xref: /freebsd/sys/dev/acpica/acpi_cpu.c (revision 30b72b6871140f0b29c64d41fc85c4c1d4d4b3f4)
1 /*-
2  * Copyright (c) 2003-2005 Nate Lawson (SDG)
3  * Copyright (c) 2001 Michael Smith
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #include "opt_acpi.h"
32 #include <sys/param.h>
33 #include <sys/bus.h>
34 #include <sys/cpu.h>
35 #include <sys/kernel.h>
36 #include <sys/malloc.h>
37 #include <sys/module.h>
38 #include <sys/pcpu.h>
39 #include <sys/power.h>
40 #include <sys/proc.h>
41 #include <sys/sched.h>
42 #include <sys/sbuf.h>
43 #include <sys/smp.h>
44 
45 #include <dev/pci/pcivar.h>
46 #include <machine/atomic.h>
47 #include <machine/bus.h>
48 #if defined(__amd64__) || defined(__i386__)
49 #include <machine/clock.h>
50 #include <machine/specialreg.h>
51 #include <machine/md_var.h>
52 #endif
53 #include <sys/rman.h>
54 
55 #include <contrib/dev/acpica/include/acpi.h>
56 #include <contrib/dev/acpica/include/accommon.h>
57 
58 #include <dev/acpica/acpivar.h>
59 
60 /*
61  * Support for ACPI Processor devices, including C[1-3] sleep states.
62  */
63 
64 /* Hooks for the ACPI CA debugging infrastructure */
65 #define _COMPONENT	ACPI_PROCESSOR
66 ACPI_MODULE_NAME("PROCESSOR")
67 
68 struct acpi_cx {
69     struct resource	*p_lvlx;	/* Register to read to enter state. */
70     uint32_t		 type;		/* C1-3 (C4 and up treated as C3). */
71     uint32_t		 trans_lat;	/* Transition latency (usec). */
72     uint32_t		 power;		/* Power consumed (mW). */
73     int			 res_type;	/* Resource type for p_lvlx. */
74     int			 res_rid;	/* Resource ID for p_lvlx. */
75     bool		 do_mwait;
76     uint32_t		 mwait_hint;
77     bool		 mwait_hw_coord;
78     bool		 mwait_bm_avoidance;
79 };
80 #define MAX_CX_STATES	 8
81 
82 struct acpi_cpu_softc {
83     device_t		 cpu_dev;
84     ACPI_HANDLE		 cpu_handle;
85     struct pcpu		*cpu_pcpu;
86     uint32_t		 cpu_acpi_id;	/* ACPI processor id */
87     uint32_t		 cpu_p_blk;	/* ACPI P_BLK location */
88     uint32_t		 cpu_p_blk_len;	/* P_BLK length (must be 6). */
89     struct acpi_cx	 cpu_cx_states[MAX_CX_STATES];
90     int			 cpu_cx_count;	/* Number of valid Cx states. */
91     int			 cpu_prev_sleep;/* Last idle sleep duration. */
92     int			 cpu_features;	/* Child driver supported features. */
93     /* Runtime state. */
94     int			 cpu_non_c2;	/* Index of lowest non-C2 state. */
95     int			 cpu_non_c3;	/* Index of lowest non-C3 state. */
96     u_int		 cpu_cx_stats[MAX_CX_STATES];/* Cx usage history. */
97     /* Values for sysctl. */
98     struct sysctl_ctx_list cpu_sysctl_ctx;
99     struct sysctl_oid	*cpu_sysctl_tree;
100     int			 cpu_cx_lowest;
101     int			 cpu_cx_lowest_lim;
102     int			 cpu_disable_idle; /* Disable entry to idle function */
103     char 		 cpu_cx_supported[64];
104 };
105 
106 struct acpi_cpu_device {
107     struct resource_list	ad_rl;
108 };
109 
110 #define CPU_GET_REG(reg, width) 					\
111     (bus_space_read_ ## width(rman_get_bustag((reg)), 			\
112 		      rman_get_bushandle((reg)), 0))
113 #define CPU_SET_REG(reg, width, val)					\
114     (bus_space_write_ ## width(rman_get_bustag((reg)), 			\
115 		       rman_get_bushandle((reg)), 0, (val)))
116 
117 #define PM_USEC(x)	 ((x) >> 2)	/* ~4 clocks per usec (3.57955 Mhz) */
118 
119 #define ACPI_NOTIFY_CX_STATES	0x81	/* _CST changed. */
120 
121 #define CPU_QUIRK_NO_C3		(1<<0)	/* C3-type states are not usable. */
122 #define CPU_QUIRK_NO_BM_CTRL	(1<<2)	/* No bus mastering control. */
123 
124 #define PCI_VENDOR_INTEL	0x8086
125 #define PCI_DEVICE_82371AB_3	0x7113	/* PIIX4 chipset for quirks. */
126 #define PCI_REVISION_A_STEP	0
127 #define PCI_REVISION_B_STEP	1
128 #define PCI_REVISION_4E		2
129 #define PCI_REVISION_4M		3
130 #define PIIX4_DEVACTB_REG	0x58
131 #define PIIX4_BRLD_EN_IRQ0	(1<<0)
132 #define PIIX4_BRLD_EN_IRQ	(1<<1)
133 #define PIIX4_BRLD_EN_IRQ8	(1<<5)
134 #define PIIX4_STOP_BREAK_MASK	(PIIX4_BRLD_EN_IRQ0 | PIIX4_BRLD_EN_IRQ | PIIX4_BRLD_EN_IRQ8)
135 #define PIIX4_PCNTRL_BST_EN	(1<<10)
136 
137 #define	CST_FFH_VENDOR_INTEL	1
138 #define	CST_FFH_INTEL_CL_C1IO	1
139 #define	CST_FFH_INTEL_CL_MWAIT	2
140 #define	CST_FFH_MWAIT_HW_COORD	0x0001
141 #define	CST_FFH_MWAIT_BM_AVOID	0x0002
142 
143 /* Allow users to ignore processor orders in MADT. */
144 static int cpu_unordered;
145 SYSCTL_INT(_debug_acpi, OID_AUTO, cpu_unordered, CTLFLAG_RDTUN,
146     &cpu_unordered, 0,
147     "Do not use the MADT to match ACPI Processor objects to CPUs.");
148 
149 /* Knob to disable acpi_cpu devices */
150 bool acpi_cpu_disabled = false;
151 
152 /* Platform hardware resource information. */
153 static uint32_t		 cpu_smi_cmd;	/* Value to write to SMI_CMD. */
154 static uint8_t		 cpu_cst_cnt;	/* Indicate we are _CST aware. */
155 static int		 cpu_quirks;	/* Indicate any hardware bugs. */
156 
157 /* Values for sysctl. */
158 static struct sysctl_ctx_list cpu_sysctl_ctx;
159 static struct sysctl_oid *cpu_sysctl_tree;
160 static int		 cpu_cx_generic;
161 static int		 cpu_cx_lowest_lim;
162 
163 static device_t		*cpu_devices;
164 static int		 cpu_ndevices;
165 static struct acpi_cpu_softc **cpu_softc;
166 ACPI_SERIAL_DECL(cpu, "ACPI CPU");
167 
168 static int	acpi_cpu_probe(device_t dev);
169 static int	acpi_cpu_attach(device_t dev);
170 static int	acpi_cpu_suspend(device_t dev);
171 static int	acpi_cpu_resume(device_t dev);
172 static int	acpi_pcpu_get_id(device_t dev, uint32_t *acpi_id,
173 		    uint32_t *cpu_id);
174 static struct resource_list *acpi_cpu_get_rlist(device_t dev, device_t child);
175 static device_t	acpi_cpu_add_child(device_t dev, u_int order, const char *name,
176 		    int unit);
177 static int	acpi_cpu_read_ivar(device_t dev, device_t child, int index,
178 		    uintptr_t *result);
179 static int	acpi_cpu_shutdown(device_t dev);
180 static void	acpi_cpu_cx_probe(struct acpi_cpu_softc *sc);
181 static void	acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc);
182 static int	acpi_cpu_cx_cst(struct acpi_cpu_softc *sc);
183 static void	acpi_cpu_startup(void *arg);
184 static void	acpi_cpu_startup_cx(struct acpi_cpu_softc *sc);
185 static void	acpi_cpu_cx_list(struct acpi_cpu_softc *sc);
