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>
34 #include <linux/acpi.h>
35 #include <acpi/processor.h>
37 #include <asm/cpufeature.h>
40 #define PREFIX "ACPI: "
42 #define ACPI_PROCESSOR_CLASS "processor"
43 #define ACPI_PROCESSOR_FILE_PERFORMANCE "performance"
44 #define _COMPONENT ACPI_PROCESSOR_COMPONENT
45 ACPI_MODULE_NAME("processor_perflib");
47 static DEFINE_MUTEX(performance_mutex
);
50 * _PPC support is implemented as a CPUfreq policy notifier:
51 * This means each time a CPUfreq driver registered also with
52 * the ACPI core is asked to change the speed policy, the maximum
53 * value is adjusted so that it is within the platform limit.
55 * Also, when a new platform limit value is detected, the CPUfreq
56 * policy is adjusted accordingly.
60 * -1 -> cpufreq low level drivers not initialized -> _PSS, etc. not called yet
62 * 0 -> cpufreq low level drivers initialized -> consider _PPC values
63 * 1 -> ignore _PPC totally -> forced by user through boot param
65 static int ignore_ppc
= -1;
66 module_param(ignore_ppc
, int, 0644);
67 MODULE_PARM_DESC(ignore_ppc
, "If the frequency of your machine gets wrongly" \
68 "limited by BIOS, this should help");
70 #define PPC_REGISTERED 1
73 static int acpi_processor_ppc_status
;
75 static int acpi_processor_ppc_notifier(struct notifier_block
*nb
,
76 unsigned long event
, void *data
)
78 struct cpufreq_policy
*policy
= data
;
79 struct acpi_processor
*pr
;
82 if (event
== CPUFREQ_START
&& ignore_ppc
<= 0) {
90 if (event
!= CPUFREQ_INCOMPATIBLE
)
93 mutex_lock(&performance_mutex
);
95 pr
= per_cpu(processors
, policy
->cpu
);
96 if (!pr
|| !pr
->performance
)
99 ppc
= (unsigned int)pr
->performance_platform_limit
;
101 if (ppc
>= pr
->performance
->state_count
)
104 cpufreq_verify_within_limits(policy
, 0,
105 pr
->performance
->states
[ppc
].
106 core_frequency
* 1000);
109 mutex_unlock(&performance_mutex
);
114 static struct notifier_block acpi_ppc_notifier_block
= {
115 .notifier_call
= acpi_processor_ppc_notifier
,
118 static int acpi_processor_get_platform_limit(struct acpi_processor
*pr
)
120 acpi_status status
= 0;
121 unsigned long long ppc
= 0;
128 * _PPC indicates the maximum state currently supported by the platform
129 * (e.g. 0 = states 0..n; 1 = states 1..n; etc.
131 status
= acpi_evaluate_integer(pr
->handle
, "_PPC", NULL
, &ppc
);
133 if (status
!= AE_NOT_FOUND
)
134 acpi_processor_ppc_status
|= PPC_IN_USE
;
136 if (ACPI_FAILURE(status
) && status
!= AE_NOT_FOUND
) {
137 ACPI_EXCEPTION((AE_INFO
, status
, "Evaluating _PPC"));
141 pr_debug("CPU %d: _PPC is %d - frequency %s limited\n", pr
->id
,
142 (int)ppc
, ppc
? "" : "not");
144 pr
->performance_platform_limit
= (int)ppc
;
149 #define ACPI_PROCESSOR_NOTIFY_PERFORMANCE 0x80
151 * acpi_processor_ppc_ost: Notify firmware the _PPC evaluation status
152 * @handle: ACPI processor handle
153 * @status: the status code of _PPC evaluation
154 * 0: success. OSPM is now using the performance state specificed.
155 * 1: failure. OSPM has not changed the number of P-states in use
157 static void acpi_processor_ppc_ost(acpi_handle handle
, int status
)
159 if (acpi_has_method(handle
, "_OST"))
160 acpi_evaluate_ost(handle
, ACPI_PROCESSOR_NOTIFY_PERFORMANCE
,
164 int acpi_processor_ppc_has_changed(struct acpi_processor
*pr
, int event_flag
)
170 * Only when it is notification event, the _OST object
171 * will be evaluated. Otherwise it is skipped.
174 acpi_processor_ppc_ost(pr
->handle
, 1);
178 ret
= acpi_processor_get_platform_limit(pr
);
180 * Only when it is notification event, the _OST object
181 * will be evaluated. Otherwise it is skipped.
