xref: /freebsd/sys/dev/acpica/acpi_cpu.c (revision 1e413cf93298b5b97441a21d9a50fdcd0ee9945e)
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[4], *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     /* UUID needed by _OSC evaluation */
272     static uint8_t cpu_oscuuid[16] = { 0x16, 0xA6, 0x77, 0x40, 0x0C, 0x29,
273 				       0xBE, 0x47, 0x9E, 0xBD, 0xD8, 0x70,
274 				       0x58, 0x71, 0x39, 0x53 };
275 
276     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
277 
278     sc = device_get_softc(dev);
279     sc->cpu_dev = dev;
280     sc->cpu_handle = acpi_get_handle(dev);
281     cpu_id = acpi_get_magic(dev);
282     cpu_softc[cpu_id] = sc;
283     pcpu_data = pcpu_find(cpu_id);
284     pcpu_data->pc_device = dev;
285     sc->cpu_pcpu = pcpu_data;
286     cpu_smi_cmd = AcpiGbl_FADT.SmiCommand;
287     cpu_cst_cnt = AcpiGbl_FADT.CstControl;
288 
289     buf.Pointer = NULL;
290     buf.Length = ACPI_ALLOCATE_BUFFER;
291     status = AcpiEvaluateObject(sc->cpu_handle, NULL, NULL, &buf);
292     if (ACPI_FAILURE(status)) {
293 	device_printf(dev, "attach failed to get Processor obj - %s\n",
294 		      AcpiFormatException(status));
295 	return (ENXIO);
296     }
297     obj = (ACPI_OBJECT *)buf.Pointer;
298     sc->cpu_p_blk = obj->Processor.PblkAddress;
299     sc->cpu_p_blk_len = obj->Processor.PblkLength;
300     sc->cpu_acpi_id = obj->Processor.ProcId;
301     AcpiOsFree(obj);
302     ACPI_DEBUG_PRINT((ACPI_DB_INFO, "acpi_cpu%d: P_BLK at %#x/%d\n",
303 		     device_get_unit(dev), sc->cpu_p_blk, sc->cpu_p_blk_len));
304 
305     /*
306      * If this is the first cpu we attach, create and initialize the generic
307      * resources that will be used by all acpi cpu devices.
308      */
309     if (device_get_unit(dev) == 0) {
310 	/* Assume we won't be using generic Cx mode by default */
311 	cpu_cx_generic = FALSE;
312 
313 	/* Install hw.acpi.cpu sysctl tree */
314 	acpi_sc = acpi_device_get_parent_softc(dev);
315 	sysctl_ctx_init(&cpu_sysctl_ctx);
316 	cpu_sysctl_tree = SYSCTL_ADD_NODE(&cpu_sysctl_ctx,
317 	    SYSCTL_CHILDREN(acpi_sc->acpi_sysctl_tree), OID_AUTO, "cpu",
318 	    CTLFLAG_RD, 0, "node for CPU children");
319 
320 	/* Queue post cpu-probing task handler */
321 	AcpiOsExecute(OSL_NOTIFY_HANDLER, acpi_cpu_startup, NULL);
322     }
323 
324     /*
325      * Before calling any CPU methods, collect child driver feature hints
326      * and notify ACPI of them.  We support unified SMP power control
327      * so advertise this ourselves.  Note this is not the same as independent
328      * SMP control where each CPU can have different settings.
329      */
330     sc->cpu_features = ACPI_CAP_SMP_SAME | ACPI_CAP_SMP_SAME_C3;
331     if (devclass_get_drivers(acpi_cpu_devclass, &drivers, &drv_count) == 0) {
332 	for (i = 0; i < drv_count; i++) {
333 	    if (ACPI_GET_FEATURES(drivers[i], &features) == 0)
334 		sc->cpu_features |= features;
335 	}
336 	free(drivers, M_TEMP);
337     }
338 
339     /*
340      * CPU capabilities are specified as a buffer of 32-bit integers:
341      * revision, count, and one or more capabilities.  The revision of
342      * "1" is not specified anywhere but seems to match Linux.
343      */
344     if (sc->cpu_features) {
345 	arglist.Pointer = arg;
346 	arglist.Count = 1;
347 	arg[0].Type = ACPI_TYPE_BUFFER;
348 	arg[0].Buffer.Length = sizeof(cap_set);
349 	arg[0].Buffer.Pointer = (uint8_t *)cap_set;
350 	cap_set[0] = 1; /* revision */
351 	cap_set[1] = 1; /* number of capabilities integers */
352 	cap_set[2] = sc->cpu_features;
353 	AcpiEvaluateObject(sc->cpu_handle, "_PDC", &arglist, NULL);
354 
355 	/*
356 	 * On some systems we need to evaluate _OSC so that the ASL
357 	 * loads the _PSS and/or _PDC methods at runtime.
