xref: /dragonfly/sys/dev/acpica/acpi_cpu_cstate.c (revision 82730a9c)
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  * $FreeBSD: src/sys/dev/acpica/acpi_cpu.c,v 1.72 2008/04/12 12:06:00 rpaulo Exp $
28  */
29 
30 #include "opt_acpi.h"
31 #include <sys/param.h>
32 #include <sys/bus.h>
33 #include <sys/kernel.h>
34 #include <sys/malloc.h>
35 #include <sys/globaldata.h>
36 #include <sys/power.h>
37 #include <sys/proc.h>
38 #include <sys/sbuf.h>
39 #include <sys/thread2.h>
40 #include <sys/serialize.h>
41 #include <sys/msgport2.h>
42 
43 #include <bus/pci/pcivar.h>
44 #include <machine/atomic.h>
45 #include <machine/globaldata.h>
46 #include <machine/md_var.h>
47 #include <machine/smp.h>
48 #include <sys/rman.h>
49 
50 #include <net/netisr2.h>
51 #include <net/netmsg2.h>
52 #include <net/if_var.h>
53 
54 #include "acpi.h"
55 #include "acpivar.h"
56 #include "acpi_cpu.h"
57 
58 /*
59  * Support for ACPI Processor devices, including C[1-3] sleep states.
60  */
61 
62 /* Hooks for the ACPI CA debugging infrastructure */
63 #define _COMPONENT	ACPI_PROCESSOR
64 ACPI_MODULE_NAME("PROCESSOR")
65 
66 struct netmsg_acpi_cst {
67 	struct netmsg_base base;
68 	struct acpi_cpu_softc *sc;
69 	int		val;
70 };
71 
72 struct acpi_cx {
73     struct resource	*p_lvlx;	/* Register to read to enter state. */
74     int			 rid;		/* rid of p_lvlx */
75     uint32_t		 type;		/* C1-3 (C4 and up treated as C3). */
76     uint32_t		 trans_lat;	/* Transition latency (usec). */
77     uint32_t		 power;		/* Power consumed (mW). */
78     int			 res_type;	/* Resource type for p_lvlx. */
79 };
80 #define MAX_CX_STATES	 8
81 
82 struct acpi_cpu_softc {
83     device_t		 cpu_dev;
84     struct acpi_cpux_softc *cpu_parent;
85     ACPI_HANDLE		 cpu_handle;
86     int			 cpu_id;
87     uint32_t		 cst_flags;	/* ACPI_CST_FLAG_ */
88     uint32_t		 cpu_p_blk;	/* ACPI P_BLK location */
89     uint32_t		 cpu_p_blk_len;	/* P_BLK length (must be 6). */
90     struct acpi_cx	 cpu_cx_states[MAX_CX_STATES];
91     int			 cpu_cx_count;	/* Number of valid Cx states. */
92     int			 cpu_prev_sleep;/* Last idle sleep duration. */
93     /* Runtime state. */
94     int			 cpu_non_c3;	/* Index of lowest non-C3 state. */
95     u_long		 cpu_cx_stats[MAX_CX_STATES];/* Cx usage history. */
96     /* Values for sysctl. */
97     int			 cpu_cx_lowest; /* Current Cx lowest */
98     int			 cpu_cx_lowest_req; /* Requested Cx lowest */
99     char 		 cpu_cx_supported[64];
100 };
101 
102 #define ACPI_CST_FLAG_PROBING	0x1
103 
104 struct acpi_cpu_device {
105     struct resource_list	ad_rl;
106 };
107 
108 #define CPU_GET_REG(reg, width) 					\
109     (bus_space_read_ ## width(rman_get_bustag((reg)), 			\
110 		      rman_get_bushandle((reg)), 0))
111 #define CPU_SET_REG(reg, width, val)					\
112     (bus_space_write_ ## width(rman_get_bustag((reg)), 			\
113 		       rman_get_bushandle((reg)), 0, (val)))
114 
115 #define PM_USEC(x)	 ((x) >> 2)	/* ~4 clocks per usec (3.57955 Mhz) */
116 
117 #define ACPI_NOTIFY_CX_STATES	0x81	/* _CST changed. */
118 
119 #define CPU_QUIRK_NO_C3		(1<<0)	/* C3-type states are not usable. */
120 #define CPU_QUIRK_NO_BM_CTRL	(1<<2)	/* No bus mastering control. */
121 
122 #define PCI_VENDOR_INTEL	0x8086
123 #define PCI_DEVICE_82371AB_3	0x7113	/* PIIX4 chipset for quirks. */
124 #define PCI_REVISION_A_STEP	0
125 #define PCI_REVISION_B_STEP	1
126 #define PCI_REVISION_4E		2
127 #define PCI_REVISION_4M		3
128 #define PIIX4_DEVACTB_REG	0x58
129 #define PIIX4_BRLD_EN_IRQ0	(1<<0)
130 #define PIIX4_BRLD_EN_IRQ	(1<<1)
131 #define PIIX4_BRLD_EN_IRQ8	(1<<5)
132 #define PIIX4_STOP_BREAK_MASK	(PIIX4_BRLD_EN_IRQ0 | PIIX4_BRLD_EN_IRQ | PIIX4_BRLD_EN_IRQ8)
133 #define PIIX4_PCNTRL_BST_EN	(1<<10)
134 
135 /* 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 static int		 cpu_cx_count;	/* Number of valid Cx states */
143 
144 /* Values for sysctl. */
145 static int		 cpu_cx_generic;
146 static int		 cpu_cx_lowest; /* Current Cx lowest */
147 static int		 cpu_cx_lowest_req; /* Requested Cx lowest */
148 static struct lwkt_serialize cpu_cx_slize = LWKT_SERIALIZE_INITIALIZER;
149 
150 /* C3 state transition */
151 static int		 cpu_c3_ncpus;
152 
153 static device_t		*cpu_devices;
154 static int		 cpu_ndevices;
155 static struct acpi_cpu_softc **cpu_softc;
156 
157 static int	acpi_cpu_cst_probe(device_t dev);
158 static int	acpi_cpu_cst_attach(device_t dev);
159 static int	acpi_cpu_cst_suspend(device_t dev);
160 static int	acpi_cpu_cst_resume(device_t dev);
161 static struct resource_list *acpi_cpu_cst_get_rlist(device_t dev,
162 		    device_t child);
163 static device_t	acpi_cpu_cst_add_child(device_t bus, device_t parent,
164 		    int order, const char *name, int unit);
165 static int	acpi_cpu_cst_read_ivar(device_t dev, device_t child,
166 		    int index, uintptr_t *result);
167 static int	acpi_cpu_cst_shutdown(device_t dev);
168 static void	acpi_cpu_cx_probe(struct acpi_cpu_softc *sc);
169 static void	acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc);
