xref: /freebsd/sys/dev/acpica/acpi_cpu.c (revision 145992504973bd16cf3518af9ba5ce185fefa82a)
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/sbuf.h>
42 #include <sys/smp.h>
43 
44 #include <dev/pci/pcivar.h>
45 #include <machine/atomic.h>
46 #include <machine/bus.h>
47 #if defined(__amd64__) || defined(__i386__)
48 #include <machine/clock.h>
49 #endif
50 #include <sys/rman.h>
51 
52 #include <contrib/dev/acpica/include/acpi.h>
53 #include <contrib/dev/acpica/include/accommon.h>
54 
55 #include <dev/acpica/acpivar.h>
56 
57 /*
58  * Support for ACPI Processor devices, including C[1-3] sleep states.
59  */
60 
61 /* Hooks for the ACPI CA debugging infrastructure */
62 #define _COMPONENT	ACPI_PROCESSOR
63 ACPI_MODULE_NAME("PROCESSOR")
64 
65 struct acpi_cx {
66     struct resource	*p_lvlx;	/* Register to read to enter state. */
67     uint32_t		 type;		/* C1-3 (C4 and up treated as C3). */
68     uint32_t		 trans_lat;	/* Transition latency (usec). */
69     uint32_t		 power;		/* Power consumed (mW). */
70     int			 res_type;	/* Resource type for p_lvlx. */
71 };
72 #define MAX_CX_STATES	 8
73 
74 struct acpi_cpu_softc {
75     device_t		 cpu_dev;
76     ACPI_HANDLE		 cpu_handle;
77     struct pcpu		*cpu_pcpu;
78     uint32_t		 cpu_acpi_id;	/* ACPI processor id */
79     uint32_t		 cpu_p_blk;	/* ACPI P_BLK location */
80     uint32_t		 cpu_p_blk_len;	/* P_BLK length (must be 6). */
81     struct acpi_cx	 cpu_cx_states[MAX_CX_STATES];
82     int			 cpu_cx_count;	/* Number of valid Cx states. */
83     int			 cpu_prev_sleep;/* Last idle sleep duration. */
84     int			 cpu_features;	/* Child driver supported features. */
85     /* Runtime state. */
86     int			 cpu_non_c3;	/* Index of lowest non-C3 state. */
87     u_int		 cpu_cx_stats[MAX_CX_STATES];/* Cx usage history. */
88     /* Values for sysctl. */
89     struct sysctl_ctx_list cpu_sysctl_ctx;
90     struct sysctl_oid	*cpu_sysctl_tree;
91     int			 cpu_cx_lowest;
92     int			 cpu_cx_lowest_lim;
93     char 		 cpu_cx_supported[64];
94     int			 cpu_rid;
95 };
96 
97 struct acpi_cpu_device {
98     struct resource_list	ad_rl;
99 };
100 
101 #define CPU_GET_REG(reg, width) 					\
102     (bus_space_read_ ## width(rman_get_bustag((reg)), 			\
103 		      rman_get_bushandle((reg)), 0))
104 #define CPU_SET_REG(reg, width, val)					\
105     (bus_space_write_ ## width(rman_get_bustag((reg)), 			\
106 		       rman_get_bushandle((reg)), 0, (val)))
107 
108 #define PM_USEC(x)	 ((x) >> 2)	/* ~4 clocks per usec (3.57955 Mhz) */
109 
110 #define ACPI_NOTIFY_CX_STATES	0x81	/* _CST changed. */
111 
112 #define CPU_QUIRK_NO_C3		(1<<0)	/* C3-type states are not usable. */
113 #define CPU_QUIRK_NO_BM_CTRL	(1<<2)	/* No bus mastering control. */
114 
115 #define PCI_VENDOR_INTEL	0x8086
116 #define PCI_DEVICE_82371AB_3	0x7113	/* PIIX4 chipset for quirks. */
117 #define PCI_REVISION_A_STEP	0
118 #define PCI_REVISION_B_STEP	1
119 #define PCI_REVISION_4E		2
120 #define PCI_REVISION_4M		3
121 #define PIIX4_DEVACTB_REG	0x58
122 #define PIIX4_BRLD_EN_IRQ0	(1<<0)
123 #define PIIX4_BRLD_EN_IRQ	(1<<1)
124 #define PIIX4_BRLD_EN_IRQ8	(1<<5)
125 #define PIIX4_STOP_BREAK_MASK	(PIIX4_BRLD_EN_IRQ0 | PIIX4_BRLD_EN_IRQ | PIIX4_BRLD_EN_IRQ8)
126 #define PIIX4_PCNTRL_BST_EN	(1<<10)
127 
128 /* Allow users to ignore processor orders in MADT. */
129 static int cpu_unordered;
130 TUNABLE_INT("debug.acpi.cpu_unordered", &cpu_unordered);
131 SYSCTL_INT(_debug_acpi, OID_AUTO, cpu_unordered, CTLFLAG_RDTUN,
132     &cpu_unordered, 0,
133     "Do not use the MADT to match ACPI Processor objects to CPUs.");
134 
135 /* Platform hardware resource information. */
136 static uint32_t		 cpu_smi_cmd;	/* Value to write to SMI_CMD. */
137 static uint8_t		 cpu_cst_cnt;	/* Indicate we are _CST aware. */
138 static int		 cpu_quirks;	/* Indicate any hardware bugs. */
139 
140 /* Runtime state. */
141 static int		 cpu_disable_idle; /* Disable entry to idle function */
142 
143 /* Values for sysctl. */
144 static struct sysctl_ctx_list cpu_sysctl_ctx;
145 static struct sysctl_oid *cpu_sysctl_tree;
146 static int		 cpu_cx_generic;
147 static int		 cpu_cx_lowest_lim;
148 
149 static device_t		*cpu_devices;
150 static int		 cpu_ndevices;
151 static struct acpi_cpu_softc **cpu_softc;
152 ACPI_SERIAL_DECL(cpu, "ACPI CPU");
153 
154 static int	acpi_cpu_probe(device_t dev);
155 static int	acpi_cpu_attach(device_t dev);
156 static int	acpi_cpu_suspend(device_t dev);
157 static int	acpi_cpu_resume(device_t dev);
158 static int	acpi_pcpu_get_id(device_t dev, uint32_t *acpi_id,
159 		    uint32_t *cpu_id);
160 static struct resource_list *acpi_cpu_get_rlist(device_t dev, device_t child);
161 static device_t	acpi_cpu_add_child(device_t dev, u_int order, const char *name,
162 		    int unit);
163 static int	acpi_cpu_read_ivar(device_t dev, device_t child, int index,
164 		    uintptr_t *result);
165 static int	acpi_cpu_shutdown(device_t dev);
166 static void	acpi_cpu_cx_probe(struct acpi_cpu_softc *sc);
167 static void	acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc);
168 static int	acpi_cpu_cx_cst(struct acpi_cpu_softc *sc);
169 static void	acpi_cpu_startup(void *arg);
170 static void	acpi_cpu_startup_cx(struct acpi_cpu_softc *sc);
171 static void	acpi_cpu_cx_list(struct acpi_cpu_softc *sc);
172 static void	acpi_cpu_idle(void);
173 static void	acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context);
174 static int	acpi_cpu_quirks(void);
175 static int	acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS);
176 static int	acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc);
