2 * Copyright (c) 2003-2005 Nate Lawson (SDG)
3 * Copyright (c) 2001 Michael Smith
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
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.
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
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
32 #include <sys/param.h>
35 #include <sys/kernel.h>
36 #include <sys/malloc.h>
37 #include <sys/module.h>
39 #include <sys/power.h>
41 #include <sys/sched.h>
45 #include <dev/pci/pcivar.h>
46 #include <machine/atomic.h>
47 #include <machine/bus.h>
48 #if defined(__amd64__) || defined(__i386__)
49 #include <machine/clock.h>
53 #include <contrib/dev/acpica/include/acpi.h>
54 #include <contrib/dev/acpica/include/accommon.h>
56 #include <dev/acpica/acpivar.h>
59 * Support for ACPI Processor devices, including C[1-3] sleep states.
62 /* Hooks for the ACPI CA debugging infrastructure */
63 #define _COMPONENT ACPI_PROCESSOR
64 ACPI_MODULE_NAME("PROCESSOR")
67 struct resource *p_lvlx; /* Register to read to enter state. */
68 uint32_t type; /* C1-3 (C4 and up treated as C3). */
69 uint32_t trans_lat; /* Transition latency (usec). */
70 uint32_t power; /* Power consumed (mW). */
71 int res_type; /* Resource type for p_lvlx. */
72 int res_rid; /* Resource ID for p_lvlx. */
74 #define MAX_CX_STATES 8
76 struct acpi_cpu_softc {
78 ACPI_HANDLE cpu_handle;
79 struct pcpu *cpu_pcpu;
80 uint32_t cpu_acpi_id; /* ACPI processor id */
81 uint32_t cpu_p_blk; /* ACPI P_BLK location */
82 uint32_t cpu_p_blk_len; /* P_BLK length (must be 6). */
83 struct acpi_cx cpu_cx_states[MAX_CX_STATES];
84 int cpu_cx_count; /* Number of valid Cx states. */
85 int cpu_prev_sleep;/* Last idle sleep duration. */
86 int cpu_features; /* Child driver supported features. */
88 int cpu_non_c3; /* Index of lowest non-C3 state. */
89 u_int cpu_cx_stats[MAX_CX_STATES];/* Cx usage history. */
90 /* Values for sysctl. */
91 struct sysctl_ctx_list cpu_sysctl_ctx;
92 struct sysctl_oid *cpu_sysctl_tree;
94 int cpu_cx_lowest_lim;
95 int cpu_disable_idle; /* Disable entry to idle function */
96 char cpu_cx_supported[64];
99 struct acpi_cpu_device {
100 struct resource_list ad_rl;
103 #define CPU_GET_REG(reg, width) \
104 (bus_space_read_ ## width(rman_get_bustag((reg)), \
105 rman_get_bushandle((reg)), 0))
106 #define CPU_SET_REG(reg, width, val) \
107 (bus_space_write_ ## width(rman_get_bustag((reg)), \
108 rman_get_bushandle((reg)), 0, (val)))
110 #define PM_USEC(x) ((x) >> 2) /* ~4 clocks per usec (3.57955 Mhz) */
112 #define ACPI_NOTIFY_CX_STATES 0x81 /* _CST changed. */
114 #define CPU_QUIRK_NO_C3 (1<<0) /* C3-type states are not usable. */
115 #define CPU_QUIRK_NO_BM_CTRL (1<<2) /* No bus mastering control. */
117 #define PCI_VENDOR_INTEL 0x8086
118 #define PCI_DEVICE_82371AB_3 0x7113 /* PIIX4 chipset for quirks. */
119 #define PCI_REVISION_A_STEP 0
120 #define PCI_REVISION_B_STEP 1
121 #define PCI_REVISION_4E 2
122 #define PCI_REVISION_4M 3
123 #define PIIX4_DEVACTB_REG 0x58
124 #define PIIX4_BRLD_EN_IRQ0 (1<<0)
125 #define PIIX4_BRLD_EN_IRQ (1<<1)
126 #define PIIX4_BRLD_EN_IRQ8 (1<<5)
127 #define PIIX4_STOP_BREAK_MASK (PIIX4_BRLD_EN_IRQ0 | PIIX4_BRLD_EN_IRQ | PIIX4_BRLD_EN_IRQ8)
128 #define PIIX4_PCNTRL_BST_EN (1<<10)
130 /* Allow users to ignore processor orders in MADT. */
131 static int cpu_unordered;
132 SYSCTL_INT(_debug_acpi, OID_AUTO, cpu_unordered, CTLFLAG_RDTUN,
134 "Do not use the MADT to match ACPI Processor objects to CPUs.");
136 /* Knob to disable acpi_cpu devices */
137 bool acpi_cpu_disabled = false;
139 /* Platform hardware resource information. */
140 static uint32_t cpu_smi_cmd; /* Value to write to SMI_CMD. */
141 static uint8_t cpu_cst_cnt; /* Indicate we are _CST aware. */
142 static int cpu_quirks; /* Indicate any hardware bugs. */
144 /* Values for sysctl. */
145 static struct sysctl_ctx_list cpu_sysctl_ctx;
146 static struct sysctl_oid *cpu_sysctl_tree;
147 static int cpu_cx_generic;
148 static int cpu_cx_lowest_lim;
150 static device_t *cpu_devices;
151 static int cpu_ndevices;
152 static struct acpi_cpu_softc **cpu_softc;
153 ACPI_SERIAL_DECL(cpu, "ACPI CPU");
155 static int acpi_cpu_probe(device_t dev);
156 static int acpi_cpu_attach(device_t dev);
157 static int acpi_cpu_suspend(device_t dev);
158 static int acpi_cpu_resume(device_t dev);
159 static int acpi_pcpu_get_id(device_t dev, uint32_t *acpi_id,
161 static struct resource_list *acpi_cpu_get_rlist(device_t dev, device_t child);
162 static device_t acpi_cpu_add_child(device_t dev, u_int order, const char *name,
164 static int acpi_cpu_read_ivar(device_t dev, device_t child, int index,
166 static int acpi_cpu_shutdown(device_t dev);
167 static void acpi_cpu_cx_probe(struct acpi_cpu_softc *sc);
