xref: /freebsd/sys/dev/acpica/acpi_cpu.c (revision aa0a1e58)
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 #if defined(__amd64__) || defined(__i386__)
48 #include <machine/clock.h>
49 #endif
50 #include <sys/rman.h>
51 
52 #include <contrib/dev/acpica/include/acpi.h>
53 #include <contrib/dev/acpica/include/accommon.h>
54 
55 #include <dev/acpica/acpivar.h>
56 
57 /*
58  * Support for ACPI Processor devices, including C[1-3] sleep states.
59  */
60 
61 /* Hooks for the ACPI CA debugging infrastructure */
62 #define _COMPONENT	ACPI_PROCESSOR
63 ACPI_MODULE_NAME("PROCESSOR")
64 
65 struct acpi_cx {
66     struct resource	*p_lvlx;	/* Register to read to enter state. */
67     uint32_t		 type;		/* C1-3 (C4 and up treated as C3). */
68     uint32_t		 trans_lat;	/* Transition latency (usec). */
69     uint32_t		 power;		/* Power consumed (mW). */
70     int			 res_type;	/* Resource type for p_lvlx. */
71 };
72 #define MAX_CX_STATES	 8
73 
74 struct acpi_cpu_softc {
75     device_t		 cpu_dev;
76     ACPI_HANDLE		 cpu_handle;
77     struct pcpu		*cpu_pcpu;
78     uint32_t		 cpu_acpi_id;	/* ACPI processor id */
79     uint32_t		 cpu_p_blk;	/* ACPI P_BLK location */
80     uint32_t		 cpu_p_blk_len;	/* P_BLK length (must be 6). */
81     struct acpi_cx	 cpu_cx_states[MAX_CX_STATES];
82     int			 cpu_cx_count;	/* Number of valid Cx states. */
83     int			 cpu_prev_sleep;/* Last idle sleep duration. */
84     int			 cpu_features;	/* Child driver supported features. */
85     /* Runtime state. */
86     int			 cpu_non_c3;	/* Index of lowest non-C3 state. */
87     u_int		 cpu_cx_stats[MAX_CX_STATES];/* Cx usage history. */
88     /* Values for sysctl. */
89     struct sysctl_ctx_list cpu_sysctl_ctx;
90     struct sysctl_oid	*cpu_sysctl_tree;
91     int			 cpu_cx_lowest;
92     char 		 cpu_cx_supported[64];
93     int			 cpu_rid;
94 };
95 
96 struct acpi_cpu_device {
97     struct resource_list	ad_rl;
98 };
99 
100 #define CPU_GET_REG(reg, width) 					\
101     (bus_space_read_ ## width(rman_get_bustag((reg)), 			\
102 		      rman_get_bushandle((reg)), 0))
103 #define CPU_SET_REG(reg, width, val)					\
104     (bus_space_write_ ## width(rman_get_bustag((reg)), 			\
105 		       rman_get_bushandle((reg)), 0, (val)))
106 
107 #define PM_USEC(x)	 ((x) >> 2)	/* ~4 clocks per usec (3.57955 Mhz) */
108 
109 #define ACPI_NOTIFY_CX_STATES	0x81	/* _CST changed. */
110 
111 #define CPU_QUIRK_NO_C3		(1<<0)	/* C3-type states are not usable. */
112 #define CPU_QUIRK_NO_BM_CTRL	(1<<2)	/* No bus mastering control. */
113 
114 #define PCI_VENDOR_INTEL	0x8086
115 #define PCI_DEVICE_82371AB_3	0x7113	/* PIIX4 chipset for quirks. */
116 #define PCI_REVISION_A_STEP	0
117 #define PCI_REVISION_B_STEP	1
118 #define PCI_REVISION_4E		2
119 #define PCI_REVISION_4M		3
120 #define PIIX4_DEVACTB_REG	0x58
121 #define PIIX4_BRLD_EN_IRQ0	(1<<0)
122 #define PIIX4_BRLD_EN_IRQ	(1<<1)
123 #define PIIX4_BRLD_EN_IRQ8	(1<<5)
124 #define PIIX4_STOP_BREAK_MASK	(PIIX4_BRLD_EN_IRQ0 | PIIX4_BRLD_EN_IRQ | PIIX4_BRLD_EN_IRQ8)
125 #define PIIX4_PCNTRL_BST_EN	(1<<10)
126 
127 /* Platform hardware resource information. */
128 static uint32_t		 cpu_smi_cmd;	/* Value to write to SMI_CMD. */
129 static uint8_t		 cpu_cst_cnt;	/* Indicate we are _CST aware. */
130 static int		 cpu_quirks;	/* Indicate any hardware bugs. */
131 
132 /* Runtime state. */
133 static int		 cpu_disable_idle; /* Disable entry to idle function */
134 static int		 cpu_cx_count;	/* Number of valid Cx states */
135 
136 /* Values for sysctl. */
137 static struct sysctl_ctx_list cpu_sysctl_ctx;
138 static struct sysctl_oid *cpu_sysctl_tree;
139 static int		 cpu_cx_generic;
140 static int		 cpu_cx_lowest;
141 
142 static device_t		*cpu_devices;
143 static int		 cpu_ndevices;
144 static struct acpi_cpu_softc **cpu_softc;
145 ACPI_SERIAL_DECL(cpu, "ACPI CPU");
146 
147 static int	acpi_cpu_probe(device_t dev);
148 static int	acpi_cpu_attach(device_t dev);
149 static int	acpi_cpu_suspend(device_t dev);
150 static int	acpi_cpu_resume(device_t dev);
151 static int	acpi_pcpu_get_id(uint32_t idx, uint32_t *acpi_id,
152 		    uint32_t *cpu_id);
153 static struct resource_list *acpi_cpu_get_rlist(device_t dev, device_t child);
154 static device_t	acpi_cpu_add_child(device_t dev, u_int order, const char *name,
155 		    int unit);
156 static int	acpi_cpu_read_ivar(device_t dev, device_t child, int index,
157 		    uintptr_t *result);
158 static int	acpi_cpu_shutdown(device_t dev);
159 static void	acpi_cpu_cx_probe(struct acpi_cpu_softc *sc);
160 static void	acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc);
161 static int	acpi_cpu_cx_cst(struct acpi_cpu_softc *sc);
162 static void	acpi_cpu_startup(void *arg);
163 static void	acpi_cpu_startup_cx(struct acpi_cpu_softc *sc);
164 static void	acpi_cpu_cx_list(struct acpi_cpu_softc *sc);
165 static void	acpi_cpu_idle(void);
166 static void	acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context);
167 static int	acpi_cpu_quirks(void);
168 static int	acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS);
169 static int	acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc, int val);
170 static int	acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
171 static int	acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
172 
173 static device_method_t acpi_cpu_methods[] = {
174     /* Device interface */
