xref: /freebsd/sys/dev/acpica/acpi_cpu.c (revision 93e779a26c651610ac6e7986d67ecc9ed2cadcbf)
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 PM_USEC(x)	 ((x) >> 2)	/* ~4 clocks per usec (3.57955 Mhz) */
118 
119 #define ACPI_NOTIFY_CX_STATES	0x81	/* _CST changed. */
120 
121 #define CPU_QUIRK_NO_C3		(1<<0)	/* C3-type states are not usable. */
122 #define CPU_QUIRK_NO_BM_CTRL	(1<<2)	/* No bus mastering control. */
123 
124 #define PCI_VENDOR_INTEL	0x8086
125 #define PCI_DEVICE_82371AB_3	0x7113	/* PIIX4 chipset for quirks. */
126 #define PCI_REVISION_A_STEP	0
127 #define PCI_REVISION_B_STEP	1
128 #define PCI_REVISION_4E		2
129 #define PCI_REVISION_4M		3
130 #define PIIX4_DEVACTB_REG	0x58
131 #define PIIX4_BRLD_EN_IRQ0	(1<<0)
132 #define PIIX4_BRLD_EN_IRQ	(1<<1)
133 #define PIIX4_BRLD_EN_IRQ8	(1<<5)
134 #define PIIX4_STOP_BREAK_MASK	(PIIX4_BRLD_EN_IRQ0 | PIIX4_BRLD_EN_IRQ | PIIX4_BRLD_EN_IRQ8)
135 #define PIIX4_PCNTRL_BST_EN	(1<<10)
136 
137 #define	CST_FFH_VENDOR_INTEL	1
138 #define	CST_FFH_INTEL_CL_C1IO	1
139 #define	CST_FFH_INTEL_CL_MWAIT	2
140 #define	CST_FFH_MWAIT_HW_COORD	0x0001
141 #define	CST_FFH_MWAIT_BM_AVOID	0x0002
142 
143 /* Allow users to ignore processor orders in MADT. */
144 static int cpu_unordered;
145 SYSCTL_INT(_debug_acpi, OID_AUTO, cpu_unordered, CTLFLAG_RDTUN,
146     &cpu_unordered, 0,
147     "Do not use the MADT to match ACPI Processor objects to CPUs.");
148 
149 /* Knob to disable acpi_cpu devices */
150 bool acpi_cpu_disabled = false;
151 
152 /* Platform hardware resource information. */
153 static uint32_t		 cpu_smi_cmd;	/* Value to write to SMI_CMD. */
154 static uint8_t		 cpu_cst_cnt;	/* Indicate we are _CST aware. */
155 static int		 cpu_quirks;	/* Indicate any hardware bugs. */
156 
157 /* Values for sysctl. */
158 static struct sysctl_ctx_list cpu_sysctl_ctx;
159 static struct sysctl_oid *cpu_sysctl_tree;
160 static int		 cpu_cx_generic;
161 static int		 cpu_cx_lowest_lim;
162 
163 static device_t		*cpu_devices;
164 static int		 cpu_ndevices;
165 static struct acpi_cpu_softc **cpu_softc;
166 ACPI_SERIAL_DECL(cpu, "ACPI CPU");
167 
168 static int	acpi_cpu_probe(device_t dev);
169 static int	acpi_cpu_attach(device_t dev);
170 static int	acpi_cpu_suspend(device_t dev);
171 static int	acpi_cpu_resume(device_t dev);
172 static int	acpi_pcpu_get_id(device_t dev, uint32_t *acpi_id,
173 		    uint32_t *cpu_id);
174 static struct resource_list *acpi_cpu_get_rlist(device_t dev, device_t child);
175 static device_t	acpi_cpu_add_child(device_t dev, u_int order, const char *name,
176 		    int unit);
177 static int	acpi_cpu_read_ivar(device_t dev, device_t child, int index,
178 		    uintptr_t *result);
179 static int	acpi_cpu_shutdown(device_t dev);
180 static void	acpi_cpu_cx_probe(struct acpi_cpu_softc *sc);
181 static void	acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc);
182 static int	acpi_cpu_cx_cst(struct acpi_cpu_softc *sc);
183 static void	acpi_cpu_startup(void *arg);
184 static void	acpi_cpu_startup_cx(struct acpi_cpu_softc *sc);
185 static void	acpi_cpu_cx_list(struct acpi_cpu_softc *sc);
186 #if defined(__i386__) || defined(__amd64__)
187 static void	acpi_cpu_idle(sbintime_t sbt);
188 #endif
189 static void	acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context);
190 static void	acpi_cpu_quirks(void);
191 static void	acpi_cpu_quirks_piix4(void);
192 static int	acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS);
193 static int	acpi_cpu_usage_counters_sysctl(SYSCTL_HANDLER_ARGS);
194 static int	acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc);
195 static int	acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
196 static int	acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
197 #if defined(__i386__) || defined(__amd64__)
198 static int	acpi_cpu_method_sysctl(SYSCTL_HANDLER_ARGS);
199 #endif
200 
201 static device_method_t acpi_cpu_methods[] = {
202     /* Device interface */
203     DEVMETHOD(device_probe,	acpi_cpu_probe),
204     DEVMETHOD(device_attach,	acpi_cpu_attach),
205     DEVMETHOD(device_detach,	bus_generic_detach),
206     DEVMETHOD(device_shutdown,	acpi_cpu_shutdown),
207     DEVMETHOD(device_suspend,	acpi_cpu_suspend),
208     DEVMETHOD(device_resume,	acpi_cpu_resume),
209 
210     /* Bus interface */
211     DEVMETHOD(bus_add_child,	acpi_cpu_add_child),
212     DEVMETHOD(bus_read_ivar,	acpi_cpu_read_ivar),
213     DEVMETHOD(bus_get_resource_list, acpi_cpu_get_rlist),
214     DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
215     DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
216     DEVMETHOD(bus_alloc_resource, bus_generic_rl_alloc_resource),
217     DEVMETHOD(bus_release_resource, bus_generic_rl_release_resource),
218     DEVMETHOD(bus_activate_resource, bus_generic_activate_resource),
219     DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource),
220     DEVMETHOD(bus_setup_intr,	bus_generic_setup_intr),
221     DEVMETHOD(bus_teardown_intr, bus_generic_teardown_intr),
222 
223     DEVMETHOD_END
224 };
225 
226 static driver_t acpi_cpu_driver = {
227     "cpu",
228     acpi_cpu_methods,
229     sizeof(struct acpi_cpu_softc),
230 };
231 
232 static devclass_t acpi_cpu_devclass;
233 DRIVER_MODULE(cpu, acpi, acpi_cpu_driver, acpi_cpu_devclass, 0, 0);
234 MODULE_DEPEND(cpu, acpi, 1, 1, 1);
235 
236 static int
237 acpi_cpu_probe(device_t dev)
238 {
239     int			   acpi_id, cpu_id;
240     ACPI_BUFFER		   buf;
241     ACPI_HANDLE		   handle;
242     ACPI_OBJECT		   *obj;
243     ACPI_STATUS		   status;
244 
245     if (acpi_disabled("cpu") || acpi_get_type(dev) != ACPI_TYPE_PROCESSOR ||
246 	    acpi_cpu_disabled)
247 	return (ENXIO);
248 
249     handle = acpi_get_handle(dev);
250     if (cpu_softc == NULL)
251 	cpu_softc = malloc(sizeof(struct acpi_cpu_softc *) *
252 	    (mp_maxid + 1), M_TEMP /* XXX */, M_WAITOK | M_ZERO);
253 
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", obj->Type);
266 	AcpiOsFree(obj);
267 	return (ENXIO);
268     }
269 
270     /*
271      * Find the processor associated with our unit.  We could use the
272      * ProcId as a key, however, some boxes do not have the same values
273      * in their Processor object as the ProcId values in the MADT.