185 acpi_processor_ppc_ost(pr
->handle
, 1);
187 acpi_processor_ppc_ost(pr
->handle
, 0);
192 return cpufreq_update_policy(pr
->id
);
195 int acpi_processor_get_bios_limit(int cpu
, unsigned int *limit
)
197 struct acpi_processor
*pr
;
199 pr
= per_cpu(processors
, cpu
);
200 if (!pr
|| !pr
->performance
|| !pr
->performance
->state_count
)
202 *limit
= pr
->performance
->states
[pr
->performance_platform_limit
].
203 core_frequency
* 1000;
206 EXPORT_SYMBOL(acpi_processor_get_bios_limit
);
208 void acpi_processor_ppc_init(void)
210 if (!cpufreq_register_notifier
211 (&acpi_ppc_notifier_block
, CPUFREQ_POLICY_NOTIFIER
))
212 acpi_processor_ppc_status
|= PPC_REGISTERED
;
215 "Warning: Processor Platform Limit not supported.\n");
218 void acpi_processor_ppc_exit(void)
220 if (acpi_processor_ppc_status
& PPC_REGISTERED
)
221 cpufreq_unregister_notifier(&acpi_ppc_notifier_block
,
222 CPUFREQ_POLICY_NOTIFIER
);
224 acpi_processor_ppc_status
&= ~PPC_REGISTERED
;
227 static int acpi_processor_get_performance_control(struct acpi_processor
*pr
)
230 acpi_status status
= 0;
231 struct acpi_buffer buffer
= { ACPI_ALLOCATE_BUFFER
, NULL
};
232 union acpi_object
*pct
= NULL
;
233 union acpi_object obj
= { 0 };
236 status
= acpi_evaluate_object(pr
->handle
, "_PCT", NULL
, &buffer
);
237 if (ACPI_FAILURE(status
)) {
238 ACPI_EXCEPTION((AE_INFO
, status
, "Evaluating _PCT"));
242 pct
= (union acpi_object
*)buffer
.pointer
;
243 if (!pct
|| (pct
->type
!= ACPI_TYPE_PACKAGE
)
244 || (pct
->package
.count
!= 2)) {
245 printk(KERN_ERR PREFIX
"Invalid _PCT data\n");
254 obj
= pct
->package
.elements
[0];
256 if ((obj
.type
!= ACPI_TYPE_BUFFER
)
257 || (obj
.buffer
.length
< sizeof(struct acpi_pct_register
))
258 || (obj
.buffer
.pointer
== NULL
)) {
259 printk(KERN_ERR PREFIX
"Invalid _PCT data (control_register)\n");
263 memcpy(&pr
->performance
->control_register
, obj
.buffer
.pointer
,
264 sizeof(struct acpi_pct_register
));
270 obj
= pct
->package
.elements
[1];
272 if ((obj
.type
!= ACPI_TYPE_BUFFER
)
273 || (obj
.buffer
.length
< sizeof(struct acpi_pct_register
))
274 || (obj
.buffer
.pointer
== NULL
)) {
275 printk(KERN_ERR PREFIX
"Invalid _PCT data (status_register)\n");
280 memcpy(&pr
->performance
->status_register
, obj
.buffer
.pointer
,
281 sizeof(struct acpi_pct_register
));
284 kfree(buffer
.pointer
);
291 * Some AMDs have 50MHz frequency multiples, but only provide 100MHz rounding
292 * in their ACPI data. Calculate the real values and fix up the _PSS data.
294 static void amd_fixup_frequency(struct acpi_processor_px
*px
, int i
)
296 u32 hi
, lo
, fid
, did
;
297 int index
= px
->control
& 0x00000007;
299 if (boot_cpu_data
.x86_vendor
!= X86_VENDOR_AMD
)
302 if ((boot_cpu_data
.x86
== 0x10 && boot_cpu_data
.x86_model
< 10)
303 || boot_cpu_data
.x86
== 0x11) {
304 rdmsr(MSR_AMD_PSTATE_DEF_BASE
+ index
, lo
, hi
);
307 * Bit 63: PstateEn. Read-write. If set, the P-state is valid.
