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>
44 #include <dev/pci/pcivar.h>
45 #include <machine/atomic.h>
46 #include <machine/bus.h>
47 #if defined(__amd64__) || defined(__i386__)
48 #include <machine/clock.h>
52 #include <contrib/dev/acpica/include/acpi.h>
53 #include <contrib/dev/acpica/include/accommon.h>
55 #include <dev/acpica/acpivar.h>
58 * Support for ACPI Processor devices, including C[1-3] sleep states.
61 /* Hooks for the ACPI CA debugging infrastructure */
62 #define _COMPONENT ACPI_PROCESSOR
63 ACPI_MODULE_NAME("PROCESSOR")
66 struct resource *p_lvlx; /* Register to read to enter state. */
67 uint32_t type; /* C1-3 (C4 and up treated as C3). */
68 uint32_t trans_lat; /* Transition latency (usec). */
69 uint32_t power; /* Power consumed (mW). */
70 int res_type; /* Resource type for p_lvlx. */
72 #define MAX_CX_STATES 8
74 struct acpi_cpu_softc {
76 ACPI_HANDLE cpu_handle;
77 struct pcpu *cpu_pcpu;
78 uint32_t cpu_acpi_id; /* ACPI processor id */
79 uint32_t cpu_p_blk; /* ACPI P_BLK location */
80 uint32_t cpu_p_blk_len; /* P_BLK length (must be 6). */
81 struct acpi_cx cpu_cx_states[MAX_CX_STATES];
82 int cpu_cx_count; /* Number of valid Cx states. */
83 int cpu_prev_sleep;/* Last idle sleep duration. */
84 int cpu_features; /* Child driver supported features. */
86 int cpu_non_c3; /* Index of lowest non-C3 state. */
87 u_int cpu_cx_stats[MAX_CX_STATES];/* Cx usage history. */
88 /* Values for sysctl. */
89 struct sysctl_ctx_list cpu_sysctl_ctx;
90 struct sysctl_oid *cpu_sysctl_tree;
92 int cpu_cx_lowest_lim;
93 char cpu_cx_supported[64];
97 struct acpi_cpu_device {
98 struct resource_list ad_rl;
101 #define CPU_GET_REG(reg, width) \
102 (bus_space_read_ ## width(rman_get_bustag((reg)), \
103 rman_get_bushandle((reg)), 0))
104 #define CPU_SET_REG(reg, width, val) \
105 (bus_space_write_ ## width(rman_get_bustag((reg)), \
106 rman_get_bushandle((reg)), 0, (val)))
108 #define PM_USEC(x) ((x) >> 2) /* ~4 clocks per usec (3.57955 Mhz) */
110 #define ACPI_NOTIFY_CX_STATES 0x81 /* _CST changed. */
112 #define CPU_QUIRK_NO_C3 (1<<0) /* C3-type states are not usable. */
113 #define CPU_QUIRK_NO_BM_CTRL (1<<2) /* No bus mastering control. */
115 #define PCI_VENDOR_INTEL 0x8086
116 #define PCI_DEVICE_82371AB_3 0x7113 /* PIIX4 chipset for quirks. */
117 #define PCI_REVISION_A_STEP 0
118 #define PCI_REVISION_B_STEP 1
119 #define PCI_REVISION_4E 2
120 #define PCI_REVISION_4M 3
121 #define PIIX4_DEVACTB_REG 0x58
122 #define PIIX4_BRLD_EN_IRQ0 (1<<0)
123 #define PIIX4_BRLD_EN_IRQ (1<<1)
124 #define PIIX4_BRLD_EN_IRQ8 (1<<5)
125 #define PIIX4_STOP_BREAK_MASK (PIIX4_BRLD_EN_IRQ0 | PIIX4_BRLD_EN_IRQ | PIIX4_BRLD_EN_IRQ8)
126 #define PIIX4_PCNTRL_BST_EN (1<<10)
128 /* Allow users to ignore processor orders in MADT. */
129 static int cpu_unordered;
130 TUNABLE_INT("debug.acpi.cpu_unordered", &cpu_unordered);
131 SYSCTL_INT(_debug_acpi, OID_AUTO, cpu_unordered, CTLFLAG_RDTUN,
133 "Do not use the MADT to match ACPI Processor objects to CPUs.");
135 /* Platform hardware resource information. */
136 static uint32_t cpu_smi_cmd; /* Value to write to SMI_CMD. */
137 static uint8_t cpu_cst_cnt; /* Indicate we are _CST aware. */
138 static int cpu_quirks; /* Indicate any hardware bugs. */
141 static int cpu_disable_idle; /* Disable entry to idle function */
143 /* Values for sysctl. */
144 static struct sysctl_ctx_list cpu_sysctl_ctx;
145 static struct sysctl_oid *cpu_sysctl_tree;
146 static int cpu_cx_generic;
147 static int cpu_cx_lowest_lim;
149 static device_t *cpu_devices;
150 static int cpu_ndevices;
151 static struct acpi_cpu_softc **cpu_softc;
152 ACPI_SERIAL_DECL(cpu, "ACPI CPU");
154 static int acpi_cpu_probe(device_t dev);