175     DEVMETHOD(device_probe,	acpi_cpu_probe),
176     DEVMETHOD(device_attach,	acpi_cpu_attach),
177     DEVMETHOD(device_detach,	bus_generic_detach),
178     DEVMETHOD(device_shutdown,	acpi_cpu_shutdown),
179     DEVMETHOD(device_suspend,	acpi_cpu_suspend),
180     DEVMETHOD(device_resume,	acpi_cpu_resume),
181 
182     /* Bus interface */
183     DEVMETHOD(bus_add_child,	acpi_cpu_add_child),
184     DEVMETHOD(bus_read_ivar,	acpi_cpu_read_ivar),
185     DEVMETHOD(bus_get_resource_list, acpi_cpu_get_rlist),
186     DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
187     DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
188     DEVMETHOD(bus_alloc_resource, bus_generic_rl_alloc_resource),
189     DEVMETHOD(bus_release_resource, bus_generic_rl_release_resource),
190     DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
191     DEVMETHOD(bus_activate_resource, bus_generic_activate_resource),
192     DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource),
193     DEVMETHOD(bus_setup_intr,	bus_generic_setup_intr),
194     DEVMETHOD(bus_teardown_intr, bus_generic_teardown_intr),
195 
196     {0, 0}
197 };
198 
199 static driver_t acpi_cpu_driver = {
200     "cpu",
201     acpi_cpu_methods,
202     sizeof(struct acpi_cpu_softc),
203 };
204 
205 static devclass_t acpi_cpu_devclass;
206 DRIVER_MODULE(cpu, acpi, acpi_cpu_driver, acpi_cpu_devclass, 0, 0);
207 MODULE_DEPEND(cpu, acpi, 1, 1, 1);
208 
209 static int
210 acpi_cpu_probe(device_t dev)
211 {
212     int			   acpi_id, cpu_id;
213     ACPI_BUFFER		   buf;
214     ACPI_HANDLE		   handle;
215     ACPI_OBJECT		   *obj;
216     ACPI_STATUS		   status;
217 
218     if (acpi_disabled("cpu") || acpi_get_type(dev) != ACPI_TYPE_PROCESSOR)
219 	return (ENXIO);
220 
221     handle = acpi_get_handle(dev);
222     if (cpu_softc == NULL)
223 	cpu_softc = malloc(sizeof(struct acpi_cpu_softc *) *
224 	    (mp_maxid + 1), M_TEMP /* XXX */, M_WAITOK | M_ZERO);
225 
226     /* Get our Processor object. */
227     buf.Pointer = NULL;
228     buf.Length = ACPI_ALLOCATE_BUFFER;
229     status = AcpiEvaluateObject(handle, NULL, NULL, &buf);
230     if (ACPI_FAILURE(status)) {
231 	device_printf(dev, "probe failed to get Processor obj - %s\n",
232 		      AcpiFormatException(status));
233 	return (ENXIO);
234     }
235     obj = (ACPI_OBJECT *)buf.Pointer;
236     if (obj->Type != ACPI_TYPE_PROCESSOR) {
237 	device_printf(dev, "Processor object has bad type %d\n", obj->Type);
238 	AcpiOsFree(obj);
239 	return (ENXIO);
240     }
241 
242     /*
243      * Find the processor associated with our unit.  We could use the
244      * ProcId as a key, however, some boxes do not have the same values
245      * in their Processor object as the ProcId values in the MADT.
246      */
247     acpi_id = obj->Processor.ProcId;
248     AcpiOsFree(obj);
249     if (acpi_pcpu_get_id(device_get_unit(dev), &acpi_id, &cpu_id) != 0)
250 	return (ENXIO);
251 
252     /*
253      * Check if we already probed this processor.  We scan the bus twice
254      * so it's possible we've already seen this one.
255      */
256     if (cpu_softc[cpu_id] != NULL)
257 	return (ENXIO);
258 
259     /* Mark this processor as in-use and save our derived id for attach. */
260     cpu_softc[cpu_id] = (void *)1;
261     acpi_set_private(dev, (void*)(intptr_t)cpu_id);
262     device_set_desc(dev, "ACPI CPU");
263 
264     return (0);
265 }
266 
267 static int
268 acpi_cpu_attach(device_t dev)
269 {
270     ACPI_BUFFER		   buf;
271     ACPI_OBJECT		   arg[4], *obj;
272     ACPI_OBJECT_LIST	   arglist;
273     struct pcpu		   *pcpu_data;
274     struct acpi_cpu_softc *sc;
275     struct acpi_softc	  *acpi_sc;
276     ACPI_STATUS		   status;
277     u_int		   features;
278     int			   cpu_id, drv_count, i;
279     driver_t 		  **drivers;
280     uint32_t		   cap_set[3];
281 
282     /* UUID needed by _OSC evaluation */
283     static uint8_t cpu_oscuuid[16] = { 0x16, 0xA6, 0x77, 0x40, 0x0C, 0x29,
284 				       0xBE, 0x47, 0x9E, 0xBD, 0xD8, 0x70,
285 				       0x58, 0x71, 0x39, 0x53 };
286 
287     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
288 
289     sc = device_get_softc(dev);
290     sc->cpu_dev = dev;
291     sc->cpu_handle = acpi_get_handle(dev);
292     cpu_id = (int)(intptr_t)acpi_get_private(dev);
293     cpu_softc[cpu_id] = sc;
294     pcpu_data = pcpu_find(cpu_id);
295     pcpu_data->pc_device = dev;
296     sc->cpu_pcpu = pcpu_data;
297     cpu_smi_cmd = AcpiGbl_FADT.SmiCommand;
298     cpu_cst_cnt = AcpiGbl_FADT.CstControl;
299 
300     buf.Pointer = NULL;
301     buf.Length = ACPI_ALLOCATE_BUFFER;
302     status = AcpiEvaluateObject(sc->cpu_handle, NULL, NULL, &buf);
303     if (ACPI_FAILURE(status)) {
304 	device_printf(dev, "attach failed to get Processor obj - %s\n",
305 		      AcpiFormatException(status));
306 	return (ENXIO);
307     }
308     obj = (ACPI_OBJECT *)buf.Pointer;
309     sc->cpu_p_blk = obj->Processor.PblkAddress;
310     sc->cpu_p_blk_len = obj->Processor.PblkLength;
311     sc->cpu_acpi_id = obj->Processor.ProcId;
312     AcpiOsFree(obj);
313     ACPI_DEBUG_PRINT((ACPI_DB_INFO, "acpi_cpu%d: P_BLK at %#x/%d\n",
314 		     device_get_unit(dev), sc->cpu_p_blk, sc->cpu_p_blk_len));
315 
316     /*
317      * If this is the first cpu we attach, create and initialize the generic
318      * resources that will be used by all acpi cpu devices.
