xref: /freebsd/sys/dev/acpica/acpi_cpu.c (revision 94cba8034ba53725c225c85e35724f0c2b13cea5)
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 #define	CPUDEV_DEVICE_ID	"ACPI0007"
144 
145 /* Allow users to ignore processor orders in MADT. */
146 static int cpu_unordered;
147 SYSCTL_INT(_debug_acpi, OID_AUTO, cpu_unordered, CTLFLAG_RDTUN,
148     &cpu_unordered, 0,
149     "Do not use the MADT to match ACPI Processor objects to CPUs.");
150 
151 /* Knob to disable acpi_cpu devices */
152 bool acpi_cpu_disabled = false;
153 
154 /* Platform hardware resource information. */
155 static uint32_t		 cpu_smi_cmd;	/* Value to write to SMI_CMD. */
156 static uint8_t		 cpu_cst_cnt;	/* Indicate we are _CST aware. */
157 static int		 cpu_quirks;	/* Indicate any hardware bugs. */
158 
159 /* Values for sysctl. */
160 static struct sysctl_ctx_list cpu_sysctl_ctx;
161 static struct sysctl_oid *cpu_sysctl_tree;
162 static int		 cpu_cx_generic;
163 static int		 cpu_cx_lowest_lim;
164 
165 static device_t		*cpu_devices;
166 static int		 cpu_ndevices;
167 static struct acpi_cpu_softc **cpu_softc;
168 ACPI_SERIAL_DECL(cpu, "ACPI CPU");
169 
170 static int	acpi_cpu_probe(device_t dev);
171 static int	acpi_cpu_attach(device_t dev);
172 static int	acpi_cpu_suspend(device_t dev);
173 static int	acpi_cpu_resume(device_t dev);
174 static int	acpi_pcpu_get_id(device_t dev, uint32_t *acpi_id,
175 		    uint32_t *cpu_id);
176 static struct resource_list *acpi_cpu_get_rlist(device_t dev, device_t child);
177 static device_t	acpi_cpu_add_child(device_t dev, u_int order, const char *name,
178 		    int unit);
179 static int	acpi_cpu_read_ivar(device_t dev, device_t child, int index,
180 		    uintptr_t *result);
181 static int	acpi_cpu_shutdown(device_t dev);
182 static void	acpi_cpu_cx_probe(struct acpi_cpu_softc *sc);
183 static void	acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc);
184 static int	acpi_cpu_cx_cst(struct acpi_cpu_softc *sc);
185 static void	acpi_cpu_startup(void *arg);
186 static void	acpi_cpu_startup_cx(struct acpi_cpu_softc *sc);
187 static void	acpi_cpu_cx_list(struct acpi_cpu_softc *sc);
188 #if defined(__i386__) || defined(__amd64__)
189 static void	acpi_cpu_idle(sbintime_t sbt);
190 #endif
191 static void	acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context);
192 static void	acpi_cpu_quirks(void);
193 static void	acpi_cpu_quirks_piix4(void);
194 static int	acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS);
195 static int	acpi_cpu_usage_counters_sysctl(SYSCTL_HANDLER_ARGS);
196 static int	acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc);
197 static int	acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
198 static int	acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
199 #if defined(__i386__) || defined(__amd64__)
200 static int	acpi_cpu_method_sysctl(SYSCTL_HANDLER_ARGS);
201 #endif
202 
203 static device_method_t acpi_cpu_methods[] = {
204     /* Device interface */
205     DEVMETHOD(device_probe,	acpi_cpu_probe),
206     DEVMETHOD(device_attach,	acpi_cpu_attach),
207     DEVMETHOD(device_detach,	bus_generic_detach),
208     DEVMETHOD(device_shutdown,	acpi_cpu_shutdown),
209     DEVMETHOD(device_suspend,	acpi_cpu_suspend),
210     DEVMETHOD(device_resume,	acpi_cpu_resume),
211 
212     /* Bus interface */
213     DEVMETHOD(bus_add_child,	acpi_cpu_add_child),
214     DEVMETHOD(bus_read_ivar,	acpi_cpu_read_ivar),
215     DEVMETHOD(bus_get_resource_list, acpi_cpu_get_rlist),
216     DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
217     DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
218     DEVMETHOD(bus_alloc_resource, bus_generic_rl_alloc_resource),
219     DEVMETHOD(bus_release_resource, bus_generic_rl_release_resource),
220     DEVMETHOD(bus_activate_resource, bus_generic_activate_resource),
221     DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource),
222     DEVMETHOD(bus_setup_intr,	bus_generic_setup_intr),
223     DEVMETHOD(bus_teardown_intr, bus_generic_teardown_intr),
224 
225     DEVMETHOD_END
226 };
227 
228 static driver_t acpi_cpu_driver = {
229     "cpu",
230     acpi_cpu_methods,
231     sizeof(struct acpi_cpu_softc),
232 };
233 
234 static devclass_t acpi_cpu_devclass;
235 DRIVER_MODULE(cpu, acpi, acpi_cpu_driver, acpi_cpu_devclass, 0, 0);
236 MODULE_DEPEND(cpu, acpi, 1, 1, 1);
237 
238 static int
239 acpi_cpu_probe(device_t dev)
240 {
241     static char		   *cpudev_ids[] = { CPUDEV_DEVICE_ID, NULL };
242     int			   acpi_id, cpu_id;
243     ACPI_BUFFER		   buf;
244     ACPI_HANDLE		   handle;
245     ACPI_OBJECT		   *obj;
246     ACPI_STATUS		   status;
247     ACPI_OBJECT_TYPE	   type;
248 
249     if (acpi_disabled("cpu") || acpi_cpu_disabled)
250 	return (ENXIO);
251     type = acpi_get_type(dev);
252     if (type != ACPI_TYPE_PROCESSOR && type != ACPI_TYPE_DEVICE)
253 	return (ENXIO);
254     if (type == ACPI_TYPE_DEVICE &&
255 	ACPI_ID_PROBE(device_get_parent(dev), dev, cpudev_ids, NULL) >= 0)
256 	return (ENXIO);
257 
258     handle = acpi_get_handle(dev);
259     if (cpu_softc == NULL)
260 	cpu_softc = malloc(sizeof(struct acpi_cpu_softc *) *
261 	    (mp_maxid + 1), M_TEMP /* XXX */, M_WAITOK | M_ZERO);
262 
263     if (type == ACPI_TYPE_PROCESSOR) {
264 	/* Get our Processor object. */
265 	buf.Pointer = NULL;
266 	buf.Length = ACPI_ALLOCATE_BUFFER;
267 	status = AcpiEvaluateObject(handle, NULL, NULL, &buf);
268 	if (ACPI_FAILURE(status)) {
269 	    device_printf(dev, "probe failed to get Processor obj - %s\n",
270 		AcpiFormatException(status));
271 	    return (ENXIO);
272 	}
273 	obj = (ACPI_OBJECT *)buf.Pointer;
274 	if (obj->Type != ACPI_TYPE_PROCESSOR) {
275 	    device_printf(dev, "Processor object has bad type %d\n",
276 		obj->Type);
277 	    AcpiOsFree(obj);
278 	    return (ENXIO);
279 	}
280 
281 	/*
282 	 * Find the processor associated with our unit.  We could use the
283 	 * ProcId as a key, however, some boxes do not have the same values
284 	 * in their Processor object as the ProcId values in the MADT.
