xref: /freebsd/sys/dev/acpica/acpi_cpu.c (revision e28a4053)
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/include/acpi.h>
50 #include <contrib/dev/acpica/include/accommon.h>
51 
52 #include <dev/acpica/acpivar.h>
53 
54 /*
55  * Support for ACPI Processor devices, including C[1-3] sleep states.
56  */
57 
58 /* Hooks for the ACPI CA debugging infrastructure */
59 #define _COMPONENT	ACPI_PROCESSOR
60 ACPI_MODULE_NAME("PROCESSOR")
61 
62 struct acpi_cx {
63     struct resource	*p_lvlx;	/* Register to read to enter state. */
64     uint32_t		 type;		/* C1-3 (C4 and up treated as C3). */
65     uint32_t		 trans_lat;	/* Transition latency (usec). */
66     uint32_t		 power;		/* Power consumed (mW). */
67     int			 res_type;	/* Resource type for p_lvlx. */
68 };
69 #define MAX_CX_STATES	 8
70 
71 struct acpi_cpu_softc {
72     device_t		 cpu_dev;
73     ACPI_HANDLE		 cpu_handle;
74     struct pcpu		*cpu_pcpu;
75     uint32_t		 cpu_acpi_id;	/* ACPI processor id */
76     uint32_t		 cpu_p_blk;	/* ACPI P_BLK location */
77     uint32_t		 cpu_p_blk_len;	/* P_BLK length (must be 6). */
78     struct acpi_cx	 cpu_cx_states[MAX_CX_STATES];
79     int			 cpu_cx_count;	/* Number of valid Cx states. */
80     int			 cpu_prev_sleep;/* Last idle sleep duration. */
81     int			 cpu_features;	/* Child driver supported features. */
82     /* Runtime state. */
83     int			 cpu_non_c3;	/* Index of lowest non-C3 state. */
84     u_int		 cpu_cx_stats[MAX_CX_STATES];/* Cx usage history. */
85     /* Values for sysctl. */
86     struct sysctl_ctx_list cpu_sysctl_ctx;
87     struct sysctl_oid	*cpu_sysctl_tree;
88     int			 cpu_cx_lowest;
89     char 		 cpu_cx_supported[64];
90     int			 cpu_rid;
91 };
92 
93 struct acpi_cpu_device {
94     struct resource_list	ad_rl;
95 };
96 
97 #define CPU_GET_REG(reg, width) 					\
98     (bus_space_read_ ## width(rman_get_bustag((reg)), 			\
99 		      rman_get_bushandle((reg)), 0))
100 #define CPU_SET_REG(reg, width, val)					\
101     (bus_space_write_ ## width(rman_get_bustag((reg)), 			\
102 		       rman_get_bushandle((reg)), 0, (val)))
103 
104 #define PM_USEC(x)	 ((x) >> 2)	/* ~4 clocks per usec (3.57955 Mhz) */
105 
106 #define ACPI_NOTIFY_CX_STATES	0x81	/* _CST changed. */
107 
108 #define CPU_QUIRK_NO_C3		(1<<0)	/* C3-type states are not usable. */
109 #define CPU_QUIRK_NO_BM_CTRL	(1<<2)	/* No bus mastering control. */
110 
111 #define PCI_VENDOR_INTEL	0x8086
112 #define PCI_DEVICE_82371AB_3	0x7113	/* PIIX4 chipset for quirks. */
113 #define PCI_REVISION_A_STEP	0
114 #define PCI_REVISION_B_STEP	1
115 #define PCI_REVISION_4E		2
116 #define PCI_REVISION_4M		3
117 #define PIIX4_DEVACTB_REG	0x58
118 #define PIIX4_BRLD_EN_IRQ0	(1<<0)
119 #define PIIX4_BRLD_EN_IRQ	(1<<1)
120 #define PIIX4_BRLD_EN_IRQ8	(1<<5)
121 #define PIIX4_STOP_BREAK_MASK	(PIIX4_BRLD_EN_IRQ0 | PIIX4_BRLD_EN_IRQ | PIIX4_BRLD_EN_IRQ8)
122 #define PIIX4_PCNTRL_BST_EN	(1<<10)
123 
124 /* Platform hardware resource information. */
125 static uint32_t		 cpu_smi_cmd;	/* Value to write to SMI_CMD. */
126 static uint8_t		 cpu_cst_cnt;	/* Indicate we are _CST aware. */
127 static int		 cpu_quirks;	/* Indicate any hardware bugs. */
128 
129 /* Runtime state. */
130 static int		 cpu_disable_idle; /* Disable entry to idle function */
131 static int		 cpu_cx_count;	/* Number of valid Cx states */
132 
133 /* Values for sysctl. */
134 static struct sysctl_ctx_list cpu_sysctl_ctx;
135 static struct sysctl_oid *cpu_sysctl_tree;
136 static int		 cpu_cx_generic;
137 static int		 cpu_cx_lowest;
138 
139 static device_t		*cpu_devices;
140 static int		 cpu_ndevices;
141 static struct acpi_cpu_softc **cpu_softc;
142 ACPI_SERIAL_DECL(cpu, "ACPI CPU");
143 
144 static int	acpi_cpu_probe(device_t dev);
145 static int	acpi_cpu_attach(device_t dev);
146 static int	acpi_cpu_suspend(device_t dev);
147 static int	acpi_cpu_resume(device_t dev);
148 static int	acpi_pcpu_get_id(uint32_t idx, uint32_t *acpi_id,
149 		    uint32_t *cpu_id);
150 static struct resource_list *acpi_cpu_get_rlist(device_t dev, device_t child);
151 static device_t	acpi_cpu_add_child(device_t dev, u_int order, const char *name,
152 		    int unit);
153 static int	acpi_cpu_read_ivar(device_t dev, device_t child, int index,
154 		    uintptr_t *result);
155 static int	acpi_cpu_shutdown(device_t dev);
156 static void	acpi_cpu_cx_probe(struct acpi_cpu_softc *sc);
157 static void	acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc);
158 static int	acpi_cpu_cx_cst(struct acpi_cpu_softc *sc);
159 static void	acpi_cpu_startup(void *arg);
160 static void	acpi_cpu_startup_cx(struct acpi_cpu_softc *sc);
161 static void	acpi_cpu_cx_list(struct acpi_cpu_softc *sc);
162 static void	acpi_cpu_idle(void);
163 static void	acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context);
164 static int	acpi_cpu_quirks(void);
165 static int	acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS);
166 static int	acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc, int val);
167 static int	acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
168 static int	acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
169 
170 static device_method_t acpi_cpu_methods[] = {
171     /* Device interface */
172     DEVMETHOD(device_probe,	acpi_cpu_probe),
173     DEVMETHOD(device_attach,	acpi_cpu_attach),
