GNU General Public License along
24 * with this program; if not, write to the Free Software Foundation, Inc.,
25 * 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA.
29 #include <linux/kernel.h>
30 #include <linux/module.h>
31 #include <linux/init.h>
32 #include <linux/cpufreq.h>
34 #ifdef CONFIG_X86
35 #include <asm/cpufeature.h>
36 #endif
38 #include <acpi/acpi_bus.h>
39 #include <acpi/acpi_drivers.h>
40 #include <acpi/processor.h>
42 #define PREFIX "ACPI: "
44 #define ACPI_PROCESSOR_CLASS "processor"
45 #define ACPI_PROCESSOR_FILE_PERFORMANCE "performance"
46 #define _COMPONENT ACPI_PROCESSOR_COMPONENT
47 ACPI_MODULE_NAME("processor_perflib");
49 static DEFINE_MUTEX(performance_mutex);
51 /* Use cpufreq debug layer for _PPC changes. */
52 #define cpufreq_printk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_CORE, \
53 "cpufreq-core", msg)
56 * _PPC support is implemented as a CPUfreq policy notifier:
57 * This means each time a CPUfreq driver registered also with
58 * the ACPI core is asked to change the speed policy, the maximum
59 * value is adjusted so that it is within the platform limit.
61 * Also, when a new platform limit value is detected, the CPUfreq
62 * policy is adjusted accordingly.
65 /* ignore_ppc:
66 * -1 -> cpufreq low level drivers not initialized -> _PSS, etc. not called yet
67 * ignore _PPC
68 * 0 -> cpufreq low level drivers initialized -> consider _PPC values
69 * 1 -> ignore _PPC totally -> forced by user through boot param
71 static int ignore_ppc = -1;
72 module_param(ignore_ppc, int, 0644);
73 MODULE_PARM_DESC(ignore_ppc, "If the frequency of your machine gets wrongly" \
74 "limited by BIOS, this should help");
76 #define PPC_REGISTERED 1
77 #define PPC_IN_USE 2
79 static int acpi_processor_ppc_status;
81 static int acpi_processor_ppc_notifier(struct notifier_block *nb,
82 unsigned long event, void *data)
84 struct cpufreq_policy *policy = data;
85 struct acpi_processor *pr;
86 unsigned int ppc = 0;
88 if (event == CPUFREQ_START && ignore_ppc <= 0) {
89 ignore_ppc = 0;
90 return 0;
93 if (ignore_ppc)
94 return 0;
96 if (event != CPUFREQ_INCOMPATIBLE)
97 return 0;
99 mutex_lock(&performance_mutex);
101 pr = per_cpu(processors, policy->cpu);
102 if (!pr || !pr->performance)
103 goto out;
105 ppc = (unsigned int)pr->performance_platform_limit;
107 if (ppc >= pr->performance->state_count)
108 goto out;
110 cpufreq_verify_within_limits(policy, 0,
111 pr->performance->states[ppc].
112 core_frequency * 1000);
114 out:
115 mutex_unlock(&performance_mutex);
117 return 0;
120 static struct notifier_block acpi_ppc_notifier_block = {
121 .notifier_call = acpi_processor_ppc_notifier,
124 static int acpi_processor_get_platform_limit(struct acpi_processor *pr)
126 acpi_status status = 0;
127 unsigned long long ppc = 0;
130 if (!pr)
131 return -EINVAL;
134 * _PPC indicates the maximum state currently supported by the platform
135 * (e.g. 0 = states 0..n; 1 = states 1..n; etc.
