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>
33 #include <linux/slab.h>
36 #include <asm/cpufeature.h>
39 #include <acpi/acpi_bus.h>
40 #include <acpi/acpi_drivers.h>
41 #include <acpi/processor.h>
43 #define PREFIX "ACPI: "
45 #define ACPI_PROCESSOR_CLASS "processor"
46 #define ACPI_PROCESSOR_FILE_PERFORMANCE "performance"
47 #define _COMPONENT ACPI_PROCESSOR_COMPONENT
48 ACPI_MODULE_NAME("processor_perflib");
50 static DEFINE_MUTEX(performance_mutex
);
52 /* Use cpufreq debug layer for _PPC changes. */
53 #define cpufreq_printk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_CORE, \
57 * _PPC support is implemented as a CPUfreq policy notifier:
58 * This means each time a CPUfreq driver registered also with
59 * the ACPI core is asked to change the speed policy, the maximum
60 * value is adjusted so that it is within the platform limit.
62 * Also, when a new platform limit value is detected, the CPUfreq
63 * policy is adjusted accordingly.
67 * -1 -> cpufreq low level drivers not initialized -> _PSS, etc. not called yet
69 * 0 -> cpufreq low level drivers initialized -> consider _PPC values
70 * 1 -> ignore _PPC totally -> forced by user through boot param
72 static int ignore_ppc
= -1;
73 module_param(ignore_ppc
, int, 0644);
74 MODULE_PARM_DESC(ignore_ppc
, "If the frequency of your machine gets wrongly" \
75 "limited by BIOS, this should help");
77 #define PPC_REGISTERED 1
80 static int acpi_processor_ppc_status
;
82 static int acpi_processor_ppc_notifier(struct notifier_block
*nb
,
83 unsigned long event
, void *data
)
85 struct cpufreq_policy
*policy
= data
;
86 struct acpi_processor
*pr
;
89 if (event
== CPUFREQ_START
&& ignore_ppc
<= 0) {
97 if (event
!= CPUFREQ_INCOMPATIBLE
)
100 mutex_lock(&performance_mutex
);
102 pr
= per_cpu(processors
, policy
->cpu
);
103 if (!pr
|| !pr
->performance
)
106 ppc
= (unsigned int)pr
->performance_platform_limit
;
108 if (ppc
>= pr
->performance
->state_count
)
111 cpufreq_verify_within_limits(policy
, 0,
112 pr
->performance
->states
[ppc
].
113 core_frequency
* 1000);
116 mutex_unlock(&performance_mutex
);
121 static struct notifier_block acpi_ppc_notifier_block
= {
122 .notifier_call
= acpi_processor_ppc_notifier
,
125 static int acpi_processor_get_platform_limit(struct acpi_processor
*pr
)
127 acpi_status status
= 0;
128 unsigned long long ppc
= 0;
135 * _PPC indicates the maximum state currently supported by the platform
136 * (e.g. 0 = states 0..n; 1 = states 1..n; etc.
138 status
= acpi_evaluate_integer(pr
->handle
, "_PPC", NULL
, &ppc
);
140 if (status
!= AE_NOT_FOUND
)
141 acpi_processor_ppc_status
|= PPC_IN_USE
;
143 if (ACPI_FAILURE(status
) && status
!= AE_NOT_FOUND
) {
144 ACPI_EXCEPTION((AE_INFO
, status
, "Evaluating _PPC"));
148 cpufreq_printk("CPU %d: _PPC is %d - frequency %s limited\n", pr
->id
,
149 (int)ppc
, ppc
? "" : "not");
151 pr
->performance_platform_limit
= (int)ppc
;
156 #define ACPI_PROCESSOR_NOTIFY_PERFORMANCE 0x80
158 * acpi_processor_ppc_ost: Notify firmware the _PPC evaluation status
159 * @handle: ACPI processor handle
160 * @status: the status code of _PPC evaluation
161 * 0: success. OSPM is now using the performance state specificed.
