d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

If x86 CPU implementation of the MWAIT instruction reasonably

Browse Source 
interacts with interrupts, query ACPI and use MWAIT for entrance into
Cx sleep states.  Support C1 "I/O then halt" mode.  See Intel'
document 302223-007 "Intelб╝ Processor Vendor-Specific ACPI Interface
Specification" for description.

Move the acpi_cpu_c1() function into x86/cpu_machdep.c and use
it instead of inlining "sti; hlt" sequence in several places.

In the acpi(4) man page, besides documenting the dev.cpu.N.cx_methods
sysctl, correct the names for dev.cpu.N.{cx_usage,cx_lowest,cx_supported}
sysctls.

Both jkim and avg have some other patches implementing the mwait
functionality; this work is unrelated.  Linux does not rely on the
ACPI to provide correct tables describing Cx modes.  Instead, the
driver has pre-defined knowledge of the CPU models, it was supplied by
Intel.

Tested by:    pho (previous versions)
Sponsored by:	The FreeBSD Foundation
This commit is contained in:
kib 2015-05-09 12:28:48 +00:00
parent 4f7fc385ca
commit 6006bf3a7d
10 changed files with 234 additions and 41 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
share/man/man4/acpi.4
View File

@ -25,7 +25,7 @@
.\"
.\" $FreeBSD$
.\"
.Dd June 23, 2014
.Dd May 9, 2015
.Dt ACPI 4
.Os
.Sh NAME
@ -69,14 +69,12 @@ them (such as
Enable dumping Debug objects without
.Cd "options ACPI_DEBUG" .
Default is 0, ignore Debug objects.
.It Va hw.acpi.acline
AC line state (1 means online, 0 means on battery power).
.It Va hw.acpi.cpu.cx_usage
.It Va dev.cpu.N.cx_usage
Debugging information listing the percent of total usage for each sleep state.
The values are reset when
.Va hw.acpi.cpu.cx_lowest
.Va dev.cpu.N.cx_lowest
is modified.
.It Va hw.acpi.cpu.cx_lowest
.It Va dev.cpu.N.cx_lowest
Lowest Cx state to use for idling the CPU.
A scheduling algorithm will select states between
.Li C1
@ -111,6 +109,11 @@ semantics as the
state.
Deeper sleeps provide more power savings but increased transition
latency when an interrupt occurs.
.It Va dev.cpu.N.cx_method
List of supported CPU idle states and their transition methods, as
directed by the firmware.
.It Va hw.acpi.acline
AC line state (1 means online, 0 means on battery power).
.It Va hw.acpi.disable_on_reboot
Disable ACPI during the reboot process.
Most systems reboot fine with ACPI still enabled, but some require
@ -374,6 +377,14 @@ typically as a child of a PCI bus.
.Pq Vt device
Supports an ACPI laptop lid switch, which typically puts a
system to sleep.
.It Li mwait
.Pq Vt feature
Do not ask firmware for available x86-vendor specific methods to enter
.Li Cx
sleep states.
Only query and use the generic I/O-based entrance method.
The knob is provided to work around inconsistencies in the tables
filled by firmware.
.It Li quirks
.Pq Vt feature
Do not honor quirks.

7
sys/amd64/acpica/acpi_machdep.c
View File

@ -87,13 +87,6 @@ acpi_machdep_quirks(int *quirks)
return (0);
}
void
acpi_cpu_c1()
{
__asm __volatile("sti; hlt");
}
/*
* Support for mapping ACPI tables during early boot. Currently this
* uses the crashdump map to map each table. However, the crashdump

1
sys/amd64/include/md_var.h
View File

@ -91,6 +91,7 @@ struct dumperinfo;
void *alloc_fpusave(int flags);
void amd64_syscall(struct thread *td, int traced);
void busdma_swi(void);
bool cpu_mwait_usable(void);
void cpu_probe_amdc1e(void);
void cpu_setregs(void);
void doreti_iret(void) __asm(__STRING(doreti_iret));