155 static int acpi_cpu_attach(device_t dev);
156 static int acpi_cpu_suspend(device_t dev);
157 static int acpi_cpu_resume(device_t dev);
158 static int acpi_pcpu_get_id(device_t dev, uint32_t *acpi_id,
160 static struct resource_list *acpi_cpu_get_rlist(device_t dev, device_t child);
161 static device_t acpi_cpu_add_child(device_t dev, u_int order, const char *name,
163 static int acpi_cpu_read_ivar(device_t dev, device_t child, int index,
165 static int acpi_cpu_shutdown(device_t dev);
166 static void acpi_cpu_cx_probe(struct acpi_cpu_softc *sc);
167 static void acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc);
168 static int acpi_cpu_cx_cst(struct acpi_cpu_softc *sc);
169 static void acpi_cpu_startup(void *arg);
170 static void acpi_cpu_startup_cx(struct acpi_cpu_softc *sc);
171 static void acpi_cpu_cx_list(struct acpi_cpu_softc *sc);
172 static void acpi_cpu_idle(void);
173 static void acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context);
174 static int acpi_cpu_quirks(void);
175 static int acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS);
176 static int acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc);
177 static int acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
178 static int acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
180 static device_method_t acpi_cpu_methods[] = {
181 /* Device interface */
182 DEVMETHOD(device_probe, acpi_cpu_probe),
183 DEVMETHOD(device_attach, acpi_cpu_attach),
184 DEVMETHOD(device_detach, bus_generic_detach),
185 DEVMETHOD(device_shutdown, acpi_cpu_shutdown),
186 DEVMETHOD(device_suspend, acpi_cpu_suspend),
187 DEVMETHOD(device_resume, acpi_cpu_resume),
190 DEVMETHOD(bus_add_child, acpi_cpu_add_child),
191 DEVMETHOD(bus_read_ivar, acpi_cpu_read_ivar),
192 DEVMETHOD(bus_get_resource_list, acpi_cpu_get_rlist),
193 DEVMETHOD(bus_get_resource, bus_generic_rl_get_resource),
194 DEVMETHOD(bus_set_resource, bus_generic_rl_set_resource),
195 DEVMETHOD(bus_alloc_resource, bus_generic_rl_alloc_resource),
196 DEVMETHOD(bus_release_resource, bus_generic_rl_release_resource),
197 DEVMETHOD(bus_activate_resource, bus_generic_activate_resource),
198 DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource),
199 DEVMETHOD(bus_setup_intr, bus_generic_setup_intr),
200 DEVMETHOD(bus_teardown_intr, bus_generic_teardown_intr),
205 static driver_t acpi_cpu_driver = {
208 sizeof(struct acpi_cpu_softc),
211 static devclass_t acpi_cpu_devclass;
212 DRIVER_MODULE(cpu, acpi, acpi_cpu_driver, acpi_cpu_devclass, 0, 0);
213 MODULE_DEPEND(cpu, acpi, 1, 1, 1);
216 acpi_cpu_probe(device_t dev)
224 if (acpi_disabled("cpu") || acpi_get_type(dev) != ACPI_TYPE_PROCESSOR)
227 handle = acpi_get_handle(dev);
228 if (cpu_softc == NULL)
229 cpu_softc = malloc(sizeof(struct acpi_cpu_softc *) *
230 (mp_maxid + 1), M_TEMP /* XXX */, M_WAITOK | M_ZERO);
232 /* Get our Processor object. */
234 buf.Length = ACPI_ALLOCATE_BUFFER;
235 status = AcpiEvaluateObject(handle, NULL, NULL, &buf);
236 if (ACPI_FAILURE(status)) {
237 device_printf(dev, "probe failed to get Processor obj - %s\n",
238 AcpiFormatException(status));
241 obj = (ACPI_OBJECT *)buf.Pointer;
242 if (obj->Type != ACPI_TYPE_PROCESSOR) {
243 device_printf(dev, "Processor object has bad type %d\n", obj->Type);
249 * Find the processor associated with our unit. We could use the
250 * ProcId as a key, however, some boxes do not have the same values
251 * in their Processor object as the ProcId values in the MADT.
253 acpi_id = obj->Processor.ProcId;
255 if (acpi_pcpu_get_id(dev, &acpi_id, &cpu_id) != 0)
259 * Check if we already probed this processor. We scan the bus twice
260 * so it's possible we've already seen this one.