285 	 */
286 	acpi_id = obj->Processor.ProcId;
287 	AcpiOsFree(obj);
288     } else {
289 	status = acpi_GetInteger(handle, "_UID", &acpi_id);
290 	if (ACPI_FAILURE(status)) {
291 	    device_printf(dev, "Device object has bad value - %s\n",
292 		AcpiFormatException(status));
293 	    return (ENXIO);
294 	}
295     }
296     if (acpi_pcpu_get_id(dev, &acpi_id, &cpu_id) != 0)
297 	return (ENXIO);
298 
299     /*
300      * Check if we already probed this processor.  We scan the bus twice
301      * so it's possible we've already seen this one.
302      */
303     if (cpu_softc[cpu_id] != NULL)
304 	return (ENXIO);
305 
306     /* Mark this processor as in-use and save our derived id for attach. */
307     cpu_softc[cpu_id] = (void *)1;
308     acpi_set_private(dev, (void*)(intptr_t)cpu_id);
309     device_set_desc(dev, "ACPI CPU");
310 
311     if (!bootverbose && device_get_unit(dev) != 0) {
312 	    device_quiet(dev);
313 	    device_quiet_children(dev);
314     }
315 
316     return (0);
317 }
318 
319 static int
320 acpi_cpu_attach(device_t dev)
321 {
322     ACPI_BUFFER		   buf;
323     ACPI_OBJECT		   arg, *obj;
324     ACPI_OBJECT_LIST	   arglist;
325     struct pcpu		   *pcpu_data;
326     struct acpi_cpu_softc *sc;
327     struct acpi_softc	  *acpi_sc;
328     ACPI_STATUS		   status;
329     u_int		   features;
330     int			   cpu_id, drv_count, i;
331     driver_t 		  **drivers;
332     uint32_t		   cap_set[3];
333 
334     /* UUID needed by _OSC evaluation */
335     static uint8_t cpu_oscuuid[16] = { 0x16, 0xA6, 0x77, 0x40, 0x0C, 0x29,
336 				       0xBE, 0x47, 0x9E, 0xBD, 0xD8, 0x70,
337 				       0x58, 0x71, 0x39, 0x53 };
338 
339     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
340 
341     sc = device_get_softc(dev);
342     sc->cpu_dev = dev;
343     sc->cpu_handle = acpi_get_handle(dev);
344     cpu_id = (int)(intptr_t)acpi_get_private(dev);
345     cpu_softc[cpu_id] = sc;
346     pcpu_data = pcpu_find(cpu_id);
347     pcpu_data->pc_device = dev;
348     sc->cpu_pcpu = pcpu_data;
349     cpu_smi_cmd = AcpiGbl_FADT.SmiCommand;
350     cpu_cst_cnt = AcpiGbl_FADT.CstControl;
351 
352     if (acpi_get_type(dev) == ACPI_TYPE_PROCESSOR) {
353 	buf.Pointer = NULL;
354 	buf.Length = ACPI_ALLOCATE_BUFFER;
355 	status = AcpiEvaluateObject(sc->cpu_handle, NULL, NULL, &buf);
356 	if (ACPI_FAILURE(status)) {
357 	    device_printf(dev, "attach failed to get Processor obj - %s\n",
358 		AcpiFormatException(status));
359 	    return (ENXIO);
360 	}
361 	obj = (ACPI_OBJECT *)buf.Pointer;
362 	sc->cpu_p_blk = obj->Processor.PblkAddress;
363 	sc->cpu_p_blk_len = obj->Processor.PblkLength;
364 	sc->cpu_acpi_id = obj->Processor.ProcId;
365 	AcpiOsFree(obj);
366     } else {
367 	KASSERT(acpi_get_type(dev) == ACPI_TYPE_DEVICE,
368 	    ("Unexpected ACPI object"));
369 	status = acpi_GetInteger(sc->cpu_handle, "_UID", &sc->cpu_acpi_id);
370 	if (ACPI_FAILURE(status)) {
371 	    device_printf(dev, "Device object has bad value - %s\n",
372 		AcpiFormatException(status));
373 	    return (ENXIO);
374 	}
375 	sc->cpu_p_blk = 0;
376 	sc->cpu_p_blk_len = 0;
377     }
378     ACPI_DEBUG_PRINT((ACPI_DB_INFO, "acpi_cpu%d: P_BLK at %#x/%d\n",
379 		     device_get_unit(dev), sc->cpu_p_blk, sc->cpu_p_blk_len));
380 
381     /*
382      * If this is the first cpu we attach, create and initialize the generic
383      * resources that will be used by all acpi cpu devices.
384      */
385     if (device_get_unit(dev) == 0) {
386 	/* Assume we won't be using generic Cx mode by default */
387 	cpu_cx_generic = FALSE;
388 
389 	/* Install hw.acpi.cpu sysctl tree */
390 	acpi_sc = acpi_device_get_parent_softc(dev);
391 	sysctl_ctx_init(&cpu_sysctl_ctx);
392 	cpu_sysctl_tree = SYSCTL_ADD_NODE(&cpu_sysctl_ctx,
393 	    SYSCTL_CHILDREN(acpi_sc->acpi_sysctl_tree), OID_AUTO, "cpu",
394 	    CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "node for CPU children");
395     }
396 
397     /*
398      * Before calling any CPU methods, collect child driver feature hints
399      * and notify ACPI of them.  We support unified SMP power control
400      * so advertise this ourselves.  Note this is not the same as independent
401      * SMP control where each CPU can have different settings.
402      */
403     sc->cpu_features = ACPI_CAP_SMP_SAME | ACPI_CAP_SMP_SAME_C3 |
404       ACPI_CAP_C1_IO_HALT;
405 
406 #if defined(__i386__) || defined(__amd64__)
407     /*
408      * Ask for MWAIT modes if not disabled and interrupts work
409      * reasonable with MWAIT.
410      */
411     if (!acpi_disabled("mwait") && cpu_mwait_usable())
412 	sc->cpu_features |= ACPI_CAP_SMP_C1_NATIVE | ACPI_CAP_SMP_C3_NATIVE;
413 #endif
414 
415     if (devclass_get_drivers(acpi_cpu_devclass, &drivers, &drv_count) == 0) {
416 	for (i = 0; i < drv_count; i++) {
417 	    if (ACPI_GET_FEATURES(drivers[i], &features) == 0)
418 		sc->cpu_features |= features;
419 	}
420 	free(drivers, M_TEMP);
421     }
422 
423     /*
424      * CPU capabilities are specified in
425      * Intel Processor Vendor-Specific ACPI Interface Specification.