162 * 1: failure. OSPM has not changed the number of P-states in use
164 static void acpi_processor_ppc_ost(acpi_handle handle
, int status
)
166 union acpi_object params
[2] = {
167 {.type
= ACPI_TYPE_INTEGER
,},
168 {.type
= ACPI_TYPE_INTEGER
,},
170 struct acpi_object_list arg_list
= {2, params
};
173 params
[0].integer
.value
= ACPI_PROCESSOR_NOTIFY_PERFORMANCE
;
174 params
[1].integer
.value
= status
;
176 /* when there is no _OST , skip it */
177 if (ACPI_FAILURE(acpi_get_handle(handle
, "_OST", &temp
)))
180 acpi_evaluate_object(handle
, "_OST", &arg_list
, NULL
);
184 int acpi_processor_ppc_has_changed(struct acpi_processor
*pr
, int event_flag
)
190 * Only when it is notification event, the _OST object
191 * will be evaluated. Otherwise it is skipped.
194 acpi_processor_ppc_ost(pr
->handle
, 1);
198 ret
= acpi_processor_get_platform_limit(pr
);
200 * Only when it is notification event, the _OST object
201 * will be evaluated. Otherwise it is skipped.
205 acpi_processor_ppc_ost(pr
->handle
, 1);
207 acpi_processor_ppc_ost(pr
->handle
, 0);
212 return cpufreq_update_policy(pr
->id
);
215 int acpi_processor_get_bios_limit(int cpu
, unsigned int *limit
)
217 struct acpi_processor
*pr
;
219 pr
= per_cpu(processors
, cpu
);
220 if (!pr
|| !pr
->performance
|| !pr
->performance
->state_count
)
222 *limit
= pr
->performance
->states
[pr
->performance_platform_limit
].
223 core_frequency
* 1000;
226 EXPORT_SYMBOL(acpi_processor_get_bios_limit
);
228 void acpi_processor_ppc_init(void)
230 if (!cpufreq_register_notifier
231 (&acpi_ppc_notifier_block
, CPUFREQ_POLICY_NOTIFIER
))
232 acpi_processor_ppc_status
|= PPC_REGISTERED
;
235 "Warning: Processor Platform Limit not supported.\n");
238 void acpi_processor_ppc_exit(void)
240 if (acpi_processor_ppc_status
& PPC_REGISTERED
)
241 cpufreq_unregister_notifier(&acpi_ppc_notifier_block
,
242 CPUFREQ_POLICY_NOTIFIER
);
244 acpi_processor_ppc_status
&= ~PPC_REGISTERED
;
247 static int acpi_processor_get_performance_control(struct acpi_processor
*pr
)
250 acpi_status status
= 0;
251 struct acpi_buffer buffer
= { ACPI_ALLOCATE_BUFFER
, NULL
};
252 union acpi_object
*pct
= NULL
;
253 union acpi_object obj
= { 0 };
256 status
= acpi_evaluate_object(pr
->handle
, "_PCT", NULL
, &buffer
);
257 if (ACPI_FAILURE(status
)) {
258 ACPI_EXCEPTION((AE_INFO
, status
, "Evaluating _PCT"));
262 pct
= (union acpi_object
*)buffer
.pointer
;
263 if (!pct
|| (pct
->type
!= ACPI_TYPE_PACKAGE
)
264 || (pct
->package
.count
!= 2)) {
265 printk(KERN_ERR PREFIX
"Invalid _PCT data\n");
274 obj
= pct
->package
.elements
[0];
276 if ((obj
.type
!= ACPI_TYPE_BUFFER
)
277 || (obj
.buffer
.length
< sizeof(struct acpi_pct_register
))
278 || (obj
.buffer
.pointer
== NULL
)) {
279 printk(KERN_ERR PREFIX
"Invalid _PCT data (control_register)\n");
283 memcpy(&pr
->performance
->control_register
, obj
.buffer
.pointer
,
284 sizeof(struct acpi_pct_register
));
290 obj
= pct
->package
.elements
[1];
292 if ((obj
.type
!= ACPI_TYPE_BUFFER
)
293 || (obj
.buffer
.length
< sizeof(struct acpi_pct_register
))
294 || (obj
.buffer
.pointer
== NULL
)) {
295 printk(KERN_ERR PREFIX
"Invalid _PCT data (status_register)\n");
300 memcpy(&pr
->performance
->status_register
, obj
.buffer
.pointer
,
301 sizeof(struct acpi_pct_register
));
304 kfree(buffer
.pointer
);
309 static int acpi_processor_get_performance_states(struct acpi_processor