170 static int	acpi_cpu_cx_cst(struct acpi_cpu_softc *sc);
171 static int	acpi_cpu_cx_cst_dispatch(struct acpi_cpu_softc *sc);
172 static void	acpi_cpu_startup(void *arg);
173 static void	acpi_cpu_startup_cx(struct acpi_cpu_softc *sc);
174 static void	acpi_cpu_cx_list(struct acpi_cpu_softc *sc);
175 static void	acpi_cpu_idle(void);
176 static void	acpi_cpu_cst_notify(device_t);
177 static int	acpi_cpu_quirks(void);
178 static int	acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS);
179 static int	acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *, int);
180 static int	acpi_cpu_set_cx_lowest_oncpu(struct acpi_cpu_softc *, int);
181 static int	acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
182 static int	acpi_cpu_cx_lowest_use_sysctl(SYSCTL_HANDLER_ARGS);
183 static int	acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
184 static int	acpi_cpu_global_cx_lowest_use_sysctl(SYSCTL_HANDLER_ARGS);
185 static void	acpi_cpu_cx_non_c3(struct acpi_cpu_softc *sc);
186 static void	acpi_cpu_global_cx_count(void);
187 
188 static void	acpi_cpu_c1(void);	/* XXX */
189 
190 static device_method_t acpi_cpu_cst_methods[] = {
191     /* Device interface */
192     DEVMETHOD(device_probe,	acpi_cpu_cst_probe),
193     DEVMETHOD(device_attach,	acpi_cpu_cst_attach),
194     DEVMETHOD(device_detach,	bus_generic_detach),
195     DEVMETHOD(device_shutdown,	acpi_cpu_cst_shutdown),
196     DEVMETHOD(device_suspend,	acpi_cpu_cst_suspend),
197     DEVMETHOD(device_resume,	acpi_cpu_cst_resume),
198 
199     /* Bus interface */
200     DEVMETHOD(bus_add_child,	acpi_cpu_cst_add_child),
201     DEVMETHOD(bus_read_ivar,	acpi_cpu_cst_read_ivar),
202     DEVMETHOD(bus_get_resource_list, acpi_cpu_cst_get_rlist),
203     DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
204     DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
205     DEVMETHOD(bus_alloc_resource, bus_generic_rl_alloc_resource),
206     DEVMETHOD(bus_release_resource, bus_generic_rl_release_resource),
207     DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
208     DEVMETHOD(bus_activate_resource, bus_generic_activate_resource),
209     DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource),
210     DEVMETHOD(bus_setup_intr,	bus_generic_setup_intr),
211     DEVMETHOD(bus_teardown_intr, bus_generic_teardown_intr),
212     DEVMETHOD_END
213 };
214 
215 static driver_t acpi_cpu_cst_driver = {
216     "cpu_cst",
217     acpi_cpu_cst_methods,
218     sizeof(struct acpi_cpu_softc),
219 };
220 
221 static devclass_t acpi_cpu_cst_devclass;
222 DRIVER_MODULE(cpu_cst, cpu, acpi_cpu_cst_driver, acpi_cpu_cst_devclass, NULL, NULL);
223 MODULE_DEPEND(cpu_cst, acpi, 1, 1, 1);
224 
225 static int
226 acpi_cpu_cst_probe(device_t dev)
227 {
228     int cpu_id;
229 
230     if (acpi_disabled("cpu_cst") || acpi_get_type(dev) != ACPI_TYPE_PROCESSOR)
231 	return (ENXIO);
232 
233     cpu_id = acpi_get_magic(dev);
234 
235     if (cpu_softc == NULL)
236 	cpu_softc = kmalloc(sizeof(struct acpi_cpu_softc *) *
237 	    SMP_MAXCPU, M_TEMP /* XXX */, M_INTWAIT | M_ZERO);
238 
239     /*
240      * Check if we already probed this processor.  We scan the bus twice
241      * so it's possible we've already seen this one.
242      */
243     if (cpu_softc[cpu_id] != NULL) {
244 	device_printf(dev, "CPU%d cstate already exist\n", cpu_id);
245 	return (ENXIO);
246     }
247 
248     /* Mark this processor as in-use and save our derived id for attach. */
249     cpu_softc[cpu_id] = (void *)1;
250     device_set_desc(dev, "ACPI CPU C-State");
251 
252     return (0);
253 }
254 
255 static int
256 acpi_cpu_cst_attach(device_t dev)
257 {
258     ACPI_BUFFER		   buf;
259     ACPI_OBJECT		   *obj;
260     struct acpi_cpu_softc *sc;
261     ACPI_STATUS		   status;
262 
263     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
264 
265     sc = device_get_softc(dev);
266     sc->cpu_dev = dev;
267     sc->cpu_parent = device_get_softc(device_get_parent(dev));
268     sc->cpu_handle = acpi_get_handle(dev);
269     sc->cpu_id = acpi_get_magic(dev);
270     cpu_softc[sc->cpu_id] = sc;
271     cpu_smi_cmd = AcpiGbl_FADT.SmiCommand;
272     cpu_cst_cnt = AcpiGbl_FADT.CstControl;
273 
274     buf.Pointer = NULL;
275     buf.Length = ACPI_ALLOCATE_BUFFER;
276     status = AcpiEvaluateObject(sc->cpu_handle, NULL, NULL, &buf);
277     if (ACPI_FAILURE(status)) {
278 	device_printf(dev, "attach failed to get Processor obj - %s\n",
279 		      AcpiFormatException(status));
280 	return (ENXIO);
281     }
282     obj = (ACPI_OBJECT *)buf.Pointer;
283     sc->cpu_p_blk = obj->Processor.PblkAddress;
284     sc->cpu_p_blk_len = obj->Processor.PblkLength;
285     AcpiOsFree(obj);
286     ACPI_DEBUG_PRINT((ACPI_DB_INFO, "acpi_cpu%d: P_BLK at %#x/%d\n",
287 		     device_get_unit(dev), sc->cpu_p_blk, sc->cpu_p_blk_len));
288 
289     /*
290      * If this is the first cpu we attach, create and initialize the generic
291      * resources that will be used by all acpi cpu devices.