426      */
427     if (sc->cpu_features) {
428 	cap_set[1] = sc->cpu_features;
429 	status = acpi_EvaluateOSC(sc->cpu_handle, cpu_oscuuid, 1, 2, cap_set,
430 	    cap_set, false);
431 	if (ACPI_SUCCESS(status)) {
432 	    if (cap_set[0] != 0)
433 		device_printf(dev, "_OSC returned status %#x\n", cap_set[0]);
434 	}
435 	else {
436 	    arglist.Pointer = &arg;
437 	    arglist.Count = 1;
438 	    arg.Type = ACPI_TYPE_BUFFER;
439 	    arg.Buffer.Length = sizeof(cap_set);
440 	    arg.Buffer.Pointer = (uint8_t *)cap_set;
441 	    cap_set[0] = 1; /* revision */
442 	    cap_set[1] = 1; /* number of capabilities integers */
443 	    cap_set[2] = sc->cpu_features;
444 	    AcpiEvaluateObject(sc->cpu_handle, "_PDC", &arglist, NULL);
445 	}
446     }
447 
448     /* Probe for Cx state support. */
449     acpi_cpu_cx_probe(sc);
450 
451     return (0);
452 }
453 
454 static void
455 acpi_cpu_postattach(void *unused __unused)
456 {
457     device_t *devices;
458     int err;
459     int i, n;
460     int attached;
461 
462     err = devclass_get_devices(acpi_cpu_devclass, &devices, &n);
463     if (err != 0) {
464 	printf("devclass_get_devices(acpi_cpu_devclass) failed\n");
465 	return;
466     }
467     attached = 0;
468     for (i = 0; i < n; i++)
469 	if (device_is_attached(devices[i]) &&
470 	    device_get_driver(devices[i]) == &acpi_cpu_driver)
471 	    attached = 1;
472     for (i = 0; i < n; i++)
473 	bus_generic_probe(devices[i]);
474     for (i = 0; i < n; i++)
475 	bus_generic_attach(devices[i]);
476     free(devices, M_TEMP);
477 
478     if (attached) {
479 #ifdef EARLY_AP_STARTUP
480 	acpi_cpu_startup(NULL);
481 #else
482 	/* Queue post cpu-probing task handler */
483 	AcpiOsExecute(OSL_NOTIFY_HANDLER, acpi_cpu_startup, NULL);
484 #endif
485     }
486 }
487 
488 SYSINIT(acpi_cpu, SI_SUB_CONFIGURE, SI_ORDER_MIDDLE,
489     acpi_cpu_postattach, NULL);
490 
491 static void
492 disable_idle(struct acpi_cpu_softc *sc)
493 {
494     cpuset_t cpuset;
495 
496     CPU_SETOF(sc->cpu_pcpu->pc_cpuid, &cpuset);
497     sc->cpu_disable_idle = TRUE;
498 
499     /*
500      * Ensure that the CPU is not in idle state or in acpi_cpu_idle().
501      * Note that this code depends on the fact that the rendezvous IPI
502      * can not penetrate context where interrupts are disabled and acpi_cpu_idle
503      * is called and executed in such a context with interrupts being re-enabled
504      * right before return.
505      */
506     smp_rendezvous_cpus(cpuset, smp_no_rendezvous_barrier, NULL,
507 	smp_no_rendezvous_barrier, NULL);
508 }
509 
510 static void
511 enable_idle(struct acpi_cpu_softc *sc)
512 {
513 
514     sc->cpu_disable_idle = FALSE;
515 }
516 
517 #if defined(__i386__) || defined(__amd64__)
518 static int
519 is_idle_disabled(struct acpi_cpu_softc *sc)
520 {
521 
522     return (sc->cpu_disable_idle);
523 }
524 #endif
525 
526 /*
527  * Disable any entry to the idle function during suspend and re-enable it
528  * during resume.
529  */
530 static int
531 acpi_cpu_suspend(device_t dev)
532 {
533     int error;
534 
535     error = bus_generic_suspend(dev);
536     if (error)
537 	return (error);
538     disable_idle(device_get_softc(dev));
539     return (0);
540 }
541 
542 static int
543 acpi_cpu_resume(device_t dev)
544 {
545 
546     enable_idle(device_get_softc(dev));
547     return (bus_generic_resume(dev));
548 }
549 
550 /*
551  * Find the processor associated with a given ACPI ID.  By default,
552  * use the MADT to map ACPI IDs to APIC IDs and use that to locate a
553  * processor.  Some systems have inconsistent ASL and MADT however.
554  * For these systems the cpu_unordered tunable can be set in which
555  * case we assume that Processor objects are listed in the same order
556  * in both the MADT and ASL.
557  */
558 static int
559 acpi_pcpu_get_id(device_t dev, uint32_t *acpi_id, uint32_t *cpu_id)
560 {
561     struct pcpu	*pc;
562     uint32_t	 i, idx;
563 
564     KASSERT(acpi_id != NULL, ("Null acpi_id"));
565     KASSERT(cpu_id != NULL, ("Null cpu_id"));
566     idx = device_get_unit(dev);
567 
568     /*
569      * If pc_acpi_id for CPU 0 is not initialized (e.g. a non-APIC
570      * UP box) use the ACPI ID from the first processor we find.
571      */
572     if (idx == 0 && mp_ncpus == 1) {
573 	pc = pcpu_find(0);
574 	if (pc->pc_acpi_id == 0xffffffff)
575 	    pc->pc_acpi_id = *acpi_id;
576 	*cpu_id = 0;
577 	return (0);
578     }
579 
580     CPU_FOREACH(i) {
581 	pc = pcpu_find(i);
582 	KASSERT(pc != NULL, ("no pcpu data for %d", i));
583 	if (cpu_unordered) {
584 	    if (idx-- == 0) {
585 		/*
586 		 * If pc_acpi_id doesn't match the ACPI ID from the
587 		 * ASL, prefer the MADT-derived value.
588 		 */
589 		if (pc->pc_acpi_id != *acpi_id)
590 		    *acpi_id = pc->pc_acpi_id;
591 		*cpu_id = pc->pc_cpuid;
592 		return (0);
593 	    }
594 	} else {
595 	    if (pc->pc_acpi_id == *acpi_id) {
596 		if (bootverbose)
597 		    device_printf(dev,
598 			"Processor %s (ACPI ID %u) -> APIC ID %d\n",
599 			acpi_name(acpi_get_handle(dev)), *acpi_id,
600 			pc->pc_cpuid);
601 		*cpu_id = pc->pc_cpuid;
602 		return (0);
603 	    }
604 	}
605     }
606 
607     if (bootverbose)
608 	printf("ACPI: Processor %s (ACPI ID %u) ignored\n",
609 	    acpi_name(acpi_get_handle(dev)), *acpi_id);
610 
611     return (ESRCH);
612 }
613 
614 static struct resource_list *
615 acpi_cpu_get_rlist(device_t dev, device_t child)
616 {
617     struct acpi_cpu_device *ad;
618 
619     ad = device_get_ivars(child);
620     if (ad == NULL)
621 	return (NULL);
622     return (&ad->ad_rl);
623 }
624 
625 static device_t
626 acpi_cpu_add_child(device_t dev, u_int order, const char *name, int unit)
627 {
628     struct acpi_cpu_device *ad;
629     device_t child;
630 
631     if ((ad = malloc(sizeof(*ad), M_TEMP, M_NOWAIT | M_ZERO)) == NULL)
632 	return (NULL);
633 
634     resource_list_init(&ad->ad_rl);
635 
636     child = device_add_child_ordered(dev, order, name, unit);
637     if (child != NULL)
638 	device_set_ivars(child, ad);
639     else
640 	free(ad, M_TEMP);
641     return (child);
642 }
643 
644 static int
645 acpi_cpu_read_ivar(device_t dev, device_t child, int index, uintptr_t *result)
646 {
647     struct acpi_cpu_softc *sc;
648 
649     sc = device_get_softc(dev);
650     switch (index) {
651     case ACPI_IVAR_HANDLE:
652 	*result = (uintptr_t)sc->cpu_handle;
653 	break;
654     case CPU_IVAR_PCPU:
655 	*result = (uintptr_t)sc->cpu_pcpu;
656 	break;
657 #if defined(__amd64__) || defined(__i386__)
658     case CPU_IVAR_NOMINAL_MHZ:
659 	if (tsc_is_invariant) {
660 	    *result = (uintptr_t)(atomic_load_acq_64(&tsc_freq) / 1000000);
661 	    break;
662 	}
663 	/* FALLTHROUGH */
664 #endif
665     default:
666 	return (ENOENT);
667     }
668     return (0);
669 }
670 
671 static int
672 acpi_cpu_shutdown(device_t dev)
673 {
674     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
675 
676     /* Allow children to shutdown first. */
677     bus_generic_shutdown(dev);
678 
679     /*
680      * Disable any entry to the idle function.