319      */
320     if (device_get_unit(dev) == 0) {
321 	/* Assume we won't be using generic Cx mode by default */
322 	cpu_cx_generic = FALSE;
323 
324 	/* Install hw.acpi.cpu sysctl tree */
325 	acpi_sc = acpi_device_get_parent_softc(dev);
326 	sysctl_ctx_init(&cpu_sysctl_ctx);
327 	cpu_sysctl_tree = SYSCTL_ADD_NODE(&cpu_sysctl_ctx,
328 	    SYSCTL_CHILDREN(acpi_sc->acpi_sysctl_tree), OID_AUTO, "cpu",
329 	    CTLFLAG_RD, 0, "node for CPU children");
330 
331 	/* Queue post cpu-probing task handler */
332 	AcpiOsExecute(OSL_NOTIFY_HANDLER, acpi_cpu_startup, NULL);
333     }
334 
335     /*
336      * Before calling any CPU methods, collect child driver feature hints
337      * and notify ACPI of them.  We support unified SMP power control
338      * so advertise this ourselves.  Note this is not the same as independent
339      * SMP control where each CPU can have different settings.
340      */
341     sc->cpu_features = ACPI_CAP_SMP_SAME | ACPI_CAP_SMP_SAME_C3;
342     if (devclass_get_drivers(acpi_cpu_devclass, &drivers, &drv_count) == 0) {
343 	for (i = 0; i < drv_count; i++) {
344 	    if (ACPI_GET_FEATURES(drivers[i], &features) == 0)
345 		sc->cpu_features |= features;
346 	}
347 	free(drivers, M_TEMP);
348     }
349 
350     /*
351      * CPU capabilities are specified in
352      * Intel Processor Vendor-Specific ACPI Interface Specification.
353      */
354     if (sc->cpu_features) {
355 	arglist.Pointer = arg;
356 	arglist.Count = 4;
357 	arg[0].Type = ACPI_TYPE_BUFFER;
358 	arg[0].Buffer.Length = sizeof(cpu_oscuuid);
359 	arg[0].Buffer.Pointer = cpu_oscuuid;	/* UUID */
360 	arg[1].Type = ACPI_TYPE_INTEGER;
361 	arg[1].Integer.Value = 1;		/* revision */
362 	arg[2].Type = ACPI_TYPE_INTEGER;
363 	arg[2].Integer.Value = 1;		/* count */
364 	arg[3].Type = ACPI_TYPE_BUFFER;
365 	arg[3].Buffer.Length = sizeof(cap_set);	/* Capabilities buffer */
366 	arg[3].Buffer.Pointer = (uint8_t *)cap_set;
367 	cap_set[0] = 0;				/* status */
368 	cap_set[1] = sc->cpu_features;
369 	status = AcpiEvaluateObject(sc->cpu_handle, "_OSC", &arglist, NULL);
370 	if (ACPI_SUCCESS(status)) {
371 	    if (cap_set[0] != 0)
372 		device_printf(dev, "_OSC returned status %#x\n", cap_set[0]);
373 	}
374 	else {
375 	    arglist.Pointer = arg;
376 	    arglist.Count = 1;
377 	    arg[0].Type = ACPI_TYPE_BUFFER;
378 	    arg[0].Buffer.Length = sizeof(cap_set);
379 	    arg[0].Buffer.Pointer = (uint8_t *)cap_set;
380 	    cap_set[0] = 1; /* revision */
381 	    cap_set[1] = 1; /* number of capabilities integers */
382 	    cap_set[2] = sc->cpu_features;
383 	    AcpiEvaluateObject(sc->cpu_handle, "_PDC", &arglist, NULL);
384 	}
385     }
386 
387     /* Probe for Cx state support. */
388     acpi_cpu_cx_probe(sc);
389 
390     return (0);
391 }
392 
393 static void
394 acpi_cpu_postattach(void *unused __unused)
395 {
396     device_t *devices;
397     int err;
398     int i, n;
399 
400     err = devclass_get_devices(acpi_cpu_devclass, &devices, &n);
401     if (err != 0) {
402 	printf("devclass_get_devices(acpi_cpu_devclass) failed\n");
403 	return;
404     }
405     for (i = 0; i < n; i++)
406 	bus_generic_probe(devices[i]);
407     for (i = 0; i < n; i++)
408 	bus_generic_attach(devices[i]);
409     free(devices, M_TEMP);
410 }
411 
412 SYSINIT(acpi_cpu, SI_SUB_CONFIGURE, SI_ORDER_MIDDLE,
413     acpi_cpu_postattach, NULL);
414 
415 /*
416  * Disable any entry to the idle function during suspend and re-enable it
417  * during resume.
418  */
419 static int
420 acpi_cpu_suspend(device_t dev)
421 {
422     int error;
423 
424     error = bus_generic_suspend(dev);
425     if (error)
426 	return (error);
427     cpu_disable_idle = TRUE;
428     return (0);
429 }
430 
431 static int
432 acpi_cpu_resume(device_t dev)
433 {
434 
435     cpu_disable_idle = FALSE;
436     return (bus_generic_resume(dev));
437 }
438 
439 /*
440  * Find the nth present CPU and return its pc_cpuid as well as set the
441  * pc_acpi_id from the most reliable source.
