xref: /dflybsd-src/sys/dev/acpica/acpi_cpu_cstate.c (revision 48aba498f4b01b8d7a6eca64ab084dc06d29f34d)
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 #include <sys/microtime_pcpu.h>
43 
44 #include <bus/pci/pcivar.h>
45 #include <machine/atomic.h>
46 #include <machine/globaldata.h>
47 #include <machine/md_var.h>
48 #include <machine/smp.h>
49 #include <sys/rman.h>
50 
51 #include <net/netisr2.h>
52 #include <net/netmsg2.h>
53 #include <net/if_var.h>
54 
55 #include "acpi.h"
56 #include "acpivar.h"
57 #include "acpi_cpu.h"
58 
59 /*
60  * Support for ACPI Processor devices, including C[1-3] sleep states.
61  */
62 
63 /* Hooks for the ACPI CA debugging infrastructure */
64 #define _COMPONENT	ACPI_PROCESSOR
65 ACPI_MODULE_NAME("PROCESSOR")
66 
67 struct netmsg_acpi_cst {
68 	struct netmsg_base base;
69 	struct acpi_cst_softc *sc;
70 	int		val;
71 };
72 
73 struct acpi_cx {
74     struct resource	*p_lvlx;	/* Register to read to enter state. */
75     int			rid;		/* rid of p_lvlx */
76     uint32_t		type;		/* C1-3 (C4 and up treated as C3). */
77     uint32_t		trans_lat;	/* Transition latency (usec). */
78     uint32_t		power;		/* Power consumed (mW). */
79     int			res_type;	/* Resource type for p_lvlx. */
80     bus_space_tag_t	btag;
81     bus_space_handle_t	bhand;
82 };
83 #define MAX_CX_STATES	 8
84 
85 struct acpi_cst_softc {
86     device_t		cst_dev;
87     struct acpi_cpux_softc *cst_parent;
88     ACPI_HANDLE		cst_handle;
89     int			cst_cpuid;
90     uint32_t		cst_flags;	/* ACPI_CST_FLAG_ */
91     uint32_t		cst_p_blk;	/* ACPI P_BLK location */
92     uint32_t		cst_p_blk_len;	/* P_BLK length (must be 6). */
93     struct acpi_cx	cst_cx_states[MAX_CX_STATES];
94     int			cst_cx_count;	/* Number of valid Cx states. */
95     int			cst_prev_sleep;	/* Last idle sleep duration. */
96     /* Runtime state. */
97     int			cst_non_c3;	/* Index of lowest non-C3 state. */
98     u_long		cst_cx_stats[MAX_CX_STATES];/* Cx usage history. */
99     /* Values for sysctl. */
100     int			cst_cx_lowest;	/* Current Cx lowest */
101     int			cst_cx_lowest_req; /* Requested Cx lowest */
102     char 		cst_cx_supported[64];
103 };
104 
105 #define ACPI_CST_FLAG_PROBING	0x1
106 
107 #define ACPI_CST_ENTER_IO(cx)	bus_space_read_1((cx)->btag, (cx)->bhand, 0)
108 
109 #define CPU_QUIRK_NO_C3		(1<<0)	/* C3-type states are not usable. */
110 #define CPU_QUIRK_NO_BM_CTRL	(1<<2)	/* No bus mastering control. */
111 
112 #define PCI_VENDOR_INTEL	0x8086
113 #define PCI_DEVICE_82371AB_3	0x7113	/* PIIX4 chipset for quirks. */
114 #define PCI_REVISION_A_STEP	0
115 #define PCI_REVISION_B_STEP	1
116 #define PCI_REVISION_4E		2
117 #define PCI_REVISION_4M		3
118 #define PIIX4_DEVACTB_REG	0x58
119 #define PIIX4_BRLD_EN_IRQ0	(1<<0)
120 #define PIIX4_BRLD_EN_IRQ	(1<<1)
121 #define PIIX4_BRLD_EN_IRQ8	(1<<5)
122 #define PIIX4_STOP_BREAK_MASK	(PIIX4_BRLD_EN_IRQ0 | PIIX4_BRLD_EN_IRQ | PIIX4_BRLD_EN_IRQ8)
123 #define PIIX4_PCNTRL_BST_EN	(1<<10)
124 
125 /* Platform hardware resource information. */
126 static uint32_t		 cpu_smi_cmd;	/* Value to write to SMI_CMD. */
127 static uint8_t		 cpu_cst_cnt;	/* Indicate we are _CST aware. */
128 static int		 cpu_quirks;	/* Indicate any hardware bugs. */
129 
130 /* Runtime state. */
131 static int		 cpu_disable_idle; /* Disable entry to idle function */
132 static int		 cpu_cx_count;	/* Number of valid Cx states */
133 
134 /* Values for sysctl. */
135 static int		 cpu_cx_generic;
136 static int		 cpu_cx_lowest; /* Current Cx lowest */
137 static int		 cpu_cx_lowest_req; /* Requested Cx lowest */
138 static struct lwkt_serialize cpu_cx_slize = LWKT_SERIALIZE_INITIALIZER;
139 
140 /* C3 state transition */
141 static int		 cpu_c3_ncpus;
142 
143 static device_t		*cpu_devices;
144 static int		 cpu_ndevices;
145 static struct acpi_cst_softc **cpu_softc;
146 
147 static int	acpi_cst_probe(device_t dev);
148 static int	acpi_cst_attach(device_t dev);
149 static int	acpi_cst_suspend(device_t dev);
150 static int	acpi_cst_resume(device_t dev);
151 static int	acpi_cst_shutdown(device_t dev);
152 
153 static void	acpi_cpu_cx_probe(struct acpi_cst_softc *sc);
154 static void	acpi_cpu_generic_cx_probe(struct acpi_cst_softc *sc);
155 static int	acpi_cpu_cx_cst(struct acpi_cst_softc *sc);
156 static int	acpi_cpu_cx_cst_dispatch(struct acpi_cst_softc *sc);
157 static void	acpi_cpu_startup(void *arg);
158 static void	acpi_cpu_startup_cx(struct acpi_cst_softc *sc);
159 static void	acpi_cpu_cx_list(struct acpi_cst_softc *sc);
160 static void	acpi_cpu_idle(void);
161 static void	acpi_cpu_cst_notify(device_t);
162 static int	acpi_cpu_quirks(void);
163 static int	acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS);
164 static int	acpi_cpu_set_cx_lowest(struct acpi_cst_softc *, int);
165 static int	acpi_cpu_set_cx_lowest_oncpu(struct acpi_cst_softc *, int);
166 static int	acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