681      */
682     disable_idle(device_get_softc(dev));
683 
684     /*
685      * CPU devices are not truly detached and remain referenced,
686      * so their resources are not freed.
687      */
688 
689     return_VALUE (0);
690 }
691 
692 static void
693 acpi_cpu_cx_probe(struct acpi_cpu_softc *sc)
694 {
695     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
696 
697     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
698     sc->cpu_prev_sleep = 1000000;
699     sc->cpu_cx_lowest = 0;
700     sc->cpu_cx_lowest_lim = 0;
701 
702     /*
703      * Check for the ACPI 2.0 _CST sleep states object. If we can't find
704      * any, we'll revert to generic FADT/P_BLK Cx control method which will
705      * be handled by acpi_cpu_startup. We need to defer to after having
706      * probed all the cpus in the system before probing for generic Cx
707      * states as we may already have found cpus with valid _CST packages
708      */
709     if (!cpu_cx_generic && acpi_cpu_cx_cst(sc) != 0) {
710 	/*
711 	 * We were unable to find a _CST package for this cpu or there
712 	 * was an error parsing it. Switch back to generic mode.
713 	 */
714 	cpu_cx_generic = TRUE;
715 	if (bootverbose)
716 	    device_printf(sc->cpu_dev, "switching to generic Cx mode\n");
717     }
718 
719     /*
720      * TODO: _CSD Package should be checked here.
721      */
722 }
723 
724 static void
725 acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc)
726 {
727     ACPI_GENERIC_ADDRESS	 gas;
728     struct acpi_cx		*cx_ptr;
729 
730     sc->cpu_cx_count = 0;
731     cx_ptr = sc->cpu_cx_states;
732 
733     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
734     sc->cpu_prev_sleep = 1000000;
735 
736     /* C1 has been required since just after ACPI 1.0 */
737     cx_ptr->type = ACPI_STATE_C1;
738     cx_ptr->trans_lat = 0;
739     cx_ptr++;
740     sc->cpu_non_c2 = sc->cpu_cx_count;
741     sc->cpu_non_c3 = sc->cpu_cx_count;
742     sc->cpu_cx_count++;
743 
744     /*
745      * The spec says P_BLK must be 6 bytes long.  However, some systems
746      * use it to indicate a fractional set of features present so we
747      * take 5 as C2.  Some may also have a value of 7 to indicate
748      * another C3 but most use _CST for this (as required) and having
749      * "only" C1-C3 is not a hardship.
750      */
751     if (sc->cpu_p_blk_len < 5)
752 	return;
753 
754     /* Validate and allocate resources for C2 (P_LVL2). */
755     gas.SpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
756     gas.BitWidth = 8;
757     if (AcpiGbl_FADT.C2Latency <= 100) {
758 	gas.Address = sc->cpu_p_blk + 4;
759 	cx_ptr->res_rid = 0;
760 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &cx_ptr->res_rid,
761 	    &gas, &cx_ptr->p_lvlx, RF_SHAREABLE);
762 	if (cx_ptr->p_lvlx != NULL) {
763 	    cx_ptr->type = ACPI_STATE_C2;
764 	    cx_ptr->trans_lat = AcpiGbl_FADT.C2Latency;
765 	    cx_ptr++;
766 	    sc->cpu_non_c3 = sc->cpu_cx_count;
767 	    sc->cpu_cx_count++;
768 	}
769     }
770     if (sc->cpu_p_blk_len < 6)
771 	return;
772 
773     /* Validate and allocate resources for C3 (P_LVL3). */
774     if (AcpiGbl_FADT.C3Latency <= 1000 && !(cpu_quirks & CPU_QUIRK_NO_C3)) {
775 	gas.Address = sc->cpu_p_blk + 5;
776 	cx_ptr->res_rid = 1;
777 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &cx_ptr->res_rid,
778 	    &gas, &cx_ptr->p_lvlx, RF_SHAREABLE);
779 	if (cx_ptr->p_lvlx != NULL) {
780 	    cx_ptr->type = ACPI_STATE_C3;
781 	    cx_ptr->trans_lat = AcpiGbl_FADT.C3Latency;
782 	    cx_ptr++;
783 	    sc->cpu_cx_count++;
784 	}
785     }
786 }
787 
788 #if defined(__i386__) || defined(__amd64__)
789 static void
790 acpi_cpu_cx_cst_mwait(struct acpi_cx *cx_ptr, uint64_t address, int accsize)
791 {
792 
793 	cx_ptr->do_mwait = true;
794 	cx_ptr->mwait_hint = address & 0xffffffff;
795 	cx_ptr->mwait_hw_coord = (accsize & CST_FFH_MWAIT_HW_COORD) != 0;
796 	cx_ptr->mwait_bm_avoidance = (accsize & CST_FFH_MWAIT_BM_AVOID) != 0;
797 }
798 #endif
799 
800 static void
801 acpi_cpu_cx_cst_free_plvlx(device_t cpu_dev, struct acpi_cx *cx_ptr)
802 {
803 
804 	if (cx_ptr->p_lvlx == NULL)
805 		return;
806 	bus_release_resource(cpu_dev, cx_ptr->res_type, cx_ptr->res_rid,
807 	    cx_ptr->p_lvlx);
808 	cx_ptr->p_lvlx = NULL;
809 }
810 
811 /*
812  * Parse a _CST package and set up its Cx states.  Since the _CST object
813  * can change dynamically, our notify handler may call this function
814  * to clean up and probe the new _CST package.
815  */
816 static int
817 acpi_cpu_cx_cst(struct acpi_cpu_softc *sc)
818 {
819     struct	 acpi_cx *cx_ptr;
820     ACPI_STATUS	 status;
821     ACPI_BUFFER	 buf;
822     ACPI_OBJECT	*top;
823     ACPI_OBJECT	*pkg;
824     uint32_t	 count;
825     int		 i;
826 #if defined(__i386__) || defined(__amd64__)
827     uint64_t	 address;
828     int		 vendor, class, accsize;
829 #endif
830 
831     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
832 
833     buf.Pointer = NULL;
834     buf.Length = ACPI_ALLOCATE_BUFFER;
835     status = AcpiEvaluateObject(sc->cpu_handle, "_CST", NULL, &buf);
836     if (ACPI_FAILURE(status))
837 	return (ENXIO);
838 
839     /* _CST is a package with a count and at least one Cx package. */
840     top = (ACPI_OBJECT *)buf.Pointer;
841     if (!ACPI_PKG_VALID(top, 2) || acpi_PkgInt32(top, 0, &count) != 0) {
842 	device_printf(sc->cpu_dev, "invalid _CST package\n");
843 	AcpiOsFree(buf.Pointer);
844 	return (ENXIO);
845     }
846     if (count != top->Package.Count - 1) {
847 	device_printf(sc->cpu_dev, "invalid _CST state count (%d != %d)\n",
848 	       count, top->Package.Count - 1);
849 	count = top->Package.Count - 1;
850     }
851     if (count > MAX_CX_STATES) {
852 	device_printf(sc->cpu_dev, "_CST has too many states (%d)\n", count);
853 	count = MAX_CX_STATES;
854     }
855 
856     sc->cpu_non_c2 = 0;
857     sc->cpu_non_c3 = 0;
858     sc->cpu_cx_count = 0;
859     cx_ptr = sc->cpu_cx_states;
860 
861     /*
862      * C1 has been required since just after ACPI 1.0.
