xref: /dragonfly/sys/dev/acpica/acpi_cpu_cstate.c (revision 896f2e3a)
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 #include <sys/cpu_topology.h>
44 
45 #include <bus/pci/pcivar.h>
46 #include <machine/atomic.h>
47 #include <machine/globaldata.h>
48 #include <machine/md_var.h>
49 #include <machine/smp.h>
50 #include <sys/rman.h>
51 
52 #include <net/netisr2.h>
53 #include <net/netmsg2.h>
54 #include <net/if_var.h>
55 
56 #include "acpi.h"
57 #include "acpivar.h"
58 #include "acpi_cpu.h"
59 #include "acpi_cpu_cstate.h"
60 
61 /*
62  * Support for ACPI Processor devices, including C[1-3+] sleep states.
63  */
64 
65 /* Hooks for the ACPICA debugging infrastructure */
66 #define _COMPONENT	ACPI_PROCESSOR
67 ACPI_MODULE_NAME("PROCESSOR")
68 
69 struct netmsg_acpi_cst {
70 	struct netmsg_base base;
71 	struct acpi_cst_softc *sc;
72 	int		val;
73 };
74 
75 #define MAX_CX_STATES	 8
76 
77 struct acpi_cst_softc {
78     device_t		cst_dev;
79     struct acpi_cpu_softc *cst_parent;
80     ACPI_HANDLE		cst_handle;
81     int			cst_cpuid;
82     uint32_t		cst_flags;	/* ACPI_CST_FLAG_ */
83     uint32_t		cst_p_blk;	/* ACPI P_BLK location */
84     uint32_t		cst_p_blk_len;	/* P_BLK length (must be 6). */
85     struct acpi_cst_cx	cst_cx_states[MAX_CX_STATES];
86     int			cst_cx_count;	/* Number of valid Cx states. */
87     int			cst_prev_sleep;	/* Last idle sleep duration. */
88     /* Runtime state. */
89     int			cst_non_c3;	/* Index of lowest non-C3 state. */
90     u_long		cst_cx_stats[MAX_CX_STATES];/* Cx usage history. */
91     /* Values for sysctl. */
92     int			cst_cx_lowest;	/* Current Cx lowest */
93     int			cst_cx_lowest_req; /* Requested Cx lowest */
94     char 		cst_cx_supported[64];
95 };
96 
97 #define ACPI_CST_FLAG_PROBING	0x1
98 #define ACPI_CST_FLAG_ATTACHED	0x2
99 /* Match C-states of other hyperthreads on the same core */
100 #define ACPI_CST_FLAG_MATCH_HT	0x4
101 
102 #define PCI_VENDOR_INTEL	0x8086
103 #define PCI_DEVICE_82371AB_3	0x7113	/* PIIX4 chipset for quirks. */
104 #define PCI_REVISION_A_STEP	0
105 #define PCI_REVISION_B_STEP	1
106 #define PCI_REVISION_4E		2
107 #define PCI_REVISION_4M		3
108 #define PIIX4_DEVACTB_REG	0x58
109 #define PIIX4_BRLD_EN_IRQ0	(1<<0)
110 #define PIIX4_BRLD_EN_IRQ	(1<<1)
111 #define PIIX4_BRLD_EN_IRQ8	(1<<5)
112 #define PIIX4_STOP_BREAK_MASK	(PIIX4_BRLD_EN_IRQ0 | \
113 				 PIIX4_BRLD_EN_IRQ | \
114 				 PIIX4_BRLD_EN_IRQ8)
115 #define PIIX4_PCNTRL_BST_EN	(1<<10)
116 
117 /* Platform hardware resource information. */
118 static uint32_t		 acpi_cst_smi_cmd; /* Value to write to SMI_CMD. */
119 static uint8_t		 acpi_cst_ctrl;	/* Indicate we are _CST aware. */
120 int		 	 acpi_cst_quirks; /* Indicate any hardware bugs. */
121 static boolean_t	 acpi_cst_use_fadt;
122 
123 /* Runtime state. */
124 static boolean_t	 acpi_cst_disable_idle;
125 					/* Disable entry to idle function */
126 static int		 acpi_cst_cx_count; /* Number of valid Cx states */
127 
128 /* Values for sysctl. */
129 static int		 acpi_cst_cx_lowest; /* Current Cx lowest */
130 static int		 acpi_cst_cx_lowest_req; /* Requested Cx lowest */
131 
132 static device_t		*acpi_cst_devices;
133 static int		 acpi_cst_ndevices;
134 static struct acpi_cst_softc **acpi_cst_softc;
135 static struct lwkt_serialize acpi_cst_slize = LWKT_SERIALIZE_INITIALIZER;
136 
137 static int	acpi_cst_probe(device_t);
138 static int	acpi_cst_attach(device_t);
139 static int	acpi_cst_suspend(device_t);
140 static int	acpi_cst_resume(device_t);
141 static int	acpi_cst_shutdown(device_t);
142 
143 static void	acpi_cst_notify(device_t);
144 static void	acpi_cst_postattach(void *);
145 static void	acpi_cst_idle(void);
146 static void	acpi_cst_copy(struct acpi_cst_softc *,
147 		    const struct acpi_cst_softc *);
148 
149 static void	acpi_cst_cx_probe(struct acpi_cst_softc *);
150 static void	acpi_cst_cx_probe_fadt(struct acpi_cst_softc *);
151 static int	acpi_cst_cx_probe_cst(struct acpi_cst_softc *, int);
152 static int	acpi_cst_cx_reprobe_cst(struct acpi_cst_softc *);
153 
154 static void	acpi_cst_startup(struct acpi_cst_softc *);
155 static void	acpi_cst_support_list(struct acpi_cst_softc *);
156 static int	acpi_cst_set_lowest(struct acpi_cst_softc *, int);
157 static int	acpi_cst_set_lowest_oncpu(struct acpi_cst_softc *, int);
158 static void	acpi_cst_non_c3(struct acpi_cst_softc *);
159 static void	acpi_cst_global_cx_count(void);
160 static int	acpi_cst_set_quirks(void);
161 static void	acpi_cst_c3_bm_rld(struct acpi_cst_softc *);
162 static void	acpi_cst_free_resource(struct acpi_cst_softc *, int);
163 static void	acpi_cst_c1_halt(void);
164 
165 static int	acpi_cst_usage_sysctl(SYSCTL_HANDLER_ARGS);
166 static int	acpi_cst_lowest_sysctl(SYSCTL_HANDLER_ARGS);
167 static int	acpi_cst_lowest_use_sysctl(SYSCTL_HANDLER_ARGS);
168 static int	acpi_cst_global_lowest_sysctl(SYSCTL_HANDLER_ARGS);
169 static int	acpi_cst_global_lowest_use_sysctl(SYSCTL_HANDLER_ARGS);
170 
171 static int	acpi_cst_cx_setup(struct acpi_cst_cx *cx);
172 static void	acpi_cst_c1_halt_enter(const struct acpi_cst_cx *);
173 static void	acpi_cst_cx_io_enter(const struct acpi_cst_cx *);
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 (acpi_cst_softc == NULL)
221 	acpi_cst_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 (acpi_cst_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     acpi_cst_softc[cpu_id] = device_get_softc(dev);
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     acpi_cst_softc[sc->cst_cpuid] = sc;
256     acpi_cst_smi_cmd = AcpiGbl_FADT.SmiCommand;
257     acpi_cst_ctrl = 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 	acpi_cst_softc[sc->cst_cpuid] = NULL;
266 	return (ENXIO);
267     }
268     obj = (ACPI_OBJECT *)buf.Pointer;
269     sc->cst_p_blk = obj->Processor.PblkAddress;
270     sc->cst_p_blk_len = obj->Processor.PblkLength;
271     AcpiOsFree(obj);
272     ACPI_DEBUG_PRINT((ACPI_DB_INFO, "cpu_cst%d: P_BLK at %#x/%d\n",
273 		     device_get_unit(dev), sc->cst_p_blk, sc->cst_p_blk_len));
274 
275     /*
276      * If this is the first cpu we attach, create and initialize the generic
277      * resources that will be used by all acpi cpu devices.