167 static int	acpi_cpu_cx_lowest_use_sysctl(SYSCTL_HANDLER_ARGS);
168 static int	acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
169 static int	acpi_cpu_global_cx_lowest_use_sysctl(SYSCTL_HANDLER_ARGS);
170 static void	acpi_cpu_cx_non_c3(struct acpi_cst_softc *sc);
171 static void	acpi_cpu_global_cx_count(void);
172 
173 static void	acpi_cpu_c1(void);	/* XXX */
174 
175 static device_method_t acpi_cst_methods[] = {
176     /* Device interface */
177     DEVMETHOD(device_probe,	acpi_cst_probe),
178     DEVMETHOD(device_attach,	acpi_cst_attach),
179     DEVMETHOD(device_detach,	bus_generic_detach),
180     DEVMETHOD(device_shutdown,	acpi_cst_shutdown),
181     DEVMETHOD(device_suspend,	acpi_cst_suspend),
182     DEVMETHOD(device_resume,	acpi_cst_resume),
183 
184     /* Bus interface */
185     DEVMETHOD(bus_add_child,	bus_generic_add_child),
186     DEVMETHOD(bus_read_ivar,	bus_generic_read_ivar),
187     DEVMETHOD(bus_get_resource_list, bus_generic_get_resource_list),
188     DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
189     DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
190     DEVMETHOD(bus_alloc_resource, bus_generic_rl_alloc_resource),
191     DEVMETHOD(bus_release_resource, bus_generic_rl_release_resource),
192     DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
193     DEVMETHOD(bus_activate_resource, bus_generic_activate_resource),
194     DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource),
195     DEVMETHOD(bus_setup_intr,	bus_generic_setup_intr),
196     DEVMETHOD(bus_teardown_intr, bus_generic_teardown_intr),
197     DEVMETHOD_END
198 };
199 
200 static driver_t acpi_cst_driver = {
201     "cpu_cst",
202     acpi_cst_methods,
203     sizeof(struct acpi_cst_softc),
204 };
205 
206 static devclass_t acpi_cst_devclass;
207 DRIVER_MODULE(cpu_cst, cpu, acpi_cst_driver, acpi_cst_devclass, NULL, NULL);
208 MODULE_DEPEND(cpu_cst, acpi, 1, 1, 1);
209 
210 static int
211 acpi_cst_probe(device_t dev)
212 {
213     int cpu_id;
214 
215     if (acpi_disabled("cpu_cst") || acpi_get_type(dev) != ACPI_TYPE_PROCESSOR)
216 	return (ENXIO);
217 
218     cpu_id = acpi_get_magic(dev);
219 
220     if (cpu_softc == NULL)
221 	cpu_softc = kmalloc(sizeof(struct acpi_cst_softc *) *
222 	    SMP_MAXCPU, M_TEMP /* XXX */, M_INTWAIT | M_ZERO);
223 
224     /*
225      * Check if we already probed this processor.  We scan the bus twice
226      * so it's possible we've already seen this one.
227      */
228     if (cpu_softc[cpu_id] != NULL) {
229 	device_printf(dev, "CPU%d cstate already exist\n", cpu_id);
230 	return (ENXIO);
231     }
232 
233     /* Mark this processor as in-use and save our derived id for attach. */
234     cpu_softc[cpu_id] = (void *)1;
235     device_set_desc(dev, "ACPI CPU C-State");
236 
237     return (0);
238 }
239 
240 static int
241 acpi_cst_attach(device_t dev)
242 {
243     ACPI_BUFFER		   buf;
244     ACPI_OBJECT		   *obj;
245     struct acpi_cst_softc *sc;
246     ACPI_STATUS		   status;
247 
248     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
249 
250     sc = device_get_softc(dev);
251     sc->cst_dev = dev;
252     sc->cst_parent = device_get_softc(device_get_parent(dev));
253     sc->cst_handle = acpi_get_handle(dev);
254     sc->cst_cpuid = acpi_get_magic(dev);
255     cpu_softc[sc->cst_cpuid] = sc;
256     cpu_smi_cmd = AcpiGbl_FADT.SmiCommand;
257     cpu_cst_cnt = AcpiGbl_FADT.CstControl;
258 
259     buf.Pointer = NULL;
260     buf.Length = ACPI_ALLOCATE_BUFFER;
261     status = AcpiEvaluateObject(sc->cst_handle, NULL, NULL, &buf);
262     if (ACPI_FAILURE(status)) {
263 	device_printf(dev, "attach failed to get Processor obj - %s\n",
264 		      AcpiFormatException(status));
265 	return (ENXIO);
266     }
267     obj = (ACPI_OBJECT *)buf.Pointer;
268     sc->cst_p_blk = obj->Processor.PblkAddress;
269     sc->cst_p_blk_len = obj->Processor.PblkLength;
270     AcpiOsFree(obj);
271     ACPI_DEBUG_PRINT((ACPI_DB_INFO, "acpi_cpu%d: P_BLK at %#x/%d\n",
272 		     device_get_unit(dev), sc->cst_p_blk, sc->cst_p_blk_len));
273 
274     /*
275      * If this is the first cpu we attach, create and initialize the generic
276      * resources that will be used by all acpi cpu devices.
277      */
278     if (device_get_unit(dev) == 0) {
279 	/* Assume we won't be using generic Cx mode by default */
280 	cpu_cx_generic = FALSE;
281 
282 	/* Queue post cpu-probing task handler */
283 	AcpiOsExecute(OSL_NOTIFY_HANDLER, acpi_cpu_startup, NULL);
284     }
285 
286     /* Probe for Cx state support. */
287     acpi_cpu_cx_probe(sc);
288 
289     /* Finally,  call identify and probe/attach for child devices. */
290     bus_generic_probe(dev);
291     bus_generic_attach(dev);
292 
293     return (0);
294 }
295 
296 /*
297  * Disable any entry to the idle function during suspend and re-enable it
298  * during resume.
