xref: /freebsd/sys/dev/acpica/acpi_cpu.c (revision 3642298923e528d795e3a30ec165d2b469e28b40)
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 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #include "opt_acpi.h"
32 #include <sys/param.h>
33 #include <sys/bus.h>
34 #include <sys/cpu.h>
35 #include <sys/kernel.h>
36 #include <sys/malloc.h>
37 #include <sys/module.h>
38 #include <sys/pcpu.h>
39 #include <sys/power.h>
40 #include <sys/proc.h>
41 #include <sys/sbuf.h>
42 #include <sys/smp.h>
43 
44 #include <dev/pci/pcivar.h>
45 #include <machine/atomic.h>
46 #include <machine/bus.h>
47 #include <sys/rman.h>
48 
49 #include <contrib/dev/acpica/acpi.h>
50 #include <dev/acpica/acpivar.h>
51 
52 /*
53  * Support for ACPI Processor devices, including C[1-3] sleep states.
54  *
55  * TODO: implement scans of all CPUs to be sure all Cx states are
56  * equivalent.
57  */
58 
59 /* Hooks for the ACPI CA debugging infrastructure */
60 #define _COMPONENT	ACPI_PROCESSOR
61 ACPI_MODULE_NAME("PROCESSOR")
62 
63 struct acpi_cx {
64     struct resource	*p_lvlx;	/* Register to read to enter state. */
65     uint32_t		 type;		/* C1-3 (C4 and up treated as C3). */
66     uint32_t		 trans_lat;	/* Transition latency (usec). */
67     uint32_t		 power;		/* Power consumed (mW). */
68     int			 res_type;	/* Resource type for p_lvlx. */
69 };
70 #define MAX_CX_STATES	 8
71 
72 struct acpi_cpu_softc {
73     device_t		 cpu_dev;
74     ACPI_HANDLE		 cpu_handle;
75     struct pcpu		*cpu_pcpu;
76     uint32_t		 cpu_acpi_id;	/* ACPI processor id */
77     uint32_t		 cpu_p_blk;	/* ACPI P_BLK location */
78     uint32_t		 cpu_p_blk_len;	/* P_BLK length (must be 6). */
79     struct acpi_cx	 cpu_cx_states[MAX_CX_STATES];
80     int			 cpu_cx_count;	/* Number of valid Cx states. */
81     int			 cpu_prev_sleep;/* Last idle sleep duration. */
82     int			 cpu_features;	/* Child driver supported features. */
83 };
84 
85 struct acpi_cpu_device {
86     struct resource_list        ad_rl;
87 };
88 
89 #define CPU_GET_REG(reg, width) 					\
90     (bus_space_read_ ## width(rman_get_bustag((reg)), 			\
91 		      rman_get_bushandle((reg)), 0))
92 #define CPU_SET_REG(reg, width, val)					\
93     (bus_space_write_ ## width(rman_get_bustag((reg)), 			\
94 		       rman_get_bushandle((reg)), 0, (val)))
95 
96 #define PM_USEC(x)	 ((x) >> 2)	/* ~4 clocks per usec (3.57955 Mhz) */
97 
98 #define ACPI_NOTIFY_CX_STATES	0x81	/* _CST changed. */
99 
100 #define CPU_QUIRK_NO_C3		(1<<0)	/* C3-type states are not usable. */
101 #define CPU_QUIRK_NO_BM_CTRL	(1<<2)	/* No bus mastering control. */
102 
103 #define PCI_VENDOR_INTEL	0x8086
104 #define PCI_DEVICE_82371AB_3	0x7113	/* PIIX4 chipset for quirks. */
105 #define PCI_REVISION_A_STEP	0
106 #define PCI_REVISION_B_STEP	1
107 #define PCI_REVISION_4E		2
108 #define PCI_REVISION_4M		3
109 
110 /* Platform hardware resource information. */
111 static uint32_t		 cpu_smi_cmd;	/* Value to write to SMI_CMD. */
112 static uint8_t		 cpu_cst_cnt;	/* Indicate we are _CST aware. */
113 static int		 cpu_rid;	/* Driver-wide resource id. */
114 static int		 cpu_quirks;	/* Indicate any hardware bugs. */
115 
116 /* Runtime state. */
117 static int		 cpu_cx_count;	/* Number of valid states */
118 static int		 cpu_non_c3;	/* Index of lowest non-C3 state. */
119 static u_int		 cpu_cx_stats[MAX_CX_STATES];/* Cx usage history. */
120 
121 /* Values for sysctl. */
122 static struct sysctl_ctx_list acpi_cpu_sysctl_ctx;
123 static struct sysctl_oid *acpi_cpu_sysctl_tree;
124 static int		 cpu_cx_lowest;
125 static char 		 cpu_cx_supported[64];
126 
127 static device_t		*cpu_devices;
128 static int		 cpu_ndevices;
129 static struct acpi_cpu_softc **cpu_softc;
130 ACPI_SERIAL_DECL(cpu, "ACPI CPU");
131 
132 static int	acpi_cpu_probe(device_t dev);
133 static int	acpi_cpu_attach(device_t dev);
134 static int	acpi_pcpu_get_id(uint32_t idx, uint32_t *acpi_id,
135 		    uint32_t *cpu_id);
136 static struct resource_list *acpi_cpu_get_rlist(device_t dev, device_t child);
