xref: /freebsd/sys/dev/acpica/acpi_cpu.c (revision 690b7ea081790eef2c890f63a4fe7e195cf51df0)
1 /*-
2  * Copyright (c) 2003-2005 Nate Lawson (SDG)
3  * Copyright (c) 2001 Michael Smith
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #include "opt_acpi.h"
32 #include <sys/param.h>
33 #include <sys/bus.h>
34 #include <sys/cpu.h>
35 #include <sys/kernel.h>
36 #include <sys/malloc.h>
37 #include <sys/module.h>
38 #include <sys/pcpu.h>
39 #include <sys/power.h>
40 #include <sys/proc.h>
41 #include <sys/sched.h>
42 #include <sys/sbuf.h>
43 #include <sys/smp.h>
44 
45 #include <dev/pci/pcivar.h>
46 #include <machine/atomic.h>
47 #include <machine/bus.h>
48 #if defined(__amd64__) || defined(__i386__)
49 #include <machine/clock.h>
50 #include <machine/specialreg.h>
51 #include <machine/md_var.h>
52 #endif
53 #include <sys/rman.h>
54 
55 #include <contrib/dev/acpica/include/acpi.h>
56 #include <contrib/dev/acpica/include/accommon.h>
57 
58 #include <dev/acpica/acpivar.h>
59 
60 /*
61  * Support for ACPI Processor devices, including C[1-3] sleep states.
62  */
63 
64 /* Hooks for the ACPI CA debugging infrastructure */
65 #define _COMPONENT	ACPI_PROCESSOR
66 ACPI_MODULE_NAME("PROCESSOR")
67 
68 struct acpi_cx {
69     struct resource	*p_lvlx;	/* Register to read to enter state. */
70     uint32_t		 type;		/* C1-3 (C4 and up treated as C3). */
71     uint32_t		 trans_lat;	/* Transition latency (usec). */
72     uint32_t		 power;		/* Power consumed (mW). */
73     int			 res_type;	/* Resource type for p_lvlx. */
74     int			 res_rid;	/* Resource ID for p_lvlx. */
75     bool		 do_mwait;
76     uint32_t		 mwait_hint;
77     bool		 mwait_hw_coord;
78     bool		 mwait_bm_avoidance;
79 };
80 #define MAX_CX_STATES	 8
81 
82 struct acpi_cpu_softc {
83     device_t		 cpu_dev;
84     ACPI_HANDLE		 cpu_handle;
85     struct pcpu		*cpu_pcpu;
86     uint32_t		 cpu_acpi_id;	/* ACPI processor id */
87     uint32_t		 cpu_p_blk;	/* ACPI P_BLK location */
88     uint32_t		 cpu_p_blk_len;	/* P_BLK length (must be 6). */
89     struct acpi_cx	 cpu_cx_states[MAX_CX_STATES];
90     int			 cpu_cx_count;	/* Number of valid Cx states. */
91     int			 cpu_prev_sleep;/* Last idle sleep duration. */
92     int			 cpu_features;	/* Child driver supported features. */
93     /* Runtime state. */
94     int			 cpu_non_c2;	/* Index of lowest non-C2 state. */
95     int			 cpu_non_c3;	/* Index of lowest non-C3 state. */
96     u_int		 cpu_cx_stats[MAX_CX_STATES];/* Cx usage history. */
97     /* Values for sysctl. */
98     struct sysctl_ctx_list cpu_sysctl_ctx;
99     struct sysctl_oid	*cpu_sysctl_tree;
100     int			 cpu_cx_lowest;
101     int			 cpu_cx_lowest_lim;
102     int			 cpu_disable_idle; /* Disable entry to idle function */
103     char 		 cpu_cx_supported[64];
104 };
105 
106 struct acpi_cpu_device {
107     struct resource_list	ad_rl;
108 };
109 
110 #define CPU_GET_REG(reg, width) 					\
111     (bus_space_read_ ## width(rman_get_bustag((reg)), 			\
112 		      rman_get_bushandle((reg)), 0))
113 #define CPU_SET_REG(reg, width, val)					\
114     (bus_space_write_ ## width(rman_get_bustag((reg)), 			\
115 		       rman_get_bushandle((reg)), 0, (val)))
116 
117 #define ACPI_NOTIFY_CX_STATES	0x81	/* _CST changed. */
118 
119 #define CPU_QUIRK_NO_C3		(1<<0)	/* C3-type states are not usable. */
120 #define CPU_QUIRK_NO_BM_CTRL	(1<<2)	/* No bus mastering control. */
121 
122 #define PCI_VENDOR_INTEL	0x8086
123 #define PCI_DEVICE_82371AB_3	0x7113	/* PIIX4 chipset for quirks. */
124 #define PCI_REVISION_A_STEP	0
125 #define PCI_REVISION_B_STEP	1
126 #define PCI_REVISION_4E		2
127 #define PCI_REVISION_4M		3
128 #define PIIX4_DEVACTB_REG	0x58
129 #define PIIX4_BRLD_EN_IRQ0	(1<<0)
130 #define PIIX4_BRLD_EN_IRQ	(1<<1)
131 #define PIIX4_BRLD_EN_IRQ8	(1<<5)
132 #define PIIX4_STOP_BREAK_MASK	(PIIX4_BRLD_EN_IRQ0 | PIIX4_BRLD_EN_IRQ | PIIX4_BRLD_EN_IRQ8)
133 #define PIIX4_PCNTRL_BST_EN	(1<<10)
134 
135 #define	CST_FFH_VENDOR_INTEL	1
136 #define	CST_FFH_INTEL_CL_C1IO	1
137 #define	CST_FFH_INTEL_CL_MWAIT	2
138 #define	CST_FFH_MWAIT_HW_COORD	0x0001
139 #define	CST_FFH_MWAIT_BM_AVOID	0x0002
140 
141 #define	CPUDEV_DEVICE_ID	"ACPI0007"
142 
143 /* Knob to disable acpi_cpu devices */
144 bool acpi_cpu_disabled = false;
145 
146 /* Platform hardware resource information. */
147 static uint32_t		 cpu_smi_cmd;	/* Value to write to SMI_CMD. */
148 static uint8_t		 cpu_cst_cnt;	/* Indicate we are _CST aware. */
149 static int		 cpu_quirks;	/* Indicate any hardware bugs. */
150 
151 /* Values for sysctl. */
152 static struct sysctl_ctx_list cpu_sysctl_ctx;
153 static struct sysctl_oid *cpu_sysctl_tree;
154 static int		 cpu_cx_generic;
155 static int		 cpu_cx_lowest_lim;
156 
157 static struct acpi_cpu_softc **cpu_softc;
158 ACPI_SERIAL_DECL(cpu, "ACPI CPU");
159 
160 static int	acpi_cpu_probe(device_t dev);
161 static int	acpi_cpu_attach(device_t dev);
162 static int	acpi_cpu_suspend(device_t dev);
163 static int	acpi_cpu_resume(device_t dev);
164 static int	acpi_pcpu_get_id(device_t dev, uint32_t acpi_id,
165 		    u_int *cpu_id);
166 static struct resource_list *acpi_cpu_get_rlist(device_t dev, device_t child);
167 static device_t	acpi_cpu_add_child(device_t dev, u_int order, const char *name,
168 		    int unit);
169 static int	acpi_cpu_read_ivar(device_t dev, device_t child, int index,
170 		    uintptr_t *result);
171 static int	acpi_cpu_shutdown(device_t dev);
172 static void	acpi_cpu_cx_probe(struct acpi_cpu_softc *sc);
173 static void	acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc);
174 static int	acpi_cpu_cx_cst(struct acpi_cpu_softc *sc);
175 static void	acpi_cpu_startup(void *arg);
176 static void	acpi_cpu_startup_cx(struct acpi_cpu_softc *sc);
177 static void	acpi_cpu_cx_list(struct acpi_cpu_softc *sc);
178 #if defined(__i386__) || defined(__amd64__)
179 static void	acpi_cpu_idle(sbintime_t sbt);
180 #endif
181 static void	acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context);
182 static void	acpi_cpu_quirks(void);
183 static void	acpi_cpu_quirks_piix4(void);
184 static int	acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS);
185 static int	acpi_cpu_usage_counters_sysctl(SYSCTL_HANDLER_ARGS);
186 static int	acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc);
187 static int	acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
188 static int	acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
189 #if defined(__i386__) || defined(__amd64__)