299  */
300 static int
301 acpi_cst_suspend(device_t dev)
302 {
303     int error;
304 
305     error = bus_generic_suspend(dev);
306     if (error)
307 	return (error);
308     cpu_disable_idle = TRUE;
309     return (0);
310 }
311 
312 static int
313 acpi_cst_resume(device_t dev)
314 {
315 
316     cpu_disable_idle = FALSE;
317     return (bus_generic_resume(dev));
318 }
319 
320 static int
321 acpi_cst_shutdown(device_t dev)
322 {
323     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
324 
325     /* Allow children to shutdown first. */
326     bus_generic_shutdown(dev);
327 
328     /*
329      * Disable any entry to the idle function.  There is a small race where
330      * an idle thread have passed this check but not gone to sleep.  This
331      * is ok since device_shutdown() does not free the softc, otherwise
332      * we'd have to be sure all threads were evicted before returning.
333      */
334     cpu_disable_idle = TRUE;
335 
336     return_VALUE (0);
337 }
338 
339 static void
340 acpi_cpu_cx_probe(struct acpi_cst_softc *sc)
341 {
342     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
343 
344     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
345     sc->cst_prev_sleep = 1000000;
346     sc->cst_cx_lowest = 0;
347     sc->cst_cx_lowest_req = 0;
348 
349     /*
350      * Check for the ACPI 2.0 _CST sleep states object. If we can't find
351      * any, we'll revert to generic FADT/P_BLK Cx control method which will
352      * be handled by acpi_cpu_startup. We need to defer to after having
353      * probed all the cpus in the system before probing for generic Cx
354      * states as we may already have found cpus with valid _CST packages
355      */
356     if (!cpu_cx_generic && acpi_cpu_cx_cst(sc) != 0) {
357 	/*
358 	 * We were unable to find a _CST package for this cpu or there
359 	 * was an error parsing it. Switch back to generic mode.
360 	 */
361 	cpu_cx_generic = TRUE;
362 	if (bootverbose)
363 	    device_printf(sc->cst_dev, "switching to generic Cx mode\n");
364     }
365 
366     /*
367      * TODO: _CSD Package should be checked here.
368      */
369 }
370 
371 static void
372 acpi_cpu_generic_cx_probe(struct acpi_cst_softc *sc)
373 {
374     ACPI_GENERIC_ADDRESS	 gas;
375     struct acpi_cx		*cx_ptr;
376 
377     sc->cst_cx_count = 0;
378     cx_ptr = sc->cst_cx_states;
379 
380     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
381     sc->cst_prev_sleep = 1000000;
382 
383     /* C1 has been required since just after ACPI 1.0 */
384     cx_ptr->type = ACPI_STATE_C1;
385     cx_ptr->trans_lat = 0;
386     cx_ptr++;
387     sc->cst_cx_count++;
388 
389     /* C2(+) is not supported on MP system */
390     if (ncpus > 1 && (AcpiGbl_FADT.Flags & ACPI_FADT_C2_MP_SUPPORTED) == 0)
391 	return;
392 
393     /*
394      * The spec says P_BLK must be 6 bytes long.  However, some systems
395      * use it to indicate a fractional set of features present so we
396      * take 5 as C2.  Some may also have a value of 7 to indicate
397      * another C3 but most use _CST for this (as required) and having
398      * "only" C1-C3 is not a hardship.
399      */
400     if (sc->cst_p_blk_len < 5)
401 	return;
402 
403     /* Validate and allocate resources for C2 (P_LVL2). */
404     gas.SpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
405     gas.BitWidth = 8;
406     if (AcpiGbl_FADT.C2Latency <= 100) {
407 	gas.Address = sc->cst_p_blk + 4;
408 
409 	cx_ptr->rid = sc->cst_parent->cpux_next_rid;
410 	acpi_bus_alloc_gas(sc->cst_dev, &cx_ptr->type, &cx_ptr->rid, &gas,
411 	    &cx_ptr->p_lvlx, RF_SHAREABLE);
412 	if (cx_ptr->p_lvlx != NULL) {
413 	    sc->cst_parent->cpux_next_rid++;
414 	    cx_ptr->type = ACPI_STATE_C2;
415 	    cx_ptr->trans_lat = AcpiGbl_FADT.C2Latency;
416 	    cx_ptr->btag = rman_get_bustag(cx_ptr->p_lvlx);
417 	    cx_ptr->bhand = rman_get_bushandle(cx_ptr->p_lvlx);
418 	    cx_ptr++;
419 	    sc->cst_cx_count++;
420 	    sc->cst_non_c3 = 1;
421 	}
422     }
423     if (sc->cst_p_blk_len < 6)
424 	return;
425 
426     /* Validate and allocate resources for C3 (P_LVL3). */
427     if (AcpiGbl_FADT.C3Latency <= 1000 && !(cpu_quirks & CPU_QUIRK_NO_C3)) {
428 	gas.Address = sc->cst_p_blk + 5;
429 
430 	cx_ptr->rid = sc->cst_parent->cpux_next_rid;
431 	acpi_bus_alloc_gas(sc->cst_dev, &cx_ptr->type, &cx_ptr->rid, &gas,
432 	    &cx_ptr->p_lvlx, RF_SHAREABLE);
433 	if (cx_ptr->p_lvlx != NULL) {
434 	    sc->cst_parent->cpux_next_rid++;
435 	    cx_ptr->type = ACPI_STATE_C3;
436 	    cx_ptr->trans_lat = AcpiGbl_FADT.C3Latency;
437 	    cx_ptr->btag = rman_get_bustag(cx_ptr->p_lvlx);
438 	    cx_ptr->bhand = rman_get_bushandle(cx_ptr->p_lvlx);
439 	    cx_ptr++;
440 	    sc->cst_cx_count++;
441 	}
442     }
443 }
444 
445 /*
446  * Parse a _CST package and set up its Cx states.  Since the _CST object
447  * can change dynamically, our notify handler may call this function
448  * to clean up and probe the new _CST package.