137 status = acpi_evaluate_integer(pr->handle, "_PPC", NULL, &ppc);
139 if (status != AE_NOT_FOUND)
140 acpi_processor_ppc_status |= PPC_IN_USE;
142 if (ACPI_FAILURE(status) && status != AE_NOT_FOUND) {
143 ACPI_EXCEPTION((AE_INFO, status, "Evaluating _PPC"));
144 return -ENODEV;
147 cpufreq_printk("CPU %d: _PPC is %d - frequency %s limited\n", pr->id,
148 (int)ppc, ppc ? "" : "not");
150 pr->performance_platform_limit = (int)ppc;
152 return 0;
155 int acpi_processor_ppc_has_changed(struct acpi_processor *pr)
157 int ret;
159 if (ignore_ppc)
160 return 0;
162 ret = acpi_processor_get_platform_limit(pr);
164 if (ret < 0)
165 return (ret);
166 else
167 return cpufreq_update_policy(pr->id);
170 void acpi_processor_ppc_init(void)
172 if (!cpufreq_register_notifier
173 (&acpi_ppc_notifier_block, CPUFREQ_POLICY_NOTIFIER))
174 acpi_processor_ppc_status |= PPC_REGISTERED;
175 else
176 printk(KERN_DEBUG
177 "Warning: Processor Platform Limit not supported.\n");
180 void acpi_processor_ppc_exit(void)
182 if (acpi_processor_ppc_status & PPC_REGISTERED)
183 cpufreq_unregister_notifier(&acpi_ppc_notifier_block,
184 CPUFREQ_POLICY_NOTIFIER);
186 acpi_processor_ppc_status &= ~PPC_REGISTERED;
189 static int acpi_processor_get_performance_control(struct acpi_processor *pr)
191 int result = 0;
192 acpi_status status = 0;
193 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
194 union acpi_object *pct = NULL;
195 union acpi_object obj = { 0 };
198 status = acpi_evaluate_object(pr->handle, "_PCT", NULL, &buffer);
199 if (ACPI_FAILURE(status)) {
200 ACPI_EXCEPTION((AE_INFO, status, "Evaluating _PCT"));
201 return -ENODEV;
204 pct = (union acpi_object *)buffer.pointer;
205 if (!pct || (pct->type != ACPI_TYPE_PACKAGE)
206 || (pct->package.count != 2)) {
207 printk(KERN_ERR PREFIX "Invalid _PCT data\n");
208 result = -EFAULT;
209 goto end;
213 * control_register
216 obj = pct->package.elements[0];
218 if ((obj.type != ACPI_TYPE_BUFFER)
219 || (obj.buffer.length < sizeof(struct acpi_pct_register))
220 || (obj.buffer.pointer == NULL)) {
221 printk(KERN_ERR PREFIX "Invalid _PCT data (control_register)\n");
222 result = -EFAULT;
223 goto end;
225 memcpy(&pr->performance->control_register, obj.buffer.pointer,
226 sizeof(struct acpi_pct_register));
229 * status_register
232 obj = pct->package.elements[1];
234 if ((obj.type != ACPI_TYPE_BUFFER)
235 || (obj.buffer.length < sizeof(struct acpi_pct_register))
236 || (obj.buffer.pointer == NULL)) {
237 printk(KERN_ERR PREFIX "Invalid _PCT data (status_register)\n");
238 result = -EFAULT;
239 goto end;
242 memcpy(&pr->performance->status_register, obj.buffer.pointer,
243 sizeof(struct acpi_pct_register));
245 end:
246 kfree(buffer.pointer);
248 return result;
251 static int acpi_processor_get_performance_states(struct acpi_processor *pr)
253 int result = 0;
254 acpi_status status = AE_OK;
255 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
256 struct acpi_buffer format = { sizeof("NNNNNN"), "NNNNNN" };
257 struct acpi_buffer state = { 0, NULL };
258 union acpi_object *pss = NULL;
259 int i;
262 status = acpi_evaluate_object(pr->handle, "_PSS", NULL, &buffer);
263 if (ACPI_FAILURE(status)) {
264 ACPI_EXCEPTION((AE_INFO, status, "Evaluating _PSS"));
265 return -ENODEV;
268 pss = buffer.pointer;
269 if (!pss || (pss->type != ACPI_TYPE_PACKAGE)) {
270 printk(KERN_ERR PREFIX "Invalid _PSS data\n");
271 result = -EFAULT;
272 goto end;
275 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d performance states\n",
276 pss->package.count));
278 pr->performance->state_count = pss->package.count;
279 pr->performance->states =
280 kmalloc(sizeof(struct acpi_processor_px) * pss->package.count,