186 #if defined(__i386__) || defined(__amd64__)
187 static void	acpi_cpu_idle(sbintime_t sbt);
188 #endif
189 static void	acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context);
190 static void	acpi_cpu_quirks(void);
191 static void	acpi_cpu_quirks_piix4(void);
192 static int	acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS);
193 static int	acpi_cpu_usage_counters_sysctl(SYSCTL_HANDLER_ARGS);
194 static int	acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc);
195 static int	acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
196 static int	acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
197 #if defined(__i386__) || defined(__amd64__)
198 static int	acpi_cpu_method_sysctl(SYSCTL_HANDLER_ARGS);
199 #endif
200 
201 static device_method_t acpi_cpu_methods[] = {
202     /* Device interface */
203     DEVMETHOD(device_probe,	acpi_cpu_probe),
204     DEVMETHOD(device_attach,	acpi_cpu_attach),
205     DEVMETHOD(device_detach,	bus_generic_detach),
206     DEVMETHOD(device_shutdown,	acpi_cpu_shutdown),
207     DEVMETHOD(device_suspend,	acpi_cpu_suspend),
208     DEVMETHOD(device_resume,	acpi_cpu_resume),
209 
210     /* Bus interface */
211     DEVMETHOD(bus_add_child,	acpi_cpu_add_child),
212     DEVMETHOD(bus_read_ivar,	acpi_cpu_read_ivar),
213     DEVMETHOD(bus_get_resource_list, acpi_cpu_get_rlist),
214     DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
215     DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
216     DEVMETHOD(bus_alloc_resource, bus_generic_rl_alloc_resource),
217     DEVMETHOD(bus_release_resource, bus_generic_rl_release_resource),
218     DEVMETHOD(bus_activate_resource, bus_generic_activate_resource),
219     DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource),
220     DEVMETHOD(bus_setup_intr,	bus_generic_setup_intr),
221     DEVMETHOD(bus_teardown_intr, bus_generic_teardown_intr),
222 
223     DEVMETHOD_END
224 };
225 
226 static driver_t acpi_cpu_driver = {
227     "cpu",
228     acpi_cpu_methods,
229     sizeof(struct acpi_cpu_softc),
230 };
231 
232 static devclass_t acpi_cpu_devclass;
233 DRIVER_MODULE(cpu, acpi, acpi_cpu_driver, acpi_cpu_devclass, 0, 0);
234 MODULE_DEPEND(cpu, acpi, 1, 1, 1);
235 
236 static int
237 acpi_cpu_probe(device_t dev)
238 {
239     int			   acpi_id, cpu_id;
240     ACPI_BUFFER		   buf;
241     ACPI_HANDLE		   handle;
242     ACPI_OBJECT		   *obj;
243     ACPI_STATUS		   status;
244 
245     if (acpi_disabled("cpu") || acpi_get_type(dev) != ACPI_TYPE_PROCESSOR ||
246 	    acpi_cpu_disabled)
247 	return (ENXIO);
248 
249     handle = acpi_get_handle(dev);
250     if (cpu_softc == NULL)
251 	cpu_softc = malloc(sizeof(struct acpi_cpu_softc *) *
252 	    (mp_maxid + 1), M_TEMP /* XXX */, M_WAITOK | M_ZERO);
253 
254     /* Get our Processor object. */
255     buf.Pointer = NULL;
256     buf.Length = ACPI_ALLOCATE_BUFFER;
257     status = AcpiEvaluateObject(handle, NULL, NULL, &buf);
258     if (ACPI_FAILURE(status)) {
259 	device_printf(dev, "probe failed to get Processor obj - %s\n",
260 		      AcpiFormatException(status));
261 	return (ENXIO);
262     }
263     obj = (ACPI_OBJECT *)buf.Pointer;
264     if (obj->Type != ACPI_TYPE_PROCESSOR) {
265 	device_printf(dev, "Processor object has bad type %d\n", obj->Type);
266 	AcpiOsFree(obj);
267 	return (ENXIO);
268     }
269 
270     /*
271      * Find the processor associated with our unit.  We could use the
272      * ProcId as a key, however, some boxes do not have the same values
273      * in their Processor object as the ProcId values in the MADT.
274      */
275     acpi_id = obj->Processor.ProcId;
276     AcpiOsFree(obj);
277     if (acpi_pcpu_get_id(dev, &acpi_id, &cpu_id) != 0)
278 	return (ENXIO);
279 
280     /*
281      * Check if we already probed this processor.  We scan the bus twice
282      * so it's possible we've already seen this one.
283      */
284     if (cpu_softc[cpu_id] != NULL)
285 	return (ENXIO);
286 
287     /* Mark this processor as in-use and save our derived id for attach. */
288     cpu_softc[cpu_id] = (void *)1;
289     acpi_set_private(dev, (void*)(intptr_t)cpu_id);
290     device_set_desc(dev, "ACPI CPU");
291 
292     return (0);
293 }
294 
295 static int
296 acpi_cpu_attach(device_t dev)
297 {
298     ACPI_BUFFER		   buf;
299     ACPI_OBJECT		   arg, *obj;
300     ACPI_OBJECT_LIST	   arglist;
301     struct pcpu		   *pcpu_data;
302     struct acpi_cpu_softc *sc;
303     struct acpi_softc	  *acpi_sc;
304     ACPI_STATUS		   status;
305     u_int		   features;
306     int			   cpu_id, drv_count, i;
307     driver_t 		  **drivers;
308     uint32_t		   cap_set[3];
309 
310     /* UUID needed by _OSC evaluation */
311     static uint8_t cpu_oscuuid[16] = { 0x16, 0xA6, 0x77, 0x40, 0x0C, 0x29,
312 				       0xBE, 0x47, 0x9E, 0xBD, 0xD8, 0x70,
313 				       0x58, 0x71, 0x39, 0x53 };
314 
315     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
316 
317     sc = device_get_softc(dev);
318     sc->cpu_dev = dev;
319     sc->cpu_handle = acpi_get_handle(dev);
320     cpu_id = (int)(intptr_t)acpi_get_private(dev);
321     cpu_softc[cpu_id] = sc;
322     pcpu_data = pcpu_find(cpu_id);
323     pcpu_data->pc_device = dev;
324     sc->cpu_pcpu = pcpu_data;
325     cpu_smi_cmd = AcpiGbl_FADT.SmiCommand;
326     cpu_cst_cnt = AcpiGbl_FADT.CstControl;
327 
328     buf.Pointer = NULL;
329     buf.Length = ACPI_ALLOCATE_BUFFER;
330     status = AcpiEvaluateObject(sc->cpu_handle, NULL, NULL, &buf);
331     if (ACPI_FAILURE(status)) {
332 	device_printf(dev, "attach failed to get Processor obj - %s\n",
333 		      AcpiFormatException(status));
334 	return (ENXIO);
335     }
336     obj = (ACPI_OBJECT *)buf.Pointer;
337     sc->cpu_p_blk = obj->Processor.PblkAddress;
338     sc->cpu_p_blk_len = obj->Processor.PblkLength;
339     sc->cpu_acpi_id = obj->Processor.ProcId;
340     AcpiOsFree(obj);
341     ACPI_DEBUG_PRINT((ACPI_DB_INFO, "acpi_cpu%d: P_BLK at %#x/%d\n",
342 		     device_get_unit(dev), sc->cpu_p_blk, sc->cpu_p_blk_len));
343 
344     /*
345      * If this is the first cpu we attach, create and initialize the generic
346      * resources that will be used by all acpi cpu devices.