137 static device_t	acpi_cpu_add_child(device_t dev, int order, const char *name,
138 		    int unit);
139 static int	acpi_cpu_read_ivar(device_t dev, device_t child, int index,
140 		    uintptr_t *result);
141 static int	acpi_cpu_shutdown(device_t dev);
142 static int	acpi_cpu_cx_probe(struct acpi_cpu_softc *sc);
143 static int	acpi_cpu_cx_cst(struct acpi_cpu_softc *sc);
144 static void	acpi_cpu_startup(void *arg);
145 static void	acpi_cpu_startup_cx(void);
146 static void	acpi_cpu_idle(void);
147 static void	acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context);
148 static int	acpi_cpu_quirks(struct acpi_cpu_softc *sc);
149 static int	acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS);
150 static int	acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
151 
152 static device_method_t acpi_cpu_methods[] = {
153     /* Device interface */
154     DEVMETHOD(device_probe,	acpi_cpu_probe),
155     DEVMETHOD(device_attach,	acpi_cpu_attach),
156     DEVMETHOD(device_detach,	bus_generic_detach),
157     DEVMETHOD(device_shutdown,	acpi_cpu_shutdown),
158     DEVMETHOD(device_suspend,	bus_generic_suspend),
159     DEVMETHOD(device_resume,	bus_generic_resume),
160 
161     /* Bus interface */
162     DEVMETHOD(bus_add_child,	acpi_cpu_add_child),
163     DEVMETHOD(bus_read_ivar,	acpi_cpu_read_ivar),
164     DEVMETHOD(bus_get_resource_list, acpi_cpu_get_rlist),
165     DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
166     DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
167     DEVMETHOD(bus_alloc_resource, bus_generic_rl_alloc_resource),
168     DEVMETHOD(bus_release_resource, bus_generic_rl_release_resource),
169     DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
170     DEVMETHOD(bus_activate_resource, bus_generic_activate_resource),
171     DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource),
172     DEVMETHOD(bus_setup_intr,	bus_generic_setup_intr),
173     DEVMETHOD(bus_teardown_intr, bus_generic_teardown_intr),
174 
175     {0, 0}
176 };
177 
178 static driver_t acpi_cpu_driver = {
179     "cpu",
180     acpi_cpu_methods,
181     sizeof(struct acpi_cpu_softc),
182 };
183 
184 static devclass_t acpi_cpu_devclass;
185 DRIVER_MODULE(cpu, acpi, acpi_cpu_driver, acpi_cpu_devclass, 0, 0);
186 MODULE_DEPEND(cpu, acpi, 1, 1, 1);
187 
188 static int
189 acpi_cpu_probe(device_t dev)
190 {
191     int			   acpi_id, cpu_id;
192     ACPI_BUFFER		   buf;
193     ACPI_HANDLE		   handle;
194     ACPI_OBJECT		   *obj;
195     ACPI_STATUS		   status;
196 
197     if (acpi_disabled("cpu") || acpi_get_type(dev) != ACPI_TYPE_PROCESSOR)
198 	return (ENXIO);
199 
200     handle = acpi_get_handle(dev);
201     if (cpu_softc == NULL)
202 	cpu_softc = malloc(sizeof(struct acpi_cpu_softc *) *
203 	    (mp_maxid + 1), M_TEMP /* XXX */, M_WAITOK | M_ZERO);
204 
205     /* Get our Processor object. */
206     buf.Pointer = NULL;
207     buf.Length = ACPI_ALLOCATE_BUFFER;
208     status = AcpiEvaluateObject(handle, NULL, NULL, &buf);
209     if (ACPI_FAILURE(status)) {
210 	device_printf(dev, "probe failed to get Processor obj - %s\n",
211 		      AcpiFormatException(status));
212 	return (ENXIO);
213     }
214     obj = (ACPI_OBJECT *)buf.Pointer;
215     if (obj->Type != ACPI_TYPE_PROCESSOR) {
216 	device_printf(dev, "Processor object has bad type %d\n", obj->Type);
217 	AcpiOsFree(obj);
218 	return (ENXIO);
219     }
220 
221     /*
222      * Find the processor associated with our unit.  We could use the
223      * ProcId as a key, however, some boxes do not have the same values
224      * in their Processor object as the ProcId values in the MADT.
225      */
226     acpi_id = obj->Processor.ProcId;
227     AcpiOsFree(obj);
228     if (acpi_pcpu_get_id(device_get_unit(dev), &acpi_id, &cpu_id) != 0)
229 	return (ENXIO);
230 
231     /*
232      * Check if we already probed this processor.  We scan the bus twice
233      * so it's possible we've already seen this one.