274      */
275     acpi_id = obj->Processor.ProcId;
276     AcpiOsFree(obj);
277     if (acpi_pcpu_get_id(dev, &acpi_id, &cpu_id) != 0)
278 	return (ENXIO);
279 
280     /*
281      * Check if we already probed this processor.  We scan the bus twice
282      * so it's possible we've already seen this one.
283      */
284     if (cpu_softc[cpu_id] != NULL)
285 	return (ENXIO);
286 
287     /* Mark this processor as in-use and save our derived id for attach. */
288     cpu_softc[cpu_id] = (void *)1;
289     acpi_set_private(dev, (void*)(intptr_t)cpu_id);
290     device_set_desc(dev, "ACPI CPU");
291 
292     return (0);
293 }
294 
295 static int
296 acpi_cpu_attach(device_t dev)
297 {
298     ACPI_BUFFER		   buf;
299     ACPI_OBJECT		   arg[4], *obj;
300     ACPI_OBJECT_LIST	   arglist;
301     struct pcpu		   *pcpu_data;
302     struct acpi_cpu_softc *sc;
303     struct acpi_softc	  *acpi_sc;
304     ACPI_STATUS		   status;
305     u_int		   features;
306     int			   cpu_id, drv_count, i;
307     driver_t 		  **drivers;
308     uint32_t		   cap_set[3];
309 
310     /* UUID needed by _OSC evaluation */
311     static uint8_t cpu_oscuuid[16] = { 0x16, 0xA6, 0x77, 0x40, 0x0C, 0x29,
312 				       0xBE, 0x47, 0x9E, 0xBD, 0xD8, 0x70,
313 				       0x58, 0x71, 0x39, 0x53 };
314 
315     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
316 
317     sc = device_get_softc(dev);
318     sc->cpu_dev = dev;
319     sc->cpu_handle = acpi_get_handle(dev);
320     cpu_id = (int)(intptr_t)acpi_get_private(dev);
321     cpu_softc[cpu_id] = sc;
322     pcpu_data = pcpu_find(cpu_id);
323     pcpu_data->pc_device = dev;
324     sc->cpu_pcpu = pcpu_data;
325     cpu_smi_cmd = AcpiGbl_FADT.SmiCommand;
326     cpu_cst_cnt = AcpiGbl_FADT.CstControl;
327 
328     buf.Pointer = NULL;
329     buf.Length = ACPI_ALLOCATE_BUFFER;
330     status = AcpiEvaluateObject(sc->cpu_handle, NULL, NULL, &buf);
331     if (ACPI_FAILURE(status)) {
332 	device_printf(dev, "attach failed to get Processor obj - %s\n",
333 		      AcpiFormatException(status));
334 	return (ENXIO);
335     }
336     obj = (ACPI_OBJECT *)buf.Pointer;
337     sc->cpu_p_blk = obj->Processor.PblkAddress;
338     sc->cpu_p_blk_len = obj->Processor.PblkLength;
339     sc->cpu_acpi_id = obj->Processor.ProcId;
340     AcpiOsFree(obj);
341     ACPI_DEBUG_PRINT((ACPI_DB_INFO, "acpi_cpu%d: P_BLK at %#x/%d\n",
342 		     device_get_unit(dev), sc->cpu_p_blk, sc->cpu_p_blk_len));
343 
344     /*
345      * If this is the first cpu we attach, create and initialize the generic
346      * resources that will be used by all acpi cpu devices.
347      */
348     if (device_get_unit(dev) == 0) {
349 	/* Assume we won't be using generic Cx mode by default */
350 	cpu_cx_generic = FALSE;
351 
352 	/* Install hw.acpi.cpu sysctl tree */
353 	acpi_sc = acpi_device_get_parent_softc(dev);
354 	sysctl_ctx_init(&cpu_sysctl_ctx);
355 	cpu_sysctl_tree = SYSCTL_ADD_NODE(&cpu_sysctl_ctx,
356 	    SYSCTL_CHILDREN(acpi_sc->acpi_sysctl_tree), OID_AUTO, "cpu",
357 	    CTLFLAG_RD, 0, "node for CPU children");
358 
359 	/* Queue post cpu-probing task handler */
360 	AcpiOsExecute(OSL_NOTIFY_HANDLER, acpi_cpu_startup, NULL);
361     }
362 
363     /*
364      * Before calling any CPU methods, collect child driver feature hints
365      * and notify ACPI of them.  We support unified SMP power control
366      * so advertise this ourselves.  Note this is not the same as independent
367      * SMP control where each CPU can have different settings.
368      */
369     sc->cpu_features = ACPI_CAP_SMP_SAME | ACPI_CAP_SMP_SAME_C3 |
370       ACPI_CAP_C1_IO_HALT;
371 
372 #if defined(__i386__) || defined(__amd64__)
373     /*
374      * Ask for MWAIT modes if not disabled and interrupts work
375      * reasonable with MWAIT.
376      */
377     if (!acpi_disabled("mwait") && cpu_mwait_usable())
378 	sc->cpu_features |= ACPI_CAP_SMP_C1_NATIVE | ACPI_CAP_SMP_C3_NATIVE;
379 #endif
380 
381     if (devclass_get_drivers(acpi_cpu_devclass, &drivers, &drv_count) == 0) {
382 	for (i = 0; i < drv_count; i++) {
383 	    if (ACPI_GET_FEATURES(drivers[i], &features) == 0)
384 		sc->cpu_features |= features;
385 	}
386 	free(drivers, M_TEMP);
387     }
388 
389     /*
390      * CPU capabilities are specified in
391      * Intel Processor Vendor-Specific ACPI Interface Specification.
