xref: /dragonfly/sys/dev/acpica/acpi_cpu_cstate.c (revision 0ca59c34)
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 int		acpi_cst_force_bmarb;
176 TUNABLE_INT("hw.acpi.cpu.cst.force_bmarb", &acpi_cst_force_bmarb);
177 
178 int		acpi_cst_force_bmsts;
179 TUNABLE_INT("hw.acpi.cpu.cst.force_bmsts", &acpi_cst_force_bmsts);
180 
181 static device_method_t acpi_cst_methods[] = {
182     /* Device interface */
183     DEVMETHOD(device_probe,	acpi_cst_probe),
184     DEVMETHOD(device_attach,	acpi_cst_attach),
185     DEVMETHOD(device_detach,	bus_generic_detach),
186     DEVMETHOD(device_shutdown,	acpi_cst_shutdown),
187     DEVMETHOD(device_suspend,	acpi_cst_suspend),
188     DEVMETHOD(device_resume,	acpi_cst_resume),
189 
190     /* Bus interface */
191     DEVMETHOD(bus_add_child,	bus_generic_add_child),
192     DEVMETHOD(bus_read_ivar,	bus_generic_read_ivar),
193     DEVMETHOD(bus_get_resource_list, bus_generic_get_resource_list),
194     DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
195     DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
196     DEVMETHOD(bus_alloc_resource, bus_generic_rl_alloc_resource),
197     DEVMETHOD(bus_release_resource, bus_generic_rl_release_resource),
198     DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
199     DEVMETHOD(bus_activate_resource, bus_generic_activate_resource),
200     DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource),
201     DEVMETHOD(bus_setup_intr,	bus_generic_setup_intr),
202     DEVMETHOD(bus_teardown_intr, bus_generic_teardown_intr),
203     DEVMETHOD_END
204 };
205 
206 static driver_t acpi_cst_driver = {
207     "cpu_cst",
208     acpi_cst_methods,
209     sizeof(struct acpi_cst_softc),
210 };
211 
212 static devclass_t acpi_cst_devclass;
213 DRIVER_MODULE(cpu_cst, cpu, acpi_cst_driver, acpi_cst_devclass, NULL, NULL);
214 MODULE_DEPEND(cpu_cst, acpi, 1, 1, 1);
215 
216 static int
217 acpi_cst_probe(device_t dev)
218 {
219     int cpu_id;
220 
221     if (acpi_disabled("cpu_cst") || acpi_get_type(dev) != ACPI_TYPE_PROCESSOR)
222 	return (ENXIO);
223 
224     cpu_id = acpi_get_magic(dev);
225 
226     if (acpi_cst_softc == NULL)
227 	acpi_cst_softc = kmalloc(sizeof(struct acpi_cst_softc *) *
228 	    SMP_MAXCPU, M_TEMP /* XXX */, M_INTWAIT | M_ZERO);
229 
230     /*
231      * Check if we already probed this processor.  We scan the bus twice
232      * so it's possible we've already seen this one.
233      */
234     if (acpi_cst_softc[cpu_id] != NULL) {
235 	device_printf(dev, "CPU%d cstate already exist\n", cpu_id);
236 	return (ENXIO);
237     }
238 
239     /* Mark this processor as in-use and save our derived id for attach. */
240     acpi_cst_softc[cpu_id] = device_get_softc(dev);
241     device_set_desc(dev, "ACPI CPU C-State");
242 
243     return (0);
244 }
245 
246 static int
247 acpi_cst_attach(device_t dev)
248 {
249     ACPI_BUFFER		   buf;
250     ACPI_OBJECT		   *obj;
251     struct acpi_cst_softc *sc;
252     ACPI_STATUS		   status;
253 
254     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
255 
256     sc = device_get_softc(dev);
257     sc->cst_dev = dev;
258     sc->cst_parent = device_get_softc(device_get_parent(dev));
259     sc->cst_handle = acpi_get_handle(dev);
260     sc->cst_cpuid = acpi_get_magic(dev);
261     acpi_cst_softc[sc->cst_cpuid] = sc;
262     acpi_cst_smi_cmd = AcpiGbl_FADT.SmiCommand;
263     acpi_cst_ctrl = AcpiGbl_FADT.CstControl;
264 
265     buf.Pointer = NULL;
266     buf.Length = ACPI_ALLOCATE_BUFFER;
267     status = AcpiEvaluateObject(sc->cst_handle, NULL, NULL, &buf);
268     if (ACPI_FAILURE(status)) {
269 	device_printf(dev, "attach failed to get Processor obj - %s\n",
270 		      AcpiFormatException(status));
271 	acpi_cst_softc[sc->cst_cpuid] = NULL;
272 	return (ENXIO);
273     }
274     obj = (ACPI_OBJECT *)buf.Pointer;
275     sc->cst_p_blk = obj->Processor.PblkAddress;
276     sc->cst_p_blk_len = obj->Processor.PblkLength;
277     AcpiOsFree(obj);
278     ACPI_DEBUG_PRINT((ACPI_DB_INFO, "cpu_cst%d: P_BLK at %#x/%d\n",
279 		     device_get_unit(dev), sc->cst_p_blk, sc->cst_p_blk_len));
280 
281     /*
282      * If this is the first cpu we attach, create and initialize the generic
283      * resources that will be used by all acpi cpu devices.
284      */
285     if (device_get_unit(dev) == 0) {
286 	/* Assume we won't be using FADT for Cx states by default */
287 	acpi_cst_use_fadt = FALSE;
288 
289 	/* Queue post cpu-probing task handler */
290 	AcpiOsExecute(OSL_NOTIFY_HANDLER, acpi_cst_postattach, NULL);
291     }
292 
293     /* Probe for Cx state support. */
294     acpi_cst_cx_probe(sc);
295 
296     sc->cst_flags |= ACPI_CST_FLAG_ATTACHED;
297 
298     return (0);
299 }
300 
301 /*
302  * Disable any entry to the idle function during suspend and re-enable it
303  * during resume.
304  */
305 static int
306 acpi_cst_suspend(device_t dev)
307 {
308     int error;
309 
310     error = bus_generic_suspend(dev);
311     if (error)
312 	return (error);
313     acpi_cst_disable_idle = TRUE;
314     return (0);
315 }
316 
317 static int
318 acpi_cst_resume(device_t dev)
319 {
320     acpi_cst_disable_idle = FALSE;
321     return (bus_generic_resume(dev));
322 }
323 
324 static int
325 acpi_cst_shutdown(device_t dev)
326 {
327     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
328 
329     /* Allow children to shutdown first. */
330     bus_generic_shutdown(dev);
331 
332     /*
333      * Disable any entry to the idle function.  There is a small race where
334      * an idle thread have passed this check but not gone to sleep.  This
335      * is ok since device_shutdown() does not free the softc, otherwise
336      * we'd have to be sure all threads were evicted before returning.
