2 * Copyright (c) 2003 Nate Lawson (SDG)
3 * Copyright (c) 2001 Michael Smith
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * $FreeBSD: src/sys/dev/acpica/acpi_cpu.c,v 1.41 2004/06/24 00:38:51 njl Exp $
28 * $DragonFly: src/sys/dev/acpica5/acpi_cpu.c,v 1.8 2004/11/21 09:07:14 y0netan1 Exp $
32 #include <sys/param.h>
34 #include <sys/kernel.h>
35 #include <sys/malloc.h>
36 #include <sys/globaldata.h>
37 #include <sys/power.h>
40 #include <sys/thread2.h>
42 #include <bus/pci/pcivar.h>
43 #include <machine/atomic.h>
44 #include <machine/bus.h>
45 #include <machine/globaldata.h>
46 #include <machine/smp.h>
53 * Support for ACPI Processor devices, including ACPI 2.0 throttling
54 * and C[1-3] sleep states.
56 * TODO: implement scans of all CPUs to be sure all Cx states are
60 /* Hooks for the ACPI CA debugging infrastructure */
61 #define _COMPONENT ACPI_PROCESSOR
62 ACPI_MODULE_NAME("PROCESSOR")
65 struct resource
*p_lvlx
; /* Register to read to enter state. */
66 uint32_t type
; /* C1-3 (C4 and up treated as C3). */
67 uint32_t trans_lat
; /* Transition latency (usec). */
68 uint32_t power
; /* Power consumed (mW). */
70 #define MAX_CX_STATES 8
72 struct acpi_cpu_softc
{
74 ACPI_HANDLE cpu_handle
;
75 uint32_t acpi_id
; /* ACPI processor id */
76 uint32_t cpu_p_blk
; /* ACPI P_BLK location */
77 uint32_t cpu_p_blk_len
; /* P_BLK length (must be 6). */
78 struct resource
*cpu_p_cnt
; /* Throttling control register */
79 struct acpi_cx cpu_cx_states
[MAX_CX_STATES
];
80 int cpu_cx_count
; /* Number of valid Cx states. */
81 int cpu_prev_sleep
;/* Last idle sleep duration. */
84 #define CPU_GET_REG(reg, width) \
85 (bus_space_read_ ## width(rman_get_bustag((reg)), \
86 rman_get_bushandle((reg)), 0))
87 #define CPU_SET_REG(reg, width, val) \
88 (bus_space_write_ ## width(rman_get_bustag((reg)), \
89 rman_get_bushandle((reg)), 0, (val)))
92 * Speeds are stored in counts, from 1 to CPU_MAX_SPEED, and
93 * reported to the user in tenths of a percent.
95 static uint32_t cpu_duty_offset
;
96 static uint32_t cpu_duty_width
;
97 #define CPU_MAX_SPEED (1 << cpu_duty_width)
98 #define CPU_SPEED_PERCENT(x) ((1000 * (x)) / CPU_MAX_SPEED)
99 #define CPU_SPEED_PRINTABLE(x) (CPU_SPEED_PERCENT(x) / 10), \
100 (CPU_SPEED_PERCENT(x) % 10)
101 #define CPU_P_CNT_THT_EN (1<<4)
102 #define PM_USEC(x) ((x) >> 2) /* ~4 clocks per usec (3.57955 Mhz) */
104 #define ACPI_CPU_NOTIFY_PERF_STATES 0x80 /* _PSS changed. */
105 #define ACPI_CPU_NOTIFY_CX_STATES 0x81 /* _CST changed. */
107 #define CPU_QUIRK_NO_C3 0x0001 /* C3-type states are not usable. */
108 #define CPU_QUIRK_NO_THROTTLE 0x0002 /* Throttling is not usable. */
110 #define PCI_VENDOR_INTEL 0x8086
111 #define PCI_DEVICE_82371AB_3 0x7113 /* PIIX4 chipset for quirks. */
112 #define PCI_REVISION_A_STEP 0
113 #define PCI_REVISION_B_STEP 1
114 #define PCI_REVISION_4E 2
115 #define PCI_REVISION_4M 3
117 /* Platform hardware resource information. */
118 static uint32_t cpu_smi_cmd
; /* Value to write to SMI_CMD. */
119 static uint8_t cpu_pstate_cnt
;/* Register to take over throttling. */
120 static uint8_t cpu_cst_cnt
; /* Indicate we are _CST aware. */
121 static int cpu_rid
; /* Driver-wide resource id. */
122 static int cpu_quirks
; /* Indicate any hardware bugs. */
125 static int cpu_cx_count
; /* Number of valid states */
126 static int cpu_non_c3
; /* Index of lowest non-C3 state. */
127 static u_int cpu_cx_stats
[MAX_CX_STATES
];/* Cx usage history. */
129 /* Values for sysctl. */
130 static uint32_t cpu_throttle_state
;
131 static uint32_t cpu_throttle_max
;
132 static uint32_t cpu_throttle_performance
;
133 static uint32_t cpu_throttle_economy
;
134 static int cpu_cx_lowest
;
135 static char cpu_cx_supported
[64];
137 static device_t
*cpu_devices
;
138 static int cpu_ndevices
;
139 static struct acpi_cpu_softc
**cpu_softc
;
141 static struct sysctl_ctx_list acpi_cpu_sysctl_ctx
;
142 static struct sysctl_oid
*acpi_cpu_sysctl_tree
;
144 static int acpi_cpu_probe(device_t dev
);
145 static int acpi_cpu_attach(device_t dev
);
146 static int acpi_pcpu_get_id(uint32_t idx
, uint32_t *acpi_id
,
148 static int acpi_cpu_shutdown(device_t dev
);
149 static int acpi_cpu_throttle_probe(struct acpi_cpu_softc
*sc
);
150 static void acpi_cpu_power_profile(void *arg
);
151 static int acpi_cpu_cx_probe(struct acpi_cpu_softc
*sc
);
