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
);
53 * _PPC support is implemented as a CPUfreq policy notifier:
54 * This means each time a CPUfreq driver registered also with
55 * the ACPI core is asked to change the speed policy, the maximum
56 * value is adjusted so that it is within the platform limit.
58 * Also, when a new platform limit value is detected, the CPUfreq
59 * policy is adjusted accordingly.
63 * -1 -> cpufreq low level drivers not initialized -> _PSS, etc. not called yet
65 * 0 -> cpufreq low level drivers initialized -> consider _PPC values
66 * 1 -> ignore _PPC totally -> forced by user through boot param
68 static int ignore_ppc
= -1;
69 module_param(ignore_ppc
, int, 0644);
70 MODULE_PARM_DESC(ignore_ppc
, "If the frequency of your machine gets wrongly" \
71 "limited by BIOS, this should help");
73 #define PPC_REGISTERED 1
76 static int acpi_processor_ppc_status
;
78 static int acpi_processor_ppc_notifier(struct notifier_block
*nb
,
79 unsigned long event
, void *data
)
81 struct cpufreq_policy
*policy
= data
;
82 struct acpi_processor
*pr
;
85 if (event
== CPUFREQ_START
&& ignore_ppc
<= 0) {
93 if (event
!= CPUFREQ_INCOMPATIBLE
)
96 mutex_lock(&performance_mutex
);
98 pr
= per_cpu(processors
, policy
->cpu
);
99 if (!pr
|| !pr
->performance
)
102 ppc
= (unsigned int)pr
->performance_platform_limit
;
104 if (ppc
>= pr
->performance
->state_count
)
107 cpufreq_verify_within_limits(policy
, 0,
108 pr
->performance
->states
[ppc
].
109 core_frequency
* 1000);
112 mutex_unlock(&performance_mutex
);
117 static struct notifier_block acpi_ppc_notifier_block
= {
118 .notifier_call
= acpi_processor_ppc_notifier
,
121 static int acpi_processor_get_platform_limit(struct acpi_processor
*pr
)
123 acpi_status status
= 0;
124 unsigned long long ppc
= 0;
131 * _PPC indicates the maximum state currently supported by the platform
132 * (e.g. 0 = states 0..n; 1 = states 1..n; etc.
134 status
= acpi_evaluate_integer(pr
->handle
, "_PPC", NULL
, &ppc
);
136 if (status
!= AE_NOT_FOUND
)
137 acpi_processor_ppc_status
|= PPC_IN_USE
;
139 if (ACPI_FAILURE(status
) && status
!= AE_NOT_FOUND
) {
140 ACPI_EXCEPTION((AE_INFO
, status
, "Evaluating _PPC"));
144 pr_debug("CPU %d: _PPC is %d - frequency %s limited\n", pr
->id
,
145 (int)ppc
, ppc
? "" : "not");
147 pr
->performance_platform_limit
= (int)ppc
;
152 #define ACPI_PROCESSOR_NOTIFY_PERFORMANCE 0x80
154 * acpi_processor_ppc_ost: Notify firmware the _PPC evaluation status
155 * @handle: ACPI processor handle
156 * @status: the status code of _PPC evaluation
157 * 0: success. OSPM is now using the performance state specificed.
158 * 1: failure. OSPM has not changed the number of P-states in use
160 static void acpi_processor_ppc_ost(acpi_handle handle
, int status
)
162 union acpi_object params
[2] = {
163 {.type
= ACPI_TYPE_INTEGER
,},
164 {.type
= ACPI_TYPE_INTEGER
,},
166 struct acpi_object_list arg_list
= {2, params
};
169 params
[0].integer
.value
= ACPI_PROCESSOR_NOTIFY_PERFORMANCE
;
170 params
[1].integer
.value
= status
;
172 /* when there is no _OST , skip it */
173 if (ACPI_FAILURE(acpi_get_handle(handle
, "_OST", &temp
)))
176 acpi_evaluate_object(handle
, "_OST", &arg_list
, NULL
);
180 int acpi_processor_ppc_has_changed(struct acpi_processor
*pr
, int event_flag
)
186 * Only when it is notification event, the _OST object
187 * will be evaluated. Otherwise it is skipped.
190 acpi_processor_ppc_ost(pr
->handle
, 1);
194 ret
= acpi_processor_get_platform_limit(pr
);
196 * Only when it is notification event, the _OST object
197 * will be evaluated. Otherwise it is skipped.