347      */
348     if (device_get_unit(dev) == 0) {
349 	/* Assume we won't be using generic Cx mode by default */
350 	cpu_cx_generic = FALSE;
351 
352 	/* Install hw.acpi.cpu sysctl tree */
353 	acpi_sc = acpi_device_get_parent_softc(dev);
354 	sysctl_ctx_init(&cpu_sysctl_ctx);
355 	cpu_sysctl_tree = SYSCTL_ADD_NODE(&cpu_sysctl_ctx,
356 	    SYSCTL_CHILDREN(acpi_sc->acpi_sysctl_tree), OID_AUTO, "cpu",
357 	    CTLFLAG_RD, 0, "node for CPU children");
358     }
359 
360     /*
361      * Before calling any CPU methods, collect child driver feature hints
362      * and notify ACPI of them.  We support unified SMP power control
363      * so advertise this ourselves.  Note this is not the same as independent
364      * SMP control where each CPU can have different settings.
365      */
366     sc->cpu_features = ACPI_CAP_SMP_SAME | ACPI_CAP_SMP_SAME_C3 |
367       ACPI_CAP_C1_IO_HALT;
368 
369 #if defined(__i386__) || defined(__amd64__)
370     /*
371      * Ask for MWAIT modes if not disabled and interrupts work
372      * reasonable with MWAIT.
373      */
374     if (!acpi_disabled("mwait") && cpu_mwait_usable())
375 	sc->cpu_features |= ACPI_CAP_SMP_C1_NATIVE | ACPI_CAP_SMP_C3_NATIVE;
376 #endif
377 
378     if (devclass_get_drivers(acpi_cpu_devclass, &drivers, &drv_count) == 0) {
379 	for (i = 0; i < drv_count; i++) {
380 	    if (ACPI_GET_FEATURES(drivers[i], &features) == 0)
381 		sc->cpu_features |= features;
382 	}
383 	free(drivers, M_TEMP);
384     }
385 
386     /*
387      * CPU capabilities are specified in
388      * Intel Processor Vendor-Specific ACPI Interface Specification.
389      */
390     if (sc->cpu_features) {
391 	cap_set[1] = sc->cpu_features;
392 	status = acpi_EvaluateOSC(sc->cpu_handle, cpu_oscuuid, 1, 2, cap_set,
393 	    cap_set, false);
394 	if (ACPI_SUCCESS(status)) {
395 	    if (cap_set[0] != 0)
396 		device_printf(dev, "_OSC returned status %#x\n", cap_set[0]);
397 	}
398 	else {
399 	    arglist.Pointer = &arg;
400 	    arglist.Count = 1;
401 	    arg.Type = ACPI_TYPE_BUFFER;
402 	    arg.Buffer.Length = sizeof(cap_set);
403 	    arg.Buffer.Pointer = (uint8_t *)cap_set;
404 	    cap_set[0] = 1; /* revision */
405 	    cap_set[1] = 1; /* number of capabilities integers */
406 	    cap_set[2] = sc->cpu_features;
407 	    AcpiEvaluateObject(sc->cpu_handle, "_PDC", &arglist, NULL);
408 	}
409     }
410 
411     /* Probe for Cx state support. */
412     acpi_cpu_cx_probe(sc);
413 
414     return (0);
415 }
416 
417 static void
418 acpi_cpu_postattach(void *unused __unused)
419 {
420     device_t *devices;
421     int err;
422     int i, n;
423     int attached;
424 
425     err = devclass_get_devices(acpi_cpu_devclass, &devices, &n);
426     if (err != 0) {
427 	printf("devclass_get_devices(acpi_cpu_devclass) failed\n");
428 	return;
429     }
430     attached = 0;
431     for (i = 0; i < n; i++)
432 	if (device_is_attached(devices[i]))
433 	    attached = 1;
434     for (i = 0; i < n; i++)
435 	bus_generic_probe(devices[i]);
436     for (i = 0; i < n; i++)
437 	bus_generic_attach(devices[i]);
438     free(devices, M_TEMP);
439 
440     if (attached) {
441 	/* Queue post cpu-probing task handler */
442 	AcpiOsExecute(OSL_NOTIFY_HANDLER, acpi_cpu_startup, NULL);
443     }
444 }
445 
446 SYSINIT(acpi_cpu, SI_SUB_CONFIGURE, SI_ORDER_MIDDLE,
447     acpi_cpu_postattach, NULL);
448 
449 static void
450 disable_idle(struct acpi_cpu_softc *sc)
451 {
452     cpuset_t cpuset;
453 
454     CPU_SETOF(sc->cpu_pcpu->pc_cpuid, &cpuset);
455     sc->cpu_disable_idle = TRUE;
456 
457     /*
458      * Ensure that the CPU is not in idle state or in acpi_cpu_idle().
459      * Note that this code depends on the fact that the rendezvous IPI
460      * can not penetrate context where interrupts are disabled and acpi_cpu_idle
461      * is called and executed in such a context with interrupts being re-enabled
462      * right before return.
463      */
464     smp_rendezvous_cpus(cpuset, smp_no_rendevous_barrier, NULL,
465 	smp_no_rendevous_barrier, NULL);
466 }
467 
468 static void
469 enable_idle(struct acpi_cpu_softc *sc)
470 {
471 
472     sc->cpu_disable_idle = FALSE;
473 }
474 
475 #if defined(__i386__) || defined(__amd64__)
476 static int
477 is_idle_disabled(struct acpi_cpu_softc *sc)
478 {
479 
480     return (sc->cpu_disable_idle);
481 }
482 #endif
483 
484 /*
485  * Disable any entry to the idle function during suspend and re-enable it
486  * during resume.
487  */
488 static int
489 acpi_cpu_suspend(device_t dev)
490 {
491     int error;
492 
493     error = bus_generic_suspend(dev);
494     if (error)
495 	return (error);
496     disable_idle(device_get_softc(dev));
497     return (0);
498 }
499 
500 static int
501 acpi_cpu_resume(device_t dev)
502 {
503 
504     enable_idle(device_get_softc(dev));
505     return (bus_generic_resume(dev));
506 }
507 
508 /*
509  * Find the processor associated with a given ACPI ID.  By default,
510  * use the MADT to map ACPI IDs to APIC IDs and use that to locate a
511  * processor.  Some systems have inconsistent ASL and MADT however.
512  * For these systems the cpu_unordered tunable can be set in which
513  * case we assume that Processor objects are listed in the same order
514  * in both the MADT and ASL.
515  */
516 static int
517 acpi_pcpu_get_id(device_t dev, uint32_t *acpi_id, uint32_t *cpu_id)
518 {
519     struct pcpu	*pc;
520     uint32_t	 i, idx;
521 
522     KASSERT(acpi_id != NULL, ("Null acpi_id"));
523     KASSERT(cpu_id != NULL, ("Null cpu_id"));
524     idx = device_get_unit(dev);
525 
526     /*
527      * If pc_acpi_id for CPU 0 is not initialized (e.g. a non-APIC
528      * UP box) use the ACPI ID from the first processor we find.
529      */
530     if (idx == 0 && mp_ncpus == 1) {
531 	pc = pcpu_find(0);
532 	if (pc->pc_acpi_id == 0xffffffff)
533 	    pc->pc_acpi_id = *acpi_id;
534 	*cpu_id = 0;
535 	return (0);
536     }
537 
538     CPU_FOREACH(i) {
539 	pc = pcpu_find(i);
540 	KASSERT(pc != NULL, ("no pcpu data for %d", i));
541 	if (cpu_unordered) {
542 	    if (idx-- == 0) {
543 		/*
544 		 * If pc_acpi_id doesn't match the ACPI ID from the
545 		 * ASL, prefer the MADT-derived value.