234      */
235     if (cpu_softc[cpu_id] != NULL)
236 	return (ENXIO);
237 
238     /* Mark this processor as in-use and save our derived id for attach. */
239     cpu_softc[cpu_id] = (void *)1;
240     acpi_set_magic(dev, cpu_id);
241     device_set_desc(dev, "ACPI CPU");
242 
243     return (0);
244 }
245 
246 static int
247 acpi_cpu_attach(device_t dev)
248 {
249     ACPI_BUFFER		   buf;
250     ACPI_OBJECT		   arg, *obj;
251     ACPI_OBJECT_LIST	   arglist;
252     struct pcpu		   *pcpu_data;
253     struct acpi_cpu_softc *sc;
254     struct acpi_softc	  *acpi_sc;
255     ACPI_STATUS		   status;
256     u_int		   features;
257     int			   cpu_id, drv_count, i;
258     driver_t 		  **drivers;
259     uint32_t		   cap_set[3];
260 
261     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
262 
263     sc = device_get_softc(dev);
264     sc->cpu_dev = dev;
265     sc->cpu_handle = acpi_get_handle(dev);
266     cpu_id = acpi_get_magic(dev);
267     cpu_softc[cpu_id] = sc;
268     pcpu_data = pcpu_find(cpu_id);
269     pcpu_data->pc_device = dev;
270     sc->cpu_pcpu = pcpu_data;
271     cpu_smi_cmd = AcpiGbl_FADT->SmiCmd;
272     cpu_cst_cnt = AcpiGbl_FADT->CstCnt;
273 
274     buf.Pointer = NULL;
275     buf.Length = ACPI_ALLOCATE_BUFFER;
276     status = AcpiEvaluateObject(sc->cpu_handle, NULL, NULL, &buf);
277     if (ACPI_FAILURE(status)) {
278 	device_printf(dev, "attach failed to get Processor obj - %s\n",
279 		      AcpiFormatException(status));
280 	return (ENXIO);
281     }
282     obj = (ACPI_OBJECT *)buf.Pointer;
283     sc->cpu_p_blk = obj->Processor.PblkAddress;
284     sc->cpu_p_blk_len = obj->Processor.PblkLength;
285     sc->cpu_acpi_id = obj->Processor.ProcId;
286     AcpiOsFree(obj);
287     ACPI_DEBUG_PRINT((ACPI_DB_INFO, "acpi_cpu%d: P_BLK at %#x/%d\n",
288 		     device_get_unit(dev), sc->cpu_p_blk, sc->cpu_p_blk_len));
289 
290     acpi_sc = acpi_device_get_parent_softc(dev);
291     sysctl_ctx_init(&acpi_cpu_sysctl_ctx);
292     acpi_cpu_sysctl_tree = SYSCTL_ADD_NODE(&acpi_cpu_sysctl_ctx,
293 	SYSCTL_CHILDREN(acpi_sc->acpi_sysctl_tree), OID_AUTO, "cpu",
294 	CTLFLAG_RD, 0, "");
295 
296     /*
297      * Before calling any CPU methods, collect child driver feature hints
298      * and notify ACPI of them.  We support unified SMP power control
299      * so advertise this ourselves.  Note this is not the same as independent
300      * SMP control where each CPU can have different settings.
301      */
302     sc->cpu_features = ACPI_CAP_SMP_SAME | ACPI_CAP_SMP_SAME_C3;
303     if (devclass_get_drivers(acpi_cpu_devclass, &drivers, &drv_count) == 0) {
304 	for (i = 0; i < drv_count; i++) {
305 	    if (ACPI_GET_FEATURES(drivers[i], &features) == 0)
306 		sc->cpu_features |= features;
307 	}
308 	free(drivers, M_TEMP);
309     }
310 
311     /*
312      * CPU capabilities are specified as a buffer of 32-bit integers:
313      * revision, count, and one or more capabilities.  The revision of
314      * "1" is not specified anywhere but seems to match Linux.  We should
315      * also support _OSC here.
316      */
317     if (sc->cpu_features) {
318 	arglist.Pointer = &arg;
319 	arglist.Count = 1;
320 	arg.Type = ACPI_TYPE_BUFFER;
321 	arg.Buffer.Length = sizeof(cap_set);
322 	arg.Buffer.Pointer = (uint8_t *)cap_set;
323 	cap_set[0] = 1; /* revision */
324 	cap_set[1] = 1; /* number of capabilities integers */
325 	cap_set[2] = sc->cpu_features;
326 	AcpiEvaluateObject(sc->cpu_handle, "_PDC", &arglist, NULL);
327     }
328 
329     /*
330      * Probe for Cx state support.  If it isn't present, free up unused
331      * resources.
332      */
333     if (acpi_cpu_cx_probe(sc) == 0) {
334 	status = AcpiInstallNotifyHandler(sc->cpu_handle, ACPI_DEVICE_NOTIFY,
335 					  acpi_cpu_notify, sc);
336 	if (device_get_unit(dev) == 0)
337 	    AcpiOsQueueForExecution(OSD_PRIORITY_LO, acpi_cpu_startup, NULL);
338     } else
339 	sysctl_ctx_free(&acpi_cpu_sysctl_ctx);
340 
341     /* Finally,  call identify and probe/attach for child devices. */
342     bus_generic_probe(dev);
343     bus_generic_attach(dev);
344 
345     return (0);
346 }
347 
348 /*
349  * Find the nth present CPU and return its pc_cpuid as well as set the
350  * pc_acpi_id from the most reliable source.