442  */
443 static int
444 acpi_pcpu_get_id(uint32_t idx, uint32_t *acpi_id, uint32_t *cpu_id)
445 {
446     struct pcpu	*pcpu_data;
447     uint32_t	 i;
448 
449     KASSERT(acpi_id != NULL, ("Null acpi_id"));
450     KASSERT(cpu_id != NULL, ("Null cpu_id"));
451     CPU_FOREACH(i) {
452 	pcpu_data = pcpu_find(i);
453 	KASSERT(pcpu_data != NULL, ("no pcpu data for %d", i));
454 	if (idx-- == 0) {
455 	    /*
456 	     * If pc_acpi_id was not initialized (e.g., a non-APIC UP box)
457 	     * override it with the value from the ASL.  Otherwise, if the
458 	     * two don't match, prefer the MADT-derived value.  Finally,
459 	     * return the pc_cpuid to reference this processor.
460 	     */
461 	    if (pcpu_data->pc_acpi_id == 0xffffffff)
462 		pcpu_data->pc_acpi_id = *acpi_id;
463 	    else if (pcpu_data->pc_acpi_id != *acpi_id)
464 		*acpi_id = pcpu_data->pc_acpi_id;
465 	    *cpu_id = pcpu_data->pc_cpuid;
466 	    return (0);
467 	}
468     }
469 
470     return (ESRCH);
471 }
472 
473 static struct resource_list *
474 acpi_cpu_get_rlist(device_t dev, device_t child)
475 {
476     struct acpi_cpu_device *ad;
477 
478     ad = device_get_ivars(child);
479     if (ad == NULL)
480 	return (NULL);
481     return (&ad->ad_rl);
482 }
483 
484 static device_t
485 acpi_cpu_add_child(device_t dev, u_int order, const char *name, int unit)
486 {
487     struct acpi_cpu_device *ad;
488     device_t child;
489 
490     if ((ad = malloc(sizeof(*ad), M_TEMP, M_NOWAIT | M_ZERO)) == NULL)
491 	return (NULL);
492 
493     resource_list_init(&ad->ad_rl);
494 
495     child = device_add_child_ordered(dev, order, name, unit);
496     if (child != NULL)
497 	device_set_ivars(child, ad);
498     else
499 	free(ad, M_TEMP);
500     return (child);
501 }
502 
503 static int
504 acpi_cpu_read_ivar(device_t dev, device_t child, int index, uintptr_t *result)
505 {
506     struct acpi_cpu_softc *sc;
507 
508     sc = device_get_softc(dev);
509     switch (index) {
510     case ACPI_IVAR_HANDLE:
511 	*result = (uintptr_t)sc->cpu_handle;
512 	break;
513     case CPU_IVAR_PCPU:
514 	*result = (uintptr_t)sc->cpu_pcpu;
515 	break;
516 #if defined(__amd64__) || defined(__i386__)
517     case CPU_IVAR_NOMINAL_MHZ:
518 	if (tsc_is_invariant) {
519 	    *result = (uintptr_t)(tsc_freq / 1000000);
520 	    break;
521 	}
522 	/* FALLTHROUGH */
523 #endif
524     default:
525 	return (ENOENT);
526     }
527     return (0);
528 }
529 
530 static int
531 acpi_cpu_shutdown(device_t dev)
532 {
533     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
534 
535     /* Allow children to shutdown first. */
536     bus_generic_shutdown(dev);
537 
538     /*
539      * Disable any entry to the idle function.  There is a small race where
540      * an idle thread have passed this check but not gone to sleep.  This
541      * is ok since device_shutdown() does not free the softc, otherwise
542      * we'd have to be sure all threads were evicted before returning.
543      */
544     cpu_disable_idle = TRUE;
545 
546     return_VALUE (0);
547 }
548 
549 static void
550 acpi_cpu_cx_probe(struct acpi_cpu_softc *sc)
551 {
552     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
553 
554     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
555     sc->cpu_prev_sleep = 1000000;
556     sc->cpu_cx_lowest = 0;
557 
558     /*
559      * Check for the ACPI 2.0 _CST sleep states object. If we can't find
560      * any, we'll revert to generic FADT/P_BLK Cx control method which will
561      * be handled by acpi_cpu_startup. We need to defer to after having
562      * probed all the cpus in the system before probing for generic Cx
563      * states as we may already have found cpus with valid _CST packages
564      */
565     if (!cpu_cx_generic && acpi_cpu_cx_cst(sc) != 0) {
566 	/*
567 	 * We were unable to find a _CST package for this cpu or there
568 	 * was an error parsing it. Switch back to generic mode.
569 	 */
570 	cpu_cx_generic = TRUE;
571 	if (bootverbose)
572 	    device_printf(sc->cpu_dev, "switching to generic Cx mode\n");
573     }
574 
575     /*
576      * TODO: _CSD Package should be checked here.
577      */
578 }
579 
580 static void
581 acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc)
582 {
583     ACPI_GENERIC_ADDRESS	 gas;
584     struct acpi_cx		*cx_ptr;
585 
586     sc->cpu_cx_count = 0;
587     cx_ptr = sc->cpu_cx_states;
588 
589     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
590     sc->cpu_prev_sleep = 1000000;
591 
592     /* C1 has been required since just after ACPI 1.0 */
593     cx_ptr->type = ACPI_STATE_C1;
594     cx_ptr->trans_lat = 0;
595     cx_ptr++;
596     sc->cpu_cx_count++;
597 
598     /*
599      * The spec says P_BLK must be 6 bytes long.  However, some systems
600      * use it to indicate a fractional set of features present so we
601      * take 5 as C2.  Some may also have a value of 7 to indicate
602      * another C3 but most use _CST for this (as required) and having
603      * "only" C1-C3 is not a hardship.
