xref: /dragonfly/sys/dev/acpica/acpi_cpu_cstate.c (revision e0b1d537)
1 /*-
2  * Copyright (c) 2003-2005 Nate Lawson (SDG)
3  * Copyright (c) 2001 Michael Smith
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  *
27  * $FreeBSD: src/sys/dev/acpica/acpi_cpu.c,v 1.72 2008/04/12 12:06:00 rpaulo Exp $
28  */
29 
30 #include "opt_acpi.h"
31 #include <sys/param.h>
32 #include <sys/bus.h>
33 #include <sys/kernel.h>
34 #include <sys/malloc.h>
35 #include <sys/globaldata.h>
36 #include <sys/power.h>
37 #include <sys/proc.h>
38 #include <sys/sbuf.h>
39 #include <sys/thread2.h>
40 #include <sys/serialize.h>
41 #include <sys/msgport2.h>
42 #include <sys/microtime_pcpu.h>
43 
44 #include <bus/pci/pcivar.h>
45 #include <machine/atomic.h>
46 #include <machine/globaldata.h>
47 #include <machine/md_var.h>
48 #include <machine/smp.h>
49 #include <sys/rman.h>
50 
51 #include <net/netisr2.h>
52 #include <net/netmsg2.h>
53 #include <net/if_var.h>
54 
55 #include "acpi.h"
56 #include "acpivar.h"
57 #include "acpi_cpu.h"
58 #include "acpi_cpu_cstate.h"
59 
60 /*
61  * Support for ACPI Processor devices, including C[1-3+] sleep states.
62  */
63 
64 /* Hooks for the ACPI CA debugging infrastructure */
65 #define _COMPONENT	ACPI_PROCESSOR
66 ACPI_MODULE_NAME("PROCESSOR")
67 
68 struct netmsg_acpi_cst {
69 	struct netmsg_base base;
70 	struct acpi_cst_softc *sc;
71 	int		val;
72 };
73 
74 #define MAX_CX_STATES	 8
75 
76 struct acpi_cst_softc {
77     device_t		cst_dev;
78     struct acpi_cpu_softc *cst_parent;
79     ACPI_HANDLE		cst_handle;
80     int			cst_cpuid;
81     uint32_t		cst_flags;	/* ACPI_CST_FLAG_ */
82     uint32_t		cst_p_blk;	/* ACPI P_BLK location */
83     uint32_t		cst_p_blk_len;	/* P_BLK length (must be 6). */
84     struct acpi_cst_cx	cst_cx_states[MAX_CX_STATES];
85     int			cst_cx_count;	/* Number of valid Cx states. */
86     int			cst_prev_sleep;	/* Last idle sleep duration. */
87     /* Runtime state. */
88     int			cst_non_c3;	/* Index of lowest non-C3 state. */
89     u_long		cst_cx_stats[MAX_CX_STATES];/* Cx usage history. */
90     /* Values for sysctl. */
91     int			cst_cx_lowest;	/* Current Cx lowest */
92     int			cst_cx_lowest_req; /* Requested Cx lowest */
93     char 		cst_cx_supported[64];
94 };
95 
96 #define ACPI_CST_FLAG_PROBING	0x1
97 
98 #define PCI_VENDOR_INTEL	0x8086
99 #define PCI_DEVICE_82371AB_3	0x7113	/* PIIX4 chipset for quirks. */
100 #define PCI_REVISION_A_STEP	0
101 #define PCI_REVISION_B_STEP	1
102 #define PCI_REVISION_4E		2
103 #define PCI_REVISION_4M		3
104 #define PIIX4_DEVACTB_REG	0x58
105 #define PIIX4_BRLD_EN_IRQ0	(1<<0)
106 #define PIIX4_BRLD_EN_IRQ	(1<<1)
107 #define PIIX4_BRLD_EN_IRQ8	(1<<5)
108 #define PIIX4_STOP_BREAK_MASK	(PIIX4_BRLD_EN_IRQ0 | \
109 				 PIIX4_BRLD_EN_IRQ | \
110 				 PIIX4_BRLD_EN_IRQ8)
111 #define PIIX4_PCNTRL_BST_EN	(1<<10)
112 
113 /* Platform hardware resource information. */
114 static uint32_t		 acpi_cst_smi_cmd; /* Value to write to SMI_CMD. */
115 static uint8_t		 acpi_cst_ctrl;	/* Indicate we are _CST aware. */
116 int		 	 acpi_cst_quirks; /* Indicate any hardware bugs. */
117 static boolean_t	 acpi_cst_use_fadt;
118 
119 /* Runtime state. */
120 static boolean_t	 acpi_cst_disable_idle;
121 					/* Disable entry to idle function */
122 static int		 acpi_cst_cx_count; /* Number of valid Cx states */
123 
124 /* Values for sysctl. */
125 static int		 acpi_cst_cx_lowest; /* Current Cx lowest */
126 static int		 acpi_cst_cx_lowest_req; /* Requested Cx lowest */
127 
128 static device_t		*acpi_cst_devices;
129 static int		 acpi_cst_ndevices;
130 static struct acpi_cst_softc **acpi_cst_softc;
131 static struct lwkt_serialize acpi_cst_slize = LWKT_SERIALIZE_INITIALIZER;
132 
133 static int	acpi_cst_probe(device_t);
134 static int	acpi_cst_attach(device_t);
135 static int	acpi_cst_suspend(device_t);
136 static int	acpi_cst_resume(device_t);
137 static int	acpi_cst_shutdown(device_t);
138 
139 static void	acpi_cst_notify(device_t);
140 static void	acpi_cst_postattach(void *);
141 static void	acpi_cst_idle(void);
142 
143 static void	acpi_cst_cx_probe(struct acpi_cst_softc *);
144 static void	acpi_cst_cx_probe_fadt(struct acpi_cst_softc *);
145 static int	acpi_cst_cx_probe_cst(struct acpi_cst_softc *, int);
146 static int	acpi_cst_cx_reprobe_cst(struct acpi_cst_softc *);
147 
148 static void	acpi_cst_startup(struct acpi_cst_softc *);
149 static void	acpi_cst_support_list(struct acpi_cst_softc *);
150 static int	acpi_cst_set_lowest(struct acpi_cst_softc *, int);
151 static int	acpi_cst_set_lowest_oncpu(struct acpi_cst_softc *, int);
152 static void	acpi_cst_non_c3(struct acpi_cst_softc *);
153 static void	acpi_cst_global_cx_count(void);
154 static int	acpi_cst_set_quirks(void);
155 static void	acpi_cst_c3_bm_rld(struct acpi_cst_softc *);
156 static void	acpi_cst_free_resource(struct acpi_cst_softc *, int);
157 static void	acpi_cst_c1_halt(void);
158 
159 static int	acpi_cst_usage_sysctl(SYSCTL_HANDLER_ARGS);
160 static int	acpi_cst_lowest_sysctl(SYSCTL_HANDLER_ARGS);
161 static int	acpi_cst_lowest_use_sysctl(SYSCTL_HANDLER_ARGS);
162 static int	acpi_cst_global_lowest_sysctl(SYSCTL_HANDLER_ARGS);
163 static int	acpi_cst_global_lowest_use_sysctl(SYSCTL_HANDLER_ARGS);
164 
165 static int	acpi_cst_cx_setup(struct acpi_cst_cx *cx);
166 static void	acpi_cst_c1_halt_enter(const struct acpi_cst_cx *);