190 static int	acpi_cpu_method_sysctl(SYSCTL_HANDLER_ARGS);
191 #endif
192 
193 static device_method_t acpi_cpu_methods[] = {
194     /* Device interface */
195     DEVMETHOD(device_probe,	acpi_cpu_probe),
196     DEVMETHOD(device_attach,	acpi_cpu_attach),
197     DEVMETHOD(device_detach,	bus_generic_detach),
198     DEVMETHOD(device_shutdown,	acpi_cpu_shutdown),
199     DEVMETHOD(device_suspend,	acpi_cpu_suspend),
200     DEVMETHOD(device_resume,	acpi_cpu_resume),
201 
202     /* Bus interface */
203     DEVMETHOD(bus_add_child,	acpi_cpu_add_child),
204     DEVMETHOD(bus_read_ivar,	acpi_cpu_read_ivar),
205     DEVMETHOD(bus_get_resource_list, acpi_cpu_get_rlist),
206     DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
207     DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
208     DEVMETHOD(bus_alloc_resource, bus_generic_rl_alloc_resource),
209     DEVMETHOD(bus_release_resource, bus_generic_rl_release_resource),
210     DEVMETHOD(bus_activate_resource, bus_generic_activate_resource),
211     DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource),
212     DEVMETHOD(bus_setup_intr,	bus_generic_setup_intr),
213     DEVMETHOD(bus_teardown_intr, bus_generic_teardown_intr),
214 
215     DEVMETHOD_END
216 };
217 
218 static driver_t acpi_cpu_driver = {
219     "cpu",
220     acpi_cpu_methods,
221     sizeof(struct acpi_cpu_softc),
222 };
223 
224 static devclass_t acpi_cpu_devclass;
225 DRIVER_MODULE(cpu, acpi, acpi_cpu_driver, acpi_cpu_devclass, 0, 0);
226 MODULE_DEPEND(cpu, acpi, 1, 1, 1);
227 
228 static int
229 acpi_cpu_probe(device_t dev)
230 {
231     static char		   *cpudev_ids[] = { CPUDEV_DEVICE_ID, NULL };
232     int			   acpi_id, cpu_id;
233     ACPI_BUFFER		   buf;
234     ACPI_HANDLE		   handle;
235     ACPI_OBJECT		   *obj;
236     ACPI_STATUS		   status;
237     ACPI_OBJECT_TYPE	   type;
238 
239     if (acpi_disabled("cpu") || acpi_cpu_disabled)
240 	return (ENXIO);
241     type = acpi_get_type(dev);
242     if (type != ACPI_TYPE_PROCESSOR && type != ACPI_TYPE_DEVICE)
243 	return (ENXIO);
244     if (type == ACPI_TYPE_DEVICE &&
245 	ACPI_ID_PROBE(device_get_parent(dev), dev, cpudev_ids, NULL) >= 0)
246 	return (ENXIO);
247 
248     handle = acpi_get_handle(dev);
249     if (cpu_softc == NULL)
250 	cpu_softc = malloc(sizeof(struct acpi_cpu_softc *) *
251 	    (mp_maxid + 1), M_TEMP /* XXX */, M_WAITOK | M_ZERO);
252 
253     if (type == ACPI_TYPE_PROCESSOR) {
254 	/* Get our Processor object. */
255 	buf.Pointer = NULL;
256 	buf.Length = ACPI_ALLOCATE_BUFFER;
257 	status = AcpiEvaluateObject(handle, NULL, NULL, &buf);
258 	if (ACPI_FAILURE(status)) {
259 	    device_printf(dev, "probe failed to get Processor obj - %s\n",
260 		AcpiFormatException(status));
261 	    return (ENXIO);
262 	}
263 	obj = (ACPI_OBJECT *)buf.Pointer;
264 	if (obj->Type != ACPI_TYPE_PROCESSOR) {
265 	    device_printf(dev, "Processor object has bad type %d\n",
266 		obj->Type);
267 	    AcpiOsFree(obj);
268 	    return (ENXIO);
269 	}
270 
271 	/*
272 	 * Find the processor associated with our unit.  We could use the
273 	 * ProcId as a key, however, some boxes do not have the same values
274 	 * in their Processor object as the ProcId values in the MADT.
275 	 */
276 	acpi_id = obj->Processor.ProcId;
277 	AcpiOsFree(obj);
278     } else {
279 	status = acpi_GetInteger(handle, "_UID", &acpi_id);
280 	if (ACPI_FAILURE(status)) {
281 	    device_printf(dev, "Device object has bad value - %s\n",
282 		AcpiFormatException(status));
283 	    return (ENXIO);
284 	}
285     }
286     if (acpi_pcpu_get_id(dev, acpi_id, &cpu_id) != 0) {
287 	if (bootverbose && (type != ACPI_TYPE_PROCESSOR || acpi_id != 255))
288 	    printf("ACPI: Processor %s (ACPI ID %u) ignored\n",
289 		acpi_name(acpi_get_handle(dev)), acpi_id);
290 	return (ENXIO);
291     }
292 
293     if (device_set_unit(dev, cpu_id) != 0)
294 	return (ENXIO);
295 
296     device_set_desc(dev, "ACPI CPU");
297 
298     if (!bootverbose && device_get_unit(dev) != 0) {
299 	    device_quiet(dev);
300 	    device_quiet_children(dev);
301     }
302 
303     return (BUS_PROBE_DEFAULT);
304 }
305 
306 static int
307 acpi_cpu_attach(device_t dev)
308 {
309     ACPI_BUFFER		   buf;
310     ACPI_OBJECT		   arg, *obj;
311     ACPI_OBJECT_LIST	   arglist;
312     struct pcpu		   *pcpu_data;
313     struct acpi_cpu_softc *sc;
314     struct acpi_softc	  *acpi_sc;
315     ACPI_STATUS		   status;
316     u_int		   features;
317     int			   cpu_id, drv_count, i;
318     driver_t 		  **drivers;
319     uint32_t		   cap_set[3];
320 
321     /* UUID needed by _OSC evaluation */
322     static uint8_t cpu_oscuuid[16] = { 0x16, 0xA6, 0x77, 0x40, 0x0C, 0x29,
323 				       0xBE, 0x47, 0x9E, 0xBD, 0xD8, 0x70,
324 				       0x58, 0x71, 0x39, 0x53 };
325 
326     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
327 
328     sc = device_get_softc(dev);
329     sc->cpu_dev = dev;
330     sc->cpu_handle = acpi_get_handle(dev);
331     cpu_id = device_get_unit(dev);
332     cpu_softc[cpu_id] = sc;
333     pcpu_data = pcpu_find(cpu_id);
334     pcpu_data->pc_device = dev;
335     sc->cpu_pcpu = pcpu_data;
336     cpu_smi_cmd = AcpiGbl_FADT.SmiCommand;
337     cpu_cst_cnt = AcpiGbl_FADT.CstControl;
338 
339     if (acpi_get_type(dev) == ACPI_TYPE_PROCESSOR) {
340 	buf.Pointer = NULL;
341 	buf.Length = ACPI_ALLOCATE_BUFFER;
342 	status = AcpiEvaluateObject(sc->cpu_handle, NULL, NULL, &buf);
343 	if (ACPI_FAILURE(status)) {
344 	    device_printf(dev, "attach failed to get Processor obj - %s\n",
345 		AcpiFormatException(status));
346 	    return (ENXIO);
347 	}
348 	obj = (ACPI_OBJECT *)buf.Pointer;
349 	sc->cpu_p_blk = obj->Processor.PblkAddress;
350 	sc->cpu_p_blk_len = obj->Processor.PblkLength;
351 	sc->cpu_acpi_id = obj->Processor.ProcId;
352 	AcpiOsFree(obj);
353     } else {
354 	KASSERT(acpi_get_type(dev) == ACPI_TYPE_DEVICE,
355 	    ("Unexpected ACPI object"));
356 	status = acpi_GetInteger(sc->cpu_handle, "_UID", &sc->cpu_acpi_id);
357 	if (ACPI_FAILURE(status)) {
358 	    device_printf(dev, "Device object has bad value - %s\n",
359 		AcpiFormatException(status));
360 	    return (ENXIO);
361 	}
362 	sc->cpu_p_blk = 0;
363 	sc->cpu_p_blk_len = 0;
364     }
365     ACPI_DEBUG_PRINT((ACPI_DB_INFO, "acpi_cpu%d: P_BLK at %#x/%d\n",
366 		     device_get_unit(dev), sc->cpu_p_blk, sc->cpu_p_blk_len));
367 
368     /*
369      * If this is the first cpu we attach, create and initialize the generic
370      * resources that will be used by all acpi cpu devices.
