xref: /freebsd/sys/dev/acpica/acpi_cpu.c (revision 94942af266ac119ede0ca836f9aa5a5ac0582938)
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_pcpu_get_id(uint32_t idx, uint32_t *acpi_id,
140 		    uint32_t *cpu_id);
141 static struct resource_list *acpi_cpu_get_rlist(device_t dev, device_t child);
142 static device_t	acpi_cpu_add_child(device_t dev, int order, const char *name,
143 		    int unit);
144 static int	acpi_cpu_read_ivar(device_t dev, device_t child, int index,
145 		    uintptr_t *result);
146 static int	acpi_cpu_shutdown(device_t dev);
147 static void	acpi_cpu_cx_probe(struct acpi_cpu_softc *sc);
148 static void	acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc);
149 static int	acpi_cpu_cx_cst(struct acpi_cpu_softc *sc);
150 static void	acpi_cpu_startup(void *arg);
151 static void	acpi_cpu_startup_cx(struct acpi_cpu_softc *sc);
152 static void	acpi_cpu_idle(void);
153 static void	acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context);
154 static int	acpi_cpu_quirks(void);
155 static int	acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS);
156 static int	acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
157 static int	acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
158 
159 static device_method_t acpi_cpu_methods[] = {
160     /* Device interface */
161     DEVMETHOD(device_probe,	acpi_cpu_probe),
162     DEVMETHOD(device_attach,	acpi_cpu_attach),
163     DEVMETHOD(device_detach,	bus_generic_detach),
164     DEVMETHOD(device_shutdown,	acpi_cpu_shutdown),
165     DEVMETHOD(device_suspend,	bus_generic_suspend),
166     DEVMETHOD(device_resume,	bus_generic_resume),
167 
168     /* Bus interface */
169     DEVMETHOD(bus_add_child,	acpi_cpu_add_child),
170     DEVMETHOD(bus_read_ivar,	acpi_cpu_read_ivar),
171     DEVMETHOD(bus_get_resource_list, acpi_cpu_get_rlist),
172     DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
173     DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
174     DEVMETHOD(bus_alloc_resource, bus_generic_rl_alloc_resource),
175     DEVMETHOD(bus_release_resource, bus_generic_rl_release_resource),
176     DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
177     DEVMETHOD(bus_activate_resource, bus_generic_activate_resource),
178     DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource),
179     DEVMETHOD(bus_setup_intr,	bus_generic_setup_intr),
180     DEVMETHOD(bus_teardown_intr, bus_generic_teardown_intr),
181 
182     {0, 0}
183 };
184 
185 static driver_t acpi_cpu_driver = {
186     "cpu",
187     acpi_cpu_methods,
188     sizeof(struct acpi_cpu_softc),
189 };
190 
191 static devclass_t acpi_cpu_devclass;
192 DRIVER_MODULE(cpu, acpi, acpi_cpu_driver, acpi_cpu_devclass, 0, 0);
193 MODULE_DEPEND(cpu, acpi, 1, 1, 1);
194 
195 static int
196 acpi_cpu_probe(device_t dev)
197 {
198     int			   acpi_id, cpu_id;
199     ACPI_BUFFER		   buf;
200     ACPI_HANDLE		   handle;
201     ACPI_OBJECT		   *obj;
202     ACPI_STATUS		   status;
203 
204     if (acpi_disabled("cpu") || acpi_get_type(dev) != ACPI_TYPE_PROCESSOR)
205 	return (ENXIO);
206 
207     handle = acpi_get_handle(dev);
208     if (cpu_softc == NULL)
209 	cpu_softc = malloc(sizeof(struct acpi_cpu_softc *) *
210 	    (mp_maxid + 1), M_TEMP /* XXX */, M_WAITOK | M_ZERO);
211 
212     /* Get our Processor object. */
213     buf.Pointer = NULL;
214     buf.Length = ACPI_ALLOCATE_BUFFER;
215     status = AcpiEvaluateObject(handle, NULL, NULL, &buf);
216     if (ACPI_FAILURE(status)) {
217 	device_printf(dev, "probe failed to get Processor obj - %s\n",
218 		      AcpiFormatException(status));
219 	return (ENXIO);
220     }
221     obj = (ACPI_OBJECT *)buf.Pointer;
222     if (obj->Type != ACPI_TYPE_PROCESSOR) {
223 	device_printf(dev, "Processor object has bad type %d\n", obj->Type);
224 	AcpiOsFree(obj);
225 	return (ENXIO);
226     }
227 
228     /*
229      * Find the processor associated with our unit.  We could use the
230      * ProcId as a key, however, some boxes do not have the same values
231      * in their Processor object as the ProcId values in the MADT.
232      */
233     acpi_id = obj->Processor.ProcId;
234     AcpiOsFree(obj);
235     if (acpi_pcpu_get_id(device_get_unit(dev), &acpi_id, &cpu_id) != 0)
236 	return (ENXIO);
237 
238     /*
239      * Check if we already probed this processor.  We scan the bus twice
240      * so it's possible we've already seen this one.