177 static int	acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
178 static int	acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
179 
180 static device_method_t acpi_cpu_methods[] = {
181     /* Device interface */
182     DEVMETHOD(device_probe,	acpi_cpu_probe),
183     DEVMETHOD(device_attach,	acpi_cpu_attach),
184     DEVMETHOD(device_detach,	bus_generic_detach),
185     DEVMETHOD(device_shutdown,	acpi_cpu_shutdown),
186     DEVMETHOD(device_suspend,	acpi_cpu_suspend),
187     DEVMETHOD(device_resume,	acpi_cpu_resume),
188 
189     /* Bus interface */
190     DEVMETHOD(bus_add_child,	acpi_cpu_add_child),
191     DEVMETHOD(bus_read_ivar,	acpi_cpu_read_ivar),
192     DEVMETHOD(bus_get_resource_list, acpi_cpu_get_rlist),
193     DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
194     DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
195     DEVMETHOD(bus_alloc_resource, bus_generic_rl_alloc_resource),
196     DEVMETHOD(bus_release_resource, bus_generic_rl_release_resource),
197     DEVMETHOD(bus_activate_resource, bus_generic_activate_resource),
198     DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource),
199     DEVMETHOD(bus_setup_intr,	bus_generic_setup_intr),
200     DEVMETHOD(bus_teardown_intr, bus_generic_teardown_intr),
201 
202     DEVMETHOD_END
203 };
204 
205 static driver_t acpi_cpu_driver = {
206     "cpu",
207     acpi_cpu_methods,
208     sizeof(struct acpi_cpu_softc),
209 };
210 
211 static devclass_t acpi_cpu_devclass;
212 DRIVER_MODULE(cpu, acpi, acpi_cpu_driver, acpi_cpu_devclass, 0, 0);
213 MODULE_DEPEND(cpu, acpi, 1, 1, 1);
214 
215 static int
216 acpi_cpu_probe(device_t dev)
217 {
218     int			   acpi_id, cpu_id;
219     ACPI_BUFFER		   buf;
220     ACPI_HANDLE		   handle;
221     ACPI_OBJECT		   *obj;
222     ACPI_STATUS		   status;
223 
224     if (acpi_disabled("cpu") || acpi_get_type(dev) != ACPI_TYPE_PROCESSOR)
225 	return (ENXIO);
226 
227     handle = acpi_get_handle(dev);
228     if (cpu_softc == NULL)
229 	cpu_softc = malloc(sizeof(struct acpi_cpu_softc *) *
230 	    (mp_maxid + 1), M_TEMP /* XXX */, M_WAITOK | M_ZERO);
231 
232     /* Get our Processor object. */
233     buf.Pointer = NULL;
234     buf.Length = ACPI_ALLOCATE_BUFFER;
235     status = AcpiEvaluateObject(handle, NULL, NULL, &buf);
236     if (ACPI_FAILURE(status)) {
237 	device_printf(dev, "probe failed to get Processor obj - %s\n",
238 		      AcpiFormatException(status));
239 	return (ENXIO);
240     }
241     obj = (ACPI_OBJECT *)buf.Pointer;
242     if (obj->Type != ACPI_TYPE_PROCESSOR) {
243 	device_printf(dev, "Processor object has bad type %d\n", obj->Type);
244 	AcpiOsFree(obj);
245 	return (ENXIO);
246     }
247 
248     /*
249      * Find the processor associated with our unit.  We could use the
250      * ProcId as a key, however, some boxes do not have the same values
251      * in their Processor object as the ProcId values in the MADT.
252      */
253     acpi_id = obj->Processor.ProcId;
254     AcpiOsFree(obj);
255     if (acpi_pcpu_get_id(dev, &acpi_id, &cpu_id) != 0)
256 	return (ENXIO);
257 
258     /*
259      * Check if we already probed this processor.  We scan the bus twice
260      * so it's possible we've already seen this one.
261      */
262     if (cpu_softc[cpu_id] != NULL)
263 	return (ENXIO);
264 
265     /* Mark this processor as in-use and save our derived id for attach. */
266     cpu_softc[cpu_id] = (void *)1;
267     acpi_set_private(dev, (void*)(intptr_t)cpu_id);
268     device_set_desc(dev, "ACPI CPU");
269 
270     return (0);
271 }
272 
273 static int
274 acpi_cpu_attach(device_t dev)
275 {
276     ACPI_BUFFER		   buf;
277     ACPI_OBJECT		   arg[4], *obj;
278     ACPI_OBJECT_LIST	   arglist;
279     struct pcpu		   *pcpu_data;
280     struct acpi_cpu_softc *sc;
281     struct acpi_softc	  *acpi_sc;
282     ACPI_STATUS		   status;
283     u_int		   features;
284     int			   cpu_id, drv_count, i;
285     driver_t 		  **drivers;
286     uint32_t		   cap_set[3];
287 
288     /* UUID needed by _OSC evaluation */
289     static uint8_t cpu_oscuuid[16] = { 0x16, 0xA6, 0x77, 0x40, 0x0C, 0x29,
290 				       0xBE, 0x47, 0x9E, 0xBD, 0xD8, 0x70,
291 				       0x58, 0x71, 0x39, 0x53 };
292 
293     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
294 
295     sc = device_get_softc(dev);
296     sc->cpu_dev = dev;
297     sc->cpu_handle = acpi_get_handle(dev);
298     cpu_id = (int)(intptr_t)acpi_get_private(dev);
299     cpu_softc[cpu_id] = sc;
300     pcpu_data = pcpu_find(cpu_id);
301     pcpu_data->pc_device = dev;
302     sc->cpu_pcpu = pcpu_data;
303     cpu_smi_cmd = AcpiGbl_FADT.SmiCommand;
304     cpu_cst_cnt = AcpiGbl_FADT.CstControl;
305 
306     buf.Pointer = NULL;
307     buf.Length = ACPI_ALLOCATE_BUFFER;
308     status = AcpiEvaluateObject(sc->cpu_handle, NULL, NULL, &buf);
309     if (ACPI_FAILURE(status)) {
310 	device_printf(dev, "attach failed to get Processor obj - %s\n",
311 		      AcpiFormatException(status));
312 	return (ENXIO);
313     }
314     obj = (ACPI_OBJECT *)buf.Pointer;
315     sc->cpu_p_blk = obj->Processor.PblkAddress;
316     sc->cpu_p_blk_len = obj->Processor.PblkLength;
317     sc->cpu_acpi_id = obj->Processor.ProcId;
318     AcpiOsFree(obj);
319     ACPI_DEBUG_PRINT((ACPI_DB_INFO, "acpi_cpu%d: P_BLK at %#x/%d\n",
320 		     device_get_unit(dev), sc->cpu_p_blk, sc->cpu_p_blk_len));
321 
322     /*
323      * If this is the first cpu we attach, create and initialize the generic
324      * resources that will be used by all acpi cpu devices.