371      */
372     if (device_get_unit(dev) == 0) {
373 	/* Assume we won't be using generic Cx mode by default */
374 	cpu_cx_generic = FALSE;
375 
376 	/* Install hw.acpi.cpu sysctl tree */
377 	acpi_sc = acpi_device_get_parent_softc(dev);
378 	sysctl_ctx_init(&cpu_sysctl_ctx);
379 	cpu_sysctl_tree = SYSCTL_ADD_NODE(&cpu_sysctl_ctx,
380 	    SYSCTL_CHILDREN(acpi_sc->acpi_sysctl_tree), OID_AUTO, "cpu",
381 	    CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "node for CPU children");
382     }
383 
384     /*
385      * Before calling any CPU methods, collect child driver feature hints
386      * and notify ACPI of them.  We support unified SMP power control
387      * so advertise this ourselves.  Note this is not the same as independent
388      * SMP control where each CPU can have different settings.
389      */
390     sc->cpu_features = ACPI_CAP_SMP_SAME | ACPI_CAP_SMP_SAME_C3 |
391       ACPI_CAP_C1_IO_HALT;
392 
393 #if defined(__i386__) || defined(__amd64__)
394     /*
395      * Ask for MWAIT modes if not disabled and interrupts work
396      * reasonable with MWAIT.
397      */
398     if (!acpi_disabled("mwait") && cpu_mwait_usable())
399 	sc->cpu_features |= ACPI_CAP_SMP_C1_NATIVE | ACPI_CAP_SMP_C3_NATIVE;
400 #endif
401 
402     if (devclass_get_drivers(device_get_devclass(dev), &drivers,
403 	&drv_count) == 0) {
404 	for (i = 0; i < drv_count; i++) {
405 	    if (ACPI_GET_FEATURES(drivers[i], &features) == 0)
406 		sc->cpu_features |= features;
407 	}
408 	free(drivers, M_TEMP);
409     }
410 
411     /*
412      * CPU capabilities are specified in
413      * Intel Processor Vendor-Specific ACPI Interface Specification.
414      */
415     if (sc->cpu_features) {
416 	cap_set[1] = sc->cpu_features;
417 	status = acpi_EvaluateOSC(sc->cpu_handle, cpu_oscuuid, 1, 2, cap_set,
418 	    cap_set, false);
419 	if (ACPI_SUCCESS(status)) {
420 	    if (cap_set[0] != 0)
421 		device_printf(dev, "_OSC returned status %#x\n", cap_set[0]);
422 	}
423 	else {
424 	    arglist.Pointer = &arg;
425 	    arglist.Count = 1;
426 	    arg.Type = ACPI_TYPE_BUFFER;
427 	    arg.Buffer.Length = sizeof(cap_set);
428 	    arg.Buffer.Pointer = (uint8_t *)cap_set;
429 	    cap_set[0] = 1; /* revision */
430 	    cap_set[1] = 1; /* number of capabilities integers */
431 	    cap_set[2] = sc->cpu_features;
432 	    AcpiEvaluateObject(sc->cpu_handle, "_PDC", &arglist, NULL);
433 	}
434     }
435 
436     /* Probe for Cx state support. */
437     acpi_cpu_cx_probe(sc);
438 
439     return (0);
440 }
441 
442 static void
443 acpi_cpu_postattach(void *unused __unused)
444 {
445     struct acpi_cpu_softc *sc;
446     int attached = 0, i;
447 
448     if (cpu_softc == NULL)
449 	return;
450 
451     bus_topo_lock();
452     CPU_FOREACH(i) {
453 	if ((sc = cpu_softc[i]) != NULL)
454 		bus_generic_probe(sc->cpu_dev);
455     }
456     CPU_FOREACH(i) {
457 	if ((sc = cpu_softc[i]) != NULL) {
458 		bus_generic_attach(sc->cpu_dev);
459 		attached = 1;
460 	}
461     }
462     bus_topo_unlock();
463 
464     if (attached) {
465 #ifdef EARLY_AP_STARTUP
466 	acpi_cpu_startup(NULL);
467 #else
468 	/* Queue post cpu-probing task handler */
469 	AcpiOsExecute(OSL_NOTIFY_HANDLER, acpi_cpu_startup, NULL);
470 #endif
471     }
472 }
473 
474 SYSINIT(acpi_cpu, SI_SUB_CONFIGURE, SI_ORDER_MIDDLE,
475     acpi_cpu_postattach, NULL);
476 
477 static void
478 disable_idle(struct acpi_cpu_softc *sc)
479 {
480     cpuset_t cpuset;
481 
482     CPU_SETOF(sc->cpu_pcpu->pc_cpuid, &cpuset);
483     sc->cpu_disable_idle = TRUE;
484 
485     /*
486      * Ensure that the CPU is not in idle state or in acpi_cpu_idle().
487      * Note that this code depends on the fact that the rendezvous IPI
488      * can not penetrate context where interrupts are disabled and acpi_cpu_idle
489      * is called and executed in such a context with interrupts being re-enabled
490      * right before return.
491      */
492     smp_rendezvous_cpus(cpuset, smp_no_rendezvous_barrier, NULL,
493 	smp_no_rendezvous_barrier, NULL);
494 }
495 
496 static void
497 enable_idle(struct acpi_cpu_softc *sc)
498 {
499 
500     sc->cpu_disable_idle = FALSE;
501 }
502 
503 #if defined(__i386__) || defined(__amd64__)
504 static int
505 is_idle_disabled(struct acpi_cpu_softc *sc)
506 {
507 
508     return (sc->cpu_disable_idle);
509 }
510 #endif
511 
512 /*
513  * Disable any entry to the idle function during suspend and re-enable it
514  * during resume.
515  */
516 static int
517 acpi_cpu_suspend(device_t dev)
518 {
519     int error;
520 
521     error = bus_generic_suspend(dev);
522     if (error)
523 	return (error);
524     disable_idle(device_get_softc(dev));
525     return (0);
526 }
527 
528 static int
529 acpi_cpu_resume(device_t dev)
530 {
531 
532     enable_idle(device_get_softc(dev));
533     return (bus_generic_resume(dev));
534 }
535 
536 /*
537  * Find the processor associated with a given ACPI ID.
538  */
539 static int
540 acpi_pcpu_get_id(device_t dev, uint32_t acpi_id, u_int *cpu_id)
541 {
542     struct pcpu	*pc;
543     u_int	 i;
544 
545     CPU_FOREACH(i) {
546 	pc = pcpu_find(i);
547 	if (pc->pc_acpi_id == acpi_id) {
548 	    *cpu_id = pc->pc_cpuid;
549 	    return (0);
550 	}
551     }
552 
553     /*
554      * If pc_acpi_id for CPU 0 is not initialized (e.g. a non-APIC
555      * UP box) use the ACPI ID from the first processor we find.