176
sys/dev/acpica/acpi_cpu.c
View File

@ -47,6 +47,8 @@ __FBSDID("$FreeBSD$");
#include <machine/bus.h>
#if defined(__amd64__) || defined(__i386__)
#include <machine/clock.h>
#include <machine/specialreg.h>
#include <machine/md_var.h>
#endif
#include <sys/rman.h>
@ -70,6 +72,10 @@ struct acpi_cx {
uint32_t power; /* Power consumed (mW). */
int res_type; /* Resource type for p_lvlx. */
int res_rid; /* Resource ID for p_lvlx. */
bool do_mwait;
uint32_t mwait_hint;
bool mwait_hw_coord;
bool mwait_bm_avoidance;
};
#define MAX_CX_STATES 8
@ -128,6 +134,12 @@ struct acpi_cpu_device {
#define PIIX4_STOP_BREAK_MASK (PIIX4_BRLD_EN_IRQ0 | PIIX4_BRLD_EN_IRQ | PIIX4_BRLD_EN_IRQ8)
#define PIIX4_PCNTRL_BST_EN (1<<10)
#define CST_FFH_VENDOR_INTEL 1
#define CST_FFH_INTEL_CL_C1IO 1
#define CST_FFH_INTEL_CL_MWAIT 2
#define CST_FFH_MWAIT_HW_COORD 0x0001
#define CST_FFH_MWAIT_BM_AVOID 0x0002
/* Allow users to ignore processor orders in MADT. */
static int cpu_unordered;
SYSCTL_INT(_debug_acpi, OID_AUTO, cpu_unordered, CTLFLAG_RDTUN,
@ -179,6 +191,9 @@ static int acpi_cpu_usage_counters_sysctl(SYSCTL_HANDLER_ARGS);
static int acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc);
static int acpi_cpu_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
static int acpi_cpu_global_cx_lowest_sysctl(SYSCTL_HANDLER_ARGS);
#if defined(__i386__) || defined(__amd64__)
static int acpi_cpu_method_sysctl(SYSCTL_HANDLER_ARGS);
#endif
static device_method_t acpi_cpu_methods[] = {
/* Device interface */
@ -348,7 +363,18 @@ acpi_cpu_attach(device_t dev)
* so advertise this ourselves. Note this is not the same as independent
* SMP control where each CPU can have different settings.
*/
sc->cpu_features = ACPI_CAP_SMP_SAME | ACPI_CAP_SMP_SAME_C3;
sc->cpu_features = ACPI_CAP_SMP_SAME | ACPI_CAP_SMP_SAME_C3 |
ACPI_CAP_C1_IO_HALT;
#if defined(__i386__) || defined(__amd64__)
/*
* Ask for MWAIT modes if not disabled and interrupts work
* reasonable with MWAIT.
*/
if (!acpi_disabled("mwait") && cpu_mwait_usable())
sc->cpu_features |= ACPI_CAP_SMP_C1_NATIVE | ACPI_CAP_SMP_C3_NATIVE;
#endif
if (devclass_get_drivers(acpi_cpu_devclass, &drivers, &drv_count) == 0) {
for (i = 0; i < drv_count; i++) {
if (ACPI_GET_FEATURES(drivers[i], &features) == 0)
@ -720,6 +746,27 @@ acpi_cpu_generic_cx_probe(struct acpi_cpu_softc *sc)
}
}
static void
acpi_cpu_cx_cst_mwait(struct acpi_cx *cx_ptr, uint64_t address, int accsize)
{
cx_ptr->do_mwait = true;
cx_ptr->mwait_hint = address & 0xffffffff;
cx_ptr->mwait_hw_coord = (accsize & CST_FFH_MWAIT_HW_COORD) != 0;
cx_ptr->mwait_bm_avoidance = (accsize & CST_FFH_MWAIT_BM_AVOID) != 0;
}
static void
acpi_cpu_cx_cst_free_plvlx(device_t cpu_dev, struct acpi_cx *cx_ptr)
{
if (cx_ptr->p_lvlx == NULL)
return;
bus_release_resource(cpu_dev, cx_ptr->res_type, cx_ptr->res_rid,
cx_ptr->p_lvlx);
cx_ptr->p_lvlx = NULL;
}
/*
* Parse a _CST package and set up its Cx states. Since the _CST object
* can change dynamically, our notify handler may call this function