449  */
450 static int
451 acpi_cpu_cx_cst(struct acpi_cst_softc *sc)
452 {
453     struct	 acpi_cx *cx_ptr;
454     ACPI_STATUS	 status;
455     ACPI_BUFFER	 buf;
456     ACPI_OBJECT	*top;
457     ACPI_OBJECT	*pkg;
458     uint32_t	 count;
459     int		 i;
460 
461     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
462 
463     buf.Pointer = NULL;
464     buf.Length = ACPI_ALLOCATE_BUFFER;
465     status = AcpiEvaluateObject(sc->cst_handle, "_CST", NULL, &buf);
466     if (ACPI_FAILURE(status))
467 	return (ENXIO);
468 
469     /* _CST is a package with a count and at least one Cx package. */
470     top = (ACPI_OBJECT *)buf.Pointer;
471     if (!ACPI_PKG_VALID(top, 2) || acpi_PkgInt32(top, 0, &count) != 0) {
472 	device_printf(sc->cst_dev, "invalid _CST package\n");
473 	AcpiOsFree(buf.Pointer);
474 	return (ENXIO);
475     }
476     if (count != top->Package.Count - 1) {
477 	device_printf(sc->cst_dev, "invalid _CST state count (%d != %d)\n",
478 	       count, top->Package.Count - 1);
479 	count = top->Package.Count - 1;
480     }
481     if (count > MAX_CX_STATES) {
482 	device_printf(sc->cst_dev, "_CST has too many states (%d)\n", count);
483 	count = MAX_CX_STATES;
484     }
485 
486     sc->cst_flags |= ACPI_CST_FLAG_PROBING;
487     cpu_sfence();
488 
489     for (i = 0; i < sc->cst_cx_count; ++i) {
490 	cx_ptr = &sc->cst_cx_states[i];
491 
492 	/* Free up any previous register. */
493 	if (cx_ptr->p_lvlx != NULL) {
494 	    bus_release_resource(sc->cst_dev, cx_ptr->res_type, cx_ptr->rid,
495 	        cx_ptr->p_lvlx);
496 	    cx_ptr->p_lvlx = NULL;
497 	}
498     }
499 
500     /* Set up all valid states. */
501     sc->cst_cx_count = 0;
502     cx_ptr = sc->cst_cx_states;
503     for (i = 0; i < count; i++) {
504 	pkg = &top->Package.Elements[i + 1];
505 	if (!ACPI_PKG_VALID(pkg, 4) ||
506 	    acpi_PkgInt32(pkg, 1, &cx_ptr->type) != 0 ||
507 	    acpi_PkgInt32(pkg, 2, &cx_ptr->trans_lat) != 0 ||
508 	    acpi_PkgInt32(pkg, 3, &cx_ptr->power) != 0) {
509 
510 	    device_printf(sc->cst_dev, "skipping invalid Cx state package\n");
511 	    continue;
512 	}
513 
514 	/* Validate the state to see if we should use it. */
515 	switch (cx_ptr->type) {
516 	case ACPI_STATE_C1:
517 	    sc->cst_non_c3 = i;
518 	    cx_ptr++;
519 	    sc->cst_cx_count++;
520 	    continue;
521 	case ACPI_STATE_C2:
522 	    sc->cst_non_c3 = i;
523 	    break;
524 	case ACPI_STATE_C3:
525 	default:
526 	    if ((cpu_quirks & CPU_QUIRK_NO_C3) != 0) {
527 
528 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
529 				 "acpi_cpu%d: C3[%d] not available.\n",
530 				 device_get_unit(sc->cst_dev), i));
531 		continue;
532 	    }
533 	    break;
534 	}
535 
536 	/* Allocate the control register for C2 or C3. */
537 	KASSERT(cx_ptr->p_lvlx == NULL, ("still has lvlx"));
538 	cx_ptr->rid = sc->cst_parent->cpux_next_rid;
539 	acpi_PkgGas(sc->cst_dev, pkg, 0, &cx_ptr->res_type, &cx_ptr->rid,
540 	    &cx_ptr->p_lvlx, RF_SHAREABLE);
541 	if (cx_ptr->p_lvlx != NULL) {
542 	    sc->cst_parent->cpux_next_rid++;
543 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
544 			     "acpi_cpu%d: Got C%d - %d latency\n",
545 			     device_get_unit(sc->cst_dev), cx_ptr->type,
546 			     cx_ptr->trans_lat));
547 	    cx_ptr->btag = rman_get_bustag(cx_ptr->p_lvlx);
548 	    cx_ptr->bhand = rman_get_bushandle(cx_ptr->p_lvlx);
549 	    cx_ptr++;
550 	    sc->cst_cx_count++;
551 	}
552     }
553     AcpiOsFree(buf.Pointer);
554 
555     /*
556      * Fix up the lowest Cx being used
557      */
558     if (sc->cst_cx_lowest_req < sc->cst_cx_count)
559 	sc->cst_cx_lowest = sc->cst_cx_lowest_req;
560     if (sc->cst_cx_lowest > sc->cst_cx_count - 1)
561 	sc->cst_cx_lowest = sc->cst_cx_count - 1;
562 
563     /*
564      * Cache the lowest non-C3 state.
565      * NOTE: must after cst_cx_lowest is set.
566      */
567     acpi_cpu_cx_non_c3(sc);
568 
569     cpu_sfence();
570     sc->cst_flags &= ~ACPI_CST_FLAG_PROBING;
571 
572     return (0);
573 }
574 
575 static void
576 acpi_cst_probe_handler(netmsg_t msg)
577 {
578     struct netmsg_acpi_cst *rmsg = (struct netmsg_acpi_cst *)msg;
579     int error;
580 
581     error = acpi_cpu_cx_cst(rmsg->sc);
582     lwkt_replymsg(&rmsg->base.lmsg, error);
583 }
584 
585 static int
586 acpi_cpu_cx_cst_dispatch(struct acpi_cst_softc *sc)
587 {
588     struct netmsg_acpi_cst msg;
589 
590     netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
591 	acpi_cst_probe_handler);
592     msg.sc = sc;
593 
594     return lwkt_domsg(netisr_cpuport(sc->cst_cpuid), &msg.base.lmsg, 0);
595 }
596 
597 /*
598  * Call this *after* all CPUs have been attached.
599  */
600 static void
601 acpi_cpu_startup(void *arg)
602 {
603     struct acpi_cst_softc *sc;
604     int i;
605 
606     /* Get set of CPU devices */
607     devclass_get_devices(acpi_cst_devclass, &cpu_devices, &cpu_ndevices);
608 
609     /*
610      * Setup any quirks that might necessary now that we have probed
611      * all the CPUs
612      */
613     acpi_cpu_quirks();
614 
615     if (cpu_cx_generic) {
616 	/*
617 	 * We are using generic Cx mode, probe for available Cx states
618 	 * for all processors.