556      */
557     if (mp_ncpus == 1) {
558 	pc = pcpu_find(0);
559 	if (pc->pc_acpi_id == 0xffffffff)
560 	    pc->pc_acpi_id = acpi_id;
561 	*cpu_id = 0;
562 	return (0);
563     }
564 
565     return (ESRCH);
566 }
567 
568 static struct resource_list *
569 acpi_cpu_get_rlist(device_t dev, device_t child)
570 {
571     struct acpi_cpu_device *ad;
572 
573     ad = device_get_ivars(child);
574     if (ad == NULL)
575 	return (NULL);
576     return (&ad->ad_rl);
577 }
578 
579 static device_t
580 acpi_cpu_add_child(device_t dev, u_int order, const char *name, int unit)
581 {
582     struct acpi_cpu_device *ad;
583     device_t child;
584 
585     if ((ad = malloc(sizeof(*ad), M_TEMP, M_NOWAIT | M_ZERO)) == NULL)
586 	return (NULL);
587 
588     resource_list_init(&ad->ad_rl);
589 
590     child = device_add_child_ordered(dev, order, name, unit);
591     if (child != NULL)
592 	device_set_ivars(child, ad);
593     else
594 	free(ad, M_TEMP);
595     return (child);
596 }
597 
598 static int
599 acpi_cpu_read_ivar(device_t dev, device_t child, int index, uintptr_t *result)
600 {
601     struct acpi_cpu_softc *sc;
602 
603     sc = device_get_softc(dev);
604     switch (index) {
605     case ACPI_IVAR_HANDLE:
606 	*result = (uintptr_t)sc->cpu_handle;
607 	break;
608     case CPU_IVAR_PCPU:
609 	*result = (uintptr_t)sc->cpu_pcpu;
610 	break;
611 #if defined(__amd64__) || defined(__i386__)
612     case CPU_IVAR_NOMINAL_MHZ:
613 	if (tsc_is_invariant) {
614 	    *result = (uintptr_t)(atomic_load_acq_64(&tsc_freq) / 1000000);
615 	    break;
616 	}
617 	/* FALLTHROUGH */
618 #endif
619     default:
620 	return (ENOENT);
621     }
622     return (0);
623 }
624 
625 static int
626 acpi_cpu_shutdown(device_t dev)
627 {
628     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
629 
630     /* Allow children to shutdown first. */
631     bus_generic_shutdown(dev);
632 
633     /*
634      * Disable any entry to the idle function.
635      */
636     disable_idle(device_get_softc(dev));
637 
638     /*
639      * CPU devices are not truly detached and remain referenced,
640      * so their resources are not freed.
641      */
642 
643     return_VALUE (0);
644 }
645 
646 static void
647 acpi_cpu_cx_probe(struct acpi_cpu_softc *sc)
648 {
649     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
650 
651     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
652     sc->cpu_prev_sleep = 1000000;
653     sc->cpu_cx_lowest = 0;
654     sc->cpu_cx_lowest_lim = 0;
655 
656     /*
657      * Check for the ACPI 2.0 _CST sleep states object. If we can't find
658      * any, we'll revert to generic FADT/P_BLK Cx control method which will
659      * be handled by acpi_cpu_startup. We need to defer to after having
660      * probed all the cpus in the system before probing for generic Cx
661      * states as we may already have found cpus with valid _CST packages
662      */
663     if (!cpu_cx_generic && acpi_cpu_cx_cst(sc) != 0) {
664 	/*
665 	 * We were unable to find a _CST package for this cpu or there
666 	 * was an error parsing it. Switch back to generic mode.
667 	 */
668 	cpu_cx_generic = TRUE;
669 	if (bootverbose)
670 	    device_printf(sc->cpu_dev, "switching to generic Cx mode\n");
671     }
672 
673     /*
674      * TODO: _CSD Package should be checked here.
675      */
676 }
677 
678 static void
679 acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc)
680 {
681     ACPI_GENERIC_ADDRESS	 gas;
682     struct acpi_cx		*cx_ptr;
683 
684     sc->cpu_cx_count = 0;
685     cx_ptr = sc->cpu_cx_states;
686 
687     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
688     sc->cpu_prev_sleep = 1000000;
689 
690     /* C1 has been required since just after ACPI 1.0 */
691     cx_ptr->type = ACPI_STATE_C1;
692     cx_ptr->trans_lat = 0;
693     cx_ptr++;
694     sc->cpu_non_c2 = sc->cpu_cx_count;
695     sc->cpu_non_c3 = sc->cpu_cx_count;
696     sc->cpu_cx_count++;
697 
698     /*
699      * The spec says P_BLK must be 6 bytes long.  However, some systems
700      * use it to indicate a fractional set of features present so we
701      * take 5 as C2.  Some may also have a value of 7 to indicate
702      * another C3 but most use _CST for this (as required) and having
703      * "only" C1-C3 is not a hardship.
704      */
705     if (sc->cpu_p_blk_len < 5)
706 	return;
707 
708     /* Validate and allocate resources for C2 (P_LVL2). */
709     gas.SpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
710     gas.BitWidth = 8;
711     if (AcpiGbl_FADT.C2Latency <= 100) {
712 	gas.Address = sc->cpu_p_blk + 4;
713 	cx_ptr->res_rid = 0;
714 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &cx_ptr->res_rid,
715 	    &gas, &cx_ptr->p_lvlx, RF_SHAREABLE);
716 	if (cx_ptr->p_lvlx != NULL) {
717 	    cx_ptr->type = ACPI_STATE_C2;
718 	    cx_ptr->trans_lat = AcpiGbl_FADT.C2Latency;
719 	    cx_ptr++;
720 	    sc->cpu_non_c3 = sc->cpu_cx_count;
721 	    sc->cpu_cx_count++;
722 	}
723     }
724     if (sc->cpu_p_blk_len < 6)
725 	return;
726 
727     /* Validate and allocate resources for C3 (P_LVL3). */
728     if (AcpiGbl_FADT.C3Latency <= 1000 && !(cpu_quirks & CPU_QUIRK_NO_C3)) {
729 	gas.Address = sc->cpu_p_blk + 5;
730 	cx_ptr->res_rid = 1;
731 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &cx_ptr->res_rid,
732 	    &gas, &cx_ptr->p_lvlx, RF_SHAREABLE);
733 	if (cx_ptr->p_lvlx != NULL) {
734 	    cx_ptr->type = ACPI_STATE_C3;
735 	    cx_ptr->trans_lat = AcpiGbl_FADT.C3Latency;
736 	    cx_ptr++;
737 	    sc->cpu_cx_count++;
738 	}
739     }
740 }
741 
742 #if defined(__i386__) || defined(__amd64__)
743 static void
744 acpi_cpu_cx_cst_mwait(struct acpi_cx *cx_ptr, uint64_t address, int accsize)
745 {
746 
747 	cx_ptr->do_mwait = true;
748 	cx_ptr->mwait_hint = address & 0xffffffff;
749 	cx_ptr->mwait_hw_coord = (accsize & CST_FFH_MWAIT_HW_COORD) != 0;
750 	cx_ptr->mwait_bm_avoidance = (accsize & CST_FFH_MWAIT_BM_AVOID) != 0;
751 }
752 #endif
753 
754 static void
755 acpi_cpu_cx_cst_free_plvlx(device_t cpu_dev, struct acpi_cx *cx_ptr)
756 {
757 
758 	if (cx_ptr->p_lvlx == NULL)
759 		return;
760 	bus_release_resource(cpu_dev, cx_ptr->res_type, cx_ptr->res_rid,
761 	    cx_ptr->p_lvlx);
762 	cx_ptr->p_lvlx = NULL;
763 }
764 
765 /*
766  * Parse a _CST package and set up its Cx states.  Since the _CST object
767  * can change dynamically, our notify handler may call this function
768  * to clean up and probe the new _CST package.
