xref: /dragonfly/sys/dev/acpica/acpi_cpu_cstate.c (revision a4c31683)
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/cpuhelper.h>
34 #include <sys/kernel.h>
35 #include <sys/malloc.h>
36 #include <sys/globaldata.h>
37 #include <sys/power.h>
38 #include <sys/proc.h>
39 #include <sys/sbuf.h>
40 #include <sys/thread2.h>
41 #include <sys/serialize.h>
42 #include <sys/msgport2.h>
43 #include <sys/microtime_pcpu.h>
44 #include <sys/cpu_topology.h>
45 
46 #include <bus/pci/pcivar.h>
47 #include <machine/atomic.h>
48 #include <machine/globaldata.h>
49 #include <machine/md_var.h>
50 #include <machine/smp.h>
51 #include <sys/rman.h>
52 
53 #include "acpi.h"
54 #include "acpivar.h"
55 #include "acpi_cpu.h"
56 #include "acpi_cpu_cstate.h"
57 
58 /*
59  * Support for ACPI Processor devices, including C[1-3+] sleep states.
60  */
61 
62 /* Hooks for the ACPICA debugging infrastructure */
63 #define _COMPONENT	ACPI_PROCESSOR
64 ACPI_MODULE_NAME("PROCESSOR")
65 
66 #define MAX_CX_STATES	 8
67 
68 struct acpi_cst_softc {
69     device_t		cst_dev;
70     struct acpi_cpu_softc *cst_parent;
71     ACPI_HANDLE		cst_handle;
72     int			cst_cpuid;
73     uint32_t		cst_flags;	/* ACPI_CST_FLAG_ */
74     uint32_t		cst_p_blk;	/* ACPI P_BLK location */
75     uint32_t		cst_p_blk_len;	/* P_BLK length (must be 6). */
76     struct acpi_cst_cx	cst_cx_states[MAX_CX_STATES];
77     int			cst_cx_count;	/* Number of valid Cx states. */
78     int			cst_prev_sleep;	/* Last idle sleep duration. */
79     /* Runtime state. */
80     int			cst_non_c3;	/* Index of lowest non-C3 state. */
81     u_long		cst_cx_stats[MAX_CX_STATES];/* Cx usage history. */
82     /* Values for sysctl. */
83     int			cst_cx_lowest;	/* Current Cx lowest */
84     int			cst_cx_lowest_req; /* Requested Cx lowest */
85     char 		cst_cx_supported[64];
86 };
87 
88 #define ACPI_CST_FLAG_PROBING	0x1
89 #define ACPI_CST_FLAG_ATTACHED	0x2
90 /* Match C-states of other hyperthreads on the same core */
91 #define ACPI_CST_FLAG_MATCH_HT	0x4
92 
93 #define PCI_VENDOR_INTEL	0x8086
94 #define PCI_DEVICE_82371AB_3	0x7113	/* PIIX4 chipset for quirks. */
95 #define PCI_REVISION_A_STEP	0
96 #define PCI_REVISION_B_STEP	1
97 #define PCI_REVISION_4E		2
98 #define PCI_REVISION_4M		3
99 #define PIIX4_DEVACTB_REG	0x58
100 #define PIIX4_BRLD_EN_IRQ0	(1<<0)
101 #define PIIX4_BRLD_EN_IRQ	(1<<1)
102 #define PIIX4_BRLD_EN_IRQ8	(1<<5)
103 #define PIIX4_STOP_BREAK_MASK	(PIIX4_BRLD_EN_IRQ0 | \
104 				 PIIX4_BRLD_EN_IRQ | \
105 				 PIIX4_BRLD_EN_IRQ8)
106 #define PIIX4_PCNTRL_BST_EN	(1<<10)
107 
108 /* Platform hardware resource information. */
109 static uint32_t		 acpi_cst_smi_cmd; /* Value to write to SMI_CMD. */
110 static uint8_t		 acpi_cst_ctrl;	/* Indicate we are _CST aware. */
111 int		 	 acpi_cst_quirks; /* Indicate any hardware bugs. */
112 static boolean_t	 acpi_cst_use_fadt;
113 
114 /* Runtime state. */
115 static boolean_t	 acpi_cst_disable_idle;
116 					/* Disable entry to idle function */
117 static int		 acpi_cst_cx_count; /* Number of valid Cx states */
118 
119 /* Values for sysctl. */
120 static int		 acpi_cst_cx_lowest; /* Current Cx lowest */
121 static int		 acpi_cst_cx_lowest_req; /* Requested Cx lowest */
122 
123 static device_t		*acpi_cst_devices;
124 static int		 acpi_cst_ndevices;
125 static struct acpi_cst_softc **acpi_cst_softc;
126 static struct lwkt_serialize acpi_cst_slize = LWKT_SERIALIZE_INITIALIZER;
127 
128 static int	acpi_cst_probe(device_t);
129 static int	acpi_cst_attach(device_t);
130 static int	acpi_cst_suspend(device_t);
131 static int	acpi_cst_resume(device_t);
132 static int	acpi_cst_shutdown(device_t);
133 
134 static void	acpi_cst_notify(device_t);
135 static void	acpi_cst_postattach(void *);
136 static void	acpi_cst_idle(void);
137 static void	acpi_cst_copy(struct acpi_cst_softc *,
138 		    const struct acpi_cst_softc *);
139 
140 static void	acpi_cst_cx_probe(struct acpi_cst_softc *);
141 static void	acpi_cst_cx_probe_fadt(struct acpi_cst_softc *);
142 static int	acpi_cst_cx_probe_cst(struct acpi_cst_softc *, int);
143 static int	acpi_cst_cx_reprobe_cst(struct acpi_cst_softc *);
144 
145 static void	acpi_cst_startup(struct acpi_cst_softc *);
146 static void	acpi_cst_support_list(struct acpi_cst_softc *);
147 static int	acpi_cst_set_lowest(struct acpi_cst_softc *, int);
148 static int	acpi_cst_set_lowest_oncpu(struct acpi_cst_softc *, int);
149 static void	acpi_cst_non_c3(struct acpi_cst_softc *);
150 static void	acpi_cst_global_cx_count(void);
151 static int	acpi_cst_set_quirks(void);
152 static void	acpi_cst_c3_bm_rld(struct acpi_cst_softc *);
153 static void	acpi_cst_free_resource(struct acpi_cst_softc *, int);
154 static void	acpi_cst_c1_halt(void);
155 
156 static int	acpi_cst_usage_sysctl(SYSCTL_HANDLER_ARGS);
157 static int	acpi_cst_lowest_sysctl(SYSCTL_HANDLER_ARGS);
158 static int	acpi_cst_lowest_use_sysctl(SYSCTL_HANDLER_ARGS);
159 static int	acpi_cst_global_lowest_sysctl(SYSCTL_HANDLER_ARGS);
160 static int	acpi_cst_global_lowest_use_sysctl(SYSCTL_HANDLER_ARGS);
161 
162 static int	acpi_cst_cx_setup(struct acpi_cst_cx *cx);
163 static void	acpi_cst_c1_halt_enter(const struct acpi_cst_cx *);
164 static void	acpi_cst_cx_io_enter(const struct acpi_cst_cx *);
165 
166 int		acpi_cst_force_bmarb;
167 TUNABLE_INT("hw.acpi.cpu.cst.force_bmarb", &acpi_cst_force_bmarb);
168 
169 int		acpi_cst_force_bmsts;
170 TUNABLE_INT("hw.acpi.cpu.cst.force_bmsts", &acpi_cst_force_bmsts);
171 
172 static device_method_t acpi_cst_methods[] = {
173     /* Device interface */
174     DEVMETHOD(device_probe,	acpi_cst_probe),
175     DEVMETHOD(device_attach,	acpi_cst_attach),
176     DEVMETHOD(device_detach,	bus_generic_detach),
177     DEVMETHOD(device_shutdown,	acpi_cst_shutdown),
178     DEVMETHOD(device_suspend,	acpi_cst_suspend),
179     DEVMETHOD(device_resume,	acpi_cst_resume),