358 	 *
359 	 * TODO: evaluate failure of _OSC.
360 	 */
361 	arglist.Pointer = arg;
362 	arglist.Count = 4;
363 	arg[0].Type = ACPI_TYPE_BUFFER;
364 	arg[0].Buffer.Length = sizeof(cpu_oscuuid);
365 	arg[0].Buffer.Pointer = cpu_oscuuid;	/* UUID */
366 	arg[1].Type = ACPI_TYPE_INTEGER;
367 	arg[1].Integer.Value = 1;		/* revision */
368 	arg[2].Type = ACPI_TYPE_INTEGER;
369 	arg[2].Integer.Value = 1;		/* count */
370 	arg[3].Type = ACPI_TYPE_BUFFER;
371 	arg[3].Buffer.Length = sizeof(cap_set);	/* Capabilities buffer */
372 	arg[3].Buffer.Pointer = (uint8_t *)cap_set;
373 	cap_set[0] = 0;
374 	AcpiEvaluateObject(sc->cpu_handle, "_OSC", &arglist, NULL);
375     }
376 
377     /* Probe for Cx state support. */
378     acpi_cpu_cx_probe(sc);
379 
380     /* Finally,  call identify and probe/attach for child devices. */
381     bus_generic_probe(dev);
382     bus_generic_attach(dev);
383 
384     return (0);
385 }
386 
387 /*
388  * Disable any entry to the idle function during suspend and re-enable it
389  * during resume.
390  */
391 static int
392 acpi_cpu_suspend(device_t dev)
393 {
394     int error;
395 
396     error = bus_generic_suspend(dev);
397     if (error)
398 	return (error);
399     cpu_disable_idle = TRUE;
400     return (0);
401 }
402 
403 static int
404 acpi_cpu_resume(device_t dev)
405 {
406 
407     cpu_disable_idle = FALSE;
408     return (bus_generic_resume(dev));
409 }
410 
411 /*
412  * Find the nth present CPU and return its pc_cpuid as well as set the
413  * pc_acpi_id from the most reliable source.
414  */
415 static int
416 acpi_pcpu_get_id(uint32_t idx, uint32_t *acpi_id, uint32_t *cpu_id)
417 {
418     struct pcpu	*pcpu_data;
419     uint32_t	 i;
420 
421     KASSERT(acpi_id != NULL, ("Null acpi_id"));
422     KASSERT(cpu_id != NULL, ("Null cpu_id"));
423     for (i = 0; i <= mp_maxid; i++) {
424 	if (CPU_ABSENT(i))
425 	    continue;
426 	pcpu_data = pcpu_find(i);
427 	KASSERT(pcpu_data != NULL, ("no pcpu data for %d", i));
428 	if (idx-- == 0) {
429 	    /*
430 	     * If pc_acpi_id was not initialized (e.g., a non-APIC UP box)
431 	     * override it with the value from the ASL.  Otherwise, if the
432 	     * two don't match, prefer the MADT-derived value.  Finally,
433 	     * return the pc_cpuid to reference this processor.
434 	     */
435 	    if (pcpu_data->pc_acpi_id == 0xffffffff)
436 		pcpu_data->pc_acpi_id = *acpi_id;
437 	    else if (pcpu_data->pc_acpi_id != *acpi_id)
438 		*acpi_id = pcpu_data->pc_acpi_id;
439 	    *cpu_id = pcpu_data->pc_cpuid;
440 	    return (0);
441 	}
442     }
443 
444     return (ESRCH);
445 }
446 
447 static struct resource_list *
448 acpi_cpu_get_rlist(device_t dev, device_t child)
449 {
450     struct acpi_cpu_device *ad;
451 
452     ad = device_get_ivars(child);
453     if (ad == NULL)
454 	return (NULL);
455     return (&ad->ad_rl);
456 }
457 
458 static device_t
459 acpi_cpu_add_child(device_t dev, int order, const char *name, int unit)
460 {
461     struct acpi_cpu_device *ad;
462     device_t child;
463 
464     if ((ad = malloc(sizeof(*ad), M_TEMP, M_NOWAIT | M_ZERO)) == NULL)
465 	return (NULL);
466 
467     resource_list_init(&ad->ad_rl);
468 
469     child = device_add_child_ordered(dev, order, name, unit);
470     if (child != NULL)
471 	device_set_ivars(child, ad);
472     else
473 	free(ad, M_TEMP);
474     return (child);
475 }
476 
477 static int
478 acpi_cpu_read_ivar(device_t dev, device_t child, int index, uintptr_t *result)
479 {
480     struct acpi_cpu_softc *sc;
481 
482     sc = device_get_softc(dev);
483     switch (index) {
484     case ACPI_IVAR_HANDLE:
485 	*result = (uintptr_t)sc->cpu_handle;
486 	break;
487     case CPU_IVAR_PCPU:
488 	*result = (uintptr_t)sc->cpu_pcpu;
489 	break;
490     default:
491 	return (ENOENT);
492     }
493     return (0);
494 }
495 
496 static int
497 acpi_cpu_shutdown(device_t dev)
498 {
499     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
500 
501     /* Allow children to shutdown first. */
502     bus_generic_shutdown(dev);
503 
504     /*
505      * Disable any entry to the idle function.  There is a small race where
506      * an idle thread have passed this check but not gone to sleep.  This
507      * is ok since device_shutdown() does not free the softc, otherwise
508      * we'd have to be sure all threads were evicted before returning.