604      */
605     if (sc->cpu_p_blk_len < 5)
606 	return;
607 
608     /* Validate and allocate resources for C2 (P_LVL2). */
609     gas.SpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
610     gas.BitWidth = 8;
611     if (AcpiGbl_FADT.C2Latency <= 100) {
612 	gas.Address = sc->cpu_p_blk + 4;
613 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &sc->cpu_rid,
614 	    &gas, &cx_ptr->p_lvlx, RF_SHAREABLE);
615 	if (cx_ptr->p_lvlx != NULL) {
616 	    sc->cpu_rid++;
617 	    cx_ptr->type = ACPI_STATE_C2;
618 	    cx_ptr->trans_lat = AcpiGbl_FADT.C2Latency;
619 	    cx_ptr++;
620 	    sc->cpu_cx_count++;
621 	}
622     }
623     if (sc->cpu_p_blk_len < 6)
624 	return;
625 
626     /* Validate and allocate resources for C3 (P_LVL3). */
627     if (AcpiGbl_FADT.C3Latency <= 1000 && !(cpu_quirks & CPU_QUIRK_NO_C3)) {
628 	gas.Address = sc->cpu_p_blk + 5;
629 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &sc->cpu_rid, &gas,
630 	    &cx_ptr->p_lvlx, RF_SHAREABLE);
631 	if (cx_ptr->p_lvlx != NULL) {
632 	    sc->cpu_rid++;
633 	    cx_ptr->type = ACPI_STATE_C3;
634 	    cx_ptr->trans_lat = AcpiGbl_FADT.C3Latency;
635 	    cx_ptr++;
636 	    sc->cpu_cx_count++;
637 	}
638     }
639 }
640 
641 /*
642  * Parse a _CST package and set up its Cx states.  Since the _CST object
643  * can change dynamically, our notify handler may call this function
644  * to clean up and probe the new _CST package.
645  */
646 static int
647 acpi_cpu_cx_cst(struct acpi_cpu_softc *sc)
648 {
649     struct	 acpi_cx *cx_ptr;
650     ACPI_STATUS	 status;
651     ACPI_BUFFER	 buf;
652     ACPI_OBJECT	*top;
653     ACPI_OBJECT	*pkg;
654     uint32_t	 count;
655     int		 i;
656 
657     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
658 
659     buf.Pointer = NULL;
660     buf.Length = ACPI_ALLOCATE_BUFFER;
661     status = AcpiEvaluateObject(sc->cpu_handle, "_CST", NULL, &buf);
662     if (ACPI_FAILURE(status))
663 	return (ENXIO);
664 
665     /* _CST is a package with a count and at least one Cx package. */
666     top = (ACPI_OBJECT *)buf.Pointer;
667     if (!ACPI_PKG_VALID(top, 2) || acpi_PkgInt32(top, 0, &count) != 0) {
668 	device_printf(sc->cpu_dev, "invalid _CST package\n");
669 	AcpiOsFree(buf.Pointer);
670 	return (ENXIO);
671     }
672     if (count != top->Package.Count - 1) {
673 	device_printf(sc->cpu_dev, "invalid _CST state count (%d != %d)\n",
674 	       count, top->Package.Count - 1);
675 	count = top->Package.Count - 1;
676     }
677     if (count > MAX_CX_STATES) {
678 	device_printf(sc->cpu_dev, "_CST has too many states (%d)\n", count);
679 	count = MAX_CX_STATES;
680     }
681 
682     sc->cpu_non_c3 = 0;
683     sc->cpu_cx_count = 0;
684     cx_ptr = sc->cpu_cx_states;
685 
686     /*
687      * C1 has been required since just after ACPI 1.0.
688      * Reserve the first slot for it.
689      */
690     cx_ptr->type = ACPI_STATE_C0;
691     cx_ptr++;
692     sc->cpu_cx_count++;
693 
694     /* Set up all valid states. */
695     for (i = 0; i < count; i++) {
696 	pkg = &top->Package.Elements[i + 1];
697 	if (!ACPI_PKG_VALID(pkg, 4) ||
698 	    acpi_PkgInt32(pkg, 1, &cx_ptr->type) != 0 ||
699 	    acpi_PkgInt32(pkg, 2, &cx_ptr->trans_lat) != 0 ||
700 	    acpi_PkgInt32(pkg, 3, &cx_ptr->power) != 0) {
701 
702 	    device_printf(sc->cpu_dev, "skipping invalid Cx state package\n");
703 	    continue;
704 	}
705 
706 	/* Validate the state to see if we should use it. */
707 	switch (cx_ptr->type) {
708 	case ACPI_STATE_C1:
709 	    if (sc->cpu_cx_states[0].type == ACPI_STATE_C0) {
710 		/* This is the first C1 state.  Use the reserved slot. */
711 		sc->cpu_cx_states[0] = *cx_ptr;
712 	    } else {
713 		sc->cpu_non_c3 = i;
714 		cx_ptr++;
715 		sc->cpu_cx_count++;
716 	    }
717 	    continue;
718 	case ACPI_STATE_C2:
719 	    sc->cpu_non_c3 = i;
720 	    break;
721 	case ACPI_STATE_C3:
722 	default:
723 	    if ((cpu_quirks & CPU_QUIRK_NO_C3) != 0) {
724 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
725 				 "acpi_cpu%d: C3[%d] not available.\n",
726 				 device_get_unit(sc->cpu_dev), i));
727 		continue;
728 	    }
729 	    break;
730 	}
731 
732 #ifdef notyet
733 	/* Free up any previous register. */
734 	if (cx_ptr->p_lvlx != NULL) {
735 	    bus_release_resource(sc->cpu_dev, 0, 0, cx_ptr->p_lvlx);
736 	    cx_ptr->p_lvlx = NULL;
737 	}
738 #endif
739 
740 	/* Allocate the control register for C2 or C3. */
741 	acpi_PkgGas(sc->cpu_dev, pkg, 0, &cx_ptr->res_type, &sc->cpu_rid,
742 	    &cx_ptr->p_lvlx, RF_SHAREABLE);
743 	if (cx_ptr->p_lvlx) {
744 	    sc->cpu_rid++;
745 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
746 			     "acpi_cpu%d: Got C%d - %d latency\n",
747 			     device_get_unit(sc->cpu_dev), cx_ptr->type,
748 			     cx_ptr->trans_lat));
749 	    cx_ptr++;
750 	    sc->cpu_cx_count++;
751 	}
752     }
753     AcpiOsFree(buf.Pointer);
754 
755     /* If C1 state was not found, we need one now. */
756     cx_ptr = sc->cpu_cx_states;
757     if (cx_ptr->type == ACPI_STATE_C0) {
758 	cx_ptr->type = ACPI_STATE_C1;
759 	cx_ptr->trans_lat = 0;
760     }
761 
762     return (0);
763 }
764 
765 /*
766  * Call this *after* all CPUs have been attached.