180 
181     /* Bus interface */
182     DEVMETHOD(bus_add_child,	bus_generic_add_child),
183     DEVMETHOD(bus_read_ivar,	bus_generic_read_ivar),
184     DEVMETHOD(bus_get_resource_list, bus_generic_get_resource_list),
185     DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
186     DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
187     DEVMETHOD(bus_alloc_resource, bus_generic_rl_alloc_resource),
188     DEVMETHOD(bus_release_resource, bus_generic_rl_release_resource),
189     DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
190     DEVMETHOD(bus_activate_resource, bus_generic_activate_resource),
191     DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource),
192     DEVMETHOD(bus_setup_intr,	bus_generic_setup_intr),
193     DEVMETHOD(bus_teardown_intr, bus_generic_teardown_intr),
194     DEVMETHOD_END
195 };
196 
197 static driver_t acpi_cst_driver = {
198     "cpu_cst",
199     acpi_cst_methods,
200     sizeof(struct acpi_cst_softc),
201 };
202 
203 static devclass_t acpi_cst_devclass;
204 DRIVER_MODULE(cpu_cst, cpu, acpi_cst_driver, acpi_cst_devclass, NULL, NULL);
205 MODULE_DEPEND(cpu_cst, acpi, 1, 1, 1);
206 
207 static int
208 acpi_cst_probe(device_t dev)
209 {
210     int cpu_id;
211 
212     if (acpi_disabled("cpu_cst") || acpi_get_type(dev) != ACPI_TYPE_PROCESSOR)
213 	return (ENXIO);
214 
215     cpu_id = acpi_get_magic(dev);
216 
217     if (acpi_cst_softc == NULL)
218 	acpi_cst_softc = kmalloc(sizeof(struct acpi_cst_softc *) *
219 	    SMP_MAXCPU, M_TEMP /* XXX */, M_INTWAIT | M_ZERO);
220 
221     /*
222      * Check if we already probed this processor.  We scan the bus twice
223      * so it's possible we've already seen this one.
224      */
225     if (acpi_cst_softc[cpu_id] != NULL) {
226 	device_printf(dev, "CPU%d cstate already exist\n", cpu_id);
227 	return (ENXIO);
228     }
229 
230     /* Mark this processor as in-use and save our derived id for attach. */
231     acpi_cst_softc[cpu_id] = device_get_softc(dev);
232     device_set_desc(dev, "ACPI CPU C-State");
233 
234     return (0);
235 }
236 
237 static int
238 acpi_cst_attach(device_t dev)
239 {
240     ACPI_BUFFER		   buf;
241     ACPI_OBJECT		   *obj;
242     struct acpi_cst_softc *sc;
243     ACPI_STATUS		   status;
244 
245     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
246 
247     sc = device_get_softc(dev);
248     sc->cst_dev = dev;
249     sc->cst_parent = device_get_softc(device_get_parent(dev));
250     sc->cst_handle = acpi_get_handle(dev);
251     sc->cst_cpuid = acpi_get_magic(dev);
252     acpi_cst_softc[sc->cst_cpuid] = sc;
253     acpi_cst_smi_cmd = AcpiGbl_FADT.SmiCommand;
254     acpi_cst_ctrl = AcpiGbl_FADT.CstControl;
255 
256     buf.Pointer = NULL;
257     buf.Length = ACPI_ALLOCATE_BUFFER;
258     status = AcpiEvaluateObject(sc->cst_handle, NULL, NULL, &buf);
259     if (ACPI_FAILURE(status)) {
260 	device_printf(dev, "attach failed to get Processor obj - %s\n",
261 		      AcpiFormatException(status));
262 	acpi_cst_softc[sc->cst_cpuid] = NULL;
263 	return (ENXIO);
264     }
265     obj = (ACPI_OBJECT *)buf.Pointer;
266     sc->cst_p_blk = obj->Processor.PblkAddress;
267     sc->cst_p_blk_len = obj->Processor.PblkLength;
268     AcpiOsFree(obj);
269     ACPI_DEBUG_PRINT((ACPI_DB_INFO, "cpu_cst%d: P_BLK at %#x/%d\n",
270 		     device_get_unit(dev), sc->cst_p_blk, sc->cst_p_blk_len));
271 
272     /*
273      * If this is the first cpu we attach, create and initialize the generic
274      * resources that will be used by all acpi cpu devices.
275      */
276     if (device_get_unit(dev) == 0) {
277 	/* Assume we won't be using FADT for Cx states by default */
278 	acpi_cst_use_fadt = FALSE;
279 
280 	/* Queue post cpu-probing task handler */
281 	AcpiOsExecute(OSL_NOTIFY_HANDLER, acpi_cst_postattach, NULL);
282     }
283 
284     /* Probe for Cx state support. */
285     acpi_cst_cx_probe(sc);
286 
287     sc->cst_flags |= ACPI_CST_FLAG_ATTACHED;
288 
289     return (0);
290 }
291 
292 /*
293  * Disable any entry to the idle function during suspend and re-enable it
294  * during resume.
295  */
296 static int
297 acpi_cst_suspend(device_t dev)
298 {
299     int error;
300 
301     error = bus_generic_suspend(dev);
302     if (error)
303 	return (error);
304     acpi_cst_disable_idle = TRUE;
305     return (0);
306 }
307 
308 static int
309 acpi_cst_resume(device_t dev)
310 {
311     acpi_cst_disable_idle = FALSE;
312     return (bus_generic_resume(dev));
313 }
314 
315 static int
316 acpi_cst_shutdown(device_t dev)
317 {
318     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
319 
320     /* Allow children to shutdown first. */
321     bus_generic_shutdown(dev);
322 
323     /*
324      * Disable any entry to the idle function.  There is a small race where
325      * an idle thread have passed this check but not gone to sleep.  This
326      * is ok since device_shutdown() does not free the softc, otherwise
327      * we'd have to be sure all threads were evicted before returning.
328      */
329     acpi_cst_disable_idle = TRUE;
330 
331     return_VALUE (0);
332 }
333 
334 static void
335 acpi_cst_cx_probe(struct acpi_cst_softc *sc)
336 {
337     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
338 
339     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
340     sc->cst_prev_sleep = 1000000;
341     sc->cst_cx_lowest = 0;
342     sc->cst_cx_lowest_req = 0;
343 
344     /*
345      * Check for the ACPI 2.0 _CST sleep states object.  If we can't find
346      * any, we'll revert to FADT/P_BLK Cx control method which will be
347      * handled by acpi_cst_postattach.  We need to defer to after having
348      * probed all the cpus in the system before probing for Cx states from
349      * FADT as we may already have found cpus with valid _CST packages.