392      */
393     if (sc->cpu_features) {
394 	arglist.Pointer = arg;
395 	arglist.Count = 4;
396 	arg[0].Type = ACPI_TYPE_BUFFER;
397 	arg[0].Buffer.Length = sizeof(cpu_oscuuid);
398 	arg[0].Buffer.Pointer = cpu_oscuuid;	/* UUID */
399 	arg[1].Type = ACPI_TYPE_INTEGER;
400 	arg[1].Integer.Value = 1;		/* revision */
401 	arg[2].Type = ACPI_TYPE_INTEGER;
402 	arg[2].Integer.Value = 1;		/* count */
403 	arg[3].Type = ACPI_TYPE_BUFFER;
404 	arg[3].Buffer.Length = sizeof(cap_set);	/* Capabilities buffer */
405 	arg[3].Buffer.Pointer = (uint8_t *)cap_set;
406 	cap_set[0] = 0;				/* status */
407 	cap_set[1] = sc->cpu_features;
408 	status = AcpiEvaluateObject(sc->cpu_handle, "_OSC", &arglist, NULL);
409 	if (ACPI_SUCCESS(status)) {
410 	    if (cap_set[0] != 0)
411 		device_printf(dev, "_OSC returned status %#x\n", cap_set[0]);
412 	}
413 	else {
414 	    arglist.Pointer = arg;
415 	    arglist.Count = 1;
416 	    arg[0].Type = ACPI_TYPE_BUFFER;
417 	    arg[0].Buffer.Length = sizeof(cap_set);
418 	    arg[0].Buffer.Pointer = (uint8_t *)cap_set;
419 	    cap_set[0] = 1; /* revision */
420 	    cap_set[1] = 1; /* number of capabilities integers */
421 	    cap_set[2] = sc->cpu_features;
422 	    AcpiEvaluateObject(sc->cpu_handle, "_PDC", &arglist, NULL);
423 	}
424     }
425 
426     /* Probe for Cx state support. */
427     acpi_cpu_cx_probe(sc);
428 
429     return (0);
430 }
431 
432 static void
433 acpi_cpu_postattach(void *unused __unused)
434 {
435     device_t *devices;
436     int err;
437     int i, n;
438 
439     err = devclass_get_devices(acpi_cpu_devclass, &devices, &n);
440     if (err != 0) {
441 	printf("devclass_get_devices(acpi_cpu_devclass) failed\n");
442 	return;
443     }
444     for (i = 0; i < n; i++)
445 	bus_generic_probe(devices[i]);
446     for (i = 0; i < n; i++)
447 	bus_generic_attach(devices[i]);
448     free(devices, M_TEMP);
449 }
450 
451 SYSINIT(acpi_cpu, SI_SUB_CONFIGURE, SI_ORDER_MIDDLE,
452     acpi_cpu_postattach, NULL);
453 
454 static void
455 disable_idle(struct acpi_cpu_softc *sc)
456 {
457     cpuset_t cpuset;
458 
459     CPU_SETOF(sc->cpu_pcpu->pc_cpuid, &cpuset);
460     sc->cpu_disable_idle = TRUE;
461 
462     /*
463      * Ensure that the CPU is not in idle state or in acpi_cpu_idle().
464      * Note that this code depends on the fact that the rendezvous IPI
465      * can not penetrate context where interrupts are disabled and acpi_cpu_idle
466      * is called and executed in such a context with interrupts being re-enabled
467      * right before return.
468      */
469     smp_rendezvous_cpus(cpuset, smp_no_rendevous_barrier, NULL,
470 	smp_no_rendevous_barrier, NULL);
471 }
472 
473 static void
474 enable_idle(struct acpi_cpu_softc *sc)
475 {
476 
477     sc->cpu_disable_idle = FALSE;
478 }
479 
480 #if defined(__i386__) || defined(__amd64__)
481 static int
482 is_idle_disabled(struct acpi_cpu_softc *sc)
483 {
484 
485     return (sc->cpu_disable_idle);
486 }
487 #endif
488 
489 /*
490  * Disable any entry to the idle function during suspend and re-enable it
491  * during resume.
492  */
493 static int
494 acpi_cpu_suspend(device_t dev)
495 {
496     int error;
497 
498     error = bus_generic_suspend(dev);
499     if (error)
500 	return (error);
501     disable_idle(device_get_softc(dev));
502     return (0);
503 }
504 
505 static int
506 acpi_cpu_resume(device_t dev)
507 {
508 
509     enable_idle(device_get_softc(dev));
510     return (bus_generic_resume(dev));
511 }
512 
513 /*
514  * Find the processor associated with a given ACPI ID.  By default,
515  * use the MADT to map ACPI IDs to APIC IDs and use that to locate a
516  * processor.  Some systems have inconsistent ASL and MADT however.
517  * For these systems the cpu_unordered tunable can be set in which
518  * case we assume that Processor objects are listed in the same order
519  * in both the MADT and ASL.
520  */
521 static int
522 acpi_pcpu_get_id(device_t dev, uint32_t *acpi_id, uint32_t *cpu_id)
523 {
524     struct pcpu	*pc;
525     uint32_t	 i, idx;
526 
527     KASSERT(acpi_id != NULL, ("Null acpi_id"));
528     KASSERT(cpu_id != NULL, ("Null cpu_id"));
529     idx = device_get_unit(dev);
530 
531     /*
532      * If pc_acpi_id for CPU 0 is not initialized (e.g. a non-APIC
533      * UP box) use the ACPI ID from the first processor we find.
534      */
535     if (idx == 0 && mp_ncpus == 1) {
536 	pc = pcpu_find(0);
537 	if (pc->pc_acpi_id == 0xffffffff)
538 	    pc->pc_acpi_id = *acpi_id;
539 	*cpu_id = 0;
540 	return (0);
541     }
542 
543     CPU_FOREACH(i) {
544 	pc = pcpu_find(i);
545 	KASSERT(pc != NULL, ("no pcpu data for %d", i));
546 	if (cpu_unordered) {
547 	    if (idx-- == 0) {
548 		/*
549 		 * If pc_acpi_id doesn't match the ACPI ID from the
550 		 * ASL, prefer the MADT-derived value.
551 		 */
552 		if (pc->pc_acpi_id != *acpi_id)
553 		    *acpi_id = pc->pc_acpi_id;
554 		*cpu_id = pc->pc_cpuid;
555 		return (0);
556 	    }
557 	} else {
558 	    if (pc->pc_acpi_id == *acpi_id) {
559 		if (bootverbose)
560 		    device_printf(dev,
561 			"Processor %s (ACPI ID %u) -> APIC ID %d\n",
562 			acpi_name(acpi_get_handle(dev)), *acpi_id,
563 			pc->pc_cpuid);
564 		*cpu_id = pc->pc_cpuid;
565 		return (0);
566 	    }
567 	}
568     }
569 
570     if (bootverbose)
571 	printf("ACPI: Processor %s (ACPI ID %u) ignored\n",
572 	    acpi_name(acpi_get_handle(dev)), *acpi_id);
573 
574     return (ESRCH);
575 }
576 
577 static struct resource_list *
578 acpi_cpu_get_rlist(device_t dev, device_t child)
579 {
580     struct acpi_cpu_device *ad;
581 
582     ad = device_get_ivars(child);
583     if (ad == NULL)
584 	return (NULL);
585     return (&ad->ad_rl);
586 }
587 
588 static device_t
589 acpi_cpu_add_child(device_t dev, u_int order, const char *name, int unit)
590 {
591     struct acpi_cpu_device *ad;
592     device_t child;
593 
594     if ((ad = malloc(sizeof(*ad), M_TEMP, M_NOWAIT | M_ZERO)) == NULL)
595 	return (NULL);
596 
597     resource_list_init(&ad->ad_rl);
598 
599     child = device_add_child_ordered(dev, order, name, unit);
600     if (child != NULL)
601 	device_set_ivars(child, ad);
602     else
603 	free(ad, M_TEMP);
604     return (child);
605 }
606 
607 static int
608 acpi_cpu_read_ivar(device_t dev, device_t child, int index, uintptr_t *result)
609 {
610     struct acpi_cpu_softc *sc;
611 
612     sc = device_get_softc(dev);
613     switch (index) {
614     case ACPI_IVAR_HANDLE:
615 	*result = (uintptr_t)sc->cpu_handle;
616 	break;
617     case CPU_IVAR_PCPU:
618 	*result = (uintptr_t)sc->cpu_pcpu;
619 	break;
620 #if defined(__amd64__) || defined(__i386__)
621     case CPU_IVAR_NOMINAL_MHZ:
622 	if (tsc_is_invariant) {
623 	    *result = (uintptr_t)(atomic_load_acq_64(&tsc_freq) / 1000000);
624 	    break;
625 	}
626 	/* FALLTHROUGH */
627 #endif
628     default:
629 	return (ENOENT);
630     }
631     return (0);
632 }
633 
634 static int
635 acpi_cpu_shutdown(device_t dev)
636 {
637     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
638 
639     /* Allow children to shutdown first. */
640     bus_generic_shutdown(dev);
641 
642     /*
643      * Disable any entry to the idle function.