152 static int acpi_cpu_cx_cst(struct acpi_cpu_softc
*sc
);
153 static void acpi_cpu_startup(void *arg
);
154 static void acpi_cpu_startup_throttling(void);
155 static void acpi_cpu_startup_cx(void);
156 static void acpi_cpu_throttle_set(uint32_t speed
);
157 static void acpi_cpu_idle(void);
158 static void acpi_cpu_c1(void);
159 static void acpi_cpu_notify(ACPI_HANDLE h
, UINT32 notify
, void *context
);
160 static int acpi_cpu_quirks(struct acpi_cpu_softc
*sc
);
161 static int acpi_cpu_throttle_sysctl(SYSCTL_HANDLER_ARGS
);
162 static int acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS
);
163 static int acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS
);
165 static device_method_t acpi_cpu_methods
[] = {
166 /* Device interface */
167 DEVMETHOD(device_probe
, acpi_cpu_probe
),
168 DEVMETHOD(device_attach
, acpi_cpu_attach
),
169 DEVMETHOD(device_shutdown
, acpi_cpu_shutdown
),
174 static driver_t acpi_cpu_driver
= {
177 sizeof(struct acpi_cpu_softc
),
180 static devclass_t acpi_cpu_devclass
;
181 DRIVER_MODULE(cpu
, acpi
, acpi_cpu_driver
, acpi_cpu_devclass
, 0, 0);
182 MODULE_DEPEND(cpu
, acpi
, 1, 1, 1);
185 acpi_cpu_probe(device_t dev
)
187 int acpi_id
, cpu_id
, cx_count
;
194 if (acpi_disabled("cpu") || acpi_get_type(dev
) != ACPI_TYPE_PROCESSOR
)
197 handle
= acpi_get_handle(dev
);
198 if (cpu_softc
== NULL
)
199 cpu_softc
= malloc(sizeof(struct acpi_cpu_softc
*) *
200 SMP_MAXCPU
, M_TEMP
/* XXX */, M_INTWAIT
| M_ZERO
);
202 /* Get our Processor object. */
204 buf
.Length
= ACPI_ALLOCATE_BUFFER
;
205 status
= AcpiEvaluateObject(handle
, NULL
, NULL
, &buf
);
206 if (ACPI_FAILURE(status
)) {
207 device_printf(dev
, "probe failed to get Processor obj - %s\n",
208 AcpiFormatException(status
));
211 obj
= (ACPI_OBJECT
*)buf
.Pointer
;
212 if (obj
->Type
!= ACPI_TYPE_PROCESSOR
) {
213 device_printf(dev
, "Processor object has bad type %d\n", obj
->Type
);
219 * Find the processor associated with our unit. We could use the
220 * ProcId as a key, however, some boxes do not have the same values
221 * in their Processor object as the ProcId values in the MADT.
223 acpi_id
= obj
->Processor
.ProcId
;
225 if (acpi_pcpu_get_id(device_get_unit(dev
), &acpi_id
, &cpu_id
) != 0)
229 * Check if we already probed this processor. We scan the bus twice
230 * so it's possible we've already seen this one.
232 if (cpu_softc
[cpu_id
] != NULL
)
235 /* Get a count of Cx states for our device string. */
238 buf
.Length
= ACPI_ALLOCATE_BUFFER
;
239 status
= AcpiEvaluateObject(handle
, "_CST", NULL
, &buf
);
240 if (ACPI_SUCCESS(status
)) {
241 obj
= (ACPI_OBJECT
*)buf
.Pointer
;
242 if (ACPI_PKG_VALID(obj
, 2))
243 acpi_PkgInt32(obj
, 0, &cx_count
);
246 if (AcpiGbl_FADT
->Plvl2Lat
<= 100)
248 if (AcpiGbl_FADT
->Plvl3Lat
<= 1000)
254 snprintf(msg
, sizeof(msg
), "ACPI CPU (%d Cx states)", cx_count
);
256 strlcpy(msg
, "ACPI CPU", sizeof(msg
));
257 device_set_desc_copy(dev
, msg
);
259 /* Mark this processor as in-use and save our derived id for attach. */
260 cpu_softc
[cpu_id
] = (void *)1;
261 acpi_set_magic(dev
, cpu_id
);
267 acpi_cpu_attach(device_t dev
)
271 struct acpi_cpu_softc
*sc
;
272 struct acpi_softc
*acpi_sc
;
276 ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__
);
280 sc
= device_get_softc(dev
);
282 sc
->cpu_handle
= acpi_get_handle(dev
);
283 cpu_softc
[acpi_get_magic(dev
)] = sc
;
286 buf
.Length
= ACPI_ALLOCATE_BUFFER
;
287 status
= AcpiEvaluateObject(sc
->cpu_handle
, NULL
, NULL
, &buf
);
288 if (ACPI_FAILURE(status
)) {
289 device_printf(dev
, "attach failed to get Processor obj - %s\n",
290 AcpiFormatException(status
));
293 obj
= (ACPI_OBJECT
*)buf
.Pointer
;
294 sc
->cpu_p_blk
= obj
->Processor
.PblkAddress
;
295 sc
->cpu_p_blk_len
= obj
->Processor
.PblkLength
;
296 sc
->acpi_id
= obj
->Processor
.ProcId
;
298 ACPI_DEBUG_PRINT((ACPI_DB_INFO
, "acpi_cpu%d: P_BLK at %#x/%d\n",
299 device_get_unit(dev
), sc
->cpu_p_blk
, sc
->cpu_p_blk_len
));
301 acpi_sc
= acpi_device_get_parent_softc(dev
);
302 sysctl_ctx_init(&acpi_cpu_sysctl_ctx
);
303 acpi_cpu_sysctl_tree
= SYSCTL_ADD_NODE(&acpi_cpu_sysctl_ctx
,
304 SYSCTL_CHILDREN(acpi_sc
->acpi_sysctl_tree
),
305 OID_AUTO
, "cpu", CTLFLAG_RD
, 0, "");
307 /* If this is the first device probed, check for quirks. */
308 if (device_get_unit(dev
) == 0)
312 * Probe for throttling and Cx state support.