767  */
768 static void
769 acpi_cpu_startup(void *arg)
770 {
771     struct acpi_cpu_softc *sc;
772     int i;
773 
774     /* Get set of CPU devices */
775     devclass_get_devices(acpi_cpu_devclass, &cpu_devices, &cpu_ndevices);
776 
777     /*
778      * Setup any quirks that might necessary now that we have probed
779      * all the CPUs
780      */
781     acpi_cpu_quirks();
782 
783     cpu_cx_count = 0;
784     if (cpu_cx_generic) {
785 	/*
786 	 * We are using generic Cx mode, probe for available Cx states
787 	 * for all processors.
788 	 */
789 	for (i = 0; i < cpu_ndevices; i++) {
790 	    sc = device_get_softc(cpu_devices[i]);
791 	    acpi_cpu_generic_cx_probe(sc);
792 	    if (sc->cpu_cx_count > cpu_cx_count)
793 		    cpu_cx_count = sc->cpu_cx_count;
794 	}
795 
796 	/*
797 	 * Find the highest Cx state common to all CPUs
798 	 * in the system, taking quirks into account.
799 	 */
800 	for (i = 0; i < cpu_ndevices; i++) {
801 	    sc = device_get_softc(cpu_devices[i]);
802 	    if (sc->cpu_cx_count < cpu_cx_count)
803 		cpu_cx_count = sc->cpu_cx_count;
804 	}
805     } else {
806 	/*
807 	 * We are using _CST mode, remove C3 state if necessary.
808 	 * Update the largest Cx state supported in the global cpu_cx_count.
809 	 * It will be used in the global Cx sysctl handler.
810 	 * As we now know for sure that we will be using _CST mode
811 	 * install our notify handler.
812 	 */
813 	for (i = 0; i < cpu_ndevices; i++) {
814 	    sc = device_get_softc(cpu_devices[i]);
815 	    if (cpu_quirks & CPU_QUIRK_NO_C3) {
816 		sc->cpu_cx_count = sc->cpu_non_c3 + 1;
817 	    }
818 	    if (sc->cpu_cx_count > cpu_cx_count)
819 		cpu_cx_count = sc->cpu_cx_count;
820 	    AcpiInstallNotifyHandler(sc->cpu_handle, ACPI_DEVICE_NOTIFY,
821 		acpi_cpu_notify, sc);
822 	}
823     }
824 
825     /* Perform Cx final initialization. */
826     for (i = 0; i < cpu_ndevices; i++) {
827 	sc = device_get_softc(cpu_devices[i]);
828 	acpi_cpu_startup_cx(sc);
829     }
830 
831     /* Add a sysctl handler to handle global Cx lowest setting */
832     SYSCTL_ADD_PROC(&cpu_sysctl_ctx, SYSCTL_CHILDREN(cpu_sysctl_tree),
833 	OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
834 	NULL, 0, acpi_cpu_global_cx_lowest_sysctl, "A",
835 	"Global lowest Cx sleep state to use");
836 
837     /* Take over idling from cpu_idle_default(). */
838     cpu_cx_lowest = 0;
839     cpu_disable_idle = FALSE;
840     cpu_idle_hook = acpi_cpu_idle;
841 }
842 
843 static void
844 acpi_cpu_cx_list(struct acpi_cpu_softc *sc)
845 {
846     struct sbuf sb;
847     int i;
848 
849     /*
850      * Set up the list of Cx states
851      */
852     sc->cpu_non_c3 = 0;
853     sbuf_new(&sb, sc->cpu_cx_supported, sizeof(sc->cpu_cx_supported),
854 	SBUF_FIXEDLEN);
855     for (i = 0; i < sc->cpu_cx_count; i++) {
856 	sbuf_printf(&sb, "C%d/%d ", i + 1, sc->cpu_cx_states[i].trans_lat);
857 	if (sc->cpu_cx_states[i].type < ACPI_STATE_C3)
858 	    sc->cpu_non_c3 = i;
859     }
860     sbuf_trim(&sb);
861     sbuf_finish(&sb);
862 }
863 
864 static void
865 acpi_cpu_startup_cx(struct acpi_cpu_softc *sc)
866 {
867     acpi_cpu_cx_list(sc);
868 
869     SYSCTL_ADD_STRING(&sc->cpu_sysctl_ctx,
870 		      SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
871 		      OID_AUTO, "cx_supported", CTLFLAG_RD,
872 		      sc->cpu_cx_supported, 0,
873 		      "Cx/microsecond values for supported Cx states");
874     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
875 		    SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
876 		    OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
877 		    (void *)sc, 0, acpi_cpu_cx_lowest_sysctl, "A",
878 		    "lowest Cx sleep state to use");
879     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
880 		    SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
881 		    OID_AUTO, "cx_usage", CTLTYPE_STRING | CTLFLAG_RD,
882 		    (void *)sc, 0, acpi_cpu_usage_sysctl, "A",
883 		    "percent usage for each Cx state");
884 
885 #ifdef notyet
886     /* Signal platform that we can handle _CST notification. */
887     if (!cpu_cx_generic && cpu_cst_cnt != 0) {
888 	ACPI_LOCK(acpi);
889 	AcpiOsWritePort(cpu_smi_cmd, cpu_cst_cnt, 8);
890 	ACPI_UNLOCK(acpi);
891     }
892 #endif
893 }
894 
895 /*
896  * Idle the CPU in the lowest state possible.  This function is called with
897  * interrupts disabled.  Note that once it re-enables interrupts, a task
898  * switch can occur so do not access shared data (i.e. the softc) after
899  * interrupts are re-enabled.
900  */
901 static void
902 acpi_cpu_idle()
903 {
904     struct	acpi_cpu_softc *sc;
905     struct	acpi_cx *cx_next;
906     uint32_t	start_time, end_time;
907     int		bm_active, cx_next_idx, i;
908 
909     /* If disabled, return immediately. */
910     if (cpu_disable_idle) {
911 	ACPI_ENABLE_IRQS();
912 	return;
913     }
914 
915     /*
916      * Look up our CPU id to get our softc.  If it's NULL, we'll use C1
917      * since there is no ACPI processor object for this CPU.  This occurs
918      * for logical CPUs in the HTT case.