351  */
352 static int
353 acpi_pcpu_get_id(uint32_t idx, uint32_t *acpi_id, uint32_t *cpu_id)
354 {
355     struct pcpu	*pcpu_data;
356     uint32_t	 i;
357 
358     KASSERT(acpi_id != NULL, ("Null acpi_id"));
359     KASSERT(cpu_id != NULL, ("Null cpu_id"));
360     for (i = 0; i <= mp_maxid; i++) {
361 	if (CPU_ABSENT(i))
362 	    continue;
363 	pcpu_data = pcpu_find(i);
364 	KASSERT(pcpu_data != NULL, ("no pcpu data for %d", i));
365 	if (idx-- == 0) {
366 	    /*
367 	     * If pc_acpi_id was not initialized (e.g., a non-APIC UP box)
368 	     * override it with the value from the ASL.  Otherwise, if the
369 	     * two don't match, prefer the MADT-derived value.  Finally,
370 	     * return the pc_cpuid to reference this processor.
371 	     */
372 	    if (pcpu_data->pc_acpi_id == 0xffffffff)
373 		 pcpu_data->pc_acpi_id = *acpi_id;
374 	    else if (pcpu_data->pc_acpi_id != *acpi_id)
375 		*acpi_id = pcpu_data->pc_acpi_id;
376 	    *cpu_id = pcpu_data->pc_cpuid;
377 	    return (0);
378 	}
379     }
380 
381     return (ESRCH);
382 }
383 
384 static struct resource_list *
385 acpi_cpu_get_rlist(device_t dev, device_t child)
386 {
387     struct acpi_cpu_device *ad;
388 
389     ad = device_get_ivars(child);
390     if (ad == NULL)
391 	return (NULL);
392     return (&ad->ad_rl);
393 }
394 
395 static device_t
396 acpi_cpu_add_child(device_t dev, int order, const char *name, int unit)
397 {
398     struct acpi_cpu_device  *ad;
399     device_t            child;
400 
401     if ((ad = malloc(sizeof(*ad), M_TEMP, M_NOWAIT | M_ZERO)) == NULL)
402         return (NULL);
403 
404     resource_list_init(&ad->ad_rl);
405 
406     child = device_add_child_ordered(dev, order, name, unit);
407     if (child != NULL)
408         device_set_ivars(child, ad);
409     else
410 	free(ad, M_TEMP);
411     return (child);
412 }
413 
414 static int
415 acpi_cpu_read_ivar(device_t dev, device_t child, int index, uintptr_t *result)
416 {
417     struct acpi_cpu_softc *sc;
418 
419     sc = device_get_softc(dev);
420     switch (index) {
421     case ACPI_IVAR_HANDLE:
422 	*result = (uintptr_t)sc->cpu_handle;
423 	break;
424     case CPU_IVAR_PCPU:
425 	*result = (uintptr_t)sc->cpu_pcpu;
426 	break;
427     default:
428 	return (ENOENT);
429     }
430     return (0);
431 }
432 
433 static int
434 acpi_cpu_shutdown(device_t dev)
435 {
436     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
437 
438     /* Allow children to shutdown first. */
439     bus_generic_shutdown(dev);
440 
441     /* Disable any entry to the idle function. */
442     cpu_cx_count = 0;
443 
444     /* Signal and wait for all processors to exit acpi_cpu_idle(). */
445     smp_rendezvous(NULL, NULL, NULL, NULL);
446 
447     return_VALUE (0);
448 }
449 
450 static int
451 acpi_cpu_cx_probe(struct acpi_cpu_softc *sc)
452 {
453     ACPI_GENERIC_ADDRESS gas;
454     struct acpi_cx	*cx_ptr;
455     int			 error;
456 
457     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
458 
459     /*
460      * Bus mastering arbitration control is needed to keep caches coherent
461      * while sleeping in C3.  If it's not present but a working flush cache
462      * instruction is present, flush the caches before entering C3 instead.
463      * Otherwise, just disable C3 completely.
464      */
465     if (AcpiGbl_FADT->V1_Pm2CntBlk == 0 || AcpiGbl_FADT->Pm2CntLen == 0) {
466 	if (AcpiGbl_FADT->WbInvd && AcpiGbl_FADT->WbInvdFlush == 0) {
467 	    cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
468 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
469 		"acpi_cpu%d: no BM control, using flush cache method\n",
470 		device_get_unit(sc->cpu_dev)));
471 	} else {
472 	    cpu_quirks |= CPU_QUIRK_NO_C3;
473 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
474 		"acpi_cpu%d: no BM control, C3 not available\n",
475 		device_get_unit(sc->cpu_dev)));
476 	}
477     }
478 
479     /*
480      * First, check for the ACPI 2.0 _CST sleep states object.
481      * If not usable, fall back to the P_BLK's P_LVL2 and P_LVL3.