313 * If none of these is present, free up unused resources.
315 thr_ret
= acpi_cpu_throttle_probe(sc
);
316 cx_ret
= acpi_cpu_cx_probe(sc
);
317 if (thr_ret
== 0 || cx_ret
== 0) {
318 status
= AcpiInstallNotifyHandler(sc
->cpu_handle
, ACPI_DEVICE_NOTIFY
,
319 acpi_cpu_notify
, sc
);
320 if (device_get_unit(dev
) == 0)
321 AcpiOsQueueForExecution(OSD_PRIORITY_LO
, acpi_cpu_startup
, NULL
);
323 sysctl_ctx_free(&acpi_cpu_sysctl_ctx
);
330 * Find the nth present CPU and return its pc_cpuid as well as set the
331 * pc_acpi_id from the most reliable source.
334 acpi_pcpu_get_id(uint32_t idx
, uint32_t *acpi_id
, uint32_t *cpu_id
)
336 struct mdglobaldata
*md
;
339 KASSERT(acpi_id
!= NULL
, ("Null acpi_id"));
340 KASSERT(cpu_id
!= NULL
, ("Null cpu_id"));
341 for (i
= 0; i
<= ncpus
; i
++) {
342 if ((smp_active_mask
& (1 << i
)) == 0)
344 md
= (struct mdglobaldata
*)globaldata_find(i
);
345 KASSERT(md
!= NULL
, ("no pcpu data for %d", i
));
348 * If pc_acpi_id was not initialized (e.g., a non-APIC UP box)
349 * override it with the value from the ASL. Otherwise, if the
350 * two don't match, prefer the MADT-derived value. Finally,
351 * return the pc_cpuid to reference this processor.
353 if (md
->gd_acpi_id
== 0xffffffff)
354 md
->gd_acpi_id
= *acpi_id
;
355 else if (md
->gd_acpi_id
!= *acpi_id
)
356 *acpi_id
= md
->gd_acpi_id
;
357 *cpu_id
= md
->mi
.gd_cpuid
;
366 acpi_cpu_shutdown(device_t dev
)
368 ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__
);
370 /* Disable any entry to the idle function. */
373 /* Signal and wait for all processors to exit acpi_cpu_idle(). */
375 /*smp_rendezvous(NULL, NULL, NULL, NULL);*/
376 KKASSERT(0); /* XXX use rendezvous */
384 acpi_cpu_throttle_probe(struct acpi_cpu_softc
*sc
)
389 ACPI_GENERIC_ADDRESS gas
;
392 ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__
);
396 /* Get throttling parameters from the FADT. 0 means not supported. */
397 if (device_get_unit(sc
->cpu_dev
) == 0) {
398 cpu_smi_cmd
= AcpiGbl_FADT
->SmiCmd
;
399 cpu_pstate_cnt
= AcpiGbl_FADT
->PstateCnt
;
400 cpu_cst_cnt
= AcpiGbl_FADT
->CstCnt
;
401 cpu_duty_offset
= AcpiGbl_FADT
->DutyOffset
;
402 cpu_duty_width
= AcpiGbl_FADT
->DutyWidth
;
404 if (cpu_duty_width
== 0 || (cpu_quirks
& CPU_QUIRK_NO_THROTTLE
) != 0)
407 /* Validate the duty offset/width. */
408 duty_end
= cpu_duty_offset
+ cpu_duty_width
- 1;
410 device_printf(sc
->cpu_dev
, "CLK_VAL field overflows P_CNT register\n");
413 if (cpu_duty_offset
<= 4 && duty_end
>= 4) {
414 device_printf(sc
->cpu_dev
, "CLK_VAL field overlaps THT_EN bit\n");
419 * If not present, fall back to using the processor's P_BLK to find
420 * the P_CNT register.
422 * Note that some systems seem to duplicate the P_BLK pointer
423 * across multiple CPUs, so not getting the resource is not fatal.
