xref: /freebsd/sys/dev/acpica/acpi_cpu.c (revision 39beb93c3f8bdbf72a61fda42300b5ebed7390c8)
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 #define PIIX4_DEVACTB_REG	0x58
117 #define PIIX4_BRLD_EN_IRQ0	(1<<0)
118 #define PIIX4_BRLD_EN_IRQ	(1<<1)
119 #define PIIX4_BRLD_EN_IRQ8	(1<<5)
120 #define PIIX4_STOP_BREAK_MASK	(PIIX4_BRLD_EN_IRQ0 | PIIX4_BRLD_EN_IRQ | PIIX4_BRLD_EN_IRQ8)
121 #define PIIX4_PCNTRL_BST_EN	(1<<10)
122 
123 /* Platform hardware resource information. */
124 static uint32_t		 cpu_smi_cmd;	/* Value to write to SMI_CMD. */
125 static uint8_t		 cpu_cst_cnt;	/* Indicate we are _CST aware. */
126 static int		 cpu_quirks;	/* Indicate any hardware bugs. */
127 
128 /* Runtime state. */
129 static int		 cpu_disable_idle; /* Disable entry to idle function */
130 static int		 cpu_cx_count;	/* Number of valid Cx states */
131 
132 /* Values for sysctl. */
133 static struct sysctl_ctx_list cpu_sysctl_ctx;
134 static struct sysctl_oid *cpu_sysctl_tree;
135 static int		 cpu_cx_generic;
136 static int		 cpu_cx_lowest;
137 
138 static device_t		*cpu_devices;
139 static int		 cpu_ndevices;
140 static struct acpi_cpu_softc **cpu_softc;
141 ACPI_SERIAL_DECL(cpu, "ACPI CPU");
142 
143 static int	acpi_cpu_probe(device_t dev);
144 static int	acpi_cpu_attach(device_t dev);
145 static int	acpi_cpu_suspend(device_t dev);
146 static int	acpi_cpu_resume(device_t dev);
147 static int	acpi_pcpu_get_id(uint32_t idx, uint32_t *acpi_id,
148 		    uint32_t *cpu_id);
149 static struct resource_list *acpi_cpu_get_rlist(device_t dev, device_t child);
150 static device_t	acpi_cpu_add_child(device_t dev, int order, const char *name,
151 		    int unit);
152 static int	acpi_cpu_read_ivar(device_t dev, device_t child, int index,
153 		    uintptr_t *result);
154 static int	acpi_cpu_shutdown(device_t dev);
155 static void	acpi_cpu_cx_probe(struct acpi_cpu_softc *sc);
156 static void	acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc);
157 static int	acpi_cpu_cx_cst(struct acpi_cpu_softc *sc);
158 static void	acpi_cpu_startup(void *arg);
159 static void	acpi_cpu_startup_cx(struct acpi_cpu_softc *sc);
160 static void	acpi_cpu_cx_list(struct acpi_cpu_softc *sc);
161 static void	acpi_cpu_idle(void);
162 static void	acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context);
163 static int	acpi_cpu_quirks(void);
164 static int	acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS);
165 static int	acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc, int val);
166 static int	acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
167 static int	acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
168 
169 static device_method_t acpi_cpu_methods[] = {
170     /* Device interface */
171     DEVMETHOD(device_probe,	acpi_cpu_probe),
172     DEVMETHOD(device_attach,	acpi_cpu_attach),
173     DEVMETHOD(device_detach,	bus_generic_detach),
174     DEVMETHOD(device_shutdown,	acpi_cpu_shutdown),
175     DEVMETHOD(device_suspend,	acpi_cpu_suspend),
176     DEVMETHOD(device_resume,	acpi_cpu_resume),
177 
178     /* Bus interface */
179     DEVMETHOD(bus_add_child,	acpi_cpu_add_child),
180     DEVMETHOD(bus_read_ivar,	acpi_cpu_read_ivar),
181     DEVMETHOD(bus_get_resource_list, acpi_cpu_get_rlist),
182     DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
183     DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
184     DEVMETHOD(bus_alloc_resource, bus_generic_rl_alloc_resource),
185     DEVMETHOD(bus_release_resource, bus_generic_rl_release_resource),
186     DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
187     DEVMETHOD(bus_activate_resource, bus_generic_activate_resource),
188     DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource),
189     DEVMETHOD(bus_setup_intr,	bus_generic_setup_intr),
190     DEVMETHOD(bus_teardown_intr, bus_generic_teardown_intr),
191 
192     {0, 0}
193 };
194 
195 static driver_t acpi_cpu_driver = {
196     "cpu",
197     acpi_cpu_methods,
198     sizeof(struct acpi_cpu_softc),
199 };
200 
201 static devclass_t acpi_cpu_devclass;
202 DRIVER_MODULE(cpu, acpi, acpi_cpu_driver, acpi_cpu_devclass, 0, 0);
203 MODULE_DEPEND(cpu, acpi, 1, 1, 1);
204 
205 static int
206 acpi_cpu_probe(device_t dev)
207 {
208     int			   acpi_id, cpu_id;
209     ACPI_BUFFER		   buf;
210     ACPI_HANDLE		   handle;
211     ACPI_OBJECT		   *obj;
212     ACPI_STATUS		   status;
213 
214     if (acpi_disabled("cpu") || acpi_get_type(dev) != ACPI_TYPE_PROCESSOR)
215 	return (ENXIO);
216 
217     handle = acpi_get_handle(dev);
218     if (cpu_softc == NULL)
219 	cpu_softc = malloc(sizeof(struct acpi_cpu_softc *) *
220 	    (mp_maxid + 1), M_TEMP /* XXX */, M_WAITOK | M_ZERO);
221 
222     /* Get our Processor object. */
223     buf.Pointer = NULL;
224     buf.Length = ACPI_ALLOCATE_BUFFER;
225     status = AcpiEvaluateObject(handle, NULL, NULL, &buf);
226     if (ACPI_FAILURE(status)) {
227 	device_printf(dev, "probe failed to get Processor obj - %s\n",
228 		      AcpiFormatException(status));
229 	return (ENXIO);
230     }
231     obj = (ACPI_OBJECT *)buf.Pointer;
232     if (obj->Type != ACPI_TYPE_PROCESSOR) {
233 	device_printf(dev, "Processor object has bad type %d\n", obj->Type);
234 	AcpiOsFree(obj);
235 	return (ENXIO);
236     }
237 
238     /*
239      * Find the processor associated with our unit.  We could use the
240      * ProcId as a key, however, some boxes do not have the same values
241      * in their Processor object as the ProcId values in the MADT.