619 	 */
620 	for (i = 0; i < cpu_ndevices; i++) {
621 	    sc = device_get_softc(cpu_devices[i]);
622 	    acpi_cpu_generic_cx_probe(sc);
623 	}
624     } else {
625 	/*
626 	 * We are using _CST mode, remove C3 state if necessary.
627 	 *
628 	 * As we now know for sure that we will be using _CST mode
629 	 * install our notify handler.
630 	 */
631 	for (i = 0; i < cpu_ndevices; i++) {
632 	    sc = device_get_softc(cpu_devices[i]);
633 	    if (cpu_quirks & CPU_QUIRK_NO_C3)
634 		sc->cst_cx_count = sc->cst_non_c3 + 1;
635 	    sc->cst_parent->cpux_cst_notify = acpi_cpu_cst_notify;
636 	}
637     }
638     acpi_cpu_global_cx_count();
639 
640     /* Perform Cx final initialization. */
641     for (i = 0; i < cpu_ndevices; i++) {
642 	sc = device_get_softc(cpu_devices[i]);
643 	acpi_cpu_startup_cx(sc);
644 
645 	if (sc->cst_parent->glob_sysctl_tree != NULL) {
646 	    struct acpi_cpux_softc *cpux = sc->cst_parent;
647 
648 	    /* Add a sysctl handler to handle global Cx lowest setting */
649 	    SYSCTL_ADD_PROC(&cpux->glob_sysctl_ctx,
650 	    		    SYSCTL_CHILDREN(cpux->glob_sysctl_tree),
651 			    OID_AUTO, "cx_lowest",
652 			    CTLTYPE_STRING | CTLFLAG_RW, NULL, 0,
653 			    acpi_cpu_global_cx_lowest_sysctl, "A",
654 			    "Requested global lowest Cx sleep state");
655 	    SYSCTL_ADD_PROC(&cpux->glob_sysctl_ctx,
656 	    		    SYSCTL_CHILDREN(cpux->glob_sysctl_tree),
657 			    OID_AUTO, "cx_lowest_use",
658 			    CTLTYPE_STRING | CTLFLAG_RD, NULL, 0,
659 			    acpi_cpu_global_cx_lowest_use_sysctl, "A",
660 			    "Global lowest Cx sleep state to use");
661 	}
662     }
663 
664     /* Take over idling from cpu_idle_default(). */
665     cpu_cx_lowest = 0;
666     cpu_cx_lowest_req = 0;
667     cpu_disable_idle = FALSE;
668     cpu_idle_hook = acpi_cpu_idle;
669 }
670 
671 static void
672 acpi_cpu_cx_list(struct acpi_cst_softc *sc)
673 {
674     struct sbuf sb;
675     int i;
676 
677     /*
678      * Set up the list of Cx states
679      */
680     sbuf_new(&sb, sc->cst_cx_supported, sizeof(sc->cst_cx_supported),
681 	SBUF_FIXEDLEN);
682     for (i = 0; i < sc->cst_cx_count; i++)
683 	sbuf_printf(&sb, "C%d/%d ", i + 1, sc->cst_cx_states[i].trans_lat);
684     sbuf_trim(&sb);
685     sbuf_finish(&sb);
686 }
687 
688 static void
689 acpi_cpu_startup_cx(struct acpi_cst_softc *sc)
690 {
691     struct acpi_cpux_softc *cpux = sc->cst_parent;
692 
693     acpi_cpu_cx_list(sc);
694 
695     SYSCTL_ADD_STRING(&cpux->pcpu_sysctl_ctx,
696 		      SYSCTL_CHILDREN(cpux->pcpu_sysctl_tree),
697 		      OID_AUTO, "cx_supported", CTLFLAG_RD,
698 		      sc->cst_cx_supported, 0,
699 		      "Cx/microsecond values for supported Cx states");
700     SYSCTL_ADD_PROC(&cpux->pcpu_sysctl_ctx,
701 		    SYSCTL_CHILDREN(cpux->pcpu_sysctl_tree),
702 		    OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
703 		    (void *)sc, 0, acpi_cpu_cx_lowest_sysctl, "A",
704 		    "requested lowest Cx sleep state");
705     SYSCTL_ADD_PROC(&cpux->pcpu_sysctl_ctx,
706 		    SYSCTL_CHILDREN(cpux->pcpu_sysctl_tree),
707 		    OID_AUTO, "cx_lowest_use", CTLTYPE_STRING | CTLFLAG_RD,
708 		    (void *)sc, 0, acpi_cpu_cx_lowest_use_sysctl, "A",
709 		    "lowest Cx sleep state to use");
710     SYSCTL_ADD_PROC(&cpux->pcpu_sysctl_ctx,
711 		    SYSCTL_CHILDREN(cpux->pcpu_sysctl_tree),
712 		    OID_AUTO, "cx_usage", CTLTYPE_STRING | CTLFLAG_RD,
713 		    (void *)sc, 0, acpi_cpu_usage_sysctl, "A",
714 		    "percent usage for each Cx state");
715 
716 #ifdef notyet
717     /* Signal platform that we can handle _CST notification. */
718     if (!cpu_cx_generic && cpu_cst_cnt != 0) {
719 	ACPI_LOCK(acpi);
720 	AcpiOsWritePort(cpu_smi_cmd, cpu_cst_cnt, 8);
721 	ACPI_UNLOCK(acpi);
722     }
723 #endif
724 }
725 
726 /*
727  * Idle the CPU in the lowest state possible.  This function is called with
728  * interrupts disabled.  Note that once it re-enables interrupts, a task
729  * switch can occur so do not access shared data (i.e. the softc) after
730  * interrupts are re-enabled.
731  */
732 static void
733 acpi_cpu_idle(void)
734 {
735     struct	acpi_cst_softc *sc;
736     struct	acpi_cx *cx_next;
737     union microtime_pcpu start, end;
738     uint64_t	dummy;
739     int		bm_active, cx_next_idx, i, tdiff;
740 
741     /* If disabled, return immediately. */
742     if (cpu_disable_idle) {
743 	ACPI_ENABLE_IRQS();
744 	return;
745     }
746 
747     /*
748      * Look up our CPU id to get our softc.  If it's NULL, we'll use C1
749      * since there is no ACPI processor object for this CPU.  This occurs
750      * for logical CPUs in the HTT case.