644      */
645     disable_idle(device_get_softc(dev));
646 
647     /*
648      * CPU devices are not truely detached and remain referenced,
649      * so their resources are not freed.
650      */
651 
652     return_VALUE (0);
653 }
654 
655 static void
656 acpi_cpu_cx_probe(struct acpi_cpu_softc *sc)
657 {
658     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
659 
660     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
661     sc->cpu_prev_sleep = 1000000;
662     sc->cpu_cx_lowest = 0;
663     sc->cpu_cx_lowest_lim = 0;
664 
665     /*
666      * Check for the ACPI 2.0 _CST sleep states object. If we can't find
667      * any, we'll revert to generic FADT/P_BLK Cx control method which will
668      * be handled by acpi_cpu_startup. We need to defer to after having
669      * probed all the cpus in the system before probing for generic Cx
670      * states as we may already have found cpus with valid _CST packages
671      */
672     if (!cpu_cx_generic && acpi_cpu_cx_cst(sc) != 0) {
673 	/*
674 	 * We were unable to find a _CST package for this cpu or there
675 	 * was an error parsing it. Switch back to generic mode.
676 	 */
677 	cpu_cx_generic = TRUE;
678 	if (bootverbose)
679 	    device_printf(sc->cpu_dev, "switching to generic Cx mode\n");
680     }
681 
682     /*
683      * TODO: _CSD Package should be checked here.
684      */
685 }
686 
687 static void
688 acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc)
689 {
690     ACPI_GENERIC_ADDRESS	 gas;
691     struct acpi_cx		*cx_ptr;
692 
693     sc->cpu_cx_count = 0;
694     cx_ptr = sc->cpu_cx_states;
695 
696     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
697     sc->cpu_prev_sleep = 1000000;
698 
699     /* C1 has been required since just after ACPI 1.0 */
700     cx_ptr->type = ACPI_STATE_C1;
701     cx_ptr->trans_lat = 0;
702     cx_ptr++;
703     sc->cpu_non_c2 = sc->cpu_cx_count;
704     sc->cpu_non_c3 = sc->cpu_cx_count;
705     sc->cpu_cx_count++;
706     cpu_deepest_sleep = 1;
707 
708     /*
709      * The spec says P_BLK must be 6 bytes long.  However, some systems
710      * use it to indicate a fractional set of features present so we
711      * take 5 as C2.  Some may also have a value of 7 to indicate
712      * another C3 but most use _CST for this (as required) and having
713      * "only" C1-C3 is not a hardship.
714      */
715     if (sc->cpu_p_blk_len < 5)
716 	return;
717 
718     /* Validate and allocate resources for C2 (P_LVL2). */
719     gas.SpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
720     gas.BitWidth = 8;
721     if (AcpiGbl_FADT.C2Latency <= 100) {
722 	gas.Address = sc->cpu_p_blk + 4;
723 	cx_ptr->res_rid = 0;
724 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &cx_ptr->res_rid,
725 	    &gas, &cx_ptr->p_lvlx, RF_SHAREABLE);
726 	if (cx_ptr->p_lvlx != NULL) {
727 	    cx_ptr->type = ACPI_STATE_C2;
728 	    cx_ptr->trans_lat = AcpiGbl_FADT.C2Latency;
729 	    cx_ptr++;
730 	    sc->cpu_non_c3 = sc->cpu_cx_count;
731 	    sc->cpu_cx_count++;
732 	    cpu_deepest_sleep = 2;
733 	}
734     }
735     if (sc->cpu_p_blk_len < 6)
736 	return;
737 
738     /* Validate and allocate resources for C3 (P_LVL3). */
739     if (AcpiGbl_FADT.C3Latency <= 1000 && !(cpu_quirks & CPU_QUIRK_NO_C3)) {
740 	gas.Address = sc->cpu_p_blk + 5;
741 	cx_ptr->res_rid = 1;
742 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &cx_ptr->res_rid,
743 	    &gas, &cx_ptr->p_lvlx, RF_SHAREABLE);
744 	if (cx_ptr->p_lvlx != NULL) {
745 	    cx_ptr->type = ACPI_STATE_C3;
746 	    cx_ptr->trans_lat = AcpiGbl_FADT.C3Latency;
747 	    cx_ptr++;
748 	    sc->cpu_cx_count++;
749 	    cpu_deepest_sleep = 3;
750 	}
751     }
752 }
753 
754 #if defined(__i386__) || defined(__amd64__)
755 static void
756 acpi_cpu_cx_cst_mwait(struct acpi_cx *cx_ptr, uint64_t address, int accsize)
757 {
758 
759 	cx_ptr->do_mwait = true;
760 	cx_ptr->mwait_hint = address & 0xffffffff;
761 	cx_ptr->mwait_hw_coord = (accsize & CST_FFH_MWAIT_HW_COORD) != 0;
762 	cx_ptr->mwait_bm_avoidance = (accsize & CST_FFH_MWAIT_BM_AVOID) != 0;
763 }
764 #endif
765 
766 static void
767 acpi_cpu_cx_cst_free_plvlx(device_t cpu_dev, struct acpi_cx *cx_ptr)
768 {
769 
770 	if (cx_ptr->p_lvlx == NULL)
771 		return;
772 	bus_release_resource(cpu_dev, cx_ptr->res_type, cx_ptr->res_rid,
773 	    cx_ptr->p_lvlx);
774 	cx_ptr->p_lvlx = NULL;
775 }
776 
777 /*
778  * Parse a _CST package and set up its Cx states.  Since the _CST object
779  * can change dynamically, our notify handler may call this function
780  * to clean up and probe the new _CST package.
781  */
782 static int
783 acpi_cpu_cx_cst(struct acpi_cpu_softc *sc)
784 {
785     struct	 acpi_cx *cx_ptr;
786     ACPI_STATUS	 status;
787     ACPI_BUFFER	 buf;
788     ACPI_OBJECT	*top;
789     ACPI_OBJECT	*pkg;
790     uint32_t	 count;
791     int		 i;
792 #if defined(__i386__) || defined(__amd64__)
793     uint64_t	 address;
794     int		 vendor, class, accsize;
795 #endif
796 
797     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
798 
799     buf.Pointer = NULL;
800     buf.Length = ACPI_ALLOCATE_BUFFER;
801     status = AcpiEvaluateObject(sc->cpu_handle, "_CST", NULL, &buf);
802     if (ACPI_FAILURE(status))
803 	return (ENXIO);
804 
805     /* _CST is a package with a count and at least one Cx package. */
806     top = (ACPI_OBJECT *)buf.Pointer;
807     if (!ACPI_PKG_VALID(top, 2) || acpi_PkgInt32(top, 0, &count) != 0) {
808 	device_printf(sc->cpu_dev, "invalid _CST package\n");
809 	AcpiOsFree(buf.Pointer);
810 	return (ENXIO);
811     }
812     if (count != top->Package.Count - 1) {
813 	device_printf(sc->cpu_dev, "invalid _CST state count (%d != %d)\n",
814 	       count, top->Package.Count - 1);
815 	count = top->Package.Count - 1;
816     }
817     if (count > MAX_CX_STATES) {
818 	device_printf(sc->cpu_dev, "_CST has too many states (%d)\n", count);
819 	count = MAX_CX_STATES;
820     }
821 
822     sc->cpu_non_c2 = 0;
823     sc->cpu_non_c3 = 0;
824     sc->cpu_cx_count = 0;
825     cx_ptr = sc->cpu_cx_states;
826 
827     /*
828      * C1 has been required since just after ACPI 1.0.