337      */
338     acpi_cst_disable_idle = TRUE;
339 
340     return_VALUE (0);
341 }
342 
343 static void
344 acpi_cst_cx_probe(struct acpi_cst_softc *sc)
345 {
346     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
347 
348     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
349     sc->cst_prev_sleep = 1000000;
350     sc->cst_cx_lowest = 0;
351     sc->cst_cx_lowest_req = 0;
352 
353     /*
354      * Check for the ACPI 2.0 _CST sleep states object.  If we can't find
355      * any, we'll revert to FADT/P_BLK Cx control method which will be
356      * handled by acpi_cst_postattach.  We need to defer to after having
357      * probed all the cpus in the system before probing for Cx states from
358      * FADT as we may already have found cpus with valid _CST packages.
359      */
360     if (!acpi_cst_use_fadt && acpi_cst_cx_probe_cst(sc, 0) != 0) {
361 	/*
362 	 * We were unable to find a _CST package for this cpu or there
363 	 * was an error parsing it. Switch back to generic mode.
364 	 */
365 	acpi_cst_use_fadt = TRUE;
366 	if (bootverbose)
367 	    device_printf(sc->cst_dev, "switching to FADT Cx mode\n");
368     }
369 
370     /*
371      * TODO: _CSD Package should be checked here.
372      */
373 }
374 
375 static void
376 acpi_cst_cx_probe_fadt(struct acpi_cst_softc *sc)
377 {
378     struct acpi_cst_cx *cx_ptr;
379     int error;
380 
381     /*
382      * Free all previously allocated resources.
383      *
384      * NITE:
385      * It is needed, since we could enter here because of other
386      * cpu's _CST probing failure.
387      */
388     acpi_cst_free_resource(sc, 0);
389 
390     sc->cst_cx_count = 0;
391     cx_ptr = sc->cst_cx_states;
392 
393     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
394     sc->cst_prev_sleep = 1000000;
395 
396     /* C1 has been required since just after ACPI 1.0 */
397     cx_ptr->gas.SpaceId = ACPI_ADR_SPACE_FIXED_HARDWARE;
398     cx_ptr->type = ACPI_STATE_C1;
399     cx_ptr->trans_lat = 0;
400     cx_ptr->enter = acpi_cst_c1_halt_enter;
401     error = acpi_cst_cx_setup(cx_ptr);
402     if (error)
403 	panic("C1 FADT HALT setup failed: %d", error);
404     cx_ptr++;
405     sc->cst_cx_count++;
406 
407     /* C2(+) is not supported on MP system */
408     if (ncpus > 1 && (AcpiGbl_FADT.Flags & ACPI_FADT_C2_MP_SUPPORTED) == 0)
409 	return;
410 
411     /*
412      * The spec says P_BLK must be 6 bytes long.  However, some systems
413      * use it to indicate a fractional set of features present so we
414      * take 5 as C2.  Some may also have a value of 7 to indicate
415      * another C3 but most use _CST for this (as required) and having
416      * "only" C1-C3 is not a hardship.
417      */
418     if (sc->cst_p_blk_len < 5)
419 	return;
420 
421     /* Validate and allocate resources for C2 (P_LVL2). */
422     if (AcpiGbl_FADT.C2Latency <= 100) {
423 	cx_ptr->gas.SpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
424 	cx_ptr->gas.BitWidth = 8;
425 	cx_ptr->gas.Address = sc->cst_p_blk + 4;
426 
427 	cx_ptr->rid = sc->cst_parent->cpu_next_rid;
428 	acpi_bus_alloc_gas(sc->cst_dev, &cx_ptr->res_type, &cx_ptr->rid,
429 	    &cx_ptr->gas, &cx_ptr->res, RF_SHAREABLE);
430 	if (cx_ptr->res != NULL) {
431 	    sc->cst_parent->cpu_next_rid++;
432 	    cx_ptr->type = ACPI_STATE_C2;
433 	    cx_ptr->trans_lat = AcpiGbl_FADT.C2Latency;
434 	    cx_ptr->enter = acpi_cst_cx_io_enter;
435 	    cx_ptr->btag = rman_get_bustag(cx_ptr->res);
436 	    cx_ptr->bhand = rman_get_bushandle(cx_ptr->res);
437 	    error = acpi_cst_cx_setup(cx_ptr);
438 	    if (error)
439 		panic("C2 FADT I/O setup failed: %d", error);
440 	    cx_ptr++;
441 	    sc->cst_cx_count++;
442 	    sc->cst_non_c3 = 1;
443 	}
444     }
445     if (sc->cst_p_blk_len < 6)
446 	return;
447 
448     /* Validate and allocate resources for C3 (P_LVL3). */
449     if (AcpiGbl_FADT.C3Latency <= 1000 &&
450         !(acpi_cst_quirks & ACPI_CST_QUIRK_NO_C3)) {
451 	cx_ptr->gas.SpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
452 	cx_ptr->gas.BitWidth = 8;
453 	cx_ptr->gas.Address = sc->cst_p_blk + 5;
454 
455 	cx_ptr->rid = sc->cst_parent->cpu_next_rid;
456 	acpi_bus_alloc_gas(sc->cst_dev, &cx_ptr->res_type, &cx_ptr->rid,
457 	    &cx_ptr->gas, &cx_ptr->res, RF_SHAREABLE);
458 	if (cx_ptr->res != NULL) {
459 	    sc->cst_parent->cpu_next_rid++;
460 	    cx_ptr->type = ACPI_STATE_C3;
461 	    cx_ptr->trans_lat = AcpiGbl_FADT.C3Latency;
462 	    cx_ptr->enter = acpi_cst_cx_io_enter;
463 	    cx_ptr->btag = rman_get_bustag(cx_ptr->res);
464 	    cx_ptr->bhand = rman_get_bushandle(cx_ptr->res);
465 	    error = acpi_cst_cx_setup(cx_ptr);
466 	    if (error)
467 		panic("C3 FADT I/O setup failed: %d", error);
468 	    cx_ptr++;
469 	    sc->cst_cx_count++;
470 	}
471     }
472 }
473 
474 static void
475 acpi_cst_copy(struct acpi_cst_softc *dst_sc,
476     const struct acpi_cst_softc *src_sc)
477 {
478     dst_sc->cst_non_c3 = src_sc->cst_non_c3;
479     dst_sc->cst_cx_count = src_sc->cst_cx_count;
480     memcpy(dst_sc->cst_cx_states, src_sc->cst_cx_states,
481         sizeof(dst_sc->cst_cx_states));
482 }
483 
484 /*
485  * Parse a _CST package and set up its Cx states.  Since the _CST object
486  * can change dynamically, our notify handler may call this function
487  * to clean up and probe the new _CST package.
