xref: /freebsd/sys/dev/acpica/acpi_cpu.c (revision 781d8f4e7638e77445043bc007d455eb8c23676c)
1 /*-
2  * Copyright (c) 2003-2005 Nate Lawson (SDG)
3  * Copyright (c) 2001 Michael Smith
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
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.
14  *
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
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #include "opt_acpi.h"
32 #include <sys/param.h>
33 #include <sys/bus.h>
34 #include <sys/cpu.h>
35 #include <sys/kernel.h>
36 #include <sys/malloc.h>
37 #include <sys/module.h>
38 #include <sys/pcpu.h>
39 #include <sys/power.h>
40 #include <sys/proc.h>
41 #include <sys/sbuf.h>
42 #include <sys/smp.h>
43 
44 #include <dev/pci/pcivar.h>
45 #include <machine/atomic.h>
46 #include <machine/bus.h>
47 #include <sys/rman.h>
48 
49 #include <contrib/dev/acpica/acpi.h>
50 #include <dev/acpica/acpivar.h>
51 
52 /*
53  * Support for ACPI Processor devices, including C[1-3] sleep states.
54  */
55 
56 /* Hooks for the ACPI CA debugging infrastructure */
57 #define _COMPONENT	ACPI_PROCESSOR
58 ACPI_MODULE_NAME("PROCESSOR")
59 
60 struct acpi_cx {
61     struct resource	*p_lvlx;	/* Register to read to enter state. */
62     uint32_t		 type;		/* C1-3 (C4 and up treated as C3). */
63     uint32_t		 trans_lat;	/* Transition latency (usec). */
64     uint32_t		 power;		/* Power consumed (mW). */
65     int			 res_type;	/* Resource type for p_lvlx. */
66 };
67 #define MAX_CX_STATES	 8
68 
69 struct acpi_cpu_softc {
70     device_t		 cpu_dev;
71     ACPI_HANDLE		 cpu_handle;
72     struct pcpu		*cpu_pcpu;
73     uint32_t		 cpu_acpi_id;	/* ACPI processor id */
74     uint32_t		 cpu_p_blk;	/* ACPI P_BLK location */
75     uint32_t		 cpu_p_blk_len;	/* P_BLK length (must be 6). */
76     struct acpi_cx	 cpu_cx_states[MAX_CX_STATES];
77     int			 cpu_cx_count;	/* Number of valid Cx states. */
78     int			 cpu_prev_sleep;/* Last idle sleep duration. */
79     int			 cpu_features;	/* Child driver supported features. */
80     /* Runtime state. */
81     int			 cpu_non_c3;	/* Index of lowest non-C3 state. */
82     int			 cpu_short_slp;	/* Count of < 1us sleeps. */
83     u_int		 cpu_cx_stats[MAX_CX_STATES];/* Cx usage history. */
84     /* Values for sysctl. */
85     struct sysctl_ctx_list cpu_sysctl_ctx;
86     struct sysctl_oid	*cpu_sysctl_tree;
87     int			 cpu_cx_lowest;
88     char 		 cpu_cx_supported[64];
89     int			 cpu_rid;
90 };
91 
92 struct acpi_cpu_device {
93     struct resource_list	ad_rl;
94 };
95 
96 #define CPU_GET_REG(reg, width) 					\
97     (bus_space_read_ ## width(rman_get_bustag((reg)), 			\
98 		      rman_get_bushandle((reg)), 0))
99 #define CPU_SET_REG(reg, width, val)					\
100     (bus_space_write_ ## width(rman_get_bustag((reg)), 			\
101 		       rman_get_bushandle((reg)), 0, (val)))
102 
103 #define PM_USEC(x)	 ((x) >> 2)	/* ~4 clocks per usec (3.57955 Mhz) */
104 
105 #define ACPI_NOTIFY_CX_STATES	0x81	/* _CST changed. */
106 
107 #define CPU_QUIRK_NO_C3		(1<<0)	/* C3-type states are not usable. */
108 #define CPU_QUIRK_NO_BM_CTRL	(1<<2)	/* No bus mastering control. */
109 
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
116 
117 /* Platform hardware resource information. */
118 static uint32_t		 cpu_smi_cmd;	/* Value to write to SMI_CMD. */
119 static uint8_t		 cpu_cst_cnt;	/* Indicate we are _CST aware. */
120 static int		 cpu_quirks;	/* Indicate any hardware bugs. */
121 
122 /* Runtime state. */
123 static int		 cpu_disable_idle; /* Disable entry to idle function */
124 static int		 cpu_cx_count;	/* Number of valid Cx states */
125 
126 /* Values for sysctl. */
127 static struct sysctl_ctx_list cpu_sysctl_ctx;
128 static struct sysctl_oid *cpu_sysctl_tree;
129 static int		 cpu_cx_generic;
130 static int		 cpu_cx_lowest;
131 
132 static device_t		*cpu_devices;
133 static int		 cpu_ndevices;
134 static struct acpi_cpu_softc **cpu_softc;
135 ACPI_SERIAL_DECL(cpu, "ACPI CPU");
136 
137 static int	acpi_cpu_probe(device_t dev);
138 static int	acpi_cpu_attach(device_t dev);
139 static int	acpi_cpu_suspend(device_t dev);
140 static int	acpi_cpu_resume(device_t dev);
141 static int	acpi_pcpu_get_id(uint32_t idx, uint32_t *acpi_id,
142 		    uint32_t *cpu_id);
143 static struct resource_list *acpi_cpu_get_rlist(device_t dev, device_t child);
144 static device_t	acpi_cpu_add_child(device_t dev, int order, const char *name,
145 		    int unit);
146 static int	acpi_cpu_read_ivar(device_t dev, device_t child, int index,
147 		    uintptr_t *result);
148 static int	acpi_cpu_shutdown(device_t dev);
149 static void	acpi_cpu_cx_probe(struct acpi_cpu_softc *sc);
150 static void	acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc);
151 static int	acpi_cpu_cx_cst(struct acpi_cpu_softc *sc);
152 static void	acpi_cpu_startup(void *arg);
153 static void	acpi_cpu_startup_cx(struct acpi_cpu_softc *sc);
154 static void	acpi_cpu_idle(void);
155 static void	acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context);
156 static int	acpi_cpu_quirks(void);
157 static int	acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS);
158 static int	acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc, int val);
159 static int	acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