350      */
351     if (!acpi_cst_use_fadt && acpi_cst_cx_probe_cst(sc, 0) != 0) {
352 	/*
353 	 * We were unable to find a _CST package for this cpu or there
354 	 * was an error parsing it. Switch back to generic mode.
355 	 */
356 	acpi_cst_use_fadt = TRUE;
357 	if (bootverbose)
358 	    device_printf(sc->cst_dev, "switching to FADT Cx mode\n");
359     }
360 
361     /*
362      * TODO: _CSD Package should be checked here.
363      */
364 }
365 
366 static void
367 acpi_cst_cx_probe_fadt(struct acpi_cst_softc *sc)
368 {
369     struct acpi_cst_cx *cx_ptr;
370     int error;
371 
372     /*
373      * Free all previously allocated resources.
374      *
375      * NITE:
376      * It is needed, since we could enter here because of other
377      * cpu's _CST probing failure.
378      */
379     acpi_cst_free_resource(sc, 0);
380 
381     sc->cst_cx_count = 0;
382     cx_ptr = sc->cst_cx_states;
383 
384     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
385     sc->cst_prev_sleep = 1000000;
386 
387     /* C1 has been required since just after ACPI 1.0 */
388     cx_ptr->gas.SpaceId = ACPI_ADR_SPACE_FIXED_HARDWARE;
389     cx_ptr->type = ACPI_STATE_C1;
390     cx_ptr->trans_lat = 0;
391     cx_ptr->enter = acpi_cst_c1_halt_enter;
392     error = acpi_cst_cx_setup(cx_ptr);
393     if (error)
394 	panic("C1 FADT HALT setup failed: %d", error);
395     cx_ptr++;
396     sc->cst_cx_count++;
397 
398     /* C2(+) is not supported on MP system */
399     if (ncpus > 1 && (AcpiGbl_FADT.Flags & ACPI_FADT_C2_MP_SUPPORTED) == 0)
400 	return;
401 
402     /*
403      * The spec says P_BLK must be 6 bytes long.  However, some systems
404      * use it to indicate a fractional set of features present so we
405      * take 5 as C2.  Some may also have a value of 7 to indicate
406      * another C3 but most use _CST for this (as required) and having
407      * "only" C1-C3 is not a hardship.
408      */
409     if (sc->cst_p_blk_len < 5)
410 	return;
411 
412     /* Validate and allocate resources for C2 (P_LVL2). */
413     if (AcpiGbl_FADT.C2Latency <= 100) {
414 	cx_ptr->gas.SpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
415 	cx_ptr->gas.BitWidth = 8;
416 	cx_ptr->gas.Address = sc->cst_p_blk + 4;
417 
418 	cx_ptr->rid = sc->cst_parent->cpu_next_rid;
419 	acpi_bus_alloc_gas(sc->cst_dev, &cx_ptr->res_type, &cx_ptr->rid,
420 	    &cx_ptr->gas, &cx_ptr->res, RF_SHAREABLE);
421 	if (cx_ptr->res != NULL) {
422 	    sc->cst_parent->cpu_next_rid++;
423 	    cx_ptr->type = ACPI_STATE_C2;
424 	    cx_ptr->trans_lat = AcpiGbl_FADT.C2Latency;
425 	    cx_ptr->enter = acpi_cst_cx_io_enter;
426 	    cx_ptr->btag = rman_get_bustag(cx_ptr->res);
427 	    cx_ptr->bhand = rman_get_bushandle(cx_ptr->res);
428 	    error = acpi_cst_cx_setup(cx_ptr);
429 	    if (error)
430 		panic("C2 FADT I/O setup failed: %d", error);
431 	    cx_ptr++;
432 	    sc->cst_cx_count++;
433 	    sc->cst_non_c3 = 1;
434 	}
435     }
436     if (sc->cst_p_blk_len < 6)
437 	return;
438 
439     /* Validate and allocate resources for C3 (P_LVL3). */
440     if (AcpiGbl_FADT.C3Latency <= 1000 &&
441         !(acpi_cst_quirks & ACPI_CST_QUIRK_NO_C3)) {
442 	cx_ptr->gas.SpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
443 	cx_ptr->gas.BitWidth = 8;
444 	cx_ptr->gas.Address = sc->cst_p_blk + 5;
445 
446 	cx_ptr->rid = sc->cst_parent->cpu_next_rid;
447 	acpi_bus_alloc_gas(sc->cst_dev, &cx_ptr->res_type, &cx_ptr->rid,
448 	    &cx_ptr->gas, &cx_ptr->res, RF_SHAREABLE);
449 	if (cx_ptr->res != NULL) {
450 	    sc->cst_parent->cpu_next_rid++;
451 	    cx_ptr->type = ACPI_STATE_C3;
452 	    cx_ptr->trans_lat = AcpiGbl_FADT.C3Latency;
453 	    cx_ptr->enter = acpi_cst_cx_io_enter;
454 	    cx_ptr->btag = rman_get_bustag(cx_ptr->res);
455 	    cx_ptr->bhand = rman_get_bushandle(cx_ptr->res);
456 	    error = acpi_cst_cx_setup(cx_ptr);
457 	    if (error)
458 		panic("C3 FADT I/O setup failed: %d", error);
459 	    cx_ptr++;
460 	    sc->cst_cx_count++;
461 	}
462     }
463 }
464 
465 static void
466 acpi_cst_copy(struct acpi_cst_softc *dst_sc,
467     const struct acpi_cst_softc *src_sc)
468 {
469     dst_sc->cst_non_c3 = src_sc->cst_non_c3;
470     dst_sc->cst_cx_count = src_sc->cst_cx_count;
471     memcpy(dst_sc->cst_cx_states, src_sc->cst_cx_states,
472         sizeof(dst_sc->cst_cx_states));
473 }
474 
475 /*
476  * Parse a _CST package and set up its Cx states.  Since the _CST object
477  * can change dynamically, our notify handler may call this function
478  * to clean up and probe the new _CST package.
479  */
480 static int
481 acpi_cst_cx_probe_cst(struct acpi_cst_softc *sc, int reprobe)
482 {
483     struct	 acpi_cst_cx *cx_ptr;
484     ACPI_STATUS	 status;
485     ACPI_BUFFER	 buf;
486     ACPI_OBJECT	*top;
487     ACPI_OBJECT	*pkg;
488     uint32_t	 count;
489     int		 i;
490 
491     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
492 
493     if (reprobe)
494 	cpuhelper_assert(sc->cst_cpuid, true);
495 
496     buf.Pointer = NULL;
497     buf.Length = ACPI_ALLOCATE_BUFFER;
498     status = AcpiEvaluateObject(sc->cst_handle, "_CST", NULL, &buf);
499     if (ACPI_FAILURE(status))
500 	return (ENXIO);
501 
502     /* _CST is a package with a count and at least one Cx package. */
503     top = (ACPI_OBJECT *)buf.Pointer;
504     if (!ACPI_PKG_VALID(top, 2) || acpi_PkgInt32(top, 0, &count) != 0) {
505 	device_printf(sc->cst_dev, "invalid _CST package\n");
506 	AcpiOsFree(buf.Pointer);
507 	return (ENXIO);
508     }
509     if (count != top->Package.Count - 1) {
510 	device_printf(sc->cst_dev, "invalid _CST state count (%d != %d)\n",
511 	       count, top->Package.Count - 1);
512 	count = top->Package.Count - 1;
513     }
514     if (count > MAX_CX_STATES) {
515 	device_printf(sc->cst_dev, "_CST has too many states (%d)\n", count);
516 	count = MAX_CX_STATES;
517     }
518 
519     sc->cst_flags |= ACPI_CST_FLAG_PROBING | ACPI_CST_FLAG_MATCH_HT;
520     cpu_sfence();
521 
522     /*
523      * Free all previously allocated resources
524      *
525      * NOTE: It is needed for _CST reprobing.