167 static void	acpi_cst_cx_io_enter(const struct acpi_cst_cx *);
168 
169 static device_method_t acpi_cst_methods[] = {
170     /* Device interface */
171     DEVMETHOD(device_probe,	acpi_cst_probe),
172     DEVMETHOD(device_attach,	acpi_cst_attach),
173     DEVMETHOD(device_detach,	bus_generic_detach),
174     DEVMETHOD(device_shutdown,	acpi_cst_shutdown),
175     DEVMETHOD(device_suspend,	acpi_cst_suspend),
176     DEVMETHOD(device_resume,	acpi_cst_resume),
177 
178     /* Bus interface */
179     DEVMETHOD(bus_add_child,	bus_generic_add_child),
180     DEVMETHOD(bus_read_ivar,	bus_generic_read_ivar),
181     DEVMETHOD(bus_get_resource_list, bus_generic_get_resource_list),
182     DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
183     DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
184     DEVMETHOD(bus_alloc_resource, bus_generic_rl_alloc_resource),
185     DEVMETHOD(bus_release_resource, bus_generic_rl_release_resource),
186     DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
187     DEVMETHOD(bus_activate_resource, bus_generic_activate_resource),
188     DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource),
189     DEVMETHOD(bus_setup_intr,	bus_generic_setup_intr),
190     DEVMETHOD(bus_teardown_intr, bus_generic_teardown_intr),
191     DEVMETHOD_END
192 };
193 
194 static driver_t acpi_cst_driver = {
195     "cpu_cst",
196     acpi_cst_methods,
197     sizeof(struct acpi_cst_softc),
198 };
199 
200 static devclass_t acpi_cst_devclass;
201 DRIVER_MODULE(cpu_cst, cpu, acpi_cst_driver, acpi_cst_devclass, NULL, NULL);
202 MODULE_DEPEND(cpu_cst, acpi, 1, 1, 1);
203 
204 static int
205 acpi_cst_probe(device_t dev)
206 {
207     int cpu_id;
208 
209     if (acpi_disabled("cpu_cst") || acpi_get_type(dev) != ACPI_TYPE_PROCESSOR)
210 	return (ENXIO);
211 
212     cpu_id = acpi_get_magic(dev);
213 
214     if (acpi_cst_softc == NULL)
215 	acpi_cst_softc = kmalloc(sizeof(struct acpi_cst_softc *) *
216 	    SMP_MAXCPU, M_TEMP /* XXX */, M_INTWAIT | M_ZERO);
217 
218     /*
219      * Check if we already probed this processor.  We scan the bus twice
220      * so it's possible we've already seen this one.
221      */
222     if (acpi_cst_softc[cpu_id] != NULL) {
223 	device_printf(dev, "CPU%d cstate already exist\n", cpu_id);
224 	return (ENXIO);
225     }
226 
227     /* Mark this processor as in-use and save our derived id for attach. */
228     acpi_cst_softc[cpu_id] = (void *)1;
229     device_set_desc(dev, "ACPI CPU C-State");
230 
231     return (0);
232 }
233 
234 static int
235 acpi_cst_attach(device_t dev)
236 {
237     ACPI_BUFFER		   buf;
238     ACPI_OBJECT		   *obj;
239     struct acpi_cst_softc *sc;
240     ACPI_STATUS		   status;
241 
242     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
243 
244     sc = device_get_softc(dev);
245     sc->cst_dev = dev;
246     sc->cst_parent = device_get_softc(device_get_parent(dev));
247     sc->cst_handle = acpi_get_handle(dev);
248     sc->cst_cpuid = acpi_get_magic(dev);
249     acpi_cst_softc[sc->cst_cpuid] = sc;
250     acpi_cst_smi_cmd = AcpiGbl_FADT.SmiCommand;
251     acpi_cst_ctrl = AcpiGbl_FADT.CstControl;
252 
253     buf.Pointer = NULL;
254     buf.Length = ACPI_ALLOCATE_BUFFER;
255     status = AcpiEvaluateObject(sc->cst_handle, NULL, NULL, &buf);
256     if (ACPI_FAILURE(status)) {
257 	device_printf(dev, "attach failed to get Processor obj - %s\n",
258 		      AcpiFormatException(status));
259 	return (ENXIO);
260     }
261     obj = (ACPI_OBJECT *)buf.Pointer;
262     sc->cst_p_blk = obj->Processor.PblkAddress;
263     sc->cst_p_blk_len = obj->Processor.PblkLength;
264     AcpiOsFree(obj);
265     ACPI_DEBUG_PRINT((ACPI_DB_INFO, "cpu_cst%d: P_BLK at %#x/%d\n",
266 		     device_get_unit(dev), sc->cst_p_blk, sc->cst_p_blk_len));
267 
268     /*
269      * If this is the first cpu we attach, create and initialize the generic
270      * resources that will be used by all acpi cpu devices.
271      */
272     if (device_get_unit(dev) == 0) {
273 	/* Assume we won't be using FADT for Cx states by default */
274 	acpi_cst_use_fadt = FALSE;
275 
276 	/* Queue post cpu-probing task handler */
277 	AcpiOsExecute(OSL_NOTIFY_HANDLER, acpi_cst_postattach, NULL);
278     }
279 
280     /* Probe for Cx state support. */
281     acpi_cst_cx_probe(sc);
282 
283     return (0);
284 }
285 
286 /*
287  * Disable any entry to the idle function during suspend and re-enable it
288  * during resume.
289  */
290 static int
291 acpi_cst_suspend(device_t dev)
292 {
293     int error;
294 
295     error = bus_generic_suspend(dev);
296     if (error)
297 	return (error);
298     acpi_cst_disable_idle = TRUE;
299     return (0);
300 }
301 
302 static int
303 acpi_cst_resume(device_t dev)
304 {
305     acpi_cst_disable_idle = FALSE;
306     return (bus_generic_resume(dev));
307 }
308 
309 static int
310 acpi_cst_shutdown(device_t dev)
311 {
312     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
313 
314     /* Allow children to shutdown first. */
315     bus_generic_shutdown(dev);
316 
317     /*
318      * Disable any entry to the idle function.  There is a small race where
319      * an idle thread have passed this check but not gone to sleep.  This
320      * is ok since device_shutdown() does not free the softc, otherwise
321      * we'd have to be sure all threads were evicted before returning.