292      */
293     if (device_get_unit(dev) == 0) {
294 	/* Assume we won't be using generic Cx mode by default */
295 	cpu_cx_generic = FALSE;
296 
297 	/* Queue post cpu-probing task handler */
298 	AcpiOsExecute(OSL_NOTIFY_HANDLER, acpi_cpu_startup, NULL);
299     }
300 
301     /* Probe for Cx state support. */
302     acpi_cpu_cx_probe(sc);
303 
304     /* Finally,  call identify and probe/attach for child devices. */
305     bus_generic_probe(dev);
306     bus_generic_attach(dev);
307 
308     return (0);
309 }
310 
311 /*
312  * Disable any entry to the idle function during suspend and re-enable it
313  * during resume.
314  */
315 static int
316 acpi_cpu_cst_suspend(device_t dev)
317 {
318     int error;
319 
320     error = bus_generic_suspend(dev);
321     if (error)
322 	return (error);
323     cpu_disable_idle = TRUE;
324     return (0);
325 }
326 
327 static int
328 acpi_cpu_cst_resume(device_t dev)
329 {
330 
331     cpu_disable_idle = FALSE;
332     return (bus_generic_resume(dev));
333 }
334 
335 static struct resource_list *
336 acpi_cpu_cst_get_rlist(device_t dev, device_t child)
337 {
338     struct acpi_cpu_device *ad;
339 
340     ad = device_get_ivars(child);
341     if (ad == NULL)
342 	return (NULL);
343     return (&ad->ad_rl);
344 }
345 
346 static device_t
347 acpi_cpu_cst_add_child(device_t bus, device_t parent, int order,
348     const char *name, int unit)
349 {
350     struct acpi_cpu_device *ad;
351     device_t child;
352 
353     if ((ad = kmalloc(sizeof(*ad), M_TEMP, M_NOWAIT | M_ZERO)) == NULL)
354 	return (NULL);
355 
356     resource_list_init(&ad->ad_rl);
357 
358     child = device_add_child_ordered(parent, order, name, unit);
359     if (child != NULL)
360 	device_set_ivars(child, ad);
361     else
362 	kfree(ad, M_TEMP);
363     return (child);
364 }
365 
366 static int
367 acpi_cpu_cst_read_ivar(device_t dev, device_t child, int index,
368     uintptr_t *result)
369 {
370     struct acpi_cpu_softc *sc;
371 
372     sc = device_get_softc(dev);
373     switch (index) {
374     case ACPI_IVAR_HANDLE:
375 	*result = (uintptr_t)sc->cpu_handle;
376 	break;
377 #if 0
378     case CPU_IVAR_PCPU:
379 	*result = (uintptr_t)sc->cpu_pcpu;
380 	break;
381 #endif
382     default:
383 	return (ENOENT);
384     }
385     return (0);
386 }
387 
388 static int
389 acpi_cpu_cst_shutdown(device_t dev)
390 {
391     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
392 
393     /* Allow children to shutdown first. */
394     bus_generic_shutdown(dev);
395 
396     /*
397      * Disable any entry to the idle function.  There is a small race where
398      * an idle thread have passed this check but not gone to sleep.  This
399      * is ok since device_shutdown() does not free the softc, otherwise
400      * we'd have to be sure all threads were evicted before returning.
401      */
402     cpu_disable_idle = TRUE;
403 
404     return_VALUE (0);
405 }
406 
407 static void
408 acpi_cpu_cx_probe(struct acpi_cpu_softc *sc)
409 {
410     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
411 
412     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
413     sc->cpu_prev_sleep = 1000000;
414     sc->cpu_cx_lowest = 0;
415     sc->cpu_cx_lowest_req = 0;
416 
417     /*
418      * Check for the ACPI 2.0 _CST sleep states object. If we can't find
419      * any, we'll revert to generic FADT/P_BLK Cx control method which will
420      * be handled by acpi_cpu_startup. We need to defer to after having
421      * probed all the cpus in the system before probing for generic Cx
422      * states as we may already have found cpus with valid _CST packages
423      */
424     if (!cpu_cx_generic && acpi_cpu_cx_cst(sc) != 0) {
425 	/*
426 	 * We were unable to find a _CST package for this cpu or there
427 	 * was an error parsing it. Switch back to generic mode.
428 	 */
429 	cpu_cx_generic = TRUE;
430 	if (bootverbose)
431 	    device_printf(sc->cpu_dev, "switching to generic Cx mode\n");
432     }
433 
434     /*
435      * TODO: _CSD Package should be checked here.
436      */
437 }
438 
439 static void
440 acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc)
441 {
442     ACPI_GENERIC_ADDRESS	 gas;
443     struct acpi_cx		*cx_ptr;
444 
445     sc->cpu_cx_count = 0;
446     cx_ptr = sc->cpu_cx_states;
447 
448     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
449     sc->cpu_prev_sleep = 1000000;
450 
451     /* C1 has been required since just after ACPI 1.0 */
452     cx_ptr->type = ACPI_STATE_C1;
453     cx_ptr->trans_lat = 0;
454     cx_ptr++;
455     sc->cpu_cx_count++;
456 
457     /*
458      * The spec says P_BLK must be 6 bytes long.  However, some systems
459      * use it to indicate a fractional set of features present so we
460      * take 5 as C2.  Some may also have a value of 7 to indicate
461      * another C3 but most use _CST for this (as required) and having
462      * "only" C1-C3 is not a hardship.
463      */
464     if (sc->cpu_p_blk_len < 5)
465 	return;
466 
467     /* Validate and allocate resources for C2 (P_LVL2). */
468     gas.SpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
469     gas.BitWidth = 8;
470     if (AcpiGbl_FADT.C2Latency <= 100) {
471 	gas.Address = sc->cpu_p_blk + 4;
472 
473 	cx_ptr->rid = sc->cpu_parent->cpux_next_rid;
474 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->type, &cx_ptr->rid, &gas,
475 	    &cx_ptr->p_lvlx, RF_SHAREABLE);
476 	if (cx_ptr->p_lvlx != NULL) {
477 	    sc->cpu_parent->cpux_next_rid++;
478 	    cx_ptr->type = ACPI_STATE_C2;
479 	    cx_ptr->trans_lat = AcpiGbl_FADT.C2Latency;
480 	    cx_ptr++;
481 	    sc->cpu_cx_count++;
482 	    sc->cpu_non_c3 = 1;
483 	}
484     }
485     if (sc->cpu_p_blk_len < 6)
486 	return;
487 
488     /* Validate and allocate resources for C3 (P_LVL3). */
489     if (AcpiGbl_FADT.C3Latency <= 1000 && !(cpu_quirks & CPU_QUIRK_NO_C3)) {
490 	gas.Address = sc->cpu_p_blk + 5;
491 
492 	cx_ptr->rid = sc->cpu_parent->cpux_next_rid;
493 	acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->type, &cx_ptr->rid, &gas,
494 	    &cx_ptr->p_lvlx, RF_SHAREABLE);
495 	if (cx_ptr->p_lvlx != NULL) {
496 	    sc->cpu_parent->cpux_next_rid++;
497 	    cx_ptr->type = ACPI_STATE_C3;
498 	    cx_ptr->trans_lat = AcpiGbl_FADT.C3Latency;
499 	    cx_ptr++;
500 	    sc->cpu_cx_count++;
501 	}
502     }
503 }
504 
505 /*
506  * Parse a _CST package and set up its Cx states.  Since the _CST object
507  * can change dynamically, our notify handler may call this function
508  * to clean up and probe the new _CST package.
