xref: /freebsd/sys/dev/acpica/acpi_cpu.c (revision 19e4f2f289fb8d2b14576a65ca9e6f931dbb6a02)
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 device_t		*cpu_devices;
158 static int		 cpu_ndevices;
159 static struct acpi_cpu_softc **cpu_softc;
160 ACPI_SERIAL_DECL(cpu, "ACPI CPU");
161 
162 static int	acpi_cpu_probe(device_t dev);
163 static int	acpi_cpu_attach(device_t dev);
164 static int	acpi_cpu_suspend(device_t dev);
165 static int	acpi_cpu_resume(device_t dev);
166 static int	acpi_pcpu_get_id(device_t dev, uint32_t acpi_id,
167 		    u_int *cpu_id);
168 static struct resource_list *acpi_cpu_get_rlist(device_t dev, device_t child);
169 static device_t	acpi_cpu_add_child(device_t dev, u_int order, const char *name,
170 		    int unit);
171 static int	acpi_cpu_read_ivar(device_t dev, device_t child, int index,
172 		    uintptr_t *result);
173 static int	acpi_cpu_shutdown(device_t dev);
174 static void	acpi_cpu_cx_probe(struct acpi_cpu_softc *sc);
175 static void	acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc);
176 static int	acpi_cpu_cx_cst(struct acpi_cpu_softc *sc);
177 static void	acpi_cpu_startup(void *arg);
178 static void	acpi_cpu_startup_cx(struct acpi_cpu_softc *sc);
179 static void	acpi_cpu_cx_list(struct acpi_cpu_softc *sc);
180 #if defined(__i386__) || defined(__amd64__)
181 static void	acpi_cpu_idle(sbintime_t sbt);
182 #endif
183 static void	acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context);
184 static void	acpi_cpu_quirks(void);
185 static void	acpi_cpu_quirks_piix4(void);
186 static int	acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS);
187 static int	acpi_cpu_usage_counters_sysctl(SYSCTL_HANDLER_ARGS);
188 static int	acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc);
189 static int	acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
190 static int	acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
191 #if defined(__i386__) || defined(__amd64__)
192 static int	acpi_cpu_method_sysctl(SYSCTL_HANDLER_ARGS);
193 #endif
194 
195 static device_method_t acpi_cpu_methods[] = {
196     /* Device interface */
197     DEVMETHOD(device_probe,	acpi_cpu_probe),
198     DEVMETHOD(device_attach,	acpi_cpu_attach),
199     DEVMETHOD(device_detach,	bus_generic_detach),
200     DEVMETHOD(device_shutdown,	acpi_cpu_shutdown),
201     DEVMETHOD(device_suspend,	acpi_cpu_suspend),
202     DEVMETHOD(device_resume,	acpi_cpu_resume),
203 
204     /* Bus interface */
205     DEVMETHOD(bus_add_child,	acpi_cpu_add_child),
206     DEVMETHOD(bus_read_ivar,	acpi_cpu_read_ivar),
207     DEVMETHOD(bus_get_resource_list, acpi_cpu_get_rlist),
208     DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
209     DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
210     DEVMETHOD(bus_alloc_resource, bus_generic_rl_alloc_resource),
211     DEVMETHOD(bus_release_resource, bus_generic_rl_release_resource),
212     DEVMETHOD(bus_activate_resource, bus_generic_activate_resource),
213     DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource),
214     DEVMETHOD(bus_setup_intr,	bus_generic_setup_intr),
215     DEVMETHOD(bus_teardown_intr, bus_generic_teardown_intr),
216 
217     DEVMETHOD_END
218 };
219 
220 static driver_t acpi_cpu_driver = {
221     "cpu",
222     acpi_cpu_methods,
223     sizeof(struct acpi_cpu_softc),
224 };
225 
226 static devclass_t acpi_cpu_devclass;
227 DRIVER_MODULE(cpu, acpi, acpi_cpu_driver, acpi_cpu_devclass, 0, 0);
228 MODULE_DEPEND(cpu, acpi, 1, 1, 1);
229 
230 static int
231 acpi_cpu_probe(device_t dev)
232 {
233     static char		   *cpudev_ids[] = { CPUDEV_DEVICE_ID, NULL };
234     int			   acpi_id, cpu_id;
235     ACPI_BUFFER		   buf;
236     ACPI_HANDLE		   handle;
237     ACPI_OBJECT		   *obj;
238     ACPI_STATUS		   status;
239     ACPI_OBJECT_TYPE	   type;
240 
241     if (acpi_disabled("cpu") || acpi_cpu_disabled)
242 	return (ENXIO);
243     type = acpi_get_type(dev);
244     if (type != ACPI_TYPE_PROCESSOR && type != ACPI_TYPE_DEVICE)
245 	return (ENXIO);
246     if (type == ACPI_TYPE_DEVICE &&
247 	ACPI_ID_PROBE(device_get_parent(dev), dev, cpudev_ids, NULL) >= 0)
248 	return (ENXIO);
249 
250     handle = acpi_get_handle(dev);
251     if (cpu_softc == NULL)
252 	cpu_softc = malloc(sizeof(struct acpi_cpu_softc *) *
253 	    (mp_maxid + 1), M_TEMP /* XXX */, M_WAITOK | M_ZERO);
254 
255     if (type == ACPI_TYPE_PROCESSOR) {
256 	/* Get our Processor object. */
257 	buf.Pointer = NULL;
258 	buf.Length = ACPI_ALLOCATE_BUFFER;
259 	status = AcpiEvaluateObject(handle, NULL, NULL, &buf);
260 	if (ACPI_FAILURE(status)) {
261 	    device_printf(dev, "probe failed to get Processor obj - %s\n",
262 		AcpiFormatException(status));
263 	    return (ENXIO);
264 	}
265 	obj = (ACPI_OBJECT *)buf.Pointer;
266 	if (obj->Type != ACPI_TYPE_PROCESSOR) {
267 	    device_printf(dev, "Processor object has bad type %d\n",
268 		obj->Type);
269 	    AcpiOsFree(obj);
270 	    return (ENXIO);
271 	}
272 
273 	/*
274 	 * Find the processor associated with our unit.  We could use the
275 	 * ProcId as a key, however, some boxes do not have the same values
276 	 * in their Processor object as the ProcId values in the MADT.