160 static int	acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
161 
162 static device_method_t acpi_cpu_methods[] = {
163     /* Device interface */
164     DEVMETHOD(device_probe,	acpi_cpu_probe),
165     DEVMETHOD(device_attach,	acpi_cpu_attach),
166     DEVMETHOD(device_detach,	bus_generic_detach),
167     DEVMETHOD(device_shutdown,	acpi_cpu_shutdown),
168     DEVMETHOD(device_suspend,	acpi_cpu_suspend),
169     DEVMETHOD(device_resume,	acpi_cpu_resume),
170 
171     /* Bus interface */
172     DEVMETHOD(bus_add_child,	acpi_cpu_add_child),
173     DEVMETHOD(bus_read_ivar,	acpi_cpu_read_ivar),
174     DEVMETHOD(bus_get_resource_list, acpi_cpu_get_rlist),
175     DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
176     DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
177     DEVMETHOD(bus_alloc_resource, bus_generic_rl_alloc_resource),
178     DEVMETHOD(bus_release_resource, bus_generic_rl_release_resource),
179     DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
180     DEVMETHOD(bus_activate_resource, bus_generic_activate_resource),
181     DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource),
182     DEVMETHOD(bus_setup_intr,	bus_generic_setup_intr),
183     DEVMETHOD(bus_teardown_intr, bus_generic_teardown_intr),
184 
185     {0, 0}
186 };
187 
188 static driver_t acpi_cpu_driver = {
189     "cpu",
190     acpi_cpu_methods,
191     sizeof(struct acpi_cpu_softc),
192 };
193 
194 static devclass_t acpi_cpu_devclass;
195 DRIVER_MODULE(cpu, acpi, acpi_cpu_driver, acpi_cpu_devclass, 0, 0);
196 MODULE_DEPEND(cpu, acpi, 1, 1, 1);
197 
198 static int
199 acpi_cpu_probe(device_t dev)
200 {
201     int			   acpi_id, cpu_id;
202     ACPI_BUFFER		   buf;
203     ACPI_HANDLE		   handle;
204     ACPI_OBJECT		   *obj;
205     ACPI_STATUS		   status;
206 
207     if (acpi_disabled("cpu") || acpi_get_type(dev) != ACPI_TYPE_PROCESSOR)
208 	return (ENXIO);
209 
210     handle = acpi_get_handle(dev);
211     if (cpu_softc == NULL)
212 	cpu_softc = malloc(sizeof(struct acpi_cpu_softc *) *
213 	    (mp_maxid + 1), M_TEMP /* XXX */, M_WAITOK | M_ZERO);
214 
215     /* Get our Processor object. */
216     buf.Pointer = NULL;
217     buf.Length = ACPI_ALLOCATE_BUFFER;
218     status = AcpiEvaluateObject(handle, NULL, NULL, &buf);
219     if (ACPI_FAILURE(status)) {
220 	device_printf(dev, "probe failed to get Processor obj - %s\n",
221 		      AcpiFormatException(status));
222 	return (ENXIO);
223     }
224     obj = (ACPI_OBJECT *)buf.Pointer;
225     if (obj->Type != ACPI_TYPE_PROCESSOR) {
226 	device_printf(dev, "Processor object has bad type %d\n", obj->Type);
227 	AcpiOsFree(obj);
228 	return (ENXIO);
229     }
230 
231     /*
232      * Find the processor associated with our unit.  We could use the
233      * ProcId as a key, however, some boxes do not have the same values
234      * in their Processor object as the ProcId values in the MADT.
235      */
236     acpi_id = obj->Processor.ProcId;
237     AcpiOsFree(obj);
238     if (acpi_pcpu_get_id(device_get_unit(dev), &acpi_id, &cpu_id) != 0)
239 	return (ENXIO);
240 
241     /*
242      * Check if we already probed this processor.  We scan the bus twice
243      * so it's possible we've already seen this one.
244      */
245     if (cpu_softc[cpu_id] != NULL)
246 	return (ENXIO);
247 
248     /* Mark this processor as in-use and save our derived id for attach. */
249     cpu_softc[cpu_id] = (void *)1;
250     acpi_set_magic(dev, cpu_id);
251     device_set_desc(dev, "ACPI CPU");
252 
253     return (0);
254 }
255 
256 static int
257 acpi_cpu_attach(device_t dev)
258 {
259     ACPI_BUFFER		   buf;
260     ACPI_OBJECT		   arg, *obj;
261     ACPI_OBJECT_LIST	   arglist;
262     struct pcpu		   *pcpu_data;
263     struct acpi_cpu_softc *sc;
264     struct acpi_softc	  *acpi_sc;
265     ACPI_STATUS		   status;
266     u_int		   features;
267     int			   cpu_id, drv_count, i;
268     driver_t 		  **drivers;
269     uint32_t		   cap_set[3];
270 
271     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
272 
273     sc = device_get_softc(dev);
274     sc->cpu_dev = dev;
275     sc->cpu_handle = acpi_get_handle(dev);
276     cpu_id = acpi_get_magic(dev);
277     cpu_softc[cpu_id] = sc;
278     pcpu_data = pcpu_find(cpu_id);
279     pcpu_data->pc_device = dev;
280     sc->cpu_pcpu = pcpu_data;
281     cpu_smi_cmd = AcpiGbl_FADT.SmiCommand;
282     cpu_cst_cnt = AcpiGbl_FADT.CstControl;
283 
284     buf.Pointer = NULL;
285     buf.Length = ACPI_ALLOCATE_BUFFER;
286     status = AcpiEvaluateObject(sc->cpu_handle, NULL, NULL, &buf);
287     if (ACPI_FAILURE(status)) {
288 	device_printf(dev, "attach failed to get Processor obj - %s\n",
289 		      AcpiFormatException(status));
290 	return (ENXIO);
291     }
292     obj = (ACPI_OBJECT *)buf.Pointer;
293     sc->cpu_p_blk = obj->Processor.PblkAddress;
294     sc->cpu_p_blk_len = obj->Processor.PblkLength;
295     sc->cpu_acpi_id = obj->Processor.ProcId;
296     AcpiOsFree(obj);
297     ACPI_DEBUG_PRINT((ACPI_DB_INFO, "acpi_cpu%d: P_BLK at %#x/%d\n",
298 		     device_get_unit(dev), sc->cpu_p_blk, sc->cpu_p_blk_len));
299 
300     /*
301      * If this is the first cpu we attach, create and initialize the generic
302      * resources that will be used by all acpi cpu devices.