174     DEVMETHOD(device_detach,	bus_generic_detach),
175     DEVMETHOD(device_shutdown,	acpi_cpu_shutdown),
176     DEVMETHOD(device_suspend,	acpi_cpu_suspend),
177     DEVMETHOD(device_resume,	acpi_cpu_resume),
178 
179     /* Bus interface */
180     DEVMETHOD(bus_add_child,	acpi_cpu_add_child),
181     DEVMETHOD(bus_read_ivar,	acpi_cpu_read_ivar),
182     DEVMETHOD(bus_get_resource_list, acpi_cpu_get_rlist),
183     DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
184     DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
185     DEVMETHOD(bus_alloc_resource, bus_generic_rl_alloc_resource),
186     DEVMETHOD(bus_release_resource, bus_generic_rl_release_resource),
187     DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
188     DEVMETHOD(bus_activate_resource, bus_generic_activate_resource),
189     DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource),
190     DEVMETHOD(bus_setup_intr,	bus_generic_setup_intr),
191     DEVMETHOD(bus_teardown_intr, bus_generic_teardown_intr),
192 
193     {0, 0}
194 };
195 
196 static driver_t acpi_cpu_driver = {
197     "cpu",
198     acpi_cpu_methods,
199     sizeof(struct acpi_cpu_softc),
200 };
201 
202 static devclass_t acpi_cpu_devclass;
203 DRIVER_MODULE(cpu, acpi, acpi_cpu_driver, acpi_cpu_devclass, 0, 0);
204 MODULE_DEPEND(cpu, acpi, 1, 1, 1);
205 
206 static int
207 acpi_cpu_probe(device_t dev)
208 {
209     int			   acpi_id, cpu_id;
210     ACPI_BUFFER		   buf;
211     ACPI_HANDLE		   handle;
212     ACPI_OBJECT		   *obj;
213     ACPI_STATUS		   status;
214 
215     if (acpi_disabled("cpu") || acpi_get_type(dev) != ACPI_TYPE_PROCESSOR)
216 	return (ENXIO);
217 
218     handle = acpi_get_handle(dev);
219     if (cpu_softc == NULL)
220 	cpu_softc = malloc(sizeof(struct acpi_cpu_softc *) *
221 	    (mp_maxid + 1), M_TEMP /* XXX */, M_WAITOK | M_ZERO);
222 
223     /* Get our Processor object. */
224     buf.Pointer = NULL;
225     buf.Length = ACPI_ALLOCATE_BUFFER;
226     status = AcpiEvaluateObject(handle, NULL, NULL, &buf);
227     if (ACPI_FAILURE(status)) {
228 	device_printf(dev, "probe failed to get Processor obj - %s\n",
229 		      AcpiFormatException(status));
230 	return (ENXIO);
231     }
232     obj = (ACPI_OBJECT *)buf.Pointer;
233     if (obj->Type != ACPI_TYPE_PROCESSOR) {
234 	device_printf(dev, "Processor object has bad type %d\n", obj->Type);
235 	AcpiOsFree(obj);
236 	return (ENXIO);
237     }
238 
239     /*
240      * Find the processor associated with our unit.  We could use the
241      * ProcId as a key, however, some boxes do not have the same values
242      * in their Processor object as the ProcId values in the MADT.
243      */
244     acpi_id = obj->Processor.ProcId;
245     AcpiOsFree(obj);
246     if (acpi_pcpu_get_id(device_get_unit(dev), &acpi_id, &cpu_id) != 0)
247 	return (ENXIO);
248 
249     /*
250      * Check if we already probed this processor.  We scan the bus twice
251      * so it's possible we've already seen this one.
252      */
253     if (cpu_softc[cpu_id] != NULL)
254 	return (ENXIO);
255 
256     /* Mark this processor as in-use and save our derived id for attach. */
257     cpu_softc[cpu_id] = (void *)1;
258     acpi_set_private(dev, (void*)(intptr_t)cpu_id);
259     device_set_desc(dev, "ACPI CPU");
260 
261     return (0);
262 }
263 
264 static int
265 acpi_cpu_attach(device_t dev)
266 {
267     ACPI_BUFFER		   buf;
268     ACPI_OBJECT		   arg[4], *obj;
269     ACPI_OBJECT_LIST	   arglist;
270     struct pcpu		   *pcpu_data;
271     struct acpi_cpu_softc *sc;
272     struct acpi_softc	  *acpi_sc;
273     ACPI_STATUS		   status;
274     u_int		   features;
275     int			   cpu_id, drv_count, i;
276     driver_t 		  **drivers;
277     uint32_t		   cap_set[3];
278 
279     /* UUID needed by _OSC evaluation */
280     static uint8_t cpu_oscuuid[16] = { 0x16, 0xA6, 0x77, 0x40, 0x0C, 0x29,
281 				       0xBE, 0x47, 0x9E, 0xBD, 0xD8, 0x70,
282 				       0x58, 0x71, 0x39, 0x53 };
283 
284     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
285 
286     sc = device_get_softc(dev);
287     sc->cpu_dev = dev;
288     sc->cpu_handle = acpi_get_handle(dev);
289     cpu_id = (int)(intptr_t)acpi_get_private(dev);
290     cpu_softc[cpu_id] = sc;
291     pcpu_data = pcpu_find(cpu_id);
292     pcpu_data->pc_device = dev;
293     sc->cpu_pcpu = pcpu_data;
294     cpu_smi_cmd = AcpiGbl_FADT.SmiCommand;
295     cpu_cst_cnt = AcpiGbl_FADT.CstControl;
296 
297     buf.Pointer = NULL;
298     buf.Length = ACPI_ALLOCATE_BUFFER;
299     status = AcpiEvaluateObject(sc->cpu_handle, NULL, NULL, &buf);
300     if (ACPI_FAILURE(status)) {
301 	device_printf(dev, "attach failed to get Processor obj - %s\n",
302 		      AcpiFormatException(status));
303 	return (ENXIO);
304     }
305     obj = (ACPI_OBJECT *)buf.Pointer;
306     sc->cpu_p_blk = obj->Processor.PblkAddress;
307     sc->cpu_p_blk_len = obj->Processor.PblkLength;
308     sc->cpu_acpi_id = obj->Processor.ProcId;
309     AcpiOsFree(obj);
310     ACPI_DEBUG_PRINT((ACPI_DB_INFO, "acpi_cpu%d: P_BLK at %#x/%d\n",
311 		     device_get_unit(dev), sc->cpu_p_blk, sc->cpu_p_blk_len));
312 
313     /*
314      * If this is the first cpu we attach, create and initialize the generic
315      * resources that will be used by all acpi cpu devices.