526      */
527     acpi_cst_free_resource(sc, 0);
528 
529     /* Set up all valid states. */
530     sc->cst_cx_count = 0;
531     cx_ptr = sc->cst_cx_states;
532     for (i = 0; i < count; i++) {
533 	int error;
534 
535 	pkg = &top->Package.Elements[i + 1];
536 	if (!ACPI_PKG_VALID(pkg, 4) ||
537 	    acpi_PkgInt32(pkg, 1, &cx_ptr->type) != 0 ||
538 	    acpi_PkgInt32(pkg, 2, &cx_ptr->trans_lat) != 0 ||
539 	    acpi_PkgInt32(pkg, 3, &cx_ptr->power) != 0) {
540 
541 	    device_printf(sc->cst_dev, "skipping invalid Cx state package\n");
542 	    continue;
543 	}
544 
545 	/* Validate the state to see if we should use it. */
546 	switch (cx_ptr->type) {
547 	case ACPI_STATE_C1:
548 	    sc->cst_non_c3 = i;
549 	    cx_ptr->enter = acpi_cst_c1_halt_enter;
550 	    error = acpi_cst_cx_setup(cx_ptr);
551 	    if (error)
552 		panic("C1 CST HALT setup failed: %d", error);
553 	    if (sc->cst_cx_count != 0) {
554 		/*
555 		 * C1 is not the first C-state; something really stupid
556 		 * is going on ...
557 		 */
558 		sc->cst_flags &= ~ACPI_CST_FLAG_MATCH_HT;
559 	    }
560 	    cx_ptr++;
561 	    sc->cst_cx_count++;
562 	    continue;
563 	case ACPI_STATE_C2:
564 	    sc->cst_non_c3 = i;
565 	    break;
566 	case ACPI_STATE_C3:
567 	default:
568 	    if ((acpi_cst_quirks & ACPI_CST_QUIRK_NO_C3) != 0) {
569 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
570 				 "cpu_cst%d: C3[%d] not available.\n",
571 				 device_get_unit(sc->cst_dev), i));
572 		continue;
573 	    }
574 	    break;
575 	}
576 
577 	/*
578 	 * Allocate the control register for C2 or C3(+).
579 	 */
580 	KASSERT(cx_ptr->res == NULL, ("still has res"));
581 	acpi_PkgRawGas(pkg, 0, &cx_ptr->gas);
582 
583 	/*
584 	 * We match number of C2/C3 for hyperthreads, only if the
585 	 * register is "Fixed Hardware", e.g. on most of the Intel
586 	 * CPUs.  We don't have much to do for the rest of the
587 	 * register types.
588 	 */
589 	if (cx_ptr->gas.SpaceId != ACPI_ADR_SPACE_FIXED_HARDWARE)
590 	    sc->cst_flags &= ~ACPI_CST_FLAG_MATCH_HT;
591 
592 	cx_ptr->rid = sc->cst_parent->cpu_next_rid;
593 	acpi_bus_alloc_gas(sc->cst_dev, &cx_ptr->res_type, &cx_ptr->rid,
594 	    &cx_ptr->gas, &cx_ptr->res, RF_SHAREABLE);
595 	if (cx_ptr->res != NULL) {
596 	    sc->cst_parent->cpu_next_rid++;
597 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
598 			     "cpu_cst%d: Got C%d - %d latency\n",
599 			     device_get_unit(sc->cst_dev), cx_ptr->type,
600 			     cx_ptr->trans_lat));
601 	    cx_ptr->enter = acpi_cst_cx_io_enter;
602 	    cx_ptr->btag = rman_get_bustag(cx_ptr->res);
603 	    cx_ptr->bhand = rman_get_bushandle(cx_ptr->res);
604 	    error = acpi_cst_cx_setup(cx_ptr);
605 	    if (error)
606 		panic("C%d CST I/O setup failed: %d", cx_ptr->type, error);
607 	    cx_ptr++;
608 	    sc->cst_cx_count++;
609 	} else {
610 	    error = acpi_cst_cx_setup(cx_ptr);
611 	    if (!error) {
612 		KASSERT(cx_ptr->enter != NULL,
613 		    ("C%d enter is not set", cx_ptr->type));
614 		cx_ptr++;
615 		sc->cst_cx_count++;
616 	    }
617 	}
618     }
619     AcpiOsFree(buf.Pointer);
620 
621     if (sc->cst_flags & ACPI_CST_FLAG_MATCH_HT) {
622 	cpumask_t mask;
623 
624 	mask = get_cpumask_from_level(sc->cst_cpuid, CORE_LEVEL);
625 	if (CPUMASK_TESTNZERO(mask)) {
626 	    int cpu;
627 
628 	    for (cpu = 0; cpu < ncpus; ++cpu) {
629 		struct acpi_cst_softc *sc1 = acpi_cst_softc[cpu];
630 
631 		if (sc1 == NULL || sc1 == sc ||
632 		    (sc1->cst_flags & ACPI_CST_FLAG_ATTACHED) == 0 ||
633 		    (sc1->cst_flags & ACPI_CST_FLAG_MATCH_HT) == 0)
634 		    continue;
635 		if (!CPUMASK_TESTBIT(mask, sc1->cst_cpuid))
636 		    continue;
637 
638 		if (sc1->cst_cx_count != sc->cst_cx_count) {
639 		    struct acpi_cst_softc *src_sc, *dst_sc;
640 
641 		    if (bootverbose) {
642 			device_printf(sc->cst_dev,
643 			    "inconstent C-state count: %d, %s has %d\n",
644 			    sc->cst_cx_count,
645 			    device_get_nameunit(sc1->cst_dev),
646 			    sc1->cst_cx_count);
647 		    }
648 		    if (sc1->cst_cx_count > sc->cst_cx_count) {
649 			src_sc = sc1;
650 			dst_sc = sc;
651 		    } else {
652 			src_sc = sc;
653 			dst_sc = sc1;
654 		    }
655 		    acpi_cst_copy(dst_sc, src_sc);
656 		}
657 	    }
658 	}
659     }
660 
661     if (reprobe) {
662 	/* If there are C3(+) states, always enable bus master wakeup */
663 	if ((acpi_cst_quirks & ACPI_CST_QUIRK_NO_BM) == 0) {
664 	    for (i = 0; i < sc->cst_cx_count; ++i) {
665 		struct acpi_cst_cx *cx = &sc->cst_cx_states[i];
666 
667 		if (cx->type >= ACPI_STATE_C3) {
668 		    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
669 		    break;
670 		}
671 	    }
672 	}
673 
674 	/* Fix up the lowest Cx being used */
675 	acpi_cst_set_lowest_oncpu(sc, sc->cst_cx_lowest_req);
676     }
677 
678     /*
679      * Cache the lowest non-C3 state.
