xref: /freebsd/sys/dev/acpica/acpi_cpu.c (revision 045c8f526484cb3b97f5fd693987f4376fa43c5f)
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/sched.h>
42 #include <sys/sbuf.h>
43 #include <sys/smp.h>
44 
45 #include <dev/pci/pcivar.h>
46 #include <machine/atomic.h>
47 #include <machine/bus.h>
48 #if defined(__amd64__) || defined(__i386__)
49 #include <machine/clock.h>
50 #include <machine/specialreg.h>
51 #include <machine/md_var.h>
52 #endif
53 #include <sys/rman.h>
54 
55 #include <contrib/dev/acpica/include/acpi.h>
56 #include <contrib/dev/acpica/include/accommon.h>
57 
58 #include <dev/acpica/acpivar.h>
59 
60 /*
61  * Support for ACPI Processor devices, including C[1-3] sleep states.
62  */
63 
64 /* Hooks for the ACPI CA debugging infrastructure */
65 #define _COMPONENT	ACPI_PROCESSOR
66 ACPI_MODULE_NAME("PROCESSOR")
67 
68 struct acpi_cx {
69     struct resource	*p_lvlx;	/* Register to read to enter state. */
70     uint32_t		 type;		/* C1-3 (C4 and up treated as C3). */
71     uint32_t		 trans_lat;	/* Transition latency (usec). */
72     uint32_t		 power;		/* Power consumed (mW). */
73     int			 res_type;	/* Resource type for p_lvlx. */
74     int			 res_rid;	/* Resource ID for p_lvlx. */
75     bool		 do_mwait;
76     uint32_t		 mwait_hint;
77     bool		 mwait_hw_coord;
78     bool		 mwait_bm_avoidance;
79 };
80 #define MAX_CX_STATES	 8
81 
82 struct acpi_cpu_softc {
83     device_t		 cpu_dev;
84     ACPI_HANDLE		 cpu_handle;
85     struct pcpu		*cpu_pcpu;
86     uint32_t		 cpu_acpi_id;	/* ACPI processor id */
87     uint32_t		 cpu_p_blk;	/* ACPI P_BLK location */
88     uint32_t		 cpu_p_blk_len;	/* P_BLK length (must be 6). */
89     struct acpi_cx	 cpu_cx_states[MAX_CX_STATES];
90     int			 cpu_cx_count;	/* Number of valid Cx states. */
91     int			 cpu_prev_sleep;/* Last idle sleep duration. */
92     int			 cpu_features;	/* Child driver supported features. */
93     /* Runtime state. */
94     int			 cpu_non_c2;	/* Index of lowest non-C2 state. */
95     int			 cpu_non_c3;	/* Index of lowest non-C3 state. */
96     u_int		 cpu_cx_stats[MAX_CX_STATES];/* Cx usage history. */
97     /* Values for sysctl. */
98     struct sysctl_ctx_list cpu_sysctl_ctx;
99     struct sysctl_oid	*cpu_sysctl_tree;
100     int			 cpu_cx_lowest;
101     int			 cpu_cx_lowest_lim;
102     int			 cpu_disable_idle; /* Disable entry to idle function */
103     char 		 cpu_cx_supported[64];
104 };
105 
106 struct acpi_cpu_device {
107     struct resource_list	ad_rl;
108 };
109 
110 #define CPU_GET_REG(reg, width) 					\
111     (bus_space_read_ ## width(rman_get_bustag((reg)), 			\
112 		      rman_get_bushandle((reg)), 0))
113 #define CPU_SET_REG(reg, width, val)					\
114     (bus_space_write_ ## width(rman_get_bustag((reg)), 			\
115 		       rman_get_bushandle((reg)), 0, (val)))
116 
117 #define ACPI_NOTIFY_CX_STATES	0x81	/* _CST changed. */
118 
119 #define CPU_QUIRK_NO_C3		(1<<0)	/* C3-type states are not usable. */
120 #define CPU_QUIRK_NO_BM_CTRL	(1<<2)	/* No bus mastering control. */
121 
122 #define PCI_VENDOR_INTEL	0x8086
123 #define PCI_DEVICE_82371AB_3	0x7113	/* PIIX4 chipset for quirks. */
124 #define PCI_REVISION_A_STEP	0
125 #define PCI_REVISION_B_STEP	1
126 #define PCI_REVISION_4E		2
127 #define PCI_REVISION_4M		3
128 #define PIIX4_DEVACTB_REG	0x58
129 #define PIIX4_BRLD_EN_IRQ0	(1<<0)
130 #define PIIX4_BRLD_EN_IRQ	(1<<1)
131 #define PIIX4_BRLD_EN_IRQ8	(1<<5)
132 #define PIIX4_STOP_BREAK_MASK	(PIIX4_BRLD_EN_IRQ0 | PIIX4_BRLD_EN_IRQ | PIIX4_BRLD_EN_IRQ8)
133 #define PIIX4_PCNTRL_BST_EN	(1<<10)
134 
135 #define	CST_FFH_VENDOR_INTEL	1
136 #define	CST_FFH_INTEL_CL_C1IO	1
137 #define	CST_FFH_INTEL_CL_MWAIT	2
138 #define	CST_FFH_MWAIT_HW_COORD	0x0001
139 #define	CST_FFH_MWAIT_BM_AVOID	0x0002
140 
141 #define	CPUDEV_DEVICE_ID	"ACPI0007"
142 
143 /* Knob to disable acpi_cpu devices */
144 bool acpi_cpu_disabled = false;
145 
146 /* Platform hardware resource information. */
147 static uint32_t		 cpu_smi_cmd;	/* Value to write to SMI_CMD. */
148 static uint8_t		 cpu_cst_cnt;	/* Indicate we are _CST aware. */
149 static int		 cpu_quirks;	/* Indicate any hardware bugs. */
150 
151 /* Values for sysctl. */
152 static struct sysctl_ctx_list cpu_sysctl_ctx;
153 static struct sysctl_oid *cpu_sysctl_tree;
154 static int		 cpu_cx_generic;
155 static int		 cpu_cx_lowest_lim;
156 
157 static struct acpi_cpu_softc **cpu_softc;
158 ACPI_SERIAL_DECL(cpu, "ACPI CPU");
159 
160 static int	acpi_cpu_probe(device_t dev);
161 static int	acpi_cpu_attach(device_t dev);
162 static int	acpi_cpu_suspend(device_t dev);
163 static int	acpi_cpu_resume(device_t dev);
164 static int	acpi_pcpu_get_id(device_t dev, uint32_t acpi_id,
165 		    u_int *cpu_id);
166 static struct resource_list *acpi_cpu_get_rlist(device_t dev, device_t child);
167 static device_t	acpi_cpu_add_child(device_t dev, u_int order, const char *name,
168 		    int unit);
169 static int	acpi_cpu_read_ivar(device_t dev, device_t child, int index,
170 		    uintptr_t *result);
171 static int	acpi_cpu_shutdown(device_t dev);
172 static void	acpi_cpu_cx_probe(struct acpi_cpu_softc *sc);
173 static void	acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc);
174 static int	acpi_cpu_cx_cst(struct acpi_cpu_softc *sc);
175 static void	acpi_cpu_startup(void *arg);
176 static void	acpi_cpu_startup_cx(struct acpi_cpu_softc *sc);
177 static void	acpi_cpu_cx_list(struct acpi_cpu_softc *sc);
178 #if defined(__i386__) || defined(__amd64__)
179 static void	acpi_cpu_idle(sbintime_t sbt);
180 #endif
181 static void	acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context);
182 static void	acpi_cpu_quirks(void);
183 static void	acpi_cpu_quirks_piix4(void);
184 static int	acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS);
185 static int	acpi_cpu_usage_counters_sysctl(SYSCTL_HANDLER_ARGS);
186 static int	acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc);
187 static int	acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
188 static int	acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
189 #if defined(__i386__) || defined(__amd64__)
190 static int	acpi_cpu_method_sysctl(SYSCTL_HANDLER_ARGS);
191 #endif
192 
193 static device_method_t acpi_cpu_methods[] = {
194     /* Device interface */
195     DEVMETHOD(device_probe,	acpi_cpu_probe),