325      */
326     if (device_get_unit(dev) == 0) {
327 	/* Assume we won't be using generic Cx mode by default */
328 	cpu_cx_generic = FALSE;
329 
330 	/* Install hw.acpi.cpu sysctl tree */
331 	acpi_sc = acpi_device_get_parent_softc(dev);
332 	sysctl_ctx_init(&cpu_sysctl_ctx);
333 	cpu_sysctl_tree = SYSCTL_ADD_NODE(&cpu_sysctl_ctx,
334 	    SYSCTL_CHILDREN(acpi_sc->acpi_sysctl_tree), OID_AUTO, "cpu",
335 	    CTLFLAG_RD, 0, "node for CPU children");
336 
337 	/* Queue post cpu-probing task handler */
338 	AcpiOsExecute(OSL_NOTIFY_HANDLER, acpi_cpu_startup, NULL);
339     }
340 
341     /*
342      * Before calling any CPU methods, collect child driver feature hints
343      * and notify ACPI of them.  We support unified SMP power control
344      * so advertise this ourselves.  Note this is not the same as independent
345      * SMP control where each CPU can have different settings.
346      */
347     sc->cpu_features = ACPI_CAP_SMP_SAME | ACPI_CAP_SMP_SAME_C3;
348     if (devclass_get_drivers(acpi_cpu_devclass, &drivers, &drv_count) == 0) {
349 	for (i = 0; i < drv_count; i++) {
350 	    if (ACPI_GET_FEATURES(drivers[i], &features) == 0)
351 		sc->cpu_features |= features;
352 	}
353 	free(drivers, M_TEMP);
354     }
355 
356     /*
357      * CPU capabilities are specified in
358      * Intel Processor Vendor-Specific ACPI Interface Specification.
359      */
360     if (sc->cpu_features) {
361 	arglist.Pointer = arg;
362 	arglist.Count = 4;
363 	arg[0].Type = ACPI_TYPE_BUFFER;
364 	arg[0].Buffer.Length = sizeof(cpu_oscuuid);
365 	arg[0].Buffer.Pointer = cpu_oscuuid;	/* UUID */
366 	arg[1].Type = ACPI_TYPE_INTEGER;
367 	arg[1].Integer.Value = 1;		/* revision */
368 	arg[2].Type = ACPI_TYPE_INTEGER;
369 	arg[2].Integer.Value = 1;		/* count */
370 	arg[3].Type = ACPI_TYPE_BUFFER;
371 	arg[3].Buffer.Length = sizeof(cap_set);	/* Capabilities buffer */
372 	arg[3].Buffer.Pointer = (uint8_t *)cap_set;
373 	cap_set[0] = 0;				/* status */
374 	cap_set[1] = sc->cpu_features;
375 	status = AcpiEvaluateObject(sc->cpu_handle, "_OSC", &arglist, NULL);
376 	if (ACPI_SUCCESS(status)) {
377 	    if (cap_set[0] != 0)
378 		device_printf(dev, "_OSC returned status %#x\n", cap_set[0]);
379 	}
380 	else {
381 	    arglist.Pointer = arg;
382 	    arglist.Count = 1;
383 	    arg[0].Type = ACPI_TYPE_BUFFER;
384 	    arg[0].Buffer.Length = sizeof(cap_set);
385 	    arg[0].Buffer.Pointer = (uint8_t *)cap_set;
386 	    cap_set[0] = 1; /* revision */
387 	    cap_set[1] = 1; /* number of capabilities integers */
388 	    cap_set[2] = sc->cpu_features;
389 	    AcpiEvaluateObject(sc->cpu_handle, "_PDC", &arglist, NULL);
390 	}
391     }
392 
393     /* Probe for Cx state support. */
394     acpi_cpu_cx_probe(sc);
395 
396     return (0);
397 }
398 
399 static void
400 acpi_cpu_postattach(void *unused __unused)
401 {
402     device_t *devices;
403     int err;
404     int i, n;
405 
406     err = devclass_get_devices(acpi_cpu_devclass, &devices, &n);
407     if (err != 0) {
408 	printf("devclass_get_devices(acpi_cpu_devclass) failed\n");
409 	return;
410     }
411     for (i = 0; i < n; i++)
412 	bus_generic_probe(devices[i]);
413     for (i = 0; i < n; i++)
414 	bus_generic_attach(devices[i]);
415     free(devices, M_TEMP);
416 }
417 
418 SYSINIT(acpi_cpu, SI_SUB_CONFIGURE, SI_ORDER_MIDDLE,
419     acpi_cpu_postattach, NULL);
420 
421 /*
422  * Disable any entry to the idle function during suspend and re-enable it
423  * during resume.
424  */
425 static int
426 acpi_cpu_suspend(device_t dev)
427 {
428     int error;
429 
430     error = bus_generic_suspend(dev);
431     if (error)
432 	return (error);
433     cpu_disable_idle = TRUE;
434     return (0);
435 }
436 
437 static int
438 acpi_cpu_resume(device_t dev)
439 {
440 
441     cpu_disable_idle = FALSE;
442     return (bus_generic_resume(dev));
443 }
444 
445 /*
446  * Find the processor associated with a given ACPI ID.  By default,
447  * use the MADT to map ACPI IDs to APIC IDs and use that to locate a
448  * processor.  Some systems have inconsistent ASL and MADT however.
449  * For these systems the cpu_unordered tunable can be set in which
450  * case we assume that Processor objects are listed in the same order
451  * in both the MADT and ASL.