426 buf
.Length
= sizeof(obj
);
427 status
= AcpiEvaluateObject(sc
->cpu_handle
, "_PTC", NULL
, &buf
);
428 if (ACPI_SUCCESS(status
)) {
429 if (obj
.Buffer
.Length
< sizeof(ACPI_GENERIC_ADDRESS
) + 3) {
430 device_printf(sc
->cpu_dev
, "_PTC buffer too small\n");
433 memcpy(&gas
, obj
.Buffer
.Pointer
+ 3, sizeof(gas
));
434 sc
->cpu_p_cnt
= acpi_bus_alloc_gas(sc
->cpu_dev
, &cpu_rid
, &gas
);
435 if (sc
->cpu_p_cnt
!= NULL
) {
436 ACPI_DEBUG_PRINT((ACPI_DB_INFO
, "acpi_cpu%d: P_CNT from _PTC\n",
437 device_get_unit(sc
->cpu_dev
)));
441 /* If _PTC not present or other failure, try the P_BLK. */
442 if (sc
->cpu_p_cnt
== NULL
) {
444 * The spec says P_BLK must be 6 bytes long. However, some
445 * systems use it to indicate a fractional set of features
446 * present so we take anything >= 4.
448 if (sc
->cpu_p_blk_len
< 4)
450 gas
.Address
= sc
->cpu_p_blk
;
451 gas
.AddressSpaceId
= ACPI_ADR_SPACE_SYSTEM_IO
;
452 gas
.RegisterBitWidth
= 32;
453 sc
->cpu_p_cnt
= acpi_bus_alloc_gas(sc
->cpu_dev
, &cpu_rid
, &gas
);
454 if (sc
->cpu_p_cnt
!= NULL
) {
455 ACPI_DEBUG_PRINT((ACPI_DB_INFO
, "acpi_cpu%d: P_CNT from P_BLK\n",
456 device_get_unit(sc
->cpu_dev
)));
458 device_printf(sc
->cpu_dev
, "Failed to attach throttling P_CNT\n");
468 acpi_cpu_cx_probe(struct acpi_cpu_softc
*sc
)
470 ACPI_GENERIC_ADDRESS gas
;
471 struct acpi_cx
*cx_ptr
;
474 ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__
);
476 /* Bus mastering arbitration control is needed for C3. */
477 if (AcpiGbl_FADT
->V1_Pm2CntBlk
== 0 || AcpiGbl_FADT
->Pm2CntLen
== 0) {
478 cpu_quirks
|= CPU_QUIRK_NO_C3
;
479 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
480 "acpi_cpu%d: No BM control, C3 disabled\n",
481 device_get_unit(sc
->cpu_dev
)));
485 * First, check for the ACPI 2.0 _CST sleep states object.
486 * If not usable, fall back to the P_BLK's P_LVL2 and P_LVL3.
488 sc
->cpu_cx_count
= 0;
489 error
= acpi_cpu_cx_cst(sc
);
491 cx_ptr
= sc
->cpu_cx_states
;
493 /* C1 has been required since just after ACPI 1.0 */
494 cx_ptr
->type
= ACPI_STATE_C1
;
495 cx_ptr
->trans_lat
= 0;
501 * The spec says P_BLK must be 6 bytes long. However, some systems
502 * use it to indicate a fractional set of features present so we
503 * take 5 as C2. Some may also have a value of 7 to indicate
504 * another C3 but most use _CST for this (as required) and having
505 * "only" C1-C3 is not a hardship.
507 if (sc
->cpu_p_blk_len
< 5)
510 /* Validate and allocate resources for C2 (P_LVL2). */
511 gas
.AddressSpaceId
= ACPI_ADR_SPACE_SYSTEM_IO
;
512 gas
.RegisterBitWidth
= 8;
513 if (AcpiGbl_FADT
->Plvl2Lat
<= 100) {
514 gas
.Address
= sc
->cpu_p_blk
+ 4;
515 cx_ptr
->p_lvlx
= acpi_bus_alloc_gas(sc
->cpu_dev
, &cpu_rid
, &gas
);
516 if (cx_ptr
->p_lvlx
!= NULL
) {
518 cx_ptr
->type
= ACPI_STATE_C2
;
519 cx_ptr
->trans_lat
= AcpiGbl_FADT
->Plvl2Lat
;
525 if (sc
->cpu_p_blk_len
< 6)
528 /* Validate and allocate resources for C3 (P_LVL3). */
529 if (AcpiGbl_FADT
->Plvl3Lat
<= 1000 &&
530 (cpu_quirks
& CPU_QUIRK_NO_C3
) == 0) {
532 gas
.Address
= sc
->cpu_p_blk
+ 5;
533 cx_ptr
->p_lvlx
= acpi_bus_alloc_gas(sc
->cpu_dev
, &cpu_rid
, &gas
);
534 if (cx_ptr
->p_lvlx
!= NULL
) {
536 cx_ptr
->type
= ACPI_STATE_C3
;
537 cx_ptr
->trans_lat
= AcpiGbl_FADT
->Plvl3Lat
;
545 /* If no valid registers were found, don't attach. */
546 if (sc
->cpu_cx_count
== 0)
549 /* Use initial sleep value of 1 sec. to start with lowest idle state. */
550 sc
->cpu_prev_sleep
= 1000000;
556 * Parse a _CST package and set up its Cx states. Since the _CST object
557 * can change dynamically, our notify handler may call this function
558 * to clean up and probe the new _CST package.