*pr
)
312 acpi_status status
= AE_OK
;
313 struct acpi_buffer buffer
= { ACPI_ALLOCATE_BUFFER
, NULL
};
314 struct acpi_buffer format
= { sizeof("NNNNNN"), "NNNNNN" };
315 struct acpi_buffer state
= { 0, NULL
};
316 union acpi_object
*pss
= NULL
;
320 status
= acpi_evaluate_object(pr
->handle
, "_PSS", NULL
, &buffer
);
321 if (ACPI_FAILURE(status
)) {
322 ACPI_EXCEPTION((AE_INFO
, status
, "Evaluating _PSS"));
326 pss
= buffer
.pointer
;
327 if (!pss
|| (pss
->type
!= ACPI_TYPE_PACKAGE
)) {
328 printk(KERN_ERR PREFIX
"Invalid _PSS data\n");
333 ACPI_DEBUG_PRINT((ACPI_DB_INFO
, "Found %d performance states\n",
334 pss
->package
.count
));
336 pr
->performance
->state_count
= pss
->package
.count
;
337 pr
->performance
->states
=
338 kmalloc(sizeof(struct acpi_processor_px
) * pss
->package
.count
,
340 if (!pr
->performance
->states
) {
345 for (i
= 0; i
< pr
->performance
->state_count
; i
++) {
347 struct acpi_processor_px
*px
= &(pr
->performance
->states
[i
]);
349 state
.length
= sizeof(struct acpi_processor_px
);
352 ACPI_DEBUG_PRINT((ACPI_DB_INFO
, "Extracting state %d\n", i
));
354 status
= acpi_extract_package(&(pss
->package
.elements
[i
]),
356 if (ACPI_FAILURE(status
)) {
357 ACPI_EXCEPTION((AE_INFO
, status
, "Invalid _PSS data"));
359 kfree(pr
->performance
->states
);
363 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
364 "State [%d]: core_frequency[%d] power[%d] transition_latency[%d] bus_master_latency[%d] control[0x%x] status[0x%x]\n",
366 (u32
) px
->core_frequency
,
368 (u32
) px
->transition_latency
,
369 (u32
) px
->bus_master_latency
,
370 (u32
) px
->control
, (u32
) px
->status
));
373 * Check that ACPI's u64 MHz will be valid as u32 KHz in cpufreq
375 if (!px
->core_frequency
||
376 ((u32
)(px
->core_frequency
* 1000) !=
377 (px
->core_frequency
* 1000))) {
378 printk(KERN_ERR FW_BUG PREFIX
379 "Invalid BIOS _PSS frequency: 0x%llx MHz\n",
382 kfree(pr
->performance
->states
);
388 kfree(buffer
.pointer
);
393 static int acpi_processor_get_performance_info(struct acpi_processor
*pr
)
396 acpi_status status
= AE_OK
;
397 acpi_handle handle
= NULL
;
399 if (!pr
|| !pr
->performance
|| !pr
->handle
)
402 status
= acpi_get_handle(pr
->handle
, "_PCT", &handle
);
403 if (ACPI_FAILURE(status
)) {
404 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
405 "ACPI-based processor performance control unavailable\n"));
409 result
= acpi_processor_get_performance_control(pr
);
413 result
= acpi_processor_get_performance_states(pr
);
417 /* We need to call _PPC once when cpufreq starts */
419 result
= acpi_processor_get_platform_limit(pr
);
424 * Having _PPC but missing frequencies (_PSS, _PCT) is a very good hint that
425 * the BIOS is older than the CPU and does not know its frequencies
429 if (ACPI_SUCCESS(acpi_get_handle(pr
->handle
, "_PPC", &handle
))){
430 if(boot_cpu_has(X86_FEATURE_EST
))
431 printk(KERN_WARNING FW_BUG
"BIOS needs update for CPU "
432 "frequency support\n");
438 int acpi_processor_notify_smm(struct module
*calling_module
)
441 static int is_done
= 0;
444 if (!(acpi_processor_ppc_status
& PPC_REGISTERED
))
447 if (!try_module_get(calling_module
))
450 /* is_done is set to negative if an error occured,
451 * and to postitive if _no_ error occured, but SMM
452 * was already notified. This avoids double notification
453 * which might lead to unexpected results...