863      * Reserve the first slot for it.
864      */
865     cx_ptr->type = ACPI_STATE_C0;
866     cx_ptr++;
867     sc->cpu_cx_count++;
868 
869     /* Set up all valid states. */
870     for (i = 0; i < count; i++) {
871 	pkg = &top->Package.Elements[i + 1];
872 	if (!ACPI_PKG_VALID(pkg, 4) ||
873 	    acpi_PkgInt32(pkg, 1, &cx_ptr->type) != 0 ||
874 	    acpi_PkgInt32(pkg, 2, &cx_ptr->trans_lat) != 0 ||
875 	    acpi_PkgInt32(pkg, 3, &cx_ptr->power) != 0) {
876 
877 	    device_printf(sc->cpu_dev, "skipping invalid Cx state package\n");
878 	    continue;
879 	}
880 
881 	/* Validate the state to see if we should use it. */
882 	switch (cx_ptr->type) {
883 	case ACPI_STATE_C1:
884 	    acpi_cpu_cx_cst_free_plvlx(sc->cpu_dev, cx_ptr);
885 #if defined(__i386__) || defined(__amd64__)
886 	    if (acpi_PkgFFH_IntelCpu(pkg, 0, &vendor, &class, &address,
887 	      &accsize) == 0 && vendor == CST_FFH_VENDOR_INTEL) {
888 		if (class == CST_FFH_INTEL_CL_C1IO) {
889 		    /* C1 I/O then Halt */
890 		    cx_ptr->res_rid = sc->cpu_cx_count;
891 		    bus_set_resource(sc->cpu_dev, SYS_RES_IOPORT,
892 		      cx_ptr->res_rid, address, 1);
893 		    cx_ptr->p_lvlx = bus_alloc_resource_any(sc->cpu_dev,
894 		      SYS_RES_IOPORT, &cx_ptr->res_rid, RF_ACTIVE |
895 		      RF_SHAREABLE);
896 		    if (cx_ptr->p_lvlx == NULL) {
897 			bus_delete_resource(sc->cpu_dev, SYS_RES_IOPORT,
898 			  cx_ptr->res_rid);
899 			device_printf(sc->cpu_dev,
900 			  "C1 I/O failed to allocate port %d, "
901 			  "degrading to C1 Halt", (int)address);
902 		    }
903 		} else if (class == CST_FFH_INTEL_CL_MWAIT) {
904 		    acpi_cpu_cx_cst_mwait(cx_ptr, address, accsize);
905 		}
906 	    }
907 #endif
908 	    if (sc->cpu_cx_states[0].type == ACPI_STATE_C0) {
909 		/* This is the first C1 state.  Use the reserved slot. */
910 		sc->cpu_cx_states[0] = *cx_ptr;
911 	    } else {
912 		sc->cpu_non_c2 = sc->cpu_cx_count;
913 		sc->cpu_non_c3 = sc->cpu_cx_count;
914 		cx_ptr++;
915 		sc->cpu_cx_count++;
916 	    }
917 	    continue;
918 	case ACPI_STATE_C2:
919 	    sc->cpu_non_c3 = sc->cpu_cx_count;
920 	    break;
921 	case ACPI_STATE_C3:
922 	default:
923 	    if ((cpu_quirks & CPU_QUIRK_NO_C3) != 0) {
924 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
925 				 "acpi_cpu%d: C3[%d] not available.\n",
926 				 device_get_unit(sc->cpu_dev), i));
927 		continue;
928 	    }
929 	    break;
930 	}
931 
932 	/* Free up any previous register. */
933 	acpi_cpu_cx_cst_free_plvlx(sc->cpu_dev, cx_ptr);
934 
935 	/* Allocate the control register for C2 or C3. */
936 #if defined(__i386__) || defined(__amd64__)
937 	if (acpi_PkgFFH_IntelCpu(pkg, 0, &vendor, &class, &address,
938 	  &accsize) == 0 && vendor == CST_FFH_VENDOR_INTEL &&
939 	  class == CST_FFH_INTEL_CL_MWAIT) {
940 	    /* Native C State Instruction use (mwait) */
941 	    acpi_cpu_cx_cst_mwait(cx_ptr, address, accsize);
942 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
943 	      "acpi_cpu%d: Got C%d/mwait - %d latency\n",
944 	      device_get_unit(sc->cpu_dev), cx_ptr->type, cx_ptr->trans_lat));
945 	    cx_ptr++;
946 	    sc->cpu_cx_count++;
947 	} else
948 #endif
949 	{
950 	    cx_ptr->res_rid = sc->cpu_cx_count;
951 	    acpi_PkgGas(sc->cpu_dev, pkg, 0, &cx_ptr->res_type,
952 		&cx_ptr->res_rid, &cx_ptr->p_lvlx, RF_SHAREABLE);
953 	    if (cx_ptr->p_lvlx) {
954 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
955 		     "acpi_cpu%d: Got C%d - %d latency\n",
956 		     device_get_unit(sc->cpu_dev), cx_ptr->type,
957 		     cx_ptr->trans_lat));
958 		cx_ptr++;
959 		sc->cpu_cx_count++;
960 	    }
961 	}
962     }
963     AcpiOsFree(buf.Pointer);
964 
965     /* If C1 state was not found, we need one now. */
966     cx_ptr = sc->cpu_cx_states;
967     if (cx_ptr->type == ACPI_STATE_C0) {
968 	cx_ptr->type = ACPI_STATE_C1;
969 	cx_ptr->trans_lat = 0;
970     }
971 
972     return (0);
973 }
974 
975 /*
976  * Call this *after* all CPUs have been attached.
977  */
978 static void
979 acpi_cpu_startup(void *arg)
980 {
981     struct acpi_cpu_softc *sc;
982     int i;
983 
984     /* Get set of CPU devices */
985     devclass_get_devices(acpi_cpu_devclass, &cpu_devices, &cpu_ndevices);
986 
987     /*
988      * Setup any quirks that might necessary now that we have probed
989      * all the CPUs
990      */
991     acpi_cpu_quirks();
992 
993     if (cpu_cx_generic) {
994 	/*
995 	 * We are using generic Cx mode, probe for available Cx states
996 	 * for all processors.
997 	 */
998 	for (i = 0; i < cpu_ndevices; i++) {
999 	    sc = device_get_softc(cpu_devices[i]);
1000 	    acpi_cpu_generic_cx_probe(sc);
1001 	}
1002     } else {
1003 	/*
1004 	 * We are using _CST mode, remove C3 state if necessary.
1005 	 * As we now know for sure that we will be using _CST mode
1006 	 * install our notify handler.