314 if (boot_cpu_data
.x86
== 0x10)
315 px
->core_frequency
= (100 * (fid
+ 0x10)) >> did
;
317 px
->core_frequency
= (100 * (fid
+ 8)) >> did
;
321 static void amd_fixup_frequency(struct acpi_processor_px
*px
, int i
) {};
324 static int acpi_processor_get_performance_states(struct acpi_processor
*pr
)
327 acpi_status status
= AE_OK
;
328 struct acpi_buffer buffer
= { ACPI_ALLOCATE_BUFFER
, NULL
};
329 struct acpi_buffer format
= { sizeof("NNNNNN"), "NNNNNN" };
330 struct acpi_buffer state
= { 0, NULL
};
331 union acpi_object
*pss
= NULL
;
333 int last_invalid
= -1;
336 status
= acpi_evaluate_object(pr
->handle
, "_PSS", NULL
, &buffer
);
337 if (ACPI_FAILURE(status
)) {
338 ACPI_EXCEPTION((AE_INFO
, status
, "Evaluating _PSS"));
342 pss
= buffer
.pointer
;
343 if (!pss
|| (pss
->type
!= ACPI_TYPE_PACKAGE
)) {
344 printk(KERN_ERR PREFIX
"Invalid _PSS data\n");
349 ACPI_DEBUG_PRINT((ACPI_DB_INFO
, "Found %d performance states\n",
350 pss
->package
.count
));
352 pr
->performance
->state_count
= pss
->package
.count
;
353 pr
->performance
->states
=
354 kmalloc(sizeof(struct acpi_processor_px
) * pss
->package
.count
,
356 if (!pr
->performance
->states
) {
361 for (i
= 0; i
< pr
->performance
->state_count
; i
++) {
363 struct acpi_processor_px
*px
= &(pr
->performance
->states
[i
]);
365 state
.length
= sizeof(struct acpi_processor_px
);
368 ACPI_DEBUG_PRINT((ACPI_DB_INFO
, "Extracting state %d\n", i
));
370 status
= acpi_extract_package(&(pss
->package
.elements
[i
]),
372 if (ACPI_FAILURE(status
)) {
373 ACPI_EXCEPTION((AE_INFO
, status
, "Invalid _PSS data"));
375 kfree(pr
->performance
->states
);
379 amd_fixup_frequency(px
, i
);
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 found for processor %d: 0x%llx MHz\n",
398 pr
->id
, px
->core_frequency
);
399 if (last_invalid
== -1)
402 if (last_invalid
!= -1) {
404 * Copy this valid entry over last_invalid entry
406 memcpy(&(pr
->performance
->states
[last_invalid
]),
407 px
, sizeof(struct acpi_processor_px
));
413 if (last_invalid
== 0) {
414 printk(KERN_ERR FW_BUG PREFIX
415 "No valid BIOS _PSS frequency found for processor %d\n", pr
->id
);
417 kfree(pr
->performance
->states
);
418 pr
->performance
->states
= NULL
;
421 if (last_invalid
> 0)
422 pr
->performance
->state_count
= last_invalid
;
425 kfree(buffer
.pointer
);
430 int acpi_processor_get_performance_info(struct acpi_processor
*pr
)
434 if (!pr
|| !pr
->performance
|| !pr
->handle
)
437 if (!acpi_has_method(pr
->handle
, "_PCT")) {
438 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
439 "ACPI-based processor performance control unavailable\n"));
443 result
= acpi_processor_get_performance_control(pr
);
447 result
= acpi_processor_get_performance_states(pr
);
451 /* We need to call _PPC once when cpufreq starts */
453 result
= acpi_processor_get_platform_limit(pr
);
458 * Having _PPC but missing frequencies (_PSS, _PCT) is a very good hint that
459 * the BIOS is older than the CPU and does not know its frequencies
463 if (acpi_has_method(pr
->handle
, "_PPC")) {
464 if(boot_cpu_has(X86_FEATURE_EST
))
465 printk(KERN_WARNING FW_BUG
"BIOS needs update for CPU "
466 "frequency support\n");
471 EXPORT_SYMBOL_GPL(acpi_processor_get_performance_info
);
472 int acpi_processor_notify_smm(struct module
*calling_module
)
475 static int is_done
= 0;
478 if (!(acpi_processor_ppc_status
& PPC_REGISTERED
))
481 if (!try_module_get(calling_module
))
484 /* is_done is set to negative if an error occurred,
485 * and to postitive if _no_ error occurred, but SMM
486 * was already notified. This avoids double notification
487 * which might lead to unexpected results...