277 	 */
278 	acpi_id = obj->Processor.ProcId;
279 	AcpiOsFree(obj);
280     } else {
281 	status = acpi_GetInteger(handle, "_UID", &acpi_id);
282 	if (ACPI_FAILURE(status)) {
283 	    device_printf(dev, "Device object has bad value - %s\n",
284 		AcpiFormatException(status));
285 	    return (ENXIO);
286 	}
287     }
288     if (acpi_pcpu_get_id(dev, acpi_id, &cpu_id) != 0) {
289 	if (bootverbose && (type != ACPI_TYPE_PROCESSOR || acpi_id != 255))
290 	    printf("ACPI: Processor %s (ACPI ID %u) ignored\n",
291 		acpi_name(acpi_get_handle(dev)), acpi_id);
292 	return (ENXIO);
293     }
294 
295     if (device_set_unit(dev, cpu_id) != 0)
296 	return (ENXIO);
297 
298     device_set_desc(dev, "ACPI CPU");
299 
300     if (!bootverbose && device_get_unit(dev) != 0) {
301 	    device_quiet(dev);
302 	    device_quiet_children(dev);
303     }
304 
305     return (0);
306 }
307 
308 static int
309 acpi_cpu_attach(device_t dev)
310 {
311     ACPI_BUFFER		   buf;
312     ACPI_OBJECT		   arg, *obj;
313     ACPI_OBJECT_LIST	   arglist;
314     struct pcpu		   *pcpu_data;
315     struct acpi_cpu_softc *sc;
316     struct acpi_softc	  *acpi_sc;
317     ACPI_STATUS		   status;
318     u_int		   features;
319     int			   cpu_id, drv_count, i;
320     driver_t 		  **drivers;
321     uint32_t		   cap_set[3];
322 
323     /* UUID needed by _OSC evaluation */
324     static uint8_t cpu_oscuuid[16] = { 0x16, 0xA6, 0x77, 0x40, 0x0C, 0x29,
325 				       0xBE, 0x47, 0x9E, 0xBD, 0xD8, 0x70,
326 				       0x58, 0x71, 0x39, 0x53 };
327 
328     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
329 
330     sc = device_get_softc(dev);
331     sc->cpu_dev = dev;
332     sc->cpu_handle = acpi_get_handle(dev);
333     cpu_id = device_get_unit(dev);
334     cpu_softc[cpu_id] = sc;
335     pcpu_data = pcpu_find(cpu_id);
336     pcpu_data->pc_device = dev;
337     sc->cpu_pcpu = pcpu_data;
338     cpu_smi_cmd = AcpiGbl_FADT.SmiCommand;
339     cpu_cst_cnt = AcpiGbl_FADT.CstControl;
340 
341     if (acpi_get_type(dev) == ACPI_TYPE_PROCESSOR) {
342 	buf.Pointer = NULL;
343 	buf.Length = ACPI_ALLOCATE_BUFFER;
344 	status = AcpiEvaluateObject(sc->cpu_handle, NULL, NULL, &buf);
345 	if (ACPI_FAILURE(status)) {
346 	    device_printf(dev, "attach failed to get Processor obj - %s\n",
347 		AcpiFormatException(status));
348 	    return (ENXIO);
349 	}
350 	obj = (ACPI_OBJECT *)buf.Pointer;
351 	sc->cpu_p_blk = obj->Processor.PblkAddress;
352 	sc->cpu_p_blk_len = obj->Processor.PblkLength;
353 	sc->cpu_acpi_id = obj->Processor.ProcId;
354 	AcpiOsFree(obj);
355     } else {
356 	KASSERT(acpi_get_type(dev) == ACPI_TYPE_DEVICE,
357 	    ("Unexpected ACPI object"));
358 	status = acpi_GetInteger(sc->cpu_handle, "_UID", &sc->cpu_acpi_id);
359 	if (ACPI_FAILURE(status)) {
360 	    device_printf(dev, "Device object has bad value - %s\n",
361 		AcpiFormatException(status));
362 	    return (ENXIO);
363 	}
364 	sc->cpu_p_blk = 0;
365 	sc->cpu_p_blk_len = 0;
366     }
367     ACPI_DEBUG_PRINT((ACPI_DB_INFO, "acpi_cpu%d: P_BLK at %#x/%d\n",
368 		     device_get_unit(dev), sc->cpu_p_blk, sc->cpu_p_blk_len));
369 
370     /*
371      * If this is the first cpu we attach, create and initialize the generic
372      * resources that will be used by all acpi cpu devices.
373      */
374     if (device_get_unit(dev) == 0) {
375 	/* Assume we won't be using generic Cx mode by default */
376 	cpu_cx_generic = FALSE;
377 
378 	/* Install hw.acpi.cpu sysctl tree */
379 	acpi_sc = acpi_device_get_parent_softc(dev);
380 	sysctl_ctx_init(&cpu_sysctl_ctx);
381 	cpu_sysctl_tree = SYSCTL_ADD_NODE(&cpu_sysctl_ctx,
382 	    SYSCTL_CHILDREN(acpi_sc->acpi_sysctl_tree), OID_AUTO, "cpu",
383 	    CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "node for CPU children");
384     }
385 
386     /*
387      * Before calling any CPU methods, collect child driver feature hints
388      * and notify ACPI of them.  We support unified SMP power control
389      * so advertise this ourselves.  Note this is not the same as independent
390      * SMP control where each CPU can have different settings.
391      */
392     sc->cpu_features = ACPI_CAP_SMP_SAME | ACPI_CAP_SMP_SAME_C3 |
393       ACPI_CAP_C1_IO_HALT;
394 
395 #if defined(__i386__) || defined(__amd64__)
396     /*
397      * Ask for MWAIT modes if not disabled and interrupts work
398      * reasonable with MWAIT.
399      */
400     if (!acpi_disabled("mwait") && cpu_mwait_usable())
401 	sc->cpu_features |= ACPI_CAP_SMP_C1_NATIVE | ACPI_CAP_SMP_C3_NATIVE;
402 #endif
403 
404     if (devclass_get_drivers(acpi_cpu_devclass, &drivers, &drv_count) == 0) {
405 	for (i = 0; i < drv_count; i++) {
406 	    if (ACPI_GET_FEATURES(drivers[i], &features) == 0)
407 		sc->cpu_features |= features;
408 	}
409 	free(drivers, M_TEMP);
410     }
411 
412     /*
413      * CPU capabilities are specified in
414      * Intel Processor Vendor-Specific ACPI Interface Specification.
415      */
416     if (sc->cpu_features) {
417 	cap_set[1] = sc->cpu_features;
418 	status = acpi_EvaluateOSC(sc->cpu_handle, cpu_oscuuid, 1, 2, cap_set,
419 	    cap_set, false);
420 	if (ACPI_SUCCESS(status)) {
421 	    if (cap_set[0] != 0)
422 		device_printf(dev, "_OSC returned status %#x\n", cap_set[0]);
423 	}
424 	else {
425 	    arglist.Pointer = &arg;
426 	    arglist.Count = 1;
427 	    arg.Type = ACPI_TYPE_BUFFER;
428 	    arg.Buffer.Length = sizeof(cap_set);
429 	    arg.Buffer.Pointer = (uint8_t *)cap_set;
430 	    cap_set[0] = 1; /* revision */
431 	    cap_set[1] = 1; /* number of capabilities integers */
432 	    cap_set[2] = sc->cpu_features;
433 	    AcpiEvaluateObject(sc->cpu_handle, "_PDC", &arglist, NULL);
434 	}
435     }
436 
437     /* Probe for Cx state support. */
438     acpi_cpu_cx_probe(sc);
439 
440     return (0);
441 }
442 
443 static void
444 acpi_cpu_postattach(void *unused __unused)
445 {
446     device_t *devices;
447     int err;
448     int i, n;
449     int attached;
450 
451     err = devclass_get_devices(acpi_cpu_devclass, &devices, &n);
452     if (err != 0) {
453 	printf("devclass_get_devices(acpi_cpu_devclass) failed\n");
454 	return;
455     }
456     attached = 0;
457     for (i = 0; i < n; i++)
458 	if (device_is_attached(devices[i]) &&
459 	    device_get_driver(devices[i]) == &acpi_cpu_driver)
460 	    attached = 1;
461     for (i = 0; i < n; i++)
462 	bus_generic_probe(devices[i]);
463     for (i = 0; i < n; i++)
464 	bus_generic_attach(devices[i]);
465     free(devices, M_TEMP);
466 
467     if (attached) {
468 #ifdef EARLY_AP_STARTUP
469 	acpi_cpu_startup(NULL);
470 #else
471 	/* Queue post cpu-probing task handler */
472 	AcpiOsExecute(OSL_NOTIFY_HANDLER, acpi_cpu_startup, NULL);
473 #endif
474     }
475 }
476 
477 SYSINIT(acpi_cpu, SI_SUB_CONFIGURE, SI_ORDER_MIDDLE,
478     acpi_cpu_postattach, NULL);
479 
480 static void
481 disable_idle(struct acpi_cpu_softc *sc)
482 {
483     cpuset_t cpuset;
484 
485     CPU_SETOF(sc->cpu_pcpu->pc_cpuid, &cpuset);
486     sc->cpu_disable_idle = TRUE;
487 
488     /*
489      * Ensure that the CPU is not in idle state or in acpi_cpu_idle().