316      */
317     if (device_get_unit(dev) == 0) {
318 	/* Assume we won't be using generic Cx mode by default */
319 	cpu_cx_generic = FALSE;
320 
321 	/* Install hw.acpi.cpu sysctl tree */
322 	acpi_sc = acpi_device_get_parent_softc(dev);
323 	sysctl_ctx_init(&cpu_sysctl_ctx);
324 	cpu_sysctl_tree = SYSCTL_ADD_NODE(&cpu_sysctl_ctx,
325 	    SYSCTL_CHILDREN(acpi_sc->acpi_sysctl_tree), OID_AUTO, "cpu",
326 	    CTLFLAG_RD, 0, "node for CPU children");
327 
328 	/* Queue post cpu-probing task handler */
329 	AcpiOsExecute(OSL_NOTIFY_HANDLER, acpi_cpu_startup, NULL);
330     }
331 
332     /*
333      * Before calling any CPU methods, collect child driver feature hints
334      * and notify ACPI of them.  We support unified SMP power control
335      * so advertise this ourselves.  Note this is not the same as independent
336      * SMP control where each CPU can have different settings.
337      */
338     sc->cpu_features = ACPI_CAP_SMP_SAME | ACPI_CAP_SMP_SAME_C3;
339     if (devclass_get_drivers(acpi_cpu_devclass, &drivers, &drv_count) == 0) {
340 	for (i = 0; i < drv_count; i++) {
341 	    if (ACPI_GET_FEATURES(drivers[i], &features) == 0)
342 		sc->cpu_features |= features;
343 	}
344 	free(drivers, M_TEMP);
345     }
346 
347     /*
348      * CPU capabilities are specified in
349      * Intel Processor Vendor-Specific ACPI Interface Specification.
350      */
351     if (sc->cpu_features) {
352 	arglist.Pointer = arg;
353 	arglist.Count = 4;
354 	arg[0].Type = ACPI_TYPE_BUFFER;
355 	arg[0].Buffer.Length = sizeof(cpu_oscuuid);
356 	arg[0].Buffer.Pointer = cpu_oscuuid;	/* UUID */
357 	arg[1].Type = ACPI_TYPE_INTEGER;
358 	arg[1].Integer.Value = 1;		/* revision */
359 	arg[2].Type = ACPI_TYPE_INTEGER;
360 	arg[2].Integer.Value = 1;		/* count */
361 	arg[3].Type = ACPI_TYPE_BUFFER;
362 	arg[3].Buffer.Length = sizeof(cap_set);	/* Capabilities buffer */
363 	arg[3].Buffer.Pointer = (uint8_t *)cap_set;
364 	cap_set[0] = 0;				/* status */
365 	cap_set[1] = sc->cpu_features;
366 	status = AcpiEvaluateObject(sc->cpu_handle, "_OSC", &arglist, NULL);
367 	if (ACPI_SUCCESS(status)) {
368 	    if (cap_set[0] != 0)
369 		device_printf(dev, "_OSC returned status %#x\n", cap_set[0]);
370 	}
371 	else {
372 	    arglist.Pointer = arg;
373 	    arglist.Count = 1;
374 	    arg[0].Type = ACPI_TYPE_BUFFER;
375 	    arg[0].Buffer.Length = sizeof(cap_set);
376 	    arg[0].Buffer.Pointer = (uint8_t *)cap_set;
377 	    cap_set[0] = 1; /* revision */
378 	    cap_set[1] = 1; /* number of capabilities integers */
379 	    cap_set[2] = sc->cpu_features;
380 	    AcpiEvaluateObject(sc->cpu_handle, "_PDC", &arglist, NULL);
381 	}
382     }
383 
384     /* Probe for Cx state support. */
385     acpi_cpu_cx_probe(sc);
386 
387     return (0);
388 }
389 
390 static void
391 acpi_cpu_postattach(void *unused __unused)
392 {
393     device_t *devices;
394     int err;
395     int i, n;
396 
397     err = devclass_get_devices(acpi_cpu_devclass, &devices, &n);
398     if (err != 0) {
399 	printf("devclass_get_devices(acpi_cpu_devclass) failed\n");
400 	return;
401     }
402     for (i = 0; i < n; i++)
403 	bus_generic_probe(devices[i]);
404     for (i = 0; i < n; i++)
405 	bus_generic_attach(devices[i]);
406     free(devices, M_TEMP);
407 }
408 
409 SYSINIT(acpi_cpu, SI_SUB_CONFIGURE, SI_ORDER_MIDDLE,
410     acpi_cpu_postattach, NULL);
411 
412 /*
413  * Disable any entry to the idle function during suspend and re-enable it
414  * during resume.
415  */
416 static int
417 acpi_cpu_suspend(device_t dev)
418 {
419     int error;
420 
421     error = bus_generic_suspend(dev);
422     if (error)
423 	return (error);
424     cpu_disable_idle = TRUE;
425     return (0);
426 }
427 
428 static int
429 acpi_cpu_resume(device_t dev)
430 {
431 
432     cpu_disable_idle = FALSE;
433     return (bus_generic_resume(dev));
434 }
435 
436 /*
437  * Find the nth present CPU and return its pc_cpuid as well as set the
438  * pc_acpi_id from the most reliable source.