769  */
770 static int
771 acpi_cpu_cx_cst(struct acpi_cpu_softc *sc)
772 {
773     struct	 acpi_cx *cx_ptr;
774     ACPI_STATUS	 status;
775     ACPI_BUFFER	 buf;
776     ACPI_OBJECT	*top;
777     ACPI_OBJECT	*pkg;
778     uint32_t	 count;
779     int		 i;
780 #if defined(__i386__) || defined(__amd64__)
781     uint64_t	 address;
782     int		 vendor, class, accsize;
783 #endif
784 
785     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
786 
787     buf.Pointer = NULL;
788     buf.Length = ACPI_ALLOCATE_BUFFER;
789     status = AcpiEvaluateObject(sc->cpu_handle, "_CST", NULL, &buf);
790     if (ACPI_FAILURE(status))
791 	return (ENXIO);
792 
793     /* _CST is a package with a count and at least one Cx package. */
794     top = (ACPI_OBJECT *)buf.Pointer;
795     if (!ACPI_PKG_VALID(top, 2) || acpi_PkgInt32(top, 0, &count) != 0) {
796 	device_printf(sc->cpu_dev, "invalid _CST package\n");
797 	AcpiOsFree(buf.Pointer);
798 	return (ENXIO);
799     }
800     if (count != top->Package.Count - 1) {
801 	device_printf(sc->cpu_dev, "invalid _CST state count (%d != %d)\n",
802 	       count, top->Package.Count - 1);
803 	count = top->Package.Count - 1;
804     }
805     if (count > MAX_CX_STATES) {
806 	device_printf(sc->cpu_dev, "_CST has too many states (%d)\n", count);
807 	count = MAX_CX_STATES;
808     }
809 
810     sc->cpu_non_c2 = 0;
811     sc->cpu_non_c3 = 0;
812     sc->cpu_cx_count = 0;
813     cx_ptr = sc->cpu_cx_states;
814 
815     /*
816      * C1 has been required since just after ACPI 1.0.
817      * Reserve the first slot for it.
818      */
819     cx_ptr->type = ACPI_STATE_C0;
820     cx_ptr++;
821     sc->cpu_cx_count++;
822 
823     /* Set up all valid states. */
824     for (i = 0; i < count; i++) {
825 	pkg = &top->Package.Elements[i + 1];
826 	if (!ACPI_PKG_VALID(pkg, 4) ||
827 	    acpi_PkgInt32(pkg, 1, &cx_ptr->type) != 0 ||
828 	    acpi_PkgInt32(pkg, 2, &cx_ptr->trans_lat) != 0 ||
829 	    acpi_PkgInt32(pkg, 3, &cx_ptr->power) != 0) {
830 	    device_printf(sc->cpu_dev, "skipping invalid Cx state package\n");
831 	    continue;
832 	}
833 
834 	/* Validate the state to see if we should use it. */
835 	switch (cx_ptr->type) {
836 	case ACPI_STATE_C1:
837 	    acpi_cpu_cx_cst_free_plvlx(sc->cpu_dev, cx_ptr);
838 #if defined(__i386__) || defined(__amd64__)
839 	    if (acpi_PkgFFH_IntelCpu(pkg, 0, &vendor, &class, &address,
840 	      &accsize) == 0 && vendor == CST_FFH_VENDOR_INTEL) {
841 		if (class == CST_FFH_INTEL_CL_C1IO) {
842 		    /* C1 I/O then Halt */
843 		    cx_ptr->res_rid = sc->cpu_cx_count;
844 		    bus_set_resource(sc->cpu_dev, SYS_RES_IOPORT,
845 		      cx_ptr->res_rid, address, 1);
846 		    cx_ptr->p_lvlx = bus_alloc_resource_any(sc->cpu_dev,
847 		      SYS_RES_IOPORT, &cx_ptr->res_rid, RF_ACTIVE |
848 		      RF_SHAREABLE);
849 		    if (cx_ptr->p_lvlx == NULL) {
850 			bus_delete_resource(sc->cpu_dev, SYS_RES_IOPORT,
851 			  cx_ptr->res_rid);
852 			device_printf(sc->cpu_dev,
853 			  "C1 I/O failed to allocate port %d, "
854 			  "degrading to C1 Halt", (int)address);
855 		    }
856 		} else if (class == CST_FFH_INTEL_CL_MWAIT) {
857 		    acpi_cpu_cx_cst_mwait(cx_ptr, address, accsize);
858 		}
859 	    }
860 #endif
861 	    if (sc->cpu_cx_states[0].type == ACPI_STATE_C0) {
862 		/* This is the first C1 state.  Use the reserved slot. */
863 		sc->cpu_cx_states[0] = *cx_ptr;
864 	    } else {
865 		sc->cpu_non_c2 = sc->cpu_cx_count;
866 		sc->cpu_non_c3 = sc->cpu_cx_count;
867 		cx_ptr++;
868 		sc->cpu_cx_count++;
869 	    }
870 	    continue;
871 	case ACPI_STATE_C2:
872 	    sc->cpu_non_c3 = sc->cpu_cx_count;
873 	    break;
874 	case ACPI_STATE_C3:
875 	default:
876 	    if ((cpu_quirks & CPU_QUIRK_NO_C3) != 0) {
877 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
878 				 "acpi_cpu%d: C3[%d] not available.\n",
879 				 device_get_unit(sc->cpu_dev), i));
880 		continue;
881 	    }
882 	    break;
883 	}
884 
885 	/* Free up any previous register. */
886 	acpi_cpu_cx_cst_free_plvlx(sc->cpu_dev, cx_ptr);
887 
888 	/* Allocate the control register for C2 or C3. */
889 #if defined(__i386__) || defined(__amd64__)
890 	if (acpi_PkgFFH_IntelCpu(pkg, 0, &vendor, &class, &address,
891 	  &accsize) == 0 && vendor == CST_FFH_VENDOR_INTEL &&
892 	  class == CST_FFH_INTEL_CL_MWAIT) {
893 	    /* Native C State Instruction use (mwait) */
894 	    acpi_cpu_cx_cst_mwait(cx_ptr, address, accsize);
895 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
896 	      "acpi_cpu%d: Got C%d/mwait - %d latency\n",
897 	      device_get_unit(sc->cpu_dev), cx_ptr->type, cx_ptr->trans_lat));
898 	    cx_ptr++;
899 	    sc->cpu_cx_count++;
900 	} else
901 #endif
902 	{
903 	    cx_ptr->res_rid = sc->cpu_cx_count;
904 	    acpi_PkgGas(sc->cpu_dev, pkg, 0, &cx_ptr->res_type,
905 		&cx_ptr->res_rid, &cx_ptr->p_lvlx, RF_SHAREABLE);
906 	    if (cx_ptr->p_lvlx) {
907 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
908 		     "acpi_cpu%d: Got C%d - %d latency\n",
909 		     device_get_unit(sc->cpu_dev), cx_ptr->type,
910 		     cx_ptr->trans_lat));
911 		cx_ptr++;
912 		sc->cpu_cx_count++;
913 	    }
914 	}
915     }
916     AcpiOsFree(buf.Pointer);
917 
918     /* If C1 state was not found, we need one now. */
919     cx_ptr = sc->cpu_cx_states;
920     if (cx_ptr->type == ACPI_STATE_C0) {
921 	cx_ptr->type = ACPI_STATE_C1;
922 	cx_ptr->trans_lat = 0;
923     }
924 
925     return (0);
926 }
927 
928 /*
929  * Call this *after* all CPUs have been attached.
930  */
931 static void
932 acpi_cpu_startup(void *arg)
933 {
934     struct acpi_cpu_softc *sc;
935     int i;
936 
937     /*
938      * Setup any quirks that might necessary now that we have probed
939      * all the CPUs
940      */
941     acpi_cpu_quirks();
942 
943     if (cpu_cx_generic) {
944 	/*
945 	 * We are using generic Cx mode, probe for available Cx states
946 	 * for all processors.