201 acpi_processor_ppc_ost(pr
->handle
, 1);
203 acpi_processor_ppc_ost(pr
->handle
, 0);
208 return cpufreq_update_policy(pr
->id
);
211 int acpi_processor_get_bios_limit(int cpu
, unsigned int *limit
)
213 struct acpi_processor
*pr
;
215 pr
= per_cpu(processors
, cpu
);
216 if (!pr
|| !pr
->performance
|| !pr
->performance
->state_count
)
218 *limit
= pr
->performance
->states
[pr
->performance_platform_limit
].
219 core_frequency
* 1000;
222 EXPORT_SYMBOL(acpi_processor_get_bios_limit
);
224 void acpi_processor_ppc_init(void)
226 if (!cpufreq_register_notifier
227 (&acpi_ppc_notifier_block
, CPUFREQ_POLICY_NOTIFIER
))
228 acpi_processor_ppc_status
|= PPC_REGISTERED
;
231 "Warning: Processor Platform Limit not supported.\n");
234 void acpi_processor_ppc_exit(void)
236 if (acpi_processor_ppc_status
& PPC_REGISTERED
)
237 cpufreq_unregister_notifier(&acpi_ppc_notifier_block
,
238 CPUFREQ_POLICY_NOTIFIER
);
240 acpi_processor_ppc_status
&= ~PPC_REGISTERED
;
244 * Do a quick check if the systems looks like it should use ACPI
245 * cpufreq. We look at a _PCT method being available, but don't
246 * do a whole lot of sanity checks.
248 void acpi_processor_load_module(struct acpi_processor
*pr
)
250 static int requested
;
251 acpi_status status
= 0;
252 struct acpi_buffer buffer
= { ACPI_ALLOCATE_BUFFER
, NULL
};
254 if (!arch_has_acpi_pdc() || requested
)
256 status
= acpi_evaluate_object(pr
->handle
, "_PCT", NULL
, &buffer
);
257 if (!ACPI_FAILURE(status
)) {
258 printk(KERN_INFO PREFIX
"Requesting acpi_cpufreq\n");
259 request_module_nowait("acpi_cpufreq");
262 kfree(buffer
.pointer
);
265 static int acpi_processor_get_performance_control(struct acpi_processor
*pr
)
268 acpi_status status
= 0;
269 struct acpi_buffer buffer
= { ACPI_ALLOCATE_BUFFER
, NULL
};
270 union acpi_object
*pct
= NULL
;
271 union acpi_object obj
= { 0 };
274 status
= acpi_evaluate_object(pr
->handle
, "_PCT", NULL
, &buffer
);
275 if (ACPI_FAILURE(status
)) {
276 ACPI_EXCEPTION((AE_INFO
, status
, "Evaluating _PCT"));
280 pct
= (union acpi_object
*)buffer
.pointer
;
281 if (!pct
|| (pct
->type
!= ACPI_TYPE_PACKAGE
)
282 || (pct
->package
.count
!= 2)) {
283 printk(KERN_ERR PREFIX
"Invalid _PCT data\n");
292 obj
= pct
->package
.elements
[0];
294 if ((obj
.type
!= ACPI_TYPE_BUFFER
)
295 || (obj
.buffer
.length
< sizeof(struct acpi_pct_register
))
296 || (obj
.buffer
.pointer
== NULL
)) {
297 printk(KERN_ERR PREFIX
"Invalid _PCT data (control_register)\n");
301 memcpy(&pr
->performance
->control_register
, obj
.buffer
.pointer
,
302 sizeof(struct acpi_pct_register
));
308 obj
= pct
->package
.elements
[1];
310 if ((obj
.type
!= ACPI_TYPE_BUFFER
)
311 || (obj
.buffer
.length
< sizeof(struct acpi_pct_register
))
312 || (obj
.buffer
.pointer
== NULL
)) {
313 printk(KERN_ERR PREFIX
"Invalid _PCT data (status_register)\n");
318 memcpy(&pr
->performance
->status_register
, obj
.buffer
.pointer
,
319 sizeof(struct acpi_pct_register
));
322 kfree(buffer
.pointer
);
327 static int acpi_processor_get_performance_states(struct acpi_processor
*pr
)
330 acpi_status status
= AE_OK
;
331 struct acpi_buffer buffer
= { ACPI_ALLOCATE_BUFFER
, NULL
};
332 struct acpi_buffer format
= { sizeof("NNNNNN"), "NNNNNN" };