322      */
323     acpi_cst_disable_idle = TRUE;
324 
325     return_VALUE (0);
326 }
327 
328 static void
329 acpi_cst_cx_probe(struct acpi_cst_softc *sc)
330 {
331     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
332 
333     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
334     sc->cst_prev_sleep = 1000000;
335     sc->cst_cx_lowest = 0;
336     sc->cst_cx_lowest_req = 0;
337 
338     /*
339      * Check for the ACPI 2.0 _CST sleep states object.  If we can't find
340      * any, we'll revert to FADT/P_BLK Cx control method which will be
341      * handled by acpi_cst_postattach.  We need to defer to after having
342      * probed all the cpus in the system before probing for Cx states from
343      * FADT as we may already have found cpus with valid _CST packages.
344      */
345     if (!acpi_cst_use_fadt && acpi_cst_cx_probe_cst(sc, 0) != 0) {
346 	/*
347 	 * We were unable to find a _CST package for this cpu or there
348 	 * was an error parsing it. Switch back to generic mode.
349 	 */
350 	acpi_cst_use_fadt = TRUE;
351 	if (bootverbose)
352 	    device_printf(sc->cst_dev, "switching to FADT Cx mode\n");
353     }
354 
355     /*
356      * TODO: _CSD Package should be checked here.
357      */
358 }
359 
360 static void
361 acpi_cst_cx_probe_fadt(struct acpi_cst_softc *sc)
362 {
363     struct acpi_cst_cx *cx_ptr;
364     int error;
365 
366     /*
367      * Free all previously allocated resources.
368      *
369      * NITE:
370      * It is needed, since we could enter here because of other
371      * cpu's _CST probing failure.
372      */
373     acpi_cst_free_resource(sc, 0);
374 
375     sc->cst_cx_count = 0;
376     cx_ptr = sc->cst_cx_states;
377 
378     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
379     sc->cst_prev_sleep = 1000000;
380 
381     /* C1 has been required since just after ACPI 1.0 */
382     cx_ptr->gas.SpaceId = ACPI_ADR_SPACE_FIXED_HARDWARE;
383     cx_ptr->type = ACPI_STATE_C1;
384     cx_ptr->trans_lat = 0;
385     cx_ptr->enter = acpi_cst_c1_halt_enter;
386     error = acpi_cst_cx_setup(cx_ptr);
387     if (error)
388 	panic("C1 FADT HALT setup failed: %d", error);
389     cx_ptr++;
390     sc->cst_cx_count++;
391 
392     /* C2(+) is not supported on MP system */
393     if (ncpus > 1 && (AcpiGbl_FADT.Flags & ACPI_FADT_C2_MP_SUPPORTED) == 0)
394 	return;
395 
396     /*
397      * The spec says P_BLK must be 6 bytes long.  However, some systems
398      * use it to indicate a fractional set of features present so we
399      * take 5 as C2.  Some may also have a value of 7 to indicate
400      * another C3 but most use _CST for this (as required) and having
401      * "only" C1-C3 is not a hardship.
402      */
403     if (sc->cst_p_blk_len < 5)
404 	return;
405 
406     /* Validate and allocate resources for C2 (P_LVL2). */
407     if (AcpiGbl_FADT.C2Latency <= 100) {
408 	cx_ptr->gas.SpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
409 	cx_ptr->gas.BitWidth = 8;
410 	cx_ptr->gas.Address = sc->cst_p_blk + 4;
411 
412 	cx_ptr->rid = sc->cst_parent->cpu_next_rid;
413 	acpi_bus_alloc_gas(sc->cst_dev, &cx_ptr->res_type, &cx_ptr->rid,
414 	    &cx_ptr->gas, &cx_ptr->res, RF_SHAREABLE);
415 	if (cx_ptr->res != NULL) {
416 	    sc->cst_parent->cpu_next_rid++;
417 	    cx_ptr->type = ACPI_STATE_C2;
418 	    cx_ptr->trans_lat = AcpiGbl_FADT.C2Latency;
419 	    cx_ptr->enter = acpi_cst_cx_io_enter;
420 	    cx_ptr->btag = rman_get_bustag(cx_ptr->res);
421 	    cx_ptr->bhand = rman_get_bushandle(cx_ptr->res);
422 	    error = acpi_cst_cx_setup(cx_ptr);
423 	    if (error)
424 		panic("C2 FADT I/O setup failed: %d", error);
425 	    cx_ptr++;
426 	    sc->cst_cx_count++;
427 	    sc->cst_non_c3 = 1;
428 	}
429     }
430     if (sc->cst_p_blk_len < 6)
431 	return;
432 
433     /* Validate and allocate resources for C3 (P_LVL3). */
434     if (AcpiGbl_FADT.C3Latency <= 1000 &&
435         !(acpi_cst_quirks & ACPI_CST_QUIRK_NO_C3)) {
436 	cx_ptr->gas.SpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
437 	cx_ptr->gas.BitWidth = 8;
438 	cx_ptr->gas.Address = sc->cst_p_blk + 5;
439 
440 	cx_ptr->rid = sc->cst_parent->cpu_next_rid;
441 	acpi_bus_alloc_gas(sc->cst_dev, &cx_ptr->res_type, &cx_ptr->rid,
442 	    &cx_ptr->gas, &cx_ptr->res, RF_SHAREABLE);
443 	if (cx_ptr->res != NULL) {
444 	    sc->cst_parent->cpu_next_rid++;
445 	    cx_ptr->type = ACPI_STATE_C3;
446 	    cx_ptr->trans_lat = AcpiGbl_FADT.C3Latency;
447 	    cx_ptr->enter = acpi_cst_cx_io_enter;
448 	    cx_ptr->btag = rman_get_bustag(cx_ptr->res);
449 	    cx_ptr->bhand = rman_get_bushandle(cx_ptr->res);
450 	    error = acpi_cst_cx_setup(cx_ptr);
451 	    if (error)
452 		panic("C3 FADT I/O setup failed: %d", error);
453 	    cx_ptr++;
454 	    sc->cst_cx_count++;
455 	}
456     }
457 }
458 
459 /*
460  * Parse a _CST package and set up its Cx states.  Since the _CST object
461  * can change dynamically, our notify handler may call this function
462  * to clean up and probe the new _CST package.