509      */
510     cpu_disable_idle = TRUE;
511 
512     return_VALUE (0);
513 }
514 
515 static void
516 acpi_cpu_cx_probe(struct acpi_cpu_softc *sc)
517 {
518     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
519 
520     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
521     sc->cpu_prev_sleep = 1000000;
522     sc->cpu_cx_lowest = 0;
523 
524     /*
525      * Check for the ACPI 2.0 _CST sleep states object. If we can't find
526      * any, we'll revert to generic FADT/P_BLK Cx control method which will
527      * be handled by acpi_cpu_startup. We need to defer to after having
528      * probed all the cpus in the system before probing for generic Cx
529      * states as we may already have found cpus with valid _CST packages
530      */
531     if (!cpu_cx_generic && acpi_cpu_cx_cst(sc) != 0) {
532 	/*
533 	 * We were unable to find a _CST package for this cpu or there
534 	 * was an error parsing it. Switch back to generic mode.
535 	 */
536 	cpu_cx_generic = TRUE;
537 	if (bootverbose)
538 	    device_printf(sc->cpu_dev, "switching to generic Cx mode\n");
539     }
540 
541     /*
542      * TODO: _CSD Package should be checked here.
543      */
544 }
545 
546 static void
547 acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc)
548 {
549     ACPI_GENERIC_ADDRESS	 gas;
550     struct acpi_cx		*cx_ptr;
551 
552     sc->cpu_cx_count = 0;
553     cx_ptr = sc->cpu_cx_states;
554 
555     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
556     sc->cpu_prev_sleep = 1000000;
557 
558     /* C1 has been required since just after ACPI 1.0 */
559     cx_ptr->type = ACPI_STATE_C1;
560     cx_ptr->trans_lat = 0;
561     cx_ptr++;
562     sc->cpu_cx_count++;
563 
564     /*
565      * The spec says P_BLK must be 6 bytes long.  However, some systems
566      * use it to indicate a fractional set of features present so we
567      * take 5 as C2.  Some may also have a value of 7 to indicate
568      * another C3 but most use _CST for this (as required) and having
569      * "only" C1-C3 is not a hardship.
570      */
571     if (sc->cpu_p_blk_len < 5)
572 	return;
573 
574     /* Validate and allocate resources for C2 (P_LVL2). */
575     gas.SpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
576     gas.BitWidth = 8;
577     if (AcpiGbl_FADT.C2Latency <= 100) {
578 	gas.Address = sc->cpu_p_blk + 4;
579 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &sc->cpu_rid,
580 	    &gas, &cx_ptr->p_lvlx, RF_SHAREABLE);
581 	if (cx_ptr->p_lvlx != NULL) {
582 	    sc->cpu_rid++;
583 	    cx_ptr->type = ACPI_STATE_C2;
584 	    cx_ptr->trans_lat = AcpiGbl_FADT.C2Latency;
585 	    cx_ptr++;
586 	    sc->cpu_cx_count++;
587 	}
588     }
589     if (sc->cpu_p_blk_len < 6)
590 	return;
591 
592     /* Validate and allocate resources for C3 (P_LVL3). */
593     if (AcpiGbl_FADT.C3Latency <= 1000) {
594 	gas.Address = sc->cpu_p_blk + 5;
595 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &sc->cpu_rid, &gas,
596 	    &cx_ptr->p_lvlx, RF_SHAREABLE);
597 	if (cx_ptr->p_lvlx != NULL) {
598 	    sc->cpu_rid++;
599 	    cx_ptr->type = ACPI_STATE_C3;
600 	    cx_ptr->trans_lat = AcpiGbl_FADT.C3Latency;
601 	    cx_ptr++;
602 	    sc->cpu_cx_count++;
603 	}
604     }
605 
606     /* Update the largest cx_count seen so far */
607     if (sc->cpu_cx_count > cpu_cx_count)
608 	cpu_cx_count = sc->cpu_cx_count;
609 }
610 
611 /*
612  * Parse a _CST package and set up its Cx states.  Since the _CST object
613  * can change dynamically, our notify handler may call this function
614  * to clean up and probe the new _CST package.