561 acpi_cpu_cx_cst(struct acpi_cpu_softc
*sc
)
563 struct acpi_cx
*cx_ptr
;
571 ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__
);
574 buf
.Length
= ACPI_ALLOCATE_BUFFER
;
575 status
= AcpiEvaluateObject(sc
->cpu_handle
, "_CST", NULL
, &buf
);
576 if (ACPI_FAILURE(status
))
579 /* _CST is a package with a count and at least one Cx package. */
580 top
= (ACPI_OBJECT
*)buf
.Pointer
;
581 if (!ACPI_PKG_VALID(top
, 2) || acpi_PkgInt32(top
, 0, &count
) != 0) {
582 device_printf(sc
->cpu_dev
, "Invalid _CST package\n");
583 AcpiOsFree(buf
.Pointer
);
586 if (count
!= top
->Package
.Count
- 1) {
587 device_printf(sc
->cpu_dev
, "Invalid _CST state count (%d != %d)\n",
588 count
, top
->Package
.Count
- 1);
589 count
= top
->Package
.Count
- 1;
591 if (count
> MAX_CX_STATES
) {
592 device_printf(sc
->cpu_dev
, "_CST has too many states (%d)\n", count
);
593 count
= MAX_CX_STATES
;
596 /* Set up all valid states. */
597 sc
->cpu_cx_count
= 0;
598 cx_ptr
= sc
->cpu_cx_states
;
599 for (i
= 0; i
< count
; i
++) {
600 pkg
= &top
->Package
.Elements
[i
+ 1];
601 if (!ACPI_PKG_VALID(pkg
, 4) ||
602 acpi_PkgInt32(pkg
, 1, &cx_ptr
->type
) != 0 ||
603 acpi_PkgInt32(pkg
, 2, &cx_ptr
->trans_lat
) != 0 ||
604 acpi_PkgInt32(pkg
, 3, &cx_ptr
->power
) != 0) {
606 device_printf(sc
->cpu_dev
, "Skipping invalid Cx state package\n");
610 /* Validate the state to see if we should use it. */
611 switch (cx_ptr
->type
) {
618 if (cx_ptr
->trans_lat
> 100) {
619 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
620 "acpi_cpu%d: C2[%d] not available.\n",
621 device_get_unit(sc
->cpu_dev
), i
));
628 if (cx_ptr
->trans_lat
> 1000 ||
629 (cpu_quirks
& CPU_QUIRK_NO_C3
) != 0) {
631 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
632 "acpi_cpu%d: C3[%d] not available.\n",
633 device_get_unit(sc
->cpu_dev
), i
));
640 /* Free up any previous register. */
641 if (cx_ptr
->p_lvlx
!= NULL
) {
642 bus_release_resource(sc
->cpu_dev
, 0, 0, cx_ptr
->p_lvlx
);
643 cx_ptr
->p_lvlx
= NULL
;
647 /* Allocate the control register for C2 or C3. */
648 acpi_PkgGas(sc
->cpu_dev
, pkg
, 0, &cpu_rid
, &cx_ptr
->p_lvlx
);
649 if (cx_ptr
->p_lvlx
!= NULL
) {
651 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
652 "acpi_cpu%d: Got C%d - %d latency\n",
653 device_get_unit(sc
->cpu_dev
), cx_ptr
->type
,
659 AcpiOsFree(buf
.Pointer
);
665 * Call this *after* all CPUs have been attached.
668 acpi_cpu_startup(void *arg
)
670 struct acpi_cpu_softc
*sc
;
673 /* Get set of CPU devices */
674 devclass_get_devices(acpi_cpu_devclass
, &cpu_devices
, &cpu_ndevices
);
677 * Make sure all the processors' Cx counts match. We should probably
678 * also check the contents of each. However, no known systems have
679 * non-matching Cx counts so we'll deal with this later.
681 count
= MAX_CX_STATES
;
682 for (i
= 0; i
< cpu_ndevices
; i
++) {
683 sc
= device_get_softc(cpu_devices
[i
]);
684 count
= min(sc
->cpu_cx_count
, count
);
686 cpu_cx_count
= count
;
688 /* Perform throttling and Cx final initialization. */
689 sc
= device_get_softc(cpu_devices
[0]);
690 if (sc
->cpu_p_cnt
!= NULL
)
691 acpi_cpu_startup_throttling();
692 if (cpu_cx_count
> 0)
693 acpi_cpu_startup_cx();
695 /* register performance profile change handler */
696 EVENTHANDLER_REGISTER(power_profile_change
, acpi_cpu_power_profile
, NULL
, 0);
700 * Power profile change hook.
702 * Uses the ACPI lock to avoid reentrancy.