278      */
279     if (device_get_unit(dev) == 0) {
280 	/* Assume we won't be using FADT for Cx states by default */
281 	acpi_cst_use_fadt = FALSE;
282 
283 	/* Queue post cpu-probing task handler */
284 	AcpiOsExecute(OSL_NOTIFY_HANDLER, acpi_cst_postattach, NULL);
285     }
286 
287     /* Probe for Cx state support. */
288     acpi_cst_cx_probe(sc);
289 
290     sc->cst_flags |= ACPI_CST_FLAG_ATTACHED;
291 
292     return (0);
293 }
294 
295 /*
296  * Disable any entry to the idle function during suspend and re-enable it
297  * during resume.
298  */
299 static int
300 acpi_cst_suspend(device_t dev)
301 {
302     int error;
303 
304     error = bus_generic_suspend(dev);
305     if (error)
306 	return (error);
307     acpi_cst_disable_idle = TRUE;
308     return (0);
309 }
310 
311 static int
312 acpi_cst_resume(device_t dev)
313 {
314     acpi_cst_disable_idle = FALSE;
315     return (bus_generic_resume(dev));
316 }
317 
318 static int
319 acpi_cst_shutdown(device_t dev)
320 {
321     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
322 
323     /* Allow children to shutdown first. */
324     bus_generic_shutdown(dev);
325 
326     /*
327      * Disable any entry to the idle function.  There is a small race where
328      * an idle thread have passed this check but not gone to sleep.  This
329      * is ok since device_shutdown() does not free the softc, otherwise
330      * we'd have to be sure all threads were evicted before returning.
331      */
332     acpi_cst_disable_idle = TRUE;
333 
334     return_VALUE (0);
335 }
336 
337 static void
338 acpi_cst_cx_probe(struct acpi_cst_softc *sc)
339 {
340     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
341 
342     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
343     sc->cst_prev_sleep = 1000000;
344     sc->cst_cx_lowest = 0;
345     sc->cst_cx_lowest_req = 0;
346 
347     /*
348      * Check for the ACPI 2.0 _CST sleep states object.  If we can't find
349      * any, we'll revert to FADT/P_BLK Cx control method which will be
350      * handled by acpi_cst_postattach.  We need to defer to after having
351      * probed all the cpus in the system before probing for Cx states from
352      * FADT as we may already have found cpus with valid _CST packages.
353      */
354     if (!acpi_cst_use_fadt && acpi_cst_cx_probe_cst(sc, 0) != 0) {
355 	/*
356 	 * We were unable to find a _CST package for this cpu or there
357 	 * was an error parsing it. Switch back to generic mode.
358 	 */
359 	acpi_cst_use_fadt = TRUE;
360 	if (bootverbose)
361 	    device_printf(sc->cst_dev, "switching to FADT Cx mode\n");
362     }
363 
364     /*
365      * TODO: _CSD Package should be checked here.
366      */
367 }
368 
369 static void
370 acpi_cst_cx_probe_fadt(struct acpi_cst_softc *sc)
371 {
372     struct acpi_cst_cx *cx_ptr;
373     int error;
374 
375     /*
376      * Free all previously allocated resources.
377      *
378      * NITE:
379      * It is needed, since we could enter here because of other
380      * cpu's _CST probing failure.
381      */
382     acpi_cst_free_resource(sc, 0);
383 
384     sc->cst_cx_count = 0;
385     cx_ptr = sc->cst_cx_states;
386 
387     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
388     sc->cst_prev_sleep = 1000000;
389 
390     /* C1 has been required since just after ACPI 1.0 */
391     cx_ptr->gas.SpaceId = ACPI_ADR_SPACE_FIXED_HARDWARE;
392     cx_ptr->type = ACPI_STATE_C1;
393     cx_ptr->trans_lat = 0;
394     cx_ptr->enter = acpi_cst_c1_halt_enter;
395     error = acpi_cst_cx_setup(cx_ptr);
396     if (error)
397 	panic("C1 FADT HALT setup failed: %d", error);
398     cx_ptr++;
399     sc->cst_cx_count++;
400 
401     /* C2(+) is not supported on MP system */
402     if (ncpus > 1 && (AcpiGbl_FADT.Flags & ACPI_FADT_C2_MP_SUPPORTED) == 0)
403 	return;
404 
405     /*
406      * The spec says P_BLK must be 6 bytes long.  However, some systems
407      * use it to indicate a fractional set of features present so we
408      * take 5 as C2.  Some may also have a value of 7 to indicate
409      * another C3 but most use _CST for this (as required) and having
410      * "only" C1-C3 is not a hardship.