333 struct acpi_buffer state
= { 0, NULL
};
334 union acpi_object
*pss
= NULL
;
338 status
= acpi_evaluate_object(pr
->handle
, "_PSS", NULL
, &buffer
);
339 if (ACPI_FAILURE(status
)) {
340 ACPI_EXCEPTION((AE_INFO
, status
, "Evaluating _PSS"));
344 pss
= buffer
.pointer
;
345 if (!pss
|| (pss
->type
!= ACPI_TYPE_PACKAGE
)) {
346 printk(KERN_ERR PREFIX
"Invalid _PSS data\n");
351 ACPI_DEBUG_PRINT((ACPI_DB_INFO
, "Found %d performance states\n",
352 pss
->package
.count
));
354 pr
->performance
->state_count
= pss
->package
.count
;
355 pr
->performance
->states
=
356 kmalloc(sizeof(struct acpi_processor_px
) * pss
->package
.count
,
358 if (!pr
->performance
->states
) {
363 for (i
= 0; i
< pr
->performance
->state_count
; i
++) {
365 struct acpi_processor_px
*px
= &(pr
->performance
->states
[i
]);
367 state
.length
= sizeof(struct acpi_processor_px
);
370 ACPI_DEBUG_PRINT((ACPI_DB_INFO
, "Extracting state %d\n", i
));
372 status
= acpi_extract_package(&(pss
->package
.elements
[i
]),
374 if (ACPI_FAILURE(status
)) {
375 ACPI_EXCEPTION((AE_INFO
, status
, "Invalid _PSS data"));
377 kfree(pr
->performance
->states
);
381 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
382 "State [%d]: core_frequency[%d] power[%d] transition_latency[%d] bus_master_latency[%d] control[0x%x] status[0x%x]\n",
384 (u32
) px
->core_frequency
,
386 (u32
) px
->transition_latency
,
387 (u32
) px
->bus_master_latency
,
388 (u32
) px
->control
, (u32
) px
->status
));
391 * Check that ACPI's u64 MHz will be valid as u32 KHz in cpufreq
393 if (!px
->core_frequency
||
394 ((u32
)(px
->core_frequency
* 1000) !=
395 (px
->core_frequency
* 1000))) {
396 printk(KERN_ERR FW_BUG PREFIX
397 "Invalid BIOS _PSS frequency: 0x%llx MHz\n",
400 kfree(pr
->performance
->states
);
406 kfree(buffer
.pointer
);
411 static int acpi_processor_get_performance_info(struct acpi_processor
*pr
)
414 acpi_status status
= AE_OK
;
415 acpi_handle handle
= NULL
;
417 if (!pr
|| !pr
->performance
|| !pr
->handle
)
420 status
= acpi_get_handle(pr
->handle
, "_PCT", &handle
);
421 if (ACPI_FAILURE(status
)) {
422 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
423 "ACPI-based processor performance control unavailable\n"));
427 result
= acpi_processor_get_performance_control(pr
);
431 result
= acpi_processor_get_performance_states(pr
);
435 /* We need to call _PPC once when cpufreq starts */
437 result
= acpi_processor_get_platform_limit(pr
);
442 * Having _PPC but missing frequencies (_PSS, _PCT) is a very good hint that
443 * the BIOS is older than the CPU and does not know its frequencies
447 if (ACPI_SUCCESS(acpi_get_handle(pr
->handle
, "_PPC", &handle
))){
448 if(boot_cpu_has(X86_FEATURE_EST
))
449 printk(KERN_WARNING FW_BUG
"BIOS needs update for CPU "
450 "frequency support\n");
456 int acpi_processor_notify_smm(struct module
*calling_module
)
459 static int is_done
= 0;
462 if (!(acpi_processor_ppc_status
& PPC_REGISTERED
))
465 if (!try_module_get(calling_module
))
468 /* is_done is set to negative if an error occurred,
469 * and to postitive if _no_ error occurred, but SMM
470 * was already notified. This avoids double notification
471 * which might lead to unexpected results...