705 acpi_cpu_power_profile(void *arg
)
711 state
= power_profile_get_state();
712 if (state
!= POWER_PROFILE_PERFORMANCE
&&
713 state
!= POWER_PROFILE_ECONOMY
) {
719 case POWER_PROFILE_PERFORMANCE
:
720 speed
= cpu_throttle_performance
;
722 case POWER_PROFILE_ECONOMY
:
723 speed
= cpu_throttle_economy
;
726 speed
= cpu_throttle_state
;
729 if (speed
!= cpu_throttle_state
)
730 acpi_cpu_throttle_set(speed
);
735 * Takes the ACPI lock to avoid fighting anyone over the SMI command
739 acpi_cpu_startup_throttling()
743 /* Initialise throttling states */
744 cpu_throttle_max
= CPU_MAX_SPEED
;
745 cpu_throttle_state
= CPU_MAX_SPEED
;
746 cpu_throttle_performance
= cpu_throttle_max
;
747 cpu_throttle_economy
= cpu_throttle_performance
/ 2;
749 SYSCTL_ADD_INT(&acpi_cpu_sysctl_ctx
,
750 SYSCTL_CHILDREN(acpi_cpu_sysctl_tree
),
751 OID_AUTO
, "throttle_max", CTLFLAG_RD
,
752 &cpu_throttle_max
, 0, "maximum CPU speed");
753 SYSCTL_ADD_PROC(&acpi_cpu_sysctl_ctx
,
754 SYSCTL_CHILDREN(acpi_cpu_sysctl_tree
),
755 OID_AUTO
, "throttle_state",
756 CTLTYPE_INT
| CTLFLAG_RW
, &cpu_throttle_state
,
757 0, acpi_cpu_throttle_sysctl
, "I", "current CPU speed");
760 * Performance/Economy throttle settings
762 SYSCTL_ADD_PROC(&acpi_cpu_sysctl_ctx
,
763 SYSCTL_CHILDREN(acpi_cpu_sysctl_tree
),
764 OID_AUTO
, "performance_speed",
765 CTLTYPE_INT
| CTLFLAG_RW
, &cpu_throttle_performance
,
766 0, acpi_cpu_throttle_sysctl
, "I", "performance CPU speed");
767 SYSCTL_ADD_PROC(&acpi_cpu_sysctl_ctx
,
768 SYSCTL_CHILDREN(acpi_cpu_sysctl_tree
),
769 OID_AUTO
, "economy_speed",
770 CTLTYPE_INT
| CTLFLAG_RW
, &cpu_throttle_economy
,
771 0, acpi_cpu_throttle_sysctl
, "I", "economy CPU speed");
773 /* If ACPI 2.0+, signal platform that we are taking over throttling. */
775 if (cpu_pstate_cnt
!= 0)
776 AcpiOsWritePort(cpu_smi_cmd
, cpu_pstate_cnt
, 8);
778 /* Set initial speed to maximum. */
779 acpi_cpu_throttle_set(cpu_throttle_max
);
782 printf("acpi_cpu: throttling enabled, %d steps (100%% to %d.%d%%), "
783 "currently %d.%d%%\n", CPU_MAX_SPEED
, CPU_SPEED_PRINTABLE(1),
784 CPU_SPEED_PRINTABLE(cpu_throttle_state
));
788 extern void (*cpu_idle_hook
)(void);
791 acpi_cpu_startup_cx()
793 struct acpi_cpu_softc
*sc
;
797 sc
= device_get_softc(cpu_devices
[0]);
798 sbuf_new(&sb
, cpu_cx_supported
, sizeof(cpu_cx_supported
), SBUF_FIXEDLEN
);
799 for (i
= 0; i
< cpu_cx_count
; i
++)
800 sbuf_printf(&sb
, "C%d/%d ", i
+ 1, sc
->cpu_cx_states
[i
].trans_lat
);
803 SYSCTL_ADD_STRING(&acpi_cpu_sysctl_ctx
,
804 SYSCTL_CHILDREN(acpi_cpu_sysctl_tree
),
805 OID_AUTO
, "cx_supported", CTLFLAG_RD
, cpu_cx_supported
,
806 0, "Cx/microsecond values for supported Cx states");
807 SYSCTL_ADD_PROC(&acpi_cpu_sysctl_ctx
,
808 SYSCTL_CHILDREN(acpi_cpu_sysctl_tree
),
809 OID_AUTO
, "cx_lowest", CTLTYPE_STRING
| CTLFLAG_RW
,
810 NULL
, 0, acpi_cpu_cx_lowest_sysctl
, "A",
811 "lowest Cx sleep state to use");
812 SYSCTL_ADD_PROC(&acpi_cpu_sysctl_ctx
,
813 SYSCTL_CHILDREN(acpi_cpu_sysctl_tree
),
814 OID_AUTO
, "cx_usage", CTLTYPE_STRING
| CTLFLAG_RD
,
815 NULL
, 0, acpi_cpu_usage_sysctl
, "A",
816 "percent usage for each Cx state");
819 /* Signal platform that we can handle _CST notification. */
820 if (cpu_cst_cnt
!= 0) {
822 AcpiOsWritePort(cpu_smi_cmd
, cpu_cst_cnt
, 8);
827 /* Take over idling from cpu_idle_default_hook(). */
829 /* XXX only set this if ncpus == 1, for now XXX */
830 cpu_idle_hook
= acpi_cpu_idle
;
834 * Set CPUs to the new state.
836 * Must be called with the ACPI lock held.
839 acpi_cpu_throttle_set(uint32_t speed
)
841 struct acpi_cpu_softc
*sc
;
843 uint32_t p_cnt
, clk_val
;
847 /* Iterate over processors */
848 for (i
= 0; i
< cpu_ndevices
; i
++) {
849 sc
= device_get_softc(cpu_devices
[i
]);
850 if (sc
->cpu_p_cnt
== NULL
)
853 /* Get the current P_CNT value and disable throttling */
854 p_cnt
= CPU_GET_REG(sc
->cpu_p_cnt
, 4);
855 p_cnt
&= ~CPU_P_CNT_THT_EN
;
856 CPU_SET_REG(sc
->cpu_p_cnt
, 4, p_cnt
);
858 /* If we're at maximum speed, that's all */
859 if (speed
< CPU_MAX_SPEED
) {
860 /* Mask the old CLK_VAL off and or-in the new value */
861 clk_val
= (CPU_MAX_SPEED
- 1) << cpu_duty_offset
;
863 p_cnt
|= (speed
<< cpu_duty_offset
);
865 /* Write the new P_CNT value and then enable throttling */
866 CPU_SET_REG(sc
->cpu_p_cnt
, 4, p_cnt
);
867 p_cnt
|= CPU_P_CNT_THT_EN
;
868 CPU_SET_REG(sc
->cpu_p_cnt
, 4, p_cnt
);
870 ACPI_VPRINT(sc
->cpu_dev
, acpi_device_get_parent_softc(sc
->cpu_dev
),
871 "set speed to %d.%d%%\n", CPU_SPEED_PRINTABLE(speed
));
873 cpu_throttle_state
= speed
;
877 * Idle the CPU in the lowest state possible. This function is called with
878 * interrupts disabled. Note that once it re-enables interrupts, a task
879 * switch can occur so do not access shared data (i.e. the softc) after
880 * interrupts are re-enabled.