168 static void acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc);
169 static int acpi_cpu_cx_cst(struct acpi_cpu_softc *sc);
170 static void acpi_cpu_startup(void *arg);
171 static void acpi_cpu_startup_cx(struct acpi_cpu_softc *sc);
172 static void acpi_cpu_cx_list(struct acpi_cpu_softc *sc);
173 static void acpi_cpu_idle(sbintime_t sbt);
174 static void acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context);
175 static int acpi_cpu_quirks(void);
176 static int acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS);
177 static int acpi_cpu_usage_counters_sysctl(SYSCTL_HANDLER_ARGS);
178 static int acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc);
179 static int acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
180 static int acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
182 static device_method_t acpi_cpu_methods[] = {
183 /* Device interface */
184 DEVMETHOD(device_probe, acpi_cpu_probe),
185 DEVMETHOD(device_attach, acpi_cpu_attach),
186 DEVMETHOD(device_detach, bus_generic_detach),
187 DEVMETHOD(device_shutdown, acpi_cpu_shutdown),
188 DEVMETHOD(device_suspend, acpi_cpu_suspend),
189 DEVMETHOD(device_resume, acpi_cpu_resume),
192 DEVMETHOD(bus_add_child, acpi_cpu_add_child),
193 DEVMETHOD(bus_read_ivar, acpi_cpu_read_ivar),
194 DEVMETHOD(bus_get_resource_list, acpi_cpu_get_rlist),
195 DEVMETHOD(bus_get_resource, bus_generic_rl_get_resource),
196 DEVMETHOD(bus_set_resource, bus_generic_rl_set_resource),
197 DEVMETHOD(bus_alloc_resource, bus_generic_rl_alloc_resource),
198 DEVMETHOD(bus_release_resource, bus_generic_rl_release_resource),
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),
207 static driver_t acpi_cpu_driver = {
210 sizeof(struct acpi_cpu_softc),
213 static devclass_t acpi_cpu_devclass;
214 DRIVER_MODULE(cpu, acpi, acpi_cpu_driver, acpi_cpu_devclass, 0, 0);
215 MODULE_DEPEND(cpu, acpi, 1, 1, 1);
218 acpi_cpu_probe(device_t dev)
226 if (acpi_disabled("cpu") || acpi_get_type(dev) != ACPI_TYPE_PROCESSOR ||
230 handle = acpi_get_handle(dev);
231 if (cpu_softc == NULL)
232 cpu_softc = malloc(sizeof(struct acpi_cpu_softc *) *
233 (mp_maxid + 1), M_TEMP /* XXX */, M_WAITOK | M_ZERO);
235 /* Get our Processor object. */
237 buf.Length = ACPI_ALLOCATE_BUFFER;
238 status = AcpiEvaluateObject(handle, NULL, NULL, &buf);
239 if (ACPI_FAILURE(status)) {
240 device_printf(dev, "probe failed to get Processor obj - %s\n",
241 AcpiFormatException(status));
244 obj = (ACPI_OBJECT *)buf.Pointer;
245 if (obj->Type != ACPI_TYPE_PROCESSOR) {
246 device_printf(dev, "Processor object has bad type %d\n", obj->Type);
252 * Find the processor associated with our unit. We could use the
253 * ProcId as a key, however, some boxes do not have the same values
254 * in their Processor object as the ProcId values in the MADT.
256 acpi_id = obj->Processor.ProcId;
258 if (acpi_pcpu_get_id(dev, &acpi_id, &cpu_id) != 0)
262 * Check if we already probed this processor. We scan the bus twice
263 * so it's possible we've already seen this one.
265 if (cpu_softc[cpu_id] != NULL)
268 /* Mark this processor as in-use and save our derived id for attach. */
269 cpu_softc[cpu_id] = (void *)1;
270 acpi_set_private(dev, (void*)(intptr_t)cpu_id);
271 device_set_desc(dev, "ACPI CPU");
277 acpi_cpu_attach(device_t dev)
280 ACPI_OBJECT arg[4], *obj;
281 ACPI_OBJECT_LIST arglist;
282 struct pcpu *pcpu_data;
283 struct acpi_cpu_softc *sc;
284 struct acpi_softc *acpi_sc;
287 int cpu_id, drv_count, i;
291 /* UUID needed by _OSC evaluation */
292 static uint8_t cpu_oscuuid[16] = { 0x16, 0xA6, 0x77, 0x40, 0x0C, 0x29,
293 0xBE, 0x47, 0x9E, 0xBD, 0xD8, 0x70,
294 0x58, 0x71, 0x39, 0x53 };
296 ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
298 sc = device_get_softc(dev);
300 sc->cpu_handle = acpi_get_handle(dev);
301 cpu_id = (int)(intptr_t)acpi_get_private(dev);
302 cpu_softc[cpu_id] = sc;
303 pcpu_data = pcpu_find(cpu_id);
304 pcpu_data->pc_device = dev;
305 sc->cpu_pcpu = pcpu_data;
306 cpu_smi_cmd = AcpiGbl_FADT.SmiCommand;
307 cpu_cst_cnt = AcpiGbl_FADT.CstControl;
310 buf.Length = ACPI_ALLOCATE_BUFFER;
311 status = AcpiEvaluateObject(sc->cpu_handle, NULL, NULL, &buf);
312 if (ACPI_FAILURE(status)) {
313 device_printf(dev, "attach failed to get Processor obj - %s\n",
314 AcpiFormatException(status));
317 obj = (ACPI_OBJECT *)buf.Pointer;
318 sc->cpu_p_blk = obj->Processor.PblkAddress;
319 sc->cpu_p_blk_len = obj->Processor.PblkLength;
320 sc->cpu_acpi_id = obj->Processor.ProcId;
322 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "acpi_cpu%d: P_BLK at %#x/%d\n",
323 device_get_unit(dev), sc->cpu_p_blk, sc->cpu_p_blk_len));
326 * If this is the first cpu we attach, create and initialize the generic
327 * resources that will be used by all acpi cpu devices.