829      * Reserve the first slot for it.
830      */
831     cx_ptr->type = ACPI_STATE_C0;
832     cx_ptr++;
833     sc->cpu_cx_count++;
834     cpu_deepest_sleep = 1;
835 
836     /* Set up all valid states. */
837     for (i = 0; i < count; i++) {
838 	pkg = &top->Package.Elements[i + 1];
839 	if (!ACPI_PKG_VALID(pkg, 4) ||
840 	    acpi_PkgInt32(pkg, 1, &cx_ptr->type) != 0 ||
841 	    acpi_PkgInt32(pkg, 2, &cx_ptr->trans_lat) != 0 ||
842 	    acpi_PkgInt32(pkg, 3, &cx_ptr->power) != 0) {
843 
844 	    device_printf(sc->cpu_dev, "skipping invalid Cx state package\n");
845 	    continue;
846 	}
847 
848 	/* Validate the state to see if we should use it. */
849 	switch (cx_ptr->type) {
850 	case ACPI_STATE_C1:
851 	    acpi_cpu_cx_cst_free_plvlx(sc->cpu_dev, cx_ptr);
852 #if defined(__i386__) || defined(__amd64__)
853 	    if (acpi_PkgFFH_IntelCpu(pkg, 0, &vendor, &class, &address,
854 	      &accsize) == 0 && vendor == CST_FFH_VENDOR_INTEL) {
855 		if (class == CST_FFH_INTEL_CL_C1IO) {
856 		    /* C1 I/O then Halt */
857 		    cx_ptr->res_rid = sc->cpu_cx_count;
858 		    bus_set_resource(sc->cpu_dev, SYS_RES_IOPORT,
859 		      cx_ptr->res_rid, address, 1);
860 		    cx_ptr->p_lvlx = bus_alloc_resource_any(sc->cpu_dev,
861 		      SYS_RES_IOPORT, &cx_ptr->res_rid, RF_ACTIVE |
862 		      RF_SHAREABLE);
863 		    if (cx_ptr->p_lvlx == NULL) {
864 			bus_delete_resource(sc->cpu_dev, SYS_RES_IOPORT,
865 			  cx_ptr->res_rid);
866 			device_printf(sc->cpu_dev,
867 			  "C1 I/O failed to allocate port %d, "
868 			  "degrading to C1 Halt", (int)address);
869 		    }
870 		} else if (class == CST_FFH_INTEL_CL_MWAIT) {
871 		    acpi_cpu_cx_cst_mwait(cx_ptr, address, accsize);
872 		}
873 	    }
874 #endif
875 	    if (sc->cpu_cx_states[0].type == ACPI_STATE_C0) {
876 		/* This is the first C1 state.  Use the reserved slot. */
877 		sc->cpu_cx_states[0] = *cx_ptr;
878 	    } else {
879 		sc->cpu_non_c2 = sc->cpu_cx_count;
880 		sc->cpu_non_c3 = sc->cpu_cx_count;
881 		cx_ptr++;
882 		sc->cpu_cx_count++;
883 	    }
884 	    continue;
885 	case ACPI_STATE_C2:
886 	    sc->cpu_non_c3 = sc->cpu_cx_count;
887 	    if (cpu_deepest_sleep < 2)
888 		    cpu_deepest_sleep = 2;
889 	    break;
890 	case ACPI_STATE_C3:
891 	default:
892 	    if ((cpu_quirks & CPU_QUIRK_NO_C3) != 0) {
893 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
894 				 "acpi_cpu%d: C3[%d] not available.\n",
895 				 device_get_unit(sc->cpu_dev), i));
896 		continue;
897 	    } else
898 		cpu_deepest_sleep = 3;
899 	    break;
900 	}
901 
902 	/* Free up any previous register. */
903 	acpi_cpu_cx_cst_free_plvlx(sc->cpu_dev, cx_ptr);
904 
905 	/* Allocate the control register for C2 or C3. */
906 #if defined(__i386__) || defined(__amd64__)
907 	if (acpi_PkgFFH_IntelCpu(pkg, 0, &vendor, &class, &address,
908 	  &accsize) == 0 && vendor == CST_FFH_VENDOR_INTEL &&
909 	  class == CST_FFH_INTEL_CL_MWAIT) {
910 	    /* Native C State Instruction use (mwait) */
911 	    acpi_cpu_cx_cst_mwait(cx_ptr, address, accsize);
912 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
913 	      "acpi_cpu%d: Got C%d/mwait - %d latency\n",
914 	      device_get_unit(sc->cpu_dev), cx_ptr->type, cx_ptr->trans_lat));
915 	    cx_ptr++;
916 	    sc->cpu_cx_count++;
917 	} else
918 #endif
919 	{
920 	    cx_ptr->res_rid = sc->cpu_cx_count;
921 	    acpi_PkgGas(sc->cpu_dev, pkg, 0, &cx_ptr->res_type,
922 		&cx_ptr->res_rid, &cx_ptr->p_lvlx, RF_SHAREABLE);
923 	    if (cx_ptr->p_lvlx) {
924 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
925 		     "acpi_cpu%d: Got C%d - %d latency\n",
926 		     device_get_unit(sc->cpu_dev), cx_ptr->type,
927 		     cx_ptr->trans_lat));
928 		cx_ptr++;
929 		sc->cpu_cx_count++;
930 	    }
931 	}
932     }
933     AcpiOsFree(buf.Pointer);
934 
935     /* If C1 state was not found, we need one now. */
936     cx_ptr = sc->cpu_cx_states;
937     if (cx_ptr->type == ACPI_STATE_C0) {
938 	cx_ptr->type = ACPI_STATE_C1;
939 	cx_ptr->trans_lat = 0;
940     }
941 
942     return (0);
943 }
944 
945 /*
946  * Call this *after* all CPUs have been attached.
947  */
948 static void
949 acpi_cpu_startup(void *arg)
950 {
951     struct acpi_cpu_softc *sc;
952     int i;
953 
954     /* Get set of CPU devices */
955     devclass_get_devices(acpi_cpu_devclass, &cpu_devices, &cpu_ndevices);
956 
957     /*
958      * Setup any quirks that might necessary now that we have probed
959      * all the CPUs
960      */
961     acpi_cpu_quirks();
962 
963     if (cpu_cx_generic) {
964 	/*
965 	 * We are using generic Cx mode, probe for available Cx states
966 	 * for all processors.
967 	 */
968 	for (i = 0; i < cpu_ndevices; i++) {
969 	    sc = device_get_softc(cpu_devices[i]);
970 	    acpi_cpu_generic_cx_probe(sc);
971 	}
972     } else {
973 	/*
974 	 * We are using _CST mode, remove C3 state if necessary.
975 	 * As we now know for sure that we will be using _CST mode
976 	 * install our notify handler.