488  */
489 static int
490 acpi_cst_cx_probe_cst(struct acpi_cst_softc *sc, int reprobe)
491 {
492     struct	 acpi_cst_cx *cx_ptr;
493     ACPI_STATUS	 status;
494     ACPI_BUFFER	 buf;
495     ACPI_OBJECT	*top;
496     ACPI_OBJECT	*pkg;
497     uint32_t	 count;
498     int		 i;
499 
500     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
501 
502 #ifdef INVARIANTS
503     if (reprobe)
504 	KKASSERT(&curthread->td_msgport == netisr_cpuport(sc->cst_cpuid));
505 #endif
506 
507     buf.Pointer = NULL;
508     buf.Length = ACPI_ALLOCATE_BUFFER;
509     status = AcpiEvaluateObject(sc->cst_handle, "_CST", NULL, &buf);
510     if (ACPI_FAILURE(status))
511 	return (ENXIO);
512 
513     /* _CST is a package with a count and at least one Cx package. */
514     top = (ACPI_OBJECT *)buf.Pointer;
515     if (!ACPI_PKG_VALID(top, 2) || acpi_PkgInt32(top, 0, &count) != 0) {
516 	device_printf(sc->cst_dev, "invalid _CST package\n");
517 	AcpiOsFree(buf.Pointer);
518 	return (ENXIO);
519     }
520     if (count != top->Package.Count - 1) {
521 	device_printf(sc->cst_dev, "invalid _CST state count (%d != %d)\n",
522 	       count, top->Package.Count - 1);
523 	count = top->Package.Count - 1;
524     }
525     if (count > MAX_CX_STATES) {
526 	device_printf(sc->cst_dev, "_CST has too many states (%d)\n", count);
527 	count = MAX_CX_STATES;
528     }
529 
530     sc->cst_flags |= ACPI_CST_FLAG_PROBING | ACPI_CST_FLAG_MATCH_HT;
531     cpu_sfence();
532 
533     /*
534      * Free all previously allocated resources
535      *
536      * NOTE: It is needed for _CST reprobing.
537      */
538     acpi_cst_free_resource(sc, 0);
539 
540     /* Set up all valid states. */
541     sc->cst_cx_count = 0;
542     cx_ptr = sc->cst_cx_states;
543     for (i = 0; i < count; i++) {
544 	int error;
545 
546 	pkg = &top->Package.Elements[i + 1];
547 	if (!ACPI_PKG_VALID(pkg, 4) ||
548 	    acpi_PkgInt32(pkg, 1, &cx_ptr->type) != 0 ||
549 	    acpi_PkgInt32(pkg, 2, &cx_ptr->trans_lat) != 0 ||
550 	    acpi_PkgInt32(pkg, 3, &cx_ptr->power) != 0) {
551 
552 	    device_printf(sc->cst_dev, "skipping invalid Cx state package\n");
553 	    continue;
554 	}
555 
556 	/* Validate the state to see if we should use it. */
557 	switch (cx_ptr->type) {
558 	case ACPI_STATE_C1:
559 	    sc->cst_non_c3 = i;
560 	    cx_ptr->enter = acpi_cst_c1_halt_enter;
561 	    error = acpi_cst_cx_setup(cx_ptr);
562 	    if (error)
563 		panic("C1 CST HALT setup failed: %d", error);
564 	    if (sc->cst_cx_count != 0) {
565 		/*
566 		 * C1 is not the first C-state; something really stupid
567 		 * is going on ...
568 		 */
569 		sc->cst_flags &= ~ACPI_CST_FLAG_MATCH_HT;
570 	    }
571 	    cx_ptr++;
572 	    sc->cst_cx_count++;
573 	    continue;
574 	case ACPI_STATE_C2:
575 	    sc->cst_non_c3 = i;
576 	    break;
577 	case ACPI_STATE_C3:
578 	default:
579 	    if ((acpi_cst_quirks & ACPI_CST_QUIRK_NO_C3) != 0) {
580 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
581 				 "cpu_cst%d: C3[%d] not available.\n",
582 				 device_get_unit(sc->cst_dev), i));
583 		continue;
584 	    }
585 	    break;
586 	}
587 
588 	/*
589 	 * Allocate the control register for C2 or C3(+).
590 	 */
591 	KASSERT(cx_ptr->res == NULL, ("still has res"));
592 	acpi_PkgRawGas(pkg, 0, &cx_ptr->gas);
593 
594 	/*
595 	 * We match number of C2/C3 for hyperthreads, only if the
596 	 * register is "Fixed Hardware", e.g. on most of the Intel
597 	 * CPUs.  We don't have much to do for the rest of the
598 	 * register types.