281 GFP_KERNEL);
282 if (!pr->performance->states) {
283 result = -ENOMEM;
284 goto end;
287 for (i = 0; i < pr->performance->state_count; i++) {
289 struct acpi_processor_px *px = &(pr->performance->states[i]);
291 state.length = sizeof(struct acpi_processor_px);
292 state.pointer = px;
294 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Extracting state %d\n", i));
296 status = acpi_extract_package(&(pss->package.elements[i]),
297 &format, &state);
298 if (ACPI_FAILURE(status)) {
299 ACPI_EXCEPTION((AE_INFO, status, "Invalid _PSS data"));
300 result = -EFAULT;
301 kfree(pr->performance->states);
302 goto end;
305 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
306 "State [%d]: core_frequency[%d] power[%d] transition_latency[%d] bus_master_latency[%d] control[0x%x] status[0x%x]\n",
308 (u32) px->core_frequency,
309 (u32) px->power,
310 (u32) px->transition_latency,
311 (u32) px->bus_master_latency,
312 (u32) px->control, (u32) px->status));
315 * Check that ACPI's u64 MHz will be valid as u32 KHz in cpufreq
317 if (!px->core_frequency ||
318 ((u32)(px->core_frequency * 1000) !=
319 (px->core_frequency * 1000))) {
320 printk(KERN_ERR FW_BUG PREFIX
321 "Invalid BIOS _PSS frequency: 0x%llx MHz\n",
322 px->core_frequency);
323 result = -EFAULT;
324 kfree(pr->performance->states);
325 goto end;
329 end:
330 kfree(buffer.pointer);
332 return result;
335 static int acpi_processor_get_performance_info(struct acpi_processor *pr)
337 int result = 0;
338 acpi_status status = AE_OK;
339 acpi_handle handle = NULL;
341 if (!pr || !pr->performance || !pr->handle)
342 return -EINVAL;
344 status = acpi_get_handle(pr->handle, "_PCT", &handle);
345 if (ACPI_FAILURE(status)) {
346 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
347 "ACPI-based processor performance control unavailable\n"));
348 return -ENODEV;
351 result = acpi_processor_get_performance_control(pr);
352 if (result)
353 goto update_bios;
355 result = acpi_processor_get_performance_states(pr);
356 if (result)
357 goto update_bios;
359 return 0;
362 * Having _PPC but missing frequencies (_PSS, _PCT) is a very good hint that
363 * the BIOS is older than the CPU and does not know its frequencies
365 update_bios:
366 #ifdef CONFIG_X86
367 if (ACPI_SUCCESS(acpi_get_handle(pr->handle, "_PPC", &handle))){
368 if(boot_cpu_has(X86_FEATURE_EST))
369 printk(KERN_WARNING FW_BUG "BIOS needs update for CPU "
370 "frequency support\n");
372 #endif
373 return result;
376 int acpi_processor_notify_smm(struct module *calling_module)
378 acpi_status status;
379 static int is_done = 0;
382 if (!(acpi_processor_ppc_status & PPC_REGISTERED))
383 return -EBUSY;
385 if (!try_module_get(calling_module))
386 return -EINVAL;
388 /* is_done is set to negative if an error occured,
389 * and to postitive if _no_ error occured, but SMM
390 * was already notified. This avoids double notification
391 * which might lead to unexpected results...
393 if (is_done > 0) {
394 module_put(calling_module);
395 return 0;
396 } else if (is_done < 0) {
397 module_put(calling_module);
398 return is_done;
401 is_done = -EIO;
403 /* Can't write pstate_control to smi_command if either value is zero */
404 if ((!acpi_gbl_FADT.smi_command) || (!acpi_gbl_FADT.pstate_control)) {
405 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No SMI port or pstate_control\n"));
406 module_put(calling_module);
407 return 0;
410 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
411 "Writing pstate_control [0x%x] to smi_command [0x%x]\n",
412 acpi_gbl_FADT.pstate_control, acpi_gbl_FADT.smi_command));
414 status = acpi_os_write_port(acpi_gbl_FADT.smi_command,