329 if (device_get_unit(dev) == 0) {
330 /* Assume we won't be using generic Cx mode by default */
331 cpu_cx_generic = FALSE;
333 /* Install hw.acpi.cpu sysctl tree */
334 acpi_sc = acpi_device_get_parent_softc(dev);
335 sysctl_ctx_init(&cpu_sysctl_ctx);
336 cpu_sysctl_tree = SYSCTL_ADD_NODE(&cpu_sysctl_ctx,
337 SYSCTL_CHILDREN(acpi_sc->acpi_sysctl_tree), OID_AUTO, "cpu",
338 CTLFLAG_RD, 0, "node for CPU children");
340 /* Queue post cpu-probing task handler */
341 AcpiOsExecute(OSL_NOTIFY_HANDLER, acpi_cpu_startup, NULL);
345 * Before calling any CPU methods, collect child driver feature hints
346 * and notify ACPI of them. We support unified SMP power control
347 * so advertise this ourselves. Note this is not the same as independent
348 * SMP control where each CPU can have different settings.
350 sc->cpu_features = ACPI_CAP_SMP_SAME | ACPI_CAP_SMP_SAME_C3;
351 if (devclass_get_drivers(acpi_cpu_devclass, &drivers, &drv_count) == 0) {
352 for (i = 0; i < drv_count; i++) {
353 if (ACPI_GET_FEATURES(drivers[i], &features) == 0)
354 sc->cpu_features |= features;
356 free(drivers, M_TEMP);
360 * CPU capabilities are specified in
361 * Intel Processor Vendor-Specific ACPI Interface Specification.
363 if (sc->cpu_features) {
364 arglist.Pointer = arg;
366 arg[0].Type = ACPI_TYPE_BUFFER;
367 arg[0].Buffer.Length = sizeof(cpu_oscuuid);
368 arg[0].Buffer.Pointer = cpu_oscuuid; /* UUID */
369 arg[1].Type = ACPI_TYPE_INTEGER;
370 arg[1].Integer.Value = 1; /* revision */
371 arg[2].Type = ACPI_TYPE_INTEGER;
372 arg[2].Integer.Value = 1; /* count */
373 arg[3].Type = ACPI_TYPE_BUFFER;
374 arg[3].Buffer.Length = sizeof(cap_set); /* Capabilities buffer */
375 arg[3].Buffer.Pointer = (uint8_t *)cap_set;
376 cap_set[0] = 0; /* status */
377 cap_set[1] = sc->cpu_features;
378 status = AcpiEvaluateObject(sc->cpu_handle, "_OSC", &arglist, NULL);
379 if (ACPI_SUCCESS(status)) {
381 device_printf(dev, "_OSC returned status %#x\n", cap_set[0]);
384 arglist.Pointer = arg;
386 arg[0].Type = ACPI_TYPE_BUFFER;
387 arg[0].Buffer.Length = sizeof(cap_set);
388 arg[0].Buffer.Pointer = (uint8_t *)cap_set;
389 cap_set[0] = 1; /* revision */
390 cap_set[1] = 1; /* number of capabilities integers */
391 cap_set[2] = sc->cpu_features;
392 AcpiEvaluateObject(sc->cpu_handle, "_PDC", &arglist, NULL);
396 /* Probe for Cx state support. */
397 acpi_cpu_cx_probe(sc);
403 acpi_cpu_postattach(void *unused __unused)
409 err = devclass_get_devices(acpi_cpu_devclass, &devices, &n);
411 printf("devclass_get_devices(acpi_cpu_devclass) failed\n");
414 for (i = 0; i < n; i++)
415 bus_generic_probe(devices[i]);
416 for (i = 0; i < n; i++)
417 bus_generic_attach(devices[i]);
418 free(devices, M_TEMP);
421 SYSINIT(acpi_cpu, SI_SUB_CONFIGURE, SI_ORDER_MIDDLE,
422 acpi_cpu_postattach, NULL);
425 disable_idle(struct acpi_cpu_softc *sc)
429 CPU_SETOF(sc->cpu_pcpu->pc_cpuid, &cpuset);
430 sc->cpu_disable_idle = TRUE;
433 * Ensure that the CPU is not in idle state or in acpi_cpu_idle().
434 * Note that this code depends on the fact that the rendezvous IPI
435 * can not penetrate context where interrupts are disabled and acpi_cpu_idle
436 * is called and executed in such a context with interrupts being re-enabled
437 * right before return.