615  */
616 static int
617 acpi_cpu_cx_cst(struct acpi_cpu_softc *sc)
618 {
619     struct	 acpi_cx *cx_ptr;
620     ACPI_STATUS	 status;
621     ACPI_BUFFER	 buf;
622     ACPI_OBJECT	*top;
623     ACPI_OBJECT	*pkg;
624     uint32_t	 count;
625     int		 i;
626 
627     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
628 
629     buf.Pointer = NULL;
630     buf.Length = ACPI_ALLOCATE_BUFFER;
631     status = AcpiEvaluateObject(sc->cpu_handle, "_CST", NULL, &buf);
632     if (ACPI_FAILURE(status))
633 	return (ENXIO);
634 
635     /* _CST is a package with a count and at least one Cx package. */
636     top = (ACPI_OBJECT *)buf.Pointer;
637     if (!ACPI_PKG_VALID(top, 2) || acpi_PkgInt32(top, 0, &count) != 0) {
638 	device_printf(sc->cpu_dev, "invalid _CST package\n");
639 	AcpiOsFree(buf.Pointer);
640 	return (ENXIO);
641     }
642     if (count != top->Package.Count - 1) {
643 	device_printf(sc->cpu_dev, "invalid _CST state count (%d != %d)\n",
644 	       count, top->Package.Count - 1);
645 	count = top->Package.Count - 1;
646     }
647     if (count > MAX_CX_STATES) {
648 	device_printf(sc->cpu_dev, "_CST has too many states (%d)\n", count);
649 	count = MAX_CX_STATES;
650     }
651 
652     /* Set up all valid states. */
653     sc->cpu_cx_count = 0;
654     cx_ptr = sc->cpu_cx_states;
655     for (i = 0; i < count; i++) {
656 	pkg = &top->Package.Elements[i + 1];
657 	if (!ACPI_PKG_VALID(pkg, 4) ||
658 	    acpi_PkgInt32(pkg, 1, &cx_ptr->type) != 0 ||
659 	    acpi_PkgInt32(pkg, 2, &cx_ptr->trans_lat) != 0 ||
660 	    acpi_PkgInt32(pkg, 3, &cx_ptr->power) != 0) {
661 
662 	    device_printf(sc->cpu_dev, "skipping invalid Cx state package\n");
663 	    continue;
664 	}
665 
666 	/* Validate the state to see if we should use it. */
667 	switch (cx_ptr->type) {
668 	case ACPI_STATE_C1:
669 	    sc->cpu_non_c3 = i;
670 	    cx_ptr++;
671 	    sc->cpu_cx_count++;
672 	    continue;
673 	case ACPI_STATE_C2:
674 	    if (cx_ptr->trans_lat > 100) {
675 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
676 				 "acpi_cpu%d: C2[%d] not available.\n",
677 				 device_get_unit(sc->cpu_dev), i));
678 		continue;
679 	    }
680 	    sc->cpu_non_c3 = i;
681 	    break;
682 	case ACPI_STATE_C3:
683 	default:
684 	    if (cx_ptr->trans_lat > 1000 ||
685 		(cpu_quirks & CPU_QUIRK_NO_C3) != 0) {
686 
687 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
688 				 "acpi_cpu%d: C3[%d] not available.\n",
689 				 device_get_unit(sc->cpu_dev), i));
690 		continue;
691 	    }
692 	    break;
693 	}
694 
695 #ifdef notyet
696 	/* Free up any previous register. */
697 	if (cx_ptr->p_lvlx != NULL) {
698 	    bus_release_resource(sc->cpu_dev, 0, 0, cx_ptr->p_lvlx);
699 	    cx_ptr->p_lvlx = NULL;
700 	}
701 #endif
702 
703 	/* Allocate the control register for C2 or C3. */
704 	acpi_PkgGas(sc->cpu_dev, pkg, 0, &cx_ptr->res_type, &sc->cpu_rid,
705 	    &cx_ptr->p_lvlx, RF_SHAREABLE);
706 	if (cx_ptr->p_lvlx) {
707 	    sc->cpu_rid++;
708 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
709 			     "acpi_cpu%d: Got C%d - %d latency\n",
710 			     device_get_unit(sc->cpu_dev), cx_ptr->type,
711 			     cx_ptr->trans_lat));
712 	    cx_ptr++;
713 	    sc->cpu_cx_count++;
714 	}
715     }
716     AcpiOsFree(buf.Pointer);
717 
718     return (0);
719 }
720 
721 /*
722  * Call this *after* all CPUs have been attached.