977 	 */
978 	for (i = 0; i < cpu_ndevices; i++) {
979 	    sc = device_get_softc(cpu_devices[i]);
980 	    if (cpu_quirks & CPU_QUIRK_NO_C3) {
981 		sc->cpu_cx_count = min(sc->cpu_cx_count, sc->cpu_non_c3 + 1);
982 	    }
983 	    AcpiInstallNotifyHandler(sc->cpu_handle, ACPI_DEVICE_NOTIFY,
984 		acpi_cpu_notify, sc);
985 	}
986     }
987 
988     /* Perform Cx final initialization. */
989     for (i = 0; i < cpu_ndevices; i++) {
990 	sc = device_get_softc(cpu_devices[i]);
991 	acpi_cpu_startup_cx(sc);
992     }
993 
994     /* Add a sysctl handler to handle global Cx lowest setting */
995     SYSCTL_ADD_PROC(&cpu_sysctl_ctx, SYSCTL_CHILDREN(cpu_sysctl_tree),
996 	OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
997 	NULL, 0, acpi_cpu_global_cx_lowest_sysctl, "A",
998 	"Global lowest Cx sleep state to use");
999 
1000     /* Take over idling from cpu_idle_default(). */
1001     cpu_cx_lowest_lim = 0;
1002     for (i = 0; i < cpu_ndevices; i++) {
1003 	sc = device_get_softc(cpu_devices[i]);
1004 	enable_idle(sc);
1005     }
1006 #if defined(__i386__) || defined(__amd64__)
1007     cpu_idle_hook = acpi_cpu_idle;
1008 #endif
1009 }
1010 
1011 static void
1012 acpi_cpu_cx_list(struct acpi_cpu_softc *sc)
1013 {
1014     struct sbuf sb;
1015     int i;
1016 
1017     /*
1018      * Set up the list of Cx states
1019      */
1020     sbuf_new(&sb, sc->cpu_cx_supported, sizeof(sc->cpu_cx_supported),
1021 	SBUF_FIXEDLEN);
1022     for (i = 0; i < sc->cpu_cx_count; i++)
1023 	sbuf_printf(&sb, "C%d/%d/%d ", i + 1, sc->cpu_cx_states[i].type,
1024 	    sc->cpu_cx_states[i].trans_lat);
1025     sbuf_trim(&sb);
1026     sbuf_finish(&sb);
1027 }
1028 
1029 static void
1030 acpi_cpu_startup_cx(struct acpi_cpu_softc *sc)
1031 {
1032     acpi_cpu_cx_list(sc);
1033 
1034     SYSCTL_ADD_STRING(&sc->cpu_sysctl_ctx,
1035 		      SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
1036 		      OID_AUTO, "cx_supported", CTLFLAG_RD,
1037 		      sc->cpu_cx_supported, 0,
1038 		      "Cx/microsecond values for supported Cx states");
1039     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
1040 		    SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
1041 		    OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
1042 		    (void *)sc, 0, acpi_cpu_cx_lowest_sysctl, "A",
1043 		    "lowest Cx sleep state to use");
1044     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
1045 		    SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
1046 		    OID_AUTO, "cx_usage", CTLTYPE_STRING | CTLFLAG_RD,
1047 		    (void *)sc, 0, acpi_cpu_usage_sysctl, "A",
1048 		    "percent usage for each Cx state");
1049     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
1050 		    SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
1051 		    OID_AUTO, "cx_usage_counters", CTLTYPE_STRING | CTLFLAG_RD,
1052 		    (void *)sc, 0, acpi_cpu_usage_counters_sysctl, "A",
1053 		    "Cx sleep state counters");
1054 #if defined(__i386__) || defined(__amd64__)
1055     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
1056 		    SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
1057 		    OID_AUTO, "cx_method", CTLTYPE_STRING | CTLFLAG_RD,
1058 		    (void *)sc, 0, acpi_cpu_method_sysctl, "A",
1059 		    "Cx entrance methods");
1060 #endif
1061 
1062     /* Signal platform that we can handle _CST notification. */
1063     if (!cpu_cx_generic && cpu_cst_cnt != 0) {
1064 	ACPI_LOCK(acpi);
1065 	AcpiOsWritePort(cpu_smi_cmd, cpu_cst_cnt, 8);
1066 	ACPI_UNLOCK(acpi);
1067     }
1068 }
1069 
1070 #if defined(__i386__) || defined(__amd64__)
1071 /*
1072  * Idle the CPU in the lowest state possible.  This function is called with
1073  * interrupts disabled.  Note that once it re-enables interrupts, a task
1074  * switch can occur so do not access shared data (i.e. the softc) after
1075  * interrupts are re-enabled.
1076  */
1077 static void
1078 acpi_cpu_idle(sbintime_t sbt)
1079 {
1080     struct	acpi_cpu_softc *sc;
1081     struct	acpi_cx *cx_next;
1082     uint64_t	cputicks;
1083     uint32_t	start_time, end_time;
1084     ACPI_STATUS	status;
1085     int		bm_active, cx_next_idx, i, us;
1086 
1087     /*
1088      * Look up our CPU id to get our softc.  If it's NULL, we'll use C1
1089      * since there is no ACPI processor object for this CPU.  This occurs
1090      * for logical CPUs in the HTT case.
1091      */
1092     sc = cpu_softc[PCPU_GET(cpuid)];
1093     if (sc == NULL) {
1094 	acpi_cpu_c1();
1095 	return;
1096     }
1097 
1098     /* If disabled, take the safe path. */
1099     if (is_idle_disabled(sc)) {
1100 	acpi_cpu_c1();
1101 	return;
1102     }
1103 
1104     /* Find the lowest state that has small enough latency. */
1105     us = sc->cpu_prev_sleep;
1106     if (sbt >= 0 && us > (sbt >> 12))
1107 	us = (sbt >> 12);
1108     cx_next_idx = 0;
1109     if (cpu_disable_c2_sleep)
1110 	i = min(sc->cpu_cx_lowest, sc->cpu_non_c2);
1111     else if (cpu_disable_c3_sleep)
1112 	i = min(sc->cpu_cx_lowest, sc->cpu_non_c3);
1113     else
1114 	i = sc->cpu_cx_lowest;
1115     for (; i >= 0; i--) {
1116 	if (sc->cpu_cx_states[i].trans_lat * 3 <= us) {
1117 	    cx_next_idx = i;
1118 	    break;
1119 	}
1120     }
1121 
1122     /*
1123      * Check for bus master activity.  If there was activity, clear
1124      * the bit and use the lowest non-C3 state.  Note that the USB
1125      * driver polling for new devices keeps this bit set all the
1126      * time if USB is loaded.
1127      */
1128     if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0 &&
1129 	cx_next_idx > sc->cpu_non_c3) {
1130 	status = AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, &bm_active);
1131 	if (ACPI_SUCCESS(status) && bm_active != 0) {
1132 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, 1);
1133 	    cx_next_idx = sc->cpu_non_c3;
1134 	}
1135     }
1136 
1137     /* Select the next state and update statistics. */
1138     cx_next = &sc->cpu_cx_states[cx_next_idx];
1139     sc->cpu_cx_stats[cx_next_idx]++;
1140     KASSERT(cx_next->type != ACPI_STATE_C0, ("acpi_cpu_idle: C0 sleep"));
1141 
1142     /*
1143      * Execute HLT (or equivalent) and wait for an interrupt.  We can't
1144      * precisely calculate the time spent in C1 since the place we wake up
1145      * is an ISR.  Assume we slept no more then half of quantum, unless
1146      * we are called inside critical section, delaying context switch.
1147      */
1148     if (cx_next->type == ACPI_STATE_C1) {
1149 	cputicks = cpu_ticks();
1150 	if (cx_next->p_lvlx != NULL) {
1151 	    /* C1 I/O then Halt */
1152 	    CPU_GET_REG(cx_next->p_lvlx, 1);
1153 	}
1154 	if (cx_next->do_mwait)
1155 	    acpi_cpu_idle_mwait(cx_next->mwait_hint);
1156 	else
1157 	    acpi_cpu_c1();
1158 	end_time = ((cpu_ticks() - cputicks) << 20) / cpu_tickrate();
1159 	if (curthread->td_critnest == 0)
1160 		end_time = min(end_time, 500000 / hz);
1161 	sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + end_time) / 4;
1162 	return;
1163     }
1164 
1165     /*
1166      * For C3, disable bus master arbitration and enable bus master wake
1167      * if BM control is available, otherwise flush the CPU cache.