241      */
242     if (cpu_softc[cpu_id] != NULL)
243 	return (ENXIO);
244 
245     /* Mark this processor as in-use and save our derived id for attach. */
246     cpu_softc[cpu_id] = (void *)1;
247     acpi_set_magic(dev, cpu_id);
248     device_set_desc(dev, "ACPI CPU");
249 
250     return (0);
251 }
252 
253 static int
254 acpi_cpu_attach(device_t dev)
255 {
256     ACPI_BUFFER		   buf;
257     ACPI_OBJECT		   arg, *obj;
258     ACPI_OBJECT_LIST	   arglist;
259     struct pcpu		   *pcpu_data;
260     struct acpi_cpu_softc *sc;
261     struct acpi_softc	  *acpi_sc;
262     ACPI_STATUS		   status;
263     u_int		   features;
264     int			   cpu_id, drv_count, i;
265     driver_t 		  **drivers;
266     uint32_t		   cap_set[3];
267 
268     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
269 
270     sc = device_get_softc(dev);
271     sc->cpu_dev = dev;
272     sc->cpu_handle = acpi_get_handle(dev);
273     cpu_id = acpi_get_magic(dev);
274     cpu_softc[cpu_id] = sc;
275     pcpu_data = pcpu_find(cpu_id);
276     pcpu_data->pc_device = dev;
277     sc->cpu_pcpu = pcpu_data;
278     cpu_smi_cmd = AcpiGbl_FADT.SmiCommand;
279     cpu_cst_cnt = AcpiGbl_FADT.CstControl;
280 
281     buf.Pointer = NULL;
282     buf.Length = ACPI_ALLOCATE_BUFFER;
283     status = AcpiEvaluateObject(sc->cpu_handle, NULL, NULL, &buf);
284     if (ACPI_FAILURE(status)) {
285 	device_printf(dev, "attach failed to get Processor obj - %s\n",
286 		      AcpiFormatException(status));
287 	return (ENXIO);
288     }
289     obj = (ACPI_OBJECT *)buf.Pointer;
290     sc->cpu_p_blk = obj->Processor.PblkAddress;
291     sc->cpu_p_blk_len = obj->Processor.PblkLength;
292     sc->cpu_acpi_id = obj->Processor.ProcId;
293     AcpiOsFree(obj);
294     ACPI_DEBUG_PRINT((ACPI_DB_INFO, "acpi_cpu%d: P_BLK at %#x/%d\n",
295 		     device_get_unit(dev), sc->cpu_p_blk, sc->cpu_p_blk_len));
296 
297     /*
298      * If this is the first cpu we attach, create and initialize the generic
299      * resources that will be used by all acpi cpu devices.
300      */
301     if (device_get_unit(dev) == 0) {
302 	/* Assume we won't be using generic Cx mode by default */
303 	cpu_cx_generic = FALSE;
304 
305 	/* Install hw.acpi.cpu sysctl tree */
306 	acpi_sc = acpi_device_get_parent_softc(dev);
307 	sysctl_ctx_init(&cpu_sysctl_ctx);
308 	cpu_sysctl_tree = SYSCTL_ADD_NODE(&cpu_sysctl_ctx,
309 	    SYSCTL_CHILDREN(acpi_sc->acpi_sysctl_tree), OID_AUTO, "cpu",
310 	    CTLFLAG_RD, 0, "node for CPU children");
311 
312 	/* Queue post cpu-probing task handler */
313 	AcpiOsExecute(OSL_NOTIFY_HANDLER, acpi_cpu_startup, NULL);
314     }
315 
316     /*
317      * Before calling any CPU methods, collect child driver feature hints
318      * and notify ACPI of them.  We support unified SMP power control
319      * so advertise this ourselves.  Note this is not the same as independent
320      * SMP control where each CPU can have different settings.
321      */
322     sc->cpu_features = ACPI_CAP_SMP_SAME | ACPI_CAP_SMP_SAME_C3;
323     if (devclass_get_drivers(acpi_cpu_devclass, &drivers, &drv_count) == 0) {
324 	for (i = 0; i < drv_count; i++) {
325 	    if (ACPI_GET_FEATURES(drivers[i], &features) == 0)
326 		sc->cpu_features |= features;
327 	}
328 	free(drivers, M_TEMP);
329     }
330 
331     /*
332      * CPU capabilities are specified as a buffer of 32-bit integers:
333      * revision, count, and one or more capabilities.  The revision of
334      * "1" is not specified anywhere but seems to match Linux.  We should
335      * also support _OSC here.
336      */
337     if (sc->cpu_features) {
338 	arglist.Pointer = &arg;
339 	arglist.Count = 1;
340 	arg.Type = ACPI_TYPE_BUFFER;
341 	arg.Buffer.Length = sizeof(cap_set);
342 	arg.Buffer.Pointer = (uint8_t *)cap_set;
343 	cap_set[0] = 1; /* revision */
344 	cap_set[1] = 1; /* number of capabilities integers */
345 	cap_set[2] = sc->cpu_features;
346 	AcpiEvaluateObject(sc->cpu_handle, "_PDC", &arglist, NULL);
347     }
348 
349     /* Probe for Cx state support. */
350     acpi_cpu_cx_probe(sc);
351 
352     /* Finally,  call identify and probe/attach for child devices. */
353     bus_generic_probe(dev);
354     bus_generic_attach(dev);
355 
356     return (0);
357 }
358 
359 /*
360  * Find the nth present CPU and return its pc_cpuid as well as set the
361  * pc_acpi_id from the most reliable source.