452  */
453 static int
454 acpi_pcpu_get_id(device_t dev, uint32_t *acpi_id, uint32_t *cpu_id)
455 {
456     struct pcpu	*pc;
457     uint32_t	 i, idx;
458 
459     KASSERT(acpi_id != NULL, ("Null acpi_id"));
460     KASSERT(cpu_id != NULL, ("Null cpu_id"));
461     idx = device_get_unit(dev);
462 
463     /*
464      * If pc_acpi_id for CPU 0 is not initialized (e.g. a non-APIC
465      * UP box) use the ACPI ID from the first processor we find.
466      */
467     if (idx == 0 && mp_ncpus == 1) {
468 	pc = pcpu_find(0);
469 	if (pc->pc_acpi_id == 0xffffffff)
470 	    pc->pc_acpi_id = *acpi_id;
471 	*cpu_id = 0;
472 	return (0);
473     }
474 
475     CPU_FOREACH(i) {
476 	pc = pcpu_find(i);
477 	KASSERT(pc != NULL, ("no pcpu data for %d", i));
478 	if (cpu_unordered) {
479 	    if (idx-- == 0) {
480 		/*
481 		 * If pc_acpi_id doesn't match the ACPI ID from the
482 		 * ASL, prefer the MADT-derived value.
483 		 */
484 		if (pc->pc_acpi_id != *acpi_id)
485 		    *acpi_id = pc->pc_acpi_id;
486 		*cpu_id = pc->pc_cpuid;
487 		return (0);
488 	    }
489 	} else {
490 	    if (pc->pc_acpi_id == *acpi_id) {
491 		if (bootverbose)
492 		    device_printf(dev,
493 			"Processor %s (ACPI ID %u) -> APIC ID %d\n",
494 			acpi_name(acpi_get_handle(dev)), *acpi_id,
495 			pc->pc_cpuid);
496 		*cpu_id = pc->pc_cpuid;
497 		return (0);
498 	    }
499 	}
500     }
501 
502     if (bootverbose)
503 	printf("ACPI: Processor %s (ACPI ID %u) ignored\n",
504 	    acpi_name(acpi_get_handle(dev)), *acpi_id);
505 
506     return (ESRCH);
507 }
508 
509 static struct resource_list *
510 acpi_cpu_get_rlist(device_t dev, device_t child)
511 {
512     struct acpi_cpu_device *ad;
513 
514     ad = device_get_ivars(child);
515     if (ad == NULL)
516 	return (NULL);
517     return (&ad->ad_rl);
518 }
519 
520 static device_t
521 acpi_cpu_add_child(device_t dev, u_int order, const char *name, int unit)
522 {
523     struct acpi_cpu_device *ad;
524     device_t child;
525 
526     if ((ad = malloc(sizeof(*ad), M_TEMP, M_NOWAIT | M_ZERO)) == NULL)
527 	return (NULL);
528 
529     resource_list_init(&ad->ad_rl);
530 
531     child = device_add_child_ordered(dev, order, name, unit);
532     if (child != NULL)
533 	device_set_ivars(child, ad);
534     else
535 	free(ad, M_TEMP);
536     return (child);
537 }
538 
539 static int
540 acpi_cpu_read_ivar(device_t dev, device_t child, int index, uintptr_t *result)
541 {
542     struct acpi_cpu_softc *sc;
543 
544     sc = device_get_softc(dev);
545     switch (index) {
546     case ACPI_IVAR_HANDLE:
547 	*result = (uintptr_t)sc->cpu_handle;
548 	break;
549     case CPU_IVAR_PCPU:
550 	*result = (uintptr_t)sc->cpu_pcpu;
551 	break;
552 #if defined(__amd64__) || defined(__i386__)
553     case CPU_IVAR_NOMINAL_MHZ:
554 	if (tsc_is_invariant) {
555 	    *result = (uintptr_t)(atomic_load_acq_64(&tsc_freq) / 1000000);
556 	    break;
557 	}
558 	/* FALLTHROUGH */
559 #endif
560     default:
561 	return (ENOENT);
562     }
563     return (0);
564 }
565 
566 static int
567 acpi_cpu_shutdown(device_t dev)
568 {
569     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
570 
571     /* Allow children to shutdown first. */
572     bus_generic_shutdown(dev);
573 
574     /*
575      * Disable any entry to the idle function.  There is a small race where
576      * an idle thread have passed this check but not gone to sleep.  This
577      * is ok since device_shutdown() does not free the softc, otherwise
578      * we'd have to be sure all threads were evicted before returning.
579      */
580     cpu_disable_idle = TRUE;
581 
582     return_VALUE (0);
583 }
584 
585 static void
586 acpi_cpu_cx_probe(struct acpi_cpu_softc *sc)
587 {
588     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
589 
590     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
591     sc->cpu_prev_sleep = 1000000;
592     sc->cpu_cx_lowest = 0;
593     sc->cpu_cx_lowest_lim = 0;
594 
595     /*
596      * Check for the ACPI 2.0 _CST sleep states object. If we can't find
597      * any, we'll revert to generic FADT/P_BLK Cx control method which will
598      * be handled by acpi_cpu_startup. We need to defer to after having
599      * probed all the cpus in the system before probing for generic Cx
600      * states as we may already have found cpus with valid _CST packages
601      */
602     if (!cpu_cx_generic && acpi_cpu_cx_cst(sc) != 0) {
603 	/*
604 	 * We were unable to find a _CST package for this cpu or there
605 	 * was an error parsing it. Switch back to generic mode.
606 	 */
607 	cpu_cx_generic = TRUE;
608 	if (bootverbose)
609 	    device_printf(sc->cpu_dev, "switching to generic Cx mode\n");
610     }
611 
612     /*
613      * TODO: _CSD Package should be checked here.
614      */
615 }
616 
617 static void
618 acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc)
619 {
620     ACPI_GENERIC_ADDRESS	 gas;
621     struct acpi_cx		*cx_ptr;
622 
623     sc->cpu_cx_count = 0;
624     cx_ptr = sc->cpu_cx_states;
625 
626     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
627     sc->cpu_prev_sleep = 1000000;
628 
629     /* C1 has been required since just after ACPI 1.0 */
630     cx_ptr->type = ACPI_STATE_C1;
631     cx_ptr->trans_lat = 0;
632     cx_ptr++;
633     sc->cpu_non_c3 = sc->cpu_cx_count;
634     sc->cpu_cx_count++;
635 
636     /*
637      * The spec says P_BLK must be 6 bytes long.  However, some systems
638      * use it to indicate a fractional set of features present so we
639      * take 5 as C2.  Some may also have a value of 7 to indicate
640      * another C3 but most use _CST for this (as required) and having
641      * "only" C1-C3 is not a hardship.