1168      */
1169     if (cx_next->type == ACPI_STATE_C3 || cx_next->mwait_bm_avoidance) {
1170 	if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
1171 	    AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 1);
1172 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
1173 	} else
1174 	    ACPI_FLUSH_CPU_CACHE();
1175     }
1176 
1177     /*
1178      * Read from P_LVLx to enter C2(+), checking time spent asleep.
1179      * Use the ACPI timer for measuring sleep time.  Since we need to
1180      * get the time very close to the CPU start/stop clock logic, this
1181      * is the only reliable time source.
1182      */
1183     if (cx_next->type == ACPI_STATE_C3) {
1184 	AcpiHwRead(&start_time, &AcpiGbl_FADT.XPmTimerBlock);
1185 	cputicks = 0;
1186     } else {
1187 	start_time = 0;
1188 	cputicks = cpu_ticks();
1189     }
1190     if (cx_next->do_mwait)
1191 	acpi_cpu_idle_mwait(cx_next->mwait_hint);
1192     else
1193 	CPU_GET_REG(cx_next->p_lvlx, 1);
1194 
1195     /*
1196      * Read the end time twice.  Since it may take an arbitrary time
1197      * to enter the idle state, the first read may be executed before
1198      * the processor has stopped.  Doing it again provides enough
1199      * margin that we are certain to have a correct value.
1200      */
1201     AcpiHwRead(&end_time, &AcpiGbl_FADT.XPmTimerBlock);
1202     if (cx_next->type == ACPI_STATE_C3) {
1203 	AcpiHwRead(&end_time, &AcpiGbl_FADT.XPmTimerBlock);
1204 	end_time = acpi_TimerDelta(end_time, start_time);
1205     } else
1206 	end_time = ((cpu_ticks() - cputicks) << 20) / cpu_tickrate();
1207 
1208     /* Enable bus master arbitration and disable bus master wakeup. */
1209     if ((cx_next->type == ACPI_STATE_C3 || cx_next->mwait_bm_avoidance) &&
1210       (cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
1211 	AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 0);
1212 	AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1213     }
1214     ACPI_ENABLE_IRQS();
1215 
1216     sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + PM_USEC(end_time)) / 4;
1217 }
1218 #endif
1219 
1220 /*
1221  * Re-evaluate the _CST object when we are notified that it changed.
1222  */
1223 static void
1224 acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context)
1225 {
1226     struct acpi_cpu_softc *sc = (struct acpi_cpu_softc *)context;
1227 
1228     if (notify != ACPI_NOTIFY_CX_STATES)
1229 	return;
1230 
1231     /*
1232      * C-state data for target CPU is going to be in flux while we execute
1233      * acpi_cpu_cx_cst, so disable entering acpi_cpu_idle.
1234      * Also, it may happen that multiple ACPI taskqueues may concurrently
1235      * execute notifications for the same CPU.  ACPI_SERIAL is used to
1236      * protect against that.
1237      */
1238     ACPI_SERIAL_BEGIN(cpu);
1239     disable_idle(sc);
1240 
1241     /* Update the list of Cx states. */
1242     acpi_cpu_cx_cst(sc);
1243     acpi_cpu_cx_list(sc);
1244     acpi_cpu_set_cx_lowest(sc);
1245 
1246     enable_idle(sc);
1247     ACPI_SERIAL_END(cpu);
1248 
1249     acpi_UserNotify("PROCESSOR", sc->cpu_handle, notify);
1250 }
1251 
1252 static void
1253 acpi_cpu_quirks(void)
1254 {
1255     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
1256 
1257     /*
1258      * Bus mastering arbitration control is needed to keep caches coherent
1259      * while sleeping in C3.  If it's not present but a working flush cache
1260      * instruction is present, flush the caches before entering C3 instead.
1261      * Otherwise, just disable C3 completely.
1262      */
1263     if (AcpiGbl_FADT.Pm2ControlBlock == 0 ||
1264 	AcpiGbl_FADT.Pm2ControlLength == 0) {
1265 	if ((AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD) &&
1266 	    (AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD_FLUSH) == 0) {
1267 	    cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1268 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1269 		"acpi_cpu: no BM control, using flush cache method\n"));
1270 	} else {
1271 	    cpu_quirks |= CPU_QUIRK_NO_C3;
1272 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1273 		"acpi_cpu: no BM control, C3 not available\n"));
1274 	}
1275     }
1276 
1277     /*
1278      * If we are using generic Cx mode, C3 on multiple CPUs requires using
1279      * the expensive flush cache instruction.
1280      */
1281     if (cpu_cx_generic && mp_ncpus > 1) {
1282 	cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1283 	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1284 	    "acpi_cpu: SMP, using flush cache mode for C3\n"));
1285     }
1286 
1287     /* Look for various quirks of the PIIX4 part. */
1288     acpi_cpu_quirks_piix4();
1289 }
1290 
1291 static void
1292 acpi_cpu_quirks_piix4(void)
1293 {
1294 #ifdef __i386__
1295     device_t acpi_dev;
1296     uint32_t val;
1297     ACPI_STATUS status;
1298 
1299     acpi_dev = pci_find_device(PCI_VENDOR_INTEL, PCI_DEVICE_82371AB_3);
1300     if (acpi_dev != NULL) {
1301 	switch (pci_get_revid(acpi_dev)) {
1302 	/*
1303 	 * Disable C3 support for all PIIX4 chipsets.  Some of these parts
1304 	 * do not report the BMIDE status to the BM status register and
1305 	 * others have a livelock bug if Type-F DMA is enabled.  Linux
1306 	 * works around the BMIDE bug by reading the BM status directly
1307 	 * but we take the simpler approach of disabling C3 for these
1308 	 * parts.
1309 	 *
1310 	 * See erratum #18 ("C3 Power State/BMIDE and Type-F DMA
1311 	 * Livelock") from the January 2002 PIIX4 specification update.
1312 	 * Applies to all PIIX4 models.
1313 	 *
1314 	 * Also, make sure that all interrupts cause a "Stop Break"
1315 	 * event to exit from C2 state.
1316 	 * Also, BRLD_EN_BM (ACPI_BITREG_BUS_MASTER_RLD in ACPI-speak)
1317 	 * should be set to zero, otherwise it causes C2 to short-sleep.
1318 	 * PIIX4 doesn't properly support C3 and bus master activity
1319 	 * need not break out of C2.