362  */
363 static int
364 acpi_pcpu_get_id(uint32_t idx, uint32_t *acpi_id, uint32_t *cpu_id)
365 {
366     struct pcpu	*pcpu_data;
367     uint32_t	 i;
368 
369     KASSERT(acpi_id != NULL, ("Null acpi_id"));
370     KASSERT(cpu_id != NULL, ("Null cpu_id"));
371     for (i = 0; i <= mp_maxid; i++) {
372 	if (CPU_ABSENT(i))
373 	    continue;
374 	pcpu_data = pcpu_find(i);
375 	KASSERT(pcpu_data != NULL, ("no pcpu data for %d", i));
376 	if (idx-- == 0) {
377 	    /*
378 	     * If pc_acpi_id was not initialized (e.g., a non-APIC UP box)
379 	     * override it with the value from the ASL.  Otherwise, if the
380 	     * two don't match, prefer the MADT-derived value.  Finally,
381 	     * return the pc_cpuid to reference this processor.
382 	     */
383 	    if (pcpu_data->pc_acpi_id == 0xffffffff)
384 		pcpu_data->pc_acpi_id = *acpi_id;
385 	    else if (pcpu_data->pc_acpi_id != *acpi_id)
386 		*acpi_id = pcpu_data->pc_acpi_id;
387 	    *cpu_id = pcpu_data->pc_cpuid;
388 	    return (0);
389 	}
390     }
391 
392     return (ESRCH);
393 }
394 
395 static struct resource_list *
396 acpi_cpu_get_rlist(device_t dev, device_t child)
397 {
398     struct acpi_cpu_device *ad;
399 
400     ad = device_get_ivars(child);
401     if (ad == NULL)
402 	return (NULL);
403     return (&ad->ad_rl);
404 }
405 
406 static device_t
407 acpi_cpu_add_child(device_t dev, int order, const char *name, int unit)
408 {
409     struct acpi_cpu_device *ad;
410     device_t child;
411 
412     if ((ad = malloc(sizeof(*ad), M_TEMP, M_NOWAIT | M_ZERO)) == NULL)
413 	return (NULL);
414 
415     resource_list_init(&ad->ad_rl);
416 
417     child = device_add_child_ordered(dev, order, name, unit);
418     if (child != NULL)
419 	device_set_ivars(child, ad);
420     else
421 	free(ad, M_TEMP);
422     return (child);
423 }
424 
425 static int
426 acpi_cpu_read_ivar(device_t dev, device_t child, int index, uintptr_t *result)
427 {
428     struct acpi_cpu_softc *sc;
429 
430     sc = device_get_softc(dev);
431     switch (index) {
432     case ACPI_IVAR_HANDLE:
433 	*result = (uintptr_t)sc->cpu_handle;
434 	break;
435     case CPU_IVAR_PCPU:
436 	*result = (uintptr_t)sc->cpu_pcpu;
437 	break;
438     default:
439 	return (ENOENT);
440     }
441     return (0);
442 }
443 
444 static int
445 acpi_cpu_shutdown(device_t dev)
446 {
447     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
448 
449     /* Allow children to shutdown first. */
450     bus_generic_shutdown(dev);
451 
452     /* Disable any entry to the idle function. */
453     cpu_disable_idle = TRUE;
454 
455     /* Signal and wait for all processors to exit acpi_cpu_idle(). */
456     smp_rendezvous(NULL, NULL, NULL, NULL);
457 
458     return_VALUE (0);
459 }
460 
461 static void
462 acpi_cpu_cx_probe(struct acpi_cpu_softc *sc)
463 {
464     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
465 
466     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
467     sc->cpu_prev_sleep = 1000000;
468     sc->cpu_cx_lowest = 0;
469 
470     /*
471      * Check for the ACPI 2.0 _CST sleep states object. If we can't find
472      * any, we'll revert to generic FADT/P_BLK Cx control method which will
473      * be handled by acpi_cpu_startup. We need to defer to after having
474      * probed all the cpus in the system before probing for generic Cx
475      * states as we may already have found cpus with valid _CST packages
476      */
477     if (!cpu_cx_generic && acpi_cpu_cx_cst(sc) != 0) {
478 	/*
479 	 * We were unable to find a _CST package for this cpu or there
480 	 * was an error parsing it. Switch back to generic mode.
481 	 */
482 	cpu_cx_generic = TRUE;
483 	if (bootverbose)
484 	    device_printf(sc->cpu_dev, "switching to generic Cx mode\n");
485     }
486 
487     /*
488      * TODO: _CSD Package should be checked here.