242      */
243     acpi_id = obj->Processor.ProcId;
244     AcpiOsFree(obj);
245     if (acpi_pcpu_get_id(device_get_unit(dev), &acpi_id, &cpu_id) != 0)
246 	return (ENXIO);
247 
248     /*
249      * Check if we already probed this processor.  We scan the bus twice
250      * so it's possible we've already seen this one.
251      */
252     if (cpu_softc[cpu_id] != NULL)
253 	return (ENXIO);
254 
255     /* Mark this processor as in-use and save our derived id for attach. */
256     cpu_softc[cpu_id] = (void *)1;
257     acpi_set_magic(dev, cpu_id);
258     device_set_desc(dev, "ACPI CPU");
259 
260     return (0);
261 }
262 
263 static int
264 acpi_cpu_attach(device_t dev)
265 {
266     ACPI_BUFFER		   buf;
267     ACPI_OBJECT		   arg[4], *obj;
268     ACPI_OBJECT_LIST	   arglist;
269     struct pcpu		   *pcpu_data;
270     struct acpi_cpu_softc *sc;
271     struct acpi_softc	  *acpi_sc;
272     ACPI_STATUS		   status;
273     u_int		   features;
274     int			   cpu_id, drv_count, i;
275     driver_t 		  **drivers;
276     uint32_t		   cap_set[3];
277 
278     /* UUID needed by _OSC evaluation */
279     static uint8_t cpu_oscuuid[16] = { 0x16, 0xA6, 0x77, 0x40, 0x0C, 0x29,
280 				       0xBE, 0x47, 0x9E, 0xBD, 0xD8, 0x70,
281 				       0x58, 0x71, 0x39, 0x53 };
282 
283     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
284 
285     sc = device_get_softc(dev);
286     sc->cpu_dev = dev;
287     sc->cpu_handle = acpi_get_handle(dev);
288     cpu_id = acpi_get_magic(dev);
289     cpu_softc[cpu_id] = sc;
290     pcpu_data = pcpu_find(cpu_id);
291     pcpu_data->pc_device = dev;
292     sc->cpu_pcpu = pcpu_data;
293     cpu_smi_cmd = AcpiGbl_FADT.SmiCommand;
294     cpu_cst_cnt = AcpiGbl_FADT.CstControl;
295 
296     buf.Pointer = NULL;
297     buf.Length = ACPI_ALLOCATE_BUFFER;
298     status = AcpiEvaluateObject(sc->cpu_handle, NULL, NULL, &buf);
299     if (ACPI_FAILURE(status)) {
300 	device_printf(dev, "attach failed to get Processor obj - %s\n",
301 		      AcpiFormatException(status));
302 	return (ENXIO);
303     }
304     obj = (ACPI_OBJECT *)buf.Pointer;
305     sc->cpu_p_blk = obj->Processor.PblkAddress;
306     sc->cpu_p_blk_len = obj->Processor.PblkLength;
307     sc->cpu_acpi_id = obj->Processor.ProcId;
308     AcpiOsFree(obj);
309     ACPI_DEBUG_PRINT((ACPI_DB_INFO, "acpi_cpu%d: P_BLK at %#x/%d\n",
310 		     device_get_unit(dev), sc->cpu_p_blk, sc->cpu_p_blk_len));
311 
312     /*
313      * If this is the first cpu we attach, create and initialize the generic
314      * resources that will be used by all acpi cpu devices.
315      */
316     if (device_get_unit(dev) == 0) {
317 	/* Assume we won't be using generic Cx mode by default */
318 	cpu_cx_generic = FALSE;
319 
320 	/* Install hw.acpi.cpu sysctl tree */
321 	acpi_sc = acpi_device_get_parent_softc(dev);
322 	sysctl_ctx_init(&cpu_sysctl_ctx);
323 	cpu_sysctl_tree = SYSCTL_ADD_NODE(&cpu_sysctl_ctx,
324 	    SYSCTL_CHILDREN(acpi_sc->acpi_sysctl_tree), OID_AUTO, "cpu",
325 	    CTLFLAG_RD, 0, "node for CPU children");
326 
327 	/* Queue post cpu-probing task handler */
328 	AcpiOsExecute(OSL_NOTIFY_HANDLER, acpi_cpu_startup, NULL);
329     }
330 
331     /*
332      * Before calling any CPU methods, collect child driver feature hints
333      * and notify ACPI of them.  We support unified SMP power control
334      * so advertise this ourselves.  Note this is not the same as independent
335      * SMP control where each CPU can have different settings.
336      */
337     sc->cpu_features = ACPI_CAP_SMP_SAME | ACPI_CAP_SMP_SAME_C3;
338     if (devclass_get_drivers(acpi_cpu_devclass, &drivers, &drv_count) == 0) {
339 	for (i = 0; i < drv_count; i++) {
340 	    if (ACPI_GET_FEATURES(drivers[i], &features) == 0)
341 		sc->cpu_features |= features;
342 	}
343 	free(drivers, M_TEMP);
344     }
345 
346     /*
347      * CPU capabilities are specified as a buffer of 32-bit integers:
348      * revision, count, and one or more capabilities.  The revision of
349      * "1" is not specified anywhere but seems to match Linux.