599 	 */
600 	if (cx_ptr->gas.SpaceId != ACPI_ADR_SPACE_FIXED_HARDWARE)
601 	    sc->cst_flags &= ~ACPI_CST_FLAG_MATCH_HT;
602 
603 	cx_ptr->rid = sc->cst_parent->cpu_next_rid;
604 	acpi_bus_alloc_gas(sc->cst_dev, &cx_ptr->res_type, &cx_ptr->rid,
605 	    &cx_ptr->gas, &cx_ptr->res, RF_SHAREABLE);
606 	if (cx_ptr->res != NULL) {
607 	    sc->cst_parent->cpu_next_rid++;
608 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
609 			     "cpu_cst%d: Got C%d - %d latency\n",
610 			     device_get_unit(sc->cst_dev), cx_ptr->type,
611 			     cx_ptr->trans_lat));
612 	    cx_ptr->enter = acpi_cst_cx_io_enter;
613 	    cx_ptr->btag = rman_get_bustag(cx_ptr->res);
614 	    cx_ptr->bhand = rman_get_bushandle(cx_ptr->res);
615 	    error = acpi_cst_cx_setup(cx_ptr);
616 	    if (error)
617 		panic("C%d CST I/O setup failed: %d", cx_ptr->type, error);
618 	    cx_ptr++;
619 	    sc->cst_cx_count++;
620 	} else {
621 	    error = acpi_cst_cx_setup(cx_ptr);
622 	    if (!error) {
623 		KASSERT(cx_ptr->enter != NULL,
624 		    ("C%d enter is not set", cx_ptr->type));
625 		cx_ptr++;
626 		sc->cst_cx_count++;
627 	    }
628 	}
629     }
630     AcpiOsFree(buf.Pointer);
631 
632     if (sc->cst_flags & ACPI_CST_FLAG_MATCH_HT) {
633 	cpumask_t mask;
634 
635 	mask = get_cpumask_from_level(sc->cst_cpuid, CORE_LEVEL);
636 	if (CPUMASK_TESTNZERO(mask)) {
637 	    int cpu;
638 
639 	    for (cpu = 0; cpu < ncpus; ++cpu) {
640 		struct acpi_cst_softc *sc1 = acpi_cst_softc[cpu];
641 
642 		if (sc1 == NULL || sc1 == sc ||
643 		    (sc1->cst_flags & ACPI_CST_FLAG_ATTACHED) == 0 ||
644 		    (sc1->cst_flags & ACPI_CST_FLAG_MATCH_HT) == 0)
645 		    continue;
646 		if (!CPUMASK_TESTBIT(mask, sc1->cst_cpuid))
647 		    continue;
648 
649 		if (sc1->cst_cx_count != sc->cst_cx_count) {
650 		    struct acpi_cst_softc *src_sc, *dst_sc;
651 
652 		    if (bootverbose) {
653 			device_printf(sc->cst_dev,
654 			    "inconstent C-state count: %d, %s has %d\n",
655 			    sc->cst_cx_count,
656 			    device_get_nameunit(sc1->cst_dev),
657 			    sc1->cst_cx_count);
658 		    }
659 		    if (sc1->cst_cx_count > sc->cst_cx_count) {
660 			src_sc = sc1;
661 			dst_sc = sc;
662 		    } else {
663 			src_sc = sc;
664 			dst_sc = sc1;
665 		    }
666 		    acpi_cst_copy(dst_sc, src_sc);
667 		}
668 	    }
669 	}
670     }
671 
672     if (reprobe) {
673 	/* If there are C3(+) states, always enable bus master wakeup */
674 	if ((acpi_cst_quirks & ACPI_CST_QUIRK_NO_BM) == 0) {
675 	    for (i = 0; i < sc->cst_cx_count; ++i) {
676 		struct acpi_cst_cx *cx = &sc->cst_cx_states[i];
677 
678 		if (cx->type >= ACPI_STATE_C3) {
679 		    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
680 		    break;
681 		}
682 	    }
683 	}
684 
685 	/* Fix up the lowest Cx being used */
686 	acpi_cst_set_lowest_oncpu(sc, sc->cst_cx_lowest_req);
687     }
688 
689     /*
690      * Cache the lowest non-C3 state.
691      * NOTE: must after cst_cx_lowest is set.
692      */
693     acpi_cst_non_c3(sc);
694 
695     cpu_sfence();
696     sc->cst_flags &= ~ACPI_CST_FLAG_PROBING;
697 
698     return (0);
699 }
700 
701 static void
702 acpi_cst_cx_reprobe_cst_handler(netmsg_t msg)
703 {
704     struct netmsg_acpi_cst *rmsg = (struct netmsg_acpi_cst *)msg;
705     int error;
706 
707     error = acpi_cst_cx_probe_cst(rmsg->sc, 1);
708     lwkt_replymsg(&rmsg->base.lmsg, error);
709 }
710 
711 static int
712 acpi_cst_cx_reprobe_cst(struct acpi_cst_softc *sc)
713 {
714     struct netmsg_acpi_cst msg;
715 
716     netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
717 	acpi_cst_cx_reprobe_cst_handler);
718     msg.sc = sc;
719 
720     return lwkt_domsg(netisr_cpuport(sc->cst_cpuid), &msg.base.lmsg, 0);
721 }
722 
723 /*
724  * Call this *after* all CPUs Cx states have been attached.
725  */
726 static void
727 acpi_cst_postattach(void *arg)
728 {
729     struct acpi_cst_softc *sc;
730     int i;
731 
732     /* Get set of Cx state devices */
733     devclass_get_devices(acpi_cst_devclass, &acpi_cst_devices,
734 	&acpi_cst_ndevices);
735 
736     /*
737      * Setup any quirks that might necessary now that we have probed
738      * all the CPUs' Cx states.
739      */
740     acpi_cst_set_quirks();
741 
742     if (acpi_cst_use_fadt) {
743 	/*
744 	 * We are using Cx mode from FADT, probe for available Cx states
745 	 * for all processors.
746 	 */
747 	for (i = 0; i < acpi_cst_ndevices; i++) {
748 	    sc = device_get_softc(acpi_cst_devices[i]);
749 	    acpi_cst_cx_probe_fadt(sc);
750 	}
751     } else {
752 	/*
753 	 * We are using _CST mode, remove C3 state if necessary.
754 	 *
755 	 * As we now know for sure that we will be using _CST mode
756 	 * install our notify handler.