489      */
490 }
491 
492 static void
493 acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc)
494 {
495     ACPI_GENERIC_ADDRESS	 gas;
496     struct acpi_cx		*cx_ptr;
497 
498     sc->cpu_cx_count = 0;
499     cx_ptr = sc->cpu_cx_states;
500 
501     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
502     sc->cpu_prev_sleep = 1000000;
503 
504     /* C1 has been required since just after ACPI 1.0 */
505     cx_ptr->type = ACPI_STATE_C1;
506     cx_ptr->trans_lat = 0;
507     cx_ptr++;
508     sc->cpu_cx_count++;
509 
510     /*
511      * The spec says P_BLK must be 6 bytes long.  However, some systems
512      * use it to indicate a fractional set of features present so we
513      * take 5 as C2.  Some may also have a value of 7 to indicate
514      * another C3 but most use _CST for this (as required) and having
515      * "only" C1-C3 is not a hardship.
516      */
517     if (sc->cpu_p_blk_len < 5)
518 	return;
519 
520     /* Validate and allocate resources for C2 (P_LVL2). */
521     gas.SpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
522     gas.BitWidth = 8;
523     if (AcpiGbl_FADT.C2Latency <= 100) {
524 	gas.Address = sc->cpu_p_blk + 4;
525 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &sc->cpu_rid,
526 	    &gas, &cx_ptr->p_lvlx, RF_SHAREABLE);
527 	if (cx_ptr->p_lvlx != NULL) {
528 	    sc->cpu_rid++;
529 	    cx_ptr->type = ACPI_STATE_C2;
530 	    cx_ptr->trans_lat = AcpiGbl_FADT.C2Latency;
531 	    cx_ptr++;
532 	    sc->cpu_cx_count++;
533 	}
534     }
535     if (sc->cpu_p_blk_len < 6)
536 	return;
537 
538     /* Validate and allocate resources for C3 (P_LVL3). */
539     if (AcpiGbl_FADT.C3Latency <= 1000) {
540 	gas.Address = sc->cpu_p_blk + 5;
541 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &sc->cpu_rid, &gas,
542 	    &cx_ptr->p_lvlx, RF_SHAREABLE);
543 	if (cx_ptr->p_lvlx != NULL) {
544 	    sc->cpu_rid++;
545 	    cx_ptr->type = ACPI_STATE_C3;
546 	    cx_ptr->trans_lat = AcpiGbl_FADT.C3Latency;
547 	    cx_ptr++;
548 	    sc->cpu_cx_count++;
549 	}
550     }
551 
552     /* Update the largest cx_count seen so far */
553     if (sc->cpu_cx_count > cpu_cx_count)
554 	cpu_cx_count = sc->cpu_cx_count;
555 }
556 
557 /*
558  * Parse a _CST package and set up its Cx states.  Since the _CST object
559  * can change dynamically, our notify handler may call this function
560  * to clean up and probe the new _CST package.
561  */
562 static int
563 acpi_cpu_cx_cst(struct acpi_cpu_softc *sc)
564 {
565     struct	 acpi_cx *cx_ptr;
566     ACPI_STATUS	 status;
567     ACPI_BUFFER	 buf;
568     ACPI_OBJECT	*top;
569     ACPI_OBJECT	*pkg;
570     uint32_t	 count;
571     int		 i;
572 
573     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
574 
575     buf.Pointer = NULL;
576     buf.Length = ACPI_ALLOCATE_BUFFER;
577     status = AcpiEvaluateObject(sc->cpu_handle, "_CST", NULL, &buf);
578     if (ACPI_FAILURE(status))
579 	return (ENXIO);
580 
581     /* _CST is a package with a count and at least one Cx package. */
582     top = (ACPI_OBJECT *)buf.Pointer;
583     if (!ACPI_PKG_VALID(top, 2) || acpi_PkgInt32(top, 0, &count) != 0) {
584 	device_printf(sc->cpu_dev, "invalid _CST package\n");
585 	AcpiOsFree(buf.Pointer);
586 	return (ENXIO);
587     }
588     if (count != top->Package.Count - 1) {
589 	device_printf(sc->cpu_dev, "invalid _CST state count (%d != %d)\n",
590 	       count, top->Package.Count - 1);
591 	count = top->Package.Count - 1;
592     }
593     if (count > MAX_CX_STATES) {
594 	device_printf(sc->cpu_dev, "_CST has too many states (%d)\n", count);