415 (u32) acpi_gbl_FADT.pstate_control, 8);
416 if (ACPI_FAILURE(status)) {
417 ACPI_EXCEPTION((AE_INFO, status,
418 "Failed to write pstate_control [0x%x] to "
419 "smi_command [0x%x]", acpi_gbl_FADT.pstate_control,
420 acpi_gbl_FADT.smi_command));
421 module_put(calling_module);
422 return status;
425 /* Success. If there's no _PPC, we need to fear nothing, so
426 * we can allow the cpufreq driver to be rmmod'ed. */
427 is_done = 1;
429 if (!(acpi_processor_ppc_status & PPC_IN_USE))
430 module_put(calling_module);
432 return 0;
435 EXPORT_SYMBOL(acpi_processor_notify_smm);
437 static int acpi_processor_get_psd(struct acpi_processor *pr)
439 int result = 0;
440 acpi_status status = AE_OK;
441 struct acpi_buffer buffer = {ACPI_ALLOCATE_BUFFER, NULL};
442 struct acpi_buffer format = {sizeof("NNNNN"), "NNNNN"};
443 struct acpi_buffer state = {0, NULL};
444 union acpi_object *psd = NULL;
445 struct acpi_psd_package *pdomain;
447 status = acpi_evaluate_object(pr->handle, "_PSD", NULL, &buffer);
448 if (ACPI_FAILURE(status)) {
449 return -ENODEV;
452 psd = buffer.pointer;
453 if (!psd || (psd->type != ACPI_TYPE_PACKAGE)) {
454 printk(KERN_ERR PREFIX "Invalid _PSD data\n");
455 result = -EFAULT;
456 goto end;
459 if (psd->package.count != 1) {
460 printk(KERN_ERR PREFIX "Invalid _PSD data\n");
461 result = -EFAULT;
462 goto end;
465 pdomain = &(pr->performance->domain_info);
467 state.length = sizeof(struct acpi_psd_package);
468 state.pointer = pdomain;
470 status = acpi_extract_package(&(psd->package.elements[0]),
471 &format, &state);
472 if (ACPI_FAILURE(status)) {
473 printk(KERN_ERR PREFIX "Invalid _PSD data\n");
474 result = -EFAULT;
475 goto end;
478 if (pdomain->num_entries != ACPI_PSD_REV0_ENTRIES) {
479 printk(KERN_ERR PREFIX "Unknown _PSD:num_entries\n");
480 result = -EFAULT;
481 goto end;
484 if (pdomain->revision != ACPI_PSD_REV0_REVISION) {
485 printk(KERN_ERR PREFIX "Unknown _PSD:revision\n");
486 result = -EFAULT;
487 goto end;
490 if (pdomain->coord_type != DOMAIN_COORD_TYPE_SW_ALL &&
491 pdomain->coord_type != DOMAIN_COORD_TYPE_SW_ANY &&
492 pdomain->coord_type != DOMAIN_COORD_TYPE_HW_ALL) {
493 printk(KERN_ERR PREFIX "Invalid _PSD:coord_type\n");
494 result = -EFAULT;
495 goto end;
497 end:
498 kfree(buffer.pointer);
499 return result;
502 int acpi_processor_preregister_performance(
503 struct acpi_processor_performance *performance)
505 int count, count_target;
506 int retval = 0;
507 unsigned int i, j;
508 cpumask_var_t covered_cpus;
509 struct acpi_processor *pr;
510 struct acpi_psd_package *pdomain;
511 struct acpi_processor *match_pr;
512 struct acpi_psd_package *match_pdomain;
514 if (!zalloc_cpumask_var(&covered_cpus, GFP_KERNEL))
515 return -ENOMEM;
517 mutex_lock(&performance_mutex);
520 * Check if another driver has already registered, and abort before
521 * changing pr->performance if it has. Check input data as well.
523 for_each_possible_cpu(i) {
524 pr = per_cpu(processors, i);
525 if (!pr) {
526 /* Look only at processors in ACPI namespace */
527 continue;
530 if (pr->performance) {
531 retval = -EBUSY;
532 goto err_out;
535 if (!performance || !per_cpu_ptr(performance, i)) {
536 retval = -EINVAL;
537 goto err_out;
541 /* Call _PSD for all CPUs */
542 for_each_possible_cpu(i) {
543 pr = per_cpu(processors, i);
544 if (!pr)
545 continue;
547 pr->performance = per_cpu_ptr(performance, i);
548 cpumask_set_cpu(i, pr->performance->shared_cpu_map);
549 if (acpi_processor_get_psd(pr)) {
550 retval = -EINVAL;
551 continue;
554 if (retval)
555 goto err_ret;
558 * Now that we have _PSD data from all CPUs, lets setup P-state
559 * domain info.