680      * NOTE: must after cst_cx_lowest is set.
681      */
682     acpi_cst_non_c3(sc);
683 
684     cpu_sfence();
685     sc->cst_flags &= ~ACPI_CST_FLAG_PROBING;
686 
687     return (0);
688 }
689 
690 static void
691 acpi_cst_cx_reprobe_cst_handler(struct cpuhelper_msg *msg)
692 {
693     int error;
694 
695     error = acpi_cst_cx_probe_cst(msg->ch_cbarg, 1);
696     cpuhelper_replymsg(msg, error);
697 }
698 
699 static int
700 acpi_cst_cx_reprobe_cst(struct acpi_cst_softc *sc)
701 {
702     struct cpuhelper_msg msg;
703 
704     cpuhelper_initmsg(&msg, &curthread->td_msgport,
705         acpi_cst_cx_reprobe_cst_handler, sc, MSGF_PRIORITY);
706     return (cpuhelper_domsg(&msg, sc->cst_cpuid));
707 }
708 
709 /*
710  * Call this *after* all CPUs Cx states have been attached.
711  */
712 static void
713 acpi_cst_postattach(void *arg)
714 {
715     struct acpi_cst_softc *sc;
716     int i;
717 
718     /* Get set of Cx state devices */
719     devclass_get_devices(acpi_cst_devclass, &acpi_cst_devices,
720 	&acpi_cst_ndevices);
721 
722     /*
723      * Setup any quirks that might necessary now that we have probed
724      * all the CPUs' Cx states.
725      */
726     acpi_cst_set_quirks();
727 
728     if (acpi_cst_use_fadt) {
729 	/*
730 	 * We are using Cx mode from FADT, probe for available Cx states
731 	 * for all processors.
732 	 */
733 	for (i = 0; i < acpi_cst_ndevices; i++) {
734 	    sc = device_get_softc(acpi_cst_devices[i]);
735 	    acpi_cst_cx_probe_fadt(sc);
736 	}
737     } else {
738 	/*
739 	 * We are using _CST mode, remove C3 state if necessary.
740 	 *
741 	 * As we now know for sure that we will be using _CST mode
742 	 * install our notify handler.
743 	 */
744 	for (i = 0; i < acpi_cst_ndevices; i++) {
745 	    sc = device_get_softc(acpi_cst_devices[i]);
746 	    if (acpi_cst_quirks & ACPI_CST_QUIRK_NO_C3) {
747 		/* Free part of unused resources */
748 		acpi_cst_free_resource(sc, sc->cst_non_c3 + 1);
749 		sc->cst_cx_count = sc->cst_non_c3 + 1;
750 	    }
751 	    sc->cst_parent->cpu_cst_notify = acpi_cst_notify;
752 	}
753     }
754     acpi_cst_global_cx_count();
755 
756     /* Perform Cx final initialization. */
757     for (i = 0; i < acpi_cst_ndevices; i++) {
758 	sc = device_get_softc(acpi_cst_devices[i]);
759 	acpi_cst_startup(sc);
760 
761 	if (sc->cst_parent->glob_sysctl_tree != NULL) {
762 	    struct acpi_cpu_softc *cpu = sc->cst_parent;
763 
764 	    /* Add a sysctl handler to handle global Cx lowest setting */
765 	    SYSCTL_ADD_PROC(&cpu->glob_sysctl_ctx,
766 	    		    SYSCTL_CHILDREN(cpu->glob_sysctl_tree),
767 			    OID_AUTO, "cx_lowest",
768 			    CTLTYPE_STRING | CTLFLAG_RW, NULL, 0,
769 			    acpi_cst_global_lowest_sysctl, "A",
770 			    "Requested global lowest Cx sleep state");
771 	    SYSCTL_ADD_PROC(&cpu->glob_sysctl_ctx,
772 	    		    SYSCTL_CHILDREN(cpu->glob_sysctl_tree),
773 			    OID_AUTO, "cx_lowest_use",
774 			    CTLTYPE_STRING | CTLFLAG_RD, NULL, 0,
775 			    acpi_cst_global_lowest_use_sysctl, "A",
776 			    "Global lowest Cx sleep state to use");
777 	}
778     }
779 
780     /* Take over idling from cpu_idle_default(). */
781     acpi_cst_cx_lowest = 0;
782     acpi_cst_cx_lowest_req = 0;
783     acpi_cst_disable_idle = FALSE;
784 
785     cpu_sfence();
786     cpu_idle_hook = acpi_cst_idle;
787 }
788 
789 static void
790 acpi_cst_support_list(struct acpi_cst_softc *sc)
791 {
792     struct sbuf sb;
793     int i;
794 
795     /*
796      * Set up the list of Cx states
797      */
798     sbuf_new(&sb, sc->cst_cx_supported, sizeof(sc->cst_cx_supported),
799 	SBUF_FIXEDLEN);
800     for (i = 0; i < sc->cst_cx_count; i++)
801 	sbuf_printf(&sb, "C%d/%d ", i + 1, sc->cst_cx_states[i].trans_lat);
802     sbuf_trim(&sb);
803     sbuf_finish(&sb);
804 }
805 
806 static void
807 acpi_cst_c3_bm_rld_handler(struct cpuhelper_msg *msg)
808 {
809 
810     AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
811     cpuhelper_replymsg(msg, 0);
812 }
813 
814 static void
815 acpi_cst_c3_bm_rld(struct acpi_cst_softc *sc)
816 {
817     struct cpuhelper_msg msg;
818 
819     cpuhelper_initmsg(&msg, &curthread->td_msgport,
820 	acpi_cst_c3_bm_rld_handler, sc, MSGF_PRIORITY);
821     cpuhelper_domsg(&msg, sc->cst_cpuid);
822 }
823 
824 static void
825 acpi_cst_startup(struct acpi_cst_softc *sc)
826 {
827     struct acpi_cpu_softc *cpu = sc->cst_parent;
828     int i, bm_rld_done = 0;
829 
830     for (i = 0; i < sc->cst_cx_count; ++i) {
831 	struct acpi_cst_cx *cx = &sc->cst_cx_states[i];
832 	int error;
833 
834 	/* If there are C3(+) states, always enable bus master wakeup */
835 	if (cx->type >= ACPI_STATE_C3 && !bm_rld_done &&
836 	    (acpi_cst_quirks & ACPI_CST_QUIRK_NO_BM) == 0) {
837 	    acpi_cst_c3_bm_rld(sc);
838 	    bm_rld_done = 1;
839 	}
840 
841 	/* Redo the Cx setup, since quirks have been changed */
842 	error = acpi_cst_cx_setup(cx);
843 	if (error)
844 	    panic("C%d startup setup failed: %d", i + 1, error);
845     }
846 
847     acpi_cst_support_list(sc);
848 
849     SYSCTL_ADD_STRING(&cpu->pcpu_sysctl_ctx,
850 		      SYSCTL_CHILDREN(cpu->pcpu_sysctl_tree),
851 		      OID_AUTO, "cx_supported", CTLFLAG_RD,
852 		      sc->cst_cx_supported, 0,
853 		      "Cx/microsecond values for supported Cx states");
854     SYSCTL_ADD_PROC(&cpu->pcpu_sysctl_ctx,
855 		    SYSCTL_CHILDREN(cpu->pcpu_sysctl_tree),
856 		    OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
857 		    (void *)sc, 0, acpi_cst_lowest_sysctl, "A",
858 		    "requested lowest Cx sleep state");
859     SYSCTL_ADD_PROC(&cpu->pcpu_sysctl_ctx,
860 		    SYSCTL_CHILDREN(cpu->pcpu_sysctl_tree),
861 		    OID_AUTO, "cx_lowest_use", CTLTYPE_STRING | CTLFLAG_RD,
862 		    (void *)sc, 0, acpi_cst_lowest_use_sysctl, "A",
863 		    "lowest Cx sleep state to use");
864     SYSCTL_ADD_PROC(&cpu->pcpu_sysctl_ctx,
865 		    SYSCTL_CHILDREN(cpu->pcpu_sysctl_tree),
866 		    OID_AUTO, "cx_usage", CTLTYPE_STRING | CTLFLAG_RD,
867 		    (void *)sc, 0, acpi_cst_usage_sysctl, "A",
868 		    "percent usage for each Cx state");
869 
870 #ifdef notyet
871     /* Signal platform that we can handle _CST notification. */
872     if (!acpi_cst_use_fadt && acpi_cst_ctrl != 0) {
873 	ACPI_LOCK(acpi);
874 	AcpiOsWritePort(acpi_cst_smi_cmd, acpi_cst_ctrl, 8);
875 	ACPI_UNLOCK(acpi);
876     }
877 #endif
878 }
879 
880 /*
881  * Idle the CPU in the lowest state possible.  This function is called with
882  * interrupts disabled.  Note that once it re-enables interrupts, a task
883  * switch can occur so do not access shared data (i.e. the softc) after
884  * interrupts are re-enabled.