919      */
920     sc = cpu_softc[PCPU_GET(cpuid)];
921     if (sc == NULL) {
922 	acpi_cpu_c1();
923 	return;
924     }
925 
926     /* Find the lowest state that has small enough latency. */
927     cx_next_idx = 0;
928 #ifndef __ia64__
929     if (cpu_disable_deep_sleep)
930 	i = min(sc->cpu_cx_lowest, sc->cpu_non_c3);
931     else
932 #endif
933 	i = sc->cpu_cx_lowest;
934     for (; i >= 0; i--) {
935 	if (sc->cpu_cx_states[i].trans_lat * 3 <= sc->cpu_prev_sleep) {
936 	    cx_next_idx = i;
937 	    break;
938 	}
939     }
940 
941     /*
942      * Check for bus master activity.  If there was activity, clear
943      * the bit and use the lowest non-C3 state.  Note that the USB
944      * driver polling for new devices keeps this bit set all the
945      * time if USB is loaded.
946      */
947     if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
948 	AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, &bm_active);
949 	if (bm_active != 0) {
950 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, 1);
951 	    cx_next_idx = min(cx_next_idx, sc->cpu_non_c3);
952 	}
953     }
954 
955     /* Select the next state and update statistics. */
956     cx_next = &sc->cpu_cx_states[cx_next_idx];
957     sc->cpu_cx_stats[cx_next_idx]++;
958     KASSERT(cx_next->type != ACPI_STATE_C0, ("acpi_cpu_idle: C0 sleep"));
959 
960     /*
961      * Execute HLT (or equivalent) and wait for an interrupt.  We can't
962      * precisely calculate the time spent in C1 since the place we wake up
963      * is an ISR.  Assume we slept no more then half of quantum, unless
964      * we are called inside critical section, delaying context switch.
965      */
966     if (cx_next->type == ACPI_STATE_C1) {
967 	AcpiHwRead(&start_time, &AcpiGbl_FADT.XPmTimerBlock);
968 	acpi_cpu_c1();
969 	AcpiHwRead(&end_time, &AcpiGbl_FADT.XPmTimerBlock);
970         end_time = PM_USEC(acpi_TimerDelta(end_time, start_time));
971         if (curthread->td_critnest == 0)
972 		end_time = min(end_time, 500000 / hz);
973 	sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + end_time) / 4;
974 	return;
975     }
976 
977     /*
978      * For C3, disable bus master arbitration and enable bus master wake
979      * if BM control is available, otherwise flush the CPU cache.
980      */
981     if (cx_next->type == ACPI_STATE_C3) {
982 	if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
983 	    AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 1);
984 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
985 	} else
986 	    ACPI_FLUSH_CPU_CACHE();
987     }
988 
989     /*
990      * Read from P_LVLx to enter C2(+), checking time spent asleep.
991      * Use the ACPI timer for measuring sleep time.  Since we need to
992      * get the time very close to the CPU start/stop clock logic, this
993      * is the only reliable time source.
994      */
995     AcpiHwRead(&start_time, &AcpiGbl_FADT.XPmTimerBlock);
996     CPU_GET_REG(cx_next->p_lvlx, 1);
997 
998     /*
999      * Read the end time twice.  Since it may take an arbitrary time
1000      * to enter the idle state, the first read may be executed before
1001      * the processor has stopped.  Doing it again provides enough
1002      * margin that we are certain to have a correct value.
1003      */
1004     AcpiHwRead(&end_time, &AcpiGbl_FADT.XPmTimerBlock);
1005     AcpiHwRead(&end_time, &AcpiGbl_FADT.XPmTimerBlock);
1006 
1007     /* Enable bus master arbitration and disable bus master wakeup. */
1008     if (cx_next->type == ACPI_STATE_C3 &&
1009 	(cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
1010 	AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 0);
1011 	AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1012     }
1013     ACPI_ENABLE_IRQS();
1014 
1015     /* Find the actual time asleep in microseconds. */
1016     end_time = acpi_TimerDelta(end_time, start_time);
1017     sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + PM_USEC(end_time)) / 4;
1018 }
1019 
1020 /*
1021  * Re-evaluate the _CST object when we are notified that it changed.
1022  *
1023  * XXX Re-evaluation disabled until locking is done.
1024  */
1025 static void
1026 acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context)
1027 {
1028     struct acpi_cpu_softc *sc = (struct acpi_cpu_softc *)context;
1029     struct acpi_cpu_softc *isc;
1030     int i;
1031 
1032     if (notify != ACPI_NOTIFY_CX_STATES)
1033 	return;
1034 
1035     /* Update the list of Cx states. */
1036     acpi_cpu_cx_cst(sc);
1037     acpi_cpu_cx_list(sc);
1038 
1039     /* Update the new lowest useable Cx state for all CPUs. */
1040     ACPI_SERIAL_BEGIN(cpu);
1041     cpu_cx_count = 0;
1042     for (i = 0; i < cpu_ndevices; i++) {
1043 	isc = device_get_softc(cpu_devices[i]);
1044 	if (isc->cpu_cx_count > cpu_cx_count)
1045 	    cpu_cx_count = isc->cpu_cx_count;
1046     }
1047     if (sc->cpu_cx_lowest < cpu_cx_lowest)
1048 	acpi_cpu_set_cx_lowest(sc, min(cpu_cx_lowest, sc->cpu_cx_count - 1));
1049     ACPI_SERIAL_END(cpu);
1050 }
1051 
1052 static int
1053 acpi_cpu_quirks(void)
1054 {
1055     device_t acpi_dev;
1056     uint32_t val;
1057 
1058     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
1059 
1060     /*
1061      * Bus mastering arbitration control is needed to keep caches coherent
1062      * while sleeping in C3.  If it's not present but a working flush cache
1063      * instruction is present, flush the caches before entering C3 instead.
1064      * Otherwise, just disable C3 completely.
1065      */
1066     if (AcpiGbl_FADT.Pm2ControlBlock == 0 ||
1067 	AcpiGbl_FADT.Pm2ControlLength == 0) {
1068 	if ((AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD) &&
1069 	    (AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD_FLUSH) == 0) {
1070 	    cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1071 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1072 		"acpi_cpu: no BM control, using flush cache method\n"));
1073 	} else {
1074 	    cpu_quirks |= CPU_QUIRK_NO_C3;
1075 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1076 		"acpi_cpu: no BM control, C3 not available\n"));
1077 	}
1078     }
1079 
1080     /*
1081      * If we are using generic Cx mode, C3 on multiple CPUs requires using
1082      * the expensive flush cache instruction.