411      */
412     if (sc->cst_p_blk_len < 5)
413 	return;
414 
415     /* Validate and allocate resources for C2 (P_LVL2). */
416     if (AcpiGbl_FADT.C2Latency <= 100) {
417 	cx_ptr->gas.SpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
418 	cx_ptr->gas.BitWidth = 8;
419 	cx_ptr->gas.Address = sc->cst_p_blk + 4;
420 
421 	cx_ptr->rid = sc->cst_parent->cpu_next_rid;
422 	acpi_bus_alloc_gas(sc->cst_dev, &cx_ptr->res_type, &cx_ptr->rid,
423 	    &cx_ptr->gas, &cx_ptr->res, RF_SHAREABLE);
424 	if (cx_ptr->res != NULL) {
425 	    sc->cst_parent->cpu_next_rid++;
426 	    cx_ptr->type = ACPI_STATE_C2;
427 	    cx_ptr->trans_lat = AcpiGbl_FADT.C2Latency;
428 	    cx_ptr->enter = acpi_cst_cx_io_enter;
429 	    cx_ptr->btag = rman_get_bustag(cx_ptr->res);
430 	    cx_ptr->bhand = rman_get_bushandle(cx_ptr->res);
431 	    error = acpi_cst_cx_setup(cx_ptr);
432 	    if (error)
433 		panic("C2 FADT I/O setup failed: %d", error);
434 	    cx_ptr++;
435 	    sc->cst_cx_count++;
436 	    sc->cst_non_c3 = 1;
437 	}
438     }
439     if (sc->cst_p_blk_len < 6)
440 	return;
441 
442     /* Validate and allocate resources for C3 (P_LVL3). */
443     if (AcpiGbl_FADT.C3Latency <= 1000 &&
444         !(acpi_cst_quirks & ACPI_CST_QUIRK_NO_C3)) {
445 	cx_ptr->gas.SpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
446 	cx_ptr->gas.BitWidth = 8;
447 	cx_ptr->gas.Address = sc->cst_p_blk + 5;
448 
449 	cx_ptr->rid = sc->cst_parent->cpu_next_rid;
450 	acpi_bus_alloc_gas(sc->cst_dev, &cx_ptr->res_type, &cx_ptr->rid,
451 	    &cx_ptr->gas, &cx_ptr->res, RF_SHAREABLE);
452 	if (cx_ptr->res != NULL) {
453 	    sc->cst_parent->cpu_next_rid++;
454 	    cx_ptr->type = ACPI_STATE_C3;
455 	    cx_ptr->trans_lat = AcpiGbl_FADT.C3Latency;
456 	    cx_ptr->enter = acpi_cst_cx_io_enter;
457 	    cx_ptr->btag = rman_get_bustag(cx_ptr->res);
458 	    cx_ptr->bhand = rman_get_bushandle(cx_ptr->res);
459 	    error = acpi_cst_cx_setup(cx_ptr);
460 	    if (error)
461 		panic("C3 FADT I/O setup failed: %d", error);
462 	    cx_ptr++;
463 	    sc->cst_cx_count++;
464 	}
465     }
466 }
467 
468 static void
469 acpi_cst_copy(struct acpi_cst_softc *dst_sc,
470     const struct acpi_cst_softc *src_sc)
471 {
472     dst_sc->cst_non_c3 = src_sc->cst_non_c3;
473     dst_sc->cst_cx_count = src_sc->cst_cx_count;
474     memcpy(dst_sc->cst_cx_states, src_sc->cst_cx_states,
475         sizeof(dst_sc->cst_cx_states));
476 }
477 
478 /*
479  * Parse a _CST package and set up its Cx states.  Since the _CST object
480  * can change dynamically, our notify handler may call this function
481  * to clean up and probe the new _CST package.
482  */
483 static int
484 acpi_cst_cx_probe_cst(struct acpi_cst_softc *sc, int reprobe)
485 {
486     struct	 acpi_cst_cx *cx_ptr;
487     ACPI_STATUS	 status;
488     ACPI_BUFFER	 buf;
489     ACPI_OBJECT	*top;
490     ACPI_OBJECT	*pkg;
491     uint32_t	 count;
492     int		 i;
493 
494     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
495 
496 #ifdef INVARIANTS
497     if (reprobe)
498 	KKASSERT(&curthread->td_msgport == netisr_cpuport(sc->cst_cpuid));
499 #endif
500 
501     buf.Pointer = NULL;
502     buf.Length = ACPI_ALLOCATE_BUFFER;
503     status = AcpiEvaluateObject(sc->cst_handle, "_CST", NULL, &buf);
504     if (ACPI_FAILURE(status))
505 	return (ENXIO);
506 
507     /* _CST is a package with a count and at least one Cx package. */
508     top = (ACPI_OBJECT *)buf.Pointer;
509     if (!ACPI_PKG_VALID(top, 2) || acpi_PkgInt32(top, 0, &count) != 0) {
510 	device_printf(sc->cst_dev, "invalid _CST package\n");
511 	AcpiOsFree(buf.Pointer);
512 	return (ENXIO);
513     }
514     if (count != top->Package.Count - 1) {
515 	device_printf(sc->cst_dev, "invalid _CST state count (%d != %d)\n",
516 	       count, top->Package.Count - 1);
517 	count = top->Package.Count - 1;
518     }
519     if (count > MAX_CX_STATES) {
520 	device_printf(sc->cst_dev, "_CST has too many states (%d)\n", count);
521 	count = MAX_CX_STATES;
522     }
523 
524     sc->cst_flags |= ACPI_CST_FLAG_PROBING | ACPI_CST_FLAG_MATCH_HT;
525     cpu_sfence();
526 
527     /*
528      * Free all previously allocated resources
529      *
530      * NOTE: It is needed for _CST reprobing.
531      */
532     acpi_cst_free_resource(sc, 0);
533 
534     /* Set up all valid states. */
535     sc->cst_cx_count = 0;
536     cx_ptr = sc->cst_cx_states;
537     for (i = 0; i < count; i++) {
538 	int error;
539 
540 	pkg = &top->Package.Elements[i + 1];
541 	if (!ACPI_PKG_VALID(pkg, 4) ||
542 	    acpi_PkgInt32(pkg, 1, &cx_ptr->type) != 0 ||
543 	    acpi_PkgInt32(pkg, 2, &cx_ptr->trans_lat) != 0 ||
544 	    acpi_PkgInt32(pkg, 3, &cx_ptr->power) != 0) {
545 
546 	    device_printf(sc->cst_dev, "skipping invalid Cx state package\n");
547 	    continue;
548 	}
549 
550 	/* Validate the state to see if we should use it. */
551 	switch (cx_ptr->type) {
552 	case ACPI_STATE_C1:
553 	    sc->cst_non_c3 = i;
554 	    cx_ptr->enter = acpi_cst_c1_halt_enter;
555 	    error = acpi_cst_cx_setup(cx_ptr);
556 	    if (error)
557 		panic("C1 CST HALT setup failed: %d", error);
558 	    if (sc->cst_cx_count != 0) {
559 		/*
560 		 * C1 is not the first C-state; something really stupid
561 		 * is going on ...
562 		 */
563 		sc->cst_flags &= ~ACPI_CST_FLAG_MATCH_HT;
564 	    }
565 	    cx_ptr++;
566 	    sc->cst_cx_count++;
567 	    continue;
568 	case ACPI_STATE_C2:
569 	    sc->cst_non_c3 = i;
570 	    break;
571 	case ACPI_STATE_C3:
572 	default:
573 	    if ((acpi_cst_quirks & ACPI_CST_QUIRK_NO_C3) != 0) {
574 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
575 				 "cpu_cst%d: C3[%d] not available.\n",
576 				 device_get_unit(sc->cst_dev), i));
577 		continue;
578 	    }
579 	    break;
580 	}
581 
582 	/*
583 	 * Allocate the control register for C2 or C3(+).