490 module_put(calling_module
);
492 } else if (is_done
< 0) {
493 module_put(calling_module
);
499 /* Can't write pstate_control to smi_command if either value is zero */
500 if ((!acpi_gbl_FADT
.smi_command
) || (!acpi_gbl_FADT
.pstate_control
)) {
501 ACPI_DEBUG_PRINT((ACPI_DB_INFO
, "No SMI port or pstate_control\n"));
502 module_put(calling_module
);
506 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
507 "Writing pstate_control [0x%x] to smi_command [0x%x]\n",
508 acpi_gbl_FADT
.pstate_control
, acpi_gbl_FADT
.smi_command
));
510 status
= acpi_os_write_port(acpi_gbl_FADT
.smi_command
,
511 (u32
) acpi_gbl_FADT
.pstate_control
, 8);
512 if (ACPI_FAILURE(status
)) {
513 ACPI_EXCEPTION((AE_INFO
, status
,
514 "Failed to write pstate_control [0x%x] to "
515 "smi_command [0x%x]", acpi_gbl_FADT
.pstate_control
,
516 acpi_gbl_FADT
.smi_command
));
517 module_put(calling_module
);
521 /* Success. If there's no _PPC, we need to fear nothing, so
522 * we can allow the cpufreq driver to be rmmod'ed. */
525 if (!(acpi_processor_ppc_status
& PPC_IN_USE
))
526 module_put(calling_module
);
531 EXPORT_SYMBOL(acpi_processor_notify_smm
);
533 static int acpi_processor_get_psd(struct acpi_processor
*pr
)
536 acpi_status status
= AE_OK
;
537 struct acpi_buffer buffer
= {ACPI_ALLOCATE_BUFFER
, NULL
};
538 struct acpi_buffer format
= {sizeof("NNNNN"), "NNNNN"};
539 struct acpi_buffer state
= {0, NULL
};
540 union acpi_object
*psd
= NULL
;
541 struct acpi_psd_package
*pdomain
;
543 status
= acpi_evaluate_object(pr
->handle
, "_PSD", NULL
, &buffer
);
544 if (ACPI_FAILURE(status
)) {
548 psd
= buffer
.pointer
;
549 if (!psd
|| (psd
->type
!= ACPI_TYPE_PACKAGE
)) {
550 printk(KERN_ERR PREFIX
"Invalid _PSD data\n");
555 if (psd
->package
.count
!= 1) {
556 printk(KERN_ERR PREFIX
"Invalid _PSD data\n");
561 pdomain
= &(pr
->performance
->domain_info
);
563 state
.length
= sizeof(struct acpi_psd_package
);
564 state
.pointer
= pdomain
;
566 status
= acpi_extract_package(&(psd
->package
.elements
[0]),
568 if (ACPI_FAILURE(status
)) {
569 printk(KERN_ERR PREFIX
"Invalid _PSD data\n");
574 if (pdomain
->num_entries
!= ACPI_PSD_REV0_ENTRIES
) {
575 printk(KERN_ERR PREFIX
"Unknown _PSD:num_entries\n");
580 if (pdomain
->revision
!= ACPI_PSD_REV0_REVISION
) {
581 printk(KERN_ERR PREFIX
"Unknown _PSD:revision\n");
586 if (pdomain
->coord_type
!= DOMAIN_COORD_TYPE_SW_ALL
&&
587 pdomain
->coord_type
!= DOMAIN_COORD_TYPE_SW_ANY
&&
588 pdomain
->coord_type
!= DOMAIN_COORD_TYPE_HW_ALL
) {
589 printk(KERN_ERR PREFIX
"Invalid _PSD:coord_type\n");
594 kfree(buffer
.pointer
);
598 int acpi_processor_preregister_performance(
599 struct acpi_processor_performance __percpu
*performance
)
604 cpumask_var_t covered_cpus
;
605 struct acpi_processor
*pr
;
606 struct acpi_psd_package
*pdomain
;
607 struct acpi_processor
*match_pr
;
608 struct acpi_psd_package
*match_pdomain
;
610 if (!zalloc_cpumask_var(&covered_cpus
, GFP_KERNEL
))
613 mutex_lock(&performance_mutex
);
616 * Check if another driver has already registered, and abort before
617 * changing pr->performance if it has. Check input data as well.