723  */
724 static void
725 acpi_cpu_startup(void *arg)
726 {
727     struct acpi_cpu_softc *sc;
728     int i;
729 
730     /* Get set of CPU devices */
731     devclass_get_devices(acpi_cpu_devclass, &cpu_devices, &cpu_ndevices);
732 
733     /*
734      * Setup any quirks that might necessary now that we have probed
735      * all the CPUs
736      */
737     acpi_cpu_quirks();
738 
739     cpu_cx_count = 0;
740     if (cpu_cx_generic) {
741 	/*
742 	 * We are using generic Cx mode, probe for available Cx states
743 	 * for all processors.
744 	 */
745 	for (i = 0; i < cpu_ndevices; i++) {
746 	    sc = device_get_softc(cpu_devices[i]);
747 	    acpi_cpu_generic_cx_probe(sc);
748 	}
749 
750 	/*
751 	 * Find the highest Cx state common to all CPUs
752 	 * in the system, taking quirks into account.
753 	 */
754 	for (i = 0; i < cpu_ndevices; i++) {
755 	    sc = device_get_softc(cpu_devices[i]);
756 	    if (sc->cpu_cx_count < cpu_cx_count)
757 		cpu_cx_count = sc->cpu_cx_count;
758 	}
759     } else {
760 	/*
761 	 * We are using _CST mode, remove C3 state if necessary.
762 	 * Update the largest Cx state supported in the global cpu_cx_count.
763 	 * It will be used in the global Cx sysctl handler.
764 	 * As we now know for sure that we will be using _CST mode
765 	 * install our notify handler.
766 	 */
767 	for (i = 0; i < cpu_ndevices; i++) {
768 	    sc = device_get_softc(cpu_devices[i]);
769 	    if (cpu_quirks && CPU_QUIRK_NO_C3) {
770 		sc->cpu_cx_count = sc->cpu_non_c3 + 1;
771 	    }
772 	    if (sc->cpu_cx_count > cpu_cx_count)
773 		cpu_cx_count = sc->cpu_cx_count;
774 	    AcpiInstallNotifyHandler(sc->cpu_handle, ACPI_DEVICE_NOTIFY,
775 		acpi_cpu_notify, sc);
776 	}
777     }
778 
779     /* Perform Cx final initialization. */
780     for (i = 0; i < cpu_ndevices; i++) {
781 	sc = device_get_softc(cpu_devices[i]);
782 	acpi_cpu_startup_cx(sc);
783     }
784 
785     /* Add a sysctl handler to handle global Cx lowest setting */
786     SYSCTL_ADD_PROC(&cpu_sysctl_ctx, SYSCTL_CHILDREN(cpu_sysctl_tree),
787 	OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
788 	NULL, 0, acpi_cpu_global_cx_lowest_sysctl, "A",
789 	"Global lowest Cx sleep state to use");
790 
791     /* Take over idling from cpu_idle_default(). */
792     cpu_cx_lowest = 0;
793     cpu_disable_idle = FALSE;
794     cpu_idle_hook = acpi_cpu_idle;
795 }
796 
797 static void
798 acpi_cpu_startup_cx(struct acpi_cpu_softc *sc)
799 {
800     struct sbuf sb;
801     int i;
802 
803     /*
804      * Set up the list of Cx states
805      */
806     sc->cpu_non_c3 = 0;
807     sbuf_new(&sb, sc->cpu_cx_supported, sizeof(sc->cpu_cx_supported),
808 	SBUF_FIXEDLEN);
809     for (i = 0; i < sc->cpu_cx_count; i++) {
810 	sbuf_printf(&sb, "C%d/%d ", i + 1, sc->cpu_cx_states[i].trans_lat);
811 	if (sc->cpu_cx_states[i].type < ACPI_STATE_C3)
812 	    sc->cpu_non_c3 = i;
813     }
814     sbuf_trim(&sb);
815     sbuf_finish(&sb);
816 
817     SYSCTL_ADD_STRING(&sc->cpu_sysctl_ctx,
818 		      SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
819 		      OID_AUTO, "cx_supported", CTLFLAG_RD,
820 		      sc->cpu_cx_supported, 0,
821 		      "Cx/microsecond values for supported Cx states");
822     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
823 		    SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
824 		    OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
825 		    (void *)sc, 0, acpi_cpu_cx_lowest_sysctl, "A",
826 		    "lowest Cx sleep state to use");
827     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
828 		    SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
829 		    OID_AUTO, "cx_usage", CTLTYPE_STRING | CTLFLAG_RD,
830 		    (void *)sc, 0, acpi_cpu_usage_sysctl, "A",
831 		    "percent usage for each Cx state");
832 
833 #ifdef notyet
834     /* Signal platform that we can handle _CST notification. */
835     if (!cpu_cx_generic && cpu_cst_cnt != 0) {
836 	ACPI_LOCK(acpi);
837 	AcpiOsWritePort(cpu_smi_cmd, cpu_cst_cnt, 8);
838 	ACPI_UNLOCK(acpi);
839     }
840 #endif
841 }
842 
843 /*
844  * Idle the CPU in the lowest state possible.  This function is called with
845  * interrupts disabled.  Note that once it re-enables interrupts, a task
846  * switch can occur so do not access shared data (i.e. the softc) after
847  * interrupts are re-enabled.