350      */
351     if (sc->cpu_features) {
352 	arglist.Pointer = arg;
353 	arglist.Count = 1;
354 	arg[0].Type = ACPI_TYPE_BUFFER;
355 	arg[0].Buffer.Length = sizeof(cap_set);
356 	arg[0].Buffer.Pointer = (uint8_t *)cap_set;
357 	cap_set[0] = 1; /* revision */
358 	cap_set[1] = 1; /* number of capabilities integers */
359 	cap_set[2] = sc->cpu_features;
360 	AcpiEvaluateObject(sc->cpu_handle, "_PDC", &arglist, NULL);
361 
362 	/*
363 	 * On some systems we need to evaluate _OSC so that the ASL
364 	 * loads the _PSS and/or _PDC methods at runtime.
365 	 *
366 	 * TODO: evaluate failure of _OSC.
367 	 */
368 	arglist.Pointer = arg;
369 	arglist.Count = 4;
370 	arg[0].Type = ACPI_TYPE_BUFFER;
371 	arg[0].Buffer.Length = sizeof(cpu_oscuuid);
372 	arg[0].Buffer.Pointer = cpu_oscuuid;	/* UUID */
373 	arg[1].Type = ACPI_TYPE_INTEGER;
374 	arg[1].Integer.Value = 1;		/* revision */
375 	arg[2].Type = ACPI_TYPE_INTEGER;
376 	arg[2].Integer.Value = 1;		/* count */
377 	arg[3].Type = ACPI_TYPE_BUFFER;
378 	arg[3].Buffer.Length = sizeof(cap_set);	/* Capabilities buffer */
379 	arg[3].Buffer.Pointer = (uint8_t *)cap_set;
380 	cap_set[0] = 0;
381 	AcpiEvaluateObject(sc->cpu_handle, "_OSC", &arglist, NULL);
382     }
383 
384     /* Probe for Cx state support. */
385     acpi_cpu_cx_probe(sc);
386 
387     /* Finally,  call identify and probe/attach for child devices. */
388     bus_generic_probe(dev);
389     bus_generic_attach(dev);
390 
391     return (0);
392 }
393 
394 /*
395  * Disable any entry to the idle function during suspend and re-enable it
396  * during resume.
397  */
398 static int
399 acpi_cpu_suspend(device_t dev)
400 {
401     int error;
402 
403     error = bus_generic_suspend(dev);
404     if (error)
405 	return (error);
406     cpu_disable_idle = TRUE;
407     return (0);
408 }
409 
410 static int
411 acpi_cpu_resume(device_t dev)
412 {
413 
414     cpu_disable_idle = FALSE;
415     return (bus_generic_resume(dev));
416 }
417 
418 /*
419  * Find the nth present CPU and return its pc_cpuid as well as set the
420  * pc_acpi_id from the most reliable source.
421  */
422 static int
423 acpi_pcpu_get_id(uint32_t idx, uint32_t *acpi_id, uint32_t *cpu_id)
424 {
425     struct pcpu	*pcpu_data;
426     uint32_t	 i;
427 
428     KASSERT(acpi_id != NULL, ("Null acpi_id"));
429     KASSERT(cpu_id != NULL, ("Null cpu_id"));
430     for (i = 0; i <= mp_maxid; i++) {
431 	if (CPU_ABSENT(i))
432 	    continue;
433 	pcpu_data = pcpu_find(i);
434 	KASSERT(pcpu_data != NULL, ("no pcpu data for %d", i));
435 	if (idx-- == 0) {
436 	    /*
437 	     * If pc_acpi_id was not initialized (e.g., a non-APIC UP box)
438 	     * override it with the value from the ASL.  Otherwise, if the
439 	     * two don't match, prefer the MADT-derived value.  Finally,
440 	     * return the pc_cpuid to reference this processor.
441 	     */
442 	    if (pcpu_data->pc_acpi_id == 0xffffffff)
443 		pcpu_data->pc_acpi_id = *acpi_id;
444 	    else if (pcpu_data->pc_acpi_id != *acpi_id)
445 		*acpi_id = pcpu_data->pc_acpi_id;
446 	    *cpu_id = pcpu_data->pc_cpuid;
447 	    return (0);
448 	}
449     }
450 
451     return (ESRCH);
452 }
453 
454 static struct resource_list *
455 acpi_cpu_get_rlist(device_t dev, device_t child)
456 {
457     struct acpi_cpu_device *ad;
458 
459     ad = device_get_ivars(child);
460     if (ad == NULL)
461 	return (NULL);
462     return (&ad->ad_rl);
463 }
464 
465 static device_t
466 acpi_cpu_add_child(device_t dev, int order, const char *name, int unit)
467 {
468     struct acpi_cpu_device *ad;
469     device_t child;
470 
471     if ((ad = malloc(sizeof(*ad), M_TEMP, M_NOWAIT | M_ZERO)) == NULL)
472 	return (NULL);
473 
474     resource_list_init(&ad->ad_rl);
475 
476     child = device_add_child_ordered(dev, order, name, unit);
477     if (child != NULL)
478 	device_set_ivars(child, ad);
479     else
480 	free(ad, M_TEMP);
481     return (child);
482 }
483 
484 static int
485 acpi_cpu_read_ivar(device_t dev, device_t child, int index, uintptr_t *result)
486 {
487     struct acpi_cpu_softc *sc;
488 
489     sc = device_get_softc(dev);
490     switch (index) {
491     case ACPI_IVAR_HANDLE:
492 	*result = (uintptr_t)sc->cpu_handle;
493 	break;
494     case CPU_IVAR_PCPU:
495 	*result = (uintptr_t)sc->cpu_pcpu;
496 	break;
497     default:
498 	return (ENOENT);
499     }
500     return (0);
501 }
502 
503 static int
504 acpi_cpu_shutdown(device_t dev)
505 {
506     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
507 
508     /* Allow children to shutdown first. */
509     bus_generic_shutdown(dev);
510 
511     /*
512      * Disable any entry to the idle function.  There is a small race where
513      * an idle thread have passed this check but not gone to sleep.  This
514      * is ok since device_shutdown() does not free the softc, otherwise
515      * we'd have to be sure all threads were evicted before returning.