463  */
464 static int
465 acpi_cst_cx_probe_cst(struct acpi_cst_softc *sc, int reprobe)
466 {
467     struct	 acpi_cst_cx *cx_ptr;
468     ACPI_STATUS	 status;
469     ACPI_BUFFER	 buf;
470     ACPI_OBJECT	*top;
471     ACPI_OBJECT	*pkg;
472     uint32_t	 count;
473     int		 i;
474 
475     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
476 
477 #ifdef INVARIANTS
478     if (reprobe)
479 	KKASSERT(&curthread->td_msgport == netisr_cpuport(sc->cst_cpuid));
480 #endif
481 
482     buf.Pointer = NULL;
483     buf.Length = ACPI_ALLOCATE_BUFFER;
484     status = AcpiEvaluateObject(sc->cst_handle, "_CST", NULL, &buf);
485     if (ACPI_FAILURE(status))
486 	return (ENXIO);
487 
488     /* _CST is a package with a count and at least one Cx package. */
489     top = (ACPI_OBJECT *)buf.Pointer;
490     if (!ACPI_PKG_VALID(top, 2) || acpi_PkgInt32(top, 0, &count) != 0) {
491 	device_printf(sc->cst_dev, "invalid _CST package\n");
492 	AcpiOsFree(buf.Pointer);
493 	return (ENXIO);
494     }
495     if (count != top->Package.Count - 1) {
496 	device_printf(sc->cst_dev, "invalid _CST state count (%d != %d)\n",
497 	       count, top->Package.Count - 1);
498 	count = top->Package.Count - 1;
499     }
500     if (count > MAX_CX_STATES) {
501 	device_printf(sc->cst_dev, "_CST has too many states (%d)\n", count);
502 	count = MAX_CX_STATES;
503     }
504 
505     sc->cst_flags |= ACPI_CST_FLAG_PROBING;
506     cpu_sfence();
507 
508     /*
509      * Free all previously allocated resources
510      *
511      * NOTE: It is needed for _CST reprobing.
512      */
513     acpi_cst_free_resource(sc, 0);
514 
515     /* Set up all valid states. */
516     sc->cst_cx_count = 0;
517     cx_ptr = sc->cst_cx_states;
518     for (i = 0; i < count; i++) {
519 	int error;
520 
521 	pkg = &top->Package.Elements[i + 1];
522 	if (!ACPI_PKG_VALID(pkg, 4) ||
523 	    acpi_PkgInt32(pkg, 1, &cx_ptr->type) != 0 ||
524 	    acpi_PkgInt32(pkg, 2, &cx_ptr->trans_lat) != 0 ||
525 	    acpi_PkgInt32(pkg, 3, &cx_ptr->power) != 0) {
526 
527 	    device_printf(sc->cst_dev, "skipping invalid Cx state package\n");
528 	    continue;
529 	}
530 
531 	/* Validate the state to see if we should use it. */
532 	switch (cx_ptr->type) {
533 	case ACPI_STATE_C1:
534 	    sc->cst_non_c3 = i;
535 	    cx_ptr->enter = acpi_cst_c1_halt_enter;
536 	    error = acpi_cst_cx_setup(cx_ptr);
537 	    if (error)
538 		panic("C1 CST HALT setup failed: %d", error);
539 	    cx_ptr++;
540 	    sc->cst_cx_count++;
541 	    continue;
542 	case ACPI_STATE_C2:
543 	    sc->cst_non_c3 = i;
544 	    break;
545 	case ACPI_STATE_C3:
546 	default:
547 	    if ((acpi_cst_quirks & ACPI_CST_QUIRK_NO_C3) != 0) {
548 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
549 				 "cpu_cst%d: C3[%d] not available.\n",
550 				 device_get_unit(sc->cst_dev), i));
551 		continue;
552 	    }
553 	    break;
554 	}
555 
556 	/*
557 	 * Allocate the control register for C2 or C3(+).
558 	 */
559 	KASSERT(cx_ptr->res == NULL, ("still has res"));
560 	acpi_PkgRawGas(pkg, 0, &cx_ptr->gas);
561 
562 	cx_ptr->rid = sc->cst_parent->cpu_next_rid;
563 	acpi_bus_alloc_gas(sc->cst_dev, &cx_ptr->res_type, &cx_ptr->rid,
564 	    &cx_ptr->gas, &cx_ptr->res, RF_SHAREABLE);
565 	if (cx_ptr->res != NULL) {
566 	    sc->cst_parent->cpu_next_rid++;
567 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
568 			     "cpu_cst%d: Got C%d - %d latency\n",
569 			     device_get_unit(sc->cst_dev), cx_ptr->type,
570 			     cx_ptr->trans_lat));
571 	    cx_ptr->enter = acpi_cst_cx_io_enter;
572 	    cx_ptr->btag = rman_get_bustag(cx_ptr->res);
573 	    cx_ptr->bhand = rman_get_bushandle(cx_ptr->res);
574 	    error = acpi_cst_cx_setup(cx_ptr);
575 	    if (error)
576 		panic("C%d CST I/O setup failed: %d", cx_ptr->type, error);
577 	    cx_ptr++;
578 	    sc->cst_cx_count++;
579 	} else {
580 	    error = acpi_cst_cx_setup(cx_ptr);
581 	    if (!error) {
582 		KASSERT(cx_ptr->enter != NULL,
583 		    ("C%d enter is not set", cx_ptr->type));
584 		cx_ptr++;
585 		sc->cst_cx_count++;
586 	    }
587 	}
588     }
589     AcpiOsFree(buf.Pointer);
590 
591     if (reprobe) {
592 	/* If there are C3(+) states, always enable bus master wakeup */
593 	if ((acpi_cst_quirks & ACPI_CST_QUIRK_NO_BM) == 0) {
594 	    for (i = 0; i < sc->cst_cx_count; ++i) {
595 		struct acpi_cst_cx *cx = &sc->cst_cx_states[i];
596 
597 		if (cx->type >= ACPI_STATE_C3) {
598 		    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
599 		    break;
600 		}
601 	    }
602 	}
603 
604 	/* Fix up the lowest Cx being used */
605 	acpi_cst_set_lowest_oncpu(sc, sc->cst_cx_lowest_req);
606     }
607 
608     /*
609      * Cache the lowest non-C3 state.
610      * NOTE: must after cst_cx_lowest is set.
611      */
612     acpi_cst_non_c3(sc);
613 
614     cpu_sfence();
615     sc->cst_flags &= ~ACPI_CST_FLAG_PROBING;
616 
617     return (0);
618 }
619 
620 static void
621 acpi_cst_cx_reprobe_cst_handler(netmsg_t msg)
622 {
623     struct netmsg_acpi_cst *rmsg = (struct netmsg_acpi_cst *)msg;
624     int error;
625 
626     error = acpi_cst_cx_probe_cst(rmsg->sc, 1);
627     lwkt_replymsg(&rmsg->base.lmsg, error);
628 }
629 
630 static int
631 acpi_cst_cx_reprobe_cst(struct acpi_cst_softc *sc)
632 {
633     struct netmsg_acpi_cst msg;
634 
635     netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
636 	acpi_cst_cx_reprobe_cst_handler);
637     msg.sc = sc;
638 
639     return lwkt_domsg(netisr_cpuport(sc->cst_cpuid), &msg.base.lmsg, 0);
640 }
641 
642 /*
643  * Call this *after* all CPUs Cx states have been attached.
644  */
645 static void
646 acpi_cst_postattach(void *arg)
647 {
648     struct acpi_cst_softc *sc;
649     int i;
650 
651     /* Get set of Cx state devices */
652     devclass_get_devices(acpi_cst_devclass, &acpi_cst_devices,
653 	&acpi_cst_ndevices);
654 
655     /*
656      * Setup any quirks that might necessary now that we have probed
657      * all the CPUs' Cx states.