848  */
849 static void
850 acpi_cpu_idle()
851 {
852     struct	acpi_cpu_softc *sc;
853     struct	acpi_cx *cx_next;
854     uint32_t	start_time, end_time;
855     int		bm_active, cx_next_idx, i;
856 
857     /* If disabled, return immediately. */
858     if (cpu_disable_idle) {
859 	ACPI_ENABLE_IRQS();
860 	return;
861     }
862 
863     /*
864      * Look up our CPU id to get our softc.  If it's NULL, we'll use C1
865      * since there is no ACPI processor object for this CPU.  This occurs
866      * for logical CPUs in the HTT case.
867      */
868     sc = cpu_softc[PCPU_GET(cpuid)];
869     if (sc == NULL) {
870 	acpi_cpu_c1();
871 	return;
872     }
873 
874     /*
875      * If we slept 100 us or more, use the lowest Cx state.  Otherwise,
876      * find the lowest state that has a latency less than or equal to
877      * the length of our last sleep.
878      */
879     cx_next_idx = sc->cpu_cx_lowest;
880     if (sc->cpu_prev_sleep < 100) {
881 	/*
882 	 * If we sleep too short all the time, this system may not implement
883 	 * C2/3 correctly (i.e. reads return immediately).  In this case,
884 	 * back off and use the next higher level.
885 	 * It seems that when you have a dual core cpu (like the Intel Core Duo)
886 	 * that both cores will get out of C3 state as soon as one of them
887 	 * requires it. This breaks the sleep detection logic as the sleep
888 	 * counter is local to each cpu. Disable the sleep logic for now as a
889 	 * workaround if there's more than one CPU. The right fix would probably
890 	 * be to add quirks for system that don't really support C3 state.
891 	 */
892 	if (mp_ncpus < 2 && sc->cpu_prev_sleep <= 1) {
893 	    sc->cpu_short_slp++;
894 	    if (sc->cpu_short_slp == 1000 && sc->cpu_cx_lowest != 0) {
895 		if (sc->cpu_non_c3 == sc->cpu_cx_lowest && sc->cpu_non_c3 != 0)
896 		    sc->cpu_non_c3--;
897 		sc->cpu_cx_lowest--;
898 		sc->cpu_short_slp = 0;
899 		device_printf(sc->cpu_dev,
900 		    "too many short sleeps, backing off to C%d\n",
901 		    sc->cpu_cx_lowest + 1);
902 	    }
903 	} else
904 	    sc->cpu_short_slp = 0;
905 
906 	for (i = sc->cpu_cx_lowest; i >= 0; i--)
907 	    if (sc->cpu_cx_states[i].trans_lat <= sc->cpu_prev_sleep) {
908 		cx_next_idx = i;
909 		break;
910 	    }
911     }
912 
913     /*
914      * Check for bus master activity.  If there was activity, clear
915      * the bit and use the lowest non-C3 state.  Note that the USB
916      * driver polling for new devices keeps this bit set all the
917      * time if USB is loaded.
918      */
919     if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
920 	AcpiGetRegister(ACPI_BITREG_BUS_MASTER_STATUS, &bm_active);
921 	if (bm_active != 0) {
922 	    AcpiSetRegister(ACPI_BITREG_BUS_MASTER_STATUS, 1);
923 	    cx_next_idx = min(cx_next_idx, sc->cpu_non_c3);
924 	}
925     }
926 
927     /* Select the next state and update statistics. */
928     cx_next = &sc->cpu_cx_states[cx_next_idx];
929     sc->cpu_cx_stats[cx_next_idx]++;
930     KASSERT(cx_next->type != ACPI_STATE_C0, ("acpi_cpu_idle: C0 sleep"));
931 
932     /*
933      * Execute HLT (or equivalent) and wait for an interrupt.  We can't
934      * calculate the time spent in C1 since the place we wake up is an
935      * ISR.  Assume we slept one quantum and return.