1007 	 */
1008 	for (i = 0; i < cpu_ndevices; i++) {
1009 	    sc = device_get_softc(cpu_devices[i]);
1010 	    if (cpu_quirks & CPU_QUIRK_NO_C3) {
1011 		sc->cpu_cx_count = min(sc->cpu_cx_count, sc->cpu_non_c3 + 1);
1012 	    }
1013 	    AcpiInstallNotifyHandler(sc->cpu_handle, ACPI_DEVICE_NOTIFY,
1014 		acpi_cpu_notify, sc);
1015 	}
1016     }
1017 
1018     /* Perform Cx final initialization. */
1019     for (i = 0; i < cpu_ndevices; i++) {
1020 	sc = device_get_softc(cpu_devices[i]);
1021 	acpi_cpu_startup_cx(sc);
1022     }
1023 
1024     /* Add a sysctl handler to handle global Cx lowest setting */
1025     SYSCTL_ADD_PROC(&cpu_sysctl_ctx, SYSCTL_CHILDREN(cpu_sysctl_tree),
1026 	OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
1027 	NULL, 0, acpi_cpu_global_cx_lowest_sysctl, "A",
1028 	"Global lowest Cx sleep state to use");
1029 
1030     /* Take over idling from cpu_idle_default(). */
1031     cpu_cx_lowest_lim = 0;
1032     for (i = 0; i < cpu_ndevices; i++) {
1033 	sc = device_get_softc(cpu_devices[i]);
1034 	enable_idle(sc);
1035     }
1036 #if defined(__i386__) || defined(__amd64__)
1037     cpu_idle_hook = acpi_cpu_idle;
1038 #endif
1039 }
1040 
1041 static void
1042 acpi_cpu_cx_list(struct acpi_cpu_softc *sc)
1043 {
1044     struct sbuf sb;
1045     int i;
1046 
1047     /*
1048      * Set up the list of Cx states
1049      */
1050     sbuf_new(&sb, sc->cpu_cx_supported, sizeof(sc->cpu_cx_supported),
1051 	SBUF_FIXEDLEN);
1052     for (i = 0; i < sc->cpu_cx_count; i++)
1053 	sbuf_printf(&sb, "C%d/%d/%d ", i + 1, sc->cpu_cx_states[i].type,
1054 	    sc->cpu_cx_states[i].trans_lat);
1055     sbuf_trim(&sb);
1056     sbuf_finish(&sb);
1057 }
1058 
1059 static void
1060 acpi_cpu_startup_cx(struct acpi_cpu_softc *sc)
1061 {
1062     acpi_cpu_cx_list(sc);
1063 
1064     SYSCTL_ADD_STRING(&sc->cpu_sysctl_ctx,
1065 		      SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
1066 		      OID_AUTO, "cx_supported", CTLFLAG_RD,
1067 		      sc->cpu_cx_supported, 0,
1068 		      "Cx/microsecond values for supported Cx states");
1069     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
1070         SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)), OID_AUTO,
1071 	"cx_lowest", CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
1072 	(void *)sc, 0, acpi_cpu_cx_lowest_sysctl, "A",
1073 	"lowest Cx sleep state to use");
1074     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
1075         SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)), OID_AUTO,
1076 	"cx_usage", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
1077 	(void *)sc, 0, acpi_cpu_usage_sysctl, "A",
1078 	"percent usage for each Cx state");
1079     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
1080         SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)), OID_AUTO,
1081 	"cx_usage_counters", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
1082 	(void *)sc, 0, acpi_cpu_usage_counters_sysctl, "A",
1083 	"Cx sleep state counters");
1084 #if defined(__i386__) || defined(__amd64__)
1085     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
1086         SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)), OID_AUTO,
1087 	"cx_method", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
1088 	(void *)sc, 0, acpi_cpu_method_sysctl, "A", "Cx entrance methods");
1089 #endif
1090 
1091     /* Signal platform that we can handle _CST notification. */
1092     if (!cpu_cx_generic && cpu_cst_cnt != 0) {
1093 	ACPI_LOCK(acpi);
1094 	AcpiOsWritePort(cpu_smi_cmd, cpu_cst_cnt, 8);
1095 	ACPI_UNLOCK(acpi);
1096     }
1097 }
1098 
1099 #if defined(__i386__) || defined(__amd64__)
1100 /*
1101  * Idle the CPU in the lowest state possible.  This function is called with
1102  * interrupts disabled.  Note that once it re-enables interrupts, a task
1103  * switch can occur so do not access shared data (i.e. the softc) after
1104  * interrupts are re-enabled.
1105  */
1106 static void
1107 acpi_cpu_idle(sbintime_t sbt)
1108 {
1109     struct	acpi_cpu_softc *sc;
1110     struct	acpi_cx *cx_next;
1111     uint64_t	cputicks;
1112     uint32_t	start_time, end_time;
1113     ACPI_STATUS	status;
1114     int		bm_active, cx_next_idx, i, us;
1115 
1116     /*
1117      * Look up our CPU id to get our softc.  If it's NULL, we'll use C1
1118      * since there is no ACPI processor object for this CPU.  This occurs
1119      * for logical CPUs in the HTT case.
1120      */
1121     sc = cpu_softc[PCPU_GET(cpuid)];
1122     if (sc == NULL) {
1123 	acpi_cpu_c1();
1124 	return;
1125     }
1126 
1127     /* If disabled, take the safe path. */
1128     if (is_idle_disabled(sc)) {
1129 	acpi_cpu_c1();
1130 	return;
1131     }
1132 
1133     /* Find the lowest state that has small enough latency. */
1134     us = sc->cpu_prev_sleep;
1135     if (sbt >= 0 && us > (sbt >> 12))
1136 	us = (sbt >> 12);
1137     cx_next_idx = 0;
1138     if (cpu_disable_c2_sleep)
1139 	i = min(sc->cpu_cx_lowest, sc->cpu_non_c2);
1140     else if (cpu_disable_c3_sleep)
1141 	i = min(sc->cpu_cx_lowest, sc->cpu_non_c3);
1142     else
1143 	i = sc->cpu_cx_lowest;
1144     for (; i >= 0; i--) {
1145 	if (sc->cpu_cx_states[i].trans_lat * 3 <= us) {
1146 	    cx_next_idx = i;
1147 	    break;
1148 	}
1149     }
1150 
1151     /*
1152      * Check for bus master activity.  If there was activity, clear
1153      * the bit and use the lowest non-C3 state.  Note that the USB
1154      * driver polling for new devices keeps this bit set all the
1155      * time if USB is loaded.
1156      */
1157     if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0 &&
1158 	cx_next_idx > sc->cpu_non_c3) {
1159 	status = AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, &bm_active);
1160 	if (ACPI_SUCCESS(status) && bm_active != 0) {
1161 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, 1);
1162 	    cx_next_idx = sc->cpu_non_c3;
1163 	}
1164     }
1165 
1166     /* Select the next state and update statistics. */
1167     cx_next = &sc->cpu_cx_states[cx_next_idx];
1168     sc->cpu_cx_stats[cx_next_idx]++;
1169     KASSERT(cx_next->type != ACPI_STATE_C0, ("acpi_cpu_idle: C0 sleep"));
1170 
1171     /*
1172      * Execute HLT (or equivalent) and wait for an interrupt.  We can't
1173      * precisely calculate the time spent in C1 since the place we wake up
1174      * is an ISR.  Assume we slept no more then half of quantum, unless
1175      * we are called inside critical section, delaying context switch.
1176      */
1177     if (cx_next->type == ACPI_STATE_C1) {
1178 	cputicks = cpu_ticks();
1179 	if (cx_next->p_lvlx != NULL) {
1180 	    /* C1 I/O then Halt */
1181 	    CPU_GET_REG(cx_next->p_lvlx, 1);
1182 	}
1183 	if (cx_next->do_mwait)
1184 	    acpi_cpu_idle_mwait(cx_next->mwait_hint);
1185 	else
1186 	    acpi_cpu_c1();
1187 	end_time = ((cpu_ticks() - cputicks) << 20) / cpu_tickrate();
1188 	if (curthread->td_critnest == 0)
1189 		end_time = min(end_time, 500000 / hz);
1190 	/* acpi_cpu_c1() returns with interrupts enabled. */
1191 	if (cx_next->do_mwait)
1192 	    ACPI_ENABLE_IRQS();
1193 	sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + end_time) / 4;
1194 	return;
1195     }
1196 
1197     /*
1198      * For C3, disable bus master arbitration and enable bus master wake
1199      * if BM control is available, otherwise flush the CPU cache.