490      * Note that this code depends on the fact that the rendezvous IPI
491      * can not penetrate context where interrupts are disabled and acpi_cpu_idle
492      * is called and executed in such a context with interrupts being re-enabled
493      * right before return.
494      */
495     smp_rendezvous_cpus(cpuset, smp_no_rendezvous_barrier, NULL,
496 	smp_no_rendezvous_barrier, NULL);
497 }
498 
499 static void
500 enable_idle(struct acpi_cpu_softc *sc)
501 {
502 
503     sc->cpu_disable_idle = FALSE;
504 }
505 
506 #if defined(__i386__) || defined(__amd64__)
507 static int
508 is_idle_disabled(struct acpi_cpu_softc *sc)
509 {
510 
511     return (sc->cpu_disable_idle);
512 }
513 #endif
514 
515 /*
516  * Disable any entry to the idle function during suspend and re-enable it
517  * during resume.
518  */
519 static int
520 acpi_cpu_suspend(device_t dev)
521 {
522     int error;
523 
524     error = bus_generic_suspend(dev);
525     if (error)
526 	return (error);
527     disable_idle(device_get_softc(dev));
528     return (0);
529 }
530 
531 static int
532 acpi_cpu_resume(device_t dev)
533 {
534 
535     enable_idle(device_get_softc(dev));
536     return (bus_generic_resume(dev));
537 }
538 
539 /*
540  * Find the processor associated with a given ACPI ID.
541  */
542 static int
543 acpi_pcpu_get_id(device_t dev, uint32_t acpi_id, u_int *cpu_id)
544 {
545     struct pcpu	*pc;
546     u_int	 i;
547 
548     CPU_FOREACH(i) {
549 	pc = pcpu_find(i);
550 	if (pc->pc_acpi_id == acpi_id) {
551 	    *cpu_id = pc->pc_cpuid;
552 	    return (0);
553 	}
554     }
555 
556     /*
557      * If pc_acpi_id for CPU 0 is not initialized (e.g. a non-APIC
558      * UP box) use the ACPI ID from the first processor we find.
559      */
560     if (mp_ncpus == 1) {
561 	pc = pcpu_find(0);
562 	if (pc->pc_acpi_id == 0xffffffff)
563 	    pc->pc_acpi_id = acpi_id;
564 	*cpu_id = 0;
565 	return (0);
566     }
567 
568     return (ESRCH);
569 }
570 
571 static struct resource_list *
572 acpi_cpu_get_rlist(device_t dev, device_t child)
573 {
574     struct acpi_cpu_device *ad;
575 
576     ad = device_get_ivars(child);
577     if (ad == NULL)
578 	return (NULL);
579     return (&ad->ad_rl);
580 }
581 
582 static device_t
583 acpi_cpu_add_child(device_t dev, u_int order, const char *name, int unit)
584 {
585     struct acpi_cpu_device *ad;
586     device_t child;
587 
588     if ((ad = malloc(sizeof(*ad), M_TEMP, M_NOWAIT | M_ZERO)) == NULL)
589 	return (NULL);
590 
591     resource_list_init(&ad->ad_rl);
592 
593     child = device_add_child_ordered(dev, order, name, unit);
594     if (child != NULL)
595 	device_set_ivars(child, ad);
596     else
597 	free(ad, M_TEMP);
598     return (child);
599 }
600 
601 static int
602 acpi_cpu_read_ivar(device_t dev, device_t child, int index, uintptr_t *result)
603 {
604     struct acpi_cpu_softc *sc;
605 
606     sc = device_get_softc(dev);
607     switch (index) {
608     case ACPI_IVAR_HANDLE:
609 	*result = (uintptr_t)sc->cpu_handle;
610 	break;
611     case CPU_IVAR_PCPU:
612 	*result = (uintptr_t)sc->cpu_pcpu;
613 	break;
614 #if defined(__amd64__) || defined(__i386__)
615     case CPU_IVAR_NOMINAL_MHZ:
616 	if (tsc_is_invariant) {
617 	    *result = (uintptr_t)(atomic_load_acq_64(&tsc_freq) / 1000000);
618 	    break;
619 	}
620 	/* FALLTHROUGH */
621 #endif
622     default:
623 	return (ENOENT);
624     }
625     return (0);
626 }
627 
628 static int
629 acpi_cpu_shutdown(device_t dev)
630 {
631     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
632 
633     /* Allow children to shutdown first. */
634     bus_generic_shutdown(dev);
635 
636     /*
637      * Disable any entry to the idle function.
638      */
639     disable_idle(device_get_softc(dev));
640 
641     /*
642      * CPU devices are not truly detached and remain referenced,
643      * so their resources are not freed.
644      */
645 
646     return_VALUE (0);
647 }
648 
649 static void
650 acpi_cpu_cx_probe(struct acpi_cpu_softc *sc)
651 {
652     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
653 
654     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
655     sc->cpu_prev_sleep = 1000000;
656     sc->cpu_cx_lowest = 0;
657     sc->cpu_cx_lowest_lim = 0;
658 
659     /*
660      * Check for the ACPI 2.0 _CST sleep states object. If we can't find
661      * any, we'll revert to generic FADT/P_BLK Cx control method which will
662      * be handled by acpi_cpu_startup. We need to defer to after having
663      * probed all the cpus in the system before probing for generic Cx
664      * states as we may already have found cpus with valid _CST packages
665      */
666     if (!cpu_cx_generic && acpi_cpu_cx_cst(sc) != 0) {
667 	/*
668 	 * We were unable to find a _CST package for this cpu or there
669 	 * was an error parsing it. Switch back to generic mode.
670 	 */
671 	cpu_cx_generic = TRUE;
672 	if (bootverbose)
673 	    device_printf(sc->cpu_dev, "switching to generic Cx mode\n");
674     }
675 
676     /*
677      * TODO: _CSD Package should be checked here.