642      */
643     if (sc->cpu_p_blk_len < 5)
644 	return;
645 
646     /* Validate and allocate resources for C2 (P_LVL2). */
647     gas.SpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
648     gas.BitWidth = 8;
649     if (AcpiGbl_FADT.C2Latency <= 100) {
650 	gas.Address = sc->cpu_p_blk + 4;
651 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &sc->cpu_rid,
652 	    &gas, &cx_ptr->p_lvlx, RF_SHAREABLE);
653 	if (cx_ptr->p_lvlx != NULL) {
654 	    sc->cpu_rid++;
655 	    cx_ptr->type = ACPI_STATE_C2;
656 	    cx_ptr->trans_lat = AcpiGbl_FADT.C2Latency;
657 	    cx_ptr++;
658 	    sc->cpu_non_c3 = sc->cpu_cx_count;
659 	    sc->cpu_cx_count++;
660 	}
661     }
662     if (sc->cpu_p_blk_len < 6)
663 	return;
664 
665     /* Validate and allocate resources for C3 (P_LVL3). */
666     if (AcpiGbl_FADT.C3Latency <= 1000 && !(cpu_quirks & CPU_QUIRK_NO_C3)) {
667 	gas.Address = sc->cpu_p_blk + 5;
668 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &sc->cpu_rid, &gas,
669 	    &cx_ptr->p_lvlx, RF_SHAREABLE);
670 	if (cx_ptr->p_lvlx != NULL) {
671 	    sc->cpu_rid++;
672 	    cx_ptr->type = ACPI_STATE_C3;
673 	    cx_ptr->trans_lat = AcpiGbl_FADT.C3Latency;
674 	    cx_ptr++;
675 	    sc->cpu_cx_count++;
676 	    cpu_can_deep_sleep = 1;
677 	}
678     }
679 }
680 
681 /*
682  * Parse a _CST package and set up its Cx states.  Since the _CST object
683  * can change dynamically, our notify handler may call this function
684  * to clean up and probe the new _CST package.
685  */
686 static int
687 acpi_cpu_cx_cst(struct acpi_cpu_softc *sc)
688 {
689     struct	 acpi_cx *cx_ptr;
690     ACPI_STATUS	 status;
691     ACPI_BUFFER	 buf;
692     ACPI_OBJECT	*top;
693     ACPI_OBJECT	*pkg;
694     uint32_t	 count;
695     int		 i;
696 
697     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
698 
699     buf.Pointer = NULL;
700     buf.Length = ACPI_ALLOCATE_BUFFER;
701     status = AcpiEvaluateObject(sc->cpu_handle, "_CST", NULL, &buf);
702     if (ACPI_FAILURE(status))
703 	return (ENXIO);
704 
705     /* _CST is a package with a count and at least one Cx package. */
706     top = (ACPI_OBJECT *)buf.Pointer;
707     if (!ACPI_PKG_VALID(top, 2) || acpi_PkgInt32(top, 0, &count) != 0) {
708 	device_printf(sc->cpu_dev, "invalid _CST package\n");
709 	AcpiOsFree(buf.Pointer);
710 	return (ENXIO);
711     }
712     if (count != top->Package.Count - 1) {
713 	device_printf(sc->cpu_dev, "invalid _CST state count (%d != %d)\n",
714 	       count, top->Package.Count - 1);
715 	count = top->Package.Count - 1;
716     }
717     if (count > MAX_CX_STATES) {
718 	device_printf(sc->cpu_dev, "_CST has too many states (%d)\n", count);
719 	count = MAX_CX_STATES;
720     }
721 
722     sc->cpu_non_c3 = 0;
723     sc->cpu_cx_count = 0;
724     cx_ptr = sc->cpu_cx_states;
725 
726     /*
727      * C1 has been required since just after ACPI 1.0.
728      * Reserve the first slot for it.
729      */
730     cx_ptr->type = ACPI_STATE_C0;
731     cx_ptr++;
732     sc->cpu_cx_count++;
733 
734     /* Set up all valid states. */
735     for (i = 0; i < count; i++) {
736 	pkg = &top->Package.Elements[i + 1];
737 	if (!ACPI_PKG_VALID(pkg, 4) ||
738 	    acpi_PkgInt32(pkg, 1, &cx_ptr->type) != 0 ||
739 	    acpi_PkgInt32(pkg, 2, &cx_ptr->trans_lat) != 0 ||
740 	    acpi_PkgInt32(pkg, 3, &cx_ptr->power) != 0) {
741 
742 	    device_printf(sc->cpu_dev, "skipping invalid Cx state package\n");
743 	    continue;
744 	}
745 
746 	/* Validate the state to see if we should use it. */
747 	switch (cx_ptr->type) {
748 	case ACPI_STATE_C1:
749 	    if (sc->cpu_cx_states[0].type == ACPI_STATE_C0) {
750 		/* This is the first C1 state.  Use the reserved slot. */
751 		sc->cpu_cx_states[0] = *cx_ptr;
752 	    } else {
753 		sc->cpu_non_c3 = sc->cpu_cx_count;
754 		cx_ptr++;
755 		sc->cpu_cx_count++;
756 	    }
757 	    continue;
758 	case ACPI_STATE_C2:
759 	    sc->cpu_non_c3 = sc->cpu_cx_count;
760 	    break;
761 	case ACPI_STATE_C3:
762 	default:
763 	    if ((cpu_quirks & CPU_QUIRK_NO_C3) != 0) {
764 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
765 				 "acpi_cpu%d: C3[%d] not available.\n",
766 				 device_get_unit(sc->cpu_dev), i));
767 		continue;
768 	    } else
769 		cpu_can_deep_sleep = 1;
770 	    break;
771 	}
772 
773 #ifdef notyet
774 	/* Free up any previous register. */
775 	if (cx_ptr->p_lvlx != NULL) {
776 	    bus_release_resource(sc->cpu_dev, 0, 0, cx_ptr->p_lvlx);
777 	    cx_ptr->p_lvlx = NULL;
778 	}
779 #endif
780 
781 	/* Allocate the control register for C2 or C3. */
782 	acpi_PkgGas(sc->cpu_dev, pkg, 0, &cx_ptr->res_type, &sc->cpu_rid,
783 	    &cx_ptr->p_lvlx, RF_SHAREABLE);
784 	if (cx_ptr->p_lvlx) {
785 	    sc->cpu_rid++;
786 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
787 			     "acpi_cpu%d: Got C%d - %d latency\n",
788 			     device_get_unit(sc->cpu_dev), cx_ptr->type,
789 			     cx_ptr->trans_lat));
790 	    cx_ptr++;
791 	    sc->cpu_cx_count++;
792 	}
793     }
794     AcpiOsFree(buf.Pointer);
795 
796     /* If C1 state was not found, we need one now. */
797     cx_ptr = sc->cpu_cx_states;
798     if (cx_ptr->type == ACPI_STATE_C0) {
799 	cx_ptr->type = ACPI_STATE_C1;
800 	cx_ptr->trans_lat = 0;
801     }
802 
803     return (0);
804 }
805 
806 /*
807  * Call this *after* all CPUs have been attached.