595 	count = MAX_CX_STATES;
596     }
597 
598     /* Set up all valid states. */
599     sc->cpu_cx_count = 0;
600     cx_ptr = sc->cpu_cx_states;
601     for (i = 0; i < count; i++) {
602 	pkg = &top->Package.Elements[i + 1];
603 	if (!ACPI_PKG_VALID(pkg, 4) ||
604 	    acpi_PkgInt32(pkg, 1, &cx_ptr->type) != 0 ||
605 	    acpi_PkgInt32(pkg, 2, &cx_ptr->trans_lat) != 0 ||
606 	    acpi_PkgInt32(pkg, 3, &cx_ptr->power) != 0) {
607 
608 	    device_printf(sc->cpu_dev, "skipping invalid Cx state package\n");
609 	    continue;
610 	}
611 
612 	/* Validate the state to see if we should use it. */
613 	switch (cx_ptr->type) {
614 	case ACPI_STATE_C1:
615 	    sc->cpu_non_c3 = i;
616 	    cx_ptr++;
617 	    sc->cpu_cx_count++;
618 	    continue;
619 	case ACPI_STATE_C2:
620 	    if (cx_ptr->trans_lat > 100) {
621 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
622 				 "acpi_cpu%d: C2[%d] not available.\n",
623 				 device_get_unit(sc->cpu_dev), i));
624 		continue;
625 	    }
626 	    sc->cpu_non_c3 = i;
627 	    break;
628 	case ACPI_STATE_C3:
629 	default:
630 	    if (cx_ptr->trans_lat > 1000 ||
631 		(cpu_quirks & CPU_QUIRK_NO_C3) != 0) {
632 
633 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
634 				 "acpi_cpu%d: C3[%d] not available.\n",
635 				 device_get_unit(sc->cpu_dev), i));
636 		continue;
637 	    }
638 	    break;
639 	}
640 
641 #ifdef notyet
642 	/* Free up any previous register. */
643 	if (cx_ptr->p_lvlx != NULL) {
644 	    bus_release_resource(sc->cpu_dev, 0, 0, cx_ptr->p_lvlx);
645 	    cx_ptr->p_lvlx = NULL;
646 	}
647 #endif
648 
649 	/* Allocate the control register for C2 or C3. */
650 	acpi_PkgGas(sc->cpu_dev, pkg, 0, &cx_ptr->res_type, &sc->cpu_rid,
651 	    &cx_ptr->p_lvlx, RF_SHAREABLE);
652 	if (cx_ptr->p_lvlx) {
653 	    sc->cpu_rid++;
654 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
655 			     "acpi_cpu%d: Got C%d - %d latency\n",
656 			     device_get_unit(sc->cpu_dev), cx_ptr->type,
657 			     cx_ptr->trans_lat));
658 	    cx_ptr++;
659 	    sc->cpu_cx_count++;
660 	}
661     }
662     AcpiOsFree(buf.Pointer);
663 
664     return (0);
665 }
666 
667 /*
668  * Call this *after* all CPUs have been attached.
669  */
670 static void
671 acpi_cpu_startup(void *arg)
672 {
673     struct acpi_cpu_softc *sc;
674     int i;
675 
676     /* Get set of CPU devices */
677     devclass_get_devices(acpi_cpu_devclass, &cpu_devices, &cpu_ndevices);
678 
679     /*
680      * Setup any quirks that might necessary now that we have probed
681      * all the CPUs
682      */
683     acpi_cpu_quirks();
684 
685     cpu_cx_count = 0;
686     if (cpu_cx_generic) {
687 	/*
688 	 * We are using generic Cx mode, probe for available Cx states
689 	 * for all processors.
690 	 */
691 	for (i = 0; i < cpu_ndevices; i++) {
692 	    sc = device_get_softc(cpu_devices[i]);
693 	    acpi_cpu_generic_cx_probe(sc);
694 	}
695 
696 	/*
697 	 * Find the highest Cx state common to all CPUs
698 	 * in the system, taking quirks into account.
699 	 */
700 	for (i = 0; i < cpu_ndevices; i++) {
701 	    sc = device_get_softc(cpu_devices[i]);
702 	    if (sc->cpu_cx_count < cpu_cx_count)
703 		cpu_cx_count = sc->cpu_cx_count;
704 	}
705     } else {
706 	/*
707 	 * We are using _CST mode, remove C3 state if necessary.
708 	 * Update the largest Cx state supported in the global cpu_cx_count.
709 	 * It will be used in the global Cx sysctl handler.
710 	 * As we now know for sure that we will be using _CST mode
711 	 * install our notify handler.