262 if (cpu_softc[cpu_id] != NULL)
265 /* Mark this processor as in-use and save our derived id for attach. */
266 cpu_softc[cpu_id] = (void *)1;
267 acpi_set_private(dev, (void*)(intptr_t)cpu_id);
268 device_set_desc(dev, "ACPI CPU");
274 acpi_cpu_attach(device_t dev)
277 ACPI_OBJECT arg[4], *obj;
278 ACPI_OBJECT_LIST arglist;
279 struct pcpu *pcpu_data;
280 struct acpi_cpu_softc *sc;
281 struct acpi_softc *acpi_sc;
284 int cpu_id, drv_count, i;
288 /* UUID needed by _OSC evaluation */
289 static uint8_t cpu_oscuuid[16] = { 0x16, 0xA6, 0x77, 0x40, 0x0C, 0x29,
290 0xBE, 0x47, 0x9E, 0xBD, 0xD8, 0x70,
291 0x58, 0x71, 0x39, 0x53 };
293 ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
295 sc = device_get_softc(dev);
297 sc->cpu_handle = acpi_get_handle(dev);
298 cpu_id = (int)(intptr_t)acpi_get_private(dev);
299 cpu_softc[cpu_id] = sc;
300 pcpu_data = pcpu_find(cpu_id);
301 pcpu_data->pc_device = dev;
302 sc->cpu_pcpu = pcpu_data;
303 cpu_smi_cmd = AcpiGbl_FADT.SmiCommand;
304 cpu_cst_cnt = AcpiGbl_FADT.CstControl;
307 buf.Length = ACPI_ALLOCATE_BUFFER;
308 status = AcpiEvaluateObject(sc->cpu_handle, NULL, NULL, &buf);
309 if (ACPI_FAILURE(status)) {
310 device_printf(dev, "attach failed to get Processor obj - %s\n",
311 AcpiFormatException(status));
314 obj = (ACPI_OBJECT *)buf.Pointer;
315 sc->cpu_p_blk = obj->Processor.PblkAddress;
316 sc->cpu_p_blk_len = obj->Processor.PblkLength;
317 sc->cpu_acpi_id = obj->Processor.ProcId;
319 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "acpi_cpu%d: P_BLK at %#x/%d\n",
320 device_get_unit(dev), sc->cpu_p_blk, sc->cpu_p_blk_len));
323 * If this is the first cpu we attach, create and initialize the generic
324 * resources that will be used by all acpi cpu devices.
326 if (device_get_unit(dev) == 0) {
327 /* Assume we won't be using generic Cx mode by default */
328 cpu_cx_generic = FALSE;
330 /* Install hw.acpi.cpu sysctl tree */
331 acpi_sc = acpi_device_get_parent_softc(dev);
332 sysctl_ctx_init(&cpu_sysctl_ctx);
333 cpu_sysctl_tree = SYSCTL_ADD_NODE(&cpu_sysctl_ctx,
334 SYSCTL_CHILDREN(acpi_sc->acpi_sysctl_tree), OID_AUTO, "cpu",
335 CTLFLAG_RD, 0, "node for CPU children");
337 /* Queue post cpu-probing task handler */
338 AcpiOsExecute(OSL_NOTIFY_HANDLER, acpi_cpu_startup, NULL);
342 * Before calling any CPU methods, collect child driver feature hints
343 * and notify ACPI of them. We support unified SMP power control
344 * so advertise this ourselves. Note this is not the same as independent
345 * SMP control where each CPU can have different settings.
347 sc->cpu_features = ACPI_CAP_SMP_SAME | ACPI_CAP_SMP_SAME_C3;
348 if (devclass_get_drivers(acpi_cpu_devclass, &drivers, &drv_count) == 0) {
349 for (i = 0; i < drv_count; i++) {
350 if (ACPI_GET_FEATURES(drivers[i], &features) == 0)
351 sc->cpu_features |= features;
353 free(drivers, M_TEMP);
357 * CPU capabilities are specified in
358 * Intel Processor Vendor-Specific ACPI Interface Specification.
360 if (sc->cpu_features) {
361 arglist.Pointer = arg;
363 arg[0].Type = ACPI_TYPE_BUFFER;
364 arg[0].Buffer.Length = sizeof(cpu_oscuuid);
365 arg[0].Buffer.Pointer = cpu_oscuuid; /* UUID */
366 arg[1].Type = ACPI_TYPE_INTEGER;
367 arg[1].Integer.Value = 1; /* revision */
368 arg[2].Type = ACPI_TYPE_INTEGER;
369 arg[2].Integer.Value = 1; /* count */
370 arg[3].Type = ACPI_TYPE_BUFFER;
371 arg[3].Buffer.Length = sizeof(cap_set); /* Capabilities buffer */
372 arg[3].Buffer.Pointer = (uint8_t *)cap_set;
373 cap_set[0] = 0; /* status */
374 cap_set[1] = sc->cpu_features;
375 status = AcpiEvaluateObject(sc->cpu_handle, "_OSC", &arglist, NULL);
376 if (ACPI_SUCCESS(status)) {
378 device_printf(dev, "_OSC returned status %#x\n", cap_set[0]);
381 arglist.Pointer = arg;
383 arg[0].Type = ACPI_TYPE_BUFFER;
384 arg[0].Buffer.Length = sizeof(cap_set);
385 arg[0].Buffer.Pointer = (uint8_t *)cap_set;
386 cap_set[0] = 1; /* revision */
387 cap_set[1] = 1; /* number of capabilities integers */
388 cap_set[2] = sc->cpu_features;
389 AcpiEvaluateObject(sc->cpu_handle, "_PDC", &arglist, NULL);
393 /* Probe for Cx state support. */
394 acpi_cpu_cx_probe(sc);
400 acpi_cpu_postattach(void *unused __unused)
406 err = devclass_get_devices(acpi_cpu_devclass, &devices, &n);
408 printf("devclass_get_devices(acpi_cpu_devclass) failed\n");
411 for (i = 0; i < n; i++)
412 bus_generic_probe(devices[i]);
413 for (i = 0; i < n; i++)
414 bus_generic_attach(devices[i]);
415 free(devices, M_TEMP);
418 SYSINIT(acpi_cpu, SI_SUB_CONFIGURE, SI_ORDER_MIDDLE,
419 acpi_cpu_postattach, NULL);
422 * Disable any entry to the idle function during suspend and re-enable it
426 acpi_cpu_suspend(device_t dev)
430 error = bus_generic_suspend(dev);
433 cpu_disable_idle = TRUE;
438 acpi_cpu_resume(device_t dev)
441 cpu_disable_idle = FALSE;
442 return (bus_generic_resume(dev));
446 * Find the processor associated with a given ACPI ID. By default,
447 * use the MADT to map ACPI IDs to APIC IDs and use that to locate a
448 * processor. Some systems have inconsistent ASL and MADT however.