584 	 */
585 	KASSERT(cx_ptr->res == NULL, ("still has res"));
586 	acpi_PkgRawGas(pkg, 0, &cx_ptr->gas);
587 
588 	/*
589 	 * We match number of C2/C3 for hyperthreads, only if the
590 	 * register is "Fixed Hardware", e.g. on most of the Intel
591 	 * CPUs.  We don't have much to do for the rest of the
592 	 * register types.
593 	 */
594 	if (cx_ptr->gas.SpaceId != ACPI_ADR_SPACE_FIXED_HARDWARE)
595 	    sc->cst_flags &= ~ACPI_CST_FLAG_MATCH_HT;
596 
597 	cx_ptr->rid = sc->cst_parent->cpu_next_rid;
598 	acpi_bus_alloc_gas(sc->cst_dev, &cx_ptr->res_type, &cx_ptr->rid,
599 	    &cx_ptr->gas, &cx_ptr->res, RF_SHAREABLE);
600 	if (cx_ptr->res != NULL) {
601 	    sc->cst_parent->cpu_next_rid++;
602 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
603 			     "cpu_cst%d: Got C%d - %d latency\n",
604 			     device_get_unit(sc->cst_dev), cx_ptr->type,
605 			     cx_ptr->trans_lat));
606 	    cx_ptr->enter = acpi_cst_cx_io_enter;
607 	    cx_ptr->btag = rman_get_bustag(cx_ptr->res);
608 	    cx_ptr->bhand = rman_get_bushandle(cx_ptr->res);
609 	    error = acpi_cst_cx_setup(cx_ptr);
610 	    if (error)
611 		panic("C%d CST I/O setup failed: %d", cx_ptr->type, error);
612 	    cx_ptr++;
613 	    sc->cst_cx_count++;
614 	} else {
615 	    error = acpi_cst_cx_setup(cx_ptr);
616 	    if (!error) {
617 		KASSERT(cx_ptr->enter != NULL,
618 		    ("C%d enter is not set", cx_ptr->type));
619 		cx_ptr++;
620 		sc->cst_cx_count++;
621 	    }
622 	}
623     }
624     AcpiOsFree(buf.Pointer);
625 
626     if (sc->cst_flags & ACPI_CST_FLAG_MATCH_HT) {
627 	cpumask_t mask;
628 
629 	mask = get_cpumask_from_level(sc->cst_cpuid, CORE_LEVEL);
630 	if (CPUMASK_TESTNZERO(mask)) {
631 	    int cpu;
632 
633 	    for (cpu = 0; cpu < ncpus; ++cpu) {
634 		struct acpi_cst_softc *sc1 = acpi_cst_softc[cpu];
635 
636 		if (sc1 == NULL || sc1 == sc ||
637 		    (sc1->cst_flags & ACPI_CST_FLAG_ATTACHED) == 0 ||
638 		    (sc1->cst_flags & ACPI_CST_FLAG_MATCH_HT) == 0)
639 		    continue;
640 		if (!CPUMASK_TESTBIT(mask, sc1->cst_cpuid))
641 		    continue;
642 
643 		if (sc1->cst_cx_count != sc->cst_cx_count) {
644 		    struct acpi_cst_softc *src_sc, *dst_sc;
645 
646 		    if (bootverbose) {
647 			device_printf(sc->cst_dev,
648 			    "inconstent C-state count: %d, %s has %d\n",
649 			    sc->cst_cx_count,
650 			    device_get_nameunit(sc1->cst_dev),
651 			    sc1->cst_cx_count);
652 		    }
653 		    if (sc1->cst_cx_count > sc->cst_cx_count) {
654 			src_sc = sc1;
655 			dst_sc = sc;
656 		    } else {
657 			src_sc = sc;
658 			dst_sc = sc1;
659 		    }
660 		    acpi_cst_copy(dst_sc, src_sc);
661 		}
662 	    }
663 	}
664     }
665 
666     if (reprobe) {
667 	/* If there are C3(+) states, always enable bus master wakeup */
668 	if ((acpi_cst_quirks & ACPI_CST_QUIRK_NO_BM) == 0) {
669 	    for (i = 0; i < sc->cst_cx_count; ++i) {
670 		struct acpi_cst_cx *cx = &sc->cst_cx_states[i];
671 
672 		if (cx->type >= ACPI_STATE_C3) {
673 		    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
674 		    break;
675 		}
676 	    }
677 	}
678 
679 	/* Fix up the lowest Cx being used */
680 	acpi_cst_set_lowest_oncpu(sc, sc->cst_cx_lowest_req);
681     }
682 
683     /*
684      * Cache the lowest non-C3 state.
685      * NOTE: must after cst_cx_lowest is set.
686      */
687     acpi_cst_non_c3(sc);
688 
689     cpu_sfence();
690     sc->cst_flags &= ~ACPI_CST_FLAG_PROBING;
691 
692     return (0);
693 }
694 
695 static void
696 acpi_cst_cx_reprobe_cst_handler(netmsg_t msg)
697 {
698     struct netmsg_acpi_cst *rmsg = (struct netmsg_acpi_cst *)msg;
699     int error;
700 
701     error = acpi_cst_cx_probe_cst(rmsg->sc, 1);
702     lwkt_replymsg(&rmsg->base.lmsg, error);
703 }
704 
705 static int
706 acpi_cst_cx_reprobe_cst(struct acpi_cst_softc *sc)
707 {
708     struct netmsg_acpi_cst msg;
709 
710     netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
711 	acpi_cst_cx_reprobe_cst_handler);
712     msg.sc = sc;
713 
714     return lwkt_domsg(netisr_cpuport(sc->cst_cpuid), &msg.base.lmsg, 0);
715 }
716 
717 /*
718  * Call this *after* all CPUs Cx states have been attached.
719  */
720 static void
721 acpi_cst_postattach(void *arg)
722 {
723     struct acpi_cst_softc *sc;
724     int i;
725 
726     /* Get set of Cx state devices */
727     devclass_get_devices(acpi_cst_devclass, &acpi_cst_devices,
728 	&acpi_cst_ndevices);
729 
730     /*
731      * Setup any quirks that might necessary now that we have probed
732      * all the CPUs' Cx states.
733      */
734     acpi_cst_set_quirks();
735 
736     if (acpi_cst_use_fadt) {
737 	/*
738 	 * We are using Cx mode from FADT, probe for available Cx states
739 	 * for all processors.
740 	 */
741 	for (i = 0; i < acpi_cst_ndevices; i++) {
742 	    sc = device_get_softc(acpi_cst_devices[i]);
743 	    acpi_cst_cx_probe_fadt(sc);
744 	}
745     } else {
746 	/*
747 	 * We are using _CST mode, remove C3 state if necessary.
748 	 *
749 	 * As we now know for sure that we will be using _CST mode
750 	 * install our notify handler.