885 struct acpi_cpu_softc
*sc
;
886 struct acpi_cx
*cx_next
;
887 uint32_t start_time
, end_time
;
888 int bm_active
, cx_next_idx
, i
;
890 /* If disabled, return immediately. */
891 if (cpu_cx_count
== 0) {
897 * Look up our CPU id to get our softc. If it's NULL, we'll use C1
898 * since there is no ACPI processor object for this CPU. This occurs
899 * for logical CPUs in the HTT case.
901 sc
= cpu_softc
[mdcpu
->mi
.gd_cpuid
];
908 * If we slept 100 us or more, use the lowest Cx state. Otherwise,
909 * find the lowest state that has a latency less than or equal to
910 * the length of our last sleep.
912 cx_next_idx
= cpu_cx_lowest
;
913 if (sc
->cpu_prev_sleep
< 100)
914 for (i
= cpu_cx_lowest
; i
>= 0; i
--)
915 if (sc
->cpu_cx_states
[i
].trans_lat
<= sc
->cpu_prev_sleep
) {
921 * Check for bus master activity. If there was activity, clear
922 * the bit and use the lowest non-C3 state. Note that the USB
923 * driver polling for new devices keeps this bit set all the
924 * time if USB is loaded.
926 AcpiGetRegister(ACPI_BITREG_BUS_MASTER_STATUS
, &bm_active
,
927 ACPI_MTX_DO_NOT_LOCK
);
928 if (bm_active
!= 0) {
929 AcpiSetRegister(ACPI_BITREG_BUS_MASTER_STATUS
, 1,
930 ACPI_MTX_DO_NOT_LOCK
);
931 cx_next_idx
= min(cx_next_idx
, cpu_non_c3
);
934 /* Select the next state and update statistics. */
935 cx_next
= &sc
->cpu_cx_states
[cx_next_idx
];
936 cpu_cx_stats
[cx_next_idx
]++;
937 KASSERT(cx_next
->type
!= ACPI_STATE_C0
, ("acpi_cpu_idle: C0 sleep"));
940 * Execute HLT (or equivalent) and wait for an interrupt. We can't
941 * calculate the time spent in C1 since the place we wake up is an
942 * ISR. Assume we slept one quantum and return.
944 if (cx_next
->type
== ACPI_STATE_C1
) {
945 sc
->cpu_prev_sleep
= 1000000 / hz
;
950 /* For C3, disable bus master arbitration and enable bus master wake. */
951 if (cx_next
->type
== ACPI_STATE_C3
) {
952 AcpiSetRegister(ACPI_BITREG_ARB_DISABLE
, 1, ACPI_MTX_DO_NOT_LOCK
);
953 AcpiSetRegister(ACPI_BITREG_BUS_MASTER_RLD
, 1, ACPI_MTX_DO_NOT_LOCK
);
957 * Read from P_LVLx to enter C2(+), checking time spent asleep.
958 * Use the ACPI timer for measuring sleep time. Since we need to
959 * get the time very close to the CPU start/stop clock logic, this
960 * is the only reliable time source.
962 AcpiHwLowLevelRead(32, &start_time
, &AcpiGbl_FADT
->XPmTmrBlk
);
963 CPU_GET_REG(cx_next
->p_lvlx
, 1);
966 * Read the end time twice. Since it may take an arbitrary time
967 * to enter the idle state, the first read may be executed before
968 * the processor has stopped. Doing it again provides enough
969 * margin that we are certain to have a correct value.
971 AcpiHwLowLevelRead(32, &end_time
, &AcpiGbl_FADT
->XPmTmrBlk
);
972 AcpiHwLowLevelRead(32, &end_time
, &AcpiGbl_FADT
->XPmTmrBlk
);
974 /* Enable bus master arbitration and disable bus master wakeup. */
975 if (cx_next
->type
== ACPI_STATE_C3
) {
976 AcpiSetRegister(ACPI_BITREG_ARB_DISABLE
, 0, ACPI_MTX_DO_NOT_LOCK
);
977 AcpiSetRegister(ACPI_BITREG_BUS_MASTER_RLD
, 0, ACPI_MTX_DO_NOT_LOCK
);
980 /* Find the actual time asleep in microseconds, minus overhead. */
981 end_time
= acpi_TimerDelta(end_time
, start_time
);
982 sc
->cpu_prev_sleep
= PM_USEC(end_time
) - cx_next
->trans_lat
;
986 /* Put the CPU in C1 in a machine-dependant way. */
991 ia64_call_pal_static(PAL_HALT_LIGHT
, 0, 0, 0);
995 if (!lwkt_runnable())
996 __asm
__volatile("sti; hlt");
998 __asm
__volatile("sti; pause");
1000 if (!lwkt_runnable())
1001 __asm
__volatile("sti; hlt");
1003 __asm
__volatile("sti");
1005 #endif /* !__ia64__ */
1009 * Re-evaluate the _PSS and _CST objects when we are notified that they
1012 * XXX Re-evaluation disabled until locking is done.