885  */
886 static void
887 acpi_cst_idle(void)
888 {
889     struct	acpi_cst_softc *sc;
890     struct	acpi_cst_cx *cx_next;
891     union microtime_pcpu start, end;
892     int		cx_next_idx, i, tdiff, bm_arb_disabled = 0;
893 
894     /* If disabled, return immediately. */
895     if (acpi_cst_disable_idle) {
896 	ACPI_ENABLE_IRQS();
897 	return;
898     }
899 
900     /*
901      * Look up our CPU id to get our softc.  If it's NULL, we'll use C1
902      * since there is no Cx state for this processor.
903      */
904     sc = acpi_cst_softc[mdcpu->mi.gd_cpuid];
905     if (sc == NULL) {
906 	acpi_cst_c1_halt();
907 	return;
908     }
909 
910     /* Still probing; use C1 */
911     if (sc->cst_flags & ACPI_CST_FLAG_PROBING) {
912 	acpi_cst_c1_halt();
913 	return;
914     }
915 
916     /* Find the lowest state that has small enough latency. */
917     cx_next_idx = 0;
918     for (i = sc->cst_cx_lowest; i >= 0; i--) {
919 	if (sc->cst_cx_states[i].trans_lat * 3 <= sc->cst_prev_sleep) {
920 	    cx_next_idx = i;
921 	    break;
922 	}
923     }
924 
925     /*
926      * Check for bus master activity if needed for the selected state.
927      * If there was activity, clear the bit and use the lowest non-C3 state.
928      */
929     cx_next = &sc->cst_cx_states[cx_next_idx];
930     if (cx_next->flags & ACPI_CST_CX_FLAG_BM_STS) {
931 	int bm_active;
932 
933 	AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, &bm_active);
934 	if (bm_active != 0) {
935 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, 1);
936 	    cx_next_idx = sc->cst_non_c3;
937 	}
938     }
939 
940     /* Select the next state and update statistics. */
941     cx_next = &sc->cst_cx_states[cx_next_idx];
942     sc->cst_cx_stats[cx_next_idx]++;
943     KASSERT(cx_next->type != ACPI_STATE_C0, ("C0 sleep"));
944 
945     /*
946      * Execute HLT (or equivalent) and wait for an interrupt.  We can't
947      * calculate the time spent in C1 since the place we wake up is an
948      * ISR.  Assume we slept half of quantum and return.
949      */
950     if (cx_next->type == ACPI_STATE_C1) {
951 	sc->cst_prev_sleep = (sc->cst_prev_sleep * 3 + 500000 / hz) / 4;
952 	cx_next->enter(cx_next);
953 	return;
954     }
955 
956     /* Execute the proper preamble before enter the selected state. */
957     if (cx_next->preamble == ACPI_CST_CX_PREAMBLE_BM_ARB) {
958 	AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 1);
959 	bm_arb_disabled = 1;
960     } else if (cx_next->preamble == ACPI_CST_CX_PREAMBLE_WBINVD) {
961 	ACPI_FLUSH_CPU_CACHE();
962     }
963 
964     /*
965      * Enter the selected state and check time spent asleep.
966      */
967     microtime_pcpu_get(&start);
968     cpu_mfence();
969 
970     cx_next->enter(cx_next);
971 
972     cpu_mfence();
973     microtime_pcpu_get(&end);
974 
975     /* Enable bus master arbitration, if it was disabled. */
976     if (bm_arb_disabled)
977 	AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 0);
978 
979     ACPI_ENABLE_IRQS();
980 
981     /* Find the actual time asleep in microseconds. */
982     tdiff = microtime_pcpu_diff(&start, &end);
983     sc->cst_prev_sleep = (sc->cst_prev_sleep * 3 + tdiff) / 4;
984 }
985 
986 /*
987  * Re-evaluate the _CST object when we are notified that it changed.
988  */
989 static void
990 acpi_cst_notify(device_t dev)
991 {
992     struct acpi_cst_softc *sc = device_get_softc(dev);
993 
994     cpuhelper_assert(mycpuid, false);
995 
996     lwkt_serialize_enter(&acpi_cst_slize);
997 
998     /* Update the list of Cx states. */
999     acpi_cst_cx_reprobe_cst(sc);
1000     acpi_cst_support_list(sc);
1001 
1002     /* Update the new lowest useable Cx state for all CPUs. */
1003     acpi_cst_global_cx_count();
1004 
1005     /*
1006      * Fix up the lowest Cx being used
1007      */
1008     if (acpi_cst_cx_lowest_req < acpi_cst_cx_count)
1009 	acpi_cst_cx_lowest = acpi_cst_cx_lowest_req;
1010     if (acpi_cst_cx_lowest > acpi_cst_cx_count - 1)
1011 	acpi_cst_cx_lowest = acpi_cst_cx_count - 1;
1012 
1013     lwkt_serialize_exit(&acpi_cst_slize);
1014 }
1015 
1016 static int
1017 acpi_cst_set_quirks(void)
1018 {
1019     device_t acpi_dev;
1020     uint32_t val;
1021 
1022     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
1023 
1024     /*
1025      * Bus mastering arbitration control is needed to keep caches coherent
1026      * while sleeping in C3.  If it's not present but a working flush cache
1027      * instruction is present, flush the caches before entering C3 instead.
1028      * Otherwise, just disable C3 completely.