751      */
752     sc = cpu_softc[mdcpu->mi.gd_cpuid];
753     if (sc == NULL) {
754 	acpi_cpu_c1();
755 	return;
756     }
757 
758     /* Still probing; use C1 */
759     if (sc->cst_flags & ACPI_CST_FLAG_PROBING) {
760 	acpi_cpu_c1();
761 	return;
762     }
763 
764     /* Find the lowest state that has small enough latency. */
765     cx_next_idx = 0;
766     for (i = sc->cst_cx_lowest; i >= 0; i--) {
767 	if (sc->cst_cx_states[i].trans_lat * 3 <= sc->cst_prev_sleep) {
768 	    cx_next_idx = i;
769 	    break;
770 	}
771     }
772 
773     /*
774      * If C3(+) is to be entered, check for bus master activity.
775      * If there was activity, clear the bit and use the lowest
776      * non-C3 state.
777      */
778     cx_next = &sc->cst_cx_states[cx_next_idx];
779     if (cx_next->type >= ACPI_STATE_C3 &&
780         (cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
781 	AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, &bm_active);
782 	if (bm_active != 0) {
783 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, 1);
784 	    cx_next_idx = min(cx_next_idx, sc->cst_non_c3);
785 	}
786     }
787 
788     /* Select the next state and update statistics. */
789     cx_next = &sc->cst_cx_states[cx_next_idx];
790     sc->cst_cx_stats[cx_next_idx]++;
791     KASSERT(cx_next->type != ACPI_STATE_C0, ("acpi_cpu_idle: C0 sleep"));
792 
793     /*
794      * Execute HLT (or equivalent) and wait for an interrupt.  We can't
795      * calculate the time spent in C1 since the place we wake up is an
796      * ISR.  Assume we slept half of quantum and return.
797      */
798     if (cx_next->type == ACPI_STATE_C1) {
799 	sc->cst_prev_sleep = (sc->cst_prev_sleep * 3 + 500000 / hz) / 4;
800 	acpi_cpu_c1();
801 	return;
802     }
803 
804     /*
805      * For C3(+), disable bus master arbitration and enable bus master wake
806      * if BM control is available, otherwise flush the CPU cache.
807      */
808     if (cx_next->type >= ACPI_STATE_C3) {
809 	if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
810 	    AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 1);
811 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
812 	} else
813 	    ACPI_FLUSH_CPU_CACHE();
814     }
815 
816     /*
817      * Read from P_LVLx to enter C2(+), checking time spent asleep.
818      */
819     microtime_pcpu_get(&start);
820     cpu_mfence();
821 
822     ACPI_CST_ENTER_IO(cx_next);
823     /*
824      * Perform a dummy I/O read.  Since it may take an arbitrary time
825      * to enter the idle state, this read makes sure that we are frozen.
826      */
827     AcpiRead(&dummy, &AcpiGbl_FADT.XPmTimerBlock);
828 
829     cpu_mfence();
830     microtime_pcpu_get(&end);
831 
832     /* Enable bus master arbitration and disable bus master wakeup. */
833     if (cx_next->type >= ACPI_STATE_C3) {
834 	if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
835 	    AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 0);
836 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
837 	}
838     }
839     ACPI_ENABLE_IRQS();
840 
841     /* Find the actual time asleep in microseconds. */
842     tdiff = microtime_pcpu_diff(&start, &end);
843     sc->cst_prev_sleep = (sc->cst_prev_sleep * 3 + tdiff) / 4;
844 }
845 
846 /*
847  * Re-evaluate the _CST object when we are notified that it changed.
848  */
849 static void
850 acpi_cpu_cst_notify(device_t dev)
851 {
852     struct acpi_cst_softc *sc = device_get_softc(dev);
853 
854     KASSERT(curthread->td_type != TD_TYPE_NETISR,
855         ("notify in netisr%d", mycpuid));
856 
857     lwkt_serialize_enter(&cpu_cx_slize);
858 
859     /* Update the list of Cx states. */
860     acpi_cpu_cx_cst_dispatch(sc);
861     acpi_cpu_cx_list(sc);
862 
863     /* Update the new lowest useable Cx state for all CPUs. */
864     acpi_cpu_global_cx_count();
865 
866     /*
867      * Fix up the lowest Cx being used
868      */
869     if (cpu_cx_lowest_req < cpu_cx_count)
870 	cpu_cx_lowest = cpu_cx_lowest_req;
871     if (cpu_cx_lowest > cpu_cx_count - 1)
872 	cpu_cx_lowest = cpu_cx_count - 1;
873 
874     lwkt_serialize_exit(&cpu_cx_slize);
875 }
876 
877 static int
878 acpi_cpu_quirks(void)
879 {
880     device_t acpi_dev;
881     uint32_t val;
882 
883     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
884 
885     /*
886      * Bus mastering arbitration control is needed to keep caches coherent
887      * while sleeping in C3.  If it's not present but a working flush cache
888      * instruction is present, flush the caches before entering C3 instead.
889      * Otherwise, just disable C3 completely.
890      */
891     if (AcpiGbl_FADT.Pm2ControlBlock == 0 ||
892 	AcpiGbl_FADT.Pm2ControlLength == 0) {
893 	if ((AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD) &&
894 	    (AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD_FLUSH) == 0) {
895 	    cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
896 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
897 		"acpi_cpu: no BM control, using flush cache method\n"));
898 	} else {
899 	    cpu_quirks |= CPU_QUIRK_NO_C3;
900 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
901 		"acpi_cpu: no BM control, C3 not available\n"));
902 	}
903     }
904 
905     /*
906      * If we are using generic Cx mode, C3 on multiple CPUs requires using
907      * the expensive flush cache instruction.
908      */
909     if (cpu_cx_generic && ncpus > 1) {
910 	cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
911 	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
912 	    "acpi_cpu: SMP, using flush cache mode for C3\n"));
913     }
914 
915     /* Look for various quirks of the PIIX4 part. */
916     acpi_dev = pci_find_device(PCI_VENDOR_INTEL, PCI_DEVICE_82371AB_3);
917     if (acpi_dev != NULL) {
918 	switch (pci_get_revid(acpi_dev)) {
919 	/*
920 	 * Disable C3 support for all PIIX4 chipsets.  Some of these parts
921 	 * do not report the BMIDE status to the BM status register and
922 	 * others have a livelock bug if Type-F DMA is enabled.  Linux
923 	 * works around the BMIDE bug by reading the BM status directly
924 	 * but we take the simpler approach of disabling C3 for these
925 	 * parts.