546 		 */
547 		if (pc->pc_acpi_id != *acpi_id)
548 		    *acpi_id = pc->pc_acpi_id;
549 		*cpu_id = pc->pc_cpuid;
550 		return (0);
551 	    }
552 	} else {
553 	    if (pc->pc_acpi_id == *acpi_id) {
554 		if (bootverbose)
555 		    device_printf(dev,
556 			"Processor %s (ACPI ID %u) -> APIC ID %d\n",
557 			acpi_name(acpi_get_handle(dev)), *acpi_id,
558 			pc->pc_cpuid);
559 		*cpu_id = pc->pc_cpuid;
560 		return (0);
561 	    }
562 	}
563     }
564 
565     if (bootverbose)
566 	printf("ACPI: Processor %s (ACPI ID %u) ignored\n",
567 	    acpi_name(acpi_get_handle(dev)), *acpi_id);
568 
569     return (ESRCH);
570 }
571 
572 static struct resource_list *
573 acpi_cpu_get_rlist(device_t dev, device_t child)
574 {
575     struct acpi_cpu_device *ad;
576 
577     ad = device_get_ivars(child);
578     if (ad == NULL)
579 	return (NULL);
580     return (&ad->ad_rl);
581 }
582 
583 static device_t
584 acpi_cpu_add_child(device_t dev, u_int order, const char *name, int unit)
585 {
586     struct acpi_cpu_device *ad;
587     device_t child;
588 
589     if ((ad = malloc(sizeof(*ad), M_TEMP, M_NOWAIT | M_ZERO)) == NULL)
590 	return (NULL);
591 
592     resource_list_init(&ad->ad_rl);
593 
594     child = device_add_child_ordered(dev, order, name, unit);
595     if (child != NULL)
596 	device_set_ivars(child, ad);
597     else
598 	free(ad, M_TEMP);
599     return (child);
600 }
601 
602 static int
603 acpi_cpu_read_ivar(device_t dev, device_t child, int index, uintptr_t *result)
604 {
605     struct acpi_cpu_softc *sc;
606 
607     sc = device_get_softc(dev);
608     switch (index) {
609     case ACPI_IVAR_HANDLE:
610 	*result = (uintptr_t)sc->cpu_handle;
611 	break;
612     case CPU_IVAR_PCPU:
613 	*result = (uintptr_t)sc->cpu_pcpu;
614 	break;
615 #if defined(__amd64__) || defined(__i386__)
616     case CPU_IVAR_NOMINAL_MHZ:
617 	if (tsc_is_invariant) {
618 	    *result = (uintptr_t)(atomic_load_acq_64(&tsc_freq) / 1000000);
619 	    break;
620 	}
621 	/* FALLTHROUGH */
622 #endif
623     default:
624 	return (ENOENT);
625     }
626     return (0);
627 }
628 
629 static int
630 acpi_cpu_shutdown(device_t dev)
631 {
632     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
633 
634     /* Allow children to shutdown first. */
635     bus_generic_shutdown(dev);
636 
637     /*
638      * Disable any entry to the idle function.
639      */
640     disable_idle(device_get_softc(dev));
641 
642     /*
643      * CPU devices are not truely detached and remain referenced,
644      * so their resources are not freed.
645      */
646 
647     return_VALUE (0);
648 }
649 
650 static void
651 acpi_cpu_cx_probe(struct acpi_cpu_softc *sc)
652 {
653     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
654 
655     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
656     sc->cpu_prev_sleep = 1000000;
657     sc->cpu_cx_lowest = 0;
658     sc->cpu_cx_lowest_lim = 0;
659 
660     /*
661      * Check for the ACPI 2.0 _CST sleep states object. If we can't find
662      * any, we'll revert to generic FADT/P_BLK Cx control method which will
663      * be handled by acpi_cpu_startup. We need to defer to after having
664      * probed all the cpus in the system before probing for generic Cx
665      * states as we may already have found cpus with valid _CST packages
666      */
667     if (!cpu_cx_generic && acpi_cpu_cx_cst(sc) != 0) {
668 	/*
669 	 * We were unable to find a _CST package for this cpu or there
670 	 * was an error parsing it. Switch back to generic mode.
671 	 */
672 	cpu_cx_generic = TRUE;
673 	if (bootverbose)
674 	    device_printf(sc->cpu_dev, "switching to generic Cx mode\n");
675     }
676 
677     /*
678      * TODO: _CSD Package should be checked here.
679      */
680 }
681 
682 static void
683 acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc)
684 {
685     ACPI_GENERIC_ADDRESS	 gas;
686     struct acpi_cx		*cx_ptr;
687 
688     sc->cpu_cx_count = 0;
689     cx_ptr = sc->cpu_cx_states;
690 
691     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
692     sc->cpu_prev_sleep = 1000000;
693 
694     /* C1 has been required since just after ACPI 1.0 */
695     cx_ptr->type = ACPI_STATE_C1;
696     cx_ptr->trans_lat = 0;
697     cx_ptr++;
698     sc->cpu_non_c2 = sc->cpu_cx_count;
699     sc->cpu_non_c3 = sc->cpu_cx_count;
700     sc->cpu_cx_count++;
701     cpu_deepest_sleep = 1;
702 
703     /*
704      * The spec says P_BLK must be 6 bytes long.  However, some systems
705      * use it to indicate a fractional set of features present so we
706      * take 5 as C2.  Some may also have a value of 7 to indicate
707      * another C3 but most use _CST for this (as required) and having
708      * "only" C1-C3 is not a hardship.
709      */
710     if (sc->cpu_p_blk_len < 5)
711 	return;
712 
713     /* Validate and allocate resources for C2 (P_LVL2). */
714     gas.SpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
715     gas.BitWidth = 8;
716     if (AcpiGbl_FADT.C2Latency <= 100) {
717 	gas.Address = sc->cpu_p_blk + 4;
718 	cx_ptr->res_rid = 0;
719 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &cx_ptr->res_rid,
720 	    &gas, &cx_ptr->p_lvlx, RF_SHAREABLE);
721 	if (cx_ptr->p_lvlx != NULL) {
722 	    cx_ptr->type = ACPI_STATE_C2;
723 	    cx_ptr->trans_lat = AcpiGbl_FADT.C2Latency;
724 	    cx_ptr++;
725 	    sc->cpu_non_c3 = sc->cpu_cx_count;
726 	    sc->cpu_cx_count++;
727 	    cpu_deepest_sleep = 2;
728 	}
729     }
730     if (sc->cpu_p_blk_len < 6)
731 	return;
732 
733     /* Validate and allocate resources for C3 (P_LVL3). */
734     if (AcpiGbl_FADT.C3Latency <= 1000 && !(cpu_quirks & CPU_QUIRK_NO_C3)) {
735 	gas.Address = sc->cpu_p_blk + 5;
736 	cx_ptr->res_rid = 1;
737 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &cx_ptr->res_rid,
738 	    &gas, &cx_ptr->p_lvlx, RF_SHAREABLE);
739 	if (cx_ptr->p_lvlx != NULL) {
740 	    cx_ptr->type = ACPI_STATE_C3;
741 	    cx_ptr->trans_lat = AcpiGbl_FADT.C3Latency;
742 	    cx_ptr++;
743 	    sc->cpu_cx_count++;
744 	    cpu_deepest_sleep = 3;
745 	}
746     }
747 }
748 
749 #if defined(__i386__) || defined(__amd64__)
750 static void
751 acpi_cpu_cx_cst_mwait(struct acpi_cx *cx_ptr, uint64_t address, int accsize)
752 {
753 
754 	cx_ptr->do_mwait = true;
755 	cx_ptr->mwait_hint = address & 0xffffffff;
756 	cx_ptr->mwait_hw_coord = (accsize & CST_FFH_MWAIT_HW_COORD) != 0;
757 	cx_ptr->mwait_bm_avoidance = (accsize & CST_FFH_MWAIT_BM_AVOID) != 0;
758 }
759 #endif
760 
761 static void
762 acpi_cpu_cx_cst_free_plvlx(device_t cpu_dev, struct acpi_cx *cx_ptr)
763 {
764 
765 	if (cx_ptr->p_lvlx == NULL)
766 		return;
767 	bus_release_resource(cpu_dev, cx_ptr->res_type, cx_ptr->res_rid,
768 	    cx_ptr->p_lvlx);
769 	cx_ptr->p_lvlx = NULL;
770 }
771 
772 /*
773  * Parse a _CST package and set up its Cx states.  Since the _CST object
774  * can change dynamically, our notify handler may call this function
775  * to clean up and probe the new _CST package.