439 smp_rendezvous_cpus(cpuset, smp_no_rendevous_barrier, NULL,
440 smp_no_rendevous_barrier, NULL);
444 enable_idle(struct acpi_cpu_softc *sc)
447 sc->cpu_disable_idle = FALSE;
451 is_idle_disabled(struct acpi_cpu_softc *sc)
454 return (sc->cpu_disable_idle);
458 * Disable any entry to the idle function during suspend and re-enable it
462 acpi_cpu_suspend(device_t dev)
466 error = bus_generic_suspend(dev);
469 disable_idle(device_get_softc(dev));
474 acpi_cpu_resume(device_t dev)
477 enable_idle(device_get_softc(dev));
478 return (bus_generic_resume(dev));
482 * Find the processor associated with a given ACPI ID. By default,
483 * use the MADT to map ACPI IDs to APIC IDs and use that to locate a
484 * processor. Some systems have inconsistent ASL and MADT however.
485 * For these systems the cpu_unordered tunable can be set in which
486 * case we assume that Processor objects are listed in the same order
487 * in both the MADT and ASL.
490 acpi_pcpu_get_id(device_t dev, uint32_t *acpi_id, uint32_t *cpu_id)
495 KASSERT(acpi_id != NULL, ("Null acpi_id"));
496 KASSERT(cpu_id != NULL, ("Null cpu_id"));
497 idx = device_get_unit(dev);
500 * If pc_acpi_id for CPU 0 is not initialized (e.g. a non-APIC
501 * UP box) use the ACPI ID from the first processor we find.
503 if (idx == 0 && mp_ncpus == 1) {
505 if (pc->pc_acpi_id == 0xffffffff)
506 pc->pc_acpi_id = *acpi_id;
513 KASSERT(pc != NULL, ("no pcpu data for %d", i));
517 * If pc_acpi_id doesn't match the ACPI ID from the
518 * ASL, prefer the MADT-derived value.
520 if (pc->pc_acpi_id != *acpi_id)
521 *acpi_id = pc->pc_acpi_id;
522 *cpu_id = pc->pc_cpuid;
526 if (pc->pc_acpi_id == *acpi_id) {
529 "Processor %s (ACPI ID %u) -> APIC ID %d\n",
530 acpi_name(acpi_get_handle(dev)), *acpi_id,
532 *cpu_id = pc->pc_cpuid;
539 printf("ACPI: Processor %s (ACPI ID %u) ignored\n",
540 acpi_name(acpi_get_handle(dev)), *acpi_id);
545 static struct resource_list *
546 acpi_cpu_get_rlist(device_t dev, device_t child)
548 struct acpi_cpu_device *ad;
550 ad = device_get_ivars(child);
557 acpi_cpu_add_child(device_t dev, u_int order, const char *name, int unit)
559 struct acpi_cpu_device *ad;
562 if ((ad = malloc(sizeof(*ad), M_TEMP, M_NOWAIT | M_ZERO)) == NULL)
565 resource_list_init(&ad->ad_rl);
567 child = device_add_child_ordered(dev, order, name, unit);
569 device_set_ivars(child, ad);
576 acpi_cpu_read_ivar(device_t dev, device_t child, int index, uintptr_t *result)
578 struct acpi_cpu_softc *sc;
580 sc = device_get_softc(dev);
582 case ACPI_IVAR_HANDLE:
583 *result = (uintptr_t)sc->cpu_handle;
586 *result = (uintptr_t)sc->cpu_pcpu;
588 #if defined(__amd64__) || defined(__i386__)
589 case CPU_IVAR_NOMINAL_MHZ:
590 if (tsc_is_invariant) {
591 *result = (uintptr_t)(atomic_load_acq_64(&tsc_freq) / 1000000);
603 acpi_cpu_shutdown(device_t dev)
605 ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
607 /* Allow children to shutdown first. */
608 bus_generic_shutdown(dev);
611 * Disable any entry to the idle function.
613 disable_idle(device_get_softc(dev));
616 * CPU devices are not truely detached and remain referenced,
617 * so their resources are not freed.
624 acpi_cpu_cx_probe(struct acpi_cpu_softc *sc)
626 ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
628 /* Use initial sleep value of 1 sec. to start with lowest idle state. */
629 sc->cpu_prev_sleep = 1000000;
630 sc->cpu_cx_lowest = 0;
631 sc->cpu_cx_lowest_lim = 0;
634 * Check for the ACPI 2.0 _CST sleep states object. If we can't find
635 * any, we'll revert to generic FADT/P_BLK Cx control method which will
636 * be handled by acpi_cpu_startup. We need to defer to after having
637 * probed all the cpus in the system before probing for generic Cx
638 * states as we may already have found cpus with valid _CST packages
640 if (!cpu_cx_generic && acpi_cpu_cx_cst(sc) != 0) {
642 * We were unable to find a _CST package for this cpu or there
643 * was an error parsing it. Switch back to generic mode.
645 cpu_cx_generic = TRUE;
647 device_printf(sc->cpu_dev, "switching to generic Cx mode\n");
651 * TODO: _CSD Package should be checked here.
656 acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc)
658 ACPI_GENERIC_ADDRESS gas;
659 struct acpi_cx *cx_ptr;
661 sc->cpu_cx_count = 0;
662 cx_ptr = sc->cpu_cx_states;
664 /* Use initial sleep value of 1 sec. to start with lowest idle state. */
665 sc->cpu_prev_sleep = 1000000;
667 /* C1 has been required since just after ACPI 1.0 */
668 cx_ptr->type = ACPI_STATE_C1;
669 cx_ptr->trans_lat = 0;
671 sc->cpu_non_c3 = sc->cpu_cx_count;
675 * The spec says P_BLK must be 6 bytes long. However, some systems
676 * use it to indicate a fractional set of features present so we
677 * take 5 as C2. Some may also have a value of 7 to indicate
678 * another C3 but most use _CST for this (as required) and having
679 * "only" C1-C3 is not a hardship.