@ -734,7 +781,8 @@ acpi_cpu_cx_cst(struct acpi_cpu_softc *sc)
ACPI_OBJECT *top;
ACPI_OBJECT *pkg;
uint32_t count;
int i;
uint64_t address;
int i, vendor, class, accsize;
ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
@ -790,6 +838,30 @@ acpi_cpu_cx_cst(struct acpi_cpu_softc *sc)
/* Validate the state to see if we should use it. */
switch (cx_ptr->type) {
case ACPI_STATE_C1:
acpi_cpu_cx_cst_free_plvlx(sc->cpu_dev, cx_ptr);
#if defined(__i386__) || defined(__amd64__)
if (acpi_PkgFFH_IntelCpu(pkg, 0, &vendor, &class, &address,
&accsize) == 0 && vendor == CST_FFH_VENDOR_INTEL) {
if (class == CST_FFH_INTEL_CL_C1IO) {
/* C1 I/O then Halt */
cx_ptr->res_rid = sc->cpu_cx_count;
bus_set_resource(sc->cpu_dev, SYS_RES_IOPORT,
cx_ptr->res_rid, address, 1);
cx_ptr->p_lvlx = bus_alloc_resource_any(sc->cpu_dev,
SYS_RES_IOPORT, &cx_ptr->res_rid, RF_ACTIVE |
RF_SHAREABLE);
if (cx_ptr->p_lvlx == NULL) {
bus_delete_resource(sc->cpu_dev, SYS_RES_IOPORT,
cx_ptr->res_rid);
device_printf(sc->cpu_dev,
"C1 I/O failed to allocate port %d, "
"degrading to C1 Halt", (int)address);
}
} else if (class == CST_FFH_INTEL_CL_MWAIT) {
acpi_cpu_cx_cst_mwait(cx_ptr, address, accsize);
}
}
#endif
if (sc->cpu_cx_states[0].type == ACPI_STATE_C0) {
/* This is the first C1 state. Use the reserved slot. */
sc->cpu_cx_states[0] = *cx_ptr;
@ -818,23 +890,34 @@ acpi_cpu_cx_cst(struct acpi_cpu_softc *sc)
}
/* Free up any previous register. */
if (cx_ptr->p_lvlx != NULL) {
bus_release_resource(sc->cpu_dev, cx_ptr->res_type, cx_ptr->res_rid,
cx_ptr->p_lvlx);
cx_ptr->p_lvlx = NULL;
}
acpi_cpu_cx_cst_free_plvlx(sc->cpu_dev, cx_ptr);
/* Allocate the control register for C2 or C3. */
cx_ptr->res_rid = sc->cpu_cx_count;
acpi_PkgGas(sc->cpu_dev, pkg, 0, &cx_ptr->res_type, &cx_ptr->res_rid,
&cx_ptr->p_lvlx, RF_SHAREABLE);
if (cx_ptr->p_lvlx) {
#if defined(__i386__) || defined(__amd64__)
if (acpi_PkgFFH_IntelCpu(pkg, 0, &vendor, &class, &address,
&accsize) == 0 && vendor == CST_FFH_VENDOR_INTEL &&
class == CST_FFH_INTEL_CL_MWAIT) {
/* Native C State Instruction use (mwait) */
acpi_cpu_cx_cst_mwait(cx_ptr, address, accsize);
ACPI_DEBUG_PRINT((ACPI_DB_INFO,
"acpi_cpu%d: Got C%d - %d latency\n",
device_get_unit(sc->cpu_dev), cx_ptr->type,
cx_ptr->trans_lat));
"acpi_cpu%d: Got C%d/mwait - %d latency\n",
device_get_unit(sc->cpu_dev), cx_ptr->type, cx_ptr->trans_lat));
cx_ptr++;
sc->cpu_cx_count++;
} else
#endif
{
cx_ptr->res_rid = sc->cpu_cx_count;
acpi_PkgGas(sc->cpu_dev, pkg, 0, &cx_ptr->res_type,
&cx_ptr->res_rid, &cx_ptr->p_lvlx, RF_SHAREABLE);
if (cx_ptr->p_lvlx) {
ACPI_DEBUG_PRINT((ACPI_DB_INFO,
"acpi_cpu%d: Got C%d - %d latency\n",
device_get_unit(sc->cpu_dev), cx_ptr->type,
cx_ptr->trans_lat));
cx_ptr++;
sc->cpu_cx_count++;
}
}
}
AcpiOsFree(buf.Pointer);
@ -956,6 +1039,13 @@ acpi_cpu_startup_cx(struct acpi_cpu_softc *sc)
OID_AUTO, "cx_usage_counters", CTLTYPE_STRING | CTLFLAG_RD,
(void *)sc, 0, acpi_cpu_usage_counters_sysctl, "A",
"Cx sleep state counters");