776  */
777 static int
778 acpi_cpu_cx_cst(struct acpi_cpu_softc *sc)
779 {
780     struct	 acpi_cx *cx_ptr;
781     ACPI_STATUS	 status;
782     ACPI_BUFFER	 buf;
783     ACPI_OBJECT	*top;
784     ACPI_OBJECT	*pkg;
785     uint32_t	 count;
786     int		 i;
787 #if defined(__i386__) || defined(__amd64__)
788     uint64_t	 address;
789     int		 vendor, class, accsize;
790 #endif
791 
792     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
793 
794     buf.Pointer = NULL;
795     buf.Length = ACPI_ALLOCATE_BUFFER;
796     status = AcpiEvaluateObject(sc->cpu_handle, "_CST", NULL, &buf);
797     if (ACPI_FAILURE(status))
798 	return (ENXIO);
799 
800     /* _CST is a package with a count and at least one Cx package. */
801     top = (ACPI_OBJECT *)buf.Pointer;
802     if (!ACPI_PKG_VALID(top, 2) || acpi_PkgInt32(top, 0, &count) != 0) {
803 	device_printf(sc->cpu_dev, "invalid _CST package\n");
804 	AcpiOsFree(buf.Pointer);
805 	return (ENXIO);
806     }
807     if (count != top->Package.Count - 1) {
808 	device_printf(sc->cpu_dev, "invalid _CST state count (%d != %d)\n",
809 	       count, top->Package.Count - 1);
810 	count = top->Package.Count - 1;
811     }
812     if (count > MAX_CX_STATES) {
813 	device_printf(sc->cpu_dev, "_CST has too many states (%d)\n", count);
814 	count = MAX_CX_STATES;
815     }
816 
817     sc->cpu_non_c2 = 0;
818     sc->cpu_non_c3 = 0;
819     sc->cpu_cx_count = 0;
820     cx_ptr = sc->cpu_cx_states;
821 
822     /*
823      * C1 has been required since just after ACPI 1.0.
824      * Reserve the first slot for it.
825      */
826     cx_ptr->type = ACPI_STATE_C0;
827     cx_ptr++;
828     sc->cpu_cx_count++;
829     cpu_deepest_sleep = 1;
830 
831     /* Set up all valid states. */
832     for (i = 0; i < count; i++) {
833 	pkg = &top->Package.Elements[i + 1];
834 	if (!ACPI_PKG_VALID(pkg, 4) ||
835 	    acpi_PkgInt32(pkg, 1, &cx_ptr->type) != 0 ||
836 	    acpi_PkgInt32(pkg, 2, &cx_ptr->trans_lat) != 0 ||
837 	    acpi_PkgInt32(pkg, 3, &cx_ptr->power) != 0) {
838 
839 	    device_printf(sc->cpu_dev, "skipping invalid Cx state package\n");
840 	    continue;
841 	}
842 
843 	/* Validate the state to see if we should use it. */
844 	switch (cx_ptr->type) {
845 	case ACPI_STATE_C1:
846 	    acpi_cpu_cx_cst_free_plvlx(sc->cpu_dev, cx_ptr);
847 #if defined(__i386__) || defined(__amd64__)
848 	    if (acpi_PkgFFH_IntelCpu(pkg, 0, &vendor, &class, &address,
849 	      &accsize) == 0 && vendor == CST_FFH_VENDOR_INTEL) {
850 		if (class == CST_FFH_INTEL_CL_C1IO) {
851 		    /* C1 I/O then Halt */
852 		    cx_ptr->res_rid = sc->cpu_cx_count;
853 		    bus_set_resource(sc->cpu_dev, SYS_RES_IOPORT,
854 		      cx_ptr->res_rid, address, 1);
855 		    cx_ptr->p_lvlx = bus_alloc_resource_any(sc->cpu_dev,
856 		      SYS_RES_IOPORT, &cx_ptr->res_rid, RF_ACTIVE |
857 		      RF_SHAREABLE);
858 		    if (cx_ptr->p_lvlx == NULL) {
859 			bus_delete_resource(sc->cpu_dev, SYS_RES_IOPORT,
860 			  cx_ptr->res_rid);
861 			device_printf(sc->cpu_dev,
862 			  "C1 I/O failed to allocate port %d, "
863 			  "degrading to C1 Halt", (int)address);
864 		    }
865 		} else if (class == CST_FFH_INTEL_CL_MWAIT) {
866 		    acpi_cpu_cx_cst_mwait(cx_ptr, address, accsize);
867 		}
868 	    }
869 #endif
870 	    if (sc->cpu_cx_states[0].type == ACPI_STATE_C0) {
871 		/* This is the first C1 state.  Use the reserved slot. */
872 		sc->cpu_cx_states[0] = *cx_ptr;
873 	    } else {
874 		sc->cpu_non_c2 = sc->cpu_cx_count;
875 		sc->cpu_non_c3 = sc->cpu_cx_count;
876 		cx_ptr++;
877 		sc->cpu_cx_count++;
878 	    }
879 	    continue;
880 	case ACPI_STATE_C2:
881 	    sc->cpu_non_c3 = sc->cpu_cx_count;
882 	    if (cpu_deepest_sleep < 2)
883 		    cpu_deepest_sleep = 2;
884 	    break;
885 	case ACPI_STATE_C3:
886 	default:
887 	    if ((cpu_quirks & CPU_QUIRK_NO_C3) != 0) {
888 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
889 				 "acpi_cpu%d: C3[%d] not available.\n",
890 				 device_get_unit(sc->cpu_dev), i));
891 		continue;
892 	    } else
893 		cpu_deepest_sleep = 3;
894 	    break;
895 	}
896 
897 	/* Free up any previous register. */
898 	acpi_cpu_cx_cst_free_plvlx(sc->cpu_dev, cx_ptr);
899 
900 	/* Allocate the control register for C2 or C3. */
901 #if defined(__i386__) || defined(__amd64__)
902 	if (acpi_PkgFFH_IntelCpu(pkg, 0, &vendor, &class, &address,
903 	  &accsize) == 0 && vendor == CST_FFH_VENDOR_INTEL &&
904 	  class == CST_FFH_INTEL_CL_MWAIT) {
905 	    /* Native C State Instruction use (mwait) */
906 	    acpi_cpu_cx_cst_mwait(cx_ptr, address, accsize);
907 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
908 	      "acpi_cpu%d: Got C%d/mwait - %d latency\n",
909 	      device_get_unit(sc->cpu_dev), cx_ptr->type, cx_ptr->trans_lat));
910 	    cx_ptr++;
911 	    sc->cpu_cx_count++;
912 	} else
913 #endif
914 	{
915 	    cx_ptr->res_rid = sc->cpu_cx_count;
916 	    acpi_PkgGas(sc->cpu_dev, pkg, 0, &cx_ptr->res_type,
917 		&cx_ptr->res_rid, &cx_ptr->p_lvlx, RF_SHAREABLE);
918 	    if (cx_ptr->p_lvlx) {
919 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
920 		     "acpi_cpu%d: Got C%d - %d latency\n",
921 		     device_get_unit(sc->cpu_dev), cx_ptr->type,
922 		     cx_ptr->trans_lat));
923 		cx_ptr++;
924 		sc->cpu_cx_count++;
925 	    }
926 	}
927     }
928     AcpiOsFree(buf.Pointer);
929 
930     /* If C1 state was not found, we need one now. */
931     cx_ptr = sc->cpu_cx_states;
932     if (cx_ptr->type == ACPI_STATE_C0) {
933 	cx_ptr->type = ACPI_STATE_C1;
934 	cx_ptr->trans_lat = 0;
935     }
936 
937     return (0);
938 }
939 
940 /*
941  * Call this *after* all CPUs have been attached.