808  */
809 static void
810 acpi_cpu_startup(void *arg)
811 {
812     struct acpi_cpu_softc *sc;
813     int i;
814 
815     /* Get set of CPU devices */
816     devclass_get_devices(acpi_cpu_devclass, &cpu_devices, &cpu_ndevices);
817 
818     /*
819      * Setup any quirks that might necessary now that we have probed
820      * all the CPUs
821      */
822     acpi_cpu_quirks();
823 
824     if (cpu_cx_generic) {
825 	/*
826 	 * We are using generic Cx mode, probe for available Cx states
827 	 * for all processors.
828 	 */
829 	for (i = 0; i < cpu_ndevices; i++) {
830 	    sc = device_get_softc(cpu_devices[i]);
831 	    acpi_cpu_generic_cx_probe(sc);
832 	}
833     } else {
834 	/*
835 	 * We are using _CST mode, remove C3 state if necessary.
836 	 * As we now know for sure that we will be using _CST mode
837 	 * install our notify handler.
838 	 */
839 	for (i = 0; i < cpu_ndevices; i++) {
840 	    sc = device_get_softc(cpu_devices[i]);
841 	    if (cpu_quirks & CPU_QUIRK_NO_C3) {
842 		sc->cpu_cx_count = sc->cpu_non_c3 + 1;
843 	    }
844 	    AcpiInstallNotifyHandler(sc->cpu_handle, ACPI_DEVICE_NOTIFY,
845 		acpi_cpu_notify, sc);
846 	}
847     }
848 
849     /* Perform Cx final initialization. */
850     for (i = 0; i < cpu_ndevices; i++) {
851 	sc = device_get_softc(cpu_devices[i]);
852 	acpi_cpu_startup_cx(sc);
853     }
854 
855     /* Add a sysctl handler to handle global Cx lowest setting */
856     SYSCTL_ADD_PROC(&cpu_sysctl_ctx, SYSCTL_CHILDREN(cpu_sysctl_tree),
857 	OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
858 	NULL, 0, acpi_cpu_global_cx_lowest_sysctl, "A",
859 	"Global lowest Cx sleep state to use");
860 
861     /* Take over idling from cpu_idle_default(). */
862     cpu_cx_lowest_lim = 0;
863     cpu_disable_idle = FALSE;
864     cpu_idle_hook = acpi_cpu_idle;
865 }
866 
867 static void
868 acpi_cpu_cx_list(struct acpi_cpu_softc *sc)
869 {
870     struct sbuf sb;
871     int i;
872 
873     /*
874      * Set up the list of Cx states
875      */
876     sbuf_new(&sb, sc->cpu_cx_supported, sizeof(sc->cpu_cx_supported),
877 	SBUF_FIXEDLEN);
878     for (i = 0; i < sc->cpu_cx_count; i++)
879 	sbuf_printf(&sb, "C%d/%d/%d ", i + 1, sc->cpu_cx_states[i].type,
880 	    sc->cpu_cx_states[i].trans_lat);
881     sbuf_trim(&sb);
882     sbuf_finish(&sb);
883 }
884 
885 static void
886 acpi_cpu_startup_cx(struct acpi_cpu_softc *sc)
887 {
888     acpi_cpu_cx_list(sc);
889 
890     SYSCTL_ADD_STRING(&sc->cpu_sysctl_ctx,
891 		      SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
892 		      OID_AUTO, "cx_supported", CTLFLAG_RD,
893 		      sc->cpu_cx_supported, 0,
894 		      "Cx/microsecond values for supported Cx states");
895     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
896 		    SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
897 		    OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
898 		    (void *)sc, 0, acpi_cpu_cx_lowest_sysctl, "A",
899 		    "lowest Cx sleep state to use");
900     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
901 		    SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
902 		    OID_AUTO, "cx_usage", CTLTYPE_STRING | CTLFLAG_RD,
903 		    (void *)sc, 0, acpi_cpu_usage_sysctl, "A",
904 		    "percent usage for each Cx state");
905 
906     /* Signal platform that we can handle _CST notification. */
907     if (!cpu_cx_generic && cpu_cst_cnt != 0) {
908 	ACPI_LOCK(acpi);
909 	AcpiOsWritePort(cpu_smi_cmd, cpu_cst_cnt, 8);
910 	ACPI_UNLOCK(acpi);
911     }
912 }
913 
914 /*
915  * Idle the CPU in the lowest state possible.  This function is called with
916  * interrupts disabled.  Note that once it re-enables interrupts, a task
917  * switch can occur so do not access shared data (i.e. the softc) after
918  * interrupts are re-enabled.
919  */
920 static void
921 acpi_cpu_idle()
922 {
923     struct	acpi_cpu_softc *sc;
924     struct	acpi_cx *cx_next;
925     uint64_t	cputicks;
926     uint32_t	start_time, end_time;
927     int		bm_active, cx_next_idx, i;
928 
929     /* If disabled, return immediately. */
930     if (cpu_disable_idle) {
931 	ACPI_ENABLE_IRQS();
932 	return;
933     }
934 
935     /*
936      * Look up our CPU id to get our softc.  If it's NULL, we'll use C1
937      * since there is no ACPI processor object for this CPU.  This occurs
938      * for logical CPUs in the HTT case.
939      */
940     sc = cpu_softc[PCPU_GET(cpuid)];
941     if (sc == NULL) {
942 	acpi_cpu_c1();
943 	return;
944     }
945 
946     /* Find the lowest state that has small enough latency. */
947     cx_next_idx = 0;
948     if (cpu_disable_deep_sleep)
949 	i = min(sc->cpu_cx_lowest, sc->cpu_non_c3);
950     else
951 	i = sc->cpu_cx_lowest;
952     for (; i >= 0; i--) {
953 	if (sc->cpu_cx_states[i].trans_lat * 3 <= sc->cpu_prev_sleep) {
954 	    cx_next_idx = i;
955 	    break;
956 	}
957     }
958 
959     /*
960      * Check for bus master activity.  If there was activity, clear
961      * the bit and use the lowest non-C3 state.  Note that the USB
962      * driver polling for new devices keeps this bit set all the
963      * time if USB is loaded.