678      */
679 }
680 
681 static void
682 acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc)
683 {
684     ACPI_GENERIC_ADDRESS	 gas;
685     struct acpi_cx		*cx_ptr;
686 
687     sc->cpu_cx_count = 0;
688     cx_ptr = sc->cpu_cx_states;
689 
690     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
691     sc->cpu_prev_sleep = 1000000;
692 
693     /* C1 has been required since just after ACPI 1.0 */
694     cx_ptr->type = ACPI_STATE_C1;
695     cx_ptr->trans_lat = 0;
696     cx_ptr++;
697     sc->cpu_non_c2 = sc->cpu_cx_count;
698     sc->cpu_non_c3 = sc->cpu_cx_count;
699     sc->cpu_cx_count++;
700 
701     /*
702      * The spec says P_BLK must be 6 bytes long.  However, some systems
703      * use it to indicate a fractional set of features present so we
704      * take 5 as C2.  Some may also have a value of 7 to indicate
705      * another C3 but most use _CST for this (as required) and having
706      * "only" C1-C3 is not a hardship.
707      */
708     if (sc->cpu_p_blk_len < 5)
709 	return;
710 
711     /* Validate and allocate resources for C2 (P_LVL2). */
712     gas.SpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
713     gas.BitWidth = 8;
714     if (AcpiGbl_FADT.C2Latency <= 100) {
715 	gas.Address = sc->cpu_p_blk + 4;
716 	cx_ptr->res_rid = 0;
717 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &cx_ptr->res_rid,
718 	    &gas, &cx_ptr->p_lvlx, RF_SHAREABLE);
719 	if (cx_ptr->p_lvlx != NULL) {
720 	    cx_ptr->type = ACPI_STATE_C2;
721 	    cx_ptr->trans_lat = AcpiGbl_FADT.C2Latency;
722 	    cx_ptr++;
723 	    sc->cpu_non_c3 = sc->cpu_cx_count;
724 	    sc->cpu_cx_count++;
725 	}
726     }
727     if (sc->cpu_p_blk_len < 6)
728 	return;
729 
730     /* Validate and allocate resources for C3 (P_LVL3). */
731     if (AcpiGbl_FADT.C3Latency <= 1000 && !(cpu_quirks & CPU_QUIRK_NO_C3)) {
732 	gas.Address = sc->cpu_p_blk + 5;
733 	cx_ptr->res_rid = 1;
734 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &cx_ptr->res_rid,
735 	    &gas, &cx_ptr->p_lvlx, RF_SHAREABLE);
736 	if (cx_ptr->p_lvlx != NULL) {
737 	    cx_ptr->type = ACPI_STATE_C3;
738 	    cx_ptr->trans_lat = AcpiGbl_FADT.C3Latency;
739 	    cx_ptr++;
740 	    sc->cpu_cx_count++;
741 	}
742     }
743 }
744 
745 #if defined(__i386__) || defined(__amd64__)
746 static void
747 acpi_cpu_cx_cst_mwait(struct acpi_cx *cx_ptr, uint64_t address, int accsize)
748 {
749 
750 	cx_ptr->do_mwait = true;
751 	cx_ptr->mwait_hint = address & 0xffffffff;
752 	cx_ptr->mwait_hw_coord = (accsize & CST_FFH_MWAIT_HW_COORD) != 0;
753 	cx_ptr->mwait_bm_avoidance = (accsize & CST_FFH_MWAIT_BM_AVOID) != 0;
754 }
755 #endif
756 
757 static void
758 acpi_cpu_cx_cst_free_plvlx(device_t cpu_dev, struct acpi_cx *cx_ptr)
759 {
760 
761 	if (cx_ptr->p_lvlx == NULL)
762 		return;
763 	bus_release_resource(cpu_dev, cx_ptr->res_type, cx_ptr->res_rid,
764 	    cx_ptr->p_lvlx);
765 	cx_ptr->p_lvlx = NULL;
766 }
767 
768 /*
769  * Parse a _CST package and set up its Cx states.  Since the _CST object
770  * can change dynamically, our notify handler may call this function
771  * to clean up and probe the new _CST package.
772  */
773 static int
774 acpi_cpu_cx_cst(struct acpi_cpu_softc *sc)
775 {
776     struct	 acpi_cx *cx_ptr;
777     ACPI_STATUS	 status;
778     ACPI_BUFFER	 buf;
779     ACPI_OBJECT	*top;
780     ACPI_OBJECT	*pkg;
781     uint32_t	 count;
782     int		 i;
783 #if defined(__i386__) || defined(__amd64__)
784     uint64_t	 address;
785     int		 vendor, class, accsize;
786 #endif
787 
788     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
789 
790     buf.Pointer = NULL;
791     buf.Length = ACPI_ALLOCATE_BUFFER;
792     status = AcpiEvaluateObject(sc->cpu_handle, "_CST", NULL, &buf);
793     if (ACPI_FAILURE(status))
794 	return (ENXIO);
795 
796     /* _CST is a package with a count and at least one Cx package. */
797     top = (ACPI_OBJECT *)buf.Pointer;
798     if (!ACPI_PKG_VALID(top, 2) || acpi_PkgInt32(top, 0, &count) != 0) {
799 	device_printf(sc->cpu_dev, "invalid _CST package\n");
800 	AcpiOsFree(buf.Pointer);
801 	return (ENXIO);
802     }
803     if (count != top->Package.Count - 1) {
804 	device_printf(sc->cpu_dev, "invalid _CST state count (%d != %d)\n",
805 	       count, top->Package.Count - 1);
806 	count = top->Package.Count - 1;
807     }
808     if (count > MAX_CX_STATES) {
809 	device_printf(sc->cpu_dev, "_CST has too many states (%d)\n", count);
810 	count = MAX_CX_STATES;
811     }
812 
813     sc->cpu_non_c2 = 0;
814     sc->cpu_non_c3 = 0;
815     sc->cpu_cx_count = 0;
816     cx_ptr = sc->cpu_cx_states;
817 
818     /*
819      * C1 has been required since just after ACPI 1.0.
820      * Reserve the first slot for it.