1320 	 */
1321 	case PCI_REVISION_A_STEP:
1322 	case PCI_REVISION_B_STEP:
1323 	case PCI_REVISION_4E:
1324 	case PCI_REVISION_4M:
1325 	    cpu_quirks |= CPU_QUIRK_NO_C3;
1326 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1327 		"acpi_cpu: working around PIIX4 bug, disabling C3\n"));
1328 
1329 	    val = pci_read_config(acpi_dev, PIIX4_DEVACTB_REG, 4);
1330 	    if ((val & PIIX4_STOP_BREAK_MASK) != PIIX4_STOP_BREAK_MASK) {
1331 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1332 		    "acpi_cpu: PIIX4: enabling IRQs to generate Stop Break\n"));
1333 	    	val |= PIIX4_STOP_BREAK_MASK;
1334 		pci_write_config(acpi_dev, PIIX4_DEVACTB_REG, val, 4);
1335 	    }
1336 	    status = AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_RLD, &val);
1337 	    if (ACPI_SUCCESS(status) && val != 0) {
1338 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1339 		    "acpi_cpu: PIIX4: reset BRLD_EN_BM\n"));
1340 		AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1341 	    }
1342 	    break;
1343 	default:
1344 	    break;
1345 	}
1346     }
1347 #endif
1348 }
1349 
1350 static int
1351 acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS)
1352 {
1353     struct acpi_cpu_softc *sc;
1354     struct sbuf	 sb;
1355     char	 buf[128];
1356     int		 i;
1357     uintmax_t	 fract, sum, whole;
1358 
1359     sc = (struct acpi_cpu_softc *) arg1;
1360     sum = 0;
1361     for (i = 0; i < sc->cpu_cx_count; i++)
1362 	sum += sc->cpu_cx_stats[i];
1363     sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1364     for (i = 0; i < sc->cpu_cx_count; i++) {
1365 	if (sum > 0) {
1366 	    whole = (uintmax_t)sc->cpu_cx_stats[i] * 100;
1367 	    fract = (whole % sum) * 100;
1368 	    sbuf_printf(&sb, "%u.%02u%% ", (u_int)(whole / sum),
1369 		(u_int)(fract / sum));
1370 	} else
1371 	    sbuf_printf(&sb, "0.00%% ");
1372     }
1373     sbuf_printf(&sb, "last %dus", sc->cpu_prev_sleep);
1374     sbuf_trim(&sb);
1375     sbuf_finish(&sb);
1376     sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1377     sbuf_delete(&sb);
1378 
1379     return (0);
1380 }
1381 
1382 /*
1383  * XXX TODO: actually add support to count each entry/exit
1384  * from the Cx states.
1385  */
1386 static int
1387 acpi_cpu_usage_counters_sysctl(SYSCTL_HANDLER_ARGS)
1388 {
1389     struct acpi_cpu_softc *sc;
1390     struct sbuf	 sb;
1391     char	 buf[128];
1392     int		 i;
1393 
1394     sc = (struct acpi_cpu_softc *) arg1;
1395 
1396     /* Print out the raw counters */
1397     sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1398 
1399     for (i = 0; i < sc->cpu_cx_count; i++) {
1400         sbuf_printf(&sb, "%u ", sc->cpu_cx_stats[i]);
1401     }
1402 
1403     sbuf_trim(&sb);
1404     sbuf_finish(&sb);
1405     sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1406     sbuf_delete(&sb);
1407 
1408     return (0);
1409 }
1410 
1411 #if defined(__i386__) || defined(__amd64__)
1412 static int
1413 acpi_cpu_method_sysctl(SYSCTL_HANDLER_ARGS)
1414 {
1415 	struct acpi_cpu_softc *sc;
1416 	struct acpi_cx *cx;
1417 	struct sbuf sb;
1418 	char buf[128];
1419 	int i;
1420 
1421 	sc = (struct acpi_cpu_softc *)arg1;
1422 	sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1423 	for (i = 0; i < sc->cpu_cx_count; i++) {
1424 		cx = &sc->cpu_cx_states[i];
1425 		sbuf_printf(&sb, "C%d/", i + 1);
1426 		if (cx->do_mwait) {
1427 			sbuf_cat(&sb, "mwait");
1428 			if (cx->mwait_hw_coord)
1429 				sbuf_cat(&sb, "/hwc");
1430 			if (cx->mwait_bm_avoidance)
1431 				sbuf_cat(&sb, "/bma");
1432 		} else if (cx->type == ACPI_STATE_C1) {
1433 			sbuf_cat(&sb, "hlt");
1434 		} else {
1435 			sbuf_cat(&sb, "io");
1436 		}
1437 		if (cx->type == ACPI_STATE_C1 && cx->p_lvlx != NULL)
1438 			sbuf_cat(&sb, "/iohlt");
1439 		sbuf_putc(&sb, ' ');
1440 	}
1441 	sbuf_trim(&sb);
1442 	sbuf_finish(&sb);
1443 	sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1444 	sbuf_delete(&sb);
1445 	return (0);
1446 }
1447 #endif
1448 
1449 static int
1450 acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc)
1451 {
1452     int i;
1453 
1454     ACPI_SERIAL_ASSERT(cpu);
1455     sc->cpu_cx_lowest = min(sc->cpu_cx_lowest_lim, sc->cpu_cx_count - 1);
1456 
1457     /* If not disabling, cache the new lowest non-C3 state. */
1458     sc->cpu_non_c3 = 0;
1459     for (i = sc->cpu_cx_lowest; i >= 0; i--) {
1460 	if (sc->cpu_cx_states[i].type < ACPI_STATE_C3) {
1461 	    sc->cpu_non_c3 = i;
1462 	    break;
1463 	}
1464     }
1465 
1466     /* Reset the statistics counters. */
1467     bzero(sc->cpu_cx_stats, sizeof(sc->cpu_cx_stats));
1468     return (0);
1469 }
1470 
1471 static int
1472 acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1473 {
1474     struct	 acpi_cpu_softc *sc;
1475     char	 state[8];
1476     int		 val, error;
1477 
1478     sc = (struct acpi_cpu_softc *) arg1;
1479     snprintf(state, sizeof(state), "C%d", sc->cpu_cx_lowest_lim + 1);
1480     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1481     if (error != 0 || req->newptr == NULL)
1482 	return (error);
1483     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1484 	return (EINVAL);
1485     if (strcasecmp(state, "Cmax") == 0)
1486 	val = MAX_CX_STATES;
1487     else {
1488 	val = (int) strtol(state + 1, NULL, 10);
1489 	if (val < 1 || val > MAX_CX_STATES)
1490 	    return (EINVAL);
1491     }
1492 
1493     ACPI_SERIAL_BEGIN(cpu);
1494     sc->cpu_cx_lowest_lim = val - 1;
1495     acpi_cpu_set_cx_lowest(sc);
1496     ACPI_SERIAL_END(cpu);
1497 
1498     return (0);
1499 }
1500 
1501 static int
1502 acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1503 {
1504     struct	acpi_cpu_softc *sc;
1505     char	state[8];
1506     int		val, error, i;
1507 
1508     snprintf(state, sizeof(state), "C%d", cpu_cx_lowest_lim + 1);
1509     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1510     if (error != 0 || req->newptr == NULL)
1511 	return (error);
1512     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1513 	return (EINVAL);
1514     if (strcasecmp(state, "Cmax") == 0)
1515 	val = MAX_CX_STATES;
1516     else {
1517 	val = (int) strtol(state + 1, NULL, 10);
1518 	if (val < 1 || val > MAX_CX_STATES)
1519 	    return (EINVAL);
1520     }
1521 
1522     /* Update the new lowest useable Cx state for all CPUs. */
1523     ACPI_SERIAL_BEGIN(cpu);
1524     cpu_cx_lowest_lim = val - 1;
1525     for (i = 0; i < cpu_ndevices; i++) {
1526 	sc = device_get_softc(cpu_devices[i]);
1527 	sc->cpu_cx_lowest_lim = cpu_cx_lowest_lim;
1528 	acpi_cpu_set_cx_lowest(sc);
1529     }
1530     ACPI_SERIAL_END(cpu);
1531 
1532     return (0);
1533 }
1534