449 * For these systems the cpu_unordered tunable can be set in which
450 * case we assume that Processor objects are listed in the same order
451 * in both the MADT and ASL.
454 acpi_pcpu_get_id(device_t dev, uint32_t *acpi_id, uint32_t *cpu_id)
459 KASSERT(acpi_id != NULL, ("Null acpi_id"));
460 KASSERT(cpu_id != NULL, ("Null cpu_id"));
461 idx = device_get_unit(dev);
464 * If pc_acpi_id for CPU 0 is not initialized (e.g. a non-APIC
465 * UP box) use the ACPI ID from the first processor we find.
467 if (idx == 0 && mp_ncpus == 1) {
469 if (pc->pc_acpi_id == 0xffffffff)
470 pc->pc_acpi_id = *acpi_id;
477 KASSERT(pc != NULL, ("no pcpu data for %d", i));
481 * If pc_acpi_id doesn't match the ACPI ID from the
482 * ASL, prefer the MADT-derived value.
484 if (pc->pc_acpi_id != *acpi_id)
485 *acpi_id = pc->pc_acpi_id;
486 *cpu_id = pc->pc_cpuid;
490 if (pc->pc_acpi_id == *acpi_id) {
493 "Processor %s (ACPI ID %u) -> APIC ID %d\n",
494 acpi_name(acpi_get_handle(dev)), *acpi_id,
496 *cpu_id = pc->pc_cpuid;
503 printf("ACPI: Processor %s (ACPI ID %u) ignored\n",
504 acpi_name(acpi_get_handle(dev)), *acpi_id);
509 static struct resource_list *
510 acpi_cpu_get_rlist(device_t dev, device_t child)
512 struct acpi_cpu_device *ad;
514 ad = device_get_ivars(child);
521 acpi_cpu_add_child(device_t dev, u_int order, const char *name, int unit)
523 struct acpi_cpu_device *ad;
526 if ((ad = malloc(sizeof(*ad), M_TEMP, M_NOWAIT | M_ZERO)) == NULL)
529 resource_list_init(&ad->ad_rl);
531 child = device_add_child_ordered(dev, order, name, unit);
533 device_set_ivars(child, ad);
540 acpi_cpu_read_ivar(device_t dev, device_t child, int index, uintptr_t *result)
542 struct acpi_cpu_softc *sc;
544 sc = device_get_softc(dev);
546 case ACPI_IVAR_HANDLE:
547 *result = (uintptr_t)sc->cpu_handle;
550 *result = (uintptr_t)sc->cpu_pcpu;
552 #if defined(__amd64__) || defined(__i386__)
553 case CPU_IVAR_NOMINAL_MHZ:
554 if (tsc_is_invariant) {
555 *result = (uintptr_t)(atomic_load_acq_64(&tsc_freq) / 1000000);
567 acpi_cpu_shutdown(device_t dev)
569 ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
571 /* Allow children to shutdown first. */
572 bus_generic_shutdown(dev);
575 * Disable any entry to the idle function. There is a small race where
576 * an idle thread have passed this check but not gone to sleep. This
577 * is ok since device_shutdown() does not free the softc, otherwise
578 * we'd have to be sure all threads were evicted before returning.
580 cpu_disable_idle = TRUE;
586 acpi_cpu_cx_probe(struct acpi_cpu_softc *sc)
588 ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
590 /* Use initial sleep value of 1 sec. to start with lowest idle state. */
591 sc->cpu_prev_sleep = 1000000;
592 sc->cpu_cx_lowest = 0;
593 sc->cpu_cx_lowest_lim = 0;
596 * Check for the ACPI 2.0 _CST sleep states object. If we can't find
597 * any, we'll revert to generic FADT/P_BLK Cx control method which will
598 * be handled by acpi_cpu_startup. We need to defer to after having
599 * probed all the cpus in the system before probing for generic Cx
600 * states as we may already have found cpus with valid _CST packages
602 if (!cpu_cx_generic && acpi_cpu_cx_cst(sc) != 0) {
604 * We were unable to find a _CST package for this cpu or there
605 * was an error parsing it. Switch back to generic mode.
607 cpu_cx_generic = TRUE;
609 device_printf(sc->cpu_dev, "switching to generic Cx mode\n");
613 * TODO: _CSD Package should be checked here.