303      */
304     if (device_get_unit(dev) == 0) {
305 	/* Assume we won't be using generic Cx mode by default */
306 	cpu_cx_generic = FALSE;
307 
308 	/* Install hw.acpi.cpu sysctl tree */
309 	acpi_sc = acpi_device_get_parent_softc(dev);
310 	sysctl_ctx_init(&cpu_sysctl_ctx);
311 	cpu_sysctl_tree = SYSCTL_ADD_NODE(&cpu_sysctl_ctx,
312 	    SYSCTL_CHILDREN(acpi_sc->acpi_sysctl_tree), OID_AUTO, "cpu",
313 	    CTLFLAG_RD, 0, "node for CPU children");
314 
315 	/* Queue post cpu-probing task handler */
316 	AcpiOsExecute(OSL_NOTIFY_HANDLER, acpi_cpu_startup, NULL);
317     }
318 
319     /*
320      * Before calling any CPU methods, collect child driver feature hints
321      * and notify ACPI of them.  We support unified SMP power control
322      * so advertise this ourselves.  Note this is not the same as independent
323      * SMP control where each CPU can have different settings.
324      */
325     sc->cpu_features = ACPI_CAP_SMP_SAME | ACPI_CAP_SMP_SAME_C3;
326     if (devclass_get_drivers(acpi_cpu_devclass, &drivers, &drv_count) == 0) {
327 	for (i = 0; i < drv_count; i++) {
328 	    if (ACPI_GET_FEATURES(drivers[i], &features) == 0)
329 		sc->cpu_features |= features;
330 	}
331 	free(drivers, M_TEMP);
332     }
333 
334     /*
335      * CPU capabilities are specified as a buffer of 32-bit integers:
336      * revision, count, and one or more capabilities.  The revision of
337      * "1" is not specified anywhere but seems to match Linux.  We should
338      * also support _OSC here.
339      */
340     if (sc->cpu_features) {
341 	arglist.Pointer = &arg;
342 	arglist.Count = 1;
343 	arg.Type = ACPI_TYPE_BUFFER;
344 	arg.Buffer.Length = sizeof(cap_set);
345 	arg.Buffer.Pointer = (uint8_t *)cap_set;
346 	cap_set[0] = 1; /* revision */
347 	cap_set[1] = 1; /* number of capabilities integers */
348 	cap_set[2] = sc->cpu_features;
349 	AcpiEvaluateObject(sc->cpu_handle, "_PDC", &arglist, NULL);
350     }
351 
352     /* Probe for Cx state support. */
353     acpi_cpu_cx_probe(sc);
354 
355     /* Finally,  call identify and probe/attach for child devices. */
356     bus_generic_probe(dev);
357     bus_generic_attach(dev);
358 
359     return (0);
360 }
361 
362 /*
363  * Disable any entry to the idle function during suspend and re-enable it
364  * during resume.
365  */
366 static int
367 acpi_cpu_suspend(device_t dev)
368 {
369     int error;
370 
371     error = bus_generic_suspend(dev);
372     if (error)
373 	return (error);
374     cpu_disable_idle = TRUE;
375     return (0);
376 }
377 
378 static int
379 acpi_cpu_resume(device_t dev)
380 {
381 
382     cpu_disable_idle = FALSE;
383     return (bus_generic_resume(dev));
384 }
385 
386 /*
387  * Find the nth present CPU and return its pc_cpuid as well as set the
388  * pc_acpi_id from the most reliable source.
389  */
390 static int
391 acpi_pcpu_get_id(uint32_t idx, uint32_t *acpi_id, uint32_t *cpu_id)
392 {
393     struct pcpu	*pcpu_data;
394     uint32_t	 i;
395 
396     KASSERT(acpi_id != NULL, ("Null acpi_id"));
397     KASSERT(cpu_id != NULL, ("Null cpu_id"));
398     for (i = 0; i <= mp_maxid; i++) {
399 	if (CPU_ABSENT(i))
400 	    continue;
401 	pcpu_data = pcpu_find(i);
402 	KASSERT(pcpu_data != NULL, ("no pcpu data for %d", i));
403 	if (idx-- == 0) {
404 	    /*
405 	     * If pc_acpi_id was not initialized (e.g., a non-APIC UP box)
406 	     * override it with the value from the ASL.  Otherwise, if the
407 	     * two don't match, prefer the MADT-derived value.  Finally,
408 	     * return the pc_cpuid to reference this processor.