936      */
937     if (cx_next->type == ACPI_STATE_C1) {
938 	sc->cpu_prev_sleep = 1000000 / hz;
939 	acpi_cpu_c1();
940 	return;
941     }
942 
943     /*
944      * For C3, disable bus master arbitration and enable bus master wake
945      * if BM control is available, otherwise flush the CPU cache.
946      */
947     if (cx_next->type == ACPI_STATE_C3) {
948 	if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
949 	    AcpiSetRegister(ACPI_BITREG_ARB_DISABLE, 1);
950 	    AcpiSetRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
951 	} else
952 	    ACPI_FLUSH_CPU_CACHE();
953     }
954 
955     /*
956      * Read from P_LVLx to enter C2(+), checking time spent asleep.
957      * Use the ACPI timer for measuring sleep time.  Since we need to
958      * get the time very close to the CPU start/stop clock logic, this
959      * is the only reliable time source.
960      */
961     AcpiHwLowLevelRead(32, &start_time, &AcpiGbl_FADT.XPmTimerBlock);
962     CPU_GET_REG(cx_next->p_lvlx, 1);
963 
964     /*
965      * Read the end time twice.  Since it may take an arbitrary time
966      * to enter the idle state, the first read may be executed before
967      * the processor has stopped.  Doing it again provides enough
968      * margin that we are certain to have a correct value.
969      */
970     AcpiHwLowLevelRead(32, &end_time, &AcpiGbl_FADT.XPmTimerBlock);
971     AcpiHwLowLevelRead(32, &end_time, &AcpiGbl_FADT.XPmTimerBlock);
972 
973     /* Enable bus master arbitration and disable bus master wakeup. */
974     if (cx_next->type == ACPI_STATE_C3 &&
975 	(cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
976 	AcpiSetRegister(ACPI_BITREG_ARB_DISABLE, 0);
977 	AcpiSetRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
978     }
979     ACPI_ENABLE_IRQS();
980 
981     /* Find the actual time asleep in microseconds, minus overhead. */
982     end_time = acpi_TimerDelta(end_time, start_time);
983     sc->cpu_prev_sleep = PM_USEC(end_time) - cx_next->trans_lat;
984 }
985 
986 /*
987  * Re-evaluate the _CST object when we are notified that it changed.
988  *
989  * XXX Re-evaluation disabled until locking is done.
990  */
991 static void
992 acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context)
993 {
994     struct acpi_cpu_softc *sc = (struct acpi_cpu_softc *)context;
995 
996     if (notify != ACPI_NOTIFY_CX_STATES)
997 	return;
998 
999     device_printf(sc->cpu_dev, "Cx states changed\n");
1000     /* acpi_cpu_cx_cst(sc); */
1001 }
1002 
1003 static int
1004 acpi_cpu_quirks(void)
1005 {
1006     device_t acpi_dev;
1007 
1008     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
1009 
1010     /*
1011      * Bus mastering arbitration control is needed to keep caches coherent
1012      * while sleeping in C3.  If it's not present but a working flush cache
1013      * instruction is present, flush the caches before entering C3 instead.
1014      * Otherwise, just disable C3 completely.
1015      */
1016     if (AcpiGbl_FADT.Pm2ControlBlock == 0 ||
1017 	AcpiGbl_FADT.Pm2ControlLength == 0) {
1018 	if ((AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD) &&
1019 	    (AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD_FLUSH) == 0) {
1020 	    cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1021 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1022 		"acpi_cpu: no BM control, using flush cache method\n"));
1023 	} else {
1024 	    cpu_quirks |= CPU_QUIRK_NO_C3;
1025 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1026 		"acpi_cpu: no BM control, C3 not available\n"));
1027 	}
1028     }
1029 
1030     /*
1031      * If we are using generic Cx mode, C3 on multiple CPUs requires using
1032      * the expensive flush cache instruction.
1033      */
1034     if (cpu_cx_generic && mp_ncpus > 1) {
1035 	cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1036 	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1037 	    "acpi_cpu: SMP, using flush cache mode for C3\n"));
1038     }
1039 
1040     /* Look for various quirks of the PIIX4 part. */
1041     acpi_dev = pci_find_device(PCI_VENDOR_INTEL, PCI_DEVICE_82371AB_3);
1042     if (acpi_dev != NULL) {
1043 	switch (pci_get_revid(acpi_dev)) {
1044 	/*
1045 	 * Disable C3 support for all PIIX4 chipsets.  Some of these parts
1046 	 * do not report the BMIDE status to the BM status register and
1047 	 * others have a livelock bug if Type-F DMA is enabled.  Linux
1048 	 * works around the BMIDE bug by reading the BM status directly
1049 	 * but we take the simpler approach of disabling C3 for these
1050 	 * parts.