482      */
483     sc->cpu_cx_count = 0;
484     error = acpi_cpu_cx_cst(sc);
485     if (error != 0) {
486 	cx_ptr = sc->cpu_cx_states;
487 
488 	/* C1 has been required since just after ACPI 1.0 */
489 	cx_ptr->type = ACPI_STATE_C1;
490 	cx_ptr->trans_lat = 0;
491 	cpu_non_c3 = 0;
492 	cx_ptr++;
493 	sc->cpu_cx_count++;
494 
495 	/*
496 	 * The spec says P_BLK must be 6 bytes long.  However, some systems
497 	 * use it to indicate a fractional set of features present so we
498 	 * take 5 as C2.  Some may also have a value of 7 to indicate
499 	 * another C3 but most use _CST for this (as required) and having
500 	 * "only" C1-C3 is not a hardship.
501 	 */
502 	if (sc->cpu_p_blk_len < 5)
503 	    goto done;
504 
505 	/* Validate and allocate resources for C2 (P_LVL2). */
506 	gas.AddressSpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
507 	gas.RegisterBitWidth = 8;
508 	if (AcpiGbl_FADT->Plvl2Lat <= 100) {
509 	    gas.Address = sc->cpu_p_blk + 4;
510 	    acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &cpu_rid, &gas,
511 		&cx_ptr->p_lvlx);
512 	    if (cx_ptr->p_lvlx != NULL) {
513 		cpu_rid++;
514 		cx_ptr->type = ACPI_STATE_C2;
515 		cx_ptr->trans_lat = AcpiGbl_FADT->Plvl2Lat;
516 		cpu_non_c3 = 1;
517 		cx_ptr++;
518 		sc->cpu_cx_count++;
519 	    }
520 	}
521 	if (sc->cpu_p_blk_len < 6)
522 	    goto done;
523 
524 	/* Validate and allocate resources for C3 (P_LVL3). */
525 	if (AcpiGbl_FADT->Plvl3Lat <= 1000 &&
526 	    (cpu_quirks & CPU_QUIRK_NO_C3) == 0) {
527 	    gas.Address = sc->cpu_p_blk + 5;
528 	    acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &cpu_rid, &gas,
529 		&cx_ptr->p_lvlx);
530 	    if (cx_ptr->p_lvlx != NULL) {
531 		cpu_rid++;
532 		cx_ptr->type = ACPI_STATE_C3;
533 		cx_ptr->trans_lat = AcpiGbl_FADT->Plvl3Lat;
534 		cx_ptr++;
535 		sc->cpu_cx_count++;
536 	    }
537 	}
538     }
539 
540 done:
541     /* If no valid registers were found, don't attach. */
542     if (sc->cpu_cx_count == 0)
543 	return (ENXIO);
544 
545     /* Use initial sleep value of 1 sec. to start with lowest idle state. */
546     sc->cpu_prev_sleep = 1000000;
547 
548     return (0);
549 }
550 
551 /*
552  * Parse a _CST package and set up its Cx states.  Since the _CST object
553  * can change dynamically, our notify handler may call this function
554  * to clean up and probe the new _CST package.
555  */
556 static int
557 acpi_cpu_cx_cst(struct acpi_cpu_softc *sc)
558 {
559     struct	 acpi_cx *cx_ptr;
560     ACPI_STATUS	 status;
561     ACPI_BUFFER	 buf;
562     ACPI_OBJECT	*top;
563     ACPI_OBJECT	*pkg;
564     uint32_t	 count;
565     int		 i;
566 
567     ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
568 
569     buf.Pointer = NULL;
570     buf.Length = ACPI_ALLOCATE_BUFFER;
571     status = AcpiEvaluateObject(sc->cpu_handle, "_CST", NULL, &buf);
572     if (ACPI_FAILURE(status))
573 	return (ENXIO);
574 
575     /* _CST is a package with a count and at least one Cx package. */
576     top = (ACPI_OBJECT *)buf.Pointer;
577     if (!ACPI_PKG_VALID(top, 2) || acpi_PkgInt32(top, 0, &count) != 0) {
578 	device_printf(sc->cpu_dev, "Invalid _CST package\n");
579 	AcpiOsFree(buf.Pointer);
580 	return (ENXIO);
581     }
582     if (count != top->Package.Count - 1) {
583 	device_printf(sc->cpu_dev, "Invalid _CST state count (%d != %d)\n",
584 	       count, top->Package.Count - 1);
585 	count = top->Package.Count - 1;
586     }
587     if (count > MAX_CX_STATES) {
588 	device_printf(sc->cpu_dev, "_CST has too many states (%d)\n", count);
589 	count = MAX_CX_STATES;
590     }
591 
592     /* Set up all valid states. */
593     sc->cpu_cx_count = 0;
594     cx_ptr = sc->cpu_cx_states;
595     for (i = 0; i < count; i++) {
596 	pkg = &top->Package.Elements[i + 1];
597 	if (!ACPI_PKG_VALID(pkg, 4) ||
598 	    acpi_PkgInt32(pkg, 1, &cx_ptr->type) != 0 ||
599 	    acpi_PkgInt32(pkg, 2, &cx_ptr->trans_lat) != 0 ||
600 	    acpi_PkgInt32(pkg, 3, &cx_ptr->power) != 0) {
601 
602 	    device_printf(sc->cpu_dev, "Skipping invalid Cx state package\n");
603 	    continue;
604 	}
605 
606 	/* Validate the state to see if we should use it. */
607 	switch (cx_ptr->type) {
608 	case ACPI_STATE_C1:
609 	    cpu_non_c3 = i;
610 	    cx_ptr++;
611 	    sc->cpu_cx_count++;
612 	    continue;
613 	case ACPI_STATE_C2:
614 	    if (cx_ptr->trans_lat > 100) {
615 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