964      */
965     if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0 &&
966 	cx_next_idx > sc->cpu_non_c3) {
967 	AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, &bm_active);
968 	if (bm_active != 0) {
969 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, 1);
970 	    cx_next_idx = sc->cpu_non_c3;
971 	}
972     }
973 
974     /* Select the next state and update statistics. */
975     cx_next = &sc->cpu_cx_states[cx_next_idx];
976     sc->cpu_cx_stats[cx_next_idx]++;
977     KASSERT(cx_next->type != ACPI_STATE_C0, ("acpi_cpu_idle: C0 sleep"));
978 
979     /*
980      * Execute HLT (or equivalent) and wait for an interrupt.  We can't
981      * precisely calculate the time spent in C1 since the place we wake up
982      * is an ISR.  Assume we slept no more then half of quantum, unless
983      * we are called inside critical section, delaying context switch.
984      */
985     if (cx_next->type == ACPI_STATE_C1) {
986 	cputicks = cpu_ticks();
987 	acpi_cpu_c1();
988 	end_time = ((cpu_ticks() - cputicks) << 20) / cpu_tickrate();
989 	if (curthread->td_critnest == 0)
990 		end_time = min(end_time, 500000 / hz);
991 	sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + end_time) / 4;
992 	return;
993     }
994 
995     /*
996      * For C3, disable bus master arbitration and enable bus master wake
997      * if BM control is available, otherwise flush the CPU cache.
998      */
999     if (cx_next->type == ACPI_STATE_C3) {
1000 	if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
1001 	    AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 1);
1002 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
1003 	} else
1004 	    ACPI_FLUSH_CPU_CACHE();
1005     }
1006 
1007     /*
1008      * Read from P_LVLx to enter C2(+), checking time spent asleep.
1009      * Use the ACPI timer for measuring sleep time.  Since we need to
1010      * get the time very close to the CPU start/stop clock logic, this
1011      * is the only reliable time source.
1012      */
1013     if (cx_next->type == ACPI_STATE_C3) {
1014 	AcpiHwRead(&start_time, &AcpiGbl_FADT.XPmTimerBlock);
1015 	cputicks = 0;
1016     } else {
1017 	start_time = 0;
1018 	cputicks = cpu_ticks();
1019     }
1020     CPU_GET_REG(cx_next->p_lvlx, 1);
1021 
1022     /*
1023      * Read the end time twice.  Since it may take an arbitrary time
1024      * to enter the idle state, the first read may be executed before
1025      * the processor has stopped.  Doing it again provides enough
1026      * margin that we are certain to have a correct value.
1027      */
1028     AcpiHwRead(&end_time, &AcpiGbl_FADT.XPmTimerBlock);
1029     if (cx_next->type == ACPI_STATE_C3) {
1030 	AcpiHwRead(&end_time, &AcpiGbl_FADT.XPmTimerBlock);
1031 	end_time = acpi_TimerDelta(end_time, start_time);
1032     } else
1033 	end_time = ((cpu_ticks() - cputicks) << 20) / cpu_tickrate();
1034 
1035     /* Enable bus master arbitration and disable bus master wakeup. */
1036     if (cx_next->type == ACPI_STATE_C3 &&
1037 	(cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
1038 	AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 0);
1039 	AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1040     }
1041     ACPI_ENABLE_IRQS();
1042 
1043     sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + PM_USEC(end_time)) / 4;
1044 }
1045 
1046 /*
1047  * Re-evaluate the _CST object when we are notified that it changed.
1048  *
1049  * XXX Re-evaluation disabled until locking is done.
1050  */
1051 static void
1052 acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context)
1053 {
1054     struct acpi_cpu_softc *sc = (struct acpi_cpu_softc *)context;
1055 
1056     if (notify != ACPI_NOTIFY_CX_STATES)
1057 	return;
1058 
1059     /* Update the list of Cx states. */
1060     acpi_cpu_cx_cst(sc);
1061     acpi_cpu_cx_list(sc);
1062 
1063     ACPI_SERIAL_BEGIN(cpu);
1064     acpi_cpu_set_cx_lowest(sc);
1065     ACPI_SERIAL_END(cpu);
1066 
1067     acpi_UserNotify("PROCESSOR", sc->cpu_handle, notify);
1068 }
1069 
1070 static int
1071 acpi_cpu_quirks(void)
1072 {
1073     device_t acpi_dev;
1074     uint32_t val;
1075 
1076     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
1077 
1078     /*
1079      * Bus mastering arbitration control is needed to keep caches coherent
1080      * while sleeping in C3.  If it's not present but a working flush cache
1081      * instruction is present, flush the caches before entering C3 instead.
1082      * Otherwise, just disable C3 completely.
1083      */
1084     if (AcpiGbl_FADT.Pm2ControlBlock == 0 ||
1085 	AcpiGbl_FADT.Pm2ControlLength == 0) {
1086 	if ((AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD) &&
1087 	    (AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD_FLUSH) == 0) {
1088 	    cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1089 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1090 		"acpi_cpu: no BM control, using flush cache method\n"));
1091 	} else {
1092 	    cpu_quirks |= CPU_QUIRK_NO_C3;
1093 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1094 		"acpi_cpu: no BM control, C3 not available\n"));
1095 	}
1096     }
1097 
1098     /*
1099      * If we are using generic Cx mode, C3 on multiple CPUs requires using
1100      * the expensive flush cache instruction.
1101      */
1102     if (cpu_cx_generic && mp_ncpus > 1) {
1103 	cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1104 	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1105 	    "acpi_cpu: SMP, using flush cache mode for C3\n"));
1106     }
1107 
1108     /* Look for various quirks of the PIIX4 part. */
1109     acpi_dev = pci_find_device(PCI_VENDOR_INTEL, PCI_DEVICE_82371AB_3);
1110     if (acpi_dev != NULL) {
1111 	switch (pci_get_revid(acpi_dev)) {
1112 	/*
1113 	 * Disable C3 support for all PIIX4 chipsets.  Some of these parts
1114 	 * do not report the BMIDE status to the BM status register and
1115 	 * others have a livelock bug if Type-F DMA is enabled.  Linux
1116 	 * works around the BMIDE bug by reading the BM status directly
1117 	 * but we take the simpler approach of disabling C3 for these
1118 	 * parts.