196     DEVMETHOD(device_attach,	acpi_cpu_attach),
197     DEVMETHOD(device_detach,	bus_generic_detach),
198     DEVMETHOD(device_shutdown,	acpi_cpu_shutdown),
199     DEVMETHOD(device_suspend,	acpi_cpu_suspend),
200     DEVMETHOD(device_resume,	acpi_cpu_resume),
201 
202     /* Bus interface */
203     DEVMETHOD(bus_add_child,	acpi_cpu_add_child),
204     DEVMETHOD(bus_read_ivar,	acpi_cpu_read_ivar),
205     DEVMETHOD(bus_get_resource_list, acpi_cpu_get_rlist),
206     DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
207     DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
208     DEVMETHOD(bus_alloc_resource, bus_generic_rl_alloc_resource),
209     DEVMETHOD(bus_release_resource, bus_generic_rl_release_resource),
210     DEVMETHOD(bus_activate_resource, bus_generic_activate_resource),
211     DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource),
212     DEVMETHOD(bus_setup_intr,	bus_generic_setup_intr),
213     DEVMETHOD(bus_teardown_intr, bus_generic_teardown_intr),
214 
215     DEVMETHOD_END
216 };
217 
218 static driver_t acpi_cpu_driver = {
219     "cpu",
220     acpi_cpu_methods,
221     sizeof(struct acpi_cpu_softc),
222 };
223 
224 static devclass_t acpi_cpu_devclass;
225 DRIVER_MODULE(cpu, acpi, acpi_cpu_driver, acpi_cpu_devclass, 0, 0);
226 MODULE_DEPEND(cpu, acpi, 1, 1, 1);
227 
228 static int
229 acpi_cpu_probe(device_t dev)
230 {
231     static char		   *cpudev_ids[] = { CPUDEV_DEVICE_ID, NULL };
232     int			   acpi_id, cpu_id;
233     ACPI_BUFFER		   buf;
234     ACPI_HANDLE		   handle;
235     ACPI_OBJECT		   *obj;
236     ACPI_STATUS		   status;
237     ACPI_OBJECT_TYPE	   type;
238 
239     if (acpi_disabled("cpu") || acpi_cpu_disabled)
240 	return (ENXIO);
241     type = acpi_get_type(dev);
242     if (type != ACPI_TYPE_PROCESSOR && type != ACPI_TYPE_DEVICE)
243 	return (ENXIO);
244     if (type == ACPI_TYPE_DEVICE &&
245 	ACPI_ID_PROBE(device_get_parent(dev), dev, cpudev_ids, NULL) >= 0)
246 	return (ENXIO);
247 
248     handle = acpi_get_handle(dev);
249     if (cpu_softc == NULL)
250 	cpu_softc = malloc(sizeof(struct acpi_cpu_softc *) *
251 	    (mp_maxid + 1), M_TEMP /* XXX */, M_WAITOK | M_ZERO);
252 
253     if (type == ACPI_TYPE_PROCESSOR) {
254 	/* Get our Processor object. */
255 	buf.Pointer = NULL;
256 	buf.Length = ACPI_ALLOCATE_BUFFER;
257 	status = AcpiEvaluateObject(handle, NULL, NULL, &buf);
258 	if (ACPI_FAILURE(status)) {
259 	    device_printf(dev, "probe failed to get Processor obj - %s\n",
260 		AcpiFormatException(status));
261 	    return (ENXIO);
262 	}
263 	obj = (ACPI_OBJECT *)buf.Pointer;
264 	if (obj->Type != ACPI_TYPE_PROCESSOR) {
265 	    device_printf(dev, "Processor object has bad type %d\n",
266 		obj->Type);
267 	    AcpiOsFree(obj);
268 	    return (ENXIO);
269 	}
270 
271 	/*
272 	 * Find the processor associated with our unit.  We could use the
273 	 * ProcId as a key, however, some boxes do not have the same values
274 	 * in their Processor object as the ProcId values in the MADT.
275 	 */
276 	acpi_id = obj->Processor.ProcId;
277 	AcpiOsFree(obj);
278     } else {
279 	status = acpi_GetInteger(handle, "_UID", &acpi_id);
280 	if (ACPI_FAILURE(status)) {
281 	    device_printf(dev, "Device object has bad value - %s\n",
282 		AcpiFormatException(status));
283 	    return (ENXIO);
284 	}
285     }
286     if (acpi_pcpu_get_id(dev, acpi_id, &cpu_id) != 0) {
287 	if (bootverbose && (type != ACPI_TYPE_PROCESSOR || acpi_id != 255))
288 	    printf("ACPI: Processor %s (ACPI ID %u) ignored\n",
289 		acpi_name(acpi_get_handle(dev)), acpi_id);
290 	return (ENXIO);
291     }
292 
293     if (device_set_unit(dev, cpu_id) != 0)
294 	return (ENXIO);
295 
296     device_set_desc(dev, "ACPI CPU");
297 
298     if (!bootverbose && device_get_unit(dev) != 0) {
299 	    device_quiet(dev);
300 	    device_quiet_children(dev);
301     }
302 
303     return (0);
304 }
305 
306 static int
307 acpi_cpu_attach(device_t dev)
308 {
309     ACPI_BUFFER		   buf;
310     ACPI_OBJECT		   arg, *obj;
311     ACPI_OBJECT_LIST	   arglist;
312     struct pcpu		   *pcpu_data;
313     struct acpi_cpu_softc *sc;
314     struct acpi_softc	  *acpi_sc;
315     ACPI_STATUS		   status;
316     u_int		   features;
317     int			   cpu_id, drv_count, i;
318     driver_t 		  **drivers;
319     uint32_t		   cap_set[3];
320 
321     /* UUID needed by _OSC evaluation */
322     static uint8_t cpu_oscuuid[16] = { 0x16, 0xA6, 0x77, 0x40, 0x0C, 0x29,
323 				       0xBE, 0x47, 0x9E, 0xBD, 0xD8, 0x70,
324 				       0x58, 0x71, 0x39, 0x53 };
325 
326     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
327 
328     sc = device_get_softc(dev);
329     sc->cpu_dev = dev;
330     sc->cpu_handle = acpi_get_handle(dev);
331     cpu_id = device_get_unit(dev);
332     cpu_softc[cpu_id] = sc;
333     pcpu_data = pcpu_find(cpu_id);
334     pcpu_data->pc_device = dev;
335     sc->cpu_pcpu = pcpu_data;
336     cpu_smi_cmd = AcpiGbl_FADT.SmiCommand;
337     cpu_cst_cnt = AcpiGbl_FADT.CstControl;
338 
339     if (acpi_get_type(dev) == ACPI_TYPE_PROCESSOR) {
340 	buf.Pointer = NULL;
341 	buf.Length = ACPI_ALLOCATE_BUFFER;
342 	status = AcpiEvaluateObject(sc->cpu_handle, NULL, NULL, &buf);
343 	if (ACPI_FAILURE(status)) {
344 	    device_printf(dev, "attach failed to get Processor obj - %s\n",
345 		AcpiFormatException(status));
346 	    return (ENXIO);
347 	}
348 	obj = (ACPI_OBJECT *)buf.Pointer;
349 	sc->cpu_p_blk = obj->Processor.PblkAddress;
350 	sc->cpu_p_blk_len = obj->Processor.PblkLength;
351 	sc->cpu_acpi_id = obj->Processor.ProcId;
352 	AcpiOsFree(obj);
353     } else {
354 	KASSERT(acpi_get_type(dev) == ACPI_TYPE_DEVICE,
355 	    ("Unexpected ACPI object"));
356 	status = acpi_GetInteger(sc->cpu_handle, "_UID", &sc->cpu_acpi_id);
357 	if (ACPI_FAILURE(status)) {
358 	    device_printf(dev, "Device object has bad value - %s\n",
359 		AcpiFormatException(status));
360 	    return (ENXIO);
361 	}
362 	sc->cpu_p_blk = 0;
363 	sc->cpu_p_blk_len = 0;
364     }
365     ACPI_DEBUG_PRINT((ACPI_DB_INFO, "acpi_cpu%d: P_BLK at %#x/%d\n",
366 		     device_get_unit(dev), sc->cpu_p_blk, sc->cpu_p_blk_len));
367 
368     /*
369      * If this is the first cpu we attach, create and initialize the generic
370      * resources that will be used by all acpi cpu devices.