439  */
440 static int
441 acpi_pcpu_get_id(uint32_t idx, uint32_t *acpi_id, uint32_t *cpu_id)
442 {
443     struct pcpu	*pcpu_data;
444     uint32_t	 i;
445 
446     KASSERT(acpi_id != NULL, ("Null acpi_id"));
447     KASSERT(cpu_id != NULL, ("Null cpu_id"));
448     CPU_FOREACH(i) {
449 	pcpu_data = pcpu_find(i);
450 	KASSERT(pcpu_data != NULL, ("no pcpu data for %d", i));
451 	if (idx-- == 0) {
452 	    /*
453 	     * If pc_acpi_id was not initialized (e.g., a non-APIC UP box)
454 	     * override it with the value from the ASL.  Otherwise, if the
455 	     * two don't match, prefer the MADT-derived value.  Finally,
456 	     * return the pc_cpuid to reference this processor.
457 	     */
458 	    if (pcpu_data->pc_acpi_id == 0xffffffff)
459 		pcpu_data->pc_acpi_id = *acpi_id;
460 	    else if (pcpu_data->pc_acpi_id != *acpi_id)
461 		*acpi_id = pcpu_data->pc_acpi_id;
462 	    *cpu_id = pcpu_data->pc_cpuid;
463 	    return (0);
464 	}
465     }
466 
467     return (ESRCH);
468 }
469 
470 static struct resource_list *
471 acpi_cpu_get_rlist(device_t dev, device_t child)
472 {
473     struct acpi_cpu_device *ad;
474 
475     ad = device_get_ivars(child);
476     if (ad == NULL)
477 	return (NULL);
478     return (&ad->ad_rl);
479 }
480 
481 static device_t
482 acpi_cpu_add_child(device_t dev, u_int order, const char *name, int unit)
483 {
484     struct acpi_cpu_device *ad;
485     device_t child;
486 
487     if ((ad = malloc(sizeof(*ad), M_TEMP, M_NOWAIT | M_ZERO)) == NULL)
488 	return (NULL);
489 
490     resource_list_init(&ad->ad_rl);
491 
492     child = device_add_child_ordered(dev, order, name, unit);
493     if (child != NULL)
494 	device_set_ivars(child, ad);
495     else
496 	free(ad, M_TEMP);
497     return (child);
498 }
499 
500 static int
501 acpi_cpu_read_ivar(device_t dev, device_t child, int index, uintptr_t *result)
502 {
503     struct acpi_cpu_softc *sc;
504 
505     sc = device_get_softc(dev);
506     switch (index) {
507     case ACPI_IVAR_HANDLE:
508 	*result = (uintptr_t)sc->cpu_handle;
509 	break;
510     case CPU_IVAR_PCPU:
511 	*result = (uintptr_t)sc->cpu_pcpu;
512 	break;
513     default:
514 	return (ENOENT);
515     }
516     return (0);
517 }
518 
519 static int
520 acpi_cpu_shutdown(device_t dev)
521 {
522     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
523 
524     /* Allow children to shutdown first. */
525     bus_generic_shutdown(dev);
526 
527     /*
528      * Disable any entry to the idle function.  There is a small race where
529      * an idle thread have passed this check but not gone to sleep.  This
530      * is ok since device_shutdown() does not free the softc, otherwise
531      * we'd have to be sure all threads were evicted before returning.
532      */
533     cpu_disable_idle = TRUE;
534 
535     return_VALUE (0);
536 }
537 
538 static void
539 acpi_cpu_cx_probe(struct acpi_cpu_softc *sc)
540 {
541     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
542 
543     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
544     sc->cpu_prev_sleep = 1000000;
545     sc->cpu_cx_lowest = 0;
546 
547     /*
548      * Check for the ACPI 2.0 _CST sleep states object. If we can't find
549      * any, we'll revert to generic FADT/P_BLK Cx control method which will
550      * be handled by acpi_cpu_startup. We need to defer to after having
551      * probed all the cpus in the system before probing for generic Cx
552      * states as we may already have found cpus with valid _CST packages
553      */
554     if (!cpu_cx_generic && acpi_cpu_cx_cst(sc) != 0) {
555 	/*
556 	 * We were unable to find a _CST package for this cpu or there
557 	 * was an error parsing it. Switch back to generic mode.
558 	 */
559 	cpu_cx_generic = TRUE;
560 	if (bootverbose)
561 	    device_printf(sc->cpu_dev, "switching to generic Cx mode\n");
562     }
563 
564     /*
565      * TODO: _CSD Package should be checked here.
566      */
567 }
568 
569 static void
570 acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc)
571 {
572     ACPI_GENERIC_ADDRESS	 gas;
573     struct acpi_cx		*cx_ptr;
574 
575     sc->cpu_cx_count = 0;
576     cx_ptr = sc->cpu_cx_states;
577 
578     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
579     sc->cpu_prev_sleep = 1000000;
580 
581     /* C1 has been required since just after ACPI 1.0 */
582     cx_ptr->type = ACPI_STATE_C1;
583     cx_ptr->trans_lat = 0;
584     cx_ptr++;
585     sc->cpu_cx_count++;
586 
587     /*
588      * The spec says P_BLK must be 6 bytes long.  However, some systems
589      * use it to indicate a fractional set of features present so we
590      * take 5 as C2.  Some may also have a value of 7 to indicate
591      * another C3 but most use _CST for this (as required) and having
592      * "only" C1-C3 is not a hardship.