516      */
517     cpu_disable_idle = TRUE;
518 
519     return_VALUE (0);
520 }
521 
522 static void
523 acpi_cpu_cx_probe(struct acpi_cpu_softc *sc)
524 {
525     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
526 
527     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
528     sc->cpu_prev_sleep = 1000000;
529     sc->cpu_cx_lowest = 0;
530 
531     /*
532      * Check for the ACPI 2.0 _CST sleep states object. If we can't find
533      * any, we'll revert to generic FADT/P_BLK Cx control method which will
534      * be handled by acpi_cpu_startup. We need to defer to after having
535      * probed all the cpus in the system before probing for generic Cx
536      * states as we may already have found cpus with valid _CST packages
537      */
538     if (!cpu_cx_generic && acpi_cpu_cx_cst(sc) != 0) {
539 	/*
540 	 * We were unable to find a _CST package for this cpu or there
541 	 * was an error parsing it. Switch back to generic mode.
542 	 */
543 	cpu_cx_generic = TRUE;
544 	if (bootverbose)
545 	    device_printf(sc->cpu_dev, "switching to generic Cx mode\n");
546     }
547 
548     /*
549      * TODO: _CSD Package should be checked here.
550      */
551 }
552 
553 static void
554 acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc)
555 {
556     ACPI_GENERIC_ADDRESS	 gas;
557     struct acpi_cx		*cx_ptr;
558 
559     sc->cpu_cx_count = 0;
560     cx_ptr = sc->cpu_cx_states;
561 
562     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
563     sc->cpu_prev_sleep = 1000000;
564 
565     /* C1 has been required since just after ACPI 1.0 */
566     cx_ptr->type = ACPI_STATE_C1;
567     cx_ptr->trans_lat = 0;
568     cx_ptr++;
569     sc->cpu_cx_count++;
570 
571     /*
572      * The spec says P_BLK must be 6 bytes long.  However, some systems
573      * use it to indicate a fractional set of features present so we
574      * take 5 as C2.  Some may also have a value of 7 to indicate
575      * another C3 but most use _CST for this (as required) and having
576      * "only" C1-C3 is not a hardship.
577      */
578     if (sc->cpu_p_blk_len < 5)
579 	return;
580 
581     /* Validate and allocate resources for C2 (P_LVL2). */
582     gas.SpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
583     gas.BitWidth = 8;
584     if (AcpiGbl_FADT.C2Latency <= 100) {
585 	gas.Address = sc->cpu_p_blk + 4;
586 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &sc->cpu_rid,
587 	    &gas, &cx_ptr->p_lvlx, RF_SHAREABLE);
588 	if (cx_ptr->p_lvlx != NULL) {
589 	    sc->cpu_rid++;
590 	    cx_ptr->type = ACPI_STATE_C2;
591 	    cx_ptr->trans_lat = AcpiGbl_FADT.C2Latency;
592 	    cx_ptr++;
593 	    sc->cpu_cx_count++;
594 	}
595     }
596     if (sc->cpu_p_blk_len < 6)
597 	return;
598 
599     /* Validate and allocate resources for C3 (P_LVL3). */
600     if (AcpiGbl_FADT.C3Latency <= 1000 && !(cpu_quirks & CPU_QUIRK_NO_C3)) {
601 	gas.Address = sc->cpu_p_blk + 5;
602 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &sc->cpu_rid, &gas,
603 	    &cx_ptr->p_lvlx, RF_SHAREABLE);
604 	if (cx_ptr->p_lvlx != NULL) {
605 	    sc->cpu_rid++;
606 	    cx_ptr->type = ACPI_STATE_C3;
607 	    cx_ptr->trans_lat = AcpiGbl_FADT.C3Latency;
608 	    cx_ptr++;
609 	    sc->cpu_cx_count++;
610 	}
611     }
612 
613     /* Update the largest cx_count seen so far */
614     if (sc->cpu_cx_count > cpu_cx_count)
615 	cpu_cx_count = sc->cpu_cx_count;
616 }
617 
618 /*
619  * Parse a _CST package and set up its Cx states.  Since the _CST object
620  * can change dynamically, our notify handler may call this function
621  * to clean up and probe the new _CST package.