751 	 */
752 	for (i = 0; i < acpi_cst_ndevices; i++) {
753 	    sc = device_get_softc(acpi_cst_devices[i]);
754 	    if (acpi_cst_quirks & ACPI_CST_QUIRK_NO_C3) {
755 		/* Free part of unused resources */
756 		acpi_cst_free_resource(sc, sc->cst_non_c3 + 1);
757 		sc->cst_cx_count = sc->cst_non_c3 + 1;
758 	    }
759 	    sc->cst_parent->cpu_cst_notify = acpi_cst_notify;
760 	}
761     }
762     acpi_cst_global_cx_count();
763 
764     /* Perform Cx final initialization. */
765     for (i = 0; i < acpi_cst_ndevices; i++) {
766 	sc = device_get_softc(acpi_cst_devices[i]);
767 	acpi_cst_startup(sc);
768 
769 	if (sc->cst_parent->glob_sysctl_tree != NULL) {
770 	    struct acpi_cpu_softc *cpu = sc->cst_parent;
771 
772 	    /* Add a sysctl handler to handle global Cx lowest setting */
773 	    SYSCTL_ADD_PROC(&cpu->glob_sysctl_ctx,
774 	    		    SYSCTL_CHILDREN(cpu->glob_sysctl_tree),
775 			    OID_AUTO, "cx_lowest",
776 			    CTLTYPE_STRING | CTLFLAG_RW, NULL, 0,
777 			    acpi_cst_global_lowest_sysctl, "A",
778 			    "Requested global lowest Cx sleep state");
779 	    SYSCTL_ADD_PROC(&cpu->glob_sysctl_ctx,
780 	    		    SYSCTL_CHILDREN(cpu->glob_sysctl_tree),
781 			    OID_AUTO, "cx_lowest_use",
782 			    CTLTYPE_STRING | CTLFLAG_RD, NULL, 0,
783 			    acpi_cst_global_lowest_use_sysctl, "A",
784 			    "Global lowest Cx sleep state to use");
785 	}
786     }
787 
788     /* Take over idling from cpu_idle_default(). */
789     acpi_cst_cx_lowest = 0;
790     acpi_cst_cx_lowest_req = 0;
791     acpi_cst_disable_idle = FALSE;
792 
793     cpu_sfence();
794     cpu_idle_hook = acpi_cst_idle;
795 }
796 
797 static void
798 acpi_cst_support_list(struct acpi_cst_softc *sc)
799 {
800     struct sbuf sb;
801     int i;
802 
803     /*
804      * Set up the list of Cx states
805      */
806     sbuf_new(&sb, sc->cst_cx_supported, sizeof(sc->cst_cx_supported),
807 	SBUF_FIXEDLEN);
808     for (i = 0; i < sc->cst_cx_count; i++)
809 	sbuf_printf(&sb, "C%d/%d ", i + 1, sc->cst_cx_states[i].trans_lat);
810     sbuf_trim(&sb);
811     sbuf_finish(&sb);
812 }
813 
814 static void
815 acpi_cst_c3_bm_rld_handler(netmsg_t msg)
816 {
817     struct netmsg_acpi_cst *rmsg = (struct netmsg_acpi_cst *)msg;
818 
819     AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
820     lwkt_replymsg(&rmsg->base.lmsg, 0);
821 }
822 
823 static void
824 acpi_cst_c3_bm_rld(struct acpi_cst_softc *sc)
825 {
826     struct netmsg_acpi_cst msg;
827 
828     netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
829 	acpi_cst_c3_bm_rld_handler);
830     msg.sc = sc;
831 
832     lwkt_domsg(netisr_cpuport(sc->cst_cpuid), &msg.base.lmsg, 0);
833 }
834 
835 static void
836 acpi_cst_startup(struct acpi_cst_softc *sc)
837 {
838     struct acpi_cpu_softc *cpu = sc->cst_parent;
839     int i, bm_rld_done = 0;
840 
841     for (i = 0; i < sc->cst_cx_count; ++i) {
842 	struct acpi_cst_cx *cx = &sc->cst_cx_states[i];
843 	int error;
844 
845 	/* If there are C3(+) states, always enable bus master wakeup */
846 	if (cx->type >= ACPI_STATE_C3 && !bm_rld_done &&
847 	    (acpi_cst_quirks & ACPI_CST_QUIRK_NO_BM) == 0) {
848 	    acpi_cst_c3_bm_rld(sc);
849 	    bm_rld_done = 1;
850 	}
851 
852 	/* Redo the Cx setup, since quirks have been changed */
853 	error = acpi_cst_cx_setup(cx);
854 	if (error)
855 	    panic("C%d startup setup failed: %d", i + 1, error);
856     }
857 
858     acpi_cst_support_list(sc);
859 
860     SYSCTL_ADD_STRING(&cpu->pcpu_sysctl_ctx,
861 		      SYSCTL_CHILDREN(cpu->pcpu_sysctl_tree),
862 		      OID_AUTO, "cx_supported", CTLFLAG_RD,
863 		      sc->cst_cx_supported, 0,
864 		      "Cx/microsecond values for supported Cx states");
865     SYSCTL_ADD_PROC(&cpu->pcpu_sysctl_ctx,
866 		    SYSCTL_CHILDREN(cpu->pcpu_sysctl_tree),
867 		    OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
868 		    (void *)sc, 0, acpi_cst_lowest_sysctl, "A",
869 		    "requested lowest Cx sleep state");
870     SYSCTL_ADD_PROC(&cpu->pcpu_sysctl_ctx,
871 		    SYSCTL_CHILDREN(cpu->pcpu_sysctl_tree),
872 		    OID_AUTO, "cx_lowest_use", CTLTYPE_STRING | CTLFLAG_RD,
873 		    (void *)sc, 0, acpi_cst_lowest_use_sysctl, "A",
874 		    "lowest Cx sleep state to use");
875     SYSCTL_ADD_PROC(&cpu->pcpu_sysctl_ctx,
876 		    SYSCTL_CHILDREN(cpu->pcpu_sysctl_tree),
877 		    OID_AUTO, "cx_usage", CTLTYPE_STRING | CTLFLAG_RD,
878 		    (void *)sc, 0, acpi_cst_usage_sysctl, "A",
879 		    "percent usage for each Cx state");
880 
881 #ifdef notyet
882     /* Signal platform that we can handle _CST notification. */
883     if (!acpi_cst_use_fadt && acpi_cst_ctrl != 0) {
884 	ACPI_LOCK(acpi);
885 	AcpiOsWritePort(acpi_cst_smi_cmd, acpi_cst_ctrl, 8);
886 	ACPI_UNLOCK(acpi);
887     }
888 #endif
889 }
890 
891 /*
892  * Idle the CPU in the lowest state possible.  This function is called with
893  * interrupts disabled.  Note that once it re-enables interrupts, a task
894  * switch can occur so do not access shared data (i.e. the softc) after
895  * interrupts are re-enabled.