593      */
594     if (sc->cpu_p_blk_len < 5)
595 	return;
596 
597     /* Validate and allocate resources for C2 (P_LVL2). */
598     gas.SpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
599     gas.BitWidth = 8;
600     if (AcpiGbl_FADT.C2Latency <= 100) {
601 	gas.Address = sc->cpu_p_blk + 4;
602 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &sc->cpu_rid,
603 	    &gas, &cx_ptr->p_lvlx, RF_SHAREABLE);
604 	if (cx_ptr->p_lvlx != NULL) {
605 	    sc->cpu_rid++;
606 	    cx_ptr->type = ACPI_STATE_C2;
607 	    cx_ptr->trans_lat = AcpiGbl_FADT.C2Latency;
608 	    cx_ptr++;
609 	    sc->cpu_cx_count++;
610 	}
611     }
612     if (sc->cpu_p_blk_len < 6)
613 	return;
614 
615     /* Validate and allocate resources for C3 (P_LVL3). */
616     if (AcpiGbl_FADT.C3Latency <= 1000 && !(cpu_quirks & CPU_QUIRK_NO_C3)) {
617 	gas.Address = sc->cpu_p_blk + 5;
618 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &sc->cpu_rid, &gas,
619 	    &cx_ptr->p_lvlx, RF_SHAREABLE);
620 	if (cx_ptr->p_lvlx != NULL) {
621 	    sc->cpu_rid++;
622 	    cx_ptr->type = ACPI_STATE_C3;
623 	    cx_ptr->trans_lat = AcpiGbl_FADT.C3Latency;
624 	    cx_ptr++;
625 	    sc->cpu_cx_count++;
626 	}
627     }
628 }
629 
630 /*
631  * Parse a _CST package and set up its Cx states.  Since the _CST object
632  * can change dynamically, our notify handler may call this function
633  * to clean up and probe the new _CST package.
634  */
635 static int
636 acpi_cpu_cx_cst(struct acpi_cpu_softc *sc)
637 {
638     struct	 acpi_cx *cx_ptr;
639     ACPI_STATUS	 status;
640     ACPI_BUFFER	 buf;
641     ACPI_OBJECT	*top;
642     ACPI_OBJECT	*pkg;
643     uint32_t	 count;
644     int		 i;
645 
646     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
647 
648     buf.Pointer = NULL;
649     buf.Length = ACPI_ALLOCATE_BUFFER;
650     status = AcpiEvaluateObject(sc->cpu_handle, "_CST", NULL, &buf);
651     if (ACPI_FAILURE(status))
652 	return (ENXIO);
653 
654     /* _CST is a package with a count and at least one Cx package. */
655     top = (ACPI_OBJECT *)buf.Pointer;
656     if (!ACPI_PKG_VALID(top, 2) || acpi_PkgInt32(top, 0, &count) != 0) {
657 	device_printf(sc->cpu_dev, "invalid _CST package\n");
658 	AcpiOsFree(buf.Pointer);
659 	return (ENXIO);
660     }
661     if (count != top->Package.Count - 1) {
662 	device_printf(sc->cpu_dev, "invalid _CST state count (%d != %d)\n",
663 	       count, top->Package.Count - 1);
664 	count = top->Package.Count - 1;
665     }
666     if (count > MAX_CX_STATES) {
667 	device_printf(sc->cpu_dev, "_CST has too many states (%d)\n", count);
668 	count = MAX_CX_STATES;
669     }
670 
671     /* Set up all valid states. */
672     sc->cpu_cx_count = 0;
673     cx_ptr = sc->cpu_cx_states;
674     for (i = 0; i < count; i++) {
675 	pkg = &top->Package.Elements[i + 1];
676 	if (!ACPI_PKG_VALID(pkg, 4) ||
677 	    acpi_PkgInt32(pkg, 1, &cx_ptr->type) != 0 ||
678 	    acpi_PkgInt32(pkg, 2, &cx_ptr->trans_lat) != 0 ||
679 	    acpi_PkgInt32(pkg, 3, &cx_ptr->power) != 0) {
680 
681 	    device_printf(sc->cpu_dev, "skipping invalid Cx state package\n");
682 	    continue;
683 	}
684 
685 	/* Validate the state to see if we should use it. */
686 	switch (cx_ptr->type) {
687 	case ACPI_STATE_C1:
688 	    sc->cpu_non_c3 = i;
689 	    cx_ptr++;
690 	    sc->cpu_cx_count++;
691 	    continue;
692 	case ACPI_STATE_C2:
693 	    sc->cpu_non_c3 = i;
694 	    break;
695 	case ACPI_STATE_C3:
696 	default:
697 	    if ((cpu_quirks & CPU_QUIRK_NO_C3) != 0) {
698 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
699 				 "acpi_cpu%d: C3[%d] not available.\n",
700 				 device_get_unit(sc->cpu_dev), i));
701 		continue;
702 	    }
703 	    break;
704 	}
705 
706 #ifdef notyet
707 	/* Free up any previous register. */
708 	if (cx_ptr->p_lvlx != NULL) {
709 	    bus_release_resource(sc->cpu_dev, 0, 0, cx_ptr->p_lvlx);
710 	    cx_ptr->p_lvlx = NULL;
711 	}
712 #endif
713 
714 	/* Allocate the control register for C2 or C3. */
715 	acpi_PkgGas(sc->cpu_dev, pkg, 0, &cx_ptr->res_type, &sc->cpu_rid,
716 	    &cx_ptr->p_lvlx, RF_SHAREABLE);
717 	if (cx_ptr->p_lvlx) {
718 	    sc->cpu_rid++;
719 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
720 			     "acpi_cpu%d: Got C%d - %d latency\n",
721 			     device_get_unit(sc->cpu_dev), cx_ptr->type,
722 			     cx_ptr->trans_lat));
723 	    cx_ptr++;
724 	    sc->cpu_cx_count++;
725 	}
726     }
727     AcpiOsFree(buf.Pointer);
728 
729     return (0);
730 }
731 
732 /*
733  * Call this *after* all CPUs have been attached.
734  */
735 static void
736 acpi_cpu_startup(void *arg)
737 {
738     struct acpi_cpu_softc *sc;
739     int i;
740 
741     /* Get set of CPU devices */
742     devclass_get_devices(acpi_cpu_devclass, &cpu_devices, &cpu_ndevices);
743 
744     /*
745      * Setup any quirks that might necessary now that we have probed
746      * all the CPUs
747      */
748     acpi_cpu_quirks();
749 
750     cpu_cx_count = 0;
751     if (cpu_cx_generic) {
752 	/*
753 	 * We are using generic Cx mode, probe for available Cx states
754 	 * for all processors.
755 	 */
756 	for (i = 0; i < cpu_ndevices; i++) {
757 	    sc = device_get_softc(cpu_devices[i]);
758 	    acpi_cpu_generic_cx_probe(sc);
759 	    if (sc->cpu_cx_count > cpu_cx_count)
760 		    cpu_cx_count = sc->cpu_cx_count;
761 	}
762 
763 	/*
764 	 * Find the highest Cx state common to all CPUs
765 	 * in the system, taking quirks into account.