561 for_each_possible_cpu(i) {
562 pr = per_cpu(processors, i);
563 if (!pr)
564 continue;
566 if (cpumask_test_cpu(i, covered_cpus))
567 continue;
569 pdomain = &(pr->performance->domain_info);
570 cpumask_set_cpu(i, pr->performance->shared_cpu_map);
571 cpumask_set_cpu(i, covered_cpus);
572 if (pdomain->num_processors <= 1)
573 continue;
575 /* Validate the Domain info */
576 count_target = pdomain->num_processors;
577 count = 1;
578 if (pdomain->coord_type == DOMAIN_COORD_TYPE_SW_ALL)
579 pr->performance->shared_type = CPUFREQ_SHARED_TYPE_ALL;
580 else if (pdomain->coord_type == DOMAIN_COORD_TYPE_HW_ALL)
581 pr->performance->shared_type = CPUFREQ_SHARED_TYPE_HW;
582 else if (pdomain->coord_type == DOMAIN_COORD_TYPE_SW_ANY)
583 pr->performance->shared_type = CPUFREQ_SHARED_TYPE_ANY;
585 for_each_possible_cpu(j) {
586 if (i == j)
587 continue;
589 match_pr = per_cpu(processors, j);
590 if (!match_pr)
591 continue;
593 match_pdomain = &(match_pr->performance->domain_info);
594 if (match_pdomain->domain != pdomain->domain)
595 continue;
597 /* Here i and j are in the same domain */
599 if (match_pdomain->num_processors != count_target) {
600 retval = -EINVAL;
601 goto err_ret;
604 if (pdomain->coord_type != match_pdomain->coord_type) {
605 retval = -EINVAL;
606 goto err_ret;
609 cpumask_set_cpu(j, covered_cpus);
610 cpumask_set_cpu(j, pr->performance->shared_cpu_map);
611 count++;
614 for_each_possible_cpu(j) {
615 if (i == j)
616 continue;
618 match_pr = per_cpu(processors, j);
619 if (!match_pr)
620 continue;
622 match_pdomain = &(match_pr->performance->domain_info);
623 if (match_pdomain->domain != pdomain->domain)
624 continue;
626 match_pr->performance->shared_type =
627 pr->performance->shared_type;
628 cpumask_copy(match_pr->performance->shared_cpu_map,
629 pr->performance->shared_cpu_map);
633 err_ret:
634 for_each_possible_cpu(i) {
635 pr = per_cpu(processors, i);
636 if (!pr || !pr->performance)
637 continue;
639 /* Assume no coordination on any error parsing domain info */
640 if (retval) {
641 cpumask_clear(pr->performance->shared_cpu_map);
642 cpumask_set_cpu(i, pr->performance->shared_cpu_map);
643 pr->performance->shared_type = CPUFREQ_SHARED_TYPE_ALL;
645 pr->performance = NULL; /* Will be set for real in register */
648 err_out:
649 mutex_unlock(&performance_mutex);
650 free_cpumask_var(covered_cpus);
651 return retval;
653 EXPORT_SYMBOL(acpi_processor_preregister_performance);
656 acpi_processor_register_performance(struct acpi_processor_performance
657 *performance, unsigned int cpu)
659 struct acpi_processor *pr;
661 if (!(acpi_processor_ppc_status & PPC_REGISTERED))
662 return -EINVAL;
664 mutex_lock(&performance_mutex);
666 pr = per_cpu(processors, cpu);
667 if (!pr) {
668 mutex_unlock(&performance_mutex);
669 return -ENODEV;
672 if (pr->performance) {
673 mutex_unlock(&performance_mutex);
674 return -EBUSY;
677 WARN_ON(!performance);
679 pr->performance = performance;
681 if (acpi_processor_get_performance_info(pr)) {
682 pr->performance = NULL;
683 mutex_unlock(&performance_mutex);
684 return -EIO;
687 mutex_unlock(&performance_mutex);
688 return 0;
691 EXPORT_SYMBOL(acpi_processor_register_performance);
693 void
694 acpi_processor_unregister_performance(struct acpi_processor_performance
695 *performance, unsigned int cpu)
697 struct acpi_processor *pr;
699 mutex_lock(&performance_mutex);
701 pr = per_cpu(processors, cpu);
702 if (!pr) {
703 mutex_unlock(&performance_mutex);
704 return;
707 if (pr->performance)
708 kfree(pr->performance->states);
709 pr->performance = NULL;
711 mutex_unlock(&performance_mutex);
713 return;
716 EXPORT_SYMBOL(acpi_processor_unregister_performance);