1029      */
1030     if (AcpiGbl_FADT.Pm2ControlBlock == 0 ||
1031 	AcpiGbl_FADT.Pm2ControlLength == 0) {
1032 	if ((AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD) &&
1033 	    (AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD_FLUSH) == 0) {
1034 	    acpi_cst_quirks |= ACPI_CST_QUIRK_NO_BM;
1035 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1036 		"cpu_cst: no BM control, using flush cache method\n"));
1037 	} else {
1038 	    acpi_cst_quirks |= ACPI_CST_QUIRK_NO_C3;
1039 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1040 		"cpu_cst: no BM control, C3 not available\n"));
1041 	}
1042     }
1043 
1044     /* Look for various quirks of the PIIX4 part. */
1045     acpi_dev = pci_find_device(PCI_VENDOR_INTEL, PCI_DEVICE_82371AB_3);
1046     if (acpi_dev != NULL) {
1047 	switch (pci_get_revid(acpi_dev)) {
1048 	/*
1049 	 * Disable C3 support for all PIIX4 chipsets.  Some of these parts
1050 	 * do not report the BMIDE status to the BM status register and
1051 	 * others have a livelock bug if Type-F DMA is enabled.  Linux
1052 	 * works around the BMIDE bug by reading the BM status directly
1053 	 * but we take the simpler approach of disabling C3 for these
1054 	 * parts.
1055 	 *
1056 	 * See erratum #18 ("C3 Power State/BMIDE and Type-F DMA
1057 	 * Livelock") from the January 2002 PIIX4 specification update.
1058 	 * Applies to all PIIX4 models.
1059 	 *
1060 	 * Also, make sure that all interrupts cause a "Stop Break"
1061 	 * event to exit from C2 state.
1062 	 * Also, BRLD_EN_BM (ACPI_BITREG_BUS_MASTER_RLD in ACPI-speak)
1063 	 * should be set to zero, otherwise it causes C2 to short-sleep.
1064 	 * PIIX4 doesn't properly support C3 and bus master activity
1065 	 * need not break out of C2.
1066 	 */
1067 	case PCI_REVISION_A_STEP:
1068 	case PCI_REVISION_B_STEP:
1069 	case PCI_REVISION_4E:
1070 	case PCI_REVISION_4M:
1071 	    acpi_cst_quirks |= ACPI_CST_QUIRK_NO_C3;
1072 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1073 		"cpu_cst: working around PIIX4 bug, disabling C3\n"));
1074 
1075 	    val = pci_read_config(acpi_dev, PIIX4_DEVACTB_REG, 4);
1076 	    if ((val & PIIX4_STOP_BREAK_MASK) != PIIX4_STOP_BREAK_MASK) {
1077 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1078 		    "cpu_cst: PIIX4: enabling IRQs to generate Stop Break\n"));
1079 	    	val |= PIIX4_STOP_BREAK_MASK;
1080 		pci_write_config(acpi_dev, PIIX4_DEVACTB_REG, val, 4);
1081 	    }
1082 	    AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_RLD, &val);
1083 	    if (val) {
1084 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1085 		    "cpu_cst: PIIX4: reset BRLD_EN_BM\n"));
1086 		AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1087 	    }
1088 	    break;
1089 	default:
1090 	    break;
1091 	}
1092     }
1093 
1094     return (0);
1095 }
1096 
1097 static int
1098 acpi_cst_usage_sysctl(SYSCTL_HANDLER_ARGS)
1099 {
1100     struct acpi_cst_softc *sc;
1101     struct sbuf	 sb;
1102     char	 buf[128];
1103     int		 i;
1104     uintmax_t	 fract, sum, whole;
1105 
1106     sc = (struct acpi_cst_softc *) arg1;
1107     sum = 0;
1108     for (i = 0; i < sc->cst_cx_count; i++)
1109 	sum += sc->cst_cx_stats[i];
1110     sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1111     for (i = 0; i < sc->cst_cx_count; i++) {
1112 	if (sum > 0) {
1113 	    whole = (uintmax_t)sc->cst_cx_stats[i] * 100;
1114 	    fract = (whole % sum) * 100;
1115 	    sbuf_printf(&sb, "%u.%02u%% ", (u_int)(whole / sum),
1116 		(u_int)(fract / sum));
1117 	} else
1118 	    sbuf_printf(&sb, "0.00%% ");
1119     }
1120     sbuf_printf(&sb, "last %dus", sc->cst_prev_sleep);
1121     sbuf_trim(&sb);
1122     sbuf_finish(&sb);
1123     sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1124     sbuf_delete(&sb);
1125 
1126     return (0);
1127 }
1128 
1129 static int
1130 acpi_cst_set_lowest_oncpu(struct acpi_cst_softc *sc, int val)
1131 {
1132     int old_lowest, error = 0, old_lowest_req;
1133     uint32_t old_type, type;
1134 
1135     KKASSERT(mycpuid == sc->cst_cpuid);
1136 
1137     old_lowest_req = sc->cst_cx_lowest_req;
1138     sc->cst_cx_lowest_req = val;
1139 
1140     if (val > sc->cst_cx_count - 1)
1141 	val = sc->cst_cx_count - 1;
1142     old_lowest = atomic_swap_int(&sc->cst_cx_lowest, val);
1143 
1144     old_type = sc->cst_cx_states[old_lowest].type;
1145     type = sc->cst_cx_states[val].type;
1146     if (old_type >= ACPI_STATE_C3 && type < ACPI_STATE_C3) {
1147 	cputimer_intr_powersave_remreq();
1148     } else if (type >= ACPI_STATE_C3 && old_type < ACPI_STATE_C3) {
1149 	error = cputimer_intr_powersave_addreq();
1150 	if (error) {
1151 	    /* Restore */
1152 	    sc->cst_cx_lowest_req = old_lowest_req;
1153 	    sc->cst_cx_lowest = old_lowest;
1154 	}
1155     }
1156 
1157     if (error)
1158 	return error;
1159 
1160     /* Cache the new lowest non-C3 state. */
1161     acpi_cst_non_c3(sc);
1162 
1163     /* Reset the statistics counters. */
1164     bzero(sc->cst_cx_stats, sizeof(sc->cst_cx_stats));
1165     return (0);
1166 }
1167 
1168 static void
1169 acpi_cst_set_lowest_handler(struct cpuhelper_msg *msg)
1170 {
1171     int error;
1172 
1173     error = acpi_cst_set_lowest_oncpu(msg->ch_cbarg, msg->ch_cbarg1);
1174     cpuhelper_replymsg(msg, error);
1175 }
1176 
1177 static int
1178 acpi_cst_set_lowest(struct acpi_cst_softc *sc, int val)
1179 {
1180     struct cpuhelper_msg msg;
1181 
1182     cpuhelper_initmsg(&msg, &curthread->td_msgport,
1183 	acpi_cst_set_lowest_handler, sc, MSGF_PRIORITY);
1184     msg.ch_cbarg1 = val;
1185 
1186     return (cpuhelper_domsg(&msg, sc->cst_cpuid));
1187 }
1188 
1189 static int
1190 acpi_cst_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1191 {
1192     struct	 acpi_cst_softc *sc;
1193     char	 state[8];
1194     int		 val, error;
1195 
1196     sc = (struct acpi_cst_softc *)arg1;