757 	 */
758 	for (i = 0; i < acpi_cst_ndevices; i++) {
759 	    sc = device_get_softc(acpi_cst_devices[i]);
760 	    if (acpi_cst_quirks & ACPI_CST_QUIRK_NO_C3) {
761 		/* Free part of unused resources */
762 		acpi_cst_free_resource(sc, sc->cst_non_c3 + 1);
763 		sc->cst_cx_count = sc->cst_non_c3 + 1;
764 	    }
765 	    sc->cst_parent->cpu_cst_notify = acpi_cst_notify;
766 	}
767     }
768     acpi_cst_global_cx_count();
769 
770     /* Perform Cx final initialization. */
771     for (i = 0; i < acpi_cst_ndevices; i++) {
772 	sc = device_get_softc(acpi_cst_devices[i]);
773 	acpi_cst_startup(sc);
774 
775 	if (sc->cst_parent->glob_sysctl_tree != NULL) {
776 	    struct acpi_cpu_softc *cpu = sc->cst_parent;
777 
778 	    /* Add a sysctl handler to handle global Cx lowest setting */
779 	    SYSCTL_ADD_PROC(&cpu->glob_sysctl_ctx,
780 	    		    SYSCTL_CHILDREN(cpu->glob_sysctl_tree),
781 			    OID_AUTO, "cx_lowest",
782 			    CTLTYPE_STRING | CTLFLAG_RW, NULL, 0,
783 			    acpi_cst_global_lowest_sysctl, "A",
784 			    "Requested global lowest Cx sleep state");
785 	    SYSCTL_ADD_PROC(&cpu->glob_sysctl_ctx,
786 	    		    SYSCTL_CHILDREN(cpu->glob_sysctl_tree),
787 			    OID_AUTO, "cx_lowest_use",
788 			    CTLTYPE_STRING | CTLFLAG_RD, NULL, 0,
789 			    acpi_cst_global_lowest_use_sysctl, "A",
790 			    "Global lowest Cx sleep state to use");
791 	}
792     }
793 
794     /* Take over idling from cpu_idle_default(). */
795     acpi_cst_cx_lowest = 0;
796     acpi_cst_cx_lowest_req = 0;
797     acpi_cst_disable_idle = FALSE;
798 
799     cpu_sfence();
800     cpu_idle_hook = acpi_cst_idle;
801 }
802 
803 static void
804 acpi_cst_support_list(struct acpi_cst_softc *sc)
805 {
806     struct sbuf sb;
807     int i;
808 
809     /*
810      * Set up the list of Cx states
811      */
812     sbuf_new(&sb, sc->cst_cx_supported, sizeof(sc->cst_cx_supported),
813 	SBUF_FIXEDLEN);
814     for (i = 0; i < sc->cst_cx_count; i++)
815 	sbuf_printf(&sb, "C%d/%d ", i + 1, sc->cst_cx_states[i].trans_lat);
816     sbuf_trim(&sb);
817     sbuf_finish(&sb);
818 }
819 
820 static void
821 acpi_cst_c3_bm_rld_handler(netmsg_t msg)
822 {
823     struct netmsg_acpi_cst *rmsg = (struct netmsg_acpi_cst *)msg;
824 
825     AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
826     lwkt_replymsg(&rmsg->base.lmsg, 0);
827 }
828 
829 static void
830 acpi_cst_c3_bm_rld(struct acpi_cst_softc *sc)
831 {
832     struct netmsg_acpi_cst msg;
833 
834     netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
835 	acpi_cst_c3_bm_rld_handler);
836     msg.sc = sc;
837 
838     lwkt_domsg(netisr_cpuport(sc->cst_cpuid), &msg.base.lmsg, 0);
839 }
840 
841 static void
842 acpi_cst_startup(struct acpi_cst_softc *sc)
843 {
844     struct acpi_cpu_softc *cpu = sc->cst_parent;
845     int i, bm_rld_done = 0;
846 
847     for (i = 0; i < sc->cst_cx_count; ++i) {
848 	struct acpi_cst_cx *cx = &sc->cst_cx_states[i];
849 	int error;
850 
851 	/* If there are C3(+) states, always enable bus master wakeup */
852 	if (cx->type >= ACPI_STATE_C3 && !bm_rld_done &&
853 	    (acpi_cst_quirks & ACPI_CST_QUIRK_NO_BM) == 0) {
854 	    acpi_cst_c3_bm_rld(sc);
855 	    bm_rld_done = 1;
856 	}
857 
858 	/* Redo the Cx setup, since quirks have been changed */
859 	error = acpi_cst_cx_setup(cx);
860 	if (error)
861 	    panic("C%d startup setup failed: %d", i + 1, error);
862     }
863 
864     acpi_cst_support_list(sc);
865 
866     SYSCTL_ADD_STRING(&cpu->pcpu_sysctl_ctx,
867 		      SYSCTL_CHILDREN(cpu->pcpu_sysctl_tree),
868 		      OID_AUTO, "cx_supported", CTLFLAG_RD,
869 		      sc->cst_cx_supported, 0,
870 		      "Cx/microsecond values for supported Cx states");
871     SYSCTL_ADD_PROC(&cpu->pcpu_sysctl_ctx,
872 		    SYSCTL_CHILDREN(cpu->pcpu_sysctl_tree),
873 		    OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
874 		    (void *)sc, 0, acpi_cst_lowest_sysctl, "A",
875 		    "requested lowest Cx sleep state");
876     SYSCTL_ADD_PROC(&cpu->pcpu_sysctl_ctx,
877 		    SYSCTL_CHILDREN(cpu->pcpu_sysctl_tree),
878 		    OID_AUTO, "cx_lowest_use", CTLTYPE_STRING | CTLFLAG_RD,
879 		    (void *)sc, 0, acpi_cst_lowest_use_sysctl, "A",
880 		    "lowest Cx sleep state to use");
881     SYSCTL_ADD_PROC(&cpu->pcpu_sysctl_ctx,
882 		    SYSCTL_CHILDREN(cpu->pcpu_sysctl_tree),
883 		    OID_AUTO, "cx_usage", CTLTYPE_STRING | CTLFLAG_RD,
884 		    (void *)sc, 0, acpi_cst_usage_sysctl, "A",
885 		    "percent usage for each Cx state");
886 
887 #ifdef notyet
888     /* Signal platform that we can handle _CST notification. */
889     if (!acpi_cst_use_fadt && acpi_cst_ctrl != 0) {
890 	ACPI_LOCK(acpi);
891 	AcpiOsWritePort(acpi_cst_smi_cmd, acpi_cst_ctrl, 8);
892 	ACPI_UNLOCK(acpi);
893     }
894 #endif
895 }
896 
897 /*
898  * Idle the CPU in the lowest state possible.  This function is called with
899  * interrupts disabled.  Note that once it re-enables interrupts, a task
900  * switch can occur so do not access shared data (i.e. the softc) after
901  * interrupts are re-enabled.
902  */
903 static void
904 acpi_cst_idle(void)
905 {
906     struct	acpi_cst_softc *sc;
907     struct	acpi_cst_cx *cx_next;
908     union microtime_pcpu start, end;
909     int		cx_next_idx, i, tdiff, bm_arb_disabled = 0;
910 
911     /* If disabled, return immediately. */
912     if (acpi_cst_disable_idle) {
913 	ACPI_ENABLE_IRQS();
914 	return;
915     }
916 
917     /*
918      * Look up our CPU id to get our softc.  If it's NULL, we'll use C1
919      * since there is no Cx state for this processor.
920      */
921     sc = acpi_cst_softc[mdcpu->mi.gd_cpuid];
922     if (sc == NULL) {
923 	acpi_cst_c1_halt();
924 	return;
925     }
926 
927     /* Still probing; use C1 */
928     if (sc->cst_flags & ACPI_CST_FLAG_PROBING) {
929 	acpi_cst_c1_halt();
930 	return;
931     }
932 
933     /* Find the lowest state that has small enough latency. */
934     cx_next_idx = 0;
935     for (i = sc->cst_cx_lowest; i >= 0; i--) {
936 	if (sc->cst_cx_states[i].trans_lat * 3 <= sc->cst_prev_sleep) {
937 	    cx_next_idx = i;
938 	    break;
939 	}
940     }
941 
942     /*
943      * Check for bus master activity if needed for the selected state.