509  */
510 static int
511 acpi_cpu_cx_cst(struct acpi_cpu_softc *sc)
512 {
513     struct	 acpi_cx *cx_ptr;
514     ACPI_STATUS	 status;
515     ACPI_BUFFER	 buf;
516     ACPI_OBJECT	*top;
517     ACPI_OBJECT	*pkg;
518     uint32_t	 count;
519     int		 i;
520 
521     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
522 
523     buf.Pointer = NULL;
524     buf.Length = ACPI_ALLOCATE_BUFFER;
525     status = AcpiEvaluateObject(sc->cpu_handle, "_CST", NULL, &buf);
526     if (ACPI_FAILURE(status))
527 	return (ENXIO);
528 
529     /* _CST is a package with a count and at least one Cx package. */
530     top = (ACPI_OBJECT *)buf.Pointer;
531     if (!ACPI_PKG_VALID(top, 2) || acpi_PkgInt32(top, 0, &count) != 0) {
532 	device_printf(sc->cpu_dev, "invalid _CST package\n");
533 	AcpiOsFree(buf.Pointer);
534 	return (ENXIO);
535     }
536     if (count != top->Package.Count - 1) {
537 	device_printf(sc->cpu_dev, "invalid _CST state count (%d != %d)\n",
538 	       count, top->Package.Count - 1);
539 	count = top->Package.Count - 1;
540     }
541     if (count > MAX_CX_STATES) {
542 	device_printf(sc->cpu_dev, "_CST has too many states (%d)\n", count);
543 	count = MAX_CX_STATES;
544     }
545 
546     sc->cst_flags |= ACPI_CST_FLAG_PROBING;
547     cpu_sfence();
548 
549     /* Set up all valid states. */
550     sc->cpu_cx_count = 0;
551     cx_ptr = sc->cpu_cx_states;
552     for (i = 0; i < count; i++) {
553 	pkg = &top->Package.Elements[i + 1];
554 	if (!ACPI_PKG_VALID(pkg, 4) ||
555 	    acpi_PkgInt32(pkg, 1, &cx_ptr->type) != 0 ||
556 	    acpi_PkgInt32(pkg, 2, &cx_ptr->trans_lat) != 0 ||
557 	    acpi_PkgInt32(pkg, 3, &cx_ptr->power) != 0) {
558 
559 	    device_printf(sc->cpu_dev, "skipping invalid Cx state package\n");
560 	    continue;
561 	}
562 
563 	/* Validate the state to see if we should use it. */
564 	switch (cx_ptr->type) {
565 	case ACPI_STATE_C1:
566 	    sc->cpu_non_c3 = i;
567 	    cx_ptr++;
568 	    sc->cpu_cx_count++;
569 	    continue;
570 	case ACPI_STATE_C2:
571 	    sc->cpu_non_c3 = i;
572 	    break;
573 	case ACPI_STATE_C3:
574 	default:
575 	    if ((cpu_quirks & CPU_QUIRK_NO_C3) != 0) {
576 
577 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
578 				 "acpi_cpu%d: C3[%d] not available.\n",
579 				 device_get_unit(sc->cpu_dev), i));
580 		continue;
581 	    }
582 	    break;
583 	}
584 
585 #ifdef notyet
586 	/* Free up any previous register. */
587 	if (cx_ptr->p_lvlx != NULL) {
588 	    bus_release_resource(sc->cpu_dev, 0, 0, cx_ptr->p_lvlx);
589 	    cx_ptr->p_lvlx = NULL;
590 	}
591 #endif
592 
593 	/* Allocate the control register for C2 or C3. */
594 	cx_ptr->rid = sc->cpu_parent->cpux_next_rid;
595 	acpi_PkgGas(sc->cpu_dev, pkg, 0, &cx_ptr->res_type, &cx_ptr->rid,
596 	    &cx_ptr->p_lvlx, RF_SHAREABLE);
597 	if (cx_ptr->p_lvlx) {
598 	    sc->cpu_parent->cpux_next_rid++;
599 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
600 			     "acpi_cpu%d: Got C%d - %d latency\n",
601 			     device_get_unit(sc->cpu_dev), cx_ptr->type,
602 			     cx_ptr->trans_lat));
603 	    cx_ptr++;
604 	    sc->cpu_cx_count++;
605 	}
606     }
607     AcpiOsFree(buf.Pointer);
608 
609     /*
610      * Fix up the lowest Cx being used
611      */
612     if (sc->cpu_cx_lowest_req < sc->cpu_cx_count)
613 	sc->cpu_cx_lowest = sc->cpu_cx_lowest_req;
614     if (sc->cpu_cx_lowest > sc->cpu_cx_count - 1)
615 	sc->cpu_cx_lowest = sc->cpu_cx_count - 1;
616 
617     /*
618      * Cache the lowest non-C3 state.
619      * NOTE: must after cpu_cx_lowest is set.
620      */
621     acpi_cpu_cx_non_c3(sc);
622 
623     cpu_sfence();
624     sc->cst_flags &= ~ACPI_CST_FLAG_PROBING;
625 
626     return (0);
627 }
628 
629 static void
630 acpi_cst_probe_handler(netmsg_t msg)
631 {
632     struct netmsg_acpi_cst *rmsg = (struct netmsg_acpi_cst *)msg;
633     int error;
634 
635     error = acpi_cpu_cx_cst(rmsg->sc);
636     lwkt_replymsg(&rmsg->base.lmsg, error);
637 }
638 
639 static int
640 acpi_cpu_cx_cst_dispatch(struct acpi_cpu_softc *sc)
641 {
642     struct netmsg_acpi_cst msg;
643 
644     netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
645 	acpi_cst_probe_handler);
646     msg.sc = sc;
647 
648     return lwkt_domsg(netisr_cpuport(sc->cpu_id), &msg.base.lmsg, 0);
649 }
650 
651 /*
652  * Call this *after* all CPUs have been attached.