371      */
372     if (device_get_unit(dev) == 0) {
373 	/* Assume we won't be using generic Cx mode by default */
374 	cpu_cx_generic = FALSE;
375 
376 	/* Install hw.acpi.cpu sysctl tree */
377 	acpi_sc = acpi_device_get_parent_softc(dev);
378 	sysctl_ctx_init(&cpu_sysctl_ctx);
379 	cpu_sysctl_tree = SYSCTL_ADD_NODE(&cpu_sysctl_ctx,
380 	    SYSCTL_CHILDREN(acpi_sc->acpi_sysctl_tree), OID_AUTO, "cpu",
381 	    CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "node for CPU children");
382     }
383 
384     /*
385      * Before calling any CPU methods, collect child driver feature hints
386      * and notify ACPI of them.  We support unified SMP power control
387      * so advertise this ourselves.  Note this is not the same as independent
388      * SMP control where each CPU can have different settings.
389      */
390     sc->cpu_features = ACPI_CAP_SMP_SAME | ACPI_CAP_SMP_SAME_C3 |
391       ACPI_CAP_C1_IO_HALT;
392 
393 #if defined(__i386__) || defined(__amd64__)
394     /*
395      * Ask for MWAIT modes if not disabled and interrupts work
396      * reasonable with MWAIT.
397      */
398     if (!acpi_disabled("mwait") && cpu_mwait_usable())
399 	sc->cpu_features |= ACPI_CAP_SMP_C1_NATIVE | ACPI_CAP_SMP_C3_NATIVE;
400 #endif
401 
402     if (devclass_get_drivers(acpi_cpu_devclass, &drivers, &drv_count) == 0) {
403 	for (i = 0; i < drv_count; i++) {
404 	    if (ACPI_GET_FEATURES(drivers[i], &features) == 0)
405 		sc->cpu_features |= features;
406 	}
407 	free(drivers, M_TEMP);
408     }
409 
410     /*
411      * CPU capabilities are specified in
412      * Intel Processor Vendor-Specific ACPI Interface Specification.
413      */
414     if (sc->cpu_features) {
415 	cap_set[1] = sc->cpu_features;
416 	status = acpi_EvaluateOSC(sc->cpu_handle, cpu_oscuuid, 1, 2, cap_set,
417 	    cap_set, false);
418 	if (ACPI_SUCCESS(status)) {
419 	    if (cap_set[0] != 0)
420 		device_printf(dev, "_OSC returned status %#x\n", cap_set[0]);
421 	}
422 	else {
423 	    arglist.Pointer = &arg;
424 	    arglist.Count = 1;
425 	    arg.Type = ACPI_TYPE_BUFFER;
426 	    arg.Buffer.Length = sizeof(cap_set);
427 	    arg.Buffer.Pointer = (uint8_t *)cap_set;
428 	    cap_set[0] = 1; /* revision */
429 	    cap_set[1] = 1; /* number of capabilities integers */
430 	    cap_set[2] = sc->cpu_features;
431 	    AcpiEvaluateObject(sc->cpu_handle, "_PDC", &arglist, NULL);
432 	}
433     }
434 
435     /* Probe for Cx state support. */
436     acpi_cpu_cx_probe(sc);
437 
438     return (0);
439 }
440 
441 static void
442 acpi_cpu_postattach(void *unused __unused)
443 {
444     struct acpi_cpu_softc *sc;
445     int attached = 0, i;
446 
447     if (cpu_softc == NULL)
448 	return;
449 
450     bus_topo_lock();
451     CPU_FOREACH(i) {
452 	if ((sc = cpu_softc[i]) != NULL)
453 		bus_generic_probe(sc->cpu_dev);
454     }
455     CPU_FOREACH(i) {
456 	if ((sc = cpu_softc[i]) != NULL) {
457 		bus_generic_attach(sc->cpu_dev);
458 		attached = 1;
459 	}
460     }
461     bus_topo_unlock();
462 
463     if (attached) {
464 #ifdef EARLY_AP_STARTUP
465 	acpi_cpu_startup(NULL);
466 #else
467 	/* Queue post cpu-probing task handler */
468 	AcpiOsExecute(OSL_NOTIFY_HANDLER, acpi_cpu_startup, NULL);
469 #endif
470     }
471 }
472 
473 SYSINIT(acpi_cpu, SI_SUB_CONFIGURE, SI_ORDER_MIDDLE,
474     acpi_cpu_postattach, NULL);
475 
476 static void
477 disable_idle(struct acpi_cpu_softc *sc)
478 {
479     cpuset_t cpuset;
480 
481     CPU_SETOF(sc->cpu_pcpu->pc_cpuid, &cpuset);
482     sc->cpu_disable_idle = TRUE;
483 
484     /*
485      * Ensure that the CPU is not in idle state or in acpi_cpu_idle().
486      * Note that this code depends on the fact that the rendezvous IPI
487      * can not penetrate context where interrupts are disabled and acpi_cpu_idle
488      * is called and executed in such a context with interrupts being re-enabled
489      * right before return.
490      */
491     smp_rendezvous_cpus(cpuset, smp_no_rendezvous_barrier, NULL,
492 	smp_no_rendezvous_barrier, NULL);
493 }
494 
495 static void
496 enable_idle(struct acpi_cpu_softc *sc)
497 {
498 
499     sc->cpu_disable_idle = FALSE;
500 }
501 
502 #if defined(__i386__) || defined(__amd64__)
503 static int
504 is_idle_disabled(struct acpi_cpu_softc *sc)
505 {
506 
507     return (sc->cpu_disable_idle);
508 }
509 #endif
510 
511 /*
512  * Disable any entry to the idle function during suspend and re-enable it
513  * during resume.
514  */
515 static int
516 acpi_cpu_suspend(device_t dev)
517 {
518     int error;
519 
520     error = bus_generic_suspend(dev);
521     if (error)
522 	return (error);
523     disable_idle(device_get_softc(dev));
524     return (0);
525 }
526 
527 static int
528 acpi_cpu_resume(device_t dev)
529 {
530 
531     enable_idle(device_get_softc(dev));
532     return (bus_generic_resume(dev));
533 }
534 
535 /*
536  * Find the processor associated with a given ACPI ID.
537  */
538 static int
539 acpi_pcpu_get_id(device_t dev, uint32_t acpi_id, u_int *cpu_id)
540 {
541     struct pcpu	*pc;
542     u_int	 i;
543 
544     CPU_FOREACH(i) {
545 	pc = pcpu_find(i);
546 	if (pc->pc_acpi_id == acpi_id) {
547 	    *cpu_id = pc->pc_cpuid;
548 	    return (0);
549 	}
550     }
551 
552     /*
553      * If pc_acpi_id for CPU 0 is not initialized (e.g. a non-APIC
554      * UP box) use the ACPI ID from the first processor we find.