947 	 */
948 	CPU_FOREACH(i) {
949 	    if ((sc = cpu_softc[i]) != NULL)
950 		acpi_cpu_generic_cx_probe(sc);
951 	}
952     } else {
953 	/*
954 	 * We are using _CST mode, remove C3 state if necessary.
955 	 * As we now know for sure that we will be using _CST mode
956 	 * install our notify handler.
957 	 */
958 	CPU_FOREACH(i) {
959 	    if ((sc = cpu_softc[i]) == NULL)
960 		continue;
961 	    if (cpu_quirks & CPU_QUIRK_NO_C3) {
962 		sc->cpu_cx_count = min(sc->cpu_cx_count, sc->cpu_non_c3 + 1);
963 	    }
964 	    AcpiInstallNotifyHandler(sc->cpu_handle, ACPI_DEVICE_NOTIFY,
965 		acpi_cpu_notify, sc);
966 	}
967     }
968 
969     /* Perform Cx final initialization. */
970     CPU_FOREACH(i) {
971 	if ((sc = cpu_softc[i]) != NULL)
972 	    acpi_cpu_startup_cx(sc);
973     }
974 
975     /* Add a sysctl handler to handle global Cx lowest setting */
976     SYSCTL_ADD_PROC(&cpu_sysctl_ctx, SYSCTL_CHILDREN(cpu_sysctl_tree),
977 	OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_MPSAFE,
978 	NULL, 0, acpi_cpu_global_cx_lowest_sysctl, "A",
979 	"Global lowest Cx sleep state to use");
980 
981     /* Take over idling from cpu_idle_default(). */
982     cpu_cx_lowest_lim = 0;
983     CPU_FOREACH(i) {
984 	if ((sc = cpu_softc[i]) != NULL)
985 	    enable_idle(sc);
986     }
987 #if defined(__i386__) || defined(__amd64__)
988     cpu_idle_hook = acpi_cpu_idle;
989 #endif
990 }
991 
992 static void
993 acpi_cpu_cx_list(struct acpi_cpu_softc *sc)
994 {
995     struct sbuf sb;
996     int i;
997 
998     /*
999      * Set up the list of Cx states
1000      */
1001     sbuf_new(&sb, sc->cpu_cx_supported, sizeof(sc->cpu_cx_supported),
1002 	SBUF_FIXEDLEN);
1003     for (i = 0; i < sc->cpu_cx_count; i++)
1004 	sbuf_printf(&sb, "C%d/%d/%d ", i + 1, sc->cpu_cx_states[i].type,
1005 	    sc->cpu_cx_states[i].trans_lat);
1006     sbuf_trim(&sb);
1007     sbuf_finish(&sb);
1008 }
1009 
1010 static void
1011 acpi_cpu_startup_cx(struct acpi_cpu_softc *sc)
1012 {
1013     acpi_cpu_cx_list(sc);
1014 
1015     SYSCTL_ADD_STRING(&sc->cpu_sysctl_ctx,
1016 		      SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
1017 		      OID_AUTO, "cx_supported", CTLFLAG_RD,
1018 		      sc->cpu_cx_supported, 0,
1019 		      "Cx/microsecond values for supported Cx states");
1020     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
1021         SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)), OID_AUTO,
1022 	"cx_lowest", CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_MPSAFE,
1023 	(void *)sc, 0, acpi_cpu_cx_lowest_sysctl, "A",
1024 	"lowest Cx sleep state to use");
1025     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
1026         SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)), OID_AUTO,
1027 	"cx_usage", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE,
1028 	(void *)sc, 0, acpi_cpu_usage_sysctl, "A",
1029 	"percent usage for each Cx state");
1030     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
1031         SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)), OID_AUTO,
1032 	"cx_usage_counters", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE,
1033 	(void *)sc, 0, acpi_cpu_usage_counters_sysctl, "A",
1034 	"Cx sleep state counters");
1035 #if defined(__i386__) || defined(__amd64__)
1036     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
1037         SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)), OID_AUTO,
1038 	"cx_method", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE,
1039 	(void *)sc, 0, acpi_cpu_method_sysctl, "A", "Cx entrance methods");
1040 #endif
1041 
1042     /* Signal platform that we can handle _CST notification. */
1043     if (!cpu_cx_generic && cpu_cst_cnt != 0) {
1044 	ACPI_LOCK(acpi);
1045 	AcpiOsWritePort(cpu_smi_cmd, cpu_cst_cnt, 8);
1046 	ACPI_UNLOCK(acpi);
1047     }
1048 }
1049 
1050 #if defined(__i386__) || defined(__amd64__)
1051 /*
1052  * Idle the CPU in the lowest state possible.  This function is called with
1053  * interrupts disabled.  Note that once it re-enables interrupts, a task
1054  * switch can occur so do not access shared data (i.e. the softc) after
1055  * interrupts are re-enabled.
1056  */
1057 static void
1058 acpi_cpu_idle(sbintime_t sbt)
1059 {
1060     struct	acpi_cpu_softc *sc;
1061     struct	acpi_cx *cx_next;
1062     uint64_t	start_ticks, end_ticks;
1063     uint32_t	start_time, end_time;
1064     ACPI_STATUS	status;
1065     int		bm_active, cx_next_idx, i, us;
1066 
1067     /*
1068      * Look up our CPU id to get our softc.  If it's NULL, we'll use C1
1069      * since there is no ACPI processor object for this CPU.  This occurs
1070      * for logical CPUs in the HTT case.
1071      */
1072     sc = cpu_softc[PCPU_GET(cpuid)];
1073     if (sc == NULL) {
1074 	acpi_cpu_c1();
1075 	return;
1076     }
1077 
1078     /* If disabled, take the safe path. */
1079     if (is_idle_disabled(sc)) {
1080 	acpi_cpu_c1();
1081 	return;
1082     }
1083 
1084     /* Find the lowest state that has small enough latency. */
1085     us = sc->cpu_prev_sleep;
1086     if (sbt >= 0 && us > (sbt >> 12))
1087 	us = (sbt >> 12);
1088     cx_next_idx = 0;
1089     if (cpu_disable_c2_sleep)
1090 	i = min(sc->cpu_cx_lowest, sc->cpu_non_c2);
1091     else if (cpu_disable_c3_sleep)
1092 	i = min(sc->cpu_cx_lowest, sc->cpu_non_c3);
1093     else
1094 	i = sc->cpu_cx_lowest;
1095     for (; i >= 0; i--) {
1096 	if (sc->cpu_cx_states[i].trans_lat * 3 <= us) {
1097 	    cx_next_idx = i;
1098 	    break;
1099 	}
1100     }
1101 
1102     /*
1103      * Check for bus master activity.  If there was activity, clear
1104      * the bit and use the lowest non-C3 state.  Note that the USB
1105      * driver polling for new devices keeps this bit set all the
1106      * time if USB is loaded.
1107      */
1108     cx_next = &sc->cpu_cx_states[cx_next_idx];
1109     if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0 &&
1110 	cx_next_idx > sc->cpu_non_c3 &&
1111 	(!cx_next->do_mwait || cx_next->mwait_bm_avoidance)) {
1112 	status = AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, &bm_active);
1113 	if (ACPI_SUCCESS(status) && bm_active != 0) {
1114 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, 1);
1115 	    cx_next_idx = sc->cpu_non_c3;
1116 	    cx_next = &sc->cpu_cx_states[cx_next_idx];
1117 	}
1118     }
1119 
1120     /* Select the next state and update statistics. */
1121     sc->cpu_cx_stats[cx_next_idx]++;
1122     KASSERT(cx_next->type != ACPI_STATE_C0, ("acpi_cpu_idle: C0 sleep"));
1123 
1124     /*
1125      * Execute HLT (or equivalent) and wait for an interrupt.  We can't
1126      * precisely calculate the time spent in C1 since the place we wake up
1127      * is an ISR.  Assume we slept no more then half of quantum, unless
1128      * we are called inside critical section, delaying context switch.