681 if (sc->cpu_p_blk_len < 5)
684 /* Validate and allocate resources for C2 (P_LVL2). */
685 gas.SpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
687 if (AcpiGbl_FADT.C2Latency <= 100) {
688 gas.Address = sc->cpu_p_blk + 4;
690 acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &cx_ptr->res_rid,
691 &gas, &cx_ptr->p_lvlx, RF_SHAREABLE);
692 if (cx_ptr->p_lvlx != NULL) {
693 cx_ptr->type = ACPI_STATE_C2;
694 cx_ptr->trans_lat = AcpiGbl_FADT.C2Latency;
696 sc->cpu_non_c3 = sc->cpu_cx_count;
700 if (sc->cpu_p_blk_len < 6)
703 /* Validate and allocate resources for C3 (P_LVL3). */
704 if (AcpiGbl_FADT.C3Latency <= 1000 && !(cpu_quirks & CPU_QUIRK_NO_C3)) {
705 gas.Address = sc->cpu_p_blk + 5;
707 acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &cx_ptr->res_rid,
708 &gas, &cx_ptr->p_lvlx, RF_SHAREABLE);
709 if (cx_ptr->p_lvlx != NULL) {
710 cx_ptr->type = ACPI_STATE_C3;
711 cx_ptr->trans_lat = AcpiGbl_FADT.C3Latency;
714 cpu_can_deep_sleep = 1;
720 * Parse a _CST package and set up its Cx states. Since the _CST object
721 * can change dynamically, our notify handler may call this function
722 * to clean up and probe the new _CST package.
725 acpi_cpu_cx_cst(struct acpi_cpu_softc *sc)
727 struct acpi_cx *cx_ptr;
735 ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
738 buf.Length = ACPI_ALLOCATE_BUFFER;
739 status = AcpiEvaluateObject(sc->cpu_handle, "_CST", NULL, &buf);
740 if (ACPI_FAILURE(status))
743 /* _CST is a package with a count and at least one Cx package. */
744 top = (ACPI_OBJECT *)buf.Pointer;
745 if (!ACPI_PKG_VALID(top, 2) || acpi_PkgInt32(top, 0, &count) != 0) {
746 device_printf(sc->cpu_dev, "invalid _CST package\n");
747 AcpiOsFree(buf.Pointer);
750 if (count != top->Package.Count - 1) {
751 device_printf(sc->cpu_dev, "invalid _CST state count (%d != %d)\n",
752 count, top->Package.Count - 1);
753 count = top->Package.Count - 1;
755 if (count > MAX_CX_STATES) {
756 device_printf(sc->cpu_dev, "_CST has too many states (%d)\n", count);
757 count = MAX_CX_STATES;
761 sc->cpu_cx_count = 0;
762 cx_ptr = sc->cpu_cx_states;
765 * C1 has been required since just after ACPI 1.0.
766 * Reserve the first slot for it.
768 cx_ptr->type = ACPI_STATE_C0;
772 /* Set up all valid states. */
773 for (i = 0; i < count; i++) {
774 pkg = &top->Package.Elements[i + 1];
775 if (!ACPI_PKG_VALID(pkg, 4) ||
776 acpi_PkgInt32(pkg, 1, &cx_ptr->type) != 0 ||
777 acpi_PkgInt32(pkg, 2, &cx_ptr->trans_lat) != 0 ||
778 acpi_PkgInt32(pkg, 3, &cx_ptr->power) != 0) {
780 device_printf(sc->cpu_dev, "skipping invalid Cx state package\n");
784 /* Validate the state to see if we should use it. */
785 switch (cx_ptr->type) {
787 if (sc->cpu_cx_states[0].type == ACPI_STATE_C0) {
788 /* This is the first C1 state. Use the reserved slot. */
789 sc->cpu_cx_states[0] = *cx_ptr;
791 sc->cpu_non_c3 = sc->cpu_cx_count;
797 sc->cpu_non_c3 = sc->cpu_cx_count;
801 if ((cpu_quirks & CPU_QUIRK_NO_C3) != 0) {
802 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
803 "acpi_cpu%d: C3[%d] not available.\n",
804 device_get_unit(sc->cpu_dev), i));
807 cpu_can_deep_sleep = 1;
811 /* Free up any previous register. */
812 if (cx_ptr->p_lvlx != NULL) {
813 bus_release_resource(sc->cpu_dev, cx_ptr->res_type, cx_ptr->res_rid,
815 cx_ptr->p_lvlx = NULL;
818 /* Allocate the control register for C2 or C3. */
819 cx_ptr->res_rid = sc->cpu_cx_count;
820 acpi_PkgGas(sc->cpu_dev, pkg, 0, &cx_ptr->res_type, &cx_ptr->res_rid,
821 &cx_ptr->p_lvlx, RF_SHAREABLE);
822 if (cx_ptr->p_lvlx) {
823 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
824 "acpi_cpu%d: Got C%d - %d latency\n",
825 device_get_unit(sc->cpu_dev), cx_ptr->type,
831 AcpiOsFree(buf.Pointer);
833 /* If C1 state was not found, we need one now. */
834 cx_ptr = sc->cpu_cx_states;
835 if (cx_ptr->type == ACPI_STATE_C0) {
836 cx_ptr->type = ACPI_STATE_C1;
837 cx_ptr->trans_lat = 0;
844 * Call this *after* all CPUs have been attached.