1083      */
1084     if (cpu_cx_generic && mp_ncpus > 1) {
1085 	cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1086 	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1087 	    "acpi_cpu: SMP, using flush cache mode for C3\n"));
1088     }
1089 
1090     /* Look for various quirks of the PIIX4 part. */
1091     acpi_dev = pci_find_device(PCI_VENDOR_INTEL, PCI_DEVICE_82371AB_3);
1092     if (acpi_dev != NULL) {
1093 	switch (pci_get_revid(acpi_dev)) {
1094 	/*
1095 	 * Disable C3 support for all PIIX4 chipsets.  Some of these parts
1096 	 * do not report the BMIDE status to the BM status register and
1097 	 * others have a livelock bug if Type-F DMA is enabled.  Linux
1098 	 * works around the BMIDE bug by reading the BM status directly
1099 	 * but we take the simpler approach of disabling C3 for these
1100 	 * parts.
1101 	 *
1102 	 * See erratum #18 ("C3 Power State/BMIDE and Type-F DMA
1103 	 * Livelock") from the January 2002 PIIX4 specification update.
1104 	 * Applies to all PIIX4 models.
1105 	 *
1106 	 * Also, make sure that all interrupts cause a "Stop Break"
1107 	 * event to exit from C2 state.
1108 	 * Also, BRLD_EN_BM (ACPI_BITREG_BUS_MASTER_RLD in ACPI-speak)
1109 	 * should be set to zero, otherwise it causes C2 to short-sleep.
1110 	 * PIIX4 doesn't properly support C3 and bus master activity
1111 	 * need not break out of C2.
1112 	 */
1113 	case PCI_REVISION_A_STEP:
1114 	case PCI_REVISION_B_STEP:
1115 	case PCI_REVISION_4E:
1116 	case PCI_REVISION_4M:
1117 	    cpu_quirks |= CPU_QUIRK_NO_C3;
1118 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1119 		"acpi_cpu: working around PIIX4 bug, disabling C3\n"));
1120 
1121 	    val = pci_read_config(acpi_dev, PIIX4_DEVACTB_REG, 4);
1122 	    if ((val & PIIX4_STOP_BREAK_MASK) != PIIX4_STOP_BREAK_MASK) {
1123 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1124 		    "acpi_cpu: PIIX4: enabling IRQs to generate Stop Break\n"));
1125 	    	val |= PIIX4_STOP_BREAK_MASK;
1126 		pci_write_config(acpi_dev, PIIX4_DEVACTB_REG, val, 4);
1127 	    }
1128 	    AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_RLD, &val);
1129 	    if (val) {
1130 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1131 		    "acpi_cpu: PIIX4: reset BRLD_EN_BM\n"));
1132 		AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1133 	    }
1134 	    break;
1135 	default:
1136 	    break;
1137 	}
1138     }
1139 
1140     return (0);
1141 }
1142 
1143 static int
1144 acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS)
1145 {
1146     struct acpi_cpu_softc *sc;
1147     struct sbuf	 sb;
1148     char	 buf[128];
1149     int		 i;
1150     uintmax_t	 fract, sum, whole;
1151 
1152     sc = (struct acpi_cpu_softc *) arg1;
1153     sum = 0;
1154     for (i = 0; i < sc->cpu_cx_count; i++)
1155 	sum += sc->cpu_cx_stats[i];
1156     sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1157     for (i = 0; i < sc->cpu_cx_count; i++) {
1158 	if (sum > 0) {
1159 	    whole = (uintmax_t)sc->cpu_cx_stats[i] * 100;
1160 	    fract = (whole % sum) * 100;
1161 	    sbuf_printf(&sb, "%u.%02u%% ", (u_int)(whole / sum),
1162 		(u_int)(fract / sum));
1163 	} else
1164 	    sbuf_printf(&sb, "0.00%% ");
1165     }
1166     sbuf_printf(&sb, "last %dus", sc->cpu_prev_sleep);
1167     sbuf_trim(&sb);
1168     sbuf_finish(&sb);
1169     sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1170     sbuf_delete(&sb);
1171 
1172     return (0);
1173 }
1174 
1175 static int
1176 acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc, int val)
1177 {
1178     int i;
1179 
1180     ACPI_SERIAL_ASSERT(cpu);
1181     sc->cpu_cx_lowest = val;
1182 
1183     /* If not disabling, cache the new lowest non-C3 state. */
1184     sc->cpu_non_c3 = 0;
1185     for (i = sc->cpu_cx_lowest; i >= 0; i--) {
1186 	if (sc->cpu_cx_states[i].type < ACPI_STATE_C3) {
1187 	    sc->cpu_non_c3 = i;
1188 	    break;
1189 	}
1190     }
1191 
1192     /* Reset the statistics counters. */
1193     bzero(sc->cpu_cx_stats, sizeof(sc->cpu_cx_stats));
1194     return (0);
1195 }
1196 
1197 static int
1198 acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1199 {
1200     struct	 acpi_cpu_softc *sc;
1201     char	 state[8];
1202     int		 val, error;
1203 
1204     sc = (struct acpi_cpu_softc *) arg1;
1205     snprintf(state, sizeof(state), "C%d", sc->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 > sc->cpu_cx_count - 1)
1213 	return (EINVAL);
1214 
1215     ACPI_SERIAL_BEGIN(cpu);
1216     acpi_cpu_set_cx_lowest(sc, val);
1217     ACPI_SERIAL_END(cpu);
1218 
1219     return (0);
1220 }
1221 
1222 static int
1223 acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1224 {
1225     struct	acpi_cpu_softc *sc;
1226     char	state[8];
1227     int		val, error, i;
1228 
1229     snprintf(state, sizeof(state), "C%d", cpu_cx_lowest + 1);
1230     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1231     if (error != 0 || req->newptr == NULL)
1232 	return (error);
1233     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1234 	return (EINVAL);
1235     val = (int) strtol(state + 1, NULL, 10) - 1;
1236     if (val < 0 || val > cpu_cx_count - 1)
1237 	return (EINVAL);
1238     cpu_cx_lowest = val;
1239 
1240     /* Update the new lowest useable Cx state for all CPUs. */
1241     ACPI_SERIAL_BEGIN(cpu);
1242     for (i = 0; i < cpu_ndevices; i++) {
1243 	sc = device_get_softc(cpu_devices[i]);
1244 	acpi_cpu_set_cx_lowest(sc, min(val, sc->cpu_cx_count - 1));
1245     }
1246     ACPI_SERIAL_END(cpu);
1247 
1248     return (0);
1249 }
1250