622  */
623 static int
624 acpi_cpu_cx_cst(struct acpi_cpu_softc *sc)
625 {
626     struct	 acpi_cx *cx_ptr;
627     ACPI_STATUS	 status;
628     ACPI_BUFFER	 buf;
629     ACPI_OBJECT	*top;
630     ACPI_OBJECT	*pkg;
631     uint32_t	 count;
632     int		 i;
633 
634     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
635 
636     buf.Pointer = NULL;
637     buf.Length = ACPI_ALLOCATE_BUFFER;
638     status = AcpiEvaluateObject(sc->cpu_handle, "_CST", NULL, &buf);
639     if (ACPI_FAILURE(status))
640 	return (ENXIO);
641 
642     /* _CST is a package with a count and at least one Cx package. */
643     top = (ACPI_OBJECT *)buf.Pointer;
644     if (!ACPI_PKG_VALID(top, 2) || acpi_PkgInt32(top, 0, &count) != 0) {
645 	device_printf(sc->cpu_dev, "invalid _CST package\n");
646 	AcpiOsFree(buf.Pointer);
647 	return (ENXIO);
648     }
649     if (count != top->Package.Count - 1) {
650 	device_printf(sc->cpu_dev, "invalid _CST state count (%d != %d)\n",
651 	       count, top->Package.Count - 1);
652 	count = top->Package.Count - 1;
653     }
654     if (count > MAX_CX_STATES) {
655 	device_printf(sc->cpu_dev, "_CST has too many states (%d)\n", count);
656 	count = MAX_CX_STATES;
657     }
658 
659     /* Set up all valid states. */
660     sc->cpu_cx_count = 0;
661     cx_ptr = sc->cpu_cx_states;
662     for (i = 0; i < count; i++) {
663 	pkg = &top->Package.Elements[i + 1];
664 	if (!ACPI_PKG_VALID(pkg, 4) ||
665 	    acpi_PkgInt32(pkg, 1, &cx_ptr->type) != 0 ||
666 	    acpi_PkgInt32(pkg, 2, &cx_ptr->trans_lat) != 0 ||
667 	    acpi_PkgInt32(pkg, 3, &cx_ptr->power) != 0) {
668 
669 	    device_printf(sc->cpu_dev, "skipping invalid Cx state package\n");
670 	    continue;
671 	}
672 
673 	/* Validate the state to see if we should use it. */
674 	switch (cx_ptr->type) {
675 	case ACPI_STATE_C1:
676 	    sc->cpu_non_c3 = i;
677 	    cx_ptr++;
678 	    sc->cpu_cx_count++;
679 	    continue;
680 	case ACPI_STATE_C2:
681 	    if (cx_ptr->trans_lat > 100) {
682 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
683 				 "acpi_cpu%d: C2[%d] not available.\n",
684 				 device_get_unit(sc->cpu_dev), i));
685 		continue;
686 	    }
687 	    sc->cpu_non_c3 = i;
688 	    break;
689 	case ACPI_STATE_C3:
690 	default:
691 	    if (cx_ptr->trans_lat > 1000 ||
692 		(cpu_quirks & CPU_QUIRK_NO_C3) != 0) {
693 
694 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
695 				 "acpi_cpu%d: C3[%d] not available.\n",
696 				 device_get_unit(sc->cpu_dev), i));
697 		continue;
698 	    }
699 	    break;
700 	}
701 
702 #ifdef notyet
703 	/* Free up any previous register. */
704 	if (cx_ptr->p_lvlx != NULL) {
705 	    bus_release_resource(sc->cpu_dev, 0, 0, cx_ptr->p_lvlx);
706 	    cx_ptr->p_lvlx = NULL;
707 	}
708 #endif
709 
710 	/* Allocate the control register for C2 or C3. */
711 	acpi_PkgGas(sc->cpu_dev, pkg, 0, &cx_ptr->res_type, &sc->cpu_rid,
712 	    &cx_ptr->p_lvlx, RF_SHAREABLE);
713 	if (cx_ptr->p_lvlx) {
714 	    sc->cpu_rid++;
715 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
716 			     "acpi_cpu%d: Got C%d - %d latency\n",
717 			     device_get_unit(sc->cpu_dev), cx_ptr->type,
718 			     cx_ptr->trans_lat));
719 	    cx_ptr++;
720 	    sc->cpu_cx_count++;
721 	}
722     }
723     AcpiOsFree(buf.Pointer);
724 
725     return (0);
726 }
727 
728 /*
729  * Call this *after* all CPUs have been attached.
730  */
731 static void
732 acpi_cpu_startup(void *arg)
733 {
734     struct acpi_cpu_softc *sc;
735     int i;
736 
737     /* Get set of CPU devices */
738     devclass_get_devices(acpi_cpu_devclass, &cpu_devices, &cpu_ndevices);
739 
740     /*
741      * Setup any quirks that might necessary now that we have probed
742      * all the CPUs
743      */
744     acpi_cpu_quirks();
745 
746     cpu_cx_count = 0;
747     if (cpu_cx_generic) {
748 	/*
749 	 * We are using generic Cx mode, probe for available Cx states
750 	 * for all processors.
751 	 */
752 	for (i = 0; i < cpu_ndevices; i++) {
753 	    sc = device_get_softc(cpu_devices[i]);
754 	    acpi_cpu_generic_cx_probe(sc);
755 	}
756 
757 	/*
758 	 * Find the highest Cx state common to all CPUs
759 	 * in the system, taking quirks into account.
760 	 */
761 	for (i = 0; i < cpu_ndevices; i++) {
762 	    sc = device_get_softc(cpu_devices[i]);
763 	    if (sc->cpu_cx_count < cpu_cx_count)
764 		cpu_cx_count = sc->cpu_cx_count;
765 	}
766     } else {
767 	/*
768 	 * We are using _CST mode, remove C3 state if necessary.
769 	 * Update the largest Cx state supported in the global cpu_cx_count.
770 	 * It will be used in the global Cx sysctl handler.
771 	 * As we now know for sure that we will be using _CST mode
772 	 * install our notify handler.