555      */
556     if (mp_ncpus == 1) {
557 	pc = pcpu_find(0);
558 	if (pc->pc_acpi_id == 0xffffffff)
559 	    pc->pc_acpi_id = acpi_id;
560 	*cpu_id = 0;
561 	return (0);
562     }
563 
564     return (ESRCH);
565 }
566 
567 static struct resource_list *
568 acpi_cpu_get_rlist(device_t dev, device_t child)
569 {
570     struct acpi_cpu_device *ad;
571 
572     ad = device_get_ivars(child);
573     if (ad == NULL)
574 	return (NULL);
575     return (&ad->ad_rl);
576 }
577 
578 static device_t
579 acpi_cpu_add_child(device_t dev, u_int order, const char *name, int unit)
580 {
581     struct acpi_cpu_device *ad;
582     device_t child;
583 
584     if ((ad = malloc(sizeof(*ad), M_TEMP, M_NOWAIT | M_ZERO)) == NULL)
585 	return (NULL);
586 
587     resource_list_init(&ad->ad_rl);
588 
589     child = device_add_child_ordered(dev, order, name, unit);
590     if (child != NULL)
591 	device_set_ivars(child, ad);
592     else
593 	free(ad, M_TEMP);
594     return (child);
595 }
596 
597 static int
598 acpi_cpu_read_ivar(device_t dev, device_t child, int index, uintptr_t *result)
599 {
600     struct acpi_cpu_softc *sc;
601 
602     sc = device_get_softc(dev);
603     switch (index) {
604     case ACPI_IVAR_HANDLE:
605 	*result = (uintptr_t)sc->cpu_handle;
606 	break;
607     case CPU_IVAR_PCPU:
608 	*result = (uintptr_t)sc->cpu_pcpu;
609 	break;
610 #if defined(__amd64__) || defined(__i386__)
611     case CPU_IVAR_NOMINAL_MHZ:
612 	if (tsc_is_invariant) {
613 	    *result = (uintptr_t)(atomic_load_acq_64(&tsc_freq) / 1000000);
614 	    break;
615 	}
616 	/* FALLTHROUGH */
617 #endif
618     default:
619 	return (ENOENT);
620     }
621     return (0);
622 }
623 
624 static int
625 acpi_cpu_shutdown(device_t dev)
626 {
627     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
628 
629     /* Allow children to shutdown first. */
630     bus_generic_shutdown(dev);
631 
632     /*
633      * Disable any entry to the idle function.
634      */
635     disable_idle(device_get_softc(dev));
636 
637     /*
638      * CPU devices are not truly detached and remain referenced,
639      * so their resources are not freed.
640      */
641 
642     return_VALUE (0);
643 }
644 
645 static void
646 acpi_cpu_cx_probe(struct acpi_cpu_softc *sc)
647 {
648     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
649 
650     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
651     sc->cpu_prev_sleep = 1000000;
652     sc->cpu_cx_lowest = 0;
653     sc->cpu_cx_lowest_lim = 0;
654 
655     /*
656      * Check for the ACPI 2.0 _CST sleep states object. If we can't find
657      * any, we'll revert to generic FADT/P_BLK Cx control method which will
658      * be handled by acpi_cpu_startup. We need to defer to after having
659      * probed all the cpus in the system before probing for generic Cx
660      * states as we may already have found cpus with valid _CST packages
661      */
662     if (!cpu_cx_generic && acpi_cpu_cx_cst(sc) != 0) {
663 	/*
664 	 * We were unable to find a _CST package for this cpu or there
665 	 * was an error parsing it. Switch back to generic mode.
666 	 */
667 	cpu_cx_generic = TRUE;
668 	if (bootverbose)
669 	    device_printf(sc->cpu_dev, "switching to generic Cx mode\n");
670     }
671 
672     /*
673      * TODO: _CSD Package should be checked here.
674      */
675 }
676 
677 static void
678 acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc)
679 {
680     ACPI_GENERIC_ADDRESS	 gas;
681     struct acpi_cx		*cx_ptr;
682 
683     sc->cpu_cx_count = 0;
684     cx_ptr = sc->cpu_cx_states;
685 
686     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
687     sc->cpu_prev_sleep = 1000000;
688 
689     /* C1 has been required since just after ACPI 1.0 */
690     cx_ptr->type = ACPI_STATE_C1;
691     cx_ptr->trans_lat = 0;
692     cx_ptr++;
693     sc->cpu_non_c2 = sc->cpu_cx_count;
694     sc->cpu_non_c3 = sc->cpu_cx_count;
695     sc->cpu_cx_count++;
696 
697     /*
698      * The spec says P_BLK must be 6 bytes long.  However, some systems
699      * use it to indicate a fractional set of features present so we
700      * take 5 as C2.  Some may also have a value of 7 to indicate
701      * another C3 but most use _CST for this (as required) and having
702      * "only" C1-C3 is not a hardship.
703      */
704     if (sc->cpu_p_blk_len < 5)
705 	return;
706 
707     /* Validate and allocate resources for C2 (P_LVL2). */
708     gas.SpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
709     gas.BitWidth = 8;
710     if (AcpiGbl_FADT.C2Latency <= 100) {
711 	gas.Address = sc->cpu_p_blk + 4;
712 	cx_ptr->res_rid = 0;
713 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &cx_ptr->res_rid,
714 	    &gas, &cx_ptr->p_lvlx, RF_SHAREABLE);
715 	if (cx_ptr->p_lvlx != NULL) {
716 	    cx_ptr->type = ACPI_STATE_C2;
717 	    cx_ptr->trans_lat = AcpiGbl_FADT.C2Latency;
718 	    cx_ptr++;
719 	    sc->cpu_non_c3 = sc->cpu_cx_count;
720 	    sc->cpu_cx_count++;
721 	}
722     }
723     if (sc->cpu_p_blk_len < 6)
724 	return;
725 
726     /* Validate and allocate resources for C3 (P_LVL3). */
727     if (AcpiGbl_FADT.C3Latency <= 1000 && !(cpu_quirks & CPU_QUIRK_NO_C3)) {
728 	gas.Address = sc->cpu_p_blk + 5;
729 	cx_ptr->res_rid = 1;
730 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &cx_ptr->res_rid,
731 	    &gas, &cx_ptr->p_lvlx, RF_SHAREABLE);
732 	if (cx_ptr->p_lvlx != NULL) {
733 	    cx_ptr->type = ACPI_STATE_C3;
734 	    cx_ptr->trans_lat = AcpiGbl_FADT.C3Latency;
735 	    cx_ptr++;
736 	    sc->cpu_cx_count++;
737 	}
738     }
739 }
740 
741 #if defined(__i386__) || defined(__amd64__)
742 static void
743 acpi_cpu_cx_cst_mwait(struct acpi_cx *cx_ptr, uint64_t address, int accsize)
744 {
745 
746 	cx_ptr->do_mwait = true;
747 	cx_ptr->mwait_hint = address & 0xffffffff;
748 	cx_ptr->mwait_hw_coord = (accsize & CST_FFH_MWAIT_HW_COORD) != 0;
749 	cx_ptr->mwait_bm_avoidance = (accsize & CST_FFH_MWAIT_BM_AVOID) != 0;
750 }
751 #endif
752 
753 static void
754 acpi_cpu_cx_cst_free_plvlx(device_t cpu_dev, struct acpi_cx *cx_ptr)
755 {
756 
757 	if (cx_ptr->p_lvlx == NULL)
758 		return;
759 	bus_release_resource(cpu_dev, cx_ptr->res_type, cx_ptr->res_rid,
760 	    cx_ptr->p_lvlx);
761 	cx_ptr->p_lvlx = NULL;
762 }
763 
764 /*
765  * Parse a _CST package and set up its Cx states.  Since the _CST object
766  * can change dynamically, our notify handler may call this function
767  * to clean up and probe the new _CST package.