712 	 */
713 	for (i = 0; i < cpu_ndevices; i++) {
714 	    sc = device_get_softc(cpu_devices[i]);
715 	    if (cpu_quirks && CPU_QUIRK_NO_C3) {
716 		sc->cpu_cx_count = sc->cpu_non_c3 + 1;
717 	    }
718 	    if (sc->cpu_cx_count > cpu_cx_count)
719 		cpu_cx_count = sc->cpu_cx_count;
720 	    AcpiInstallNotifyHandler(sc->cpu_handle, ACPI_DEVICE_NOTIFY,
721 		acpi_cpu_notify, sc);
722 	}
723     }
724 
725     /* Perform Cx final initialization. */
726     for (i = 0; i < cpu_ndevices; i++) {
727 	sc = device_get_softc(cpu_devices[i]);
728 	acpi_cpu_startup_cx(sc);
729     }
730 
731     /* Add a sysctl handler to handle global Cx lowest setting */
732     SYSCTL_ADD_PROC(&cpu_sysctl_ctx, SYSCTL_CHILDREN(cpu_sysctl_tree),
733 	OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
734 	NULL, 0, acpi_cpu_global_cx_lowest_sysctl, "A",
735 	"Global lowest Cx sleep state to use");
736 
737     /* Take over idling from cpu_idle_default(). */
738     cpu_cx_lowest = 0;
739     cpu_disable_idle = FALSE;
740     cpu_idle_hook = acpi_cpu_idle;
741 }
742 
743 static void
744 acpi_cpu_startup_cx(struct acpi_cpu_softc *sc)
745 {
746     struct sbuf sb;
747     int i;
748 
749     /*
750      * Set up the list of Cx states
751      */
752     sc->cpu_non_c3 = 0;
753     sbuf_new(&sb, sc->cpu_cx_supported, sizeof(sc->cpu_cx_supported),
754 	SBUF_FIXEDLEN);
755     for (i = 0; i < sc->cpu_cx_count; i++) {
756 	sbuf_printf(&sb, "C%d/%d ", i + 1, sc->cpu_cx_states[i].trans_lat);
757 	if (sc->cpu_cx_states[i].type < ACPI_STATE_C3)
758 	    sc->cpu_non_c3 = i;
759     }
760     sbuf_trim(&sb);
761     sbuf_finish(&sb);
762 
763     SYSCTL_ADD_STRING(&sc->cpu_sysctl_ctx,
764 		      SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
765 		      OID_AUTO, "cx_supported", CTLFLAG_RD,
766 		      sc->cpu_cx_supported, 0,
767 		      "Cx/microsecond values for supported Cx states");
768     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
769 		    SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
770 		    OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
771 		    (void *)sc, 0, acpi_cpu_cx_lowest_sysctl, "A",
772 		    "lowest Cx sleep state to use");
773     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
774 		    SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
775 		    OID_AUTO, "cx_usage", CTLTYPE_STRING | CTLFLAG_RD,
776 		    (void *)sc, 0, acpi_cpu_usage_sysctl, "A",
777 		    "percent usage for each Cx state");
778 
779 #ifdef notyet
780     /* Signal platform that we can handle _CST notification. */
781     if (!cpu_cx_generic && cpu_cst_cnt != 0) {
782 	ACPI_LOCK(acpi);
783 	AcpiOsWritePort(cpu_smi_cmd, cpu_cst_cnt, 8);
784 	ACPI_UNLOCK(acpi);
785     }
786 #endif
787 }
788 
789 /*
790  * Idle the CPU in the lowest state possible.  This function is called with
791  * interrupts disabled.  Note that once it re-enables interrupts, a task
792  * switch can occur so do not access shared data (i.e. the softc) after
793  * interrupts are re-enabled.
794  */
795 static void
796 acpi_cpu_idle()
797 {
798     struct	acpi_cpu_softc *sc;
799     struct	acpi_cx *cx_next;
800     uint32_t	start_time, end_time;
801     int		bm_active, cx_next_idx, i;
802 
803     /* If disabled, return immediately. */
804     if (cpu_disable_idle) {
805 	ACPI_ENABLE_IRQS();
806 	return;
807     }
808 
809     /*
810      * Look up our CPU id to get our softc.  If it's NULL, we'll use C1
811      * since there is no ACPI processor object for this CPU.  This occurs
812      * for logical CPUs in the HTT case.
813      */
814     sc = cpu_softc[PCPU_GET(cpuid)];
815     if (sc == NULL) {
816 	acpi_cpu_c1();
817 	return;
818     }
819 
820     /*
821      * If we slept 100 us or more, use the lowest Cx state.  Otherwise,
822      * find the lowest state that has a latency less than or equal to
823      * the length of our last sleep.
824      */
825     cx_next_idx = sc->cpu_cx_lowest;
826     if (sc->cpu_prev_sleep < 100) {
827 	/*
828 	 * If we sleep too short all the time, this system may not implement
829 	 * C2/3 correctly (i.e. reads return immediately).  In this case,
830 	 * back off and use the next higher level.
831 	 * It seems that when you have a dual core cpu (like the Intel Core Duo)
832 	 * that both cores will get out of C3 state as soon as one of them
833 	 * requires it. This breaks the sleep detection logic as the sleep
834 	 * counter is local to each cpu. Disable the sleep logic for now as a
835 	 * workaround if there's more than one CPU. The right fix would probably
836 	 * be to add quirks for system that don't really support C3 state.
837 	 */
838 	if (mp_ncpus < 2 && sc->cpu_prev_sleep <= 1) {
839 	    sc->cpu_short_slp++;
840 	    if (sc->cpu_short_slp == 1000 && sc->cpu_cx_lowest != 0) {
841 		if (sc->cpu_non_c3 == sc->cpu_cx_lowest && sc->cpu_non_c3 != 0)
842 		    sc->cpu_non_c3--;
843 		sc->cpu_cx_lowest--;
844 		sc->cpu_short_slp = 0;
845 		device_printf(sc->cpu_dev,
846 		    "too many short sleeps, backing off to C%d\n",
847 		    sc->cpu_cx_lowest + 1);
848 	    }
849 	} else
850 	    sc->cpu_short_slp = 0;
851 
852 	for (i = sc->cpu_cx_lowest; i >= 0; i--)
853 	    if (sc->cpu_cx_states[i].trans_lat <= sc->cpu_prev_sleep) {
854 		cx_next_idx = i;
855 		break;
856 	    }
857     }
858 
859     /*
860      * Check for bus master activity.  If there was activity, clear
861      * the bit and use the lowest non-C3 state.  Note that the USB
862      * driver polling for new devices keeps this bit set all the
863      * time if USB is loaded.