1129      */
1130     if (cx_next->type == ACPI_STATE_C1) {
1131 	start_ticks = cpu_ticks();
1132 	if (cx_next->p_lvlx != NULL) {
1133 	    /* C1 I/O then Halt */
1134 	    CPU_GET_REG(cx_next->p_lvlx, 1);
1135 	}
1136 	if (cx_next->do_mwait)
1137 	    acpi_cpu_idle_mwait(cx_next->mwait_hint);
1138 	else
1139 	    acpi_cpu_c1();
1140 	end_ticks = cpu_ticks();
1141 	/* acpi_cpu_c1() returns with interrupts enabled. */
1142 	if (cx_next->do_mwait)
1143 	    ACPI_ENABLE_IRQS();
1144 	end_time = ((end_ticks - start_ticks) << 20) / cpu_tickrate();
1145 	if (!cx_next->do_mwait && curthread->td_critnest == 0)
1146 		end_time = min(end_time, 500000 / hz);
1147 	sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + end_time) / 4;
1148 	return;
1149     }
1150 
1151     /*
1152      * For C3, disable bus master arbitration and enable bus master wake
1153      * if BM control is available, otherwise flush the CPU cache.
1154      */
1155     if (cx_next->type == ACPI_STATE_C3) {
1156 	if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
1157 	    AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 1);
1158 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
1159 	} else
1160 	    ACPI_FLUSH_CPU_CACHE();
1161     }
1162 
1163     /*
1164      * Read from P_LVLx to enter C2(+), checking time spent asleep.
1165      * Use the ACPI timer for measuring sleep time.  Since we need to
1166      * get the time very close to the CPU start/stop clock logic, this
1167      * is the only reliable time source.
1168      */
1169     if (cx_next->type == ACPI_STATE_C3) {
1170 	AcpiGetTimer(&start_time);
1171 	start_ticks = 0;
1172     } else {
1173 	start_time = 0;
1174 	start_ticks = cpu_ticks();
1175     }
1176     if (cx_next->do_mwait) {
1177 	acpi_cpu_idle_mwait(cx_next->mwait_hint);
1178     } else {
1179 	CPU_GET_REG(cx_next->p_lvlx, 1);
1180 	/*
1181 	 * Read the end time twice.  Since it may take an arbitrary time
1182 	 * to enter the idle state, the first read may be executed before
1183 	 * the processor has stopped.  Doing it again provides enough
1184 	 * margin that we are certain to have a correct value.
1185 	 */
1186 	AcpiGetTimer(&end_time);
1187     }
1188 
1189     if (cx_next->type == ACPI_STATE_C3)
1190 	AcpiGetTimer(&end_time);
1191     else
1192 	end_ticks = cpu_ticks();
1193 
1194     /* Enable bus master arbitration and disable bus master wakeup. */
1195     if (cx_next->type == ACPI_STATE_C3 &&
1196       (cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
1197 	AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 0);
1198 	AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1199     }
1200     ACPI_ENABLE_IRQS();
1201 
1202     if (cx_next->type == ACPI_STATE_C3)
1203 	AcpiGetTimerDuration(start_time, end_time, &end_time);
1204     else
1205 	end_time = ((end_ticks - start_ticks) << 20) / cpu_tickrate();
1206     sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + end_time) / 4;
1207 }
1208 #endif
1209 
1210 /*
1211  * Re-evaluate the _CST object when we are notified that it changed.
1212  */
1213 static void
1214 acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context)
1215 {
1216     struct acpi_cpu_softc *sc = (struct acpi_cpu_softc *)context;
1217 
1218     if (notify != ACPI_NOTIFY_CX_STATES)
1219 	return;
1220 
1221     /*
1222      * C-state data for target CPU is going to be in flux while we execute
1223      * acpi_cpu_cx_cst, so disable entering acpi_cpu_idle.
1224      * Also, it may happen that multiple ACPI taskqueues may concurrently
1225      * execute notifications for the same CPU.  ACPI_SERIAL is used to
1226      * protect against that.
1227      */
1228     ACPI_SERIAL_BEGIN(cpu);
1229     disable_idle(sc);
1230 
1231     /* Update the list of Cx states. */
1232     acpi_cpu_cx_cst(sc);
1233     acpi_cpu_cx_list(sc);
1234     acpi_cpu_set_cx_lowest(sc);
1235 
1236     enable_idle(sc);
1237     ACPI_SERIAL_END(cpu);
1238 
1239     acpi_UserNotify("PROCESSOR", sc->cpu_handle, notify);
1240 }
1241 
1242 static void
1243 acpi_cpu_quirks(void)
1244 {
1245     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
1246 
1247     /*
1248      * Bus mastering arbitration control is needed to keep caches coherent
1249      * while sleeping in C3.  If it's not present but a working flush cache
1250      * instruction is present, flush the caches before entering C3 instead.
1251      * Otherwise, just disable C3 completely.
1252      */
1253     if (AcpiGbl_FADT.Pm2ControlBlock == 0 ||
1254 	AcpiGbl_FADT.Pm2ControlLength == 0) {
1255 	if ((AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD) &&
1256 	    (AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD_FLUSH) == 0) {
1257 	    cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1258 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1259 		"acpi_cpu: no BM control, using flush cache method\n"));
1260 	} else {
1261 	    cpu_quirks |= CPU_QUIRK_NO_C3;
1262 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1263 		"acpi_cpu: no BM control, C3 not available\n"));
1264 	}
1265     }
1266 
1267     /*
1268      * If we are using generic Cx mode, C3 on multiple CPUs requires using
1269      * the expensive flush cache instruction.
1270      */
1271     if (cpu_cx_generic && mp_ncpus > 1) {
1272 	cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1273 	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1274 	    "acpi_cpu: SMP, using flush cache mode for C3\n"));
1275     }
1276 
1277     /* Look for various quirks of the PIIX4 part. */
1278     acpi_cpu_quirks_piix4();
1279 }
1280 
1281 static void
1282 acpi_cpu_quirks_piix4(void)
1283 {
1284 #ifdef __i386__
1285     device_t acpi_dev;
1286     uint32_t val;
1287     ACPI_STATUS status;
1288 
1289     acpi_dev = pci_find_device(PCI_VENDOR_INTEL, PCI_DEVICE_82371AB_3);
1290     if (acpi_dev != NULL) {
1291 	switch (pci_get_revid(acpi_dev)) {
1292 	/*
1293 	 * Disable C3 support for all PIIX4 chipsets.  Some of these parts
1294 	 * do not report the BMIDE status to the BM status register and
1295 	 * others have a livelock bug if Type-F DMA is enabled.  Linux
1296 	 * works around the BMIDE bug by reading the BM status directly
1297 	 * but we take the simpler approach of disabling C3 for these
1298 	 * parts.
1299 	 *
1300 	 * See erratum #18 ("C3 Power State/BMIDE and Type-F DMA
1301 	 * Livelock") from the January 2002 PIIX4 specification update.
1302 	 * Applies to all PIIX4 models.
1303 	 *
1304 	 * Also, make sure that all interrupts cause a "Stop Break"
1305 	 * event to exit from C2 state.
1306 	 * Also, BRLD_EN_BM (ACPI_BITREG_BUS_MASTER_RLD in ACPI-speak)
1307 	 * should be set to zero, otherwise it causes C2 to short-sleep.
1308 	 * PIIX4 doesn't properly support C3 and bus master activity
1309 	 * need not break out of C2.