1197     ksnprintf(state, sizeof(state), "C%d", sc->cst_cx_lowest_req + 1);
1198     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1199     if (error != 0 || req->newptr == NULL)
1200 	return (error);
1201     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1202 	return (EINVAL);
1203     val = (int) strtol(state + 1, NULL, 10) - 1;
1204     if (val < 0)
1205 	return (EINVAL);
1206 
1207     lwkt_serialize_enter(&acpi_cst_slize);
1208     error = acpi_cst_set_lowest(sc, val);
1209     lwkt_serialize_exit(&acpi_cst_slize);
1210 
1211     return error;
1212 }
1213 
1214 static int
1215 acpi_cst_lowest_use_sysctl(SYSCTL_HANDLER_ARGS)
1216 {
1217     struct	 acpi_cst_softc *sc;
1218     char	 state[8];
1219 
1220     sc = (struct acpi_cst_softc *)arg1;
1221     ksnprintf(state, sizeof(state), "C%d", sc->cst_cx_lowest + 1);
1222     return sysctl_handle_string(oidp, state, sizeof(state), req);
1223 }
1224 
1225 static int
1226 acpi_cst_global_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1227 {
1228     struct	acpi_cst_softc *sc;
1229     char	state[8];
1230     int		val, error, i;
1231 
1232     ksnprintf(state, sizeof(state), "C%d", acpi_cst_cx_lowest_req + 1);
1233     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1234     if (error != 0 || req->newptr == NULL)
1235 	return (error);
1236     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1237 	return (EINVAL);
1238     val = (int) strtol(state + 1, NULL, 10) - 1;
1239     if (val < 0)
1240 	return (EINVAL);
1241 
1242     lwkt_serialize_enter(&acpi_cst_slize);
1243 
1244     acpi_cst_cx_lowest_req = val;
1245     acpi_cst_cx_lowest = val;
1246     if (acpi_cst_cx_lowest > acpi_cst_cx_count - 1)
1247 	acpi_cst_cx_lowest = acpi_cst_cx_count - 1;
1248 
1249     /* Update the new lowest useable Cx state for all CPUs. */
1250     for (i = 0; i < acpi_cst_ndevices; i++) {
1251 	sc = device_get_softc(acpi_cst_devices[i]);
1252 	error = acpi_cst_set_lowest(sc, val);
1253 	if (error) {
1254 	    KKASSERT(i == 0);
1255 	    break;
1256 	}
1257     }
1258 
1259     lwkt_serialize_exit(&acpi_cst_slize);
1260 
1261     return error;
1262 }
1263 
1264 static int
1265 acpi_cst_global_lowest_use_sysctl(SYSCTL_HANDLER_ARGS)
1266 {
1267     char	state[8];
1268 
1269     ksnprintf(state, sizeof(state), "C%d", acpi_cst_cx_lowest + 1);
1270     return sysctl_handle_string(oidp, state, sizeof(state), req);
1271 }
1272 
1273 /*
1274  * Put the CPU in C1 in a machine-dependant way.
1275  * XXX: shouldn't be here!
1276  */
1277 static void
1278 acpi_cst_c1_halt(void)
1279 {
1280     splz();
1281     if ((mycpu->gd_reqflags & RQF_IDLECHECK_WK_MASK) == 0)
1282         __asm __volatile("sti; hlt");
1283     else
1284         __asm __volatile("sti; pause");
1285 }
1286 
1287 static void
1288 acpi_cst_non_c3(struct acpi_cst_softc *sc)
1289 {
1290     int i;
1291 
1292     sc->cst_non_c3 = 0;
1293     for (i = sc->cst_cx_lowest; i >= 0; i--) {
1294 	if (sc->cst_cx_states[i].type < ACPI_STATE_C3) {
1295 	    sc->cst_non_c3 = i;
1296 	    break;
1297 	}
1298     }
1299     if (bootverbose)
1300 	device_printf(sc->cst_dev, "non-C3 %d\n", sc->cst_non_c3);
1301 }
1302 
1303 /*
1304  * Update the largest Cx state supported in the global acpi_cst_cx_count.
1305  * It will be used in the global Cx sysctl handler.
1306  */
1307 static void
1308 acpi_cst_global_cx_count(void)
1309 {
1310     struct acpi_cst_softc *sc;
1311     int i;
1312 
1313     if (acpi_cst_ndevices == 0) {
1314 	acpi_cst_cx_count = 0;
1315 	return;
1316     }
1317 
1318     sc = device_get_softc(acpi_cst_devices[0]);
1319     acpi_cst_cx_count = sc->cst_cx_count;
1320 
1321     for (i = 1; i < acpi_cst_ndevices; i++) {
1322 	struct acpi_cst_softc *sc = device_get_softc(acpi_cst_devices[i]);
1323 
1324 	if (sc->cst_cx_count < acpi_cst_cx_count)
1325 	    acpi_cst_cx_count = sc->cst_cx_count;
1326     }
1327     if (bootverbose)
1328 	kprintf("cpu_cst: global Cx count %d\n", acpi_cst_cx_count);
1329 }
1330 
1331 static void
1332 acpi_cst_c1_halt_enter(const struct acpi_cst_cx *cx __unused)
1333 {
1334     acpi_cst_c1_halt();
1335 }
1336 
1337 static void
1338 acpi_cst_cx_io_enter(const struct acpi_cst_cx *cx)
1339 {
1340     uint64_t dummy;
1341 
1342     /*
1343      * Read I/O to enter this Cx state
1344      */
1345     bus_space_read_1(cx->btag, cx->bhand, 0);
1346     /*
1347      * Perform a dummy I/O read.  Since it may take an arbitrary time
1348      * to enter the idle state, this read makes sure that we are frozen.
1349      */
1350     AcpiRead(&dummy, &AcpiGbl_FADT.XPmTimerBlock);
1351 }
1352 
1353 static int
1354 acpi_cst_cx_setup(struct acpi_cst_cx *cx)
1355 {
1356     cx->flags &= ~ACPI_CST_CX_FLAG_BM_STS;
1357     cx->preamble = ACPI_CST_CX_PREAMBLE_NONE;
1358 
1359     if (cx->type >= ACPI_STATE_C3) {
1360 	/*
1361 	 * Set the required operations for entering C3(+) state.
1362 	 * Later acpi_cst_md_cx_setup() may fix them up.
1363 	 */
1364 
1365 	/*
1366 	 * Always check BM_STS.
1367 	 */
1368 	if ((acpi_cst_quirks & ACPI_CST_QUIRK_NO_BM) == 0)
1369 	    cx->flags |= ACPI_CST_CX_FLAG_BM_STS;
1370 
1371 	/*
1372 	 * According to the ACPI specification, bus master arbitration
1373 	 * is only available on UP system.  For MP system, cache flushing
1374 	 * is required.
1375 	 */
1376 	if (ncpus == 1 && (acpi_cst_quirks & ACPI_CST_QUIRK_NO_BM) == 0)
1377 	    cx->preamble = ACPI_CST_CX_PREAMBLE_BM_ARB;
1378 	else
1379 	    cx->preamble = ACPI_CST_CX_PREAMBLE_WBINVD;
1380     }
1381     return acpi_cst_md_cx_setup(cx);
1382 }
1383 
1384 static void
1385 acpi_cst_free_resource(struct acpi_cst_softc *sc, int start)
1386 {
1387     int i;
1388 
1389     for (i = start; i < MAX_CX_STATES; ++i) {
1390 	struct acpi_cst_cx *cx = &sc->cst_cx_states[i];
1391 
1392 	if (cx->res != NULL)
1393 	    bus_release_resource(sc->cst_dev, cx->res_type, cx->rid, cx->res);
1394 	memset(cx, 0, sizeof(*cx));
1395     }
1396 }
1397