847 acpi_cpu_startup(void *arg)
849 struct acpi_cpu_softc *sc;
852 /* Get set of CPU devices */
853 devclass_get_devices(acpi_cpu_devclass, &cpu_devices, &cpu_ndevices);
856 * Setup any quirks that might necessary now that we have probed
861 if (cpu_cx_generic) {
863 * We are using generic Cx mode, probe for available Cx states
864 * for all processors.
866 for (i = 0; i < cpu_ndevices; i++) {
867 sc = device_get_softc(cpu_devices[i]);
868 acpi_cpu_generic_cx_probe(sc);
872 * We are using _CST mode, remove C3 state if necessary.
873 * As we now know for sure that we will be using _CST mode
874 * install our notify handler.
876 for (i = 0; i < cpu_ndevices; i++) {
877 sc = device_get_softc(cpu_devices[i]);
878 if (cpu_quirks & CPU_QUIRK_NO_C3) {
879 sc->cpu_cx_count = sc->cpu_non_c3 + 1;
881 AcpiInstallNotifyHandler(sc->cpu_handle, ACPI_DEVICE_NOTIFY,
882 acpi_cpu_notify, sc);
886 /* Perform Cx final initialization. */
887 for (i = 0; i < cpu_ndevices; i++) {
888 sc = device_get_softc(cpu_devices[i]);
889 acpi_cpu_startup_cx(sc);
892 /* Add a sysctl handler to handle global Cx lowest setting */
893 SYSCTL_ADD_PROC(&cpu_sysctl_ctx, SYSCTL_CHILDREN(cpu_sysctl_tree),
894 OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
895 NULL, 0, acpi_cpu_global_cx_lowest_sysctl, "A",
896 "Global lowest Cx sleep state to use");
898 /* Take over idling from cpu_idle_default(). */
899 cpu_cx_lowest_lim = 0;
900 for (i = 0; i < cpu_ndevices; i++) {
901 sc = device_get_softc(cpu_devices[i]);
904 cpu_idle_hook = acpi_cpu_idle;
908 acpi_cpu_cx_list(struct acpi_cpu_softc *sc)
914 * Set up the list of Cx states
916 sbuf_new(&sb, sc->cpu_cx_supported, sizeof(sc->cpu_cx_supported),
918 for (i = 0; i < sc->cpu_cx_count; i++)
919 sbuf_printf(&sb, "C%d/%d/%d ", i + 1, sc->cpu_cx_states[i].type,
920 sc->cpu_cx_states[i].trans_lat);
926 acpi_cpu_startup_cx(struct acpi_cpu_softc *sc)
928 acpi_cpu_cx_list(sc);
930 SYSCTL_ADD_STRING(&sc->cpu_sysctl_ctx,
931 SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
932 OID_AUTO, "cx_supported", CTLFLAG_RD,
933 sc->cpu_cx_supported, 0,
934 "Cx/microsecond values for supported Cx states");
935 SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
936 SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
937 OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
938 (void *)sc, 0, acpi_cpu_cx_lowest_sysctl, "A",
939 "lowest Cx sleep state to use");
940 SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
941 SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
942 OID_AUTO, "cx_usage", CTLTYPE_STRING | CTLFLAG_RD,
943 (void *)sc, 0, acpi_cpu_usage_sysctl, "A",
944 "percent usage for each Cx state");
945 SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
946 SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
947 OID_AUTO, "cx_usage_counters", CTLTYPE_STRING | CTLFLAG_RD,
948 (void *)sc, 0, acpi_cpu_usage_counters_sysctl, "A",
949 "Cx sleep state counters");
951 /* Signal platform that we can handle _CST notification. */
952 if (!cpu_cx_generic && cpu_cst_cnt != 0) {
954 AcpiOsWritePort(cpu_smi_cmd, cpu_cst_cnt, 8);
960 * Idle the CPU in the lowest state possible. This function is called with
961 * interrupts disabled. Note that once it re-enables interrupts, a task
962 * switch can occur so do not access shared data (i.e. the softc) after
963 * interrupts are re-enabled.
966 acpi_cpu_idle(sbintime_t sbt)
968 struct acpi_cpu_softc *sc;
969 struct acpi_cx *cx_next;
971 uint32_t start_time, end_time;
972 int bm_active, cx_next_idx, i, us;
975 * Look up our CPU id to get our softc. If it's NULL, we'll use C1
976 * since there is no ACPI processor object for this CPU. This occurs
977 * for logical CPUs in the HTT case.
979 sc = cpu_softc[PCPU_GET(cpuid)];
985 /* If disabled, take the safe path. */
986 if (is_idle_disabled(sc)) {
991 /* Find the lowest state that has small enough latency. */
992 us = sc->cpu_prev_sleep;
993 if (sbt >= 0 && us > (sbt >> 12))
996 if (cpu_disable_deep_sleep)
997 i = min(sc->cpu_cx_lowest, sc->cpu_non_c3);
999 i = sc->cpu_cx_lowest;
1000 for (; i >= 0; i--) {
1001 if (sc->cpu_cx_states[i].trans_lat * 3 <= us) {
1008 * Check for bus master activity. If there was activity, clear
1009 * the bit and use the lowest non-C3 state. Note that the USB
1010 * driver polling for new devices keeps this bit set all the
1011 * time if USB is loaded.