#if defined(__i386__) || defined(__amd64__)
SYSCTL_ADD_PROC(&sc->cpu_sysctl_ctx,
SYSCTL_CHILDREN(device_get_sysctl_tree(sc->cpu_dev)),
OID_AUTO, "cx_method", CTLTYPE_STRING | CTLFLAG_RD,
(void *)sc, 0, acpi_cpu_method_sysctl, "A",
"Cx entrance methods");
#endif
/* Signal platform that we can handle _CST notification. */
if (!cpu_cx_generic && cpu_cst_cnt != 0) {
@ -1043,7 +1133,14 @@ acpi_cpu_idle(sbintime_t sbt)
*/
if (cx_next->type == ACPI_STATE_C1) {
cputicks = cpu_ticks();
acpi_cpu_c1();
if (cx_next->p_lvlx != NULL) {
/* C1 I/O then Halt */
CPU_GET_REG(cx_next->p_lvlx, 1);
}
if (cx_next->do_mwait)
acpi_cpu_idle_mwait(cx_next->mwait_hint);
else
acpi_cpu_c1();
end_time = ((cpu_ticks() - cputicks) << 20) / cpu_tickrate();
if (curthread->td_critnest == 0)
end_time = min(end_time, 500000 / hz);
@ -1055,7 +1152,7 @@ acpi_cpu_idle(sbintime_t sbt)
* For C3, disable bus master arbitration and enable bus master wake
* if BM control is available, otherwise flush the CPU cache.
*/
if (cx_next->type == ACPI_STATE_C3) {
if (cx_next->type == ACPI_STATE_C3 || cx_next->mwait_bm_avoidance) {
if ((cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 1);
AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 1);
@ -1076,7 +1173,10 @@ acpi_cpu_idle(sbintime_t sbt)
start_time = 0;
cputicks = cpu_ticks();
}
CPU_GET_REG(cx_next->p_lvlx, 1);
if (cx_next->do_mwait)
acpi_cpu_idle_mwait(cx_next->mwait_hint);
else
CPU_GET_REG(cx_next->p_lvlx, 1);
/*
* Read the end time twice. Since it may take an arbitrary time
@ -1092,8 +1192,8 @@ acpi_cpu_idle(sbintime_t sbt)
end_time = ((cpu_ticks() - cputicks) << 20) / cpu_tickrate();
/* Enable bus master arbitration and disable bus master wakeup. */
if (cx_next->type == ACPI_STATE_C3 &&
(cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
if ((cx_next->type == ACPI_STATE_C3 || cx_next->mwait_bm_avoidance) &&
(cpu_quirks & CPU_QUIRK_NO_BM_CTRL) == 0) {
AcpiWriteBitRegister(ACPI_BITREG_ARB_DISABLE, 0);
AcpiWriteBitRegister(ACPI_BITREG_BUS_MASTER_RLD, 0);
}
@ -1286,6 +1386,44 @@ acpi_cpu_usage_counters_sysctl(SYSCTL_HANDLER_ARGS)
return (0);
}
#if defined(__i386__) || defined(__amd64__)
static int
acpi_cpu_method_sysctl(SYSCTL_HANDLER_ARGS)
{
struct acpi_cpu_softc *sc;
struct acpi_cx *cx;
struct sbuf sb;
char buf[128];
int i;
sc = (struct acpi_cpu_softc *)arg1;
sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
for (i = 0; i < sc->cpu_cx_count; i++) {
cx = &sc->cpu_cx_states[i];
sbuf_printf(&sb, "C%d/", i + 1);
if (cx->do_mwait) {
sbuf_cat(&sb, "mwait");
if (cx->mwait_hw_coord)
sbuf_cat(&sb, "/hwc");
if (cx->mwait_bm_avoidance)
sbuf_cat(&sb, "/bma");
} else if (cx->type == ACPI_STATE_C1) {
sbuf_cat(&sb, "hlt");
} else {
sbuf_cat(&sb, "io");
}
if (cx->type == ACPI_STATE_C1 && cx->p_lvlx != NULL)
sbuf_cat(&sb, "/iohlt");
sbuf_putc(&sb, ' ');
}
sbuf_trim(&sb);
sbuf_finish(&sb);
sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
sbuf_delete(&sb);
return (0);
}
#endif
static int
acpi_cpu_set_cx_lowest(struct acpi_cpu_softc *sc)
{