773 	 */
774 	for (i = 0; i < cpu_ndevices; i++) {
775 	    sc = device_get_softc(cpu_devices[i]);
776 	    if (cpu_quirks & CPU_QUIRK_NO_C3) {
777 		sc->cpu_cx_count = sc->cpu_non_c3 + 1;
778 	    }
779 	    if (sc->cpu_cx_count > cpu_cx_count)
780 		cpu_cx_count = sc->cpu_cx_count;
781 	    AcpiInstallNotifyHandler(sc->cpu_handle, ACPI_DEVICE_NOTIFY,
782 		acpi_cpu_notify, sc);
783 	}
784     }
785 
786     /* Perform Cx final initialization. */
787     for (i = 0; i < cpu_ndevices; i++) {
788 	sc = device_get_softc(cpu_devices[i]);
789 	acpi_cpu_startup_cx(sc);
790     }
791 
792     /* Add a sysctl handler to handle global Cx lowest setting */
793     SYSCTL_ADD_PROC(&cpu_sysctl_ctx, SYSCTL_CHILDREN(cpu_sysctl_tree),
794 	OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
795 	NULL, 0, acpi_cpu_global_cx_lowest_sysctl, "A",
796 	"Global lowest Cx sleep state to use");
797 
798     /* Take over idling from cpu_idle_default(). */
799     cpu_cx_lowest = 0;
800     cpu_disable_idle = FALSE;
801     cpu_idle_hook = acpi_cpu_idle;
802 }
803 
804 static void
805 acpi_cpu_cx_list(struct acpi_cpu_softc *sc)
806 {
807     struct sbuf sb;
808     int i;
809 
810     /*
811      * Set up the list of Cx states
812      */
813     sc->cpu_non_c3 = 0;
814     sbuf_new(&sb, sc->cpu_cx_supported, sizeof(sc->cpu_cx_supported),
815 	SBUF_FIXEDLEN);
816     for (i = 0; i < sc->cpu_cx_count; i++) {
817 	sbuf_printf(&sb, "C%d/%d ", i + 1, sc->cpu_cx_states[i].trans_lat);
818 	if (sc->cpu_cx_states[i].type < ACPI_STATE_C3)
819 	    sc->cpu_non_c3 = i;
820     }
821     sbuf_trim(&sb);
822     sbuf_finish(&sb);
823 }
824 
825 static void
826 acpi_cpu_startup_cx(struct acpi_cpu_softc *sc)
827 {
828     acpi_cpu_cx_list(sc);
829 
830     SYSCTL_ADD_STRING(&sc->cpu_sysctl_ctx,
831 		      SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
832 		      OID_AUTO, "cx_supported", CTLFLAG_RD,
833 		      sc->cpu_cx_supported, 0,
834 		      "Cx/microsecond values for supported Cx states");
835     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
836 		    SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
837 		    OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
838 		    (void *)sc, 0, acpi_cpu_cx_lowest_sysctl, "A",
839 		    "lowest Cx sleep state to use");
840     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
841 		    SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
842 		    OID_AUTO, "cx_usage", CTLTYPE_STRING | CTLFLAG_RD,
843 		    (void *)sc, 0, acpi_cpu_usage_sysctl, "A",
844 		    "percent usage for each Cx state");
845 
846 #ifdef notyet
847     /* Signal platform that we can handle _CST notification. */
848     if (!cpu_cx_generic && cpu_cst_cnt != 0) {
849 	ACPI_LOCK(acpi);
850 	AcpiOsWritePort(cpu_smi_cmd, cpu_cst_cnt, 8);
851 	ACPI_UNLOCK(acpi);
852     }
853 #endif
854 }
855 
856 /*
857  * Idle the CPU in the lowest state possible.  This function is called with
858  * interrupts disabled.  Note that once it re-enables interrupts, a task
859  * switch can occur so do not access shared data (i.e. the softc) after
860  * interrupts are re-enabled.
861  */
862 static void
863 acpi_cpu_idle()
864 {
865     struct	acpi_cpu_softc *sc;
866     struct	acpi_cx *cx_next;
867     uint32_t	start_time, end_time;
868     int		bm_active, cx_next_idx, i;
869 
870     /* If disabled, return immediately. */
871     if (cpu_disable_idle) {
872 	ACPI_ENABLE_IRQS();
873 	return;
874     }
875 
876     /*
877      * Look up our CPU id to get our softc.  If it's NULL, we'll use C1
878      * since there is no ACPI processor object for this CPU.  This occurs
879      * for logical CPUs in the HTT case.
880      */
881     sc = cpu_softc[PCPU_GET(cpuid)];
882     if (sc == NULL) {
883 	acpi_cpu_c1();
884 	return;
885     }
886 
887     /*
888      * If we slept 100 us or more, use the lowest Cx state.  Otherwise,
889      * find the lowest state that has a latency less than or equal to
890      * the length of our last sleep.
891      */
892     cx_next_idx = sc->cpu_cx_lowest;
893     if (sc->cpu_prev_sleep < 100) {
894 	/*
895 	 * If we sleep too short all the time, this system may not implement
896 	 * C2/3 correctly (i.e. reads return immediately).  In this case,
897 	 * back off and use the next higher level.
898 	 * It seems that when you have a dual core cpu (like the Intel Core Duo)
899 	 * that both cores will get out of C3 state as soon as one of them
900 	 * requires it. This breaks the sleep detection logic as the sleep
901 	 * counter is local to each cpu. Disable the sleep logic for now as a
902 	 * workaround if there's more than one CPU. The right fix would probably
903 	 * be to add quirks for system that don't really support C3 state.
904 	 */
905 	if (mp_ncpus < 2 && sc->cpu_prev_sleep <= 1) {
906 	    sc->cpu_short_slp++;
907 	    if (sc->cpu_short_slp == 1000 && sc->cpu_cx_lowest != 0) {
908 		if (sc->cpu_non_c3 == sc->cpu_cx_lowest && sc->cpu_non_c3 != 0)
909 		    sc->cpu_non_c3--;
910 		sc->cpu_cx_lowest--;
911 		sc->cpu_short_slp = 0;
912 		device_printf(sc->cpu_dev,
913 		    "too many short sleeps, backing off to C%d\n",
914 		    sc->cpu_cx_lowest + 1);
915 	    }
916 	} else
917 	    sc->cpu_short_slp = 0;
918 
919 	for (i = sc->cpu_cx_lowest; i >= 0; i--)
920 	    if (sc->cpu_cx_states[i].trans_lat <= sc->cpu_prev_sleep) {
921 		cx_next_idx = i;
922 		break;
923 	    }
924     }
925 
926     /*
927      * Check for bus master activity.  If there was activity, clear
928      * the bit and use the lowest non-C3 state.  Note that the USB
929      * driver polling for new devices keeps this bit set all the
930      * time if USB is loaded.