409 	     */
410 	    if (pcpu_data->pc_acpi_id == 0xffffffff)
411 		pcpu_data->pc_acpi_id = *acpi_id;
412 	    else if (pcpu_data->pc_acpi_id != *acpi_id)
413 		*acpi_id = pcpu_data->pc_acpi_id;
414 	    *cpu_id = pcpu_data->pc_cpuid;
415 	    return (0);
416 	}
417     }
418 
419     return (ESRCH);
420 }
421 
422 static struct resource_list *
423 acpi_cpu_get_rlist(device_t dev, device_t child)
424 {
425     struct acpi_cpu_device *ad;
426 
427     ad = device_get_ivars(child);
428     if (ad == NULL)
429 	return (NULL);
430     return (&ad->ad_rl);
431 }
432 
433 static device_t
434 acpi_cpu_add_child(device_t dev, int order, const char *name, int unit)
435 {
436     struct acpi_cpu_device *ad;
437     device_t child;
438 
439     if ((ad = malloc(sizeof(*ad), M_TEMP, M_NOWAIT | M_ZERO)) == NULL)
440 	return (NULL);
441 
442     resource_list_init(&ad->ad_rl);
443 
444     child = device_add_child_ordered(dev, order, name, unit);
445     if (child != NULL)
446 	device_set_ivars(child, ad);
447     else
448 	free(ad, M_TEMP);
449     return (child);
450 }
451 
452 static int
453 acpi_cpu_read_ivar(device_t dev, device_t child, int index, uintptr_t *result)
454 {
455     struct acpi_cpu_softc *sc;
456 
457     sc = device_get_softc(dev);
458     switch (index) {
459     case ACPI_IVAR_HANDLE:
460 	*result = (uintptr_t)sc->cpu_handle;
461 	break;
462     case CPU_IVAR_PCPU:
463 	*result = (uintptr_t)sc->cpu_pcpu;
464 	break;
465     default:
466 	return (ENOENT);
467     }
468     return (0);
469 }
470 
471 static int
472 acpi_cpu_shutdown(device_t dev)
473 {
474     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
475 
476     /* Allow children to shutdown first. */
477     bus_generic_shutdown(dev);
478 
479     /* Disable any entry to the idle function. */
480     cpu_disable_idle = TRUE;
481 
482     /* Signal and wait for all processors to exit acpi_cpu_idle(). */
483     smp_rendezvous(NULL, NULL, NULL, NULL);
484 
485     return_VALUE (0);
486 }
487 
488 static void
489 acpi_cpu_cx_probe(struct acpi_cpu_softc *sc)
490 {
491     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
492 
493     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
494     sc->cpu_prev_sleep = 1000000;
495     sc->cpu_cx_lowest = 0;
496 
497     /*
498      * Check for the ACPI 2.0 _CST sleep states object. If we can't find
499      * any, we'll revert to generic FADT/P_BLK Cx control method which will
500      * be handled by acpi_cpu_startup. We need to defer to after having
501      * probed all the cpus in the system before probing for generic Cx
502      * states as we may already have found cpus with valid _CST packages
503      */
504     if (!cpu_cx_generic && acpi_cpu_cx_cst(sc) != 0) {
505 	/*
506 	 * We were unable to find a _CST package for this cpu or there
507 	 * was an error parsing it. Switch back to generic mode.
508 	 */
509 	cpu_cx_generic = TRUE;
510 	if (bootverbose)
511 	    device_printf(sc->cpu_dev, "switching to generic Cx mode\n");
512     }
513 
514     /*
515      * TODO: _CSD Package should be checked here.
516      */
517 }
518 
519 static void
520 acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc)
521 {
522     ACPI_GENERIC_ADDRESS	 gas;
523     struct acpi_cx		*cx_ptr;
524 
525     sc->cpu_cx_count = 0;
526     cx_ptr = sc->cpu_cx_states;
527 
528     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
529     sc->cpu_prev_sleep = 1000000;
530 
531     /* C1 has been required since just after ACPI 1.0 */
532     cx_ptr->type = ACPI_STATE_C1;
533     cx_ptr->trans_lat = 0;
534     cx_ptr++;
535     sc->cpu_cx_count++;
536 
537     /*
538      * The spec says P_BLK must be 6 bytes long.  However, some systems
539      * use it to indicate a fractional set of features present so we
540      * take 5 as C2.  Some may also have a value of 7 to indicate
541      * another C3 but most use _CST for this (as required) and having
542      * "only" C1-C3 is not a hardship.
543      */
544     if (sc->cpu_p_blk_len < 5)
545 	return;
546 
547     /* Validate and allocate resources for C2 (P_LVL2). */
548     gas.SpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
549     gas.BitWidth = 8;
550     if (AcpiGbl_FADT.C2Latency <= 100) {
551 	gas.Address = sc->cpu_p_blk + 4;
552 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &sc->cpu_rid,
553 	    &gas, &cx_ptr->p_lvlx, RF_SHAREABLE);
554 	if (cx_ptr->p_lvlx != NULL) {
555 	    sc->cpu_rid++;
556 	    cx_ptr->type = ACPI_STATE_C2;
557 	    cx_ptr->trans_lat = AcpiGbl_FADT.C2Latency;
558 	    cx_ptr++;
559 	    sc->cpu_cx_count++;
560 	}
561     }
562     if (sc->cpu_p_blk_len < 6)
563 	return;
564 
565     /* Validate and allocate resources for C3 (P_LVL3). */
566     if (AcpiGbl_FADT.C3Latency <= 1000) {
567 	gas.Address = sc->cpu_p_blk + 5;
568 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &sc->cpu_rid, &gas,
569 	    &cx_ptr->p_lvlx, RF_SHAREABLE);
570 	if (cx_ptr->p_lvlx != NULL) {
571 	    sc->cpu_rid++;
572 	    cx_ptr->type = ACPI_STATE_C3;
573 	    cx_ptr->trans_lat = AcpiGbl_FADT.C3Latency;
574 	    cx_ptr++;
575 	    sc->cpu_cx_count++;
576 	}
577     }
578 
579     /* Update the largest cx_count seen so far */
580     if (sc->cpu_cx_count > cpu_cx_count)
581 	cpu_cx_count = sc->cpu_cx_count;
582 }
583 
584 /*
585  * Parse a _CST package and set up its Cx states.  Since the _CST object
586  * can change dynamically, our notify handler may call this function
587  * to clean up and probe the new _CST package.