926 	 *
927 	 * See erratum #18 ("C3 Power State/BMIDE and Type-F DMA
928 	 * Livelock") from the January 2002 PIIX4 specification update.
929 	 * Applies to all PIIX4 models.
930 	 *
931 	 * Also, make sure that all interrupts cause a "Stop Break"
932 	 * event to exit from C2 state.
933 	 * Also, BRLD_EN_BM (ACPI_BITREG_BUS_MASTER_RLD in ACPI-speak)
934 	 * should be set to zero, otherwise it causes C2 to short-sleep.
935 	 * PIIX4 doesn't properly support C3 and bus master activity
936 	 * need not break out of C2.
937 	 */
938 	case PCI_REVISION_A_STEP:
939 	case PCI_REVISION_B_STEP:
940 	case PCI_REVISION_4E:
941 	case PCI_REVISION_4M:
942 	    cpu_quirks |= CPU_QUIRK_NO_C3;
943 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
944 		"acpi_cpu: working around PIIX4 bug, disabling C3\n"));
945 
946 	    val = pci_read_config(acpi_dev, PIIX4_DEVACTB_REG, 4);
947 	    if ((val & PIIX4_STOP_BREAK_MASK) != PIIX4_STOP_BREAK_MASK) {
948 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
949 		    "acpi_cpu: PIIX4: enabling IRQs to generate Stop Break\n"));
950 	    	val |= PIIX4_STOP_BREAK_MASK;
951 		pci_write_config(acpi_dev, PIIX4_DEVACTB_REG, val, 4);
952 	    }
953 	    AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_RLD, &val);
954 	    if (val) {
955 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
956 		    "acpi_cpu: PIIX4: reset BRLD_EN_BM\n"));
957 		AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
958 	    }
959 	    break;
960 	default:
961 	    break;
962 	}
963     }
964 
965     return (0);
966 }
967 
968 static int
969 acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS)
970 {
971     struct acpi_cst_softc *sc;
972     struct sbuf	 sb;
973     char	 buf[128];
974     int		 i;
975     uintmax_t	 fract, sum, whole;
976 
977     sc = (struct acpi_cst_softc *) arg1;
978     sum = 0;
979     for (i = 0; i < sc->cst_cx_count; i++)
980 	sum += sc->cst_cx_stats[i];
981     sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
982     for (i = 0; i < sc->cst_cx_count; i++) {
983 	if (sum > 0) {
984 	    whole = (uintmax_t)sc->cst_cx_stats[i] * 100;
985 	    fract = (whole % sum) * 100;
986 	    sbuf_printf(&sb, "%u.%02u%% ", (u_int)(whole / sum),
987 		(u_int)(fract / sum));
988 	} else
989 	    sbuf_printf(&sb, "0.00%% ");
990     }
991     sbuf_printf(&sb, "last %dus", sc->cst_prev_sleep);
992     sbuf_trim(&sb);
993     sbuf_finish(&sb);
994     sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
995     sbuf_delete(&sb);
996 
997     return (0);
998 }
999 
1000 static int
1001 acpi_cpu_set_cx_lowest_oncpu(struct acpi_cst_softc *sc, int val)
1002 {
1003     int old_lowest, error = 0, old_lowest_req;
1004     uint32_t old_type, type;
1005 
1006     KKASSERT(mycpuid == sc->cst_cpuid);
1007 
1008     old_lowest_req = sc->cst_cx_lowest_req;
1009     sc->cst_cx_lowest_req = val;
1010 
1011     if (val > sc->cst_cx_count - 1)
1012 	val = sc->cst_cx_count - 1;
1013     old_lowest = atomic_swap_int(&sc->cst_cx_lowest, val);
1014 
1015     old_type = sc->cst_cx_states[old_lowest].type;
1016     type = sc->cst_cx_states[val].type;
1017     if (old_type >= ACPI_STATE_C3 && type < ACPI_STATE_C3) {
1018 	KKASSERT(cpu_c3_ncpus > 0);
1019 	if (atomic_fetchadd_int(&cpu_c3_ncpus, -1) == 1) {
1020 	    /*
1021 	     * All of the CPUs exit C3 state, use a better
1022 	     * one shot timer.
1023 	     */
1024 	    error = cputimer_intr_select_caps(CPUTIMER_INTR_CAP_NONE);
1025 	    KKASSERT(!error || error == ERESTART);
1026 	    if (error == ERESTART) {
1027 		if (bootverbose)
1028 		    kprintf("exit C3, restart intr cputimer\n");
1029 		cputimer_intr_restart();
1030 	    }
1031     	}
1032     } else if (type >= ACPI_STATE_C3 && old_type < ACPI_STATE_C3) {
1033 	if (atomic_fetchadd_int(&cpu_c3_ncpus, 1) == 0) {
1034 	    /*
1035 	     * When the first CPU enters C3(+) state, switch
1036 	     * to an one shot timer, which could handle
1037 	     * C3(+) state, i.e. the timer will not hang.