1200      */
1201     if (cx_next->type == ACPI_STATE_C3 || cx_next->mwait_bm_avoidance) {
1202 	if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
1203 	    AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 1);
1204 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
1205 	} else
1206 	    ACPI_FLUSH_CPU_CACHE();
1207     }
1208 
1209     /*
1210      * Read from P_LVLx to enter C2(+), checking time spent asleep.
1211      * Use the ACPI timer for measuring sleep time.  Since we need to
1212      * get the time very close to the CPU start/stop clock logic, this
1213      * is the only reliable time source.
1214      */
1215     if (cx_next->type == ACPI_STATE_C3) {
1216 	AcpiGetTimer(&start_time);
1217 	cputicks = 0;
1218     } else {
1219 	start_time = 0;
1220 	cputicks = cpu_ticks();
1221     }
1222     if (cx_next->do_mwait)
1223 	acpi_cpu_idle_mwait(cx_next->mwait_hint);
1224     else
1225 	CPU_GET_REG(cx_next->p_lvlx, 1);
1226 
1227     /*
1228      * Read the end time twice.  Since it may take an arbitrary time
1229      * to enter the idle state, the first read may be executed before
1230      * the processor has stopped.  Doing it again provides enough
1231      * margin that we are certain to have a correct value.
1232      */
1233     AcpiGetTimer(&end_time);
1234     if (cx_next->type == ACPI_STATE_C3) {
1235 	AcpiGetTimer(&end_time);
1236 	AcpiGetTimerDuration(start_time, end_time, &end_time);
1237     } else
1238 	end_time = ((cpu_ticks() - cputicks) << 20) / cpu_tickrate();
1239 
1240     /* Enable bus master arbitration and disable bus master wakeup. */
1241     if ((cx_next->type == ACPI_STATE_C3 || cx_next->mwait_bm_avoidance) &&
1242       (cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
1243 	AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 0);
1244 	AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1245     }
1246     ACPI_ENABLE_IRQS();
1247 
1248     sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + PM_USEC(end_time)) / 4;
1249 }
1250 #endif
1251 
1252 /*
1253  * Re-evaluate the _CST object when we are notified that it changed.
1254  */
1255 static void
1256 acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context)
1257 {
1258     struct acpi_cpu_softc *sc = (struct acpi_cpu_softc *)context;
1259 
1260     if (notify != ACPI_NOTIFY_CX_STATES)
1261 	return;
1262 
1263     /*
1264      * C-state data for target CPU is going to be in flux while we execute
1265      * acpi_cpu_cx_cst, so disable entering acpi_cpu_idle.
1266      * Also, it may happen that multiple ACPI taskqueues may concurrently
1267      * execute notifications for the same CPU.  ACPI_SERIAL is used to
1268      * protect against that.
1269      */
1270     ACPI_SERIAL_BEGIN(cpu);
1271     disable_idle(sc);
1272 
1273     /* Update the list of Cx states. */
1274     acpi_cpu_cx_cst(sc);
1275     acpi_cpu_cx_list(sc);
1276     acpi_cpu_set_cx_lowest(sc);
1277 
1278     enable_idle(sc);
1279     ACPI_SERIAL_END(cpu);
1280 
1281     acpi_UserNotify("PROCESSOR", sc->cpu_handle, notify);
1282 }
1283 
1284 static void
1285 acpi_cpu_quirks(void)
1286 {
1287     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
1288 
1289     /*
1290      * Bus mastering arbitration control is needed to keep caches coherent
1291      * while sleeping in C3.  If it's not present but a working flush cache
1292      * instruction is present, flush the caches before entering C3 instead.
1293      * Otherwise, just disable C3 completely.
1294      */
1295     if (AcpiGbl_FADT.Pm2ControlBlock == 0 ||
1296 	AcpiGbl_FADT.Pm2ControlLength == 0) {
1297 	if ((AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD) &&
1298 	    (AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD_FLUSH) == 0) {
1299 	    cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1300 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1301 		"acpi_cpu: no BM control, using flush cache method\n"));
1302 	} else {
1303 	    cpu_quirks |= CPU_QUIRK_NO_C3;
1304 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1305 		"acpi_cpu: no BM control, C3 not available\n"));
1306 	}
1307     }
1308 
1309     /*
1310      * If we are using generic Cx mode, C3 on multiple CPUs requires using
1311      * the expensive flush cache instruction.
1312      */
1313     if (cpu_cx_generic && mp_ncpus > 1) {
1314 	cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1315 	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1316 	    "acpi_cpu: SMP, using flush cache mode for C3\n"));
1317     }
1318 
1319     /* Look for various quirks of the PIIX4 part. */
1320     acpi_cpu_quirks_piix4();
1321 }
1322 
1323 static void
1324 acpi_cpu_quirks_piix4(void)
1325 {
1326 #ifdef __i386__
1327     device_t acpi_dev;
1328     uint32_t val;
1329     ACPI_STATUS status;
1330 
1331     acpi_dev = pci_find_device(PCI_VENDOR_INTEL, PCI_DEVICE_82371AB_3);
1332     if (acpi_dev != NULL) {
1333 	switch (pci_get_revid(acpi_dev)) {
1334 	/*
1335 	 * Disable C3 support for all PIIX4 chipsets.  Some of these parts
1336 	 * do not report the BMIDE status to the BM status register and
1337 	 * others have a livelock bug if Type-F DMA is enabled.  Linux
1338 	 * works around the BMIDE bug by reading the BM status directly
1339 	 * but we take the simpler approach of disabling C3 for these
1340 	 * parts.
1341 	 *
1342 	 * See erratum #18 ("C3 Power State/BMIDE and Type-F DMA
1343 	 * Livelock") from the January 2002 PIIX4 specification update.
1344 	 * Applies to all PIIX4 models.
1345 	 *
1346 	 * Also, make sure that all interrupts cause a "Stop Break"
1347 	 * event to exit from C2 state.
1348 	 * Also, BRLD_EN_BM (ACPI_BITREG_BUS_MASTER_RLD in ACPI-speak)
1349 	 * should be set to zero, otherwise it causes C2 to short-sleep.
1350 	 * PIIX4 doesn't properly support C3 and bus master activity
1351 	 * need not break out of C2.