618 acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc)
620 ACPI_GENERIC_ADDRESS gas;
621 struct acpi_cx *cx_ptr;
623 sc->cpu_cx_count = 0;
624 cx_ptr = sc->cpu_cx_states;
626 /* Use initial sleep value of 1 sec. to start with lowest idle state. */
627 sc->cpu_prev_sleep = 1000000;
629 /* C1 has been required since just after ACPI 1.0 */
630 cx_ptr->type = ACPI_STATE_C1;
631 cx_ptr->trans_lat = 0;
636 * The spec says P_BLK must be 6 bytes long. However, some systems
637 * use it to indicate a fractional set of features present so we
638 * take 5 as C2. Some may also have a value of 7 to indicate
639 * another C3 but most use _CST for this (as required) and having
640 * "only" C1-C3 is not a hardship.
642 if (sc->cpu_p_blk_len < 5)
645 /* Validate and allocate resources for C2 (P_LVL2). */
646 gas.SpaceId = ACPI_ADR_SPACE_SYSTEM_IO;
648 if (AcpiGbl_FADT.C2Latency <= 100) {
649 gas.Address = sc->cpu_p_blk + 4;
650 acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &sc->cpu_rid,
651 &gas, &cx_ptr->p_lvlx, RF_SHAREABLE);
652 if (cx_ptr->p_lvlx != NULL) {
654 cx_ptr->type = ACPI_STATE_C2;
655 cx_ptr->trans_lat = AcpiGbl_FADT.C2Latency;
660 if (sc->cpu_p_blk_len < 6)
663 /* Validate and allocate resources for C3 (P_LVL3). */
664 if (AcpiGbl_FADT.C3Latency <= 1000 && !(cpu_quirks & CPU_QUIRK_NO_C3)) {
665 gas.Address = sc->cpu_p_blk + 5;
666 acpi_bus_alloc_gas(sc->cpu_dev, &cx_ptr->res_type, &sc->cpu_rid, &gas,
667 &cx_ptr->p_lvlx, RF_SHAREABLE);
668 if (cx_ptr->p_lvlx != NULL) {
670 cx_ptr->type = ACPI_STATE_C3;
671 cx_ptr->trans_lat = AcpiGbl_FADT.C3Latency;
679 * Parse a _CST package and set up its Cx states. Since the _CST object
680 * can change dynamically, our notify handler may call this function
681 * to clean up and probe the new _CST package.
684 acpi_cpu_cx_cst(struct acpi_cpu_softc *sc)
686 struct acpi_cx *cx_ptr;
694 ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
697 buf.Length = ACPI_ALLOCATE_BUFFER;
698 status = AcpiEvaluateObject(sc->cpu_handle, "_CST", NULL, &buf);
699 if (ACPI_FAILURE(status))
702 /* _CST is a package with a count and at least one Cx package. */
703 top = (ACPI_OBJECT *)buf.Pointer;
704 if (!ACPI_PKG_VALID(top, 2) || acpi_PkgInt32(top, 0, &count) != 0) {
705 device_printf(sc->cpu_dev, "invalid _CST package\n");
706 AcpiOsFree(buf.Pointer);
709 if (count != top->Package.Count - 1) {
710 device_printf(sc->cpu_dev, "invalid _CST state count (%d != %d)\n",
711 count, top->Package.Count - 1);
712 count = top->Package.Count - 1;
714 if (count > MAX_CX_STATES) {
715 device_printf(sc->cpu_dev, "_CST has too many states (%d)\n", count);
716 count = MAX_CX_STATES;
720 sc->cpu_cx_count = 0;
721 cx_ptr = sc->cpu_cx_states;
724 * C1 has been required since just after ACPI 1.0.
725 * Reserve the first slot for it.
727 cx_ptr->type = ACPI_STATE_C0;
731 /* Set up all valid states. */
732 for (i = 0; i < count; i++) {
733 pkg = &top->Package.Elements[i + 1];
734 if (!ACPI_PKG_VALID(pkg, 4) ||
735 acpi_PkgInt32(pkg, 1, &cx_ptr->type) != 0 ||
736 acpi_PkgInt32(pkg, 2, &cx_ptr->trans_lat) != 0 ||
737 acpi_PkgInt32(pkg, 3, &cx_ptr->power) != 0) {
739 device_printf(sc->cpu_dev, "skipping invalid Cx state package\n");
743 /* Validate the state to see if we should use it. */
744 switch (cx_ptr->type) {
746 if (sc->cpu_cx_states[0].type == ACPI_STATE_C0) {
747 /* This is the first C1 state. Use the reserved slot. */
748 sc->cpu_cx_states[0] = *cx_ptr;
760 if ((cpu_quirks & CPU_QUIRK_NO_C3) != 0) {
761 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
762 "acpi_cpu%d: C3[%d] not available.\n",
763 device_get_unit(sc->cpu_dev), i));
770 /* Free up any previous register. */
771 if (cx_ptr->p_lvlx != NULL) {
772 bus_release_resource(sc->cpu_dev, 0, 0, cx_ptr->p_lvlx);
773 cx_ptr->p_lvlx = NULL;
777 /* Allocate the control register for C2 or C3. */
778 acpi_PkgGas(sc->cpu_dev, pkg, 0, &cx_ptr->res_type, &sc->cpu_rid,
779 &cx_ptr->p_lvlx, RF_SHAREABLE);
780 if (cx_ptr->p_lvlx) {
782 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
783 "acpi_cpu%d: Got C%d - %d latency\n",
784 device_get_unit(sc->cpu_dev), cx_ptr->type,
790 AcpiOsFree(buf.Pointer);
792 /* If C1 state was not found, we need one now. */
793 cx_ptr = sc->cpu_cx_states;
794 if (cx_ptr->type == ACPI_STATE_C0) {
795 cx_ptr->type = ACPI_STATE_C1;
796 cx_ptr->trans_lat = 0;
803 * Call this *after* all CPUs have been attached.