616 				 "acpi_cpu%d: C2[%d] not available.\n",
617 				 device_get_unit(sc->cpu_dev), i));
618 		continue;
619 	    }
620 	    cpu_non_c3 = i;
621 	    break;
622 	case ACPI_STATE_C3:
623 	default:
624 	    if (cx_ptr->trans_lat > 1000 ||
625 		(cpu_quirks & CPU_QUIRK_NO_C3) != 0) {
626 
627 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
628 				 "acpi_cpu%d: C3[%d] not available.\n",
629 				 device_get_unit(sc->cpu_dev), i));
630 		continue;
631 	    }
632 	    break;
633 	}
634 
635 #ifdef notyet
636 	/* Free up any previous register. */
637 	if (cx_ptr->p_lvlx != NULL) {
638 	    bus_release_resource(sc->cpu_dev, 0, 0, cx_ptr->p_lvlx);
639 	    cx_ptr->p_lvlx = NULL;
640 	}
641 #endif
642 
643 	/* Allocate the control register for C2 or C3. */
644 	acpi_PkgGas(sc->cpu_dev, pkg, 0, &cx_ptr->res_type, &cpu_rid,
645 	    &cx_ptr->p_lvlx);
646 	if (cx_ptr->p_lvlx) {
647 	    cpu_rid++;
648 	    ACPI_DEBUG_PRINT((ACPI_DB_INFO,
649 			     "acpi_cpu%d: Got C%d - %d latency\n",
650 			     device_get_unit(sc->cpu_dev), cx_ptr->type,
651 			     cx_ptr->trans_lat));
652 	    cx_ptr++;
653 	    sc->cpu_cx_count++;
654 	}
655     }
656     AcpiOsFree(buf.Pointer);
657 
658     return (0);
659 }
660 
661 /*
662  * Call this *after* all CPUs have been attached.
663  */
664 static void
665 acpi_cpu_startup(void *arg)
666 {
667     struct acpi_cpu_softc *sc;
668     int count, i;
669 
670     /* Get set of CPU devices */
671     devclass_get_devices(acpi_cpu_devclass, &cpu_devices, &cpu_ndevices);
672 
673     /* Check for quirks via the first CPU device. */
674     sc = device_get_softc(cpu_devices[0]);
675     acpi_cpu_quirks(sc);
676 
677     /*
678      * Make sure all the processors' Cx counts match.  We should probably
679      * also check the contents of each.  However, no known systems have
680      * non-matching Cx counts so we'll deal with this later.
681      */
682     count = MAX_CX_STATES;
683     for (i = 0; i < cpu_ndevices; i++) {
684 	sc = device_get_softc(cpu_devices[i]);
685 	count = min(sc->cpu_cx_count, count);
686     }
687     cpu_cx_count = count;
688 
689     /* Perform Cx final initialization. */
690     sc = device_get_softc(cpu_devices[0]);
691     if (cpu_cx_count > 0)
692 	acpi_cpu_startup_cx();
693 }
694 
695 static void
696 acpi_cpu_startup_cx()
697 {
698     struct acpi_cpu_softc *sc;
699     struct sbuf sb;
700     int i;
701 
702     /*
703      * Set up the list of Cx states, eliminating C3 states by truncating
704      * cpu_cx_count if quirks indicate C3 is not usable.
705      */
706     sc = device_get_softc(cpu_devices[0]);
707     sbuf_new(&sb, cpu_cx_supported, sizeof(cpu_cx_supported), SBUF_FIXEDLEN);
708     for (i = 0; i < cpu_cx_count; i++) {
709 	if ((cpu_quirks & CPU_QUIRK_NO_C3) == 0 ||
710 	    sc->cpu_cx_states[i].type != ACPI_STATE_C3)
711 	    sbuf_printf(&sb, "C%d/%d ", i + 1, sc->cpu_cx_states[i].trans_lat);
712 	else
713 	    cpu_cx_count = i;
714     }
715     sbuf_trim(&sb);
716     sbuf_finish(&sb);
717     SYSCTL_ADD_STRING(&acpi_cpu_sysctl_ctx,
718 		      SYSCTL_CHILDREN(acpi_cpu_sysctl_tree),
719 		      OID_AUTO, "cx_supported", CTLFLAG_RD, cpu_cx_supported,
720 		      0, "Cx/microsecond values for supported Cx states");
721     SYSCTL_ADD_PROC(&acpi_cpu_sysctl_ctx,
722 		    SYSCTL_CHILDREN(acpi_cpu_sysctl_tree),
723 		    OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
724 		    NULL, 0, acpi_cpu_cx_lowest_sysctl, "A",
725 		    "lowest Cx sleep state to use");
726     SYSCTL_ADD_PROC(&acpi_cpu_sysctl_ctx,
727 		    SYSCTL_CHILDREN(acpi_cpu_sysctl_tree),
728 		    OID_AUTO, "cx_usage", CTLTYPE_STRING | CTLFLAG_RD,
729 		    NULL, 0, acpi_cpu_usage_sysctl, "A",
730 		    "percent usage for each Cx state");
731 
732 #ifdef notyet
733     /* Signal platform that we can handle _CST notification. */
734     if (cpu_cst_cnt != 0) {
735 	ACPI_LOCK(acpi);
736 	AcpiOsWritePort(cpu_smi_cmd, cpu_cst_cnt, 8);
737 	ACPI_UNLOCK(acpi);
738     }
739 #endif
740 
741     /* Take over idling from cpu_idle_default(). */
742     cpu_idle_hook = acpi_cpu_idle;
743 }
744 
745 /*
746  * Idle the CPU in the lowest state possible.  This function is called with
747  * interrupts disabled.  Note that once it re-enables interrupts, a task
748  * switch can occur so do not access shared data (i.e. the softc) after
749  * interrupts are re-enabled.