766 	 */
767 	for (i = 0; i < cpu_ndevices; i++) {
768 	    sc = device_get_softc(cpu_devices[i]);
769 	    if (sc->cpu_cx_count < cpu_cx_count)
770 		cpu_cx_count = sc->cpu_cx_count;
771 	}
772     } else {
773 	/*
774 	 * We are using _CST mode, remove C3 state if necessary.
775 	 * Update the largest Cx state supported in the global cpu_cx_count.
776 	 * It will be used in the global Cx sysctl handler.
777 	 * As we now know for sure that we will be using _CST mode
778 	 * install our notify handler.
779 	 */
780 	for (i = 0; i < cpu_ndevices; i++) {
781 	    sc = device_get_softc(cpu_devices[i]);
782 	    if (cpu_quirks & CPU_QUIRK_NO_C3) {
783 		sc->cpu_cx_count = sc->cpu_non_c3 + 1;
784 	    }
785 	    if (sc->cpu_cx_count > cpu_cx_count)
786 		cpu_cx_count = sc->cpu_cx_count;
787 	    AcpiInstallNotifyHandler(sc->cpu_handle, ACPI_DEVICE_NOTIFY,
788 		acpi_cpu_notify, sc);
789 	}
790     }
791 
792     /* Perform Cx final initialization. */
793     for (i = 0; i < cpu_ndevices; i++) {
794 	sc = device_get_softc(cpu_devices[i]);
795 	acpi_cpu_startup_cx(sc);
796     }
797 
798     /* Add a sysctl handler to handle global Cx lowest setting */
799     SYSCTL_ADD_PROC(&cpu_sysctl_ctx, SYSCTL_CHILDREN(cpu_sysctl_tree),
800 	OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
801 	NULL, 0, acpi_cpu_global_cx_lowest_sysctl, "A",
802 	"Global lowest Cx sleep state to use");
803 
804     /* Take over idling from cpu_idle_default(). */
805     cpu_cx_lowest = 0;
806     cpu_disable_idle = FALSE;
807     cpu_idle_hook = acpi_cpu_idle;
808 }
809 
810 static void
811 acpi_cpu_cx_list(struct acpi_cpu_softc *sc)
812 {
813     struct sbuf sb;
814     int i;
815 
816     /*
817      * Set up the list of Cx states
818      */
819     sc->cpu_non_c3 = 0;
820     sbuf_new(&sb, sc->cpu_cx_supported, sizeof(sc->cpu_cx_supported),
821 	SBUF_FIXEDLEN);
822     for (i = 0; i < sc->cpu_cx_count; i++) {
823 	sbuf_printf(&sb, "C%d/%d ", i + 1, sc->cpu_cx_states[i].trans_lat);
824 	if (sc->cpu_cx_states[i].type < ACPI_STATE_C3)
825 	    sc->cpu_non_c3 = i;
826     }
827     sbuf_trim(&sb);
828     sbuf_finish(&sb);
829 }
830 
831 static void
832 acpi_cpu_startup_cx(struct acpi_cpu_softc *sc)
833 {
834     acpi_cpu_cx_list(sc);
835 
836     SYSCTL_ADD_STRING(&sc->cpu_sysctl_ctx,
837 		      SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
838 		      OID_AUTO, "cx_supported", CTLFLAG_RD,
839 		      sc->cpu_cx_supported, 0,
840 		      "Cx/microsecond values for supported Cx states");
841     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
842 		    SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
843 		    OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
844 		    (void *)sc, 0, acpi_cpu_cx_lowest_sysctl, "A",
845 		    "lowest Cx sleep state to use");
846     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
847 		    SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
848 		    OID_AUTO, "cx_usage", CTLTYPE_STRING | CTLFLAG_RD,
849 		    (void *)sc, 0, acpi_cpu_usage_sysctl, "A",
850 		    "percent usage for each Cx state");
851 
852 #ifdef notyet
853     /* Signal platform that we can handle _CST notification. */
854     if (!cpu_cx_generic && cpu_cst_cnt != 0) {
855 	ACPI_LOCK(acpi);
856 	AcpiOsWritePort(cpu_smi_cmd, cpu_cst_cnt, 8);
857 	ACPI_UNLOCK(acpi);
858     }
859 #endif
860 }
861 
862 /*
863  * Idle the CPU in the lowest state possible.  This function is called with
864  * interrupts disabled.  Note that once it re-enables interrupts, a task
865  * switch can occur so do not access shared data (i.e. the softc) after
866  * interrupts are re-enabled.
867  */
868 static void
869 acpi_cpu_idle()
870 {
871     struct	acpi_cpu_softc *sc;
872     struct	acpi_cx *cx_next;
873     uint32_t	start_time, end_time;
874     int		bm_active, cx_next_idx, i;
875 
876     /* If disabled, return immediately. */
877     if (cpu_disable_idle) {
878 	ACPI_ENABLE_IRQS();
879 	return;
880     }
881 
882     /*
883      * Look up our CPU id to get our softc.  If it's NULL, we'll use C1
884      * since there is no ACPI processor object for this CPU.  This occurs
885      * for logical CPUs in the HTT case.
886      */
887     sc = cpu_softc[PCPU_GET(cpuid)];
888     if (sc == NULL) {
889 	acpi_cpu_c1();
890 	return;
891     }
892 
893     /* Find the lowest state that has small enough latency. */
894     cx_next_idx = 0;
895 #ifndef __ia64__
896     if (cpu_disable_deep_sleep)
897 	i = min(sc->cpu_cx_lowest, sc->cpu_non_c3);
898     else
899 #endif
900 	i = sc->cpu_cx_lowest;
901     for (; i >= 0; i--) {
902 	if (sc->cpu_cx_states[i].trans_lat * 3 <= sc->cpu_prev_sleep) {
903 	    cx_next_idx = i;
904 	    break;
905 	}
906     }
907 
908     /*
909      * Check for bus master activity.  If there was activity, clear
910      * the bit and use the lowest non-C3 state.  Note that the USB
911      * driver polling for new devices keeps this bit set all the
912      * time if USB is loaded.