658      */
659     acpi_cst_set_quirks();
660 
661     if (acpi_cst_use_fadt) {
662 	/*
663 	 * We are using Cx mode from FADT, probe for available Cx states
664 	 * for all processors.
665 	 */
666 	for (i = 0; i < acpi_cst_ndevices; i++) {
667 	    sc = device_get_softc(acpi_cst_devices[i]);
668 	    acpi_cst_cx_probe_fadt(sc);
669 	}
670     } else {
671 	/*
672 	 * We are using _CST mode, remove C3 state if necessary.
673 	 *
674 	 * As we now know for sure that we will be using _CST mode
675 	 * install our notify handler.
676 	 */
677 	for (i = 0; i < acpi_cst_ndevices; i++) {
678 	    sc = device_get_softc(acpi_cst_devices[i]);
679 	    if (acpi_cst_quirks & ACPI_CST_QUIRK_NO_C3) {
680 		/* Free part of unused resources */
681 		acpi_cst_free_resource(sc, sc->cst_non_c3 + 1);
682 		sc->cst_cx_count = sc->cst_non_c3 + 1;
683 	    }
684 	    sc->cst_parent->cpu_cst_notify = acpi_cst_notify;
685 	}
686     }
687     acpi_cst_global_cx_count();
688 
689     /* Perform Cx final initialization. */
690     for (i = 0; i < acpi_cst_ndevices; i++) {
691 	sc = device_get_softc(acpi_cst_devices[i]);
692 	acpi_cst_startup(sc);
693 
694 	if (sc->cst_parent->glob_sysctl_tree != NULL) {
695 	    struct acpi_cpu_softc *cpu = sc->cst_parent;
696 
697 	    /* Add a sysctl handler to handle global Cx lowest setting */
698 	    SYSCTL_ADD_PROC(&cpu->glob_sysctl_ctx,
699 	    		    SYSCTL_CHILDREN(cpu->glob_sysctl_tree),
700 			    OID_AUTO, "cx_lowest",
701 			    CTLTYPE_STRING | CTLFLAG_RW, NULL, 0,
702 			    acpi_cst_global_lowest_sysctl, "A",
703 			    "Requested global lowest Cx sleep state");
704 	    SYSCTL_ADD_PROC(&cpu->glob_sysctl_ctx,
705 	    		    SYSCTL_CHILDREN(cpu->glob_sysctl_tree),
706 			    OID_AUTO, "cx_lowest_use",
707 			    CTLTYPE_STRING | CTLFLAG_RD, NULL, 0,
708 			    acpi_cst_global_lowest_use_sysctl, "A",
709 			    "Global lowest Cx sleep state to use");
710 	}
711     }
712 
713     /* Take over idling from cpu_idle_default(). */
714     acpi_cst_cx_lowest = 0;
715     acpi_cst_cx_lowest_req = 0;
716     acpi_cst_disable_idle = FALSE;
717 
718     cpu_sfence();
719     cpu_idle_hook = acpi_cst_idle;
720 }
721 
722 static void
723 acpi_cst_support_list(struct acpi_cst_softc *sc)
724 {
725     struct sbuf sb;
726     int i;
727 
728     /*
729      * Set up the list of Cx states
730      */
731     sbuf_new(&sb, sc->cst_cx_supported, sizeof(sc->cst_cx_supported),
732 	SBUF_FIXEDLEN);
733     for (i = 0; i < sc->cst_cx_count; i++)
734 	sbuf_printf(&sb, "C%d/%d ", i + 1, sc->cst_cx_states[i].trans_lat);
735     sbuf_trim(&sb);
736     sbuf_finish(&sb);
737 }
738 
739 static void
740 acpi_cst_c3_bm_rld_handler(netmsg_t msg)
741 {
742     struct netmsg_acpi_cst *rmsg = (struct netmsg_acpi_cst *)msg;
743 
744     AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
745     lwkt_replymsg(&rmsg->base.lmsg, 0);
746 }
747 
748 static void
749 acpi_cst_c3_bm_rld(struct acpi_cst_softc *sc)
750 {
751     struct netmsg_acpi_cst msg;
752 
753     netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
754 	acpi_cst_c3_bm_rld_handler);
755     msg.sc = sc;
756 
757     lwkt_domsg(netisr_cpuport(sc->cst_cpuid), &msg.base.lmsg, 0);
758 }
759 
760 static void
761 acpi_cst_startup(struct acpi_cst_softc *sc)
762 {
763     struct acpi_cpu_softc *cpu = sc->cst_parent;
764     int i, bm_rld_done = 0;
765 
766     for (i = 0; i < sc->cst_cx_count; ++i) {
767 	struct acpi_cst_cx *cx = &sc->cst_cx_states[i];
768 	int error;
769 
770 	/* If there are C3(+) states, always enable bus master wakeup */
771 	if (cx->type >= ACPI_STATE_C3 && !bm_rld_done &&
772 	    (acpi_cst_quirks & ACPI_CST_QUIRK_NO_BM) == 0) {
773 	    acpi_cst_c3_bm_rld(sc);
774 	    bm_rld_done = 1;
775 	}
776 
777 	/* Redo the Cx setup, since quirks have been changed */
778 	error = acpi_cst_cx_setup(cx);
779 	if (error)
780 	    panic("C%d startup setup failed: %d", i + 1, error);
781     }
782 
783     acpi_cst_support_list(sc);
784 
785     SYSCTL_ADD_STRING(&cpu->pcpu_sysctl_ctx,
786 		      SYSCTL_CHILDREN(cpu->pcpu_sysctl_tree),
787 		      OID_AUTO, "cx_supported", CTLFLAG_RD,
788 		      sc->cst_cx_supported, 0,
789 		      "Cx/microsecond values for supported Cx states");
790     SYSCTL_ADD_PROC(&cpu->pcpu_sysctl_ctx,
791 		    SYSCTL_CHILDREN(cpu->pcpu_sysctl_tree),
792 		    OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
793 		    (void *)sc, 0, acpi_cst_lowest_sysctl, "A",
794 		    "requested lowest Cx sleep state");
795     SYSCTL_ADD_PROC(&cpu->pcpu_sysctl_ctx,
796 		    SYSCTL_CHILDREN(cpu->pcpu_sysctl_tree),
797 		    OID_AUTO, "cx_lowest_use", CTLTYPE_STRING | CTLFLAG_RD,
798 		    (void *)sc, 0, acpi_cst_lowest_use_sysctl, "A",
799 		    "lowest Cx sleep state to use");
800     SYSCTL_ADD_PROC(&cpu->pcpu_sysctl_ctx,
801 		    SYSCTL_CHILDREN(cpu->pcpu_sysctl_tree),
802 		    OID_AUTO, "cx_usage", CTLTYPE_STRING | CTLFLAG_RD,
803 		    (void *)sc, 0, acpi_cst_usage_sysctl, "A",
804 		    "percent usage for each Cx state");
805 
806 #ifdef notyet
807     /* Signal platform that we can handle _CST notification. */
808     if (!acpi_cst_use_fadt && acpi_cst_ctrl != 0) {
809 	ACPI_LOCK(acpi);
810 	AcpiOsWritePort(acpi_cst_smi_cmd, acpi_cst_ctrl, 8);
811 	ACPI_UNLOCK(acpi);
812     }
813 #endif
814 }
815 
816 /*
817  * Idle the CPU in the lowest state possible.  This function is called with
818  * interrupts disabled.  Note that once it re-enables interrupts, a task
819  * switch can occur so do not access shared data (i.e. the softc) after
820  * interrupts are re-enabled.