474 module_put(calling_module
);
476 } else if (is_done
< 0) {
477 module_put(calling_module
);
483 /* Can't write pstate_control to smi_command if either value is zero */
484 if ((!acpi_gbl_FADT
.smi_command
) || (!acpi_gbl_FADT
.pstate_control
)) {
485 ACPI_DEBUG_PRINT((ACPI_DB_INFO
, "No SMI port or pstate_control\n"));
486 module_put(calling_module
);
490 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
491 "Writing pstate_control [0x%x] to smi_command [0x%x]\n",
492 acpi_gbl_FADT
.pstate_control
, acpi_gbl_FADT
.smi_command
));
494 status
= acpi_os_write_port(acpi_gbl_FADT
.smi_command
,
495 (u32
) acpi_gbl_FADT
.pstate_control
, 8);
496 if (ACPI_FAILURE(status
)) {
497 ACPI_EXCEPTION((AE_INFO
, status
,
498 "Failed to write pstate_control [0x%x] to "
499 "smi_command [0x%x]", acpi_gbl_FADT
.pstate_control
,
500 acpi_gbl_FADT
.smi_command
));
501 module_put(calling_module
);
505 /* Success. If there's no _PPC, we need to fear nothing, so
506 * we can allow the cpufreq driver to be rmmod'ed. */
509 if (!(acpi_processor_ppc_status
& PPC_IN_USE
))
510 module_put(calling_module
);
515 EXPORT_SYMBOL(acpi_processor_notify_smm
);
517 static int acpi_processor_get_psd(struct acpi_processor
*pr
)
520 acpi_status status
= AE_OK
;
521 struct acpi_buffer buffer
= {ACPI_ALLOCATE_BUFFER
, NULL
};
522 struct acpi_buffer format
= {sizeof("NNNNN"), "NNNNN"};
523 struct acpi_buffer state
= {0, NULL
};
524 union acpi_object
*psd
= NULL
;
525 struct acpi_psd_package
*pdomain
;
527 status
= acpi_evaluate_object(pr
->handle
, "_PSD", NULL
, &buffer
);
528 if (ACPI_FAILURE(status
)) {
532 psd
= buffer
.pointer
;
533 if (!psd
|| (psd
->type
!= ACPI_TYPE_PACKAGE
)) {
534 printk(KERN_ERR PREFIX
"Invalid _PSD data\n");
539 if (psd
->package
.count
!= 1) {
540 printk(KERN_ERR PREFIX
"Invalid _PSD data\n");
545 pdomain
= &(pr
->performance
->domain_info
);
547 state
.length
= sizeof(struct acpi_psd_package
);
548 state
.pointer
= pdomain
;
550 status
= acpi_extract_package(&(psd
->package
.elements
[0]),
552 if (ACPI_FAILURE(status
)) {
553 printk(KERN_ERR PREFIX
"Invalid _PSD data\n");
558 if (pdomain
->num_entries
!= ACPI_PSD_REV0_ENTRIES
) {
559 printk(KERN_ERR PREFIX
"Unknown _PSD:num_entries\n");
564 if (pdomain
->revision
!= ACPI_PSD_REV0_REVISION
) {
565 printk(KERN_ERR PREFIX
"Unknown _PSD:revision\n");
570 if (pdomain
->coord_type
!= DOMAIN_COORD_TYPE_SW_ALL
&&
571 pdomain
->coord_type
!= DOMAIN_COORD_TYPE_SW_ANY
&&
572 pdomain
->coord_type
!= DOMAIN_COORD_TYPE_HW_ALL
) {
573 printk(KERN_ERR PREFIX
"Invalid _PSD:coord_type\n");
578 kfree(buffer
.pointer
);
582 int acpi_processor_preregister_performance(
583 struct acpi_processor_performance __percpu
*performance
)
585 int count
, count_target
;
588 cpumask_var_t covered_cpus
;
589 struct acpi_processor
*pr
;
590 struct acpi_psd_package
*pdomain
;
591 struct acpi_processor
*match_pr
;
592 struct acpi_psd_package
*match_pdomain
;
594 if (!zalloc_cpumask_var(&covered_cpus
, GFP_KERNEL
))
597 mutex_lock(&performance_mutex
);
600 * Check if another driver has already registered, and abort before
601 * changing pr->performance if it has. Check input data as well.