896  */
897 static void
898 acpi_cst_idle(void)
899 {
900     struct	acpi_cst_softc *sc;
901     struct	acpi_cst_cx *cx_next;
902     union microtime_pcpu start, end;
903     int		cx_next_idx, i, tdiff, bm_arb_disabled = 0;
904 
905     /* If disabled, return immediately. */
906     if (acpi_cst_disable_idle) {
907 	ACPI_ENABLE_IRQS();
908 	return;
909     }
910 
911     /*
912      * Look up our CPU id to get our softc.  If it's NULL, we'll use C1
913      * since there is no Cx state for this processor.
914      */
915     sc = acpi_cst_softc[mdcpu->mi.gd_cpuid];
916     if (sc == NULL) {
917 	acpi_cst_c1_halt();
918 	return;
919     }
920 
921     /* Still probing; use C1 */
922     if (sc->cst_flags & ACPI_CST_FLAG_PROBING) {
923 	acpi_cst_c1_halt();
924 	return;
925     }
926 
927     /* Find the lowest state that has small enough latency. */
928     cx_next_idx = 0;
929     for (i = sc->cst_cx_lowest; i >= 0; i--) {
930 	if (sc->cst_cx_states[i].trans_lat * 3 <= sc->cst_prev_sleep) {
931 	    cx_next_idx = i;
932 	    break;
933 	}
934     }
935 
936     /*
937      * Check for bus master activity if needed for the selected state.
938      * If there was activity, clear the bit and use the lowest non-C3 state.
939      */
940     cx_next = &sc->cst_cx_states[cx_next_idx];
941     if (cx_next->flags & ACPI_CST_CX_FLAG_BM_STS) {
942 	int bm_active;
943 
944 	AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, &bm_active);
945 	if (bm_active != 0) {
946 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, 1);
947 	    cx_next_idx = sc->cst_non_c3;
948 	}
949     }
950 
951     /* Select the next state and update statistics. */
952     cx_next = &sc->cst_cx_states[cx_next_idx];
953     sc->cst_cx_stats[cx_next_idx]++;
954     KASSERT(cx_next->type != ACPI_STATE_C0, ("C0 sleep"));
955 
956     /*
957      * Execute HLT (or equivalent) and wait for an interrupt.  We can't
958      * calculate the time spent in C1 since the place we wake up is an
959      * ISR.  Assume we slept half of quantum and return.
960      */
961     if (cx_next->type == ACPI_STATE_C1) {
962 	sc->cst_prev_sleep = (sc->cst_prev_sleep * 3 + 500000 / hz) / 4;
963 	cx_next->enter(cx_next);
964 	return;
965     }
966 
967     /* Execute the proper preamble before enter the selected state. */
968     if (cx_next->preamble == ACPI_CST_CX_PREAMBLE_BM_ARB) {
969 	AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 1);
970 	bm_arb_disabled = 1;
971     } else if (cx_next->preamble == ACPI_CST_CX_PREAMBLE_WBINVD) {
972 	ACPI_FLUSH_CPU_CACHE();
973     }
974 
975     /*
976      * Enter the selected state and check time spent asleep.
977      */
978     microtime_pcpu_get(&start);
979     cpu_mfence();
980 
981     cx_next->enter(cx_next);
982 
983     cpu_mfence();
984     microtime_pcpu_get(&end);
985 
986     /* Enable bus master arbitration, if it was disabled. */
987     if (bm_arb_disabled)
988 	AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 0);
989 
990     ACPI_ENABLE_IRQS();
991 
992     /* Find the actual time asleep in microseconds. */
993     tdiff = microtime_pcpu_diff(&start, &end);
994     sc->cst_prev_sleep = (sc->cst_prev_sleep * 3 + tdiff) / 4;
995 }
996 
997 /*
998  * Re-evaluate the _CST object when we are notified that it changed.
999  */
1000 static void
1001 acpi_cst_notify(device_t dev)
1002 {
1003     struct acpi_cst_softc *sc = device_get_softc(dev);
1004 
1005     KASSERT(curthread->td_type != TD_TYPE_NETISR,
1006         ("notify in netisr%d", mycpuid));
1007 
1008     lwkt_serialize_enter(&acpi_cst_slize);
1009 
1010     /* Update the list of Cx states. */
1011     acpi_cst_cx_reprobe_cst(sc);
1012     acpi_cst_support_list(sc);
1013 
1014     /* Update the new lowest useable Cx state for all CPUs. */
1015     acpi_cst_global_cx_count();
1016 
1017     /*
1018      * Fix up the lowest Cx being used
1019      */
1020     if (acpi_cst_cx_lowest_req < acpi_cst_cx_count)
1021 	acpi_cst_cx_lowest = acpi_cst_cx_lowest_req;
1022     if (acpi_cst_cx_lowest > acpi_cst_cx_count - 1)
1023 	acpi_cst_cx_lowest = acpi_cst_cx_count - 1;
1024 
1025     lwkt_serialize_exit(&acpi_cst_slize);
1026 }
1027 
1028 static int
1029 acpi_cst_set_quirks(void)
1030 {
1031     device_t acpi_dev;
1032     uint32_t val;
1033 
1034     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
1035 
1036     /*
1037      * Bus mastering arbitration control is needed to keep caches coherent
1038      * while sleeping in C3.  If it's not present but a working flush cache
1039      * instruction is present, flush the caches before entering C3 instead.
1040      * Otherwise, just disable C3 completely.
1041      */
1042     if (AcpiGbl_FADT.Pm2ControlBlock == 0 ||
1043 	AcpiGbl_FADT.Pm2ControlLength == 0) {
1044 	if ((AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD) &&
1045 	    (AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD_FLUSH) == 0) {
1046 	    acpi_cst_quirks |= ACPI_CST_QUIRK_NO_BM;
1047 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1048 		"cpu_cst: no BM control, using flush cache method\n"));
1049 	} else {
1050 	    acpi_cst_quirks |= ACPI_CST_QUIRK_NO_C3;
1051 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1052 		"cpu_cst: no BM control, C3 not available\n"));
1053 	}
1054     }
1055 
1056     /* Look for various quirks of the PIIX4 part. */
1057     acpi_dev = pci_find_device(PCI_VENDOR_INTEL, PCI_DEVICE_82371AB_3);
1058     if (acpi_dev != NULL) {
1059 	switch (pci_get_revid(acpi_dev)) {
1060 	/*
1061 	 * Disable C3 support for all PIIX4 chipsets.  Some of these parts
1062 	 * do not report the BMIDE status to the BM status register and
1063 	 * others have a livelock bug if Type-F DMA is enabled.  Linux
1064 	 * works around the BMIDE bug by reading the BM status directly
1065 	 * but we take the simpler approach of disabling C3 for these
1066 	 * parts.