619 for_each_possible_cpu(i
) {
620 pr
= per_cpu(processors
, i
);
622 /* Look only at processors in ACPI namespace */
626 if (pr
->performance
) {
631 if (!performance
|| !per_cpu_ptr(performance
, i
)) {
637 /* Call _PSD for all CPUs */
638 for_each_possible_cpu(i
) {
639 pr
= per_cpu(processors
, i
);
643 pr
->performance
= per_cpu_ptr(performance
, i
);
644 cpumask_set_cpu(i
, pr
->performance
->shared_cpu_map
);
645 if (acpi_processor_get_psd(pr
)) {
654 * Now that we have _PSD data from all CPUs, lets setup P-state
657 for_each_possible_cpu(i
) {
658 pr
= per_cpu(processors
, i
);
662 if (cpumask_test_cpu(i
, covered_cpus
))
665 pdomain
= &(pr
->performance
->domain_info
);
666 cpumask_set_cpu(i
, pr
->performance
->shared_cpu_map
);
667 cpumask_set_cpu(i
, covered_cpus
);
668 if (pdomain
->num_processors
<= 1)
671 /* Validate the Domain info */
672 count_target
= pdomain
->num_processors
;
673 if (pdomain
->coord_type
== DOMAIN_COORD_TYPE_SW_ALL
)
674 pr
->performance
->shared_type
= CPUFREQ_SHARED_TYPE_ALL
;
675 else if (pdomain
->coord_type
== DOMAIN_COORD_TYPE_HW_ALL
)
676 pr
->performance
->shared_type
= CPUFREQ_SHARED_TYPE_HW
;
677 else if (pdomain
->coord_type
== DOMAIN_COORD_TYPE_SW_ANY
)
678 pr
->performance
->shared_type
= CPUFREQ_SHARED_TYPE_ANY
;
680 for_each_possible_cpu(j
) {
684 match_pr
= per_cpu(processors
, j
);
688 match_pdomain
= &(match_pr
->performance
->domain_info
);
689 if (match_pdomain
->domain
!= pdomain
->domain
)
692 /* Here i and j are in the same domain */
694 if (match_pdomain
->num_processors
!= count_target
) {
699 if (pdomain
->coord_type
!= match_pdomain
->coord_type
) {
704 cpumask_set_cpu(j
, covered_cpus
);
705 cpumask_set_cpu(j
, pr
->performance
->shared_cpu_map
);
708 for_each_possible_cpu(j
) {
712 match_pr
= per_cpu(processors
, j
);
716 match_pdomain
= &(match_pr
->performance
->domain_info
);
717 if (match_pdomain
->domain
!= pdomain
->domain
)
720 match_pr
->performance
->shared_type
=
721 pr
->performance
->shared_type
;
722 cpumask_copy(match_pr
->performance
->shared_cpu_map
,
723 pr
->performance
->shared_cpu_map
);
728 for_each_possible_cpu(i
) {
729 pr
= per_cpu(processors
, i
);
730 if (!pr
|| !pr
->performance
)
733 /* Assume no coordination on any error parsing domain info */
735 cpumask_clear(pr
->performance
->shared_cpu_map
);
736 cpumask_set_cpu(i
, pr
->performance
->shared_cpu_map
);
737 pr
->performance
->shared_type
= CPUFREQ_SHARED_TYPE_ALL
;
739 pr
->performance
= NULL
; /* Will be set for real in register */
743 mutex_unlock(&performance_mutex
);
744 free_cpumask_var(covered_cpus
);
747 EXPORT_SYMBOL(acpi_processor_preregister_performance
);
750 acpi_processor_register_performance(struct acpi_processor_performance
751 *performance
, unsigned int cpu
)
753 struct acpi_processor
*pr
;
755 if (!(acpi_processor_ppc_status
& PPC_REGISTERED
))
758 mutex_lock(&performance_mutex
);
760 pr
= per_cpu(processors
, cpu
);
762 mutex_unlock(&performance_mutex
);
766 if (pr
->performance
) {
767 mutex_unlock(&performance_mutex
);
771 WARN_ON(!performance
);
773 pr
->performance
= performance
;
775 if (acpi_processor_get_performance_info(pr
)) {
776 pr
->performance
= NULL
;
777 mutex_unlock(&performance_mutex
);
781 mutex_unlock(&performance_mutex
);
785 EXPORT_SYMBOL(acpi_processor_register_performance
);
788 acpi_processor_unregister_performance(struct acpi_processor_performance
789 *performance
, unsigned int cpu
)
791 struct acpi_processor
*pr
;
793 mutex_lock(&performance_mutex
);
795 pr
= per_cpu(processors
, cpu
);
797 mutex_unlock(&performance_mutex
);
802 kfree(pr
->performance
->states
);
803 pr
->performance
= NULL
;
805 mutex_unlock(&performance_mutex
);
810 EXPORT_SYMBOL(acpi_processor_unregister_performance
);