768  */
769 static int
770 acpi_cpu_cx_cst(struct acpi_cpu_softc *sc)
771 {
772     struct	 acpi_cx *cx_ptr;
773     ACPI_STATUS	 status;
774     ACPI_BUFFER	 buf;
775     ACPI_OBJECT	*top;
776     ACPI_OBJECT	*pkg;
777     uint32_t	 count;
778     int		 i;
779 #if defined(__i386__) || defined(__amd64__)
780     uint64_t	 address;
781     int		 vendor, class, accsize;
782 #endif
783 
784     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
785 
786     buf.Pointer = NULL;
787     buf.Length = ACPI_ALLOCATE_BUFFER;
788     status = AcpiEvaluateObject(sc->cpu_handle, "_CST", NULL, &buf);
789     if (ACPI_FAILURE(status))
790 	return (ENXIO);
791 
792     /* _CST is a package with a count and at least one Cx package. */
793     top = (ACPI_OBJECT *)buf.Pointer;
794     if (!ACPI_PKG_VALID(top, 2) || acpi_PkgInt32(top, 0, &count) != 0) {
795 	device_printf(sc->cpu_dev, "invalid _CST package\n");
796 	AcpiOsFree(buf.Pointer);
797 	return (ENXIO);
798     }
799     if (count != top->Package.Count - 1) {
800 	device_printf(sc->cpu_dev, "invalid _CST state count (%d != %d)\n",
801 	       count, top->Package.Count - 1);
802 	count = top->Package.Count - 1;
803     }
804     if (count > MAX_CX_STATES) {
805 	device_printf(sc->cpu_dev, "_CST has too many states (%d)\n", count);
806 	count = MAX_CX_STATES;
807     }
808 
809     sc->cpu_non_c2 = 0;
810     sc->cpu_non_c3 = 0;
811     sc->cpu_cx_count = 0;
812     cx_ptr = sc->cpu_cx_states;
813 
814     /*
815      * C1 has been required since just after ACPI 1.0.
816      * Reserve the first slot for it.
817      */
818     cx_ptr->type = ACPI_STATE_C0;
819     cx_ptr++;
820     sc->cpu_cx_count++;
821 
822     /* Set up all valid states. */
823     for (i = 0; i < count; i++) {
824 	pkg = &top->Package.Elements[i + 1];
825 	if (!ACPI_PKG_VALID(pkg, 4) ||
826 	    acpi_PkgInt32(pkg, 1, &cx_ptr->type) != 0 ||
827 	    acpi_PkgInt32(pkg, 2, &cx_ptr->trans_lat) != 0 ||
828 	    acpi_PkgInt32(pkg, 3, &cx_ptr->power) != 0) {
829 	    device_printf(sc->cpu_dev, "skipping invalid Cx state package\n");
830 	    continue;
831 	}
832 
833 	/* Validate the state to see if we should use it. */
834 	switch (cx_ptr->type) {
835 	case ACPI_STATE_C1:
836 	    acpi_cpu_cx_cst_free_plvlx(sc->cpu_dev, cx_ptr);
837 #if defined(__i386__) || defined(__amd64__)
838 	    if (acpi_PkgFFH_IntelCpu(pkg, 0, &vendor, &class, &address,
839 	      &accsize) == 0 && vendor == CST_FFH_VENDOR_INTEL) {
840 		if (class == CST_FFH_INTEL_CL_C1IO) {
841 		    /* C1 I/O then Halt */
842 		    cx_ptr->res_rid = sc->cpu_cx_count;
843 		    bus_set_resource(sc->cpu_dev, SYS_RES_IOPORT,
844 		      cx_ptr->res_rid, address, 1);
845 		    cx_ptr->p_lvlx = bus_alloc_resource_any(sc->cpu_dev,
846 		      SYS_RES_IOPORT, &cx_ptr->res_rid, RF_ACTIVE |
847 		      RF_SHAREABLE);
848 		    if (cx_ptr->p_lvlx == NULL) {
849 			bus_delete_resource(sc->cpu_dev, SYS_RES_IOPORT,
850 			  cx_ptr->res_rid);
851 			device_printf(sc->cpu_dev,
852 			  "C1 I/O failed to allocate port %d, "
853 			  "degrading to C1 Halt", (int)address);
854 		    }
855 		} else if (class == CST_FFH_INTEL_CL_MWAIT) {
856 		    acpi_cpu_cx_cst_mwait(cx_ptr, address, accsize);
857 		}
858 	    }
859 #endif
860 	    if (sc->cpu_cx_states[0].type == ACPI_STATE_C0) {
861 		/* This is the first C1 state.  Use the reserved slot. */
862 		sc->cpu_cx_states[0] = *cx_ptr;
863 	    } else {
864 		sc->cpu_non_c2 = sc->cpu_cx_count;
865 		sc->cpu_non_c3 = sc->cpu_cx_count;
866 		cx_ptr++;
867 		sc->cpu_cx_count++;
868 	    }
869 	    continue;
870 	case ACPI_STATE_C2:
871 	    sc->cpu_non_c3 = sc->cpu_cx_count;
872 	    break;
873 	case ACPI_STATE_C3:
874 	default:
875 	    if ((cpu_quirks & CPU_QUIRK_NO_C3) != 0) {
876 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
877 				 "acpi_cpu%d: C3[%d] not available.\n",
878 				 device_get_unit(sc->cpu_dev), i));
879 		continue;
880 	    }
881 	    break;
882 	}
883 
884 	/* Free up any previous register. */
885 	acpi_cpu_cx_cst_free_plvlx(sc->cpu_dev, cx_ptr);
886 
887 	/* Allocate the control register for C2 or C3. */
888 #if defined(__i386__) || defined(__amd64__)
889 	if (acpi_PkgFFH_IntelCpu(pkg, 0, &vendor, &class, &address,
890 	  &accsize) == 0 && vendor == CST_FFH_VENDOR_INTEL &&
891 	  class == CST_FFH_INTEL_CL_MWAIT) {
892 	    /* Native C State Instruction use (mwait) */
893 	    acpi_cpu_cx_cst_mwait(cx_ptr, address, accsize);
894 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
895 	      "acpi_cpu%d: Got C%d/mwait - %d latency\n",
896 	      device_get_unit(sc->cpu_dev), cx_ptr->type, cx_ptr->trans_lat));
897 	    cx_ptr++;
898 	    sc->cpu_cx_count++;
899 	} else
900 #endif
901 	{
902 	    cx_ptr->res_rid = sc->cpu_cx_count;
903 	    acpi_PkgGas(sc->cpu_dev, pkg, 0, &cx_ptr->res_type,
904 		&cx_ptr->res_rid, &cx_ptr->p_lvlx, RF_SHAREABLE);
905 	    if (cx_ptr->p_lvlx) {
906 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
907 		     "acpi_cpu%d: Got C%d - %d latency\n",
908 		     device_get_unit(sc->cpu_dev), cx_ptr->type,
909 		     cx_ptr->trans_lat));
910 		cx_ptr++;
911 		sc->cpu_cx_count++;
912 	    }
913 	}
914     }
915     AcpiOsFree(buf.Pointer);
916 
917     /* If C1 state was not found, we need one now. */
918     cx_ptr = sc->cpu_cx_states;
919     if (cx_ptr->type == ACPI_STATE_C0) {
920 	cx_ptr->type = ACPI_STATE_C1;
921 	cx_ptr->trans_lat = 0;
922     }
923 
924     return (0);
925 }
926 
927 /*
928  * Call this *after* all CPUs have been attached.
929  */
930 static void
931 acpi_cpu_startup(void *arg)
932 {
933     struct acpi_cpu_softc *sc;
934     int i;
935 
936     /*
937      * Setup any quirks that might necessary now that we have probed
938      * all the CPUs
939      */
940     acpi_cpu_quirks();
941 
942     if (cpu_cx_generic) {
943 	/*
944 	 * We are using generic Cx mode, probe for available Cx states
945 	 * for all processors.