821      */
822     cx_ptr->type = ACPI_STATE_C0;
823     cx_ptr++;
824     sc->cpu_cx_count++;
825 
826     /* Set up all valid states. */
827     for (i = 0; i < count; i++) {
828 	pkg = &top->Package.Elements[i + 1];
829 	if (!ACPI_PKG_VALID(pkg, 4) ||
830 	    acpi_PkgInt32(pkg, 1, &cx_ptr->type) != 0 ||
831 	    acpi_PkgInt32(pkg, 2, &cx_ptr->trans_lat) != 0 ||
832 	    acpi_PkgInt32(pkg, 3, &cx_ptr->power) != 0) {
833 	    device_printf(sc->cpu_dev, "skipping invalid Cx state package\n");
834 	    continue;
835 	}
836 
837 	/* Validate the state to see if we should use it. */
838 	switch (cx_ptr->type) {
839 	case ACPI_STATE_C1:
840 	    acpi_cpu_cx_cst_free_plvlx(sc->cpu_dev, cx_ptr);
841 #if defined(__i386__) || defined(__amd64__)
842 	    if (acpi_PkgFFH_IntelCpu(pkg, 0, &vendor, &class, &address,
843 	      &accsize) == 0 && vendor == CST_FFH_VENDOR_INTEL) {
844 		if (class == CST_FFH_INTEL_CL_C1IO) {
845 		    /* C1 I/O then Halt */
846 		    cx_ptr->res_rid = sc->cpu_cx_count;
847 		    bus_set_resource(sc->cpu_dev, SYS_RES_IOPORT,
848 		      cx_ptr->res_rid, address, 1);
849 		    cx_ptr->p_lvlx = bus_alloc_resource_any(sc->cpu_dev,
850 		      SYS_RES_IOPORT, &cx_ptr->res_rid, RF_ACTIVE |
851 		      RF_SHAREABLE);
852 		    if (cx_ptr->p_lvlx == NULL) {
853 			bus_delete_resource(sc->cpu_dev, SYS_RES_IOPORT,
854 			  cx_ptr->res_rid);
855 			device_printf(sc->cpu_dev,
856 			  "C1 I/O failed to allocate port %d, "
857 			  "degrading to C1 Halt", (int)address);
858 		    }
859 		} else if (class == CST_FFH_INTEL_CL_MWAIT) {
860 		    acpi_cpu_cx_cst_mwait(cx_ptr, address, accsize);
861 		}
862 	    }
863 #endif
864 	    if (sc->cpu_cx_states[0].type == ACPI_STATE_C0) {
865 		/* This is the first C1 state.  Use the reserved slot. */
866 		sc->cpu_cx_states[0] = *cx_ptr;
867 	    } else {
868 		sc->cpu_non_c2 = sc->cpu_cx_count;
869 		sc->cpu_non_c3 = sc->cpu_cx_count;
870 		cx_ptr++;
871 		sc->cpu_cx_count++;
872 	    }
873 	    continue;
874 	case ACPI_STATE_C2:
875 	    sc->cpu_non_c3 = sc->cpu_cx_count;
876 	    break;
877 	case ACPI_STATE_C3:
878 	default:
879 	    if ((cpu_quirks & CPU_QUIRK_NO_C3) != 0) {
880 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
881 				 "acpi_cpu%d: C3[%d] not available.\n",
882 				 device_get_unit(sc->cpu_dev), i));
883 		continue;
884 	    }
885 	    break;
886 	}
887 
888 	/* Free up any previous register. */
889 	acpi_cpu_cx_cst_free_plvlx(sc->cpu_dev, cx_ptr);
890 
891 	/* Allocate the control register for C2 or C3. */
892 #if defined(__i386__) || defined(__amd64__)
893 	if (acpi_PkgFFH_IntelCpu(pkg, 0, &vendor, &class, &address,
894 	  &accsize) == 0 && vendor == CST_FFH_VENDOR_INTEL &&
895 	  class == CST_FFH_INTEL_CL_MWAIT) {
896 	    /* Native C State Instruction use (mwait) */
897 	    acpi_cpu_cx_cst_mwait(cx_ptr, address, accsize);
898 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
899 	      "acpi_cpu%d: Got C%d/mwait - %d latency\n",
900 	      device_get_unit(sc->cpu_dev), cx_ptr->type, cx_ptr->trans_lat));
901 	    cx_ptr++;
902 	    sc->cpu_cx_count++;
903 	} else
904 #endif
905 	{
906 	    cx_ptr->res_rid = sc->cpu_cx_count;
907 	    acpi_PkgGas(sc->cpu_dev, pkg, 0, &cx_ptr->res_type,
908 		&cx_ptr->res_rid, &cx_ptr->p_lvlx, RF_SHAREABLE);
909 	    if (cx_ptr->p_lvlx) {
910 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
911 		     "acpi_cpu%d: Got C%d - %d latency\n",
912 		     device_get_unit(sc->cpu_dev), cx_ptr->type,
913 		     cx_ptr->trans_lat));
914 		cx_ptr++;
915 		sc->cpu_cx_count++;
916 	    }
917 	}
918     }
919     AcpiOsFree(buf.Pointer);
920 
921     /* If C1 state was not found, we need one now. */
922     cx_ptr = sc->cpu_cx_states;
923     if (cx_ptr->type == ACPI_STATE_C0) {
924 	cx_ptr->type = ACPI_STATE_C1;
925 	cx_ptr->trans_lat = 0;
926     }
927 
928     return (0);
929 }
930 
931 /*
932  * Call this *after* all CPUs have been attached.
933  */
934 static void
935 acpi_cpu_startup(void *arg)
936 {
937     struct acpi_cpu_softc *sc;
938     int i;
939 
940     /* Get set of CPU devices */
941     devclass_get_devices(acpi_cpu_devclass, &cpu_devices, &cpu_ndevices);
942 
943     /*
944      * Setup any quirks that might necessary now that we have probed
945      * all the CPUs
946      */
947     acpi_cpu_quirks();
948 
949     if (cpu_cx_generic) {
950 	/*
951 	 * We are using generic Cx mode, probe for available Cx states
952 	 * for all processors.
953 	 */
954 	for (i = 0; i < cpu_ndevices; i++) {
955 	    sc = device_get_softc(cpu_devices[i]);
956 	    acpi_cpu_generic_cx_probe(sc);
957 	}
958     } else {
959 	/*
960 	 * We are using _CST mode, remove C3 state if necessary.
961 	 * As we now know for sure that we will be using _CST mode
962 	 * install our notify handler.