913      */
914     if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
915 	AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, &bm_active);
916 	if (bm_active != 0) {
917 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, 1);
918 	    cx_next_idx = min(cx_next_idx, sc->cpu_non_c3);
919 	}
920     }
921 
922     /* Select the next state and update statistics. */
923     cx_next = &sc->cpu_cx_states[cx_next_idx];
924     sc->cpu_cx_stats[cx_next_idx]++;
925     KASSERT(cx_next->type != ACPI_STATE_C0, ("acpi_cpu_idle: C0 sleep"));
926 
927     /*
928      * Execute HLT (or equivalent) and wait for an interrupt.  We can't
929      * precisely calculate the time spent in C1 since the place we wake up
930      * is an ISR.  Assume we slept no more then half of quantum, unless
931      * we are called inside critical section, delaying context switch.
932      */
933     if (cx_next->type == ACPI_STATE_C1) {
934 	AcpiHwRead(&start_time, &AcpiGbl_FADT.XPmTimerBlock);
935 	acpi_cpu_c1();
936 	AcpiHwRead(&end_time, &AcpiGbl_FADT.XPmTimerBlock);
937         end_time = PM_USEC(acpi_TimerDelta(end_time, start_time));
938         if (curthread->td_critnest == 0)
939 		end_time = min(end_time, 500000 / hz);
940 	sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + end_time) / 4;
941 	return;
942     }
943 
944     /*
945      * For C3, disable bus master arbitration and enable bus master wake
946      * if BM control is available, otherwise flush the CPU cache.
947      */
948     if (cx_next->type == ACPI_STATE_C3) {
949 	if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
950 	    AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 1);
951 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
952 	} else
953 	    ACPI_FLUSH_CPU_CACHE();
954     }
955 
956     /*
957      * Read from P_LVLx to enter C2(+), checking time spent asleep.
958      * Use the ACPI timer for measuring sleep time.  Since we need to
959      * get the time very close to the CPU start/stop clock logic, this
960      * is the only reliable time source.
961      */
962     AcpiHwRead(&start_time, &AcpiGbl_FADT.XPmTimerBlock);
963     CPU_GET_REG(cx_next->p_lvlx, 1);
964 
965     /*
966      * Read the end time twice.  Since it may take an arbitrary time
967      * to enter the idle state, the first read may be executed before
968      * the processor has stopped.  Doing it again provides enough
969      * margin that we are certain to have a correct value.
970      */
971     AcpiHwRead(&end_time, &AcpiGbl_FADT.XPmTimerBlock);
972     AcpiHwRead(&end_time, &AcpiGbl_FADT.XPmTimerBlock);
973 
974     /* Enable bus master arbitration and disable bus master wakeup. */
975     if (cx_next->type == ACPI_STATE_C3 &&
976 	(cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
977 	AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 0);
978 	AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
979     }
980     ACPI_ENABLE_IRQS();
981 
982     /* Find the actual time asleep in microseconds. */
983     end_time = acpi_TimerDelta(end_time, start_time);
984     sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + PM_USEC(end_time)) / 4;
985 }
986 
987 /*
988  * Re-evaluate the _CST object when we are notified that it changed.
989  *
990  * XXX Re-evaluation disabled until locking is done.
991  */
992 static void
993 acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context)
994 {
995     struct acpi_cpu_softc *sc = (struct acpi_cpu_softc *)context;
996     struct acpi_cpu_softc *isc;
997     int i;
998 
999     if (notify != ACPI_NOTIFY_CX_STATES)
1000 	return;
1001 
1002     /* Update the list of Cx states. */
1003     acpi_cpu_cx_cst(sc);
1004     acpi_cpu_cx_list(sc);
1005 
1006     /* Update the new lowest useable Cx state for all CPUs. */
1007     ACPI_SERIAL_BEGIN(cpu);
1008     cpu_cx_count = 0;
1009     for (i = 0; i < cpu_ndevices; i++) {
1010 	isc = device_get_softc(cpu_devices[i]);
1011 	if (isc->cpu_cx_count > cpu_cx_count)
1012 	    cpu_cx_count = isc->cpu_cx_count;
1013     }
1014     if (sc->cpu_cx_lowest < cpu_cx_lowest)
1015 	acpi_cpu_set_cx_lowest(sc, min(cpu_cx_lowest, sc->cpu_cx_count - 1));
1016     ACPI_SERIAL_END(cpu);
1017 }
1018 
1019 static int
1020 acpi_cpu_quirks(void)
1021 {
1022     device_t acpi_dev;
1023     uint32_t val;
1024 
1025     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
1026 
1027     /*
1028      * Bus mastering arbitration control is needed to keep caches coherent
1029      * while sleeping in C3.  If it's not present but a working flush cache
1030      * instruction is present, flush the caches before entering C3 instead.
1031      * Otherwise, just disable C3 completely.
1032      */
1033     if (AcpiGbl_FADT.Pm2ControlBlock == 0 ||
1034 	AcpiGbl_FADT.Pm2ControlLength == 0) {
1035 	if ((AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD) &&
1036 	    (AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD_FLUSH) == 0) {
1037 	    cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1038 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1039 		"acpi_cpu: no BM control, using flush cache method\n"));
1040 	} else {
1041 	    cpu_quirks |= CPU_QUIRK_NO_C3;
1042 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1043 		"acpi_cpu: no BM control, C3 not available\n"));
1044 	}
1045     }
1046 
1047     /*
1048      * If we are using generic Cx mode, C3 on multiple CPUs requires using
1049      * the expensive flush cache instruction.
1050      */
1051     if (cpu_cx_generic && mp_ncpus > 1) {
1052 	cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1053 	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1054 	    "acpi_cpu: SMP, using flush cache mode for C3\n"));
1055     }
1056 
1057     /* Look for various quirks of the PIIX4 part. */
1058     acpi_dev = pci_find_device(PCI_VENDOR_INTEL, PCI_DEVICE_82371AB_3);
1059     if (acpi_dev != NULL) {
1060 	switch (pci_get_revid(acpi_dev)) {
1061 	/*
1062 	 * Disable C3 support for all PIIX4 chipsets.  Some of these parts
1063 	 * do not report the BMIDE status to the BM status register and
1064 	 * others have a livelock bug if Type-F DMA is enabled.  Linux
1065 	 * works around the BMIDE bug by reading the BM status directly
1066 	 * but we take the simpler approach of disabling C3 for these
1067 	 * parts.