653  */
654 static void
655 acpi_cpu_startup(void *arg)
656 {
657     struct acpi_cpu_softc *sc;
658     int i;
659 
660     /* Get set of CPU devices */
661     devclass_get_devices(acpi_cpu_cst_devclass, &cpu_devices, &cpu_ndevices);
662 
663     /*
664      * Setup any quirks that might necessary now that we have probed
665      * all the CPUs
666      */
667     acpi_cpu_quirks();
668 
669     if (cpu_cx_generic) {
670 	/*
671 	 * We are using generic Cx mode, probe for available Cx states
672 	 * for all processors.
673 	 */
674 	for (i = 0; i < cpu_ndevices; i++) {
675 	    sc = device_get_softc(cpu_devices[i]);
676 	    acpi_cpu_generic_cx_probe(sc);
677 	}
678     } else {
679 	/*
680 	 * We are using _CST mode, remove C3 state if necessary.
681 	 *
682 	 * As we now know for sure that we will be using _CST mode
683 	 * install our notify handler.
684 	 */
685 	for (i = 0; i < cpu_ndevices; i++) {
686 	    sc = device_get_softc(cpu_devices[i]);
687 	    if (cpu_quirks & CPU_QUIRK_NO_C3)
688 		sc->cpu_cx_count = sc->cpu_non_c3 + 1;
689 	    sc->cpu_parent->cpux_cst_notify = acpi_cpu_cst_notify;
690 	}
691     }
692     acpi_cpu_global_cx_count();
693 
694     /* Perform Cx final initialization. */
695     for (i = 0; i < cpu_ndevices; i++) {
696 	sc = device_get_softc(cpu_devices[i]);
697 	acpi_cpu_startup_cx(sc);
698 
699 	if (sc->cpu_parent->glob_sysctl_tree != NULL) {
700 	    struct acpi_cpux_softc *cpux = sc->cpu_parent;
701 
702 	    /* Add a sysctl handler to handle global Cx lowest setting */
703 	    SYSCTL_ADD_PROC(&cpux->glob_sysctl_ctx,
704 	    		    SYSCTL_CHILDREN(cpux->glob_sysctl_tree),
705 			    OID_AUTO, "cx_lowest",
706 			    CTLTYPE_STRING | CTLFLAG_RW, NULL, 0,
707 			    acpi_cpu_global_cx_lowest_sysctl, "A",
708 			    "Requested global lowest Cx sleep state");
709 	    SYSCTL_ADD_PROC(&cpux->glob_sysctl_ctx,
710 	    		    SYSCTL_CHILDREN(cpux->glob_sysctl_tree),
711 			    OID_AUTO, "cx_lowest_use",
712 			    CTLTYPE_STRING | CTLFLAG_RD, NULL, 0,
713 			    acpi_cpu_global_cx_lowest_use_sysctl, "A",
714 			    "Global lowest Cx sleep state to use");
715 	}
716     }
717 
718     /* Take over idling from cpu_idle_default(). */
719     cpu_cx_lowest = 0;
720     cpu_cx_lowest_req = 0;
721     cpu_disable_idle = FALSE;
722     cpu_idle_hook = acpi_cpu_idle;
723 }
724 
725 static void
726 acpi_cpu_cx_list(struct acpi_cpu_softc *sc)
727 {
728     struct sbuf sb;
729     int i;
730 
731     /*
732      * Set up the list of Cx states
733      */
734     sbuf_new(&sb, sc->cpu_cx_supported, sizeof(sc->cpu_cx_supported),
735 	SBUF_FIXEDLEN);
736     for (i = 0; i < sc->cpu_cx_count; i++)
737 	sbuf_printf(&sb, "C%d/%d ", i + 1, sc->cpu_cx_states[i].trans_lat);
738     sbuf_trim(&sb);
739     sbuf_finish(&sb);
740 }
741 
742 static void
743 acpi_cpu_startup_cx(struct acpi_cpu_softc *sc)
744 {
745     struct acpi_cpux_softc *cpux = sc->cpu_parent;
746 
747     acpi_cpu_cx_list(sc);
748 
749     SYSCTL_ADD_STRING(&cpux->pcpu_sysctl_ctx,
750 		      SYSCTL_CHILDREN(cpux->pcpu_sysctl_tree),
751 		      OID_AUTO, "cx_supported", CTLFLAG_RD,
752 		      sc->cpu_cx_supported, 0,
753 		      "Cx/microsecond values for supported Cx states");
754     SYSCTL_ADD_PROC(&cpux->pcpu_sysctl_ctx,
755 		    SYSCTL_CHILDREN(cpux->pcpu_sysctl_tree),
756 		    OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
757 		    (void *)sc, 0, acpi_cpu_cx_lowest_sysctl, "A",
758 		    "requested lowest Cx sleep state");
759     SYSCTL_ADD_PROC(&cpux->pcpu_sysctl_ctx,
760 		    SYSCTL_CHILDREN(cpux->pcpu_sysctl_tree),
761 		    OID_AUTO, "cx_lowest_use", CTLTYPE_STRING | CTLFLAG_RD,
762 		    (void *)sc, 0, acpi_cpu_cx_lowest_use_sysctl, "A",
763 		    "lowest Cx sleep state to use");
764     SYSCTL_ADD_PROC(&cpux->pcpu_sysctl_ctx,
765 		    SYSCTL_CHILDREN(cpux->pcpu_sysctl_tree),
766 		    OID_AUTO, "cx_usage", CTLTYPE_STRING | CTLFLAG_RD,
767 		    (void *)sc, 0, acpi_cpu_usage_sysctl, "A",
768 		    "percent usage for each Cx state");
769 
770 #ifdef notyet
771     /* Signal platform that we can handle _CST notification. */
772     if (!cpu_cx_generic && cpu_cst_cnt != 0) {
773 	ACPI_LOCK(acpi);
774 	AcpiOsWritePort(cpu_smi_cmd, cpu_cst_cnt, 8);
775 	ACPI_UNLOCK(acpi);
776     }
777 #endif
778 }
779 
780 /*
781  * Idle the CPU in the lowest state possible.  This function is called with
782  * interrupts disabled.  Note that once it re-enables interrupts, a task
783  * switch can occur so do not access shared data (i.e. the softc) after
784  * interrupts are re-enabled.