963 	 */
964 	for (i = 0; i < cpu_ndevices; i++) {
965 	    sc = device_get_softc(cpu_devices[i]);
966 	    if (cpu_quirks & CPU_QUIRK_NO_C3) {
967 		sc->cpu_cx_count = min(sc->cpu_cx_count, sc->cpu_non_c3 + 1);
968 	    }
969 	    AcpiInstallNotifyHandler(sc->cpu_handle, ACPI_DEVICE_NOTIFY,
970 		acpi_cpu_notify, sc);
971 	}
972     }
973 
974     /* Perform Cx final initialization. */
975     for (i = 0; i < cpu_ndevices; i++) {
976 	sc = device_get_softc(cpu_devices[i]);
977 	acpi_cpu_startup_cx(sc);
978     }
979 
980     /* Add a sysctl handler to handle global Cx lowest setting */
981     SYSCTL_ADD_PROC(&cpu_sysctl_ctx, SYSCTL_CHILDREN(cpu_sysctl_tree),
982 	OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
983 	NULL, 0, acpi_cpu_global_cx_lowest_sysctl, "A",
984 	"Global lowest Cx sleep state to use");
985 
986     /* Take over idling from cpu_idle_default(). */
987     cpu_cx_lowest_lim = 0;
988     for (i = 0; i < cpu_ndevices; i++) {
989 	sc = device_get_softc(cpu_devices[i]);
990 	enable_idle(sc);
991     }
992 #if defined(__i386__) || defined(__amd64__)
993     cpu_idle_hook = acpi_cpu_idle;
994 #endif
995 }
996 
997 static void
998 acpi_cpu_cx_list(struct acpi_cpu_softc *sc)
999 {
1000     struct sbuf sb;
1001     int i;
1002 
1003     /*
1004      * Set up the list of Cx states
1005      */
1006     sbuf_new(&sb, sc->cpu_cx_supported, sizeof(sc->cpu_cx_supported),
1007 	SBUF_FIXEDLEN);
1008     for (i = 0; i < sc->cpu_cx_count; i++)
1009 	sbuf_printf(&sb, "C%d/%d/%d ", i + 1, sc->cpu_cx_states[i].type,
1010 	    sc->cpu_cx_states[i].trans_lat);
1011     sbuf_trim(&sb);
1012     sbuf_finish(&sb);
1013 }
1014 
1015 static void
1016 acpi_cpu_startup_cx(struct acpi_cpu_softc *sc)
1017 {
1018     acpi_cpu_cx_list(sc);
1019 
1020     SYSCTL_ADD_STRING(&sc->cpu_sysctl_ctx,
1021 		      SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
1022 		      OID_AUTO, "cx_supported", CTLFLAG_RD,
1023 		      sc->cpu_cx_supported, 0,
1024 		      "Cx/microsecond values for supported Cx states");
1025     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
1026         SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)), OID_AUTO,
1027 	"cx_lowest", CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
1028 	(void *)sc, 0, acpi_cpu_cx_lowest_sysctl, "A",
1029 	"lowest Cx sleep state to use");
1030     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
1031         SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)), OID_AUTO,
1032 	"cx_usage", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
1033 	(void *)sc, 0, acpi_cpu_usage_sysctl, "A",
1034 	"percent usage for each Cx state");
1035     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
1036         SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)), OID_AUTO,
1037 	"cx_usage_counters", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
1038 	(void *)sc, 0, acpi_cpu_usage_counters_sysctl, "A",
1039 	"Cx sleep state counters");
1040 #if defined(__i386__) || defined(__amd64__)
1041     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
1042         SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)), OID_AUTO,
1043 	"cx_method", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
1044 	(void *)sc, 0, acpi_cpu_method_sysctl, "A", "Cx entrance methods");
1045 #endif
1046 
1047     /* Signal platform that we can handle _CST notification. */
1048     if (!cpu_cx_generic && cpu_cst_cnt != 0) {
1049 	ACPI_LOCK(acpi);
1050 	AcpiOsWritePort(cpu_smi_cmd, cpu_cst_cnt, 8);
1051 	ACPI_UNLOCK(acpi);
1052     }
1053 }
1054 
1055 #if defined(__i386__) || defined(__amd64__)
1056 /*
1057  * Idle the CPU in the lowest state possible.  This function is called with
1058  * interrupts disabled.  Note that once it re-enables interrupts, a task
1059  * switch can occur so do not access shared data (i.e. the softc) after
1060  * interrupts are re-enabled.
1061  */
1062 static void
1063 acpi_cpu_idle(sbintime_t sbt)
1064 {
1065     struct	acpi_cpu_softc *sc;
1066     struct	acpi_cx *cx_next;
1067     uint64_t	start_ticks, end_ticks;
1068     uint32_t	start_time, end_time;
1069     ACPI_STATUS	status;
1070     int		bm_active, cx_next_idx, i, us;
1071 
1072     /*
1073      * Look up our CPU id to get our softc.  If it's NULL, we'll use C1
1074      * since there is no ACPI processor object for this CPU.  This occurs
1075      * for logical CPUs in the HTT case.
1076      */
1077     sc = cpu_softc[PCPU_GET(cpuid)];
1078     if (sc == NULL) {
1079 	acpi_cpu_c1();
1080 	return;
1081     }
1082 
1083     /* If disabled, take the safe path. */
1084     if (is_idle_disabled(sc)) {
1085 	acpi_cpu_c1();
1086 	return;
1087     }
1088 
1089     /* Find the lowest state that has small enough latency. */
1090     us = sc->cpu_prev_sleep;
1091     if (sbt >= 0 && us > (sbt >> 12))
1092 	us = (sbt >> 12);
1093     cx_next_idx = 0;
1094     if (cpu_disable_c2_sleep)
1095 	i = min(sc->cpu_cx_lowest, sc->cpu_non_c2);
1096     else if (cpu_disable_c3_sleep)
1097 	i = min(sc->cpu_cx_lowest, sc->cpu_non_c3);
1098     else
1099 	i = sc->cpu_cx_lowest;
1100     for (; i >= 0; i--) {
1101 	if (sc->cpu_cx_states[i].trans_lat * 3 <= us) {
1102 	    cx_next_idx = i;
1103 	    break;
1104 	}
1105     }
1106 
1107     /*
1108      * Check for bus master activity.  If there was activity, clear
1109      * the bit and use the lowest non-C3 state.  Note that the USB
1110      * driver polling for new devices keeps this bit set all the
1111      * time if USB is loaded.
1112      */
1113     cx_next = &sc->cpu_cx_states[cx_next_idx];
1114     if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0 &&
1115 	cx_next_idx > sc->cpu_non_c3 &&
1116 	(!cx_next->do_mwait || cx_next->mwait_bm_avoidance)) {
1117 	status = AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, &bm_active);
1118 	if (ACPI_SUCCESS(status) && bm_active != 0) {
1119 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, 1);
1120 	    cx_next_idx = sc->cpu_non_c3;
1121 	    cx_next = &sc->cpu_cx_states[cx_next_idx];
1122 	}
1123     }
1124 
1125     /* Select the next state and update statistics. */
1126     sc->cpu_cx_stats[cx_next_idx]++;
1127     KASSERT(cx_next->type != ACPI_STATE_C0, ("acpi_cpu_idle: C0 sleep"));
1128 
1129     /*
1130      * Execute HLT (or equivalent) and wait for an interrupt.  We can't
1131      * precisely calculate the time spent in C1 since the place we wake up
1132      * is an ISR.  Assume we slept no more then half of quantum, unless
1133      * we are called inside critical section, delaying context switch.
1134      */
1135     if (cx_next->type == ACPI_STATE_C1) {
1136 	start_ticks = cpu_ticks();
1137 	if (cx_next->p_lvlx != NULL) {
1138 	    /* C1 I/O then Halt */
1139 	    CPU_GET_REG(cx_next->p_lvlx, 1);
1140 	}
1141 	if (cx_next->do_mwait)
1142 	    acpi_cpu_idle_mwait(cx_next->mwait_hint);
1143 	else
1144 	    acpi_cpu_c1();
1145 	end_ticks = cpu_ticks();
1146 	/* acpi_cpu_c1() returns with interrupts enabled. */
1147 	if (cx_next->do_mwait)
1148 	    ACPI_ENABLE_IRQS();
1149 	end_time = ((end_ticks - start_ticks) << 20) / cpu_tickrate();
1150 	if (!cx_next->do_mwait && curthread->td_critnest == 0)
1151 		end_time = min(end_time, 500000 / hz);
1152 	sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + end_time) / 4;
1153 	return;
1154     }
1155 
1156     /*
1157      * For C3, disable bus master arbitration and enable bus master wake
1158      * if BM control is available, otherwise flush the CPU cache.