931      */
932     if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
933 	AcpiGetRegister(ACPI_BITREG_BUS_MASTER_STATUS, &bm_active);
934 	if (bm_active != 0) {
935 	    AcpiSetRegister(ACPI_BITREG_BUS_MASTER_STATUS, 1);
936 	    cx_next_idx = min(cx_next_idx, sc->cpu_non_c3);
937 	}
938     }
939 
940     /* Select the next state and update statistics. */
941     cx_next = &sc->cpu_cx_states[cx_next_idx];
942     sc->cpu_cx_stats[cx_next_idx]++;
943     KASSERT(cx_next->type != ACPI_STATE_C0, ("acpi_cpu_idle: C0 sleep"));
944 
945     /*
946      * Execute HLT (or equivalent) and wait for an interrupt.  We can't
947      * calculate the time spent in C1 since the place we wake up is an
948      * ISR.  Assume we slept one quantum and return.
949      */
950     if (cx_next->type == ACPI_STATE_C1) {
951 	sc->cpu_prev_sleep = 1000000 / hz;
952 	acpi_cpu_c1();
953 	return;
954     }
955 
956     /*
957      * For C3, disable bus master arbitration and enable bus master wake
958      * if BM control is available, otherwise flush the CPU cache.
959      */
960     if (cx_next->type == ACPI_STATE_C3) {
961 	if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
962 	    AcpiSetRegister(ACPI_BITREG_ARB_DISABLE, 1);
963 	    AcpiSetRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
964 	} else
965 	    ACPI_FLUSH_CPU_CACHE();
966     }
967 
968     /*
969      * Read from P_LVLx to enter C2(+), checking time spent asleep.
970      * Use the ACPI timer for measuring sleep time.  Since we need to
971      * get the time very close to the CPU start/stop clock logic, this
972      * is the only reliable time source.
973      */
974     AcpiHwLowLevelRead(32, &start_time, &AcpiGbl_FADT.XPmTimerBlock);
975     CPU_GET_REG(cx_next->p_lvlx, 1);
976 
977     /*
978      * Read the end time twice.  Since it may take an arbitrary time
979      * to enter the idle state, the first read may be executed before
980      * the processor has stopped.  Doing it again provides enough
981      * margin that we are certain to have a correct value.
982      */
983     AcpiHwLowLevelRead(32, &end_time, &AcpiGbl_FADT.XPmTimerBlock);
984     AcpiHwLowLevelRead(32, &end_time, &AcpiGbl_FADT.XPmTimerBlock);
985 
986     /* Enable bus master arbitration and disable bus master wakeup. */
987     if (cx_next->type == ACPI_STATE_C3 &&
988 	(cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
989 	AcpiSetRegister(ACPI_BITREG_ARB_DISABLE, 0);
990 	AcpiSetRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
991     }
992     ACPI_ENABLE_IRQS();
993 
994     /* Find the actual time asleep in microseconds, minus overhead. */
995     end_time = acpi_TimerDelta(end_time, start_time);
996     sc->cpu_prev_sleep = PM_USEC(end_time) - cx_next->trans_lat;
997 }
998 
999 /*
1000  * Re-evaluate the _CST object when we are notified that it changed.
1001  *
1002  * XXX Re-evaluation disabled until locking is done.
1003  */
1004 static void
1005 acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context)
1006 {
1007     struct acpi_cpu_softc *sc = (struct acpi_cpu_softc *)context;
1008     struct acpi_cpu_softc *isc;
1009     int i;
1010 
1011     if (notify != ACPI_NOTIFY_CX_STATES)
1012 	return;
1013 
1014     /* Update the list of Cx states. */
1015     acpi_cpu_cx_cst(sc);
1016     acpi_cpu_cx_list(sc);
1017 
1018     /* Update the new lowest useable Cx state for all CPUs. */
1019     ACPI_SERIAL_BEGIN(cpu);
1020     cpu_cx_count = 0;
1021     for (i = 0; i < cpu_ndevices; i++) {
1022 	isc = device_get_softc(cpu_devices[i]);
1023 	if (isc->cpu_cx_count > cpu_cx_count)
1024 	    cpu_cx_count = isc->cpu_cx_count;
1025     }
1026     ACPI_SERIAL_END(cpu);
1027 }
1028 
1029 static int
1030 acpi_cpu_quirks(void)
1031 {
1032     device_t acpi_dev;
1033     uint32_t val;
1034 
1035     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
1036 
1037     /*
1038      * Bus mastering arbitration control is needed to keep caches coherent
1039      * while sleeping in C3.  If it's not present but a working flush cache
1040      * instruction is present, flush the caches before entering C3 instead.
1041      * Otherwise, just disable C3 completely.
1042      */
1043     if (AcpiGbl_FADT.Pm2ControlBlock == 0 ||
1044 	AcpiGbl_FADT.Pm2ControlLength == 0) {
1045 	if ((AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD) &&
1046 	    (AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD_FLUSH) == 0) {
1047 	    cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1048 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1049 		"acpi_cpu: no BM control, using flush cache method\n"));
1050 	} else {
1051 	    cpu_quirks |= CPU_QUIRK_NO_C3;
1052 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1053 		"acpi_cpu: no BM control, C3 not available\n"));
1054 	}
1055     }
1056 
1057     /*
1058      * If we are using generic Cx mode, C3 on multiple CPUs requires using
1059      * the expensive flush cache instruction.
1060      */
1061     if (cpu_cx_generic && mp_ncpus > 1) {
1062 	cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1063 	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1064 	    "acpi_cpu: SMP, using flush cache mode for C3\n"));
1065     }
1066 
1067     /* Look for various quirks of the PIIX4 part. */
1068     acpi_dev = pci_find_device(PCI_VENDOR_INTEL, PCI_DEVICE_82371AB_3);
1069     if (acpi_dev != NULL) {
1070 	switch (pci_get_revid(acpi_dev)) {
1071 	/*
1072 	 * Disable C3 support for all PIIX4 chipsets.  Some of these parts
1073 	 * do not report the BMIDE status to the BM status register and
1074 	 * others have a livelock bug if Type-F DMA is enabled.  Linux
1075 	 * works around the BMIDE bug by reading the BM status directly
1076 	 * but we take the simpler approach of disabling C3 for these
1077 	 * parts.