806 acpi_cpu_startup(void *arg)
808 struct acpi_cpu_softc *sc;
811 /* Get set of CPU devices */
812 devclass_get_devices(acpi_cpu_devclass, &cpu_devices, &cpu_ndevices);
815 * Setup any quirks that might necessary now that we have probed
820 if (cpu_cx_generic) {
822 * We are using generic Cx mode, probe for available Cx states
823 * for all processors.
825 for (i = 0; i < cpu_ndevices; i++) {
826 sc = device_get_softc(cpu_devices[i]);
827 acpi_cpu_generic_cx_probe(sc);
831 * We are using _CST mode, remove C3 state if necessary.
832 * As we now know for sure that we will be using _CST mode
833 * install our notify handler.
835 for (i = 0; i < cpu_ndevices; i++) {
836 sc = device_get_softc(cpu_devices[i]);
837 if (cpu_quirks & CPU_QUIRK_NO_C3) {
838 sc->cpu_cx_count = sc->cpu_non_c3 + 1;
840 AcpiInstallNotifyHandler(sc->cpu_handle, ACPI_DEVICE_NOTIFY,
841 acpi_cpu_notify, sc);
845 /* Perform Cx final initialization. */
846 for (i = 0; i < cpu_ndevices; i++) {
847 sc = device_get_softc(cpu_devices[i]);
848 acpi_cpu_startup_cx(sc);
851 /* Add a sysctl handler to handle global Cx lowest setting */
852 SYSCTL_ADD_PROC(&cpu_sysctl_ctx, SYSCTL_CHILDREN(cpu_sysctl_tree),
853 OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
854 NULL, 0, acpi_cpu_global_cx_lowest_sysctl, "A",
855 "Global lowest Cx sleep state to use");
857 /* Take over idling from cpu_idle_default(). */
858 cpu_cx_lowest_lim = 0;
859 cpu_disable_idle = FALSE;
860 cpu_idle_hook = acpi_cpu_idle;
864 acpi_cpu_cx_list(struct acpi_cpu_softc *sc)
870 * Set up the list of Cx states
873 sbuf_new(&sb, sc->cpu_cx_supported, sizeof(sc->cpu_cx_supported),
875 for (i = 0; i < sc->cpu_cx_count; i++) {
876 sbuf_printf(&sb, "C%d/%d ", i + 1, sc->cpu_cx_states[i].trans_lat);
877 if (sc->cpu_cx_states[i].type < ACPI_STATE_C3)
880 cpu_can_deep_sleep = 1;
887 acpi_cpu_startup_cx(struct acpi_cpu_softc *sc)
889 acpi_cpu_cx_list(sc);
891 SYSCTL_ADD_STRING(&sc->cpu_sysctl_ctx,
892 SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
893 OID_AUTO, "cx_supported", CTLFLAG_RD,
894 sc->cpu_cx_supported, 0,
895 "Cx/microsecond values for supported Cx states");
896 SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
897 SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
898 OID_AUTO, "cx_lowest", CTLTYPE_STRING | CTLFLAG_RW,
899 (void *)sc, 0, acpi_cpu_cx_lowest_sysctl, "A",
900 "lowest Cx sleep state to use");
901 SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
902 SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
903 OID_AUTO, "cx_usage", CTLTYPE_STRING | CTLFLAG_RD,
904 (void *)sc, 0, acpi_cpu_usage_sysctl, "A",
905 "percent usage for each Cx state");
908 /* Signal platform that we can handle _CST notification. */
909 if (!cpu_cx_generic && cpu_cst_cnt != 0) {
911 AcpiOsWritePort(cpu_smi_cmd, cpu_cst_cnt, 8);
918 * Idle the CPU in the lowest state possible. This function is called with
919 * interrupts disabled. Note that once it re-enables interrupts, a task
920 * switch can occur so do not access shared data (i.e. the softc) after
921 * interrupts are re-enabled.
926 struct acpi_cpu_softc *sc;
927 struct acpi_cx *cx_next;
928 uint32_t start_time, end_time;
929 int bm_active, cx_next_idx, i;
931 /* If disabled, return immediately. */
932 if (cpu_disable_idle) {
938 * Look up our CPU id to get our softc. If it's NULL, we'll use C1
939 * since there is no ACPI processor object for this CPU. This occurs
940 * for logical CPUs in the HTT case.
942 sc = cpu_softc[PCPU_GET(cpuid)];
948 /* Find the lowest state that has small enough latency. */
950 if (cpu_disable_deep_sleep)
951 i = min(sc->cpu_cx_lowest, sc->cpu_non_c3);
953 i = sc->cpu_cx_lowest;
954 for (; i >= 0; i--) {
955 if (sc->cpu_cx_states[i].trans_lat * 3 <= sc->cpu_prev_sleep) {
962 * Check for bus master activity. If there was activity, clear
963 * the bit and use the lowest non-C3 state. Note that the USB
964 * driver polling for new devices keeps this bit set all the
965 * time if USB is loaded.