1159      */
1160     if (cx_next->type == ACPI_STATE_C3) {
1161 	if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
1162 	    AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 1);
1163 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
1164 	} else
1165 	    ACPI_FLUSH_CPU_CACHE();
1166     }
1167 
1168     /*
1169      * Read from P_LVLx to enter C2(+), checking time spent asleep.
1170      * Use the ACPI timer for measuring sleep time.  Since we need to
1171      * get the time very close to the CPU start/stop clock logic, this
1172      * is the only reliable time source.
1173      */
1174     if (cx_next->type == ACPI_STATE_C3) {
1175 	AcpiGetTimer(&start_time);
1176 	start_ticks = 0;
1177     } else {
1178 	start_time = 0;
1179 	start_ticks = cpu_ticks();
1180     }
1181     if (cx_next->do_mwait) {
1182 	acpi_cpu_idle_mwait(cx_next->mwait_hint);
1183     } else {
1184 	CPU_GET_REG(cx_next->p_lvlx, 1);
1185 	/*
1186 	 * Read the end time twice.  Since it may take an arbitrary time
1187 	 * to enter the idle state, the first read may be executed before
1188 	 * the processor has stopped.  Doing it again provides enough
1189 	 * margin that we are certain to have a correct value.
1190 	 */
1191 	AcpiGetTimer(&end_time);
1192     }
1193 
1194     if (cx_next->type == ACPI_STATE_C3)
1195 	AcpiGetTimer(&end_time);
1196     else
1197 	end_ticks = cpu_ticks();
1198 
1199     /* Enable bus master arbitration and disable bus master wakeup. */
1200     if (cx_next->type == ACPI_STATE_C3 &&
1201       (cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
1202 	AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 0);
1203 	AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1204     }
1205     ACPI_ENABLE_IRQS();
1206 
1207     if (cx_next->type == ACPI_STATE_C3)
1208 	AcpiGetTimerDuration(start_time, end_time, &end_time);
1209     else
1210 	end_time = ((end_ticks - start_ticks) << 20) / cpu_tickrate();
1211     sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + end_time) / 4;
1212 }
1213 #endif
1214 
1215 /*
1216  * Re-evaluate the _CST object when we are notified that it changed.
1217  */
1218 static void
1219 acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context)
1220 {
1221     struct acpi_cpu_softc *sc = (struct acpi_cpu_softc *)context;
1222 
1223     if (notify != ACPI_NOTIFY_CX_STATES)
1224 	return;
1225 
1226     /*
1227      * C-state data for target CPU is going to be in flux while we execute
1228      * acpi_cpu_cx_cst, so disable entering acpi_cpu_idle.
1229      * Also, it may happen that multiple ACPI taskqueues may concurrently
1230      * execute notifications for the same CPU.  ACPI_SERIAL is used to
1231      * protect against that.
1232      */
1233     ACPI_SERIAL_BEGIN(cpu);
1234     disable_idle(sc);
1235 
1236     /* Update the list of Cx states. */
1237     acpi_cpu_cx_cst(sc);
1238     acpi_cpu_cx_list(sc);
1239     acpi_cpu_set_cx_lowest(sc);
1240 
1241     enable_idle(sc);
1242     ACPI_SERIAL_END(cpu);
1243 
1244     acpi_UserNotify("PROCESSOR", sc->cpu_handle, notify);
1245 }
1246 
1247 static void
1248 acpi_cpu_quirks(void)
1249 {
1250     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
1251 
1252     /*
1253      * Bus mastering arbitration control is needed to keep caches coherent
1254      * while sleeping in C3.  If it's not present but a working flush cache
1255      * instruction is present, flush the caches before entering C3 instead.
1256      * Otherwise, just disable C3 completely.
1257      */
1258     if (AcpiGbl_FADT.Pm2ControlBlock == 0 ||
1259 	AcpiGbl_FADT.Pm2ControlLength == 0) {
1260 	if ((AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD) &&
1261 	    (AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD_FLUSH) == 0) {
1262 	    cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1263 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1264 		"acpi_cpu: no BM control, using flush cache method\n"));
1265 	} else {
1266 	    cpu_quirks |= CPU_QUIRK_NO_C3;
1267 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1268 		"acpi_cpu: no BM control, C3 not available\n"));
1269 	}
1270     }
1271 
1272     /*
1273      * If we are using generic Cx mode, C3 on multiple CPUs requires using
1274      * the expensive flush cache instruction.
1275      */
1276     if (cpu_cx_generic && mp_ncpus > 1) {
1277 	cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1278 	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1279 	    "acpi_cpu: SMP, using flush cache mode for C3\n"));
1280     }
1281 
1282     /* Look for various quirks of the PIIX4 part. */
1283     acpi_cpu_quirks_piix4();
1284 }
1285 
1286 static void
1287 acpi_cpu_quirks_piix4(void)
1288 {
1289 #ifdef __i386__
1290     device_t acpi_dev;
1291     uint32_t val;
1292     ACPI_STATUS status;
1293 
1294     acpi_dev = pci_find_device(PCI_VENDOR_INTEL, PCI_DEVICE_82371AB_3);
1295     if (acpi_dev != NULL) {
1296 	switch (pci_get_revid(acpi_dev)) {
1297 	/*
1298 	 * Disable C3 support for all PIIX4 chipsets.  Some of these parts
1299 	 * do not report the BMIDE status to the BM status register and
1300 	 * others have a livelock bug if Type-F DMA is enabled.  Linux
1301 	 * works around the BMIDE bug by reading the BM status directly
1302 	 * but we take the simpler approach of disabling C3 for these
1303 	 * parts.
1304 	 *
1305 	 * See erratum #18 ("C3 Power State/BMIDE and Type-F DMA
1306 	 * Livelock") from the January 2002 PIIX4 specification update.
1307 	 * Applies to all PIIX4 models.
1308 	 *
1309 	 * Also, make sure that all interrupts cause a "Stop Break"
1310 	 * event to exit from C2 state.
1311 	 * Also, BRLD_EN_BM (ACPI_BITREG_BUS_MASTER_RLD in ACPI-speak)
1312 	 * should be set to zero, otherwise it causes C2 to short-sleep.
1313 	 * PIIX4 doesn't properly support C3 and bus master activity
1314 	 * need not break out of C2.