821  */
822 static void
823 acpi_cst_idle(void)
824 {
825     struct	acpi_cst_softc *sc;
826     struct	acpi_cst_cx *cx_next;
827     union microtime_pcpu start, end;
828     int		cx_next_idx, i, tdiff, bm_arb_disabled = 0;
829 
830     /* If disabled, return immediately. */
831     if (acpi_cst_disable_idle) {
832 	ACPI_ENABLE_IRQS();
833 	return;
834     }
835 
836     /*
837      * Look up our CPU id to get our softc.  If it's NULL, we'll use C1
838      * since there is no Cx state for this processor.
839      */
840     sc = acpi_cst_softc[mdcpu->mi.gd_cpuid];
841     if (sc == NULL) {
842 	acpi_cst_c1_halt();
843 	return;
844     }
845 
846     /* Still probing; use C1 */
847     if (sc->cst_flags & ACPI_CST_FLAG_PROBING) {
848 	acpi_cst_c1_halt();
849 	return;
850     }
851 
852     /* Find the lowest state that has small enough latency. */
853     cx_next_idx = 0;
854     for (i = sc->cst_cx_lowest; i >= 0; i--) {
855 	if (sc->cst_cx_states[i].trans_lat * 3 <= sc->cst_prev_sleep) {
856 	    cx_next_idx = i;
857 	    break;
858 	}
859     }
860 
861     /*
862      * Check for bus master activity if needed for the selected state.
863      * If there was activity, clear the bit and use the lowest non-C3 state.
864      */
865     cx_next = &sc->cst_cx_states[cx_next_idx];
866     if (cx_next->flags & ACPI_CST_CX_FLAG_BM_STS) {
867 	int bm_active;
868 
869 	AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, &bm_active);
870 	if (bm_active != 0) {
871 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, 1);
872 	    cx_next_idx = sc->cst_non_c3;
873 	}
874     }
875 
876     /* Select the next state and update statistics. */
877     cx_next = &sc->cst_cx_states[cx_next_idx];
878     sc->cst_cx_stats[cx_next_idx]++;
879     KASSERT(cx_next->type != ACPI_STATE_C0, ("C0 sleep"));
880 
881     /*
882      * Execute HLT (or equivalent) and wait for an interrupt.  We can't
883      * calculate the time spent in C1 since the place we wake up is an
884      * ISR.  Assume we slept half of quantum and return.
885      */
886     if (cx_next->type == ACPI_STATE_C1) {
887 	sc->cst_prev_sleep = (sc->cst_prev_sleep * 3 + 500000 / hz) / 4;
888 	cx_next->enter(cx_next);
889 	return;
890     }
891 
892     /* Execute the proper preamble before enter the selected state. */
893     if (cx_next->preamble == ACPI_CST_CX_PREAMBLE_BM_ARB) {
894 	AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 1);
895 	bm_arb_disabled = 1;
896     } else if (cx_next->preamble == ACPI_CST_CX_PREAMBLE_WBINVD) {
897 	ACPI_FLUSH_CPU_CACHE();
898     }
899 
900     /*
901      * Enter the selected state and check time spent asleep.
902      */
903     microtime_pcpu_get(&start);
904     cpu_mfence();
905 
906     cx_next->enter(cx_next);
907 
908     cpu_mfence();
909     microtime_pcpu_get(&end);
910 
911     /* Enable bus master arbitration, if it was disabled. */
912     if (bm_arb_disabled)
913 	AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 0);
914 
915     ACPI_ENABLE_IRQS();
916 
917     /* Find the actual time asleep in microseconds. */
918     tdiff = microtime_pcpu_diff(&start, &end);
919     sc->cst_prev_sleep = (sc->cst_prev_sleep * 3 + tdiff) / 4;
920 }
921 
922 /*
923  * Re-evaluate the _CST object when we are notified that it changed.
924  */
925 static void
926 acpi_cst_notify(device_t dev)
927 {
928     struct acpi_cst_softc *sc = device_get_softc(dev);
929 
930     KASSERT(curthread->td_type != TD_TYPE_NETISR,
931         ("notify in netisr%d", mycpuid));
932 
933     lwkt_serialize_enter(&acpi_cst_slize);
934 
935     /* Update the list of Cx states. */
936     acpi_cst_cx_reprobe_cst(sc);
937     acpi_cst_support_list(sc);
938 
939     /* Update the new lowest useable Cx state for all CPUs. */
940     acpi_cst_global_cx_count();
941 
942     /*
943      * Fix up the lowest Cx being used
944      */
945     if (acpi_cst_cx_lowest_req < acpi_cst_cx_count)
946 	acpi_cst_cx_lowest = acpi_cst_cx_lowest_req;
947     if (acpi_cst_cx_lowest > acpi_cst_cx_count - 1)
948 	acpi_cst_cx_lowest = acpi_cst_cx_count - 1;
949 
950     lwkt_serialize_exit(&acpi_cst_slize);
951 }
952 
953 static int
954 acpi_cst_set_quirks(void)
955 {
956     device_t acpi_dev;
957     uint32_t val;
958 
959     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
960 
961     /*
962      * Bus mastering arbitration control is needed to keep caches coherent
963      * while sleeping in C3.  If it's not present but a working flush cache
964      * instruction is present, flush the caches before entering C3 instead.
965      * Otherwise, just disable C3 completely.
966      */
967     if (AcpiGbl_FADT.Pm2ControlBlock == 0 ||
968 	AcpiGbl_FADT.Pm2ControlLength == 0) {
969 	if ((AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD) &&
970 	    (AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD_FLUSH) == 0) {
971 	    acpi_cst_quirks |= ACPI_CST_QUIRK_NO_BM;
972 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
973 		"cpu_cst: no BM control, using flush cache method\n"));
974 	} else {
975 	    acpi_cst_quirks |= ACPI_CST_QUIRK_NO_C3;
976 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
977 		"cpu_cst: no BM control, C3 not available\n"));
978 	}
979     }
980 
981     /* Look for various quirks of the PIIX4 part. */
982     acpi_dev = pci_find_device(PCI_VENDOR_INTEL, PCI_DEVICE_82371AB_3);
983     if (acpi_dev != NULL) {
984 	switch (pci_get_revid(acpi_dev)) {
985 	/*
986 	 * Disable C3 support for all PIIX4 chipsets.  Some of these parts
987 	 * do not report the BMIDE status to the BM status register and
988 	 * others have a livelock bug if Type-F DMA is enabled.  Linux
989 	 * works around the BMIDE bug by reading the BM status directly
990 	 * but we take the simpler approach of disabling C3 for these
991 	 * parts.