750  */
751 static void
752 acpi_cpu_idle()
753 {
754     struct	acpi_cpu_softc *sc;
755     struct	acpi_cx *cx_next;
756     uint32_t	start_time, end_time;
757     int		bm_active, cx_next_idx, i;
758 
759     /* If disabled, return immediately. */
760     if (cpu_cx_count == 0) {
761 	ACPI_ENABLE_IRQS();
762 	return;
763     }
764 
765     /*
766      * Look up our CPU id to get our softc.  If it's NULL, we'll use C1
767      * since there is no ACPI processor object for this CPU.  This occurs
768      * for logical CPUs in the HTT case.
769      */
770     sc = cpu_softc[PCPU_GET(cpuid)];
771     if (sc == NULL) {
772 	acpi_cpu_c1();
773 	return;
774     }
775 
776     /*
777      * If we slept 100 us or more, use the lowest Cx state.  Otherwise,
778      * find the lowest state that has a latency less than or equal to
779      * the length of our last sleep.
780      */
781     cx_next_idx = cpu_cx_lowest;
782     if (sc->cpu_prev_sleep < 100)
783 	for (i = cpu_cx_lowest; i >= 0; i--)
784 	    if (sc->cpu_cx_states[i].trans_lat <= sc->cpu_prev_sleep) {
785 		cx_next_idx = i;
786 		break;
787 	    }
788 
789     /*
790      * Check for bus master activity.  If there was activity, clear
791      * the bit and use the lowest non-C3 state.  Note that the USB
792      * driver polling for new devices keeps this bit set all the
793      * time if USB is loaded.
794      */
795     if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
796 	AcpiGetRegister(ACPI_BITREG_BUS_MASTER_STATUS, &bm_active,
797 	    ACPI_MTX_DO_NOT_LOCK);
798 	if (bm_active != 0) {
799 	    AcpiSetRegister(ACPI_BITREG_BUS_MASTER_STATUS, 1,
800 		ACPI_MTX_DO_NOT_LOCK);
801 	    cx_next_idx = min(cx_next_idx, cpu_non_c3);
802 	}
803     }
804 
805     /* Select the next state and update statistics. */
806     cx_next = &sc->cpu_cx_states[cx_next_idx];
807     cpu_cx_stats[cx_next_idx]++;
808     KASSERT(cx_next->type != ACPI_STATE_C0, ("acpi_cpu_idle: C0 sleep"));
809 
810     /*
811      * Execute HLT (or equivalent) and wait for an interrupt.  We can't
812      * calculate the time spent in C1 since the place we wake up is an
813      * ISR.  Assume we slept one quantum and return.
814      */
815     if (cx_next->type == ACPI_STATE_C1) {
816 	sc->cpu_prev_sleep = 1000000 / hz;
817 	acpi_cpu_c1();
818 	return;
819     }
820 
821     /*
822      * For C3, disable bus master arbitration and enable bus master wake
823      * if BM control is available, otherwise flush the CPU cache.
824      */
825     if (cx_next->type == ACPI_STATE_C3) {
826 	if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
827 	    AcpiSetRegister(ACPI_BITREG_ARB_DISABLE, 1, ACPI_MTX_DO_NOT_LOCK);
828 	    AcpiSetRegister(ACPI_BITREG_BUS_MASTER_RLD, 1,
829 		ACPI_MTX_DO_NOT_LOCK);
830 	} else
831 	    ACPI_FLUSH_CPU_CACHE();
832     }
833 
834     /*
835      * Read from P_LVLx to enter C2(+), checking time spent asleep.
836      * Use the ACPI timer for measuring sleep time.  Since we need to
837      * get the time very close to the CPU start/stop clock logic, this
838      * is the only reliable time source.
839      */
840     AcpiHwLowLevelRead(32, &start_time, &AcpiGbl_FADT->XPmTmrBlk);
841     CPU_GET_REG(cx_next->p_lvlx, 1);
842 
843     /*
844      * Read the end time twice.  Since it may take an arbitrary time
845      * to enter the idle state, the first read may be executed before
846      * the processor has stopped.  Doing it again provides enough
847      * margin that we are certain to have a correct value.