1015 acpi_cpu_notify(ACPI_HANDLE h
, UINT32 notify
, void *context
)
1017 struct acpi_cpu_softc
*sc
= (struct acpi_cpu_softc
*)context
;
1020 case ACPI_CPU_NOTIFY_PERF_STATES
:
1021 device_printf(sc
->cpu_dev
, "Performance states changed\n");
1022 /* acpi_cpu_px_available(sc); */
1024 case ACPI_CPU_NOTIFY_CX_STATES
:
1025 device_printf(sc
->cpu_dev
, "Cx states changed\n");
1026 /* acpi_cpu_cx_cst(sc); */
1029 device_printf(sc
->cpu_dev
, "Unknown notify %#x\n", notify
);
1035 acpi_cpu_quirks(struct acpi_cpu_softc
*sc
)
1039 * C3 is not supported on multiple CPUs since this would require
1040 * flushing all caches which is currently too expensive.
1043 cpu_quirks
|= CPU_QUIRK_NO_C3
;
1046 /* Look for various quirks of the PIIX4 part. */
1047 acpi_dev
= pci_find_device(PCI_VENDOR_INTEL
, PCI_DEVICE_82371AB_3
);
1048 if (acpi_dev
!= NULL
) {
1049 switch (pci_get_revid(acpi_dev
)) {
1051 * Disable throttling control on PIIX4 A and B-step.
1052 * See specification changes #13 ("Manual Throttle Duty Cycle")
1053 * and #14 ("Enabling and Disabling Manual Throttle"), plus
1054 * erratum #5 ("STPCLK# Deassertion Time") from the January
1055 * 2002 PIIX4 specification update. Note that few (if any)
1056 * mobile systems ever used this part.
1058 case PCI_REVISION_A_STEP
:
1059 case PCI_REVISION_B_STEP
:
1060 cpu_quirks
|= CPU_QUIRK_NO_THROTTLE
;
1063 * Disable C3 support for all PIIX4 chipsets. Some of these parts
1064 * do not report the BMIDE status to the BM status register and
1065 * others have a livelock bug if Type-F DMA is enabled. Linux
1066 * works around the BMIDE bug by reading the BM status directly
1067 * but we take the simpler approach of disabling C3 for these
1070 * See erratum #18 ("C3 Power State/BMIDE and Type-F DMA
1071 * Livelock") from the January 2002 PIIX4 specification update.
1072 * Applies to all PIIX4 models.
1074 case PCI_REVISION_4E
:
1075 case PCI_REVISION_4M
:
1076 cpu_quirks
|= CPU_QUIRK_NO_C3
;
1087 /* Handle changes in the CPU throttling setting. */
1089 acpi_cpu_throttle_sysctl(SYSCTL_HANDLER_ARGS
)
1096 argp
= (uint32_t *)oidp
->oid_arg1
;
1098 error
= sysctl_handle_int(oidp
, &arg
, 0, req
);
1100 /* Error or no new value */
1101 if (error
!= 0 || req
->newptr
== NULL
)
1103 if (arg
< 1 || arg
> cpu_throttle_max
)
1106 /* If throttling changed, notify the BIOS of the new rate. */
1110 acpi_cpu_throttle_set(arg
);
1118 acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS
)
1123 uintmax_t fract
, sum
, whole
;
1126 for (i
= 0; i
< cpu_cx_count
; i
++)
1127 sum
+= cpu_cx_stats
[i
];
1128 sbuf_new(&sb
, buf
, sizeof(buf
), SBUF_FIXEDLEN
);
1129 for (i
= 0; i
< cpu_cx_count
; i
++) {
1131 whole
= (uintmax_t)cpu_cx_stats
[i
] * 100;
1132 fract
= (whole
% sum
) * 100;
1133 sbuf_printf(&sb
, "%u.%02u%% ", (u_int
)(whole
/ sum
),
1134 (u_int
)(fract
/ sum
));
1136 sbuf_printf(&sb
, "0%% ");
1140 sysctl_handle_string(oidp
, sbuf_data(&sb
), sbuf_len(&sb
), req
);
1147 acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS
)
1149 struct acpi_cpu_softc
*sc
;
1153 sc
= device_get_softc(cpu_devices
[0]);
1154 snprintf(state
, sizeof(state
), "C%d", cpu_cx_lowest
+ 1);
1155 error
= sysctl_handle_string(oidp
, state
, sizeof(state
), req
);
1156 if (error
!= 0 || req
->newptr
== NULL
)
1158 if (strlen(state
) < 2 || toupper(state
[0]) != 'C')
1160 val
= (int) strtol(state
+ 1, NULL
, 10) - 1;
1161 if (val
< 0 || val
> cpu_cx_count
- 1)
1164 cpu_cx_lowest
= val
;
1166 /* If not disabling, cache the new lowest non-C3 state. */
1168 for (i
= cpu_cx_lowest
; i
>= 0; i
--) {
1169 if (sc
->cpu_cx_states
[i
].type
< ACPI_STATE_C3
) {
1175 /* Reset the statistics counters. */
1176 bzero(cpu_cx_stats
, sizeof(cpu_cx_stats
));