1068 	 *
1069 	 * See erratum #18 ("C3 Power State/BMIDE and Type-F DMA
1070 	 * Livelock") from the January 2002 PIIX4 specification update.
1071 	 * Applies to all PIIX4 models.
1072 	 *
1073 	 * Also, make sure that all interrupts cause a "Stop Break"
1074 	 * event to exit from C2 state.
1075 	 * Also, BRLD_EN_BM (ACPI_BITREG_BUS_MASTER_RLD in ACPI-speak)
1076 	 * should be set to zero, otherwise it causes C2 to short-sleep.
1077 	 * PIIX4 doesn't properly support C3 and bus master activity
1078 	 * need not break out of C2.
1079 	 */
1080 	case PCI_REVISION_A_STEP:
1081 	case PCI_REVISION_B_STEP:
1082 	case PCI_REVISION_4E:
1083 	case PCI_REVISION_4M:
1084 	    cpu_quirks |= CPU_QUIRK_NO_C3;
1085 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1086 		"acpi_cpu: working around PIIX4 bug, disabling C3\n"));
1087 
1088 	    val = pci_read_config(acpi_dev, PIIX4_DEVACTB_REG, 4);
1089 	    if ((val & PIIX4_STOP_BREAK_MASK) != PIIX4_STOP_BREAK_MASK) {
1090 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1091 		    "acpi_cpu: PIIX4: enabling IRQs to generate Stop Break\n"));
1092 	    	val |= PIIX4_STOP_BREAK_MASK;
1093 		pci_write_config(acpi_dev, PIIX4_DEVACTB_REG, val, 4);
1094 	    }
1095 	    AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_RLD, &val);
1096 	    if (val) {
1097 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1098 		    "acpi_cpu: PIIX4: reset BRLD_EN_BM\n"));
1099 		AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1100 	    }
1101 	    break;
1102 	default:
1103 	    break;
1104 	}
1105     }
1106 
1107     return (0);
1108 }
1109 
1110 static int
1111 acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS)
1112 {
1113     struct acpi_cpu_softc *sc;
1114     struct sbuf	 sb;
1115     char	 buf[128];
1116     int		 i;
1117     uintmax_t	 fract, sum, whole;
1118 
1119     sc = (struct acpi_cpu_softc *) arg1;
1120     sum = 0;
1121     for (i = 0; i < sc->cpu_cx_count; i++)
1122 	sum += sc->cpu_cx_stats[i];
1123     sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1124     for (i = 0; i < sc->cpu_cx_count; i++) {
1125 	if (sum > 0) {
1126 	    whole = (uintmax_t)sc->cpu_cx_stats[i] * 100;
1127 	    fract = (whole % sum) * 100;
1128 	    sbuf_printf(&sb, "%u.%02u%% ", (u_int)(whole / sum),
1129 		(u_int)(fract / sum));
1130 	} else
1131 	    sbuf_printf(&sb, "0.00%% ");
1132     }
1133     sbuf_printf(&sb, "last %dus", sc->cpu_prev_sleep);
1134     sbuf_trim(&sb);
1135     sbuf_finish(&sb);
1136     sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1137     sbuf_delete(&sb);
1138 
1139     return (0);
1140 }
1141 
1142 static int
1143 acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc, int val)
1144 {
1145     int i;
1146 
1147     ACPI_SERIAL_ASSERT(cpu);
1148     sc->cpu_cx_lowest = val;
1149 
1150     /* If not disabling, cache the new lowest non-C3 state. */
1151     sc->cpu_non_c3 = 0;
1152     for (i = sc->cpu_cx_lowest; i >= 0; i--) {
1153 	if (sc->cpu_cx_states[i].type < ACPI_STATE_C3) {
1154 	    sc->cpu_non_c3 = i;
1155 	    break;
1156 	}
1157     }
1158 
1159     /* Reset the statistics counters. */
1160     bzero(sc->cpu_cx_stats, sizeof(sc->cpu_cx_stats));
1161     return (0);
1162 }
1163 
1164 static int
1165 acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1166 {
1167     struct	 acpi_cpu_softc *sc;
1168     char	 state[8];
1169     int		 val, error;
1170 
1171     sc = (struct acpi_cpu_softc *) arg1;
1172     snprintf(state, sizeof(state), "C%d", sc->cpu_cx_lowest + 1);
1173     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1174     if (error != 0 || req->newptr == NULL)
1175 	return (error);
1176     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1177 	return (EINVAL);
1178     val = (int) strtol(state + 1, NULL, 10) - 1;
1179     if (val < 0 || val > sc->cpu_cx_count - 1)
1180 	return (EINVAL);
1181 
1182     ACPI_SERIAL_BEGIN(cpu);
1183     acpi_cpu_set_cx_lowest(sc, val);
1184     ACPI_SERIAL_END(cpu);
1185 
1186     return (0);
1187 }
1188 
1189 static int
1190 acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1191 {
1192     struct	acpi_cpu_softc *sc;
1193     char	state[8];
1194     int		val, error, i;
1195 
1196     snprintf(state, sizeof(state), "C%d", cpu_cx_lowest + 1);
1197     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1198     if (error != 0 || req->newptr == NULL)
1199 	return (error);
1200     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1201 	return (EINVAL);
1202     val = (int) strtol(state + 1, NULL, 10) - 1;
1203     if (val < 0 || val > cpu_cx_count - 1)
1204 	return (EINVAL);
1205     cpu_cx_lowest = val;
1206 
1207     /* Update the new lowest useable Cx state for all CPUs. */
1208     ACPI_SERIAL_BEGIN(cpu);
1209     for (i = 0; i < cpu_ndevices; i++) {
1210 	sc = device_get_softc(cpu_devices[i]);
1211 	acpi_cpu_set_cx_lowest(sc, min(val, sc->cpu_cx_count - 1));
1212     }
1213     ACPI_SERIAL_END(cpu);
1214 
1215     return (0);
1216 }
1217