864      */
865     if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
866 	AcpiGetRegister(ACPI_BITREG_BUS_MASTER_STATUS, &bm_active);
867 	if (bm_active != 0) {
868 	    AcpiSetRegister(ACPI_BITREG_BUS_MASTER_STATUS, 1);
869 	    cx_next_idx = min(cx_next_idx, sc->cpu_non_c3);
870 	}
871     }
872 
873     /* Select the next state and update statistics. */
874     cx_next = &sc->cpu_cx_states[cx_next_idx];
875     sc->cpu_cx_stats[cx_next_idx]++;
876     KASSERT(cx_next->type != ACPI_STATE_C0, ("acpi_cpu_idle: C0 sleep"));
877 
878     /*
879      * Execute HLT (or equivalent) and wait for an interrupt.  We can't
880      * calculate the time spent in C1 since the place we wake up is an
881      * ISR.  Assume we slept one quantum and return.
882      */
883     if (cx_next->type == ACPI_STATE_C1) {
884 	sc->cpu_prev_sleep = 1000000 / hz;
885 	acpi_cpu_c1();
886 	return;
887     }
888 
889     /*
890      * For C3, disable bus master arbitration and enable bus master wake
891      * if BM control is available, otherwise flush the CPU cache.
892      */
893     if (cx_next->type == ACPI_STATE_C3) {
894 	if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
895 	    AcpiSetRegister(ACPI_BITREG_ARB_DISABLE, 1);
896 	    AcpiSetRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
897 	} else
898 	    ACPI_FLUSH_CPU_CACHE();
899     }
900 
901     /*
902      * Read from P_LVLx to enter C2(+), checking time spent asleep.
903      * Use the ACPI timer for measuring sleep time.  Since we need to
904      * get the time very close to the CPU start/stop clock logic, this
905      * is the only reliable time source.
906      */
907     AcpiHwLowLevelRead(32, &start_time, &AcpiGbl_FADT.XPmTimerBlock);
908     CPU_GET_REG(cx_next->p_lvlx, 1);
909 
910     /*
911      * Read the end time twice.  Since it may take an arbitrary time
912      * to enter the idle state, the first read may be executed before
913      * the processor has stopped.  Doing it again provides enough
914      * margin that we are certain to have a correct value.
915      */
916     AcpiHwLowLevelRead(32, &end_time, &AcpiGbl_FADT.XPmTimerBlock);
917     AcpiHwLowLevelRead(32, &end_time, &AcpiGbl_FADT.XPmTimerBlock);
918 
919     /* Enable bus master arbitration and disable bus master wakeup. */
920     if (cx_next->type == ACPI_STATE_C3 &&
921 	(cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
922 	AcpiSetRegister(ACPI_BITREG_ARB_DISABLE, 0);
923 	AcpiSetRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
924     }
925     ACPI_ENABLE_IRQS();
926 
927     /* Find the actual time asleep in microseconds, minus overhead. */
928     end_time = acpi_TimerDelta(end_time, start_time);
929     sc->cpu_prev_sleep = PM_USEC(end_time) - cx_next->trans_lat;
930 }
931 
932 /*
933  * Re-evaluate the _CST object when we are notified that it changed.
934  *
935  * XXX Re-evaluation disabled until locking is done.
936  */
937 static void
938 acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context)
939 {
940     struct acpi_cpu_softc *sc = (struct acpi_cpu_softc *)context;
941 
942     if (notify != ACPI_NOTIFY_CX_STATES)
943 	return;
944 
945     device_printf(sc->cpu_dev, "Cx states changed\n");
946     /* acpi_cpu_cx_cst(sc); */
947 }
948 
949 static int
950 acpi_cpu_quirks(void)
951 {
952     device_t acpi_dev;
953 
954     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
955 
956     /*
957      * Bus mastering arbitration control is needed to keep caches coherent
958      * while sleeping in C3.  If it's not present but a working flush cache
959      * instruction is present, flush the caches before entering C3 instead.
960      * Otherwise, just disable C3 completely.
961      */
962     if (AcpiGbl_FADT.Pm2ControlBlock == 0 ||
963 	AcpiGbl_FADT.Pm2ControlLength == 0) {
964 	if ((AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD) &&
965 	    (AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD_FLUSH) == 0) {
966 	    cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
967 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
968 		"acpi_cpu: no BM control, using flush cache method\n"));
969 	} else {
970 	    cpu_quirks |= CPU_QUIRK_NO_C3;
971 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
972 		"acpi_cpu: no BM control, C3 not available\n"));
973 	}
974     }
975 
976     /*
977      * If we are using generic Cx mode, C3 on multiple CPUs requires using
978      * the expensive flush cache instruction.