785  */
786 static void
787 acpi_cpu_idle(void)
788 {
789     struct	acpi_cpu_softc *sc;
790     struct	acpi_cx *cx_next;
791     uint64_t	start_time, end_time;
792     int		bm_active, cx_next_idx, i;
793 
794     /* If disabled, return immediately. */
795     if (cpu_disable_idle) {
796 	ACPI_ENABLE_IRQS();
797 	return;
798     }
799 
800     /*
801      * Look up our CPU id to get our softc.  If it's NULL, we'll use C1
802      * since there is no ACPI processor object for this CPU.  This occurs
803      * for logical CPUs in the HTT case.
804      */
805     sc = cpu_softc[mdcpu->mi.gd_cpuid];
806     if (sc == NULL) {
807 	acpi_cpu_c1();
808 	return;
809     }
810 
811     /* Still probing; use C1 */
812     if (sc->cst_flags & ACPI_CST_FLAG_PROBING) {
813 	acpi_cpu_c1();
814 	return;
815     }
816 
817     /* Find the lowest state that has small enough latency. */
818     cx_next_idx = 0;
819     for (i = sc->cpu_cx_lowest; i >= 0; i--) {
820 	if (sc->cpu_cx_states[i].trans_lat * 3 <= sc->cpu_prev_sleep) {
821 	    cx_next_idx = i;
822 	    break;
823 	}
824     }
825 
826     /*
827      * Check for bus master activity.  If there was activity, clear
828      * the bit and use the lowest non-C3 state.  Note that the USB
829      * driver polling for new devices keeps this bit set all the
830      * time if USB is loaded.
831      */
832     if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
833 	AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, &bm_active);
834 	if (bm_active != 0) {
835 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, 1);
836 	    cx_next_idx = min(cx_next_idx, sc->cpu_non_c3);
837 	}
838     }
839 
840     /* Select the next state and update statistics. */
841     cx_next = &sc->cpu_cx_states[cx_next_idx];
842     sc->cpu_cx_stats[cx_next_idx]++;
843     KASSERT(cx_next->type != ACPI_STATE_C0, ("acpi_cpu_idle: C0 sleep"));
844 
845     /*
846      * Execute HLT (or equivalent) and wait for an interrupt.  We can't
847      * calculate the time spent in C1 since the place we wake up is an
848      * ISR.  Assume we slept half of quantum and return.
849      */
850     if (cx_next->type == ACPI_STATE_C1) {
851 	sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + 500000 / hz) / 4;
852 	acpi_cpu_c1();
853 	return;
854     }
855 
856     /*
857      * For C3(+), disable bus master arbitration and enable bus master wake
858      * if BM control is available, otherwise flush the CPU cache.
859      */
860     if (cx_next->type >= ACPI_STATE_C3) {
861 	if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
862 	    AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 1);
863 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
864 	} else
865 	    ACPI_FLUSH_CPU_CACHE();
866     }
867 
868     /*
869      * Read from P_LVLx to enter C2(+), checking time spent asleep.
870      * Use the ACPI timer for measuring sleep time.  Since we need to
871      * get the time very close to the CPU start/stop clock logic, this
872      * is the only reliable time source.
873      */
874     AcpiRead(&start_time, &AcpiGbl_FADT.XPmTimerBlock);
875     CPU_GET_REG(cx_next->p_lvlx, 1);
876 
877     /*
878      * Read the end time twice.  Since it may take an arbitrary time
879      * to enter the idle state, the first read may be executed before
880      * the processor has stopped.  Doing it again provides enough
881      * margin that we are certain to have a correct value.
882      */
883     AcpiRead(&end_time, &AcpiGbl_FADT.XPmTimerBlock);
884     AcpiRead(&end_time, &AcpiGbl_FADT.XPmTimerBlock);
885 
886     /* Enable bus master arbitration and disable bus master wakeup. */
887     if (cx_next->type >= ACPI_STATE_C3) {
888 	if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
889 	    AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 0);
890 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
891 	}
892     }
893     ACPI_ENABLE_IRQS();
894 
895     /* Find the actual time asleep in microseconds. */
896     end_time = acpi_TimerDelta(end_time, start_time);
897     sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + PM_USEC(end_time)) / 4;
898 }
899 
900 /*
901  * Re-evaluate the _CST object when we are notified that it changed.
902  */
903 static void
904 acpi_cpu_cst_notify(device_t dev)
905 {
906     struct acpi_cpu_softc *sc = device_get_softc(dev);
907 
908     KASSERT(curthread->td_type != TD_TYPE_NETISR,
909         ("notify in netisr%d", mycpuid));
910 
911     lwkt_serialize_enter(&cpu_cx_slize);
912 
913     /* Update the list of Cx states. */
914     acpi_cpu_cx_cst_dispatch(sc);
915     acpi_cpu_cx_list(sc);
916 
917     /* Update the new lowest useable Cx state for all CPUs. */
918     acpi_cpu_global_cx_count();
919 
920     /*
921      * Fix up the lowest Cx being used
922      */
923     if (cpu_cx_lowest_req < cpu_cx_count)
924 	cpu_cx_lowest = cpu_cx_lowest_req;
925     if (cpu_cx_lowest > cpu_cx_count - 1)
926 	cpu_cx_lowest = cpu_cx_count - 1;
927 
928     lwkt_serialize_exit(&cpu_cx_slize);
929 }
930 
931 static int
932 acpi_cpu_quirks(void)
933 {
934     device_t acpi_dev;
935     uint32_t val;
936 
937     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
938 
939     /*
940      * Bus mastering arbitration control is needed to keep caches coherent
941      * while sleeping in C3.  If it's not present but a working flush cache
942      * instruction is present, flush the caches before entering C3 instead.
943      * Otherwise, just disable C3 completely.
944      */
945     if (AcpiGbl_FADT.Pm2ControlBlock == 0 ||
946 	AcpiGbl_FADT.Pm2ControlLength == 0) {
947 	if ((AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD) &&
948 	    (AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD_FLUSH) == 0) {
949 	    cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
950 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
951 		"acpi_cpu: no BM control, using flush cache method\n"));
952 	} else {
953 	    cpu_quirks |= CPU_QUIRK_NO_C3;
954 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
955 		"acpi_cpu: no BM control, C3 not available\n"));
956 	}
957     }
958 
959     /*
960      * If we are using generic Cx mode, C3 on multiple CPUs requires using
961      * the expensive flush cache instruction.