603 for_each_possible_cpu(i
) {
604 pr
= per_cpu(processors
, i
);
606 /* Look only at processors in ACPI namespace */
610 if (pr
->performance
) {
615 if (!performance
|| !per_cpu_ptr(performance
, i
)) {
621 /* Call _PSD for all CPUs */
622 for_each_possible_cpu(i
) {
623 pr
= per_cpu(processors
, i
);
627 pr
->performance
= per_cpu_ptr(performance
, i
);
628 cpumask_set_cpu(i
, pr
->performance
->shared_cpu_map
);
629 if (acpi_processor_get_psd(pr
)) {
638 * Now that we have _PSD data from all CPUs, lets setup P-state
641 for_each_possible_cpu(i
) {
642 pr
= per_cpu(processors
, i
);
646 if (cpumask_test_cpu(i
, covered_cpus
))
649 pdomain
= &(pr
->performance
->domain_info
);
650 cpumask_set_cpu(i
, pr
->performance
->shared_cpu_map
);
651 cpumask_set_cpu(i
, covered_cpus
);
652 if (pdomain
->num_processors
<= 1)
655 /* Validate the Domain info */
656 count_target
= pdomain
->num_processors
;
658 if (pdomain
->coord_type
== DOMAIN_COORD_TYPE_SW_ALL
)
659 pr
->performance
->shared_type
= CPUFREQ_SHARED_TYPE_ALL
;
660 else if (pdomain
->coord_type
== DOMAIN_COORD_TYPE_HW_ALL
)
661 pr
->performance
->shared_type
= CPUFREQ_SHARED_TYPE_HW
;
662 else if (pdomain
->coord_type
== DOMAIN_COORD_TYPE_SW_ANY
)
663 pr
->performance
->shared_type
= CPUFREQ_SHARED_TYPE_ANY
;
665 for_each_possible_cpu(j
) {
669 match_pr
= per_cpu(processors
, j
);
673 match_pdomain
= &(match_pr
->performance
->domain_info
);
674 if (match_pdomain
->domain
!= pdomain
->domain
)
677 /* Here i and j are in the same domain */
679 if (match_pdomain
->num_processors
!= count_target
) {
684 if (pdomain
->coord_type
!= match_pdomain
->coord_type
) {
689 cpumask_set_cpu(j
, covered_cpus
);
690 cpumask_set_cpu(j
, pr
->performance
->shared_cpu_map
);
694 for_each_possible_cpu(j
) {
698 match_pr
= per_cpu(processors
, j
);
702 match_pdomain
= &(match_pr
->performance
->domain_info
);
703 if (match_pdomain
->domain
!= pdomain
->domain
)
706 match_pr
->performance
->shared_type
=
707 pr
->performance
->shared_type
;
708 cpumask_copy(match_pr
->performance
->shared_cpu_map
,
709 pr
->performance
->shared_cpu_map
);
714 for_each_possible_cpu(i
) {
715 pr
= per_cpu(processors
, i
);
716 if (!pr
|| !pr
->performance
)
719 /* Assume no coordination on any error parsing domain info */
721 cpumask_clear(pr
->performance
->shared_cpu_map
);
722 cpumask_set_cpu(i
, pr
->performance
->shared_cpu_map
);
723 pr
->performance
->shared_type
= CPUFREQ_SHARED_TYPE_ALL
;
725 pr
->performance
= NULL
; /* Will be set for real in register */
729 mutex_unlock(&performance_mutex
);
730 free_cpumask_var(covered_cpus
);
733 EXPORT_SYMBOL(acpi_processor_preregister_performance
);
736 acpi_processor_register_performance(struct acpi_processor_performance
737 *performance
, unsigned int cpu
)
739 struct acpi_processor
*pr
;
741 if (!(acpi_processor_ppc_status
& PPC_REGISTERED
))
744 mutex_lock(&performance_mutex
);
746 pr
= per_cpu(processors
, cpu
);
748 mutex_unlock(&performance_mutex
);
752 if (pr
->performance
) {
753 mutex_unlock(&performance_mutex
);
757 WARN_ON(!performance
);
759 pr
->performance
= performance
;
761 if (acpi_processor_get_performance_info(pr
)) {
762 pr
->performance
= NULL
;
763 mutex_unlock(&performance_mutex
);
767 mutex_unlock(&performance_mutex
);
771 EXPORT_SYMBOL(acpi_processor_register_performance
);
774 acpi_processor_unregister_performance(struct acpi_processor_performance
775 *performance
, unsigned int cpu
)
777 struct acpi_processor
*pr
;
779 mutex_lock(&performance_mutex
);
781 pr
= per_cpu(processors
, cpu
);
783 mutex_unlock(&performance_mutex
);
788 kfree(pr
->performance
->states
);
789 pr
->performance
= NULL
;
791 mutex_unlock(&performance_mutex
);
796 EXPORT_SYMBOL(acpi_processor_unregister_performance
);