22
sys/dev/acpica/acpi_package.c
View File

@ -120,6 +120,28 @@ acpi_PkgGas(device_t dev, ACPI_OBJECT *res, int idx, int *type, int *rid,
return (acpi_bus_alloc_gas(dev, type, rid, &gas, dst, flags));
}
int
acpi_PkgFFH_IntelCpu(ACPI_OBJECT *res, int idx, int *vendor, int *class,
uint64_t *address, int *accsize)
{
ACPI_GENERIC_ADDRESS gas;
ACPI_OBJECT *obj;
obj = &res->Package.Elements[idx];
if (obj == NULL || obj->Type != ACPI_TYPE_BUFFER ||
obj->Buffer.Length < sizeof(ACPI_GENERIC_ADDRESS) + 3)
return (EINVAL);
memcpy(&gas, obj->Buffer.Pointer + 3, sizeof(gas));
if (gas.SpaceId != ACPI_ADR_SPACE_FIXED_HARDWARE)
return (ERESTART);
*vendor = gas.BitWidth;
*class = gas.BitOffset;
*address = gas.Address;
*accsize = gas.AccessWidth;
return (0);
}
ACPI_HANDLE
acpi_GetReference(ACPI_HANDLE scope, ACPI_OBJECT *obj)
{

2
sys/dev/acpica/acpivar.h
View File

@ -467,6 +467,8 @@ int acpi_PkgInt32(ACPI_OBJECT *res, int idx, uint32_t *dst);
int acpi_PkgStr(ACPI_OBJECT *res, int idx, void *dst, size_t size);
int acpi_PkgGas(device_t dev, ACPI_OBJECT *res, int idx, int *type,
int *rid, struct resource **dst, u_int flags);
int acpi_PkgFFH_IntelCpu(ACPI_OBJECT *res, int idx, int *vendor,
int *class, uint64_t *address, int *accsize);
ACPI_HANDLE acpi_GetReference(ACPI_HANDLE scope, ACPI_OBJECT *obj);
/*

7
sys/i386/acpica/acpi_machdep.c
View File

@ -106,13 +106,6 @@ acpi_machdep_quirks(int *quirks)
return (0);
}
void
acpi_cpu_c1()
{
__asm __volatile("sti; hlt");
}
/*
* Support for mapping ACPI tables during early boot. This abuses the
* crashdump map because the kernel cannot allocate KVA in

1
sys/i386/include/md_var.h
View File

@ -97,6 +97,7 @@ struct dumperinfo;
void *alloc_fpusave(int flags);
void bcopyb(const void *from, void *to, size_t len);
void busdma_swi(void);
bool cpu_mwait_usable(void);
void cpu_probe_amdc1e(void);
void cpu_setregs(void);
void cpu_switch_load_gs(void) __asm(__STRING(cpu_switch_load_gs));

1
sys/x86/include/acpica_machdep.h
View File

@ -74,6 +74,7 @@ enum intr_polarity;
void acpi_SetDefaultIntrModel(int model);
void acpi_cpu_c1(void);
void acpi_cpu_idle_mwait(uint32_t mwait_hint);
void *acpi_map_table(vm_paddr_t pa, const char *sig);
void acpi_unmap_table(void *table);
vm_paddr_t acpi_find_table(const char *sig);

35
sys/x86/x86/cpu_machdep.c
View File

@ -90,6 +90,7 @@ __FBSDID("$FreeBSD$");
#ifdef SMP
#include <machine/smp.h>
#endif
#include <x86/acpica_machdep.h>
#include <vm/vm.h>
#include <vm/vm_extern.h>
@ -121,6 +122,27 @@ cpu_flush_dcache(void *ptr, size_t len)
/* Not applicable */
}
void
acpi_cpu_c1(void)
{
__asm __volatile("sti; hlt");
}
void
acpi_cpu_idle_mwait(uint32_t mwait_hint)
{
int *state;
state = (int *)PCPU_PTR(monitorbuf);
/*
* XXXKIB. Software coordination mode should be supported,
* but all Intel CPUs provide hardware coordination.
*/
cpu_monitor(state, 0, 0);
cpu_mwait(MWAIT_INTRBREAK, mwait_hint);
}
/* Get current clock frequency for the given cpu id. */
int
cpu_est_clockrate(int cpu_id, uint64_t *rate)
@ -194,6 +216,15 @@ cpu_halt(void)
halt();
}
bool
cpu_mwait_usable(void)
{
return ((cpu_feature2 & CPUID2_MON) != 0 && ((cpu_mon_mwait_flags &
(CPUID5_MON_MWAIT_EXT | CPUID5_MWAIT_INTRBREAK)) ==
(CPUID5_MON_MWAIT_EXT | CPUID5_MWAIT_INTRBREAK)));
}
void (*cpu_idle_hook)(sbintime_t) = NULL; /* ACPI idle hook. */
static int cpu_ident_amdc1e = 0; /* AMD C1E supported. */
static int idle_mwait = 1; /* Use MONITOR/MWAIT for short idle. */
@ -220,7 +251,7 @@ cpu_idle_acpi(sbintime_t sbt)
else if (cpu_idle_hook)
cpu_idle_hook(sbt);
else
__asm __volatile("sti; hlt");
acpi_cpu_c1();
*state = STATE_RUNNING;
}
#endif /* !PC98 */
@ -253,7 +284,7 @@ cpu_idle_hlt(sbintime_t sbt)
if (sched_runnable())
enable_intr();
else
__asm __volatile("sti; hlt");
acpi_cpu_c1();
*state = STATE_RUNNING;
}