962      */
963     if (cpu_cx_generic && ncpus > 1) {
964 	cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
965 	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
966 	    "acpi_cpu: SMP, using flush cache mode for C3\n"));
967     }
968 
969     /* Look for various quirks of the PIIX4 part. */
970     acpi_dev = pci_find_device(PCI_VENDOR_INTEL, PCI_DEVICE_82371AB_3);
971     if (acpi_dev != NULL) {
972 	switch (pci_get_revid(acpi_dev)) {
973 	/*
974 	 * Disable C3 support for all PIIX4 chipsets.  Some of these parts
975 	 * do not report the BMIDE status to the BM status register and
976 	 * others have a livelock bug if Type-F DMA is enabled.  Linux
977 	 * works around the BMIDE bug by reading the BM status directly
978 	 * but we take the simpler approach of disabling C3 for these
979 	 * parts.
980 	 *
981 	 * See erratum #18 ("C3 Power State/BMIDE and Type-F DMA
982 	 * Livelock") from the January 2002 PIIX4 specification update.
983 	 * Applies to all PIIX4 models.
984 	 *
985 	 * Also, make sure that all interrupts cause a "Stop Break"
986 	 * event to exit from C2 state.
987 	 * Also, BRLD_EN_BM (ACPI_BITREG_BUS_MASTER_RLD in ACPI-speak)
988 	 * should be set to zero, otherwise it causes C2 to short-sleep.
989 	 * PIIX4 doesn't properly support C3 and bus master activity
990 	 * need not break out of C2.
991 	 */
992 	case PCI_REVISION_A_STEP:
993 	case PCI_REVISION_B_STEP:
994 	case PCI_REVISION_4E:
995 	case PCI_REVISION_4M:
996 	    cpu_quirks |= CPU_QUIRK_NO_C3;
997 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
998 		"acpi_cpu: working around PIIX4 bug, disabling C3\n"));
999 
1000 	    val = pci_read_config(acpi_dev, PIIX4_DEVACTB_REG, 4);
1001 	    if ((val & PIIX4_STOP_BREAK_MASK) != PIIX4_STOP_BREAK_MASK) {
1002 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1003 		    "acpi_cpu: PIIX4: enabling IRQs to generate Stop Break\n"));
1004 	    	val |= PIIX4_STOP_BREAK_MASK;
1005 		pci_write_config(acpi_dev, PIIX4_DEVACTB_REG, val, 4);
1006 	    }
1007 	    AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_RLD, &val);
1008 	    if (val) {
1009 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1010 		    "acpi_cpu: PIIX4: reset BRLD_EN_BM\n"));
1011 		AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1012 	    }
1013 	    break;
1014 	default:
1015 	    break;
1016 	}
1017     }
1018 
1019     return (0);
1020 }
1021 
1022 static int
1023 acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS)
1024 {
1025     struct acpi_cpu_softc *sc;
1026     struct sbuf	 sb;
1027     char	 buf[128];
1028     int		 i;
1029     uintmax_t	 fract, sum, whole;
1030 
1031     sc = (struct acpi_cpu_softc *) arg1;
1032     sum = 0;
1033     for (i = 0; i < sc->cpu_cx_count; i++)
1034 	sum += sc->cpu_cx_stats[i];
1035     sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1036     for (i = 0; i < sc->cpu_cx_count; i++) {
1037 	if (sum > 0) {
1038 	    whole = (uintmax_t)sc->cpu_cx_stats[i] * 100;
1039 	    fract = (whole % sum) * 100;
1040 	    sbuf_printf(&sb, "%u.%02u%% ", (u_int)(whole / sum),
1041 		(u_int)(fract / sum));
1042 	} else
1043 	    sbuf_printf(&sb, "0.00%% ");
1044     }
1045     sbuf_printf(&sb, "last %dus", sc->cpu_prev_sleep);
1046     sbuf_trim(&sb);
1047     sbuf_finish(&sb);
1048     sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1049     sbuf_delete(&sb);
1050 
1051     return (0);
1052 }
1053 
1054 static int
1055 acpi_cpu_set_cx_lowest_oncpu(struct acpi_cpu_softc *sc, int val)
1056 {
1057     int old_lowest, error = 0;
1058     uint32_t old_type, type;
1059 
1060     KKASSERT(mycpuid == sc->cpu_id);
1061 
1062     sc->cpu_cx_lowest_req = val;
1063     if (val > sc->cpu_cx_count - 1)
1064 	val = sc->cpu_cx_count - 1;
1065     old_lowest = atomic_swap_int(&sc->cpu_cx_lowest, val);
1066 
1067     old_type = sc->cpu_cx_states[old_lowest].type;
1068     type = sc->cpu_cx_states[val].type;
1069     if (old_type >= ACPI_STATE_C3 && type < ACPI_STATE_C3) {
1070 	KKASSERT(cpu_c3_ncpus > 0);
1071 	if (atomic_fetchadd_int(&cpu_c3_ncpus, -1) == 1) {
1072 	    /*
1073 	     * All of the CPUs exit C3 state, use a better
1074 	     * one shot timer.
1075 	     */
1076 	    error = cputimer_intr_select_caps(CPUTIMER_INTR_CAP_NONE);
1077 	    KKASSERT(!error || error == ERESTART);
1078 	    if (error == ERESTART) {
1079 		if (bootverbose)
1080 		    kprintf("exit C3, restart intr cputimer\n");
1081 		cputimer_intr_restart();
1082 	    }
1083     	}
1084     } else if (type >= ACPI_STATE_C3 && old_type < ACPI_STATE_C3) {
1085 	if (atomic_fetchadd_int(&cpu_c3_ncpus, 1) == 0) {
1086 	    /*
1087 	     * When the first CPU enters C3(+) state, switch
1088 	     * to an one shot timer, which could handle
1089 	     * C3(+) state, i.e. the timer will not hang.