1051 	 *
1052 	 * See erratum #18 ("C3 Power State/BMIDE and Type-F DMA
1053 	 * Livelock") from the January 2002 PIIX4 specification update.
1054 	 * Applies to all PIIX4 models.
1055 	 */
1056 	case PCI_REVISION_4E:
1057 	case PCI_REVISION_4M:
1058 	    cpu_quirks |= CPU_QUIRK_NO_C3;
1059 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1060 		"acpi_cpu: working around PIIX4 bug, disabling C3\n"));
1061 	    break;
1062 	default:
1063 	    break;
1064 	}
1065     }
1066 
1067     return (0);
1068 }
1069 
1070 static int
1071 acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS)
1072 {
1073     struct acpi_cpu_softc *sc;
1074     struct sbuf	 sb;
1075     char	 buf[128];
1076     int		 i;
1077     uintmax_t	 fract, sum, whole;
1078 
1079     sc = (struct acpi_cpu_softc *) arg1;
1080     sum = 0;
1081     for (i = 0; i < sc->cpu_cx_count; i++)
1082 	sum += sc->cpu_cx_stats[i];
1083     sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1084     for (i = 0; i < sc->cpu_cx_count; i++) {
1085 	if (sum > 0) {
1086 	    whole = (uintmax_t)sc->cpu_cx_stats[i] * 100;
1087 	    fract = (whole % sum) * 100;
1088 	    sbuf_printf(&sb, "%u.%02u%% ", (u_int)(whole / sum),
1089 		(u_int)(fract / sum));
1090 	} else
1091 	    sbuf_printf(&sb, "0%% ");
1092     }
1093     sbuf_trim(&sb);
1094     sbuf_finish(&sb);
1095     sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1096     sbuf_delete(&sb);
1097 
1098     return (0);
1099 }
1100 
1101 static int
1102 acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc, int val)
1103 {
1104     int i;
1105 
1106     ACPI_SERIAL_ASSERT(cpu);
1107     sc->cpu_cx_lowest = val;
1108 
1109     /* If not disabling, cache the new lowest non-C3 state. */
1110     sc->cpu_non_c3 = 0;
1111     for (i = sc->cpu_cx_lowest; i >= 0; i--) {
1112 	if (sc->cpu_cx_states[i].type < ACPI_STATE_C3) {
1113 	    sc->cpu_non_c3 = i;
1114 	    break;
1115 	}
1116     }
1117 
1118     /* Reset the statistics counters. */
1119     bzero(sc->cpu_cx_stats, sizeof(sc->cpu_cx_stats));
1120     return (0);
1121 }
1122 
1123 static int
1124 acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1125 {
1126     struct	 acpi_cpu_softc *sc;
1127     char	 state[8];
1128     int		 val, error;
1129 
1130     sc = (struct acpi_cpu_softc *) arg1;
1131     snprintf(state, sizeof(state), "C%d", sc->cpu_cx_lowest + 1);
1132     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1133     if (error != 0 || req->newptr == NULL)
1134 	return (error);
1135     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1136 	return (EINVAL);
1137     val = (int) strtol(state + 1, NULL, 10) - 1;
1138     if (val < 0 || val > sc->cpu_cx_count - 1)
1139 	return (EINVAL);
1140 
1141     ACPI_SERIAL_BEGIN(cpu);
1142     acpi_cpu_set_cx_lowest(sc, val);
1143     ACPI_SERIAL_END(cpu);
1144 
1145     return (0);
1146 }
1147 
1148 static int
1149 acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1150 {
1151     struct	acpi_cpu_softc *sc;
1152     char	state[8];
1153     int		val, error, i;
1154 
1155     snprintf(state, sizeof(state), "C%d", cpu_cx_lowest + 1);
1156     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1157     if (error != 0 || req->newptr == NULL)
1158 	return (error);
1159     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1160 	return (EINVAL);
1161     val = (int) strtol(state + 1, NULL, 10) - 1;
1162     if (val < 0 || val > cpu_cx_count - 1)
1163 	return (EINVAL);
1164     cpu_cx_lowest = val;
1165 
1166     /* Update the new lowest useable Cx state for all CPUs. */
1167     ACPI_SERIAL_BEGIN(cpu);
1168     for (i = 0; i < cpu_ndevices; i++) {
1169 	sc = device_get_softc(cpu_devices[i]);
1170 	acpi_cpu_set_cx_lowest(sc, val);
1171     }
1172     ACPI_SERIAL_END(cpu);
1173 
1174     return (0);
1175 }
1176