944      * If there was activity, clear the bit and use the lowest non-C3 state.
945      */
946     cx_next = &sc->cst_cx_states[cx_next_idx];
947     if (cx_next->flags & ACPI_CST_CX_FLAG_BM_STS) {
948 	int bm_active;
949 
950 	AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, &bm_active);
951 	if (bm_active != 0) {
952 	    AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, 1);
953 	    cx_next_idx = sc->cst_non_c3;
954 	}
955     }
956 
957     /* Select the next state and update statistics. */
958     cx_next = &sc->cst_cx_states[cx_next_idx];
959     sc->cst_cx_stats[cx_next_idx]++;
960     KASSERT(cx_next->type != ACPI_STATE_C0, ("C0 sleep"));
961 
962     /*
963      * Execute HLT (or equivalent) and wait for an interrupt.  We can't
964      * calculate the time spent in C1 since the place we wake up is an
965      * ISR.  Assume we slept half of quantum and return.
966      */
967     if (cx_next->type == ACPI_STATE_C1) {
968 	sc->cst_prev_sleep = (sc->cst_prev_sleep * 3 + 500000 / hz) / 4;
969 	cx_next->enter(cx_next);
970 	return;
971     }
972 
973     /* Execute the proper preamble before enter the selected state. */
974     if (cx_next->preamble == ACPI_CST_CX_PREAMBLE_BM_ARB) {
975 	AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 1);
976 	bm_arb_disabled = 1;
977     } else if (cx_next->preamble == ACPI_CST_CX_PREAMBLE_WBINVD) {
978 	ACPI_FLUSH_CPU_CACHE();
979     }
980 
981     /*
982      * Enter the selected state and check time spent asleep.
983      */
984     microtime_pcpu_get(&start);
985     cpu_mfence();
986 
987     cx_next->enter(cx_next);
988 
989     cpu_mfence();
990     microtime_pcpu_get(&end);
991 
992     /* Enable bus master arbitration, if it was disabled. */
993     if (bm_arb_disabled)
994 	AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 0);
995 
996     ACPI_ENABLE_IRQS();
997 
998     /* Find the actual time asleep in microseconds. */
999     tdiff = microtime_pcpu_diff(&start, &end);
1000     sc->cst_prev_sleep = (sc->cst_prev_sleep * 3 + tdiff) / 4;
1001 }
1002 
1003 /*
1004  * Re-evaluate the _CST object when we are notified that it changed.
1005  */
1006 static void
1007 acpi_cst_notify(device_t dev)
1008 {
1009     struct acpi_cst_softc *sc = device_get_softc(dev);
1010 
1011     KASSERT(curthread->td_type != TD_TYPE_NETISR,
1012         ("notify in netisr%d", mycpuid));
1013 
1014     lwkt_serialize_enter(&acpi_cst_slize);
1015 
1016     /* Update the list of Cx states. */
1017     acpi_cst_cx_reprobe_cst(sc);
1018     acpi_cst_support_list(sc);
1019 
1020     /* Update the new lowest useable Cx state for all CPUs. */
1021     acpi_cst_global_cx_count();
1022 
1023     /*
1024      * Fix up the lowest Cx being used
1025      */
1026     if (acpi_cst_cx_lowest_req < acpi_cst_cx_count)
1027 	acpi_cst_cx_lowest = acpi_cst_cx_lowest_req;
1028     if (acpi_cst_cx_lowest > acpi_cst_cx_count - 1)
1029 	acpi_cst_cx_lowest = acpi_cst_cx_count - 1;
1030 
1031     lwkt_serialize_exit(&acpi_cst_slize);
1032 }
1033 
1034 static int
1035 acpi_cst_set_quirks(void)
1036 {
1037     device_t acpi_dev;
1038     uint32_t val;
1039 
1040     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
1041 
1042     /*
1043      * Bus mastering arbitration control is needed to keep caches coherent
1044      * while sleeping in C3.  If it's not present but a working flush cache
1045      * instruction is present, flush the caches before entering C3 instead.
1046      * Otherwise, just disable C3 completely.
1047      */
1048     if (AcpiGbl_FADT.Pm2ControlBlock == 0 ||
1049 	AcpiGbl_FADT.Pm2ControlLength == 0) {
1050 	if ((AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD) &&
1051 	    (AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD_FLUSH) == 0) {
1052 	    acpi_cst_quirks |= ACPI_CST_QUIRK_NO_BM;
1053 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1054 		"cpu_cst: no BM control, using flush cache method\n"));
1055 	} else {
1056 	    acpi_cst_quirks |= ACPI_CST_QUIRK_NO_C3;
1057 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1058 		"cpu_cst: no BM control, C3 not available\n"));
1059 	}
1060     }
1061 
1062     /* Look for various quirks of the PIIX4 part. */
1063     acpi_dev = pci_find_device(PCI_VENDOR_INTEL, PCI_DEVICE_82371AB_3);
1064     if (acpi_dev != NULL) {
1065 	switch (pci_get_revid(acpi_dev)) {
1066 	/*
1067 	 * Disable C3 support for all PIIX4 chipsets.  Some of these parts
1068 	 * do not report the BMIDE status to the BM status register and
1069 	 * others have a livelock bug if Type-F DMA is enabled.  Linux
1070 	 * works around the BMIDE bug by reading the BM status directly
1071 	 * but we take the simpler approach of disabling C3 for these
1072 	 * parts.
1073 	 *
1074 	 * See erratum #18 ("C3 Power State/BMIDE and Type-F DMA
1075 	 * Livelock") from the January 2002 PIIX4 specification update.
1076 	 * Applies to all PIIX4 models.
1077 	 *
1078 	 * Also, make sure that all interrupts cause a "Stop Break"
1079 	 * event to exit from C2 state.
1080 	 * Also, BRLD_EN_BM (ACPI_BITREG_BUS_MASTER_RLD in ACPI-speak)
1081 	 * should be set to zero, otherwise it causes C2 to short-sleep.
1082 	 * PIIX4 doesn't properly support C3 and bus master activity
1083 	 * need not break out of C2.