456 module_put(calling_module
);
458 } else if (is_done
< 0) {
459 module_put(calling_module
);
465 /* Can't write pstate_control to smi_command if either value is zero */
466 if ((!acpi_gbl_FADT
.smi_command
) || (!acpi_gbl_FADT
.pstate_control
)) {
467 ACPI_DEBUG_PRINT((ACPI_DB_INFO
, "No SMI port or pstate_control\n"));
468 module_put(calling_module
);
472 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
473 "Writing pstate_control [0x%x] to smi_command [0x%x]\n",
474 acpi_gbl_FADT
.pstate_control
, acpi_gbl_FADT
.smi_command
));
476 status
= acpi_os_write_port(acpi_gbl_FADT
.smi_command
,
477 (u32
) acpi_gbl_FADT
.pstate_control
, 8);
478 if (ACPI_FAILURE(status
)) {
479 ACPI_EXCEPTION((AE_INFO
, status
,
480 "Failed to write pstate_control [0x%x] to "
481 "smi_command [0x%x]", acpi_gbl_FADT
.pstate_control
,
482 acpi_gbl_FADT
.smi_command
));
483 module_put(calling_module
);
487 /* Success. If there's no _PPC, we need to fear nothing, so
488 * we can allow the cpufreq driver to be rmmod'ed. */
491 if (!(acpi_processor_ppc_status
& PPC_IN_USE
))
492 module_put(calling_module
);
497 EXPORT_SYMBOL(acpi_processor_notify_smm
);
499 static int acpi_processor_get_psd(struct acpi_processor
*pr
)
502 acpi_status status
= AE_OK
;
503 struct acpi_buffer buffer
= {ACPI_ALLOCATE_BUFFER
, NULL
};
504 struct acpi_buffer format
= {sizeof("NNNNN"), "NNNNN"};
505 struct acpi_buffer state
= {0, NULL
};
506 union acpi_object
*psd
= NULL
;
507 struct acpi_psd_package
*pdomain
;
509 status
= acpi_evaluate_object(pr
->handle
, "_PSD", NULL
, &buffer
);
510 if (ACPI_FAILURE(status
)) {
514 psd
= buffer
.pointer
;
515 if (!psd
|| (psd
->type
!= ACPI_TYPE_PACKAGE
)) {
516 printk(KERN_ERR PREFIX
"Invalid _PSD data\n");
521 if (psd
->package
.count
!= 1) {
522 printk(KERN_ERR PREFIX
"Invalid _PSD data\n");
527 pdomain
= &(pr
->performance
->domain_info
);
529 state
.length
= sizeof(struct acpi_psd_package
);
530 state
.pointer
= pdomain
;
532 status
= acpi_extract_package(&(psd
->package
.elements
[0]),
534 if (ACPI_FAILURE(status
)) {
535 printk(KERN_ERR PREFIX
"Invalid _PSD data\n");
540 if (pdomain
->num_entries
!= ACPI_PSD_REV0_ENTRIES
) {
541 printk(KERN_ERR PREFIX
"Unknown _PSD:num_entries\n");
546 if (pdomain
->revision
!= ACPI_PSD_REV0_REVISION
) {
547 printk(KERN_ERR PREFIX
"Unknown _PSD:revision\n");
552 if (pdomain
->coord_type
!= DOMAIN_COORD_TYPE_SW_ALL
&&
553 pdomain
->coord_type
!= DOMAIN_COORD_TYPE_SW_ANY
&&
554 pdomain
->coord_type
!= DOMAIN_COORD_TYPE_HW_ALL
) {
555 printk(KERN_ERR PREFIX
"Invalid _PSD:coord_type\n");
560 kfree(buffer
.pointer
);
564 int acpi_processor_preregister_performance(
565 struct acpi_processor_performance __percpu
*performance
)
567 int count
, count_target
;
570 cpumask_var_t covered_cpus
;
571 struct acpi_processor
*pr
;
572 struct acpi_psd_package
*pdomain
;
573 struct acpi_processor
*match_pr
;
574 struct acpi_psd_package
*match_pdomain
;
576 if (!zalloc_cpumask_var(&covered_cpus
, GFP_KERNEL
))
579 mutex_lock(&performance_mutex
);
582 * Check if another driver has already registered, and abort before
583 * changing pr->performance if it has. Check input data as well.