979      */
980     if (cpu_cx_generic && mp_ncpus > 1) {
981 	cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
982 	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
983 	    "acpi_cpu: SMP, using flush cache mode for C3\n"));
984     }
985 
986     /* Look for various quirks of the PIIX4 part. */
987     acpi_dev = pci_find_device(PCI_VENDOR_INTEL, PCI_DEVICE_82371AB_3);
988     if (acpi_dev != NULL) {
989 	switch (pci_get_revid(acpi_dev)) {
990 	/*
991 	 * Disable C3 support for all PIIX4 chipsets.  Some of these parts
992 	 * do not report the BMIDE status to the BM status register and
993 	 * others have a livelock bug if Type-F DMA is enabled.  Linux
994 	 * works around the BMIDE bug by reading the BM status directly
995 	 * but we take the simpler approach of disabling C3 for these
996 	 * parts.
997 	 *
998 	 * See erratum #18 ("C3 Power State/BMIDE and Type-F DMA
999 	 * Livelock") from the January 2002 PIIX4 specification update.
1000 	 * Applies to all PIIX4 models.
1001 	 */
1002 	case PCI_REVISION_4E:
1003 	case PCI_REVISION_4M:
1004 	    cpu_quirks |= CPU_QUIRK_NO_C3;
1005 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1006 		"acpi_cpu: working around PIIX4 bug, disabling C3\n"));
1007 	    break;
1008 	default:
1009 	    break;
1010 	}
1011     }
1012 
1013     return (0);
1014 }
1015 
1016 static int
1017 acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS)
1018 {
1019     struct acpi_cpu_softc *sc;
1020     struct sbuf	 sb;
1021     char	 buf[128];
1022     int		 i;
1023     uintmax_t	 fract, sum, whole;
1024 
1025     sc = (struct acpi_cpu_softc *) arg1;
1026     sum = 0;
1027     for (i = 0; i < sc->cpu_cx_count; i++)
1028 	sum += sc->cpu_cx_stats[i];
1029     sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1030     for (i = 0; i < sc->cpu_cx_count; i++) {
1031 	if (sum > 0) {
1032 	    whole = (uintmax_t)sc->cpu_cx_stats[i] * 100;
1033 	    fract = (whole % sum) * 100;
1034 	    sbuf_printf(&sb, "%u.%02u%% ", (u_int)(whole / sum),
1035 		(u_int)(fract / sum));
1036 	} else
1037 	    sbuf_printf(&sb, "0%% ");
1038     }
1039     sbuf_trim(&sb);
1040     sbuf_finish(&sb);
1041     sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1042     sbuf_delete(&sb);
1043 
1044     return (0);
1045 }
1046 
1047 static int
1048 acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1049 {
1050     struct	 acpi_cpu_softc *sc;
1051     char	 state[8];
1052     int		 val, error, i;
1053 
1054     sc = (struct acpi_cpu_softc *) arg1;
1055     snprintf(state, sizeof(state), "C%d", sc->cpu_cx_lowest + 1);
1056     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1057     if (error != 0 || req->newptr == NULL)
1058 	return (error);
1059     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1060 	return (EINVAL);
1061     val = (int) strtol(state + 1, NULL, 10) - 1;
1062     if (val < 0 || val > sc->cpu_cx_count - 1)
1063 	return (EINVAL);
1064 
1065     ACPI_SERIAL_BEGIN(cpu);
1066     sc->cpu_cx_lowest = val;
1067 
1068     /* If not disabling, cache the new lowest non-C3 state. */
1069     sc->cpu_non_c3 = 0;
1070     for (i = sc->cpu_cx_lowest; i >= 0; i--) {
1071 	if (sc->cpu_cx_states[i].type < ACPI_STATE_C3) {
1072 	    sc->cpu_non_c3 = i;
1073 	    break;
1074 	}
1075     }
1076 
1077     /* Reset the statistics counters. */
1078     bzero(sc->cpu_cx_stats, sizeof(sc->cpu_cx_stats));
1079     ACPI_SERIAL_END(cpu);
1080 
1081     return (0);
1082 }
1083 
1084 static int
1085 acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1086 {
1087     struct	acpi_cpu_softc *sc;
1088     char	state[8];
1089     int		val, error, i, j;
1090 
1091     snprintf(state, sizeof(state), "C%d", cpu_cx_lowest + 1);
1092     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1093     if (error != 0 || req->newptr == NULL)
1094 	return (error);
1095     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1096 	return (EINVAL);
1097     val = (int) strtol(state + 1, NULL, 10) - 1;
1098     if (val < 0 || val > cpu_cx_count - 1)
1099 	return (EINVAL);
1100 
1101     cpu_cx_lowest = val;
1102 
1103     /*
1104      * Update the new lowest useable Cx state for all CPUs
1105      */
1106     ACPI_SERIAL_BEGIN(cpu);
1107     for (i = 0; i < cpu_ndevices; i++) {
1108 	sc = device_get_softc(cpu_devices[i]);
1109 	sc->cpu_cx_lowest = cpu_cx_lowest;
1110 	sc->cpu_non_c3 = 0;
1111 	for (j = sc->cpu_cx_lowest; j >= 0; j++) {
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     }
1121     ACPI_SERIAL_END(cpu);
1122 
1123     return (0);
1124 }
1125