942  */
943 static void
944 acpi_cpu_startup(void *arg)
945 {
946     struct acpi_cpu_softc *sc;
947     int i;
948 
949     /* Get set of CPU devices */
950     devclass_get_devices(acpi_cpu_devclass, &cpu_devices, &cpu_ndevices);
951 
952     /*
953      * Setup any quirks that might necessary now that we have probed
954      * all the CPUs
955      */
956     acpi_cpu_quirks();
957 
958     if (cpu_cx_generic) {
959 	/*
960 	 * We are using generic Cx mode, probe for available Cx states
961 	 * for all processors.
962 	 */
963 	for (i = 0; i < cpu_ndevices; i++) {
964 	    sc = device_get_softc(cpu_devices[i]);
965 	    acpi_cpu_generic_cx_probe(sc);
966 	}
967     } else {
968 	/*
969 	 * We are using _CST mode, remove C3 state if necessary.
970 	 * As we now know for sure that we will be using _CST mode
971 	 * install our notify handler.
972 	 */
973 	for (i = 0; i < cpu_ndevices; i++) {
974 	    sc = device_get_softc(cpu_devices[i]);
975 	    if (cpu_quirks & CPU_QUIRK_NO_C3) {
976 		sc->cpu_cx_count = min(sc->cpu_cx_count, sc->cpu_non_c3 + 1);
977 	    }
978 	    AcpiInstallNotifyHandler(sc->cpu_handle, ACPI_DEVICE_NOTIFY,
979 		acpi_cpu_notify, sc);
980 	}
981     }
982 
983     /* Perform Cx final initialization. */
984     for (i = 0; i < cpu_ndevices; i++) {
985 	sc = device_get_softc(cpu_devices[i]);
986 	acpi_cpu_startup_cx(sc);
987     }
988 
989     /* Add a sysctl handler to handle global Cx lowest setting */
990     SYSCTL_ADD_PROC(&cpu_sysctl_ctx, SYSCTL_CHILDREN(cpu_sysctl_tree),
991 	OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
992 	NULL, 0, acpi_cpu_global_cx_lowest_sysctl, "A",
993 	"Global lowest Cx sleep state to use");
994 
995     /* Take over idling from cpu_idle_default(). */
996     cpu_cx_lowest_lim = 0;
997     for (i = 0; i < cpu_ndevices; i++) {
998 	sc = device_get_softc(cpu_devices[i]);
999 	enable_idle(sc);
1000     }
1001 #if defined(__i386__) || defined(__amd64__)
1002     cpu_idle_hook = acpi_cpu_idle;
1003 #endif
1004 }
1005 
1006 static void
1007 acpi_cpu_cx_list(struct acpi_cpu_softc *sc)
1008 {
1009     struct sbuf sb;
1010     int i;
1011 
1012     /*
1013      * Set up the list of Cx states
1014      */
1015     sbuf_new(&sb, sc->cpu_cx_supported, sizeof(sc->cpu_cx_supported),
1016 	SBUF_FIXEDLEN);
1017     for (i = 0; i < sc->cpu_cx_count; i++)
1018 	sbuf_printf(&sb, "C%d/%d/%d ", i + 1, sc->cpu_cx_states[i].type,
1019 	    sc->cpu_cx_states[i].trans_lat);
1020     sbuf_trim(&sb);
1021     sbuf_finish(&sb);
1022 }
1023 
1024 static void
1025 acpi_cpu_startup_cx(struct acpi_cpu_softc *sc)
1026 {
1027     acpi_cpu_cx_list(sc);
1028 
1029     SYSCTL_ADD_STRING(&sc->cpu_sysctl_ctx,
1030 		      SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
1031 		      OID_AUTO, "cx_supported", CTLFLAG_RD,
1032 		      sc->cpu_cx_supported, 0,
1033 		      "Cx/microsecond values for supported Cx states");
1034     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
1035 		    SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
1036 		    OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
1037 		    (void *)sc, 0, acpi_cpu_cx_lowest_sysctl, "A",
1038 		    "lowest Cx sleep state to use");
1039     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
1040 		    SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
1041 		    OID_AUTO, "cx_usage", CTLTYPE_STRING | CTLFLAG_RD,
1042 		    (void *)sc, 0, acpi_cpu_usage_sysctl, "A",
1043 		    "percent usage for each Cx state");
1044     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
1045 		    SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
1046 		    OID_AUTO, "cx_usage_counters", CTLTYPE_STRING | CTLFLAG_RD,
1047 		    (void *)sc, 0, acpi_cpu_usage_counters_sysctl, "A",
1048 		    "Cx sleep state counters");
1049 #if defined(__i386__) || defined(__amd64__)
1050     SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
1051 		    SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
1052 		    OID_AUTO, "cx_method", CTLTYPE_STRING | CTLFLAG_RD,
1053 		    (void *)sc, 0, acpi_cpu_method_sysctl, "A",
1054 		    "Cx entrance methods");
1055 #endif
1056 
1057     /* Signal platform that we can handle _CST notification. */
1058     if (!cpu_cx_generic && cpu_cst_cnt != 0) {
1059 	ACPI_LOCK(acpi);
1060 	AcpiOsWritePort(cpu_smi_cmd, cpu_cst_cnt, 8);
1061 	ACPI_UNLOCK(acpi);
1062     }
1063 }
1064 
1065 #if defined(__i386__) || defined(__amd64__)
1066 /*
1067  * Idle the CPU in the lowest state possible.  This function is called with
1068  * interrupts disabled.  Note that once it re-enables interrupts, a task
1069  * switch can occur so do not access shared data (i.e. the softc) after
1070  * interrupts are re-enabled.
1071  */
1072 static void
1073 acpi_cpu_idle(sbintime_t sbt)
1074 {
1075     struct	acpi_cpu_softc *sc;
1076     struct	acpi_cx *cx_next;
1077     uint64_t	cputicks;
1078     uint32_t	start_time, end_time;
1079     ACPI_STATUS	status;
1080     int		bm_active, cx_next_idx, i, us;
1081 
1082     /*
1083      * Look up our CPU id to get our softc.  If it's NULL, we'll use C1
1084      * since there is no ACPI processor object for this CPU.  This occurs
1085      * for logical CPUs in the HTT case.
1086      */
1087     sc = cpu_softc[PCPU_GET(cpuid)];
1088     if (sc == NULL) {
1089 	acpi_cpu_c1();
1090 	return;
1091     }
1092 
1093     /* If disabled, take the safe path. */
1094     if (is_idle_disabled(sc)) {
1095 	acpi_cpu_c1();
1096 	return;
1097     }
1098 
1099     /* Find the lowest state that has small enough latency. */
1100     us = sc->cpu_prev_sleep;
1101     if (sbt >= 0 && us > (sbt >> 12))
1102 	us = (sbt >> 12);
1103     cx_next_idx = 0;
1104     if (cpu_disable_c2_sleep)
1105 	i = min(sc->cpu_cx_lowest, sc->cpu_non_c2);
1106     else if (cpu_disable_c3_sleep)
1107 	i = min(sc->cpu_cx_lowest, sc->cpu_non_c3);
1108     else
1109 	i = sc->cpu_cx_lowest;
1110     for (; i >= 0; i--) {
1111 	if (sc->cpu_cx_states[i].trans_lat * 3 <= us) {
1112 	    cx_next_idx = i;
1113 	    break;
1114 	}
1115     }
1116 
1117     /*
1118      * Check for bus master activity.  If there was activity, clear
1119      * the bit and use the lowest non-C3 state.  Note that the USB
1120      * driver polling for new devices keeps this bit set all the
1121      * time if USB is loaded.