1067 	 *
1068 	 * See erratum #18 ("C3 Power State/BMIDE and Type-F DMA
1069 	 * Livelock") from the January 2002 PIIX4 specification update.
1070 	 * Applies to all PIIX4 models.
1071 	 *
1072 	 * Also, make sure that all interrupts cause a "Stop Break"
1073 	 * event to exit from C2 state.
1074 	 * Also, BRLD_EN_BM (ACPI_BITREG_BUS_MASTER_RLD in ACPI-speak)
1075 	 * should be set to zero, otherwise it causes C2 to short-sleep.
1076 	 * PIIX4 doesn't properly support C3 and bus master activity
1077 	 * need not break out of C2.
1078 	 */
1079 	case PCI_REVISION_A_STEP:
1080 	case PCI_REVISION_B_STEP:
1081 	case PCI_REVISION_4E:
1082 	case PCI_REVISION_4M:
1083 	    acpi_cst_quirks |= ACPI_CST_QUIRK_NO_C3;
1084 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1085 		"cpu_cst: working around PIIX4 bug, disabling C3\n"));
1086 
1087 	    val = pci_read_config(acpi_dev, PIIX4_DEVACTB_REG, 4);
1088 	    if ((val & PIIX4_STOP_BREAK_MASK) != PIIX4_STOP_BREAK_MASK) {
1089 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1090 		    "cpu_cst: PIIX4: enabling IRQs to generate Stop Break\n"));
1091 	    	val |= PIIX4_STOP_BREAK_MASK;
1092 		pci_write_config(acpi_dev, PIIX4_DEVACTB_REG, val, 4);
1093 	    }
1094 	    AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_RLD, &val);
1095 	    if (val) {
1096 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1097 		    "cpu_cst: PIIX4: reset BRLD_EN_BM\n"));
1098 		AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1099 	    }
1100 	    break;
1101 	default:
1102 	    break;
1103 	}
1104     }
1105 
1106     return (0);
1107 }
1108 
1109 static int
1110 acpi_cst_usage_sysctl(SYSCTL_HANDLER_ARGS)
1111 {
1112     struct acpi_cst_softc *sc;
1113     struct sbuf	 sb;
1114     char	 buf[128];
1115     int		 i;
1116     uintmax_t	 fract, sum, whole;
1117 
1118     sc = (struct acpi_cst_softc *) arg1;
1119     sum = 0;
1120     for (i = 0; i < sc->cst_cx_count; i++)
1121 	sum += sc->cst_cx_stats[i];
1122     sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1123     for (i = 0; i < sc->cst_cx_count; i++) {
1124 	if (sum > 0) {
1125 	    whole = (uintmax_t)sc->cst_cx_stats[i] * 100;
1126 	    fract = (whole % sum) * 100;
1127 	    sbuf_printf(&sb, "%u.%02u%% ", (u_int)(whole / sum),
1128 		(u_int)(fract / sum));
1129 	} else
1130 	    sbuf_printf(&sb, "0.00%% ");
1131     }
1132     sbuf_printf(&sb, "last %dus", sc->cst_prev_sleep);
1133     sbuf_trim(&sb);
1134     sbuf_finish(&sb);
1135     sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1136     sbuf_delete(&sb);
1137 
1138     return (0);
1139 }
1140 
1141 static int
1142 acpi_cst_set_lowest_oncpu(struct acpi_cst_softc *sc, int val)
1143 {
1144     int old_lowest, error = 0, old_lowest_req;
1145     uint32_t old_type, type;
1146 
1147     KKASSERT(mycpuid == sc->cst_cpuid);
1148 
1149     old_lowest_req = sc->cst_cx_lowest_req;
1150     sc->cst_cx_lowest_req = val;
1151 
1152     if (val > sc->cst_cx_count - 1)
1153 	val = sc->cst_cx_count - 1;
1154     old_lowest = atomic_swap_int(&sc->cst_cx_lowest, val);
1155 
1156     old_type = sc->cst_cx_states[old_lowest].type;
1157     type = sc->cst_cx_states[val].type;
1158     if (old_type >= ACPI_STATE_C3 && type < ACPI_STATE_C3) {
1159 	cputimer_intr_powersave_remreq();
1160     } else if (type >= ACPI_STATE_C3 && old_type < ACPI_STATE_C3) {
1161 	error = cputimer_intr_powersave_addreq();
1162 	if (error) {
1163 	    /* Restore */
1164 	    sc->cst_cx_lowest_req = old_lowest_req;
1165 	    sc->cst_cx_lowest = old_lowest;
1166 	}
1167     }
1168 
1169     if (error)
1170 	return error;
1171 
1172     /* Cache the new lowest non-C3 state. */
1173     acpi_cst_non_c3(sc);
1174 
1175     /* Reset the statistics counters. */
1176     bzero(sc->cst_cx_stats, sizeof(sc->cst_cx_stats));
1177     return (0);
1178 }
1179 
1180 static void
1181 acpi_cst_set_lowest_handler(netmsg_t msg)
1182 {
1183     struct netmsg_acpi_cst *rmsg = (struct netmsg_acpi_cst *)msg;
1184     int error;
1185 
1186     error = acpi_cst_set_lowest_oncpu(rmsg->sc, rmsg->val);
1187     lwkt_replymsg(&rmsg->base.lmsg, error);
1188 }
1189 
1190 static int
1191 acpi_cst_set_lowest(struct acpi_cst_softc *sc, int val)
1192 {
1193     struct netmsg_acpi_cst msg;
1194 
1195     netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
1196 	acpi_cst_set_lowest_handler);
1197     msg.sc = sc;
1198     msg.val = val;
1199 
1200     return lwkt_domsg(netisr_cpuport(sc->cst_cpuid), &msg.base.lmsg, 0);
1201 }
1202 
1203 static int
1204 acpi_cst_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1205 {
1206     struct	 acpi_cst_softc *sc;
1207     char	 state[8];
1208     int		 val, error;
1209 
1210     sc = (struct acpi_cst_softc *)arg1;
1211     ksnprintf(state, sizeof(state), "C%d", sc->cst_cx_lowest_req + 1);
1212     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1213     if (error != 0 || req->newptr == NULL)
1214 	return (error);
1215     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1216 	return (EINVAL);
1217     val = (int) strtol(state + 1, NULL, 10) - 1;
1218     if (val < 0)
1219 	return (EINVAL);
1220 
1221     lwkt_serialize_enter(&acpi_cst_slize);
1222     error = acpi_cst_set_lowest(sc, val);
1223     lwkt_serialize_exit(&acpi_cst_slize);
1224 
1225     return error;
1226 }
1227 
1228 static int