1084 	 */
1085 	case PCI_REVISION_A_STEP:
1086 	case PCI_REVISION_B_STEP:
1087 	case PCI_REVISION_4E:
1088 	case PCI_REVISION_4M:
1089 	    acpi_cst_quirks |= ACPI_CST_QUIRK_NO_C3;
1090 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1091 		"cpu_cst: working around PIIX4 bug, disabling C3\n"));
1092 
1093 	    val = pci_read_config(acpi_dev, PIIX4_DEVACTB_REG, 4);
1094 	    if ((val & PIIX4_STOP_BREAK_MASK) != PIIX4_STOP_BREAK_MASK) {
1095 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1096 		    "cpu_cst: PIIX4: enabling IRQs to generate Stop Break\n"));
1097 	    	val |= PIIX4_STOP_BREAK_MASK;
1098 		pci_write_config(acpi_dev, PIIX4_DEVACTB_REG, val, 4);
1099 	    }
1100 	    AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_RLD, &val);
1101 	    if (val) {
1102 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1103 		    "cpu_cst: PIIX4: reset BRLD_EN_BM\n"));
1104 		AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1105 	    }
1106 	    break;
1107 	default:
1108 	    break;
1109 	}
1110     }
1111 
1112     return (0);
1113 }
1114 
1115 static int
1116 acpi_cst_usage_sysctl(SYSCTL_HANDLER_ARGS)
1117 {
1118     struct acpi_cst_softc *sc;
1119     struct sbuf	 sb;
1120     char	 buf[128];
1121     int		 i;
1122     uintmax_t	 fract, sum, whole;
1123 
1124     sc = (struct acpi_cst_softc *) arg1;
1125     sum = 0;
1126     for (i = 0; i < sc->cst_cx_count; i++)
1127 	sum += sc->cst_cx_stats[i];
1128     sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1129     for (i = 0; i < sc->cst_cx_count; i++) {
1130 	if (sum > 0) {
1131 	    whole = (uintmax_t)sc->cst_cx_stats[i] * 100;
1132 	    fract = (whole % sum) * 100;
1133 	    sbuf_printf(&sb, "%u.%02u%% ", (u_int)(whole / sum),
1134 		(u_int)(fract / sum));
1135 	} else
1136 	    sbuf_printf(&sb, "0.00%% ");
1137     }
1138     sbuf_printf(&sb, "last %dus", sc->cst_prev_sleep);
1139     sbuf_trim(&sb);
1140     sbuf_finish(&sb);
1141     sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1142     sbuf_delete(&sb);
1143 
1144     return (0);
1145 }
1146 
1147 static int
1148 acpi_cst_set_lowest_oncpu(struct acpi_cst_softc *sc, int val)
1149 {
1150     int old_lowest, error = 0, old_lowest_req;
1151     uint32_t old_type, type;
1152 
1153     KKASSERT(mycpuid == sc->cst_cpuid);
1154 
1155     old_lowest_req = sc->cst_cx_lowest_req;
1156     sc->cst_cx_lowest_req = val;
1157 
1158     if (val > sc->cst_cx_count - 1)
1159 	val = sc->cst_cx_count - 1;
1160     old_lowest = atomic_swap_int(&sc->cst_cx_lowest, val);
1161 
1162     old_type = sc->cst_cx_states[old_lowest].type;
1163     type = sc->cst_cx_states[val].type;
1164     if (old_type >= ACPI_STATE_C3 && type < ACPI_STATE_C3) {
1165 	cputimer_intr_powersave_remreq();
1166     } else if (type >= ACPI_STATE_C3 && old_type < ACPI_STATE_C3) {
1167 	error = cputimer_intr_powersave_addreq();
1168 	if (error) {
1169 	    /* Restore */
1170 	    sc->cst_cx_lowest_req = old_lowest_req;
1171 	    sc->cst_cx_lowest = old_lowest;
1172 	}
1173     }
1174 
1175     if (error)
1176 	return error;
1177 
1178     /* Cache the new lowest non-C3 state. */
1179     acpi_cst_non_c3(sc);
1180 
1181     /* Reset the statistics counters. */
1182     bzero(sc->cst_cx_stats, sizeof(sc->cst_cx_stats));
1183     return (0);
1184 }
1185 
1186 static void
1187 acpi_cst_set_lowest_handler(netmsg_t msg)
1188 {
1189     struct netmsg_acpi_cst *rmsg = (struct netmsg_acpi_cst *)msg;
1190     int error;
1191 
1192     error = acpi_cst_set_lowest_oncpu(rmsg->sc, rmsg->val);
1193     lwkt_replymsg(&rmsg->base.lmsg, error);
1194 }
1195 
1196 static int
1197 acpi_cst_set_lowest(struct acpi_cst_softc *sc, int val)
1198 {
1199     struct netmsg_acpi_cst msg;
1200 
1201     netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
1202 	acpi_cst_set_lowest_handler);
1203     msg.sc = sc;
1204     msg.val = val;
1205 
1206     return lwkt_domsg(netisr_cpuport(sc->cst_cpuid), &msg.base.lmsg, 0);
1207 }
1208 
1209 static int
1210 acpi_cst_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1211 {
1212     struct	 acpi_cst_softc *sc;
1213     char	 state[8];
1214     int		 val, error;
1215 
1216     sc = (struct acpi_cst_softc *)arg1;
1217     ksnprintf(state, sizeof(state), "C%d", sc->cst_cx_lowest_req + 1);
1218     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1219     if (error != 0 || req->newptr == NULL)
1220 	return (error);
1221     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1222 	return (EINVAL);
1223     val = (int) strtol(state + 1, NULL, 10) - 1;
1224     if (val < 0)
1225 	return (EINVAL);
1226 
1227     lwkt_serialize_enter(&acpi_cst_slize);
1228     error = acpi_cst_set_lowest(sc, val);
1229     lwkt_serialize_exit(&acpi_cst_slize);
1230 
1231     return error;
1232 }
1233 
1234 static int
1235 acpi_cst_lowest_use_sysctl(SYSCTL_HANDLER_ARGS)
1236 {
1237     struct	 acpi_cst_softc *sc;
1238     char	 state[8];
1239 
1240     sc = (struct acpi_cst_softc *)arg1;
1241     ksnprintf(state, sizeof(state), "C%d", sc->cst_cx_lowest + 1);
1242     return sysctl_handle_string(oidp, state, sizeof(state), req);
1243 }
1244 
1245 static int
1246 acpi_cst_global_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1247 {
1248     struct	acpi_cst_softc *sc;
1249     char	state[8];
1250     int		val, error, i;
1251 
1252     ksnprintf(state, sizeof(state), "C%d", acpi_cst_cx_lowest_req + 1);
1253     error = sysctl_handle_string(oidp, state, sizeof(state), req);
1254     if (error != 0 || req->newptr == NULL)
1255 	return (error);
1256     if (strlen(state) < 2 || toupper(state[0]) != 'C')
1257 	return (EINVAL);
1258     val = (int) strtol(state + 1, NULL, 10) - 1;
1259     if (val < 0)
1260 	return (EINVAL);
1261 
1262     lwkt_serialize_enter(&acpi_cst_slize);
1263 
1264     acpi_cst_cx_lowest_req = val;
1265     acpi_cst_cx_lowest = val;
1266     if (acpi_cst_cx_lowest > acpi_cst_cx_count - 1)
1267 	acpi_cst_cx_lowest = acpi_cst_cx_count - 1;
1268 
1269     /* Update the new lowest useable Cx state for all CPUs. */
1270     for (i = 0; i < acpi_cst_ndevices; i++) {
1271 	sc = device_get_softc(acpi_cst_devices[i]);
1272 	error = acpi_cst_set_lowest(sc, val);
1273 	if (error) {
1274 	    KKASSERT(i == 0);
1275 	    break;
1276 	}
1277     }
1278 
1279     lwkt_serialize_exit(&acpi_cst_slize);
1280 
1281     return error;
1282 }
1283 
1284 static int
1285 acpi_cst_global_lowest_use_sysctl(SYSCTL_HANDLER_ARGS)
1286 {
1287     char	state[8];
1288 
1289     ksnprintf(state, sizeof(state), "C%d", acpi_cst_cx_lowest + 1);
1290     return sysctl_handle_string(oidp, state, sizeof(state), req);
1291 }
1292 
1293 /*
1294  * Put the CPU in C1 in a machine-dependant way.