1119 	 *
1120 	 * See erratum #18 ("C3 Power State/BMIDE and Type-F DMA
1121 	 * Livelock") from the January 2002 PIIX4 specification update.
1122 	 * Applies to all PIIX4 models.
1123 	 *
1124 	 * Also, make sure that all interrupts cause a "Stop Break"
1125 	 * event to exit from C2 state.
1126 	 * Also, BRLD_EN_BM (ACPI_BITREG_BUS_MASTER_RLD in ACPI-speak)
1127 	 * should be set to zero, otherwise it causes C2 to short-sleep.
1128 	 * PIIX4 doesn't properly support C3 and bus master activity
1129 	 * need not break out of C2.
1130 	 */
1131 	case PCI_REVISION_A_STEP:
1132 	case PCI_REVISION_B_STEP:
1133 	case PCI_REVISION_4E:
1134 	case PCI_REVISION_4M:
1135 	    cpu_quirks |= CPU_QUIRK_NO_C3;
1136 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1137 		"acpi_cpu: working around PIIX4 bug, disabling C3\n"));
1138 
1139 	    val = pci_read_config(acpi_dev, PIIX4_DEVACTB_REG, 4);
1140 	    if ((val & PIIX4_STOP_BREAK_MASK) != PIIX4_STOP_BREAK_MASK) {
1141 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1142 		    "acpi_cpu: PIIX4: enabling IRQs to generate Stop Break\n"));
1143 	    	val |= PIIX4_STOP_BREAK_MASK;
1144 		pci_write_config(acpi_dev, PIIX4_DEVACTB_REG, val, 4);
1145 	    }
1146 	    AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_RLD, &val);
1147 	    if (val) {
1148 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1149 		    "acpi_cpu: PIIX4: reset BRLD_EN_BM\n"));
1150 		AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1151 	    }
1152 	    break;
1153 	default:
1154 	    break;
1155 	}
1156     }
1157 
1158     return (0);
1159 }
1160 
1161 static int
1162 acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS)
1163 {
1164     struct acpi_cpu_softc *sc;
1165     struct sbuf	 sb;
1166     char	 buf[128];
1167     int		 i;
1168     uintmax_t	 fract, sum, whole;
1169 
1170     sc = (struct acpi_cpu_softc *) arg1;
1171     sum = 0;
1172     for (i = 0; i < sc->cpu_cx_count; i++)
1173 	sum += sc->cpu_cx_stats[i];
1174     sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1175     for (i = 0; i < sc->cpu_cx_count; i++) {
1176 	if (sum > 0) {
1177 	    whole = (uintmax_t)sc->cpu_cx_stats[i] * 100;
1178 	    fract = (whole % sum) * 100;
1179 	    sbuf_printf(&sb, "%u.%02u%% ", (u_int)(whole / sum),
1180 		(u_int)(fract / sum));
1181 	} else
1182 	    sbuf_printf(&sb, "0.00%% ");
1183     }
1184     sbuf_printf(&sb, "last %dus", sc->cpu_prev_sleep);
1185     sbuf_trim(&sb);
1186     sbuf_finish(&sb);
1187     sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1188     sbuf_delete(&sb);
1189 
1190     return (0);
1191 }
1192 
1193 static int
1194 acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc)
1195 {
1196     int i;
1197 
1198     ACPI_SERIAL_ASSERT(cpu);
1199     sc->cpu_cx_lowest = min(sc->cpu_cx_lowest_lim, sc->cpu_cx_count - 1);
1200 
1201     /* If not disabling, cache the new lowest non-C3 state. */
1202     sc->cpu_non_c3 = 0;
1203     for (i = sc->cpu_cx_lowest; i >= 0; i--) {
1204 	if (sc->cpu_cx_states[i].type < ACPI_STATE_C3) {
1205 	    sc->cpu_non_c3 = i;
1206 	    break;
1207 	}
1208     }
1209 
1210     /* Reset the statistics counters. */
1211     bzero(sc->cpu_cx_stats, sizeof(sc->cpu_cx_stats));
1212     return (0);
1213 }
1214 
1215 static int
1216 acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1217 {
1218     struct	 acpi_cpu_softc *sc;
1219     char	 state[8];
1220     int		 val, error;
1221 
1222     sc = (struct acpi_cpu_softc *) arg1;
1223     snprintf(state, sizeof(state), "C%d", sc->cpu_cx_lowest_lim + 1);
1224     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1225     if (error != 0 || req->newptr == NULL)
1226 	return (error);
1227     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1228 	return (EINVAL);
1229     if (strcasecmp(state, "Cmax") == 0)
1230 	val = MAX_CX_STATES;
1231     else {
1232 	val = (int) strtol(state + 1, NULL, 10);
1233 	if (val < 1 || val > MAX_CX_STATES)
1234 	    return (EINVAL);
1235     }
1236 
1237     ACPI_SERIAL_BEGIN(cpu);
1238     sc->cpu_cx_lowest_lim = val - 1;
1239     acpi_cpu_set_cx_lowest(sc);
1240     ACPI_SERIAL_END(cpu);
1241 
1242     return (0);
1243 }
1244 
1245 static int
1246 acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1247 {
1248     struct	acpi_cpu_softc *sc;
1249     char	state[8];
1250     int		val, error, i;
1251 
1252     snprintf(state, sizeof(state), "C%d", cpu_cx_lowest_lim + 1);
1253     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1254     if (error != 0 || req->newptr == NULL)
1255 	return (error);
1256     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1257 	return (EINVAL);
1258     if (strcasecmp(state, "Cmax") == 0)
1259 	val = MAX_CX_STATES;
1260     else {
1261 	val = (int) strtol(state + 1, NULL, 10);
1262 	if (val < 1 || val > MAX_CX_STATES)
1263 	    return (EINVAL);
1264     }
1265 
1266     /* Update the new lowest useable Cx state for all CPUs. */
1267     ACPI_SERIAL_BEGIN(cpu);
1268     cpu_cx_lowest_lim = val - 1;
1269     for (i = 0; i < cpu_ndevices; i++) {
1270 	sc = device_get_softc(cpu_devices[i]);
1271 	sc->cpu_cx_lowest_lim = cpu_cx_lowest_lim;
1272 	acpi_cpu_set_cx_lowest(sc);
1273     }
1274     ACPI_SERIAL_END(cpu);
1275 
1276     return (0);
1277 }
1278