1013 if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0 &&
1014 cx_next_idx > sc->cpu_non_c3) {
1015 AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, &bm_active);
1016 if (bm_active != 0) {
1017 AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, 1);
1018 cx_next_idx = sc->cpu_non_c3;
1022 /* Select the next state and update statistics. */
1023 cx_next = &sc->cpu_cx_states[cx_next_idx];
1024 sc->cpu_cx_stats[cx_next_idx]++;
1025 KASSERT(cx_next->type != ACPI_STATE_C0, ("acpi_cpu_idle: C0 sleep"));
1028 * Execute HLT (or equivalent) and wait for an interrupt. We can't
1029 * precisely calculate the time spent in C1 since the place we wake up
1030 * is an ISR. Assume we slept no more then half of quantum, unless
1031 * we are called inside critical section, delaying context switch.
1033 if (cx_next->type == ACPI_STATE_C1) {
1034 cputicks = cpu_ticks();
1036 end_time = ((cpu_ticks() - cputicks) << 20) / cpu_tickrate();
1037 if (curthread->td_critnest == 0)
1038 end_time = min(end_time, 500000 / hz);
1039 sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + end_time) / 4;
1044 * For C3, disable bus master arbitration and enable bus master wake
1045 * if BM control is available, otherwise flush the CPU cache.
1047 if (cx_next->type == ACPI_STATE_C3) {
1048 if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
1049 AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 1);
1050 AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
1052 ACPI_FLUSH_CPU_CACHE();
1056 * Read from P_LVLx to enter C2(+), checking time spent asleep.
1057 * Use the ACPI timer for measuring sleep time. Since we need to
1058 * get the time very close to the CPU start/stop clock logic, this
1059 * is the only reliable time source.
1061 if (cx_next->type == ACPI_STATE_C3) {
1062 AcpiHwRead(&start_time, &AcpiGbl_FADT.XPmTimerBlock);
1066 cputicks = cpu_ticks();
1068 CPU_GET_REG(cx_next->p_lvlx, 1);
1071 * Read the end time twice. Since it may take an arbitrary time
1072 * to enter the idle state, the first read may be executed before
1073 * the processor has stopped. Doing it again provides enough
1074 * margin that we are certain to have a correct value.
1076 AcpiHwRead(&end_time, &AcpiGbl_FADT.XPmTimerBlock);
1077 if (cx_next->type == ACPI_STATE_C3) {
1078 AcpiHwRead(&end_time, &AcpiGbl_FADT.XPmTimerBlock);
1079 end_time = acpi_TimerDelta(end_time, start_time);
1081 end_time = ((cpu_ticks() - cputicks) << 20) / cpu_tickrate();
1083 /* Enable bus master arbitration and disable bus master wakeup. */
1084 if (cx_next->type == ACPI_STATE_C3 &&
1085 (cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
1086 AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 0);
1087 AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1091 sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + PM_USEC(end_time)) / 4;
1095 * Re-evaluate the _CST object when we are notified that it changed.
1098 acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context)
1100 struct acpi_cpu_softc *sc = (struct acpi_cpu_softc *)context;
1102 if (notify != ACPI_NOTIFY_CX_STATES)
1106 * C-state data for target CPU is going to be in flux while we execute
1107 * acpi_cpu_cx_cst, so disable entering acpi_cpu_idle.
1108 * Also, it may happen that multiple ACPI taskqueues may concurrently
1109 * execute notifications for the same CPU. ACPI_SERIAL is used to
1110 * protect against that.
1112 ACPI_SERIAL_BEGIN(cpu);
1115 /* Update the list of Cx states. */
1116 acpi_cpu_cx_cst(sc);
1117 acpi_cpu_cx_list(sc);
1118 acpi_cpu_set_cx_lowest(sc);
1121 ACPI_SERIAL_END(cpu);
1123 acpi_UserNotify("PROCESSOR", sc->cpu_handle, notify);
1127 acpi_cpu_quirks(void)
1132 ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
1135 * Bus mastering arbitration control is needed to keep caches coherent
1136 * while sleeping in C3. If it's not present but a working flush cache
1137 * instruction is present, flush the caches before entering C3 instead.
1138 * Otherwise, just disable C3 completely.
1140 if (AcpiGbl_FADT.Pm2ControlBlock == 0 ||
1141 AcpiGbl_FADT.Pm2ControlLength == 0) {
1142 if ((AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD) &&
1143 (AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD_FLUSH) == 0) {
1144 cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1145 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1146 "acpi_cpu: no BM control, using flush cache method\n"));
1148 cpu_quirks |= CPU_QUIRK_NO_C3;
1149 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1150 "acpi_cpu: no BM control, C3 not available\n"));
1155 * If we are using generic Cx mode, C3 on multiple CPUs requires using
1156 * the expensive flush cache instruction.
1158 if (cpu_cx_generic && mp_ncpus > 1) {
1159 cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1160 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1161 "acpi_cpu: SMP, using flush cache mode for C3\n"));
1164 /* Look for various quirks of the PIIX4 part. */
1165 acpi_dev = pci_find_device(PCI_VENDOR_INTEL, PCI_DEVICE_82371AB_3);
1166 if (acpi_dev != NULL) {
1167 switch (pci_get_revid(acpi_dev)) {
1169 * Disable C3 support for all PIIX4 chipsets. Some of these parts
1170 * do not report the BMIDE status to the BM status register and
1171 * others have a livelock bug if Type-F DMA is enabled. Linux
1172 * works around the BMIDE bug by reading the BM status directly
1173 * but we take the simpler approach of disabling C3 for these
1176 * See erratum #18 ("C3 Power State/BMIDE and Type-F DMA
1177 * Livelock") from the January 2002 PIIX4 specification update.