1315 	 */
1316 	case PCI_REVISION_A_STEP:
1317 	case PCI_REVISION_B_STEP:
1318 	case PCI_REVISION_4E:
1319 	case PCI_REVISION_4M:
1320 	    cpu_quirks |= CPU_QUIRK_NO_C3;
1321 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1322 		"acpi_cpu: working around PIIX4 bug, disabling C3\n"));
1323 
1324 	    val = pci_read_config(acpi_dev, PIIX4_DEVACTB_REG, 4);
1325 	    if ((val & PIIX4_STOP_BREAK_MASK) != PIIX4_STOP_BREAK_MASK) {
1326 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1327 		    "acpi_cpu: PIIX4: enabling IRQs to generate Stop Break\n"));
1328 	    	val |= PIIX4_STOP_BREAK_MASK;
1329 		pci_write_config(acpi_dev, PIIX4_DEVACTB_REG, val, 4);
1330 	    }
1331 	    status = AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_RLD, &val);
1332 	    if (ACPI_SUCCESS(status) && val != 0) {
1333 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1334 		    "acpi_cpu: PIIX4: reset BRLD_EN_BM\n"));
1335 		AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1336 	    }
1337 	    break;
1338 	default:
1339 	    break;
1340 	}
1341     }
1342 #endif
1343 }
1344 
1345 static int
1346 acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS)
1347 {
1348     struct acpi_cpu_softc *sc;
1349     struct sbuf	 sb;
1350     char	 buf[128];
1351     int		 i;
1352     uintmax_t	 fract, sum, whole;
1353 
1354     sc = (struct acpi_cpu_softc *) arg1;
1355     sum = 0;
1356     for (i = 0; i < sc->cpu_cx_count; i++)
1357 	sum += sc->cpu_cx_stats[i];
1358     sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1359     for (i = 0; i < sc->cpu_cx_count; i++) {
1360 	if (sum > 0) {
1361 	    whole = (uintmax_t)sc->cpu_cx_stats[i] * 100;
1362 	    fract = (whole % sum) * 100;
1363 	    sbuf_printf(&sb, "%u.%02u%% ", (u_int)(whole / sum),
1364 		(u_int)(fract / sum));
1365 	} else
1366 	    sbuf_printf(&sb, "0.00%% ");
1367     }
1368     sbuf_printf(&sb, "last %dus", sc->cpu_prev_sleep);
1369     sbuf_trim(&sb);
1370     sbuf_finish(&sb);
1371     sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1372     sbuf_delete(&sb);
1373 
1374     return (0);
1375 }
1376 
1377 /*
1378  * XXX TODO: actually add support to count each entry/exit
1379  * from the Cx states.
1380  */
1381 static int
1382 acpi_cpu_usage_counters_sysctl(SYSCTL_HANDLER_ARGS)
1383 {
1384     struct acpi_cpu_softc *sc;
1385     struct sbuf	 sb;
1386     char	 buf[128];
1387     int		 i;
1388 
1389     sc = (struct acpi_cpu_softc *) arg1;
1390 
1391     /* Print out the raw counters */
1392     sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1393 
1394     for (i = 0; i < sc->cpu_cx_count; i++) {
1395         sbuf_printf(&sb, "%u ", sc->cpu_cx_stats[i]);
1396     }
1397 
1398     sbuf_trim(&sb);
1399     sbuf_finish(&sb);
1400     sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1401     sbuf_delete(&sb);
1402 
1403     return (0);
1404 }
1405 
1406 #if defined(__i386__) || defined(__amd64__)
1407 static int
1408 acpi_cpu_method_sysctl(SYSCTL_HANDLER_ARGS)
1409 {
1410 	struct acpi_cpu_softc *sc;
1411 	struct acpi_cx *cx;
1412 	struct sbuf sb;
1413 	char buf[128];
1414 	int i;
1415 
1416 	sc = (struct acpi_cpu_softc *)arg1;
1417 	sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1418 	for (i = 0; i < sc->cpu_cx_count; i++) {
1419 		cx = &sc->cpu_cx_states[i];
1420 		sbuf_printf(&sb, "C%d/", i + 1);
1421 		if (cx->do_mwait) {
1422 			sbuf_cat(&sb, "mwait");
1423 			if (cx->mwait_hw_coord)
1424 				sbuf_cat(&sb, "/hwc");
1425 			if (cx->mwait_bm_avoidance)
1426 				sbuf_cat(&sb, "/bma");
1427 		} else if (cx->type == ACPI_STATE_C1) {
1428 			sbuf_cat(&sb, "hlt");
1429 		} else {
1430 			sbuf_cat(&sb, "io");
1431 		}
1432 		if (cx->type == ACPI_STATE_C1 && cx->p_lvlx != NULL)
1433 			sbuf_cat(&sb, "/iohlt");
1434 		sbuf_putc(&sb, ' ');
1435 	}
1436 	sbuf_trim(&sb);
1437 	sbuf_finish(&sb);
1438 	sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1439 	sbuf_delete(&sb);
1440 	return (0);
1441 }
1442 #endif
1443 
1444 static int
1445 acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc)
1446 {
1447     int i;
1448 
1449     ACPI_SERIAL_ASSERT(cpu);
1450     sc->cpu_cx_lowest = min(sc->cpu_cx_lowest_lim, sc->cpu_cx_count - 1);
1451 
1452     /* If not disabling, cache the new lowest non-C3 state. */
1453     sc->cpu_non_c3 = 0;
1454     for (i = sc->cpu_cx_lowest; i >= 0; i--) {
1455 	if (sc->cpu_cx_states[i].type < ACPI_STATE_C3) {
1456 	    sc->cpu_non_c3 = i;
1457 	    break;
1458 	}
1459     }
1460 
1461     /* Reset the statistics counters. */
1462     bzero(sc->cpu_cx_stats, sizeof(sc->cpu_cx_stats));
1463     return (0);
1464 }
1465 
1466 static int
1467 acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1468 {
1469     struct	 acpi_cpu_softc *sc;
1470     char	 state[8];
1471     int		 val, error;
1472 
1473     sc = (struct acpi_cpu_softc *) arg1;
1474     snprintf(state, sizeof(state), "C%d", sc->cpu_cx_lowest_lim + 1);
1475     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1476     if (error != 0 || req->newptr == NULL)
1477 	return (error);
1478     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1479 	return (EINVAL);
1480     if (strcasecmp(state, "Cmax") == 0)
1481 	val = MAX_CX_STATES;
1482     else {
1483 	val = (int) strtol(state + 1, NULL, 10);
1484 	if (val < 1 || val > MAX_CX_STATES)
1485 	    return (EINVAL);
1486     }
1487 
1488     ACPI_SERIAL_BEGIN(cpu);
1489     sc->cpu_cx_lowest_lim = val - 1;
1490     acpi_cpu_set_cx_lowest(sc);
1491     ACPI_SERIAL_END(cpu);
1492 
1493     return (0);
1494 }
1495 
1496 static int
1497 acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1498 {
1499     struct	acpi_cpu_softc *sc;
1500     char	state[8];
1501     int		val, error, i;
1502 
1503     snprintf(state, sizeof(state), "C%d", cpu_cx_lowest_lim + 1);
1504     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1505     if (error != 0 || req->newptr == NULL)
1506 	return (error);
1507     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1508 	return (EINVAL);
1509     if (strcasecmp(state, "Cmax") == 0)
1510 	val = MAX_CX_STATES;
1511     else {
1512 	val = (int) strtol(state + 1, NULL, 10);
1513 	if (val < 1 || val > MAX_CX_STATES)
1514 	    return (EINVAL);
1515     }
1516 
1517     /* Update the new lowest useable Cx state for all CPUs. */
1518     ACPI_SERIAL_BEGIN(cpu);
1519     cpu_cx_lowest_lim = val - 1;
1520     for (i = 0; i < cpu_ndevices; i++) {
1521 	sc = device_get_softc(cpu_devices[i]);
1522 	sc->cpu_cx_lowest_lim = cpu_cx_lowest_lim;
1523 	acpi_cpu_set_cx_lowest(sc);
1524     }
1525     ACPI_SERIAL_END(cpu);
1526 
1527     return (0);
1528 }
1529