967 if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
968 AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, &bm_active);
969 if (bm_active != 0) {
970 AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_STATUS, 1);
971 cx_next_idx = min(cx_next_idx, sc->cpu_non_c3);
975 /* Select the next state and update statistics. */
976 cx_next = &sc->cpu_cx_states[cx_next_idx];
977 sc->cpu_cx_stats[cx_next_idx]++;
978 KASSERT(cx_next->type != ACPI_STATE_C0, ("acpi_cpu_idle: C0 sleep"));
981 * Execute HLT (or equivalent) and wait for an interrupt. We can't
982 * precisely calculate the time spent in C1 since the place we wake up
983 * is an ISR. Assume we slept no more then half of quantum, unless
984 * we are called inside critical section, delaying context switch.
986 if (cx_next->type == ACPI_STATE_C1) {
987 AcpiHwRead(&start_time, &AcpiGbl_FADT.XPmTimerBlock);
989 AcpiHwRead(&end_time, &AcpiGbl_FADT.XPmTimerBlock);
990 end_time = PM_USEC(acpi_TimerDelta(end_time, start_time));
991 if (curthread->td_critnest == 0)
992 end_time = min(end_time, 500000 / hz);
993 sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + end_time) / 4;
998 * For C3, disable bus master arbitration and enable bus master wake
999 * if BM control is available, otherwise flush the CPU cache.
1001 if (cx_next->type == ACPI_STATE_C3) {
1002 if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
1003 AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 1);
1004 AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
1006 ACPI_FLUSH_CPU_CACHE();
1010 * Read from P_LVLx to enter C2(+), checking time spent asleep.
1011 * Use the ACPI timer for measuring sleep time. Since we need to
1012 * get the time very close to the CPU start/stop clock logic, this
1013 * is the only reliable time source.
1015 AcpiHwRead(&start_time, &AcpiGbl_FADT.XPmTimerBlock);
1016 CPU_GET_REG(cx_next->p_lvlx, 1);
1019 * Read the end time twice. Since it may take an arbitrary time
1020 * to enter the idle state, the first read may be executed before
1021 * the processor has stopped. Doing it again provides enough
1022 * margin that we are certain to have a correct value.
1024 AcpiHwRead(&end_time, &AcpiGbl_FADT.XPmTimerBlock);
1025 AcpiHwRead(&end_time, &AcpiGbl_FADT.XPmTimerBlock);
1027 /* Enable bus master arbitration and disable bus master wakeup. */
1028 if (cx_next->type == ACPI_STATE_C3 &&
1029 (cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
1030 AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 0);
1031 AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1035 /* Find the actual time asleep in microseconds. */
1036 end_time = acpi_TimerDelta(end_time, start_time);
1037 sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + PM_USEC(end_time)) / 4;
1041 * Re-evaluate the _CST object when we are notified that it changed.
1043 * XXX Re-evaluation disabled until locking is done.
1046 acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context)
1048 struct acpi_cpu_softc *sc = (struct acpi_cpu_softc *)context;
1050 if (notify != ACPI_NOTIFY_CX_STATES)
1053 /* Update the list of Cx states. */
1054 acpi_cpu_cx_cst(sc);
1055 acpi_cpu_cx_list(sc);
1057 ACPI_SERIAL_BEGIN(cpu);
1058 acpi_cpu_set_cx_lowest(sc);
1059 ACPI_SERIAL_END(cpu);
1063 acpi_cpu_quirks(void)
1068 ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
1071 * Bus mastering arbitration control is needed to keep caches coherent
1072 * while sleeping in C3. If it's not present but a working flush cache
1073 * instruction is present, flush the caches before entering C3 instead.
1074 * Otherwise, just disable C3 completely.
1076 if (AcpiGbl_FADT.Pm2ControlBlock == 0 ||
1077 AcpiGbl_FADT.Pm2ControlLength == 0) {
1078 if ((AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD) &&
1079 (AcpiGbl_FADT.Flags & ACPI_FADT_WBINVD_FLUSH) == 0) {
1080 cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1081 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1082 "acpi_cpu: no BM control, using flush cache method\n"));
1084 cpu_quirks |= CPU_QUIRK_NO_C3;
1085 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1086 "acpi_cpu: no BM control, C3 not available\n"));
1091 * If we are using generic Cx mode, C3 on multiple CPUs requires using
1092 * the expensive flush cache instruction.
1094 if (cpu_cx_generic && mp_ncpus > 1) {
1095 cpu_quirks |= CPU_QUIRK_NO_BM_CTRL;
1096 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1097 "acpi_cpu: SMP, using flush cache mode for C3\n"));
1100 /* Look for various quirks of the PIIX4 part. */
1101 acpi_dev = pci_find_device(PCI_VENDOR_INTEL, PCI_DEVICE_82371AB_3);
1102 if (acpi_dev != NULL) {
1103 switch (pci_get_revid(acpi_dev)) {
1105 * Disable C3 support for all PIIX4 chipsets. Some of these parts
1106 * do not report the BMIDE status to the BM status register and
1107 * others have a livelock bug if Type-F DMA is enabled. Linux
1108 * works around the BMIDE bug by reading the BM status directly
1109 * but we take the simpler approach of disabling C3 for these
1112 * See erratum #18 ("C3 Power State/BMIDE and Type-F DMA
1113 * Livelock") from the January 2002 PIIX4 specification update.