1178 * Applies to all PIIX4 models.
1180 * Also, make sure that all interrupts cause a "Stop Break"
1181 * event to exit from C2 state.
1182 * Also, BRLD_EN_BM (ACPI_BITREG_BUS_MASTER_RLD in ACPI-speak)
1183 * should be set to zero, otherwise it causes C2 to short-sleep.
1184 * PIIX4 doesn't properly support C3 and bus master activity
1185 * need not break out of C2.
1187 case PCI_REVISION_A_STEP:
1188 case PCI_REVISION_B_STEP:
1189 case PCI_REVISION_4E:
1190 case PCI_REVISION_4M:
1191 cpu_quirks |= CPU_QUIRK_NO_C3;
1192 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1193 "acpi_cpu: working around PIIX4 bug, disabling C3\n"));
1195 val = pci_read_config(acpi_dev, PIIX4_DEVACTB_REG, 4);
1196 if ((val & PIIX4_STOP_BREAK_MASK) != PIIX4_STOP_BREAK_MASK) {
1197 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1198 "acpi_cpu: PIIX4: enabling IRQs to generate Stop Break\n"));
1199 val |= PIIX4_STOP_BREAK_MASK;
1200 pci_write_config(acpi_dev, PIIX4_DEVACTB_REG, val, 4);
1202 AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_RLD, &val);
1204 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1205 "acpi_cpu: PIIX4: reset BRLD_EN_BM\n"));
1206 AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1218 acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS)
1220 struct acpi_cpu_softc *sc;
1224 uintmax_t fract, sum, whole;
1226 sc = (struct acpi_cpu_softc *) arg1;
1228 for (i = 0; i < sc->cpu_cx_count; i++)
1229 sum += sc->cpu_cx_stats[i];
1230 sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1231 for (i = 0; i < sc->cpu_cx_count; i++) {
1233 whole = (uintmax_t)sc->cpu_cx_stats[i] * 100;
1234 fract = (whole % sum) * 100;
1235 sbuf_printf(&sb, "%u.%02u%% ", (u_int)(whole / sum),
1236 (u_int)(fract / sum));
1238 sbuf_printf(&sb, "0.00%% ");
1240 sbuf_printf(&sb, "last %dus", sc->cpu_prev_sleep);
1243 sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1250 * XXX TODO: actually add support to count each entry/exit
1251 * from the Cx states.
1254 acpi_cpu_usage_counters_sysctl(SYSCTL_HANDLER_ARGS)
1256 struct acpi_cpu_softc *sc;
1261 sc = (struct acpi_cpu_softc *) arg1;
1263 /* Print out the raw counters */
1264 sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1266 for (i = 0; i < sc->cpu_cx_count; i++) {
1267 sbuf_printf(&sb, "%u ", sc->cpu_cx_stats[i]);
1272 sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1279 acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc)
1283 ACPI_SERIAL_ASSERT(cpu);
1284 sc->cpu_cx_lowest = min(sc->cpu_cx_lowest_lim, sc->cpu_cx_count - 1);
1286 /* If not disabling, cache the new lowest non-C3 state. */
1288 for (i = sc->cpu_cx_lowest; i >= 0; i--) {
1289 if (sc->cpu_cx_states[i].type < ACPI_STATE_C3) {
1295 /* Reset the statistics counters. */
1296 bzero(sc->cpu_cx_stats, sizeof(sc->cpu_cx_stats));
1301 acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1303 struct acpi_cpu_softc *sc;
1307 sc = (struct acpi_cpu_softc *) arg1;
1308 snprintf(state, sizeof(state), "C%d", sc->cpu_cx_lowest_lim + 1);
1309 error = sysctl_handle_string(oidp, state, sizeof(state), req);
1310 if (error != 0 || req->newptr == NULL)
1312 if (strlen(state) < 2 || toupper(state[0]) != 'C')
1314 if (strcasecmp(state, "Cmax") == 0)
1315 val = MAX_CX_STATES;
1317 val = (int) strtol(state + 1, NULL, 10);
1318 if (val < 1 || val > MAX_CX_STATES)
1322 ACPI_SERIAL_BEGIN(cpu);
1323 sc->cpu_cx_lowest_lim = val - 1;
1324 acpi_cpu_set_cx_lowest(sc);
1325 ACPI_SERIAL_END(cpu);
1331 acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1333 struct acpi_cpu_softc *sc;
1337 snprintf(state, sizeof(state), "C%d", cpu_cx_lowest_lim + 1);
1338 error = sysctl_handle_string(oidp, state, sizeof(state), req);
1339 if (error != 0 || req->newptr == NULL)
1341 if (strlen(state) < 2 || toupper(state[0]) != 'C')
1343 if (strcasecmp(state, "Cmax") == 0)
1344 val = MAX_CX_STATES;
1346 val = (int) strtol(state + 1, NULL, 10);
1347 if (val < 1 || val > MAX_CX_STATES)
1351 /* Update the new lowest useable Cx state for all CPUs. */
1352 ACPI_SERIAL_BEGIN(cpu);
1353 cpu_cx_lowest_lim = val - 1;
1354 for (i = 0; i < cpu_ndevices; i++) {
1355 sc = device_get_softc(cpu_devices[i]);
1356 sc->cpu_cx_lowest_lim = cpu_cx_lowest_lim;
1357 acpi_cpu_set_cx_lowest(sc);
1359 ACPI_SERIAL_END(cpu);