992 	 *
993 	 * See erratum #18 ("C3 Power State/BMIDE and Type-F DMA
994 	 * Livelock") from the January 2002 PIIX4 specification update.
995 	 * Applies to all PIIX4 models.
996 	 *
997 	 * Also, make sure that all interrupts cause a "Stop Break"
998 	 * event to exit from C2 state.
999 	 * Also, BRLD_EN_BM (ACPI_BITREG_BUS_MASTER_RLD in ACPI-speak)
1000 	 * should be set to zero, otherwise it causes C2 to short-sleep.
1001 	 * PIIX4 doesn't properly support C3 and bus master activity
1002 	 * need not break out of C2.
1003 	 */
1004 	case PCI_REVISION_A_STEP:
1005 	case PCI_REVISION_B_STEP:
1006 	case PCI_REVISION_4E:
1007 	case PCI_REVISION_4M:
1008 	    acpi_cst_quirks |= ACPI_CST_QUIRK_NO_C3;
1009 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1010 		"cpu_cst: working around PIIX4 bug, disabling C3\n"));
1011 
1012 	    val = pci_read_config(acpi_dev, PIIX4_DEVACTB_REG, 4);
1013 	    if ((val & PIIX4_STOP_BREAK_MASK) != PIIX4_STOP_BREAK_MASK) {
1014 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1015 		    "cpu_cst: PIIX4: enabling IRQs to generate Stop Break\n"));
1016 	    	val |= PIIX4_STOP_BREAK_MASK;
1017 		pci_write_config(acpi_dev, PIIX4_DEVACTB_REG, val, 4);
1018 	    }
1019 	    AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_RLD, &val);
1020 	    if (val) {
1021 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1022 		    "cpu_cst: PIIX4: reset BRLD_EN_BM\n"));
1023 		AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1024 	    }
1025 	    break;
1026 	default:
1027 	    break;
1028 	}
1029     }
1030 
1031     return (0);
1032 }
1033 
1034 static int
1035 acpi_cst_usage_sysctl(SYSCTL_HANDLER_ARGS)
1036 {
1037     struct acpi_cst_softc *sc;
1038     struct sbuf	 sb;
1039     char	 buf[128];
1040     int		 i;
1041     uintmax_t	 fract, sum, whole;
1042 
1043     sc = (struct acpi_cst_softc *) arg1;
1044     sum = 0;
1045     for (i = 0; i < sc->cst_cx_count; i++)
1046 	sum += sc->cst_cx_stats[i];
1047     sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1048     for (i = 0; i < sc->cst_cx_count; i++) {
1049 	if (sum > 0) {
1050 	    whole = (uintmax_t)sc->cst_cx_stats[i] * 100;
1051 	    fract = (whole % sum) * 100;
1052 	    sbuf_printf(&sb, "%u.%02u%% ", (u_int)(whole / sum),
1053 		(u_int)(fract / sum));
1054 	} else
1055 	    sbuf_printf(&sb, "0.00%% ");
1056     }
1057     sbuf_printf(&sb, "last %dus", sc->cst_prev_sleep);
1058     sbuf_trim(&sb);
1059     sbuf_finish(&sb);
1060     sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1061     sbuf_delete(&sb);
1062 
1063     return (0);
1064 }
1065 
1066 static int
1067 acpi_cst_set_lowest_oncpu(struct acpi_cst_softc *sc, int val)
1068 {
1069     int old_lowest, error = 0, old_lowest_req;
1070     uint32_t old_type, type;
1071 
1072     KKASSERT(mycpuid == sc->cst_cpuid);
1073 
1074     old_lowest_req = sc->cst_cx_lowest_req;
1075     sc->cst_cx_lowest_req = val;
1076 
1077     if (val > sc->cst_cx_count - 1)
1078 	val = sc->cst_cx_count - 1;
1079     old_lowest = atomic_swap_int(&sc->cst_cx_lowest, val);
1080 
1081     old_type = sc->cst_cx_states[old_lowest].type;
1082     type = sc->cst_cx_states[val].type;
1083     if (old_type >= ACPI_STATE_C3 && type < ACPI_STATE_C3) {
1084 	cputimer_intr_powersave_remreq();
1085     } else if (type >= ACPI_STATE_C3 && old_type < ACPI_STATE_C3) {
1086 	error = cputimer_intr_powersave_addreq();
1087 	if (error) {
1088 	    /* Restore */
1089 	    sc->cst_cx_lowest_req = old_lowest_req;
1090 	    sc->cst_cx_lowest = old_lowest;
1091 	}
1092     }
1093 
1094     if (error)
1095 	return error;
1096 
1097     /* Cache the new lowest non-C3 state. */
1098     acpi_cst_non_c3(sc);
1099 
1100     /* Reset the statistics counters. */
1101     bzero(sc->cst_cx_stats, sizeof(sc->cst_cx_stats));
1102     return (0);
1103 }
1104 
1105 static void
1106 acpi_cst_set_lowest_handler(netmsg_t msg)
1107 {
1108     struct netmsg_acpi_cst *rmsg = (struct netmsg_acpi_cst *)msg;
1109     int error;
1110 
1111     error = acpi_cst_set_lowest_oncpu(rmsg->sc, rmsg->val);
1112     lwkt_replymsg(&rmsg->base.lmsg, error);
1113 }
1114 
1115 static int
1116 acpi_cst_set_lowest(struct acpi_cst_softc *sc, int val)
1117 {
1118     struct netmsg_acpi_cst msg;
1119 
1120     netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
1121 	acpi_cst_set_lowest_handler);
1122     msg.sc = sc;
1123     msg.val = val;
1124 
1125     return lwkt_domsg(netisr_cpuport(sc->cst_cpuid), &msg.base.lmsg, 0);
1126 }
1127 
1128 static int
1129 acpi_cst_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1130 {
1131     struct	 acpi_cst_softc *sc;
1132     char	 state[8];
1133     int		 val, error;
1134 
1135     sc = (struct acpi_cst_softc *)arg1;
1136     ksnprintf(state, sizeof(state), "C%d", sc->cst_cx_lowest_req + 1);
1137     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1138     if (error != 0 || req->newptr == NULL)
1139 	return (error);
1140     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1141 	return (EINVAL);
1142     val = (int) strtol(state + 1, NULL, 10) - 1;
1143     if (val < 0)
1144 	return (EINVAL);
1145 
1146     lwkt_serialize_enter(&acpi_cst_slize);
1147     error = acpi_cst_set_lowest(sc, val);
1148     lwkt_serialize_exit(&acpi_cst_slize);
1149 
1150     return error;
1151 }
1152 
1153 static int