946 	 */
947 	CPU_FOREACH(i) {
948 	    if ((sc = cpu_softc[i]) != NULL)
949 		acpi_cpu_generic_cx_probe(sc);
950 	}
951     } else {
952 	/*
953 	 * We are using _CST mode, remove C3 state if necessary.
954 	 * As we now know for sure that we will be using _CST mode
955 	 * install our notify handler.
956 	 */
957 	CPU_FOREACH(i) {
958 	    if ((sc = cpu_softc[i]) == NULL)
959 		continue;
960 	    if (cpu_quirks & CPU_QUIRK_NO_C3) {
961 		sc->cpu_cx_count = min(sc->cpu_cx_count, sc->cpu_non_c3 + 1);
962 	    }
963 	    AcpiInstallNotifyHandler(sc->cpu_handle, ACPI_DEVICE_NOTIFY,
964 		acpi_cpu_notify, sc);
965 	}
966     }
967 
968     /* Perform Cx final initialization. */
969     CPU_FOREACH(i) {
970 	if ((sc = cpu_softc[i]) != NULL)
971 	    acpi_cpu_startup_cx(sc);
972     }
973 
974     /* Add a sysctl handler to handle global Cx lowest setting */
975     SYSCTL_ADD_PROC(&cpu_sysctl_ctx, SYSCTL_CHILDREN(cpu_sysctl_tree),
976 	OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_MPSAFE,
977 	NULL, 0, acpi_cpu_global_cx_lowest_sysctl, "A",
978 	"Global lowest Cx sleep state to use");
979 
980     /* Take over idling from cpu_idle_default(). */
981     cpu_cx_lowest_lim = 0;
982     CPU_FOREACH(i) {
983 	if ((sc = cpu_softc[i]) != NULL)
984 	    enable_idle(sc);
985     }
986 #if defined(__i386__) || defined(__amd64__)
987     cpu_idle_hook = acpi_cpu_idle;
988 #endif
989 }
990 
991 static void
992 acpi_cpu_cx_list(struct acpi_cpu_softc *sc)
993 {
994     struct sbuf sb;
995     int i;
996 
997     /*
998      * Set up the list of Cx states
999      */
1000     sbuf_new(&sb, sc->cpu_cx_supported, sizeof(sc->cpu_cx_supported),
1001 	SBUF_FIXEDLEN);
1002     for (i = 0; i < sc->cpu_cx_count; i++)
1003 	sbuf_printf(&sb, "C%d/%d/%d ", i + 1, sc->cpu_cx_states[i].type,
1004 	    sc->cpu_cx_states[i].trans_lat);
1005     sbuf_trim(&sb);
1006     sbuf_finish(&sb);
1007 }
1008 
1009 static void
1010 acpi_cpu_startup_cx(struct acpi_cpu_softc *sc)
1011 {
1012     acpi_cpu_cx_list(sc);
1013 
1014     SYSCTL_ADD_STRING(&sc->cpu_sysctl_ctx,
1015 		      SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
1016 		      OID_AUTO, "cx_supported", CTLFLAG_RD,
1017 		      sc->cpu_cx_supported, 0,
1018 		      "Cx/microsecond values for supported Cx states");
1019     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
1020         SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)), OID_AUTO,
1021 	"cx_lowest", CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_MPSAFE,
1022 	(void *)sc, 0, acpi_cpu_cx_lowest_sysctl, "A",
1023 	"lowest Cx sleep state to use");
1024     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
1025         SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)), OID_AUTO,
1026 	"cx_usage", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE,
1027 	(void *)sc, 0, acpi_cpu_usage_sysctl, "A",
1028 	"percent usage for each Cx state");
1029     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
1030         SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)), OID_AUTO,
1031 	"cx_usage_counters", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE,
1032 	(void *)sc, 0, acpi_cpu_usage_counters_sysctl, "A",
1033 	"Cx sleep state counters");
1034 #if defined(__i386__) || defined(__amd64__)
1035     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
1036         SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)), OID_AUTO,
1037 	"cx_method", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE,
1038 	(void *)sc, 0, acpi_cpu_method_sysctl, "A", "Cx entrance methods");
1039 #endif
1040 
1041     /* Signal platform that we can handle _CST notification. */
1042     if (!cpu_cx_generic && cpu_cst_cnt != 0) {
1043 	ACPI_LOCK(acpi);
1044 	AcpiOsWritePort(cpu_smi_cmd, cpu_cst_cnt, 8);
1045 	ACPI_UNLOCK(acpi);
1046     }
1047 }
1048 
1049 #if defined(__i386__) || defined(__amd64__)
1050 /*
1051  * Idle the CPU in the lowest state possible.  This function is called with
1052  * interrupts disabled.  Note that once it re-enables interrupts, a task
1053  * switch can occur so do not access shared data (i.e. the softc) after
1054  * interrupts are re-enabled.
1055  */
1056 static void
1057 acpi_cpu_idle(sbintime_t sbt)
1058 {
1059     struct	acpi_cpu_softc *sc;
1060     struct	acpi_cx *cx_next;
1061     uint64_t	start_ticks, end_ticks;
1062     uint32_t	start_time, end_time;
1063     ACPI_STATUS	status;
1064     int		bm_active, cx_next_idx, i, us;
1065 
1066     /*
1067      * Look up our CPU id to get our softc.  If it's NULL, we'll use C1
1068      * since there is no ACPI processor object for this CPU.  This occurs
1069      * for logical CPUs in the HTT case.
1070      */
1071     sc = cpu_softc[PCPU_GET(cpuid)];
1072     if (sc == NULL) {
1073 	acpi_cpu_c1();
1074 	return;
1075     }
1076 
1077     /* If disabled, take the safe path. */
1078     if (is_idle_disabled(sc)) {
1079 	acpi_cpu_c1();
1080 	return;
1081     }
1082 
1083     /* Find the lowest state that has small enough latency. */
1084     us = sc->cpu_prev_sleep;
1085     if (sbt >= 0 && us > (sbt >> 12))
1086 	us = (sbt >> 12);
1087     cx_next_idx = 0;
1088     if (cpu_disable_c2_sleep)
1089 	i = min(sc->cpu_cx_lowest, sc->cpu_non_c2);
1090     else if (cpu_disable_c3_sleep)
1091 	i = min(sc->cpu_cx_lowest, sc->cpu_non_c3);
1092     else
1093 	i = sc->cpu_cx_lowest;
1094     for (; i >= 0; i--) {
1095 	if (sc->cpu_cx_states[i].trans_lat * 3 <= us) {
1096 	    cx_next_idx = i;
1097 	    break;
1098 	}
1099     }
1100 
1101     /*
1102      * Check for bus master activity.  If there was activity, clear
1103      * the bit and use the lowest non-C3 state.  Note that the USB
1104      * driver polling for new devices keeps this bit set all the
1105      * time if USB is loaded.
1106      */
1107     cx_next = &sc->cpu_cx_states[cx_next_idx];
1108     if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0 &&
1109 	cx_next_idx > sc->cpu_non_c3 &&
1110 	(!cx_next->do_mwait || cx_next->mwait_bm_avoidance)) {
1111 	status = AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, &bm_active);
1112 	if (ACPI_SUCCESS(status) && bm_active != 0) {
1113 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, 1);
1114 	    cx_next_idx = sc->cpu_non_c3;
1115 	    cx_next = &sc->cpu_cx_states[cx_next_idx];
1116 	}
1117     }
1118 
1119     /* Select the next state and update statistics. */
1120     sc->cpu_cx_stats[cx_next_idx]++;
1121     KASSERT(cx_next->type != ACPI_STATE_C0, ("acpi_cpu_idle: C0 sleep"));
1122 
1123     /*
1124      * Execute HLT (or equivalent) and wait for an interrupt.  We can't
1125      * precisely calculate the time spent in C1 since the place we wake up
1126      * is an ISR.  Assume we slept no more then half of quantum, unless
1127      * we are called inside critical section, delaying context switch.