588  */
589 static int
590 acpi_cpu_cx_cst(struct acpi_cpu_softc *sc)
591 {
592     struct	 acpi_cx *cx_ptr;
593     ACPI_STATUS	 status;
594     ACPI_BUFFER	 buf;
595     ACPI_OBJECT	*top;
596     ACPI_OBJECT	*pkg;
597     uint32_t	 count;
598     int		 i;
599 
600     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
601 
602     buf.Pointer = NULL;
603     buf.Length = ACPI_ALLOCATE_BUFFER;
604     status = AcpiEvaluateObject(sc->cpu_handle, "_CST", NULL, &buf);
605     if (ACPI_FAILURE(status))
606 	return (ENXIO);
607 
608     /* _CST is a package with a count and at least one Cx package. */
609     top = (ACPI_OBJECT *)buf.Pointer;
610     if (!ACPI_PKG_VALID(top, 2) || acpi_PkgInt32(top, 0, &count) != 0) {
611 	device_printf(sc->cpu_dev, "invalid _CST package\n");
612 	AcpiOsFree(buf.Pointer);
613 	return (ENXIO);
614     }
615     if (count != top->Package.Count - 1) {
616 	device_printf(sc->cpu_dev, "invalid _CST state count (%d != %d)\n",
617 	       count, top->Package.Count - 1);
618 	count = top->Package.Count - 1;
619     }
620     if (count > MAX_CX_STATES) {
621 	device_printf(sc->cpu_dev, "_CST has too many states (%d)\n", count);
622 	count = MAX_CX_STATES;
623     }
624 
625     /* Set up all valid states. */
626     sc->cpu_cx_count = 0;
627     cx_ptr = sc->cpu_cx_states;
628     for (i = 0; i < count; i++) {
629 	pkg = &top->Package.Elements[i + 1];
630 	if (!ACPI_PKG_VALID(pkg, 4) ||
631 	    acpi_PkgInt32(pkg, 1, &cx_ptr->type) != 0 ||
632 	    acpi_PkgInt32(pkg, 2, &cx_ptr->trans_lat) != 0 ||
633 	    acpi_PkgInt32(pkg, 3, &cx_ptr->power) != 0) {
634 
635 	    device_printf(sc->cpu_dev, "skipping invalid Cx state package\n");
636 	    continue;
637 	}
638 
639 	/* Validate the state to see if we should use it. */
640 	switch (cx_ptr->type) {
641 	case ACPI_STATE_C1:
642 	    sc->cpu_non_c3 = i;
643 	    cx_ptr++;
644 	    sc->cpu_cx_count++;
645 	    continue;
646 	case ACPI_STATE_C2:
647 	    if (cx_ptr->trans_lat > 100) {
648 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
649 				 "acpi_cpu%d: C2[%d] not available.\n",
650 				 device_get_unit(sc->cpu_dev), i));
651 		continue;
652 	    }
653 	    sc->cpu_non_c3 = i;
654 	    break;
655 	case ACPI_STATE_C3:
656 	default:
657 	    if (cx_ptr->trans_lat > 1000 ||
658 		(cpu_quirks & CPU_QUIRK_NO_C3) != 0) {
659 
660 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
661 				 "acpi_cpu%d: C3[%d] not available.\n",
662 				 device_get_unit(sc->cpu_dev), i));
663 		continue;
664 	    }
665 	    break;
666 	}
667 
668 #ifdef notyet
669 	/* Free up any previous register. */
670 	if (cx_ptr->p_lvlx != NULL) {
671 	    bus_release_resource(sc->cpu_dev, 0, 0, cx_ptr->p_lvlx);
672 	    cx_ptr->p_lvlx = NULL;
673 	}
674 #endif
675 
676 	/* Allocate the control register for C2 or C3. */
677 	acpi_PkgGas(sc->cpu_dev, pkg, 0, &cx_ptr->res_type, &sc->cpu_rid,
678 	    &cx_ptr->p_lvlx, RF_SHAREABLE);
679 	if (cx_ptr->p_lvlx) {
680 	    sc->cpu_rid++;
681 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
682 			     "acpi_cpu%d: Got C%d - %d latency\n",
683 			     device_get_unit(sc->cpu_dev), cx_ptr->type,
684 			     cx_ptr->trans_lat));
685 	    cx_ptr++;
686 	    sc->cpu_cx_count++;
687 	}
688     }
689     AcpiOsFree(buf.Pointer);
690 
691     return (0);
692 }
693 
694 /*
695  * Call this *after* all CPUs have been attached.
696  */
697 static void
698 acpi_cpu_startup(void *arg)
699 {
700     struct acpi_cpu_softc *sc;
701     int i;
702 
703     /* Get set of CPU devices */
704     devclass_get_devices(acpi_cpu_devclass, &cpu_devices, &cpu_ndevices);
705 
706     /*
707      * Setup any quirks that might necessary now that we have probed
708      * all the CPUs
709      */
710     acpi_cpu_quirks();
711 
712     cpu_cx_count = 0;
713     if (cpu_cx_generic) {
714 	/*
715 	 * We are using generic Cx mode, probe for available Cx states
716 	 * for all processors.
717 	 */
718 	for (i = 0; i < cpu_ndevices; i++) {
719 	    sc = device_get_softc(cpu_devices[i]);
720 	    acpi_cpu_generic_cx_probe(sc);
721 	}
722 
723 	/*
724 	 * Find the highest Cx state common to all CPUs
725 	 * in the system, taking quirks into account.