1295  * XXX: shouldn't be here!
1296  */
1297 static void
1298 acpi_cst_c1_halt(void)
1299 {
1300     splz();
1301     if ((mycpu->gd_reqflags & RQF_IDLECHECK_WK_MASK) == 0)
1302         __asm __volatile("sti; hlt");
1303     else
1304         __asm __volatile("sti; pause");
1305 }
1306 
1307 static void
1308 acpi_cst_non_c3(struct acpi_cst_softc *sc)
1309 {
1310     int i;
1311 
1312     sc->cst_non_c3 = 0;
1313     for (i = sc->cst_cx_lowest; i >= 0; i--) {
1314 	if (sc->cst_cx_states[i].type < ACPI_STATE_C3) {
1315 	    sc->cst_non_c3 = i;
1316 	    break;
1317 	}
1318     }
1319     if (bootverbose)
1320 	device_printf(sc->cst_dev, "non-C3 %d\n", sc->cst_non_c3);
1321 }
1322 
1323 /*
1324  * Update the largest Cx state supported in the global acpi_cst_cx_count.
1325  * It will be used in the global Cx sysctl handler.
1326  */
1327 static void
1328 acpi_cst_global_cx_count(void)
1329 {
1330     struct acpi_cst_softc *sc;
1331     int i;
1332 
1333     if (acpi_cst_ndevices == 0) {
1334 	acpi_cst_cx_count = 0;
1335 	return;
1336     }
1337 
1338     sc = device_get_softc(acpi_cst_devices[0]);
1339     acpi_cst_cx_count = sc->cst_cx_count;
1340 
1341     for (i = 1; i < acpi_cst_ndevices; i++) {
1342 	struct acpi_cst_softc *sc = device_get_softc(acpi_cst_devices[i]);
1343 
1344 	if (sc->cst_cx_count < acpi_cst_cx_count)
1345 	    acpi_cst_cx_count = sc->cst_cx_count;
1346     }
1347     if (bootverbose)
1348 	kprintf("cpu_cst: global Cx count %d\n", acpi_cst_cx_count);
1349 }
1350 
1351 static void
1352 acpi_cst_c1_halt_enter(const struct acpi_cst_cx *cx __unused)
1353 {
1354     acpi_cst_c1_halt();
1355 }
1356 
1357 static void
1358 acpi_cst_cx_io_enter(const struct acpi_cst_cx *cx)
1359 {
1360     uint64_t dummy;
1361 
1362     /*
1363      * Read I/O to enter this Cx state
1364      */
1365     bus_space_read_1(cx->btag, cx->bhand, 0);
1366     /*
1367      * Perform a dummy I/O read.  Since it may take an arbitrary time
1368      * to enter the idle state, this read makes sure that we are frozen.
1369      */
1370     AcpiRead(&dummy, &AcpiGbl_FADT.XPmTimerBlock);
1371 }
1372 
1373 static int
1374 acpi_cst_cx_setup(struct acpi_cst_cx *cx)
1375 {
1376     cx->flags &= ~ACPI_CST_CX_FLAG_BM_STS;
1377     cx->preamble = ACPI_CST_CX_PREAMBLE_NONE;
1378 
1379     if (cx->type >= ACPI_STATE_C3) {
1380 	/*
1381 	 * Set the required operations for entering C3(+) state.
1382 	 * Later acpi_cst_md_cx_setup() may fix them up.
1383 	 */
1384 
1385 	/*
1386 	 * Always check BM_STS.
1387 	 */
1388 	if ((acpi_cst_quirks & ACPI_CST_QUIRK_NO_BM) == 0)
1389 	    cx->flags |= ACPI_CST_CX_FLAG_BM_STS;
1390 
1391 	/*
1392 	 * According to the ACPI specification, bus master arbitration
1393 	 * is only available on UP system.  For MP system, cache flushing
1394 	 * is required.
1395 	 */
1396 	if (ncpus == 1 && (acpi_cst_quirks & ACPI_CST_QUIRK_NO_BM) == 0)
1397 	    cx->preamble = ACPI_CST_CX_PREAMBLE_BM_ARB;
1398 	else
1399 	    cx->preamble = ACPI_CST_CX_PREAMBLE_WBINVD;
1400     }
1401     return acpi_cst_md_cx_setup(cx);
1402 }
1403 
1404 static void
1405 acpi_cst_free_resource(struct acpi_cst_softc *sc, int start)
1406 {
1407     int i;
1408 
1409     for (i = start; i < MAX_CX_STATES; ++i) {
1410 	struct acpi_cst_cx *cx = &sc->cst_cx_states[i];
1411 
1412 	if (cx->res != NULL)
1413 	    bus_release_resource(sc->cst_dev, cx->res_type, cx->rid, cx->res);
1414 	memset(cx, 0, sizeof(*cx));
1415     }
1416 }
1417