1229 acpi_cst_lowest_use_sysctl(SYSCTL_HANDLER_ARGS)
1230 {
1231     struct	 acpi_cst_softc *sc;
1232     char	 state[8];
1233 
1234     sc = (struct acpi_cst_softc *)arg1;
1235     ksnprintf(state, sizeof(state), "C%d", sc->cst_cx_lowest + 1);
1236     return sysctl_handle_string(oidp, state, sizeof(state), req);
1237 }
1238 
1239 static int
1240 acpi_cst_global_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1241 {
1242     struct	acpi_cst_softc *sc;
1243     char	state[8];
1244     int		val, error, i;
1245 
1246     ksnprintf(state, sizeof(state), "C%d", acpi_cst_cx_lowest_req + 1);
1247     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1248     if (error != 0 || req->newptr == NULL)
1249 	return (error);
1250     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1251 	return (EINVAL);
1252     val = (int) strtol(state + 1, NULL, 10) - 1;
1253     if (val < 0)
1254 	return (EINVAL);
1255 
1256     lwkt_serialize_enter(&acpi_cst_slize);
1257 
1258     acpi_cst_cx_lowest_req = val;
1259     acpi_cst_cx_lowest = val;
1260     if (acpi_cst_cx_lowest > acpi_cst_cx_count - 1)
1261 	acpi_cst_cx_lowest = acpi_cst_cx_count - 1;
1262 
1263     /* Update the new lowest useable Cx state for all CPUs. */
1264     for (i = 0; i < acpi_cst_ndevices; i++) {
1265 	sc = device_get_softc(acpi_cst_devices[i]);
1266 	error = acpi_cst_set_lowest(sc, val);
1267 	if (error) {
1268 	    KKASSERT(i == 0);
1269 	    break;
1270 	}
1271     }
1272 
1273     lwkt_serialize_exit(&acpi_cst_slize);
1274 
1275     return error;
1276 }
1277 
1278 static int
1279 acpi_cst_global_lowest_use_sysctl(SYSCTL_HANDLER_ARGS)
1280 {
1281     char	state[8];
1282 
1283     ksnprintf(state, sizeof(state), "C%d", acpi_cst_cx_lowest + 1);
1284     return sysctl_handle_string(oidp, state, sizeof(state), req);
1285 }
1286 
1287 /*
1288  * Put the CPU in C1 in a machine-dependant way.
1289  * XXX: shouldn't be here!
1290  */
1291 static void
1292 acpi_cst_c1_halt(void)
1293 {
1294     splz();
1295     if ((mycpu->gd_reqflags & RQF_IDLECHECK_WK_MASK) == 0)
1296         __asm __volatile("sti; hlt");
1297     else
1298         __asm __volatile("sti; pause");
1299 }
1300 
1301 static void
1302 acpi_cst_non_c3(struct acpi_cst_softc *sc)
1303 {
1304     int i;
1305 
1306     sc->cst_non_c3 = 0;
1307     for (i = sc->cst_cx_lowest; i >= 0; i--) {
1308 	if (sc->cst_cx_states[i].type < ACPI_STATE_C3) {
1309 	    sc->cst_non_c3 = i;
1310 	    break;
1311 	}
1312     }
1313     if (bootverbose)
1314 	device_printf(sc->cst_dev, "non-C3 %d\n", sc->cst_non_c3);
1315 }
1316 
1317 /*
1318  * Update the largest Cx state supported in the global acpi_cst_cx_count.
1319  * It will be used in the global Cx sysctl handler.
1320  */
1321 static void
1322 acpi_cst_global_cx_count(void)
1323 {
1324     struct acpi_cst_softc *sc;
1325     int i;
1326 
1327     if (acpi_cst_ndevices == 0) {
1328 	acpi_cst_cx_count = 0;
1329 	return;
1330     }
1331 
1332     sc = device_get_softc(acpi_cst_devices[0]);
1333     acpi_cst_cx_count = sc->cst_cx_count;
1334 
1335     for (i = 1; i < acpi_cst_ndevices; i++) {
1336 	struct acpi_cst_softc *sc = device_get_softc(acpi_cst_devices[i]);
1337 
1338 	if (sc->cst_cx_count < acpi_cst_cx_count)
1339 	    acpi_cst_cx_count = sc->cst_cx_count;
1340     }
1341     if (bootverbose)
1342 	kprintf("cpu_cst: global Cx count %d\n", acpi_cst_cx_count);
1343 }
1344 
1345 static void
1346 acpi_cst_c1_halt_enter(const struct acpi_cst_cx *cx __unused)
1347 {
1348     acpi_cst_c1_halt();
1349 }
1350 
1351 static void
1352 acpi_cst_cx_io_enter(const struct acpi_cst_cx *cx)
1353 {
1354     uint64_t dummy;
1355 
1356     /*
1357      * Read I/O to enter this Cx state
1358      */
1359     bus_space_read_1(cx->btag, cx->bhand, 0);
1360     /*
1361      * Perform a dummy I/O read.  Since it may take an arbitrary time
1362      * to enter the idle state, this read makes sure that we are frozen.
1363      */
1364     AcpiRead(&dummy, &AcpiGbl_FADT.XPmTimerBlock);
1365 }
1366 
1367 static int
1368 acpi_cst_cx_setup(struct acpi_cst_cx *cx)
1369 {
1370     cx->flags &= ~ACPI_CST_CX_FLAG_BM_STS;
1371     cx->preamble = ACPI_CST_CX_PREAMBLE_NONE;
1372 
1373     if (cx->type >= ACPI_STATE_C3) {
1374 	/*
1375 	 * Set the required operations for entering C3(+) state.
1376 	 * Later acpi_cst_md_cx_setup() may fix them up.
1377 	 */
1378 
1379 	/*
1380 	 * Always check BM_STS.
1381 	 */
1382 	if ((acpi_cst_quirks & ACPI_CST_QUIRK_NO_BM) == 0)
1383 	    cx->flags |= ACPI_CST_CX_FLAG_BM_STS;
1384 
1385 	/*
1386 	 * According to the ACPI specification, bus master arbitration
1387 	 * is only available on UP system.  For MP system, cache flushing
1388 	 * is required.
1389 	 */
1390 	if (ncpus == 1 && (acpi_cst_quirks & ACPI_CST_QUIRK_NO_BM) == 0)
1391 	    cx->preamble = ACPI_CST_CX_PREAMBLE_BM_ARB;
1392 	else
1393 	    cx->preamble = ACPI_CST_CX_PREAMBLE_WBINVD;
1394     }
1395     return acpi_cst_md_cx_setup(cx);
1396 }
1397 
1398 static void
1399 acpi_cst_free_resource(struct acpi_cst_softc *sc, int start)
1400 {
1401     int i;
1402 
1403     for (i = start; i < MAX_CX_STATES; ++i) {
1404 	struct acpi_cst_cx *cx = &sc->cst_cx_states[i];
1405 
1406 	if (cx->res != NULL)
1407 	    bus_release_resource(sc->cst_dev, cx->res_type, cx->rid, cx->res);
1408 	memset(cx, 0, sizeof(*cx));
1409     }
1410 }
1411