freebsd-nq/sys/dev/acpica/acpi_perf.c

596 lines
16 KiB
C
Raw Normal View History

/*-
* Copyright (c) 2003-2005 Nate Lawson (SDG)
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_acpi.h"
#include <sys/param.h>
#include <sys/kernel.h>
#include <sys/proc.h>
#include <sys/sched.h>
#include <sys/bus.h>
#include <sys/cpu.h>
#include <sys/power.h>
#include <sys/malloc.h>
#include <sys/module.h>
#include <sys/sbuf.h>
#include <sys/pcpu.h>
#include <machine/bus.h>
#include <machine/resource.h>
#include <sys/rman.h>
2009-06-05 18:44:36 +00:00
#include <contrib/dev/acpica/include/acpi.h>
#include <dev/acpica/acpivar.h>
#include "cpufreq_if.h"
/*
* Support for ACPI processor performance states (Px) according to
* section 8.3.3 of the ACPI 2.0c specification.
*/
struct acpi_px {
uint32_t core_freq;
uint32_t power;
uint32_t trans_lat;
uint32_t bm_lat;
uint32_t ctrl_val;
uint32_t sts_val;
};
/* Offsets in struct cf_setting array for storing driver-specific values. */
#define PX_SPEC_CONTROL 0
#define PX_SPEC_STATUS 1
#define MAX_PX_STATES 16
struct acpi_perf_softc {
device_t dev;
ACPI_HANDLE handle;
struct resource *perf_ctrl; /* Set new performance state. */
int perf_ctrl_type; /* Resource type for perf_ctrl. */
struct resource *perf_status; /* Check that transition succeeded. */
int perf_sts_type; /* Resource type for perf_status. */
struct acpi_px *px_states; /* ACPI perf states. */
uint32_t px_count; /* Total number of perf states. */
uint32_t px_max_avail; /* Lowest index state available. */
int px_curr_state; /* Active state index. */
int px_rid;
int info_only; /* Can we set new states? */
};
#define PX_GET_REG(reg) \
(bus_space_read_4(rman_get_bustag((reg)), \
rman_get_bushandle((reg)), 0))
#define PX_SET_REG(reg, val) \
(bus_space_write_4(rman_get_bustag((reg)), \
rman_get_bushandle((reg)), 0, (val)))
#define ACPI_NOTIFY_PERF_STATES 0x80 /* _PSS changed. */
static void acpi_perf_identify(driver_t *driver, device_t parent);
static int acpi_perf_probe(device_t dev);
static int acpi_perf_attach(device_t dev);
static int acpi_perf_detach(device_t dev);
static int acpi_perf_evaluate(device_t dev);
static int acpi_px_to_set(device_t dev, struct acpi_px *px,
struct cf_setting *set);
static void acpi_px_available(struct acpi_perf_softc *sc);
static void acpi_px_startup(void *arg);
static void acpi_px_notify(ACPI_HANDLE h, UINT32 notify, void *context);
static int acpi_px_settings(device_t dev, struct cf_setting *sets,
int *count);
static int acpi_px_set(device_t dev, const struct cf_setting *set);
static int acpi_px_get(device_t dev, struct cf_setting *set);
static int acpi_px_type(device_t dev, int *type);
static device_method_t acpi_perf_methods[] = {
/* Device interface */
DEVMETHOD(device_identify, acpi_perf_identify),
DEVMETHOD(device_probe, acpi_perf_probe),
DEVMETHOD(device_attach, acpi_perf_attach),
DEVMETHOD(device_detach, acpi_perf_detach),
/* cpufreq interface */
DEVMETHOD(cpufreq_drv_set, acpi_px_set),
DEVMETHOD(cpufreq_drv_get, acpi_px_get),
DEVMETHOD(cpufreq_drv_type, acpi_px_type),
DEVMETHOD(cpufreq_drv_settings, acpi_px_settings),
DEVMETHOD_END
};
static driver_t acpi_perf_driver = {
"acpi_perf",
acpi_perf_methods,
sizeof(struct acpi_perf_softc),
};
static devclass_t acpi_perf_devclass;
DRIVER_MODULE(acpi_perf, cpu, acpi_perf_driver, acpi_perf_devclass, 0, 0);
MODULE_DEPEND(acpi_perf, acpi, 1, 1, 1);
static MALLOC_DEFINE(M_ACPIPERF, "acpi_perf", "ACPI Performance states");
static void
acpi_perf_identify(driver_t *driver, device_t parent)
{
ACPI_HANDLE handle;
device_t dev;
/* Make sure we're not being doubly invoked. */
if (device_find_child(parent, "acpi_perf", -1) != NULL)
return;
/* Get the handle for the Processor object and check for perf states. */
handle = acpi_get_handle(parent);
if (handle == NULL)
return;
if (ACPI_FAILURE(AcpiEvaluateObject(handle, "_PSS", NULL, NULL)))
return;
/*
* Add a child to every CPU that has the right methods. In future
* versions of the ACPI spec, CPUs can have different settings.