1078 	 *
1079 	 * See erratum #18 ("C3 Power State/BMIDE and Type-F DMA
1080 	 * Livelock") from the January 2002 PIIX4 specification update.
1081 	 * Applies to all PIIX4 models.
1082 	 *
1083 	 * Also, make sure that all interrupts cause a "Stop Break"
1084 	 * event to exit from C2 state.
1085 	 * Also, BRLD_EN_BM (ACPI_BITREG_BUS_MASTER_RLD in ACPI-speak)
1086 	 * should be set to zero, otherwise it causes C2 to short-sleep.
1087 	 * PIIX4 doesn't properly support C3 and bus master activity
1088 	 * need not break out of C2.
1089 	 */
1090 	case PCI_REVISION_A_STEP:
1091 	case PCI_REVISION_B_STEP:
1092 	case PCI_REVISION_4E:
1093 	case PCI_REVISION_4M:
1094 	    cpu_quirks |= CPU_QUIRK_NO_C3;
1095 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1096 		"acpi_cpu: working around PIIX4 bug, disabling C3\n"));
1097 
1098 	    val = pci_read_config(acpi_dev, PIIX4_DEVACTB_REG, 4);
1099 	    if ((val & PIIX4_STOP_BREAK_MASK) != PIIX4_STOP_BREAK_MASK) {
1100 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1101 		    "acpi_cpu: PIIX4: enabling IRQs to generate Stop Break\n"));
1102 	    	val |= PIIX4_STOP_BREAK_MASK;
1103 		pci_write_config(acpi_dev, PIIX4_DEVACTB_REG, val, 4);
1104 	    }
1105 	    AcpiGetRegister(ACPI_BITREG_BUS_MASTER_RLD, &val);
1106 	    if (val) {
1107 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1108 		    "acpi_cpu: PIIX4: reset BRLD_EN_BM\n"));
1109 		AcpiSetRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1110 	    }
1111 	    break;
1112 	default:
1113 	    break;
1114 	}
1115     }
1116 
1117     return (0);
1118 }
1119 
1120 static int
1121 acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS)
1122 {
1123     struct acpi_cpu_softc *sc;
1124     struct sbuf	 sb;
1125     char	 buf[128];
1126     int		 i;
1127     uintmax_t	 fract, sum, whole;
1128 
1129     sc = (struct acpi_cpu_softc *) arg1;
1130     sum = 0;
1131     for (i = 0; i < sc->cpu_cx_count; i++)
1132 	sum += sc->cpu_cx_stats[i];
1133     sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1134     for (i = 0; i < sc->cpu_cx_count; i++) {
1135 	if (sum > 0) {
1136 	    whole = (uintmax_t)sc->cpu_cx_stats[i] * 100;
1137 	    fract = (whole % sum) * 100;
1138 	    sbuf_printf(&sb, "%u.%02u%% ", (u_int)(whole / sum),
1139 		(u_int)(fract / sum));
1140 	} else
1141 	    sbuf_printf(&sb, "0%% ");
1142     }
1143     sbuf_trim(&sb);
1144     sbuf_finish(&sb);
1145     sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1146     sbuf_delete(&sb);
1147 
1148     return (0);
1149 }
1150 
1151 static int
1152 acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc, int val)
1153 {
1154     int i;
1155 
1156     ACPI_SERIAL_ASSERT(cpu);
1157     sc->cpu_cx_lowest = val;
1158 
1159     /* If not disabling, cache the new lowest non-C3 state. */
1160     sc->cpu_non_c3 = 0;
1161     for (i = sc->cpu_cx_lowest; i >= 0; i--) {
1162 	if (sc->cpu_cx_states[i].type < ACPI_STATE_C3) {
1163 	    sc->cpu_non_c3 = i;
1164 	    break;
1165 	}
1166     }
1167 
1168     /* Reset the statistics counters. */
1169     bzero(sc->cpu_cx_stats, sizeof(sc->cpu_cx_stats));
1170     return (0);
1171 }
1172 
1173 static int
1174 acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1175 {
1176     struct	 acpi_cpu_softc *sc;
1177     char	 state[8];
1178     int		 val, error;
1179 
1180     sc = (struct acpi_cpu_softc *) arg1;
1181     snprintf(state, sizeof(state), "C%d", sc->cpu_cx_lowest + 1);
1182     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1183     if (error != 0 || req->newptr == NULL)
1184 	return (error);
1185     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1186 	return (EINVAL);
1187     val = (int) strtol(state + 1, NULL, 10) - 1;
1188     if (val < 0 || val > sc->cpu_cx_count - 1)
1189 	return (EINVAL);
1190 
1191     ACPI_SERIAL_BEGIN(cpu);
1192     acpi_cpu_set_cx_lowest(sc, val);
1193     ACPI_SERIAL_END(cpu);
1194 
1195     return (0);
1196 }
1197 
1198 static int
1199 acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1200 {
1201     struct	acpi_cpu_softc *sc;
1202     char	state[8];
1203     int		val, error, i;
1204 
1205     snprintf(state, sizeof(state), "C%d", cpu_cx_lowest + 1);
1206     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1207     if (error != 0 || req->newptr == NULL)
1208 	return (error);
1209     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1210 	return (EINVAL);
1211     val = (int) strtol(state + 1, NULL, 10) - 1;
1212     if (val < 0 || val > cpu_cx_count - 1)
1213 	return (EINVAL);
1214     cpu_cx_lowest = val;
1215 
1216     /* Update the new lowest useable Cx state for all CPUs. */
1217     ACPI_SERIAL_BEGIN(cpu);
1218     for (i = 0; i < cpu_ndevices; i++) {
1219 	sc = device_get_softc(cpu_devices[i]);
1220 	acpi_cpu_set_cx_lowest(sc, val);
1221     }
1222     ACPI_SERIAL_END(cpu);
1223 
1224     return (0);
1225 }
1226