1114 * Applies to all PIIX4 models.
1116 * Also, make sure that all interrupts cause a "Stop Break"
1117 * event to exit from C2 state.
1118 * Also, BRLD_EN_BM (ACPI_BITREG_BUS_MASTER_RLD in ACPI-speak)
1119 * should be set to zero, otherwise it causes C2 to short-sleep.
1120 * PIIX4 doesn't properly support C3 and bus master activity
1121 * need not break out of C2.
1123 case PCI_REVISION_A_STEP:
1124 case PCI_REVISION_B_STEP:
1125 case PCI_REVISION_4E:
1126 case PCI_REVISION_4M:
1127 cpu_quirks |= CPU_QUIRK_NO_C3;
1128 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1129 "acpi_cpu: working around PIIX4 bug, disabling C3\n"));
1131 val = pci_read_config(acpi_dev, PIIX4_DEVACTB_REG, 4);
1132 if ((val & PIIX4_STOP_BREAK_MASK) != PIIX4_STOP_BREAK_MASK) {
1133 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1134 "acpi_cpu: PIIX4: enabling IRQs to generate Stop Break\n"));
1135 val |= PIIX4_STOP_BREAK_MASK;
1136 pci_write_config(acpi_dev, PIIX4_DEVACTB_REG, val, 4);
1138 AcpiReadBitRegister(ACPI_BITREG_BUS_MASTER_RLD, &val);
1140 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1141 "acpi_cpu: PIIX4: reset BRLD_EN_BM\n"));
1142 AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
1154 acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS)
1156 struct acpi_cpu_softc *sc;
1160 uintmax_t fract, sum, whole;
1162 sc = (struct acpi_cpu_softc *) arg1;
1164 for (i = 0; i < sc->cpu_cx_count; i++)
1165 sum += sc->cpu_cx_stats[i];
1166 sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
1167 for (i = 0; i < sc->cpu_cx_count; i++) {
1169 whole = (uintmax_t)sc->cpu_cx_stats[i] * 100;
1170 fract = (whole % sum) * 100;
1171 sbuf_printf(&sb, "%u.%02u%% ", (u_int)(whole / sum),
1172 (u_int)(fract / sum));
1174 sbuf_printf(&sb, "0.00%% ");
1176 sbuf_printf(&sb, "last %dus", sc->cpu_prev_sleep);
1179 sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
1186 acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc)
1190 ACPI_SERIAL_ASSERT(cpu);
1191 sc->cpu_cx_lowest = min(sc->cpu_cx_lowest_lim, sc->cpu_cx_count - 1);
1193 /* If not disabling, cache the new lowest non-C3 state. */
1195 for (i = sc->cpu_cx_lowest; i >= 0; i--) {
1196 if (sc->cpu_cx_states[i].type < ACPI_STATE_C3) {
1202 /* Reset the statistics counters. */
1203 bzero(sc->cpu_cx_stats, sizeof(sc->cpu_cx_stats));
1208 acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1210 struct acpi_cpu_softc *sc;
1214 sc = (struct acpi_cpu_softc *) arg1;
1215 snprintf(state, sizeof(state), "C%d", sc->cpu_cx_lowest_lim + 1);
1216 error = sysctl_handle_string(oidp, state, sizeof(state), req);
1217 if (error != 0 || req->newptr == NULL)
1219 if (strlen(state) < 2 || toupper(state[0]) != 'C')
1221 if (strcasecmp(state, "Cmax") == 0)
1222 val = MAX_CX_STATES;
1224 val = (int) strtol(state + 1, NULL, 10);
1225 if (val < 1 || val > MAX_CX_STATES)
1229 ACPI_SERIAL_BEGIN(cpu);
1230 sc->cpu_cx_lowest_lim = val - 1;
1231 acpi_cpu_set_cx_lowest(sc);
1232 ACPI_SERIAL_END(cpu);
1238 acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS)
1240 struct acpi_cpu_softc *sc;
1244 snprintf(state, sizeof(state), "C%d", cpu_cx_lowest_lim + 1);
1245 error = sysctl_handle_string(oidp, state, sizeof(state), req);
1246 if (error != 0 || req->newptr == NULL)
1248 if (strlen(state) < 2 || toupper(state[0]) != 'C')
1250 if (strcasecmp(state, "Cmax") == 0)
1251 val = MAX_CX_STATES;
1253 val = (int) strtol(state + 1, NULL, 10);
1254 if (val < 1 || val > MAX_CX_STATES)
1258 /* Update the new lowest useable Cx state for all CPUs. */
1259 ACPI_SERIAL_BEGIN(cpu);
1260 cpu_cx_lowest_lim = val - 1;
1261 for (i = 0; i < cpu_ndevices; i++) {
1262 sc = device_get_softc(cpu_devices[i]);
1263 sc->cpu_cx_lowest_lim = cpu_cx_lowest_lim;
1264 acpi_cpu_set_cx_lowest(sc);
1266 ACPI_SERIAL_END(cpu);