1122      */
1123     if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0 &&
1124 	cx_next_idx > sc->cpu_non_c3) {
1125 	status = AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, &bm_active);
1126 	if (ACPI_SUCCESS(status) && bm_active != 0) {
1127 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, 1);
1128 	    cx_next_idx = sc->cpu_non_c3;
1129 	}
1130     }
1131 
1132     /* Select the next state and update statistics. */
1133     cx_next = &sc->cpu_cx_states[cx_next_idx];
1134     sc->cpu_cx_stats[cx_next_idx]++;
1135     KASSERT(cx_next->type != ACPI_STATE_C0, ("acpi_cpu_idle: C0 sleep"));
1136 
1137     /*
1138      * Execute HLT (or equivalent) and wait for an interrupt.  We can't
1139      * precisely calculate the time spent in C1 since the place we wake up
1140      * is an ISR.  Assume we slept no more then half of quantum, unless
1141      * we are called inside critical section, delaying context switch.
1142      */
1143     if (cx_next->type == ACPI_STATE_C1) {
1144 	cputicks = cpu_ticks();
1145 	if (cx_next->p_lvlx != NULL) {
1146 	    /* C1 I/O then Halt */
1147 	    CPU_GET_REG(cx_next->p_lvlx, 1);
1148 	}
1149 	if (cx_next->do_mwait)
1150 	    acpi_cpu_idle_mwait(cx_next->mwait_hint);
1151 	else
1152 	    acpi_cpu_c1();
1153 	end_time = ((cpu_ticks() - cputicks) << 20) / cpu_tickrate();
1154 	if (curthread->td_critnest == 0)
1155 		end_time = min(end_time, 500000 / hz);
1156 	sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + end_time) / 4;
1157 	return;
1158     }
1159 
1160     /*
1161      * For C3, disable bus master arbitration and enable bus master wake
1162      * if BM control is available, otherwise flush the CPU cache.
1163      */
1164     if (cx_next->type == ACPI_STATE_C3 || cx_next->mwait_bm_avoidance) {
1165 	if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
1166 	    AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 1);
1167 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
1168 	} else
1169 	    ACPI_FLUSH_CPU_CACHE();
1170     }
1171 
1172     /*
1173      * Read from P_LVLx to enter C2(+), checking time spent asleep.
1174      * Use the ACPI timer for measuring sleep time.  Since we need to
1175      * get the time very close to the CPU start/stop clock logic, this
1176      * is the only reliable time source.
1177      */
1178     if (cx_next->type == ACPI_STATE_C3) {
1179 	AcpiHwRead(&start_time, &AcpiGbl_FADT.XPmTimerBlock);
1180 	cputicks = 0;
1181     } else {
1182 	start_time = 0;
1183 	cputicks = cpu_ticks();
1184     }
1185     if (cx_next->do_mwait)
1186 	acpi_cpu_idle_mwait(cx_next->mwait_hint);
1187     else
1188 	CPU_GET_REG(cx_next->p_lvlx, 1);
1189 
1190     /*
1191      * Read the end time twice.  Since it may take an arbitrary time
1192      * to enter the idle state, the first read may be executed before
1193      * the processor has stopped.  Doing it again provides enough
1194      * margin that we are certain to have a correct value.
1195      */
1196     AcpiHwRead(&end_time, &AcpiGbl_FADT.XPmTimerBlock);
1197     if (cx_next->type == ACPI_STATE_C3) {
1198 	AcpiHwRead(&end_time, &AcpiGbl_FADT.XPmTimerBlock);
1199 	end_time = acpi_TimerDelta(end_time, start_time);
1200     } else
1201 	end_time = ((cpu_ticks() - cputicks) << 20) / cpu_tickrate();
1202 
1203     /* Enable bus master arbitration and disable bus master wakeup. */
1204     if ((cx_next->type == ACPI_STATE_C3 || cx_next->mwait_bm_avoidance) &&
1205       (cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
1206 	AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 0);
1207 	AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1208     }
1209     ACPI_ENABLE_IRQS();
1210 
1211     sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + PM_USEC(end_time)) / 4;
1212 }
1213 #endif
1214 
1215 /*
1216  * Re-evaluate the _CST object when we are notified that it changed.
1217  */
1218 static void
1219 acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context)
1220 {
1221     struct acpi_cpu_softc *sc = (struct acpi_cpu_softc *)context;
1222 
1223     if (notify != ACPI_NOTIFY_CX_STATES)
1224 	return;
1225 
1226     /*
1227      * C-state data for target CPU is going to be in flux while we execute
1228      * acpi_cpu_cx_cst, so disable entering acpi_cpu_idle.
1229      * Also, it may happen that multiple ACPI taskqueues may concurrently
1230      * execute notifications for the same CPU.  ACPI_SERIAL is used to
1231      * protect against that.
1232      */
1233     ACPI_SERIAL_BEGIN(cpu);
1234     disable_idle(sc);
1235 
1236     /* Update the list of Cx states. */
1237     acpi_cpu_cx_cst(sc);
1238     acpi_cpu_cx_list(sc);
1239     acpi_cpu_set_cx_lowest(sc);
1240 
1241     enable_idle(sc);
1242     ACPI_SERIAL_END(cpu);
1243 
1244     acpi_UserNotify("PROCESSOR", sc->cpu_handle, notify);
1245 }
1246 
1247 static void
1248 acpi_cpu_quirks(void)
1249 {
1250     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
1251 
1252     /*
1253      * Bus mastering arbitration control is needed to keep caches coherent
1254      * while sleeping in C3.  If it's not present but a working flush cache
1255      * instruction is present, flush the caches before entering C3 instead.
1256      * Otherwise, just disable C3 completely.
1257      */
1258     if (AcpiGbl_FADT.Pm2ControlBlock == 0 ||
1259 	AcpiGbl_FADT.Pm2ControlLength == 0) {
1260 	if ((AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD) &&
1261 	    (AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD_FLUSH) == 0) {
1262 	    cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1263 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1264 		"acpi_cpu: no BM control, using flush cache method\n"));
1265 	} else {
1266 	    cpu_quirks |= CPU_QUIRK_NO_C3;
1267 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1268 		"acpi_cpu: no BM control, C3 not available\n"));
1269 	}
1270     }
1271 
1272     /*
1273      * If we are using generic Cx mode, C3 on multiple CPUs requires using
1274      * the expensive flush cache instruction.
1275      */
1276     if (cpu_cx_generic && mp_ncpus > 1) {
1277 	cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1278 	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1279 	    "acpi_cpu: SMP, using flush cache mode for C3\n"));
1280     }
1281 
1282     /* Look for various quirks of the PIIX4 part. */
1283     acpi_cpu_quirks_piix4();
1284 }
1285 
1286 static void
1287 acpi_cpu_quirks_piix4(void)
1288 {
1289 #ifdef __i386__
1290     device_t acpi_dev;
1291     uint32_t val;
1292     ACPI_STATUS status;
1293 
1294     acpi_dev = pci_find_device(PCI_VENDOR_INTEL, PCI_DEVICE_82371AB_3);
1295     if (acpi_dev != NULL) {
1296 	switch (pci_get_revid(acpi_dev)) {
1297 	/*
1298 	 * Disable C3 support for all PIIX4 chipsets.  Some of these parts
1299 	 * do not report the BMIDE status to the BM status register and
1300 	 * others have a livelock bug if Type-F DMA is enabled.  Linux
1301 	 * works around the BMIDE bug by reading the BM status directly
1302 	 * but we take the simpler approach of disabling C3 for these
1303 	 * parts.
1304 	 *
1305 	 * See erratum #18 ("C3 Power State/BMIDE and Type-F DMA
1306 	 * Livelock") from the January 2002 PIIX4 specification update.
1307 	 * Applies to all PIIX4 models.
1308 	 *
1309 	 * Also, make sure that all interrupts cause a "Stop Break"
1310 	 * event to exit from C2 state.
1311 	 * Also, BRLD_EN_BM (ACPI_BITREG_BUS_MASTER_RLD in ACPI-speak)
1312 	 * should be set to zero, otherwise it causes C2 to short-sleep.
1313 	 * PIIX4 doesn't properly support C3 and bus master activity
1314 	 * need not break out of C2.