1038 	     */
1039 	    error = cputimer_intr_select_caps(CPUTIMER_INTR_CAP_PS);
1040 	    if (error == ERESTART) {
1041 		if (bootverbose)
1042 		    kprintf("enter C3, restart intr cputimer\n");
1043 		cputimer_intr_restart();
1044 	    } else if (error) {
1045 		kprintf("no suitable intr cputimer found\n");
1046 
1047 		/* Restore */
1048 		sc->cst_cx_lowest_req = old_lowest_req;
1049 		sc->cst_cx_lowest = old_lowest;
1050 		atomic_fetchadd_int(&cpu_c3_ncpus, -1);
1051 	    }
1052 	}
1053     }
1054 
1055     if (error)
1056 	return error;
1057 
1058     /* Cache the new lowest non-C3 state. */
1059     acpi_cpu_cx_non_c3(sc);
1060 
1061     /* Reset the statistics counters. */
1062     bzero(sc->cst_cx_stats, sizeof(sc->cst_cx_stats));
1063     return (0);
1064 }
1065 
1066 static void
1067 acpi_cst_set_lowest_handler(netmsg_t msg)
1068 {
1069     struct netmsg_acpi_cst *rmsg = (struct netmsg_acpi_cst *)msg;
1070     int error;
1071 
1072     error = acpi_cpu_set_cx_lowest_oncpu(rmsg->sc, rmsg->val);
1073     lwkt_replymsg(&rmsg->base.lmsg, error);
1074 }
1075 
1076 static int
1077 acpi_cpu_set_cx_lowest(struct acpi_cst_softc *sc, int val)
1078 {
1079     struct netmsg_acpi_cst msg;
1080 
1081     netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
1082 	acpi_cst_set_lowest_handler);
1083     msg.sc = sc;
1084     msg.val = val;
1085 
1086     return lwkt_domsg(netisr_cpuport(sc->cst_cpuid), &msg.base.lmsg, 0);
1087 }
1088 
1089 static int
1090 acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1091 {
1092     struct	 acpi_cst_softc *sc;
1093     char	 state[8];
1094     int		 val, error;
1095 
1096     sc = (struct acpi_cst_softc *)arg1;
1097     ksnprintf(state, sizeof(state), "C%d", sc->cst_cx_lowest_req + 1);
1098     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1099     if (error != 0 || req->newptr == NULL)
1100 	return (error);
1101     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1102 	return (EINVAL);
1103     val = (int) strtol(state + 1, NULL, 10) - 1;
1104     if (val < 0)
1105 	return (EINVAL);
1106 
1107     lwkt_serialize_enter(&cpu_cx_slize);
1108     error = acpi_cpu_set_cx_lowest(sc, val);
1109     lwkt_serialize_exit(&cpu_cx_slize);
1110 
1111     return error;
1112 }
1113 
1114 static int
1115 acpi_cpu_cx_lowest_use_sysctl(SYSCTL_HANDLER_ARGS)
1116 {
1117     struct	 acpi_cst_softc *sc;
1118     char	 state[8];
1119 
1120     sc = (struct acpi_cst_softc *)arg1;
1121     ksnprintf(state, sizeof(state), "C%d", sc->cst_cx_lowest + 1);
1122     return sysctl_handle_string(oidp, state, sizeof(state), req);
1123 }
1124 
1125 static int
1126 acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1127 {
1128     struct	acpi_cst_softc *sc;
1129     char	state[8];
1130     int		val, error, i;
1131 
1132     ksnprintf(state, sizeof(state), "C%d", cpu_cx_lowest_req + 1);
1133     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1134     if (error != 0 || req->newptr == NULL)
1135 	return (error);
1136     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1137 	return (EINVAL);
1138     val = (int) strtol(state + 1, NULL, 10) - 1;
1139     if (val < 0)
1140 	return (EINVAL);
1141 
1142     lwkt_serialize_enter(&cpu_cx_slize);
1143 
1144     cpu_cx_lowest_req = val;
1145     cpu_cx_lowest = val;
1146     if (cpu_cx_lowest > cpu_cx_count - 1)
1147 	cpu_cx_lowest = cpu_cx_count - 1;
1148 
1149     /* Update the new lowest useable Cx state for all CPUs. */
1150     for (i = 0; i < cpu_ndevices; i++) {
1151 	sc = device_get_softc(cpu_devices[i]);
1152 	error = acpi_cpu_set_cx_lowest(sc, val);
1153 	if (error) {
1154 	    KKASSERT(i == 0);
1155 	    break;
1156 	}
1157     }
1158 
1159     lwkt_serialize_exit(&cpu_cx_slize);
1160 
1161     return error;
1162 }
1163 
1164 static int
1165 acpi_cpu_global_cx_lowest_use_sysctl(SYSCTL_HANDLER_ARGS)
1166 {
1167     char	state[8];
1168 
1169     ksnprintf(state, sizeof(state), "C%d", cpu_cx_lowest + 1);
1170     return sysctl_handle_string(oidp, state, sizeof(state), req);
1171 }
1172 
1173 /*
1174  * Put the CPU in C1 in a machine-dependant way.
1175  * XXX: shouldn't be here!
1176  */
1177 static void
1178 acpi_cpu_c1(void)
1179 {
1180 #ifdef __ia64__
1181     ia64_call_pal_static(PAL_HALT_LIGHT, 0, 0, 0);
1182 #else
1183     splz();
1184     if ((mycpu->gd_reqflags & RQF_IDLECHECK_WK_MASK) == 0)
1185         __asm __volatile("sti; hlt");
1186     else
1187         __asm __volatile("sti; pause");
1188 #endif /* !__ia64__ */
1189 }
1190 
1191 static void
1192 acpi_cpu_cx_non_c3(struct acpi_cst_softc *sc)
1193 {
1194     int i;
1195 
1196     sc->cst_non_c3 = 0;
1197     for (i = sc->cst_cx_lowest; i >= 0; i--) {
1198 	if (sc->cst_cx_states[i].type < ACPI_STATE_C3) {
1199 	    sc->cst_non_c3 = i;
1200 	    break;
1201 	}
1202     }
1203     if (bootverbose)
1204 	device_printf(sc->cst_dev, "non-C3 %d\n", sc->cst_non_c3);
1205 }
1206 
1207 /*
1208  * Update the largest Cx state supported in the global cpu_cx_count.
1209  * It will be used in the global Cx sysctl handler.
1210  */
1211 static void
1212 acpi_cpu_global_cx_count(void)
1213 {
1214     struct acpi_cst_softc *sc;
1215     int i;
1216 
1217     if (cpu_ndevices == 0) {
1218 	cpu_cx_count = 0;
1219 	return;
1220     }
1221 
1222     sc = device_get_softc(cpu_devices[0]);
1223     cpu_cx_count = sc->cst_cx_count;
1224 
1225     for (i = 1; i < cpu_ndevices; i++) {
1226 	struct acpi_cst_softc *sc = device_get_softc(cpu_devices[i]);
1227 
1228 	if (sc->cst_cx_count < cpu_cx_count)
1229 	    cpu_cx_count = sc->cst_cx_count;
1230     }
1231     if (bootverbose)
1232 	kprintf("cpu_cst: global Cx count %d\n", cpu_cx_count);
1233 }
1234