585 for_each_possible_cpu(i
) {
586 pr
= per_cpu(processors
, i
);
588 /* Look only at processors in ACPI namespace */
592 if (pr
->performance
) {
597 if (!performance
|| !per_cpu_ptr(performance
, i
)) {
603 /* Call _PSD for all CPUs */
604 for_each_possible_cpu(i
) {
605 pr
= per_cpu(processors
, i
);
609 pr
->performance
= per_cpu_ptr(performance
, i
);
610 cpumask_set_cpu(i
, pr
->performance
->shared_cpu_map
);
611 if (acpi_processor_get_psd(pr
)) {
620 * Now that we have _PSD data from all CPUs, lets setup P-state
623 for_each_possible_cpu(i
) {
624 pr
= per_cpu(processors
, i
);
628 if (cpumask_test_cpu(i
, covered_cpus
))
631 pdomain
= &(pr
->performance
->domain_info
);
632 cpumask_set_cpu(i
, pr
->performance
->shared_cpu_map
);
633 cpumask_set_cpu(i
, covered_cpus
);
634 if (pdomain
->num_processors
<= 1)
637 /* Validate the Domain info */
638 count_target
= pdomain
->num_processors
;
640 if (pdomain
->coord_type
== DOMAIN_COORD_TYPE_SW_ALL
)
641 pr
->performance
->shared_type
= CPUFREQ_SHARED_TYPE_ALL
;
642 else if (pdomain
->coord_type
== DOMAIN_COORD_TYPE_HW_ALL
)
643 pr
->performance
->shared_type
= CPUFREQ_SHARED_TYPE_HW
;
644 else if (pdomain
->coord_type
== DOMAIN_COORD_TYPE_SW_ANY
)
645 pr
->performance
->shared_type
= CPUFREQ_SHARED_TYPE_ANY
;
647 for_each_possible_cpu(j
) {
651 match_pr
= per_cpu(processors
, j
);
655 match_pdomain
= &(match_pr
->performance
->domain_info
);
656 if (match_pdomain
->domain
!= pdomain
->domain
)
659 /* Here i and j are in the same domain */
661 if (match_pdomain
->num_processors
!= count_target
) {
666 if (pdomain
->coord_type
!= match_pdomain
->coord_type
) {
671 cpumask_set_cpu(j
, covered_cpus
);
672 cpumask_set_cpu(j
, pr
->performance
->shared_cpu_map
);
676 for_each_possible_cpu(j
) {
680 match_pr
= per_cpu(processors
, j
);
684 match_pdomain
= &(match_pr
->performance
->domain_info
);
685 if (match_pdomain
->domain
!= pdomain
->domain
)
688 match_pr
->performance
->shared_type
=
689 pr
->performance
->shared_type
;
690 cpumask_copy(match_pr
->performance
->shared_cpu_map
,
691 pr
->performance
->shared_cpu_map
);
696 for_each_possible_cpu(i
) {
697 pr
= per_cpu(processors
, i
);
698 if (!pr
|| !pr
->performance
)
701 /* Assume no coordination on any error parsing domain info */
703 cpumask_clear(pr
->performance
->shared_cpu_map
);
704 cpumask_set_cpu(i
, pr
->performance
->shared_cpu_map
);
705 pr
->performance
->shared_type
= CPUFREQ_SHARED_TYPE_ALL
;
707 pr
->performance
= NULL
; /* Will be set for real in register */
711 mutex_unlock(&performance_mutex
);
712 free_cpumask_var(covered_cpus
);
715 EXPORT_SYMBOL(acpi_processor_preregister_performance
);
718 acpi_processor_register_performance(struct acpi_processor_performance
719 *performance
, unsigned int cpu
)
721 struct acpi_processor
*pr
;
723 if (!(acpi_processor_ppc_status
& PPC_REGISTERED
))
726 mutex_lock(&performance_mutex
);
728 pr
= per_cpu(processors
, cpu
);
730 mutex_unlock(&performance_mutex
);
734 if (pr
->performance
) {
735 mutex_unlock(&performance_mutex
);
739 WARN_ON(!performance
);
741 pr
->performance
= performance
;
743 if (acpi_processor_get_performance_info(pr
)) {
744 pr
->performance
= NULL
;
745 mutex_unlock(&performance_mutex
);
749 mutex_unlock(&performance_mutex
);
753 EXPORT_SYMBOL(acpi_processor_register_performance
);
756 acpi_processor_unregister_performance(struct acpi_processor_performance
757 *performance
, unsigned int cpu
)
759 struct acpi_processor
*pr
;
761 mutex_lock(&performance_mutex
);
763 pr
= per_cpu(processors
, cpu
);
765 mutex_unlock(&performance_mutex
);
770 kfree(pr
->performance
->states
);
771 pr
->performance
= NULL
;
773 mutex_unlock(&performance_mutex
);
778 EXPORT_SYMBOL(acpi_processor_unregister_performance
);