848      */
849     AcpiHwLowLevelRead(32, &end_time, &AcpiGbl_FADT->XPmTmrBlk);
850     AcpiHwLowLevelRead(32, &end_time, &AcpiGbl_FADT->XPmTmrBlk);
851 
852     /* Enable bus master arbitration and disable bus master wakeup. */
853     if (cx_next->type == ACPI_STATE_C3 &&
854 	(cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
855 	AcpiSetRegister(ACPI_BITREG_ARB_DISABLE, 0, ACPI_MTX_DO_NOT_LOCK);
856 	AcpiSetRegister(ACPI_BITREG_BUS_MASTER_RLD, 0, ACPI_MTX_DO_NOT_LOCK);
857     }
858 
859     /* Find the actual time asleep in microseconds, minus overhead. */
860     end_time = acpi_TimerDelta(end_time, start_time);
861     sc->cpu_prev_sleep = PM_USEC(end_time) - cx_next->trans_lat;
862     ACPI_ENABLE_IRQS();
863 }
864 
865 /*
866  * Re-evaluate the _CST object when we are notified that it changed.
867  *
868  * XXX Re-evaluation disabled until locking is done.
869  */
870 static void
871 acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context)
872 {
873     struct acpi_cpu_softc *sc = (struct acpi_cpu_softc *)context;
874 
875     if (notify != ACPI_NOTIFY_CX_STATES)
876 	return;
877 
878     device_printf(sc->cpu_dev, "Cx states changed\n");
879     /* acpi_cpu_cx_cst(sc); */
880 }
881 
882 static int
883 acpi_cpu_quirks(struct acpi_cpu_softc *sc)
884 {
885     device_t acpi_dev;
886 
887     /*
888      * C3 on multiple CPUs requires using the expensive flush cache
889      * instruction.
890      */
891     if (mp_ncpus > 1)
892 	cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
893 
894     /* Look for various quirks of the PIIX4 part. */
895     acpi_dev = pci_find_device(PCI_VENDOR_INTEL, PCI_DEVICE_82371AB_3);
896     if (acpi_dev != NULL) {
897 	switch (pci_get_revid(acpi_dev)) {
898 	/*
899 	 * Disable C3 support for all PIIX4 chipsets.  Some of these parts
900 	 * do not report the BMIDE status to the BM status register and
901 	 * others have a livelock bug if Type-F DMA is enabled.  Linux
902 	 * works around the BMIDE bug by reading the BM status directly
903 	 * but we take the simpler approach of disabling C3 for these
904 	 * parts.
905 	 *
906 	 * See erratum #18 ("C3 Power State/BMIDE and Type-F DMA
907 	 * Livelock") from the January 2002 PIIX4 specification update.
908 	 * Applies to all PIIX4 models.
909 	 */
910 	case PCI_REVISION_4E:
911 	case PCI_REVISION_4M:
912 	    cpu_quirks |= CPU_QUIRK_NO_C3;
913 	    break;
914 	default:
915 	    break;
916 	}
917     }
918 
919     return (0);
920 }
921 
922 static int
923 acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS)
924 {
925     struct sbuf	 sb;
926     char	 buf[128];
927     int		 i;
928     uintmax_t	 fract, sum, whole;
929 
930     sum = 0;
931     for (i = 0; i < cpu_cx_count; i++)
932 	sum += cpu_cx_stats[i];
933     sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
934     for (i = 0; i < cpu_cx_count; i++) {
935 	if (sum > 0) {
936 	    whole = (uintmax_t)cpu_cx_stats[i] * 100;
937 	    fract = (whole % sum) * 100;
938 	    sbuf_printf(&sb, "%u.%02u%% ", (u_int)(whole / sum),
939 		(u_int)(fract / sum));
940 	} else
941 	    sbuf_printf(&sb, "0%% ");
942     }
943     sbuf_trim(&sb);
944     sbuf_finish(&sb);
945     sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
946     sbuf_delete(&sb);
947 
948     return (0);
949 }
950 
951 static int
952 acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
953 {
954     struct	 acpi_cpu_softc *sc;
955     char	 state[8];
956     int		 val, error, i;
957 
958     sc = device_get_softc(cpu_devices[0]);
959     snprintf(state, sizeof(state), "C%d", cpu_cx_lowest + 1);
960     error = sysctl_handle_string(oidp, state, sizeof(state), req);
961     if (error != 0 || req->newptr == NULL)
962 	return (error);
963     if (strlen(state) < 2 || toupper(state[0]) != 'C')
964 	return (EINVAL);
965     val = (int) strtol(state + 1, NULL, 10) - 1;
966     if (val < 0 || val > cpu_cx_count - 1)
967 	return (EINVAL);
968 
969     ACPI_SERIAL_BEGIN(cpu);
970     cpu_cx_lowest = val;
971 
972     /* If not disabling, cache the new lowest non-C3 state. */
973     cpu_non_c3 = 0;
974     for (i = cpu_cx_lowest; i >= 0; i--) {
975 	if (sc->cpu_cx_states[i].type < ACPI_STATE_C3) {
976 	    cpu_non_c3 = i;
977 	    break;
978 	}
979     }
980 
981     /* Reset the statistics counters. */
982     bzero(cpu_cx_stats, sizeof(cpu_cx_stats));
983     ACPI_SERIAL_END(cpu);
984 
985     return (0);
986 }
987