1352 	 */
1353 	case PCI_REVISION_A_STEP:
1354 	case PCI_REVISION_B_STEP:
1355 	case PCI_REVISION_4E:
1356 	case PCI_REVISION_4M:
1357 	    cpu_quirks |= CPU_QUIRK_NO_C3;
1358 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1359 		"acpi_cpu: working around PIIX4 bug, disabling C3\n"));
1360 
1361 	    val = pci_read_config(acpi_dev, PIIX4_DEVACTB_REG, 4);
1362 	    if ((val & PIIX4_STOP_BREAK_MASK) != PIIX4_STOP_BREAK_MASK) {
1363 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1364 		    "acpi_cpu: PIIX4: enabling IRQs to generate Stop Break\n"));
1365 	    	val |= PIIX4_STOP_BREAK_MASK;
1366 		pci_write_config(acpi_dev, PIIX4_DEVACTB_REG, val, 4);
1367 	    }
1368 	    status = AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_RLD, &val);
1369 	    if (ACPI_SUCCESS(status) && val != 0) {
1370 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1371 		    "acpi_cpu: PIIX4: reset BRLD_EN_BM\n"));
1372 		AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1373 	    }
1374 	    break;
1375 	default:
1376 	    break;
1377 	}
1378     }
1379 #endif
1380 }
1381 
1382 static int
1383 acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS)
1384 {
1385     struct acpi_cpu_softc *sc;
1386     struct sbuf	 sb;
1387     char	 buf[128];
1388     int		 i;
1389     uintmax_t	 fract, sum, whole;
1390 
1391     sc = (struct acpi_cpu_softc *) arg1;
1392     sum = 0;
1393     for (i = 0; i < sc->cpu_cx_count; i++)
1394 	sum += sc->cpu_cx_stats[i];
1395     sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1396     for (i = 0; i < sc->cpu_cx_count; i++) {
1397 	if (sum > 0) {
1398 	    whole = (uintmax_t)sc->cpu_cx_stats[i] * 100;
1399 	    fract = (whole % sum) * 100;
1400 	    sbuf_printf(&sb, "%u.%02u%% ", (u_int)(whole / sum),
1401 		(u_int)(fract / sum));
1402 	} else
1403 	    sbuf_printf(&sb, "0.00%% ");
1404     }
1405     sbuf_printf(&sb, "last %dus", sc->cpu_prev_sleep);
1406     sbuf_trim(&sb);
1407     sbuf_finish(&sb);
1408     sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1409     sbuf_delete(&sb);
1410 
1411     return (0);
1412 }
1413 
1414 /*
1415  * XXX TODO: actually add support to count each entry/exit
1416  * from the Cx states.
1417  */
1418 static int
1419 acpi_cpu_usage_counters_sysctl(SYSCTL_HANDLER_ARGS)
1420 {
1421     struct acpi_cpu_softc *sc;
1422     struct sbuf	 sb;
1423     char	 buf[128];
1424     int		 i;
1425 
1426     sc = (struct acpi_cpu_softc *) arg1;
1427 
1428     /* Print out the raw counters */
1429     sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1430 
1431     for (i = 0; i < sc->cpu_cx_count; i++) {
1432         sbuf_printf(&sb, "%u ", sc->cpu_cx_stats[i]);
1433     }
1434 
1435     sbuf_trim(&sb);
1436     sbuf_finish(&sb);
1437     sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1438     sbuf_delete(&sb);
1439 
1440     return (0);
1441 }
1442 
1443 #if defined(__i386__) || defined(__amd64__)
1444 static int
1445 acpi_cpu_method_sysctl(SYSCTL_HANDLER_ARGS)
1446 {
1447 	struct acpi_cpu_softc *sc;
1448 	struct acpi_cx *cx;
1449 	struct sbuf sb;
1450 	char buf[128];
1451 	int i;
1452 
1453 	sc = (struct acpi_cpu_softc *)arg1;
1454 	sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1455 	for (i = 0; i < sc->cpu_cx_count; i++) {
1456 		cx = &sc->cpu_cx_states[i];
1457 		sbuf_printf(&sb, "C%d/", i + 1);
1458 		if (cx->do_mwait) {
1459 			sbuf_cat(&sb, "mwait");
1460 			if (cx->mwait_hw_coord)
1461 				sbuf_cat(&sb, "/hwc");
1462 			if (cx->mwait_bm_avoidance)
1463 				sbuf_cat(&sb, "/bma");
1464 		} else if (cx->type == ACPI_STATE_C1) {
1465 			sbuf_cat(&sb, "hlt");
1466 		} else {
1467 			sbuf_cat(&sb, "io");
1468 		}
1469 		if (cx->type == ACPI_STATE_C1 && cx->p_lvlx != NULL)
1470 			sbuf_cat(&sb, "/iohlt");
1471 		sbuf_putc(&sb, ' ');
1472 	}
1473 	sbuf_trim(&sb);
1474 	sbuf_finish(&sb);
1475 	sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1476 	sbuf_delete(&sb);
1477 	return (0);
1478 }
1479 #endif
1480 
1481 static int
1482 acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc)
1483 {
1484     int i;
1485 
1486     ACPI_SERIAL_ASSERT(cpu);
1487     sc->cpu_cx_lowest = min(sc->cpu_cx_lowest_lim, sc->cpu_cx_count - 1);
1488 
1489     /* If not disabling, cache the new lowest non-C3 state. */
1490     sc->cpu_non_c3 = 0;
1491     for (i = sc->cpu_cx_lowest; i >= 0; i--) {
1492 	if (sc->cpu_cx_states[i].type < ACPI_STATE_C3) {
1493 	    sc->cpu_non_c3 = i;
1494 	    break;
1495 	}
1496     }
1497 
1498     /* Reset the statistics counters. */
1499     bzero(sc->cpu_cx_stats, sizeof(sc->cpu_cx_stats));
1500     return (0);
1501 }
1502 
1503 static int
1504 acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1505 {
1506     struct	 acpi_cpu_softc *sc;
1507     char	 state[8];
1508     int		 val, error;
1509 
1510     sc = (struct acpi_cpu_softc *) arg1;
1511     snprintf(state, sizeof(state), "C%d", sc->cpu_cx_lowest_lim + 1);
1512     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1513     if (error != 0 || req->newptr == NULL)
1514 	return (error);
1515     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1516 	return (EINVAL);
1517     if (strcasecmp(state, "Cmax") == 0)
1518 	val = MAX_CX_STATES;
1519     else {
1520 	val = (int) strtol(state + 1, NULL, 10);
1521 	if (val < 1 || val > MAX_CX_STATES)
1522 	    return (EINVAL);
1523     }
1524 
1525     ACPI_SERIAL_BEGIN(cpu);
1526     sc->cpu_cx_lowest_lim = val - 1;
1527     acpi_cpu_set_cx_lowest(sc);
1528     ACPI_SERIAL_END(cpu);
1529 
1530     return (0);
1531 }
1532 
1533 static int
1534 acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1535 {
1536     struct	acpi_cpu_softc *sc;
1537     char	state[8];
1538     int		val, error, i;
1539 
1540     snprintf(state, sizeof(state), "C%d", cpu_cx_lowest_lim + 1);
1541     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1542     if (error != 0 || req->newptr == NULL)
1543 	return (error);
1544     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1545 	return (EINVAL);
1546     if (strcasecmp(state, "Cmax") == 0)
1547 	val = MAX_CX_STATES;
1548     else {
1549 	val = (int) strtol(state + 1, NULL, 10);
1550 	if (val < 1 || val > MAX_CX_STATES)
1551 	    return (EINVAL);
1552     }
1553 
1554     /* Update the new lowest useable Cx state for all CPUs. */
1555     ACPI_SERIAL_BEGIN(cpu);
1556     cpu_cx_lowest_lim = val - 1;
1557     for (i = 0; i < cpu_ndevices; i++) {
1558 	sc = device_get_softc(cpu_devices[i]);
1559 	sc->cpu_cx_lowest_lim = cpu_cx_lowest_lim;
1560 	acpi_cpu_set_cx_lowest(sc);
1561     }
1562     ACPI_SERIAL_END(cpu);
1563 
1564     return (0);
1565 }
1566