726 	 */
727 	for (i = 0; i < cpu_ndevices; i++) {
728 	    sc = device_get_softc(cpu_devices[i]);
729 	    if (sc->cpu_cx_count < cpu_cx_count)
730 		cpu_cx_count = sc->cpu_cx_count;
731 	}
732     } else {
733 	/*
734 	 * We are using _CST mode, remove C3 state if necessary.
735 	 * Update the largest Cx state supported in the global cpu_cx_count.
736 	 * It will be used in the global Cx sysctl handler.
737 	 * As we now know for sure that we will be using _CST mode
738 	 * install our notify handler.
739 	 */
740 	for (i = 0; i < cpu_ndevices; i++) {
741 	    sc = device_get_softc(cpu_devices[i]);
742 	    if (cpu_quirks && CPU_QUIRK_NO_C3) {
743 		sc->cpu_cx_count = sc->cpu_non_c3 + 1;
744 	    }
745 	    if (sc->cpu_cx_count > cpu_cx_count)
746 		cpu_cx_count = sc->cpu_cx_count;
747 	    AcpiInstallNotifyHandler(sc->cpu_handle, ACPI_DEVICE_NOTIFY,
748 		acpi_cpu_notify, sc);
749 	}
750     }
751 
752     /* Perform Cx final initialization. */
753     for (i = 0; i < cpu_ndevices; i++) {
754 	sc = device_get_softc(cpu_devices[i]);
755 	acpi_cpu_startup_cx(sc);
756     }
757 
758     /* Add a sysctl handler to handle global Cx lowest setting */
759     SYSCTL_ADD_PROC(&cpu_sysctl_ctx, SYSCTL_CHILDREN(cpu_sysctl_tree),
760 	OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
761 	NULL, 0, acpi_cpu_global_cx_lowest_sysctl, "A",
762 	"Global lowest Cx sleep state to use");
763 
764     /* Take over idling from cpu_idle_default(). */
765     cpu_cx_lowest = 0;
766     cpu_disable_idle = FALSE;
767     cpu_idle_hook = acpi_cpu_idle;
768 }
769 
770 static void
771 acpi_cpu_startup_cx(struct acpi_cpu_softc *sc)
772 {
773     struct sbuf sb;
774     int i;
775 
776     /*
777      * Set up the list of Cx states
778      */
779     sc->cpu_non_c3 = 0;
780     sbuf_new(&sb, sc->cpu_cx_supported, sizeof(sc->cpu_cx_supported),
781 	SBUF_FIXEDLEN);
782     for (i = 0; i < sc->cpu_cx_count; i++) {
783 	sbuf_printf(&sb, "C%d/%d ", i + 1, sc->cpu_cx_states[i].trans_lat);
784 	if (sc->cpu_cx_states[i].type < ACPI_STATE_C3)
785 	    sc->cpu_non_c3 = i;
786     }
787     sbuf_trim(&sb);
788     sbuf_finish(&sb);
789 
790     SYSCTL_ADD_STRING(&sc->cpu_sysctl_ctx,
791 		      SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
792 		      OID_AUTO, "cx_supported", CTLFLAG_RD,
793 		      sc->cpu_cx_supported, 0,
794 		      "Cx/microsecond values for supported Cx states");
795     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
796 		    SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
797 		    OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
798 		    (void *)sc, 0, acpi_cpu_cx_lowest_sysctl, "A",
799 		    "lowest Cx sleep state to use");
800     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
801 		    SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
802 		    OID_AUTO, "cx_usage", CTLTYPE_STRING | CTLFLAG_RD,
803 		    (void *)sc, 0, acpi_cpu_usage_sysctl, "A",
804 		    "percent usage for each Cx state");
805 
806 #ifdef notyet
807     /* Signal platform that we can handle _CST notification. */
808     if (!cpu_cx_generic && cpu_cst_cnt != 0) {
809 	ACPI_LOCK(acpi);
810 	AcpiOsWritePort(cpu_smi_cmd, cpu_cst_cnt, 8);
811 	ACPI_UNLOCK(acpi);
812     }
813 #endif
814 }
815 
816 /*
817  * Idle the CPU in the lowest state possible.  This function is called with
818  * interrupts disabled.  Note that once it re-enables interrupts, a task
819  * switch can occur so do not access shared data (i.e. the softc) after
820  * interrupts are re-enabled.
821  */
822 static void
823 acpi_cpu_idle()
824 {
825     struct	acpi_cpu_softc *sc;
826     struct	acpi_cx *cx_next;
827     uint32_t	start_time, end_time;
828     int		bm_active, cx_next_idx, i;
829 
830     /* If disabled, return immediately. */
831     if (cpu_disable_idle) {
832 	ACPI_ENABLE_IRQS();
833 	return;
834     }
835 
836     /*
837      * Look up our CPU id to get our softc.  If it's NULL, we'll use C1
838      * since there is no ACPI processor object for this CPU.  This occurs
839      * for logical CPUs in the HTT case.
840      */
841     sc = cpu_softc[PCPU_GET(cpuid)];
842     if (sc == NULL) {
843 	acpi_cpu_c1();
844 	return;
845     }
846 
847     /*
848      * If we slept 100 us or more, use the lowest Cx state.  Otherwise,
849      * find the lowest state that has a latency less than or equal to
850      * the length of our last sleep.
851      */
852     cx_next_idx = sc->cpu_cx_lowest;
853     if (sc->cpu_prev_sleep < 100) {
854 	/*
855 	 * If we sleep too short all the time, this system may not implement
856 	 * C2/3 correctly (i.e. reads return immediately).  In this case,
857 	 * back off and use the next higher level.