1128      */
1129     if (cx_next->type == ACPI_STATE_C1) {
1130 	start_ticks = cpu_ticks();
1131 	if (cx_next->p_lvlx != NULL) {
1132 	    /* C1 I/O then Halt */
1133 	    CPU_GET_REG(cx_next->p_lvlx, 1);
1134 	}
1135 	if (cx_next->do_mwait)
1136 	    acpi_cpu_idle_mwait(cx_next->mwait_hint);
1137 	else
1138 	    acpi_cpu_c1();
1139 	end_ticks = cpu_ticks();
1140 	/* acpi_cpu_c1() returns with interrupts enabled. */
1141 	if (cx_next->do_mwait)
1142 	    ACPI_ENABLE_IRQS();
1143 	end_time = ((end_ticks - start_ticks) << 20) / cpu_tickrate();
1144 	if (!cx_next->do_mwait && curthread->td_critnest == 0)
1145 		end_time = min(end_time, 500000 / hz);
1146 	sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + end_time) / 4;
1147 	return;
1148     }
1149 
1150     /*
1151      * For C3, disable bus master arbitration and enable bus master wake
1152      * if BM control is available, otherwise flush the CPU cache.
1153      */
1154     if (cx_next->type == ACPI_STATE_C3) {
1155 	if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
1156 	    AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 1);
1157 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
1158 	} else
1159 	    ACPI_FLUSH_CPU_CACHE();
1160     }
1161 
1162     /*
1163      * Read from P_LVLx to enter C2(+), checking time spent asleep.
1164      * Use the ACPI timer for measuring sleep time.  Since we need to
1165      * get the time very close to the CPU start/stop clock logic, this
1166      * is the only reliable time source.
1167      */
1168     if (cx_next->type == ACPI_STATE_C3) {
1169 	AcpiGetTimer(&start_time);
1170 	start_ticks = 0;
1171     } else {
1172 	start_time = 0;
1173 	start_ticks = cpu_ticks();
1174     }
1175     if (cx_next->do_mwait) {
1176 	acpi_cpu_idle_mwait(cx_next->mwait_hint);
1177     } else {
1178 	CPU_GET_REG(cx_next->p_lvlx, 1);
1179 	/*
1180 	 * Read the end time twice.  Since it may take an arbitrary time
1181 	 * to enter the idle state, the first read may be executed before
1182 	 * the processor has stopped.  Doing it again provides enough
1183 	 * margin that we are certain to have a correct value.
1184 	 */
1185 	AcpiGetTimer(&end_time);
1186     }
1187 
1188     if (cx_next->type == ACPI_STATE_C3)
1189 	AcpiGetTimer(&end_time);
1190     else
1191 	end_ticks = cpu_ticks();
1192 
1193     /* Enable bus master arbitration and disable bus master wakeup. */
1194     if (cx_next->type == ACPI_STATE_C3 &&
1195       (cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
1196 	AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 0);
1197 	AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1198     }
1199     ACPI_ENABLE_IRQS();
1200 
1201     if (cx_next->type == ACPI_STATE_C3)
1202 	AcpiGetTimerDuration(start_time, end_time, &end_time);
1203     else
1204 	end_time = ((end_ticks - start_ticks) << 20) / cpu_tickrate();
1205     sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + end_time) / 4;
1206 }
1207 #endif
1208 
1209 /*
1210  * Re-evaluate the _CST object when we are notified that it changed.
1211  */
1212 static void
1213 acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context)
1214 {
1215     struct acpi_cpu_softc *sc = (struct acpi_cpu_softc *)context;
1216 
1217     if (notify != ACPI_NOTIFY_CX_STATES)
1218 	return;
1219 
1220     /*
1221      * C-state data for target CPU is going to be in flux while we execute
1222      * acpi_cpu_cx_cst, so disable entering acpi_cpu_idle.
1223      * Also, it may happen that multiple ACPI taskqueues may concurrently
1224      * execute notifications for the same CPU.  ACPI_SERIAL is used to
1225      * protect against that.
1226      */
1227     ACPI_SERIAL_BEGIN(cpu);
1228     disable_idle(sc);
1229 
1230     /* Update the list of Cx states. */
1231     acpi_cpu_cx_cst(sc);
1232     acpi_cpu_cx_list(sc);
1233     acpi_cpu_set_cx_lowest(sc);
1234 
1235     enable_idle(sc);
1236     ACPI_SERIAL_END(cpu);
1237 
1238     acpi_UserNotify("PROCESSOR", sc->cpu_handle, notify);
1239 }
1240 
1241 static void
1242 acpi_cpu_quirks(void)
1243 {
1244     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
1245 
1246     /*
1247      * Bus mastering arbitration control is needed to keep caches coherent
1248      * while sleeping in C3.  If it's not present but a working flush cache
1249      * instruction is present, flush the caches before entering C3 instead.
1250      * Otherwise, just disable C3 completely.
1251      */
1252     if (AcpiGbl_FADT.Pm2ControlBlock == 0 ||
1253 	AcpiGbl_FADT.Pm2ControlLength == 0) {
1254 	if ((AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD) &&
1255 	    (AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD_FLUSH) == 0) {
1256 	    cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1257 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1258 		"acpi_cpu: no BM control, using flush cache method\n"));
1259 	} else {
1260 	    cpu_quirks |= CPU_QUIRK_NO_C3;
1261 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1262 		"acpi_cpu: no BM control, C3 not available\n"));
1263 	}
1264     }
1265 
1266     /*
1267      * If we are using generic Cx mode, C3 on multiple CPUs requires using
1268      * the expensive flush cache instruction.
1269      */
1270     if (cpu_cx_generic && mp_ncpus > 1) {
1271 	cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1272 	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1273 	    "acpi_cpu: SMP, using flush cache mode for C3\n"));
1274     }
1275 
1276     /* Look for various quirks of the PIIX4 part. */
1277     acpi_cpu_quirks_piix4();
1278 }
1279 
1280 static void
1281 acpi_cpu_quirks_piix4(void)
1282 {
1283 #ifdef __i386__
1284     device_t acpi_dev;
1285     uint32_t val;
1286     ACPI_STATUS status;
1287 
1288     acpi_dev = pci_find_device(PCI_VENDOR_INTEL, PCI_DEVICE_82371AB_3);
1289     if (acpi_dev != NULL) {
1290 	switch (pci_get_revid(acpi_dev)) {
1291 	/*
1292 	 * Disable C3 support for all PIIX4 chipsets.  Some of these parts
1293 	 * do not report the BMIDE status to the BM status register and
1294 	 * others have a livelock bug if Type-F DMA is enabled.  Linux
1295 	 * works around the BMIDE bug by reading the BM status directly
1296 	 * but we take the simpler approach of disabling C3 for these
1297 	 * parts.
1298 	 *
1299 	 * See erratum #18 ("C3 Power State/BMIDE and Type-F DMA
1300 	 * Livelock") from the January 2002 PIIX4 specification update.
1301 	 * Applies to all PIIX4 models.
1302 	 *
1303 	 * Also, make sure that all interrupts cause a "Stop Break"
1304 	 * event to exit from C2 state.
1305 	 * Also, BRLD_EN_BM (ACPI_BITREG_BUS_MASTER_RLD in ACPI-speak)
1306 	 * should be set to zero, otherwise it causes C2 to short-sleep.
1307 	 * PIIX4 doesn't properly support C3 and bus master activity
1308 	 * need not break out of C2.