1090 	     */
1091 	    error = cputimer_intr_select_caps(CPUTIMER_INTR_CAP_PS);
1092 	    if (error == ERESTART) {
1093 		if (bootverbose)
1094 		    kprintf("enter C3, restart intr cputimer\n");
1095 		cputimer_intr_restart();
1096 	    } else if (error) {
1097 		kprintf("no suitable intr cputimer found\n");
1098 
1099 		/* Restore */
1100 		sc->cpu_cx_lowest = old_lowest;
1101 		atomic_fetchadd_int(&cpu_c3_ncpus, -1);
1102 	    }
1103 	}
1104     }
1105 
1106     if (error)
1107 	return error;
1108 
1109     /* Cache the new lowest non-C3 state. */
1110     acpi_cpu_cx_non_c3(sc);
1111 
1112     /* Reset the statistics counters. */
1113     bzero(sc->cpu_cx_stats, sizeof(sc->cpu_cx_stats));
1114     return (0);
1115 }
1116 
1117 static void
1118 acpi_cst_set_lowest_handler(netmsg_t msg)
1119 {
1120     struct netmsg_acpi_cst *rmsg = (struct netmsg_acpi_cst *)msg;
1121     int error;
1122 
1123     error = acpi_cpu_set_cx_lowest_oncpu(rmsg->sc, rmsg->val);
1124     lwkt_replymsg(&rmsg->base.lmsg, error);
1125 }
1126 
1127 static int
1128 acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc, int val)
1129 {
1130     struct netmsg_acpi_cst msg;
1131 
1132     netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
1133 	acpi_cst_set_lowest_handler);
1134     msg.sc = sc;
1135     msg.val = val;
1136 
1137     return lwkt_domsg(netisr_cpuport(sc->cpu_id), &msg.base.lmsg, 0);
1138 }
1139 
1140 static int
1141 acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1142 {
1143     struct	 acpi_cpu_softc *sc;
1144     char	 state[8];
1145     int		 val, error;
1146 
1147     sc = (struct acpi_cpu_softc *)arg1;
1148     ksnprintf(state, sizeof(state), "C%d", sc->cpu_cx_lowest_req + 1);
1149     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1150     if (error != 0 || req->newptr == NULL)
1151 	return (error);
1152     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1153 	return (EINVAL);
1154     val = (int) strtol(state + 1, NULL, 10) - 1;
1155     if (val < 0)
1156 	return (EINVAL);
1157 
1158     lwkt_serialize_enter(&cpu_cx_slize);
1159     error = acpi_cpu_set_cx_lowest(sc, val);
1160     lwkt_serialize_exit(&cpu_cx_slize);
1161 
1162     return error;
1163 }
1164 
1165 static int
1166 acpi_cpu_cx_lowest_use_sysctl(SYSCTL_HANDLER_ARGS)
1167 {
1168     struct	 acpi_cpu_softc *sc;
1169     char	 state[8];
1170 
1171     sc = (struct acpi_cpu_softc *)arg1;
1172     ksnprintf(state, sizeof(state), "C%d", sc->cpu_cx_lowest + 1);
1173     return sysctl_handle_string(oidp, state, sizeof(state), req);
1174 }
1175 
1176 static int
1177 acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1178 {
1179     struct	acpi_cpu_softc *sc;
1180     char	state[8];
1181     int		val, error, i;
1182 
1183     ksnprintf(state, sizeof(state), "C%d", cpu_cx_lowest_req + 1);
1184     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1185     if (error != 0 || req->newptr == NULL)
1186 	return (error);
1187     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1188 	return (EINVAL);
1189     val = (int) strtol(state + 1, NULL, 10) - 1;
1190     if (val < 0)
1191 	return (EINVAL);
1192 
1193     lwkt_serialize_enter(&cpu_cx_slize);
1194 
1195     cpu_cx_lowest_req = val;
1196     cpu_cx_lowest = val;
1197     if (cpu_cx_lowest > cpu_cx_count - 1)
1198 	cpu_cx_lowest = cpu_cx_count - 1;
1199 
1200     /* Update the new lowest useable Cx state for all CPUs. */
1201     for (i = 0; i < cpu_ndevices; i++) {
1202 	sc = device_get_softc(cpu_devices[i]);
1203 	error = acpi_cpu_set_cx_lowest(sc, val);
1204 	if (error) {
1205 	    KKASSERT(i == 0);
1206 	    break;
1207 	}
1208     }
1209 
1210     lwkt_serialize_exit(&cpu_cx_slize);
1211 
1212     return error;
1213 }
1214 
1215 static int
1216 acpi_cpu_global_cx_lowest_use_sysctl(SYSCTL_HANDLER_ARGS)
1217 {
1218     char	state[8];
1219 
1220     ksnprintf(state, sizeof(state), "C%d", cpu_cx_lowest + 1);
1221     return sysctl_handle_string(oidp, state, sizeof(state), req);
1222 }
1223 
1224 /*
1225  * Put the CPU in C1 in a machine-dependant way.
1226  * XXX: shouldn't be here!
1227  */
1228 static void
1229 acpi_cpu_c1(void)
1230 {
1231 #ifdef __ia64__
1232     ia64_call_pal_static(PAL_HALT_LIGHT, 0, 0, 0);
1233 #else
1234     splz();
1235     if ((mycpu->gd_reqflags & RQF_IDLECHECK_WK_MASK) == 0)
1236         __asm __volatile("sti; hlt");
1237     else
1238         __asm __volatile("sti; pause");
1239 #endif /* !__ia64__ */
1240 }
1241 
1242 static void
1243 acpi_cpu_cx_non_c3(struct acpi_cpu_softc *sc)
1244 {
1245     int i;
1246 
1247     sc->cpu_non_c3 = 0;
1248     for (i = sc->cpu_cx_lowest; i >= 0; i--) {
1249 	if (sc->cpu_cx_states[i].type < ACPI_STATE_C3) {
1250 	    sc->cpu_non_c3 = i;
1251 	    break;
1252 	}
1253     }
1254     if (bootverbose)
1255 	device_printf(sc->cpu_dev, "non-C3 %d\n", sc->cpu_non_c3);
1256 }
1257 
1258 /*
1259  * Update the largest Cx state supported in the global cpu_cx_count.
1260  * It will be used in the global Cx sysctl handler.
1261  */
1262 static void
1263 acpi_cpu_global_cx_count(void)
1264 {
1265     struct acpi_cpu_softc *sc;
1266     int i;
1267 
1268     if (cpu_ndevices == 0) {
1269 	cpu_cx_count = 0;
1270 	return;
1271     }
1272 
1273     sc = device_get_softc(cpu_devices[0]);
1274     cpu_cx_count = sc->cpu_cx_count;
1275 
1276     for (i = 1; i < cpu_ndevices; i++) {
1277 	struct acpi_cpu_softc *sc = device_get_softc(cpu_devices[i]);
1278 
1279 	if (sc->cpu_cx_count < cpu_cx_count)
1280 	    cpu_cx_count = sc->cpu_cx_count;
1281     }
1282     if (bootverbose)
1283 	kprintf("cpu_cst: global Cx count %d\n", cpu_cx_count);
1284 }
1285