858 	 * It seems that when you have a dual core cpu (like the Intel Core Duo)
859 	 * that both cores will get out of C3 state as soon as one of them
860 	 * requires it. This breaks the sleep detection logic as the sleep
861 	 * counter is local to each cpu. Disable the sleep logic for now as a
862 	 * workaround if there's more than one CPU. The right fix would probably
863 	 * be to add quirks for system that don't really support C3 state.
864 	 */
865 	if (mp_ncpus < 2 && sc->cpu_prev_sleep <= 1) {
866 	    sc->cpu_short_slp++;
867 	    if (sc->cpu_short_slp == 1000 && sc->cpu_cx_lowest != 0) {
868 		if (sc->cpu_non_c3 == sc->cpu_cx_lowest && sc->cpu_non_c3 != 0)
869 		    sc->cpu_non_c3--;
870 		sc->cpu_cx_lowest--;
871 		sc->cpu_short_slp = 0;
872 		device_printf(sc->cpu_dev,
873 		    "too many short sleeps, backing off to C%d\n",
874 		    sc->cpu_cx_lowest + 1);
875 	    }
876 	} else
877 	    sc->cpu_short_slp = 0;
878 
879 	for (i = sc->cpu_cx_lowest; i >= 0; i--)
880 	    if (sc->cpu_cx_states[i].trans_lat <= sc->cpu_prev_sleep) {
881 		cx_next_idx = i;
882 		break;
883 	    }
884     }
885 
886     /*
887      * Check for bus master activity.  If there was activity, clear
888      * the bit and use the lowest non-C3 state.  Note that the USB
889      * driver polling for new devices keeps this bit set all the
890      * time if USB is loaded.
891      */
892     if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
893 	AcpiGetRegister(ACPI_BITREG_BUS_MASTER_STATUS, &bm_active);
894 	if (bm_active != 0) {
895 	    AcpiSetRegister(ACPI_BITREG_BUS_MASTER_STATUS, 1);
896 	    cx_next_idx = min(cx_next_idx, sc->cpu_non_c3);
897 	}
898     }
899 
900     /* Select the next state and update statistics. */
901     cx_next = &sc->cpu_cx_states[cx_next_idx];
902     sc->cpu_cx_stats[cx_next_idx]++;
903     KASSERT(cx_next->type != ACPI_STATE_C0, ("acpi_cpu_idle: C0 sleep"));
904 
905     /*
906      * Execute HLT (or equivalent) and wait for an interrupt.  We can't
907      * calculate the time spent in C1 since the place we wake up is an
908      * ISR.  Assume we slept one quantum and return.
909      */
910     if (cx_next->type == ACPI_STATE_C1) {
911 	sc->cpu_prev_sleep = 1000000 / hz;
912 	acpi_cpu_c1();
913 	return;
914     }
915 
916     /*
917      * For C3, disable bus master arbitration and enable bus master wake
918      * if BM control is available, otherwise flush the CPU cache.
919      */
920     if (cx_next->type == ACPI_STATE_C3) {
921 	if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
922 	    AcpiSetRegister(ACPI_BITREG_ARB_DISABLE, 1);
923 	    AcpiSetRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
924 	} else
925 	    ACPI_FLUSH_CPU_CACHE();
926     }
927 
928     /*
929      * Read from P_LVLx to enter C2(+), checking time spent asleep.
930      * Use the ACPI timer for measuring sleep time.  Since we need to
931      * get the time very close to the CPU start/stop clock logic, this
932      * is the only reliable time source.
933      */
934     AcpiHwLowLevelRead(32, &start_time, &AcpiGbl_FADT.XPmTimerBlock);
935     CPU_GET_REG(cx_next->p_lvlx, 1);
936 
937     /*
938      * Read the end time twice.  Since it may take an arbitrary time
939      * to enter the idle state, the first read may be executed before
940      * the processor has stopped.  Doing it again provides enough
941      * margin that we are certain to have a correct value.
942      */
943     AcpiHwLowLevelRead(32, &end_time, &AcpiGbl_FADT.XPmTimerBlock);
944     AcpiHwLowLevelRead(32, &end_time, &AcpiGbl_FADT.XPmTimerBlock);
945 
946     /* Enable bus master arbitration and disable bus master wakeup. */
947     if (cx_next->type == ACPI_STATE_C3 &&
948 	(cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
949 	AcpiSetRegister(ACPI_BITREG_ARB_DISABLE, 0);
950 	AcpiSetRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
951     }
952     ACPI_ENABLE_IRQS();
953 
954     /* Find the actual time asleep in microseconds, minus overhead. */
955     end_time = acpi_TimerDelta(end_time, start_time);
956     sc->cpu_prev_sleep = PM_USEC(end_time) - cx_next->trans_lat;
957 }
958 
959 /*
960  * Re-evaluate the _CST object when we are notified that it changed.
961  *
962  * XXX Re-evaluation disabled until locking is done.
963  */
964 static void
965 acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context)
966 {
967     struct acpi_cpu_softc *sc = (struct acpi_cpu_softc *)context;
968 
969     if (notify != ACPI_NOTIFY_CX_STATES)
970 	return;
971 
972     device_printf(sc->cpu_dev, "Cx states changed\n");
973     /* acpi_cpu_cx_cst(sc); */
974 }
975 
976 static int
977 acpi_cpu_quirks(void)
978 {
979     device_t acpi_dev;
980 
981     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
982 
983     /*
984      * Bus mastering arbitration control is needed to keep caches coherent
985      * while sleeping in C3.  If it's not present but a working flush cache
986      * instruction is present, flush the caches before entering C3 instead.
987      * Otherwise, just disable C3 completely.