1315 	 */
1316 	case PCI_REVISION_A_STEP:
1317 	case PCI_REVISION_B_STEP:
1318 	case PCI_REVISION_4E:
1319 	case PCI_REVISION_4M:
1320 	    cpu_quirks |= CPU_QUIRK_NO_C3;
1321 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1322 		"acpi_cpu: working around PIIX4 bug, disabling C3\n"));
1323 
1324 	    val = pci_read_config(acpi_dev, PIIX4_DEVACTB_REG, 4);
1325 	    if ((val & PIIX4_STOP_BREAK_MASK) != PIIX4_STOP_BREAK_MASK) {
1326 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1327 		    "acpi_cpu: PIIX4: enabling IRQs to generate Stop Break\n"));
1328 	    	val |= PIIX4_STOP_BREAK_MASK;
1329 		pci_write_config(acpi_dev, PIIX4_DEVACTB_REG, val, 4);
1330 	    }
1331 	    status = AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_RLD, &val);
1332 	    if (ACPI_SUCCESS(status) && val != 0) {
1333 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1334 		    "acpi_cpu: PIIX4: reset BRLD_EN_BM\n"));
1335 		AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1336 	    }
1337 	    break;
1338 	default:
1339 	    break;
1340 	}
1341     }
1342 #endif
1343 }
1344 
1345 static int
1346 acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS)
1347 {
1348     struct acpi_cpu_softc *sc;
1349     struct sbuf	 sb;
1350     char	 buf[128];
1351     int		 i;
1352     uintmax_t	 fract, sum, whole;
1353 
1354     sc = (struct acpi_cpu_softc *) arg1;
1355     sum = 0;
1356     for (i = 0; i < sc->cpu_cx_count; i++)
1357 	sum += sc->cpu_cx_stats[i];
1358     sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1359     for (i = 0; i < sc->cpu_cx_count; i++) {
1360 	if (sum > 0) {
1361 	    whole = (uintmax_t)sc->cpu_cx_stats[i] * 100;
1362 	    fract = (whole % sum) * 100;
1363 	    sbuf_printf(&sb, "%u.%02u%% ", (u_int)(whole / sum),
1364 		(u_int)(fract / sum));
1365 	} else
1366 	    sbuf_printf(&sb, "0.00%% ");
1367     }
1368     sbuf_printf(&sb, "last %dus", sc->cpu_prev_sleep);
1369     sbuf_trim(&sb);
1370     sbuf_finish(&sb);
1371     sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1372     sbuf_delete(&sb);
1373 
1374     return (0);
1375 }
1376 
1377 /*
1378  * XXX TODO: actually add support to count each entry/exit
1379  * from the Cx states.
1380  */
1381 static int
1382 acpi_cpu_usage_counters_sysctl(SYSCTL_HANDLER_ARGS)
1383 {
1384     struct acpi_cpu_softc *sc;
1385     struct sbuf	 sb;
1386     char	 buf[128];
1387     int		 i;
1388 
1389     sc = (struct acpi_cpu_softc *) arg1;
1390 
1391     /* Print out the raw counters */
1392     sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1393 
1394     for (i = 0; i < sc->cpu_cx_count; i++) {
1395         sbuf_printf(&sb, "%u ", sc->cpu_cx_stats[i]);
1396     }
1397 
1398     sbuf_trim(&sb);
1399     sbuf_finish(&sb);
1400     sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1401     sbuf_delete(&sb);
1402 
1403     return (0);
1404 }
1405 
1406 #if defined(__i386__) || defined(__amd64__)
1407 static int
1408 acpi_cpu_method_sysctl(SYSCTL_HANDLER_ARGS)
1409 {
1410 	struct acpi_cpu_softc *sc;
1411 	struct acpi_cx *cx;
1412 	struct sbuf sb;
1413 	char buf[128];
1414 	int i;
1415 
1416 	sc = (struct acpi_cpu_softc *)arg1;
1417 	sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1418 	for (i = 0; i < sc->cpu_cx_count; i++) {
1419 		cx = &sc->cpu_cx_states[i];
1420 		sbuf_printf(&sb, "C%d/", i + 1);
1421 		if (cx->do_mwait) {
1422 			sbuf_cat(&sb, "mwait");
1423 			if (cx->mwait_hw_coord)
1424 				sbuf_cat(&sb, "/hwc");
1425 			if (cx->mwait_bm_avoidance)
1426 				sbuf_cat(&sb, "/bma");
1427 		} else if (cx->type == ACPI_STATE_C1) {
1428 			sbuf_cat(&sb, "hlt");
1429 		} else {
1430 			sbuf_cat(&sb, "io");
1431 		}
1432 		if (cx->type == ACPI_STATE_C1 && cx->p_lvlx != NULL)
1433 			sbuf_cat(&sb, "/iohlt");
1434 		sbuf_putc(&sb, ' ');
1435 	}
1436 	sbuf_trim(&sb);
1437 	sbuf_finish(&sb);
1438 	sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1439 	sbuf_delete(&sb);
1440 	return (0);
1441 }
1442 #endif
1443 
1444 static int
1445 acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc)
1446 {
1447     int i;
1448 
1449     ACPI_SERIAL_ASSERT(cpu);
1450     sc->cpu_cx_lowest = min(sc->cpu_cx_lowest_lim, sc->cpu_cx_count - 1);
1451 
1452     /* If not disabling, cache the new lowest non-C3 state. */
1453     sc->cpu_non_c3 = 0;
1454     for (i = sc->cpu_cx_lowest; i >= 0; i--) {
1455 	if (sc->cpu_cx_states[i].type < ACPI_STATE_C3) {
1456 	    sc->cpu_non_c3 = i;
1457 	    break;
1458 	}
1459     }
1460 
1461     /* Reset the statistics counters. */
1462     bzero(sc->cpu_cx_stats, sizeof(sc->cpu_cx_stats));
1463     return (0);
1464 }
1465 
1466 static int
1467 acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1468 {
1469     struct	 acpi_cpu_softc *sc;
1470     char	 state[8];
1471     int		 val, error;
1472 
1473     sc = (struct acpi_cpu_softc *) arg1;
1474     snprintf(state, sizeof(state), "C%d", sc->cpu_cx_lowest_lim + 1);
1475     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1476     if (error != 0 || req->newptr == NULL)
1477 	return (error);
1478     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1479 	return (EINVAL);
1480     if (strcasecmp(state, "Cmax") == 0)
1481 	val = MAX_CX_STATES;
1482     else {
1483 	val = (int) strtol(state + 1, NULL, 10);
1484 	if (val < 1 || val > MAX_CX_STATES)
1485 	    return (EINVAL);
1486     }
1487 
1488     ACPI_SERIAL_BEGIN(cpu);
1489     sc->cpu_cx_lowest_lim = val - 1;
1490     acpi_cpu_set_cx_lowest(sc);
1491     ACPI_SERIAL_END(cpu);
1492 
1493     return (0);
1494 }
1495 
1496 static int
1497 acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1498 {
1499     struct	acpi_cpu_softc *sc;
1500     char	state[8];
1501     int		val, error, i;
1502 
1503     snprintf(state, sizeof(state), "C%d", cpu_cx_lowest_lim + 1);
1504     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1505     if (error != 0 || req->newptr == NULL)
1506 	return (error);
1507     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1508 	return (EINVAL);
1509     if (strcasecmp(state, "Cmax") == 0)
1510 	val = MAX_CX_STATES;
1511     else {
1512 	val = (int) strtol(state + 1, NULL, 10);
1513 	if (val < 1 || val > MAX_CX_STATES)
1514 	    return (EINVAL);
1515     }
1516 
1517     /* Update the new lowest useable Cx state for all CPUs. */
1518     ACPI_SERIAL_BEGIN(cpu);
1519     cpu_cx_lowest_lim = val - 1;
1520     for (i = 0; i < cpu_ndevices; i++) {
1521 	sc = device_get_softc(cpu_devices[i]);
1522 	sc->cpu_cx_lowest_lim = cpu_cx_lowest_lim;
1523 	acpi_cpu_set_cx_lowest(sc);
1524     }
1525     ACPI_SERIAL_END(cpu);
1526 
1527     return (0);
1528 }
1529