1309 	 */
1310 	case PCI_REVISION_A_STEP:
1311 	case PCI_REVISION_B_STEP:
1312 	case PCI_REVISION_4E:
1313 	case PCI_REVISION_4M:
1314 	    cpu_quirks |= CPU_QUIRK_NO_C3;
1315 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1316 		"acpi_cpu: working around PIIX4 bug, disabling C3\n"));
1317 
1318 	    val = pci_read_config(acpi_dev, PIIX4_DEVACTB_REG, 4);
1319 	    if ((val & PIIX4_STOP_BREAK_MASK) != PIIX4_STOP_BREAK_MASK) {
1320 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1321 		    "acpi_cpu: PIIX4: enabling IRQs to generate Stop Break\n"));
1322 	    	val |= PIIX4_STOP_BREAK_MASK;
1323 		pci_write_config(acpi_dev, PIIX4_DEVACTB_REG, val, 4);
1324 	    }
1325 	    status = AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_RLD, &val);
1326 	    if (ACPI_SUCCESS(status) && val != 0) {
1327 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1328 		    "acpi_cpu: PIIX4: reset BRLD_EN_BM\n"));
1329 		AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1330 	    }
1331 	    break;
1332 	default:
1333 	    break;
1334 	}
1335     }
1336 #endif
1337 }
1338 
1339 static int
1340 acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS)
1341 {
1342 	struct acpi_cpu_softc *sc = (struct acpi_cpu_softc *)arg1;
1343 	struct sbuf	 sb;
1344 	char		 buf[128];
1345 	int		 error, i;
1346 	uintmax_t	 fract, sum, whole;
1347 
1348 	sbuf_new_for_sysctl(&sb, buf, sizeof(buf), req);
1349 	sum = 0;
1350 	for (i = 0; i < sc->cpu_cx_count; i++)
1351 		sum += sc->cpu_cx_stats[i];
1352 	for (i = 0; i < sc->cpu_cx_count; i++) {
1353 		if (sum > 0) {
1354 			whole = (uintmax_t)sc->cpu_cx_stats[i] * 100;
1355 			fract = (whole % sum) * 100;
1356 			sbuf_printf(&sb, "%u.%02u%% ", (u_int)(whole / sum),
1357 			    (u_int)(fract / sum));
1358 		} else
1359 			sbuf_printf(&sb, "0.00%% ");
1360 	}
1361 	sbuf_printf(&sb, "last %dus", sc->cpu_prev_sleep);
1362 	error = sbuf_finish(&sb);
1363 	sbuf_delete(&sb);
1364 	return (error);
1365 }
1366 
1367 /*
1368  * XXX TODO: actually add support to count each entry/exit
1369  * from the Cx states.
1370  */
1371 static int
1372 acpi_cpu_usage_counters_sysctl(SYSCTL_HANDLER_ARGS)
1373 {
1374 	struct acpi_cpu_softc *sc = (struct acpi_cpu_softc *)arg1;
1375 	struct sbuf	 sb;
1376 	char		 buf[128];
1377 	int		 error, i;
1378 
1379 	sbuf_new_for_sysctl(&sb, buf, sizeof(buf), req);
1380 	for (i = 0; i < sc->cpu_cx_count; i++) {
1381 		if (i > 0)
1382 			sbuf_putc(&sb, ' ');
1383 		sbuf_printf(&sb, "%u", sc->cpu_cx_stats[i]);
1384 	}
1385 	error = sbuf_finish(&sb);
1386 	sbuf_delete(&sb);
1387 	return (error);
1388 }
1389 
1390 #if defined(__i386__) || defined(__amd64__)
1391 static int
1392 acpi_cpu_method_sysctl(SYSCTL_HANDLER_ARGS)
1393 {
1394 	struct acpi_cpu_softc *sc = (struct acpi_cpu_softc *)arg1;
1395 	struct acpi_cx *cx;
1396 	struct sbuf sb;
1397 	char buf[128];
1398 	int error, i;
1399 
1400 	sbuf_new_for_sysctl(&sb, buf, sizeof(buf), req);
1401 	for (i = 0; i < sc->cpu_cx_count; i++) {
1402 		cx = &sc->cpu_cx_states[i];
1403 		if (i > 0)
1404 			sbuf_putc(&sb, ' ');
1405 		sbuf_printf(&sb, "C%d/", i + 1);
1406 		if (cx->do_mwait) {
1407 			sbuf_cat(&sb, "mwait");
1408 			if (cx->mwait_hw_coord)
1409 				sbuf_cat(&sb, "/hwc");
1410 			if (cx->mwait_bm_avoidance)
1411 				sbuf_cat(&sb, "/bma");
1412 		} else if (cx->type == ACPI_STATE_C1) {
1413 			sbuf_cat(&sb, "hlt");
1414 		} else {
1415 			sbuf_cat(&sb, "io");
1416 		}
1417 		if (cx->type == ACPI_STATE_C1 && cx->p_lvlx != NULL)
1418 			sbuf_cat(&sb, "/iohlt");
1419 	}
1420 	error = sbuf_finish(&sb);
1421 	sbuf_delete(&sb);
1422 	return (error);
1423 }
1424 #endif
1425 
1426 static int
1427 acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc)
1428 {
1429     int i;
1430 
1431     ACPI_SERIAL_ASSERT(cpu);
1432     sc->cpu_cx_lowest = min(sc->cpu_cx_lowest_lim, sc->cpu_cx_count - 1);
1433 
1434     /* If not disabling, cache the new lowest non-C3 state. */
1435     sc->cpu_non_c3 = 0;
1436     for (i = sc->cpu_cx_lowest; i >= 0; i--) {
1437 	if (sc->cpu_cx_states[i].type < ACPI_STATE_C3) {
1438 	    sc->cpu_non_c3 = i;
1439 	    break;
1440 	}
1441     }
1442 
1443     /* Reset the statistics counters. */
1444     bzero(sc->cpu_cx_stats, sizeof(sc->cpu_cx_stats));
1445     return (0);
1446 }
1447 
1448 static int
1449 acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1450 {
1451     struct	 acpi_cpu_softc *sc;
1452     char	 state[8];
1453     int		 val, error;
1454 
1455     sc = (struct acpi_cpu_softc *) arg1;
1456     snprintf(state, sizeof(state), "C%d", sc->cpu_cx_lowest_lim + 1);
1457     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1458     if (error != 0 || req->newptr == NULL)
1459 	return (error);
1460     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1461 	return (EINVAL);
1462     if (strcasecmp(state, "Cmax") == 0)
1463 	val = MAX_CX_STATES;
1464     else {
1465 	val = (int) strtol(state + 1, NULL, 10);
1466 	if (val < 1 || val > MAX_CX_STATES)
1467 	    return (EINVAL);
1468     }
1469 
1470     ACPI_SERIAL_BEGIN(cpu);
1471     sc->cpu_cx_lowest_lim = val - 1;
1472     acpi_cpu_set_cx_lowest(sc);
1473     ACPI_SERIAL_END(cpu);
1474 
1475     return (0);
1476 }
1477 
1478 static int
1479 acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1480 {
1481     struct	acpi_cpu_softc *sc;
1482     char	state[8];
1483     int		val, error, i;
1484 
1485     snprintf(state, sizeof(state), "C%d", cpu_cx_lowest_lim + 1);
1486     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1487     if (error != 0 || req->newptr == NULL)
1488 	return (error);
1489     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1490 	return (EINVAL);
1491     if (strcasecmp(state, "Cmax") == 0)
1492 	val = MAX_CX_STATES;
1493     else {
1494 	val = (int) strtol(state + 1, NULL, 10);
1495 	if (val < 1 || val > MAX_CX_STATES)
1496 	    return (EINVAL);
1497     }
1498 
1499     /* Update the new lowest useable Cx state for all CPUs. */
1500     ACPI_SERIAL_BEGIN(cpu);
1501     cpu_cx_lowest_lim = val - 1;
1502     CPU_FOREACH(i) {
1503 	if ((sc = cpu_softc[i]) == NULL)
1504 	    continue;
1505 	sc->cpu_cx_lowest_lim = cpu_cx_lowest_lim;
1506 	acpi_cpu_set_cx_lowest(sc);
1507     }
1508     ACPI_SERIAL_END(cpu);
1509 
1510     return (0);
1511 }
1512