988      */
989     if (AcpiGbl_FADT.Pm2ControlBlock == 0 ||
990 	AcpiGbl_FADT.Pm2ControlLength == 0) {
991 	if ((AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD) &&
992 	    (AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD_FLUSH) == 0) {
993 	    cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
994 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
995 		"acpi_cpu: no BM control, using flush cache method\n"));
996 	} else {
997 	    cpu_quirks |= CPU_QUIRK_NO_C3;
998 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
999 		"acpi_cpu: no BM control, C3 not available\n"));
1000 	}
1001     }
1002 
1003     /*
1004      * If we are using generic Cx mode, C3 on multiple CPUs requires using
1005      * the expensive flush cache instruction.
1006      */
1007     if (cpu_cx_generic && mp_ncpus > 1) {
1008 	cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1009 	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1010 	    "acpi_cpu: SMP, using flush cache mode for C3\n"));
1011     }
1012 
1013     /* Look for various quirks of the PIIX4 part. */
1014     acpi_dev = pci_find_device(PCI_VENDOR_INTEL, PCI_DEVICE_82371AB_3);
1015     if (acpi_dev != NULL) {
1016 	switch (pci_get_revid(acpi_dev)) {
1017 	/*
1018 	 * Disable C3 support for all PIIX4 chipsets.  Some of these parts
1019 	 * do not report the BMIDE status to the BM status register and
1020 	 * others have a livelock bug if Type-F DMA is enabled.  Linux
1021 	 * works around the BMIDE bug by reading the BM status directly
1022 	 * but we take the simpler approach of disabling C3 for these
1023 	 * parts.
1024 	 *
1025 	 * See erratum #18 ("C3 Power State/BMIDE and Type-F DMA
1026 	 * Livelock") from the January 2002 PIIX4 specification update.
1027 	 * Applies to all PIIX4 models.
1028 	 */
1029 	case PCI_REVISION_4E:
1030 	case PCI_REVISION_4M:
1031 	    cpu_quirks |= CPU_QUIRK_NO_C3;
1032 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1033 		"acpi_cpu: working around PIIX4 bug, disabling C3\n"));
1034 	    break;
1035 	default:
1036 	    break;
1037 	}
1038     }
1039 
1040     return (0);
1041 }
1042 
1043 static int
1044 acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS)
1045 {
1046     struct acpi_cpu_softc *sc;
1047     struct sbuf	 sb;
1048     char	 buf[128];
1049     int		 i;
1050     uintmax_t	 fract, sum, whole;
1051 
1052     sc = (struct acpi_cpu_softc *) arg1;
1053     sum = 0;
1054     for (i = 0; i < sc->cpu_cx_count; i++)
1055 	sum += sc->cpu_cx_stats[i];
1056     sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1057     for (i = 0; i < sc->cpu_cx_count; i++) {
1058 	if (sum > 0) {
1059 	    whole = (uintmax_t)sc->cpu_cx_stats[i] * 100;
1060 	    fract = (whole % sum) * 100;
1061 	    sbuf_printf(&sb, "%u.%02u%% ", (u_int)(whole / sum),
1062 		(u_int)(fract / sum));
1063 	} else
1064 	    sbuf_printf(&sb, "0%% ");
1065     }
1066     sbuf_trim(&sb);
1067     sbuf_finish(&sb);
1068     sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1069     sbuf_delete(&sb);
1070 
1071     return (0);
1072 }
1073 
1074 static int
1075 acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc, int val)
1076 {
1077     int i;
1078 
1079     ACPI_SERIAL_ASSERT(cpu);
1080     sc->cpu_cx_lowest = val;
1081 
1082     /* If not disabling, cache the new lowest non-C3 state. */
1083     sc->cpu_non_c3 = 0;
1084     for (i = sc->cpu_cx_lowest; i >= 0; i--) {
1085 	if (sc->cpu_cx_states[i].type < ACPI_STATE_C3) {
1086 	    sc->cpu_non_c3 = i;
1087 	    break;
1088 	}
1089     }
1090 
1091     /* Reset the statistics counters. */
1092     bzero(sc->cpu_cx_stats, sizeof(sc->cpu_cx_stats));
1093     return (0);
1094 }
1095 
1096 static int
1097 acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1098 {
1099     struct	 acpi_cpu_softc *sc;
1100     char	 state[8];
1101     int		 val, error;
1102 
1103     sc = (struct acpi_cpu_softc *) arg1;
1104     snprintf(state, sizeof(state), "C%d", sc->cpu_cx_lowest + 1);
1105     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1106     if (error != 0 || req->newptr == NULL)
1107 	return (error);
1108     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1109 	return (EINVAL);
1110     val = (int) strtol(state + 1, NULL, 10) - 1;
1111     if (val < 0 || val > sc->cpu_cx_count - 1)
1112 	return (EINVAL);
1113 
1114     ACPI_SERIAL_BEGIN(cpu);
1115     acpi_cpu_set_cx_lowest(sc, val);
1116     ACPI_SERIAL_END(cpu);
1117 
1118     return (0);
1119 }
1120 
1121 static int
1122 acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1123 {
1124     struct	acpi_cpu_softc *sc;
1125     char	state[8];
1126     int		val, error, i;
1127 
1128     snprintf(state, sizeof(state), "C%d", cpu_cx_lowest + 1);
1129     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1130     if (error != 0 || req->newptr == NULL)
1131 	return (error);
1132     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1133 	return (EINVAL);
1134     val = (int) strtol(state + 1, NULL, 10) - 1;
1135     if (val < 0 || val > cpu_cx_count - 1)
1136 	return (EINVAL);
1137     cpu_cx_lowest = val;
1138 
1139     /* Update the new lowest useable Cx state for all CPUs. */
1140     ACPI_SERIAL_BEGIN(cpu);
1141     for (i = 0; i < cpu_ndevices; i++) {
1142 	sc = device_get_softc(cpu_devices[i]);
1143 	acpi_cpu_set_cx_lowest(sc, val);
1144     }
1145     ACPI_SERIAL_END(cpu);
1146 
1147     return (0);
1148 }
1149