1154 acpi_cst_lowest_use_sysctl(SYSCTL_HANDLER_ARGS)
1155 {
1156     struct	 acpi_cst_softc *sc;
1157     char	 state[8];
1158 
1159     sc = (struct acpi_cst_softc *)arg1;
1160     ksnprintf(state, sizeof(state), "C%d", sc->cst_cx_lowest + 1);
1161     return sysctl_handle_string(oidp, state, sizeof(state), req);
1162 }
1163 
1164 static int
1165 acpi_cst_global_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1166 {
1167     struct	acpi_cst_softc *sc;
1168     char	state[8];
1169     int		val, error, i;
1170 
1171     ksnprintf(state, sizeof(state), "C%d", acpi_cst_cx_lowest_req + 1);
1172     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1173     if (error != 0 || req->newptr == NULL)
1174 	return (error);
1175     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1176 	return (EINVAL);
1177     val = (int) strtol(state + 1, NULL, 10) - 1;
1178     if (val < 0)
1179 	return (EINVAL);
1180 
1181     lwkt_serialize_enter(&acpi_cst_slize);
1182 
1183     acpi_cst_cx_lowest_req = val;
1184     acpi_cst_cx_lowest = val;
1185     if (acpi_cst_cx_lowest > acpi_cst_cx_count - 1)
1186 	acpi_cst_cx_lowest = acpi_cst_cx_count - 1;
1187 
1188     /* Update the new lowest useable Cx state for all CPUs. */
1189     for (i = 0; i < acpi_cst_ndevices; i++) {
1190 	sc = device_get_softc(acpi_cst_devices[i]);
1191 	error = acpi_cst_set_lowest(sc, val);
1192 	if (error) {
1193 	    KKASSERT(i == 0);
1194 	    break;
1195 	}
1196     }
1197 
1198     lwkt_serialize_exit(&acpi_cst_slize);
1199 
1200     return error;
1201 }
1202 
1203 static int
1204 acpi_cst_global_lowest_use_sysctl(SYSCTL_HANDLER_ARGS)
1205 {
1206     char	state[8];
1207 
1208     ksnprintf(state, sizeof(state), "C%d", acpi_cst_cx_lowest + 1);
1209     return sysctl_handle_string(oidp, state, sizeof(state), req);
1210 }
1211 
1212 /*
1213  * Put the CPU in C1 in a machine-dependant way.
1214  * XXX: shouldn't be here!
1215  */
1216 static void
1217 acpi_cst_c1_halt(void)
1218 {
1219     splz();
1220     if ((mycpu->gd_reqflags & RQF_IDLECHECK_WK_MASK) == 0)
1221         __asm __volatile("sti; hlt");
1222     else
1223         __asm __volatile("sti; pause");
1224 }
1225 
1226 static void
1227 acpi_cst_non_c3(struct acpi_cst_softc *sc)
1228 {
1229     int i;
1230 
1231     sc->cst_non_c3 = 0;
1232     for (i = sc->cst_cx_lowest; i >= 0; i--) {
1233 	if (sc->cst_cx_states[i].type < ACPI_STATE_C3) {
1234 	    sc->cst_non_c3 = i;
1235 	    break;
1236 	}
1237     }
1238     if (bootverbose)
1239 	device_printf(sc->cst_dev, "non-C3 %d\n", sc->cst_non_c3);
1240 }
1241 
1242 /*
1243  * Update the largest Cx state supported in the global acpi_cst_cx_count.
1244  * It will be used in the global Cx sysctl handler.
1245  */
1246 static void
1247 acpi_cst_global_cx_count(void)
1248 {
1249     struct acpi_cst_softc *sc;
1250     int i;
1251 
1252     if (acpi_cst_ndevices == 0) {
1253 	acpi_cst_cx_count = 0;
1254 	return;
1255     }
1256 
1257     sc = device_get_softc(acpi_cst_devices[0]);
1258     acpi_cst_cx_count = sc->cst_cx_count;
1259 
1260     for (i = 1; i < acpi_cst_ndevices; i++) {
1261 	struct acpi_cst_softc *sc = device_get_softc(acpi_cst_devices[i]);
1262 
1263 	if (sc->cst_cx_count < acpi_cst_cx_count)
1264 	    acpi_cst_cx_count = sc->cst_cx_count;
1265     }
1266     if (bootverbose)
1267 	kprintf("cpu_cst: global Cx count %d\n", acpi_cst_cx_count);
1268 }
1269 
1270 static void
1271 acpi_cst_c1_halt_enter(const struct acpi_cst_cx *cx __unused)
1272 {
1273     acpi_cst_c1_halt();
1274 }
1275 
1276 static void
1277 acpi_cst_cx_io_enter(const struct acpi_cst_cx *cx)
1278 {
1279     uint64_t dummy;
1280 
1281     /*
1282      * Read I/O to enter this Cx state
1283      */
1284     bus_space_read_1(cx->btag, cx->bhand, 0);
1285     /*
1286      * Perform a dummy I/O read.  Since it may take an arbitrary time
1287      * to enter the idle state, this read makes sure that we are frozen.
1288      */
1289     AcpiRead(&dummy, &AcpiGbl_FADT.XPmTimerBlock);
1290 }
1291 
1292 static int
1293 acpi_cst_cx_setup(struct acpi_cst_cx *cx)
1294 {
1295     cx->flags &= ~ACPI_CST_CX_FLAG_BM_STS;
1296     cx->preamble = ACPI_CST_CX_PREAMBLE_NONE;
1297 
1298     if (cx->type >= ACPI_STATE_C3) {
1299 	/*
1300 	 * Set the required operations for entering C3(+) state.
1301 	 * Later acpi_cst_md_cx_setup() may fix them up.
1302 	 */
1303 
1304 	/*
1305 	 * Always check BM_STS.
1306 	 */
1307 	if ((acpi_cst_quirks & ACPI_CST_QUIRK_NO_BM) == 0)
1308 	    cx->flags |= ACPI_CST_CX_FLAG_BM_STS;
1309 
1310 	/*
1311 	 * According to the ACPI specification, bus master arbitration
1312 	 * is only available on UP system.  For MP system, cache flushing
1313 	 * is required.
1314 	 */
1315 	if (ncpus == 1 && (acpi_cst_quirks & ACPI_CST_QUIRK_NO_BM) == 0)
1316 	    cx->preamble = ACPI_CST_CX_PREAMBLE_BM_ARB;
1317 	else
1318 	    cx->preamble = ACPI_CST_CX_PREAMBLE_WBINVD;
1319     }
1320     return acpi_cst_md_cx_setup(cx);
1321 }
1322 
1323 static void
1324 acpi_cst_free_resource(struct acpi_cst_softc *sc, int start)
1325 {
1326     int i;
1327 
1328     for (i = start; i < MAX_CX_STATES; ++i) {
1329 	struct acpi_cst_cx *cx = &sc->cst_cx_states[i];
1330 
1331 	if (cx->res != NULL)
1332 	    bus_release_resource(sc->cst_dev, cx->res_type, cx->rid, cx->res);
1333 	memset(cx, 0, sizeof(*cx));
1334     }
1335 }
1336