1310 	 */
1311 	case PCI_REVISION_A_STEP:
1312 	case PCI_REVISION_B_STEP:
1313 	case PCI_REVISION_4E:
1314 	case PCI_REVISION_4M:
1315 	    cpu_quirks |= CPU_QUIRK_NO_C3;
1316 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1317 		"acpi_cpu: working around PIIX4 bug, disabling C3\n"));
1318 
1319 	    val = pci_read_config(acpi_dev, PIIX4_DEVACTB_REG, 4);
1320 	    if ((val & PIIX4_STOP_BREAK_MASK) != PIIX4_STOP_BREAK_MASK) {
1321 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1322 		    "acpi_cpu: PIIX4: enabling IRQs to generate Stop Break\n"));
1323 	    	val |= PIIX4_STOP_BREAK_MASK;
1324 		pci_write_config(acpi_dev, PIIX4_DEVACTB_REG, val, 4);
1325 	    }
1326 	    status = AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_RLD, &val);
1327 	    if (ACPI_SUCCESS(status) && val != 0) {
1328 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1329 		    "acpi_cpu: PIIX4: reset BRLD_EN_BM\n"));
1330 		AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1331 	    }
1332 	    break;
1333 	default:
1334 	    break;
1335 	}
1336     }
1337 #endif
1338 }
1339 
1340 static int
1341 acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS)
1342 {
1343 	struct acpi_cpu_softc *sc = (struct acpi_cpu_softc *)arg1;
1344 	struct sbuf	 sb;
1345 	char		 buf[128];
1346 	int		 error, i;
1347 	uintmax_t	 fract, sum, whole;
1348 
1349 	sbuf_new_for_sysctl(&sb, buf, sizeof(buf), req);
1350 	sum = 0;
1351 	for (i = 0; i < sc->cpu_cx_count; i++)
1352 		sum += sc->cpu_cx_stats[i];
1353 	for (i = 0; i < sc->cpu_cx_count; i++) {
1354 		if (sum > 0) {
1355 			whole = (uintmax_t)sc->cpu_cx_stats[i] * 100;
1356 			fract = (whole % sum) * 100;
1357 			sbuf_printf(&sb, "%u.%02u%% ", (u_int)(whole / sum),
1358 			    (u_int)(fract / sum));
1359 		} else
1360 			sbuf_printf(&sb, "0.00%% ");
1361 	}
1362 	sbuf_printf(&sb, "last %dus", sc->cpu_prev_sleep);
1363 	error = sbuf_finish(&sb);
1364 	sbuf_delete(&sb);
1365 	return (error);
1366 }
1367 
1368 /*
1369  * XXX TODO: actually add support to count each entry/exit
1370  * from the Cx states.
1371  */
1372 static int
1373 acpi_cpu_usage_counters_sysctl(SYSCTL_HANDLER_ARGS)
1374 {
1375 	struct acpi_cpu_softc *sc = (struct acpi_cpu_softc *)arg1;
1376 	struct sbuf	 sb;
1377 	char		 buf[128];
1378 	int		 error, i;
1379 
1380 	sbuf_new_for_sysctl(&sb, buf, sizeof(buf), req);
1381 	for (i = 0; i < sc->cpu_cx_count; i++) {
1382 		if (i > 0)
1383 			sbuf_putc(&sb, ' ');
1384 		sbuf_printf(&sb, "%u", sc->cpu_cx_stats[i]);
1385 	}
1386 	error = sbuf_finish(&sb);
1387 	sbuf_delete(&sb);
1388 	return (error);
1389 }
1390 
1391 #if defined(__i386__) || defined(__amd64__)
1392 static int
1393 acpi_cpu_method_sysctl(SYSCTL_HANDLER_ARGS)
1394 {
1395 	struct acpi_cpu_softc *sc = (struct acpi_cpu_softc *)arg1;
1396 	struct acpi_cx *cx;
1397 	struct sbuf sb;
1398 	char buf[128];
1399 	int error, i;
1400 
1401 	sbuf_new_for_sysctl(&sb, buf, sizeof(buf), req);
1402 	for (i = 0; i < sc->cpu_cx_count; i++) {
1403 		cx = &sc->cpu_cx_states[i];
1404 		if (i > 0)
1405 			sbuf_putc(&sb, ' ');
1406 		sbuf_printf(&sb, "C%d/", i + 1);
1407 		if (cx->do_mwait) {
1408 			sbuf_cat(&sb, "mwait");
1409 			if (cx->mwait_hw_coord)
1410 				sbuf_cat(&sb, "/hwc");
1411 			if (cx->mwait_bm_avoidance)
1412 				sbuf_cat(&sb, "/bma");
1413 		} else if (cx->type == ACPI_STATE_C1) {
1414 			sbuf_cat(&sb, "hlt");
1415 		} else {
1416 			sbuf_cat(&sb, "io");
1417 		}
1418 		if (cx->type == ACPI_STATE_C1 && cx->p_lvlx != NULL)
1419 			sbuf_cat(&sb, "/iohlt");
1420 	}
1421 	error = sbuf_finish(&sb);
1422 	sbuf_delete(&sb);
1423 	return (error);
1424 }
1425 #endif
1426 
1427 static int
1428 acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc)
1429 {
1430     int i;
1431 
1432     ACPI_SERIAL_ASSERT(cpu);
1433     sc->cpu_cx_lowest = min(sc->cpu_cx_lowest_lim, sc->cpu_cx_count - 1);
1434 
1435     /* If not disabling, cache the new lowest non-C3 state. */
1436     sc->cpu_non_c3 = 0;
1437     for (i = sc->cpu_cx_lowest; i >= 0; i--) {
1438 	if (sc->cpu_cx_states[i].type < ACPI_STATE_C3) {
1439 	    sc->cpu_non_c3 = i;
1440 	    break;
1441 	}
1442     }
1443 
1444     /* Reset the statistics counters. */
1445     bzero(sc->cpu_cx_stats, sizeof(sc->cpu_cx_stats));
1446     return (0);
1447 }
1448 
1449 static int
1450 acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1451 {
1452     struct	 acpi_cpu_softc *sc;
1453     char	 state[8];
1454     int		 val, error;
1455 
1456     sc = (struct acpi_cpu_softc *) arg1;
1457     snprintf(state, sizeof(state), "C%d", sc->cpu_cx_lowest_lim + 1);
1458     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1459     if (error != 0 || req->newptr == NULL)
1460 	return (error);
1461     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1462 	return (EINVAL);
1463     if (strcasecmp(state, "Cmax") == 0)
1464 	val = MAX_CX_STATES;
1465     else {
1466 	val = (int) strtol(state + 1, NULL, 10);
1467 	if (val < 1 || val > MAX_CX_STATES)
1468 	    return (EINVAL);
1469     }
1470 
1471     ACPI_SERIAL_BEGIN(cpu);
1472     sc->cpu_cx_lowest_lim = val - 1;
1473     acpi_cpu_set_cx_lowest(sc);
1474     ACPI_SERIAL_END(cpu);
1475 
1476     return (0);
1477 }
1478 
1479 static int
1480 acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1481 {
1482     struct	acpi_cpu_softc *sc;
1483     char	state[8];
1484     int		val, error, i;
1485 
1486     snprintf(state, sizeof(state), "C%d", cpu_cx_lowest_lim + 1);
1487     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1488     if (error != 0 || req->newptr == NULL)
1489 	return (error);
1490     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1491 	return (EINVAL);
1492     if (strcasecmp(state, "Cmax") == 0)
1493 	val = MAX_CX_STATES;
1494     else {
1495 	val = (int) strtol(state + 1, NULL, 10);
1496 	if (val < 1 || val > MAX_CX_STATES)
1497 	    return (EINVAL);
1498     }
1499 
1500     /* Update the new lowest useable Cx state for all CPUs. */
1501     ACPI_SERIAL_BEGIN(cpu);
1502     cpu_cx_lowest_lim = val - 1;
1503     CPU_FOREACH(i) {
1504 	if ((sc = cpu_softc[i]) == NULL)
1505 	    continue;
1506 	sc->cpu_cx_lowest_lim = cpu_cx_lowest_lim;
1507 	acpi_cpu_set_cx_lowest(sc);
1508     }
1509     ACPI_SERIAL_END(cpu);
1510 
1511     return (0);
1512 }
1513