* We probe this child now so that other devices that depend
* on it (i.e., for info about supported states) will see it.
*/
if ((dev = BUS_ADD_CHILD(parent, 0, "acpi_perf", -1)) != NULL)
device_probe_and_attach(dev);
else
device_printf(parent, "add acpi_perf child failed\n");
}
static int
acpi_perf_probe(device_t dev)
{
ACPI_HANDLE handle;
ACPI_OBJECT *pkg;
struct resource *res;
ACPI_BUFFER buf;
int error, rid, type;
if (resource_disabled("acpi_perf", 0))
return (ENXIO);
/*
* Check the performance state registers. If they are of type
* "functional fixed hardware", we attach quietly since we will
* only be providing information on settings to other drivers.
*/
error = ENXIO;
handle = acpi_get_handle(dev);
buf.Pointer = NULL;
buf.Length = ACPI_ALLOCATE_BUFFER;
if (ACPI_FAILURE(AcpiEvaluateObject(handle, "_PCT", NULL, &buf)))
return (error);
pkg = (ACPI_OBJECT *)buf.Pointer;
if (ACPI_PKG_VALID(pkg, 2)) {
rid = 0;
Re-work Cx handling to be per-cpu and asymmetrical, fixing support on modern dual-core systems as well. - Parse the _CST packages for each cpu and track all the states individually, on a per-cpu basis. - Revert to generic FADT/P_BLK based Cx control if the _CST package is not present on all cpus. In that case, the new driver will still support per-cpu Cx state handling. The driver will determine the highest Cx level that can be supported by all the cpus and configure the available Cx state based on that. - Fixed the case where multiple cpus in the system share the same registers for Cx state handling. To do that, added a new flag parameter to the acpi_PkgGas and acpi_bus_alloc_gas functions that enable the caller to add the RF_SHAREABLE flag. This flag could also be useful to other callers (acpi_throttle?) in the tree but this change is not yet made. - For Core Duo cpus, both cores seems to be taken out of C3 state when any one of the cores need to transition out. This broke the short sleep detection logic. It is disabled now if there is more than one cpu in the system for now as it fixed it in my case. This quirk may need to be re-enabled later differently. - Added support to control cx_lowest on a per-cpu basis. There is still a generic cx_lowest to enable changing cx_lowest for all cpus with a single sysctl and for ease of use. Sample output for the new sysctl: dev.cpu.0.cx_supported: C1/1 C2/1 C3/57 dev.cpu.0.cx_lowest: C3 dev.cpu.0.cx_usage: 0.00% 43.16% 56.83% dev.cpu.1.cx_supported: C1/1 C2/1 C3/57 dev.cpu.1.cx_lowest: C3 dev.cpu.1.cx_usage: 0.00% 45.65% 54.34% hw.acpi.cpu.cx_lowest: C3 This work was done by Stephane E. Potvin with some simple reworking by myself. Thank you. Submitted by: Stephane E. Potvin <sepotvin / videotron.ca> MFC after: 2 weeks
2007-01-07 21:53:42 +00:00
error = acpi_PkgGas(dev, pkg, 0, &type, &rid, &res, 0);
switch (error) {
case 0:
bus_release_resource(dev, type, rid, res);
bus_delete_resource(dev, type, rid);
device_set_desc(dev, "ACPI CPU Frequency Control");
break;
case EOPNOTSUPP:
device_quiet(dev);
error = 0;
break;
}
}
AcpiOsFree(buf.Pointer);
return (error);
}
static int
acpi_perf_attach(device_t dev)
{
struct acpi_perf_softc *sc;
sc = device_get_softc(dev);
sc->dev = dev;
sc->handle = acpi_get_handle(dev);
sc->px_max_avail = 0;
sc->px_curr_state = CPUFREQ_VAL_UNKNOWN;
if (acpi_perf_evaluate(dev) != 0)
return (ENXIO);
2007-03-22 18:16:43 +00:00
AcpiOsExecute(OSL_NOTIFY_HANDLER, acpi_px_startup, NULL);
if (!sc->info_only)
cpufreq_register(dev);
return (0);
}
static int
acpi_perf_detach(device_t dev)
{
/* TODO: teardown registers, remove notify handler. */
return (ENXIO);
}
/* Probe and setup any valid performance states (Px). */
static int
acpi_perf_evaluate(device_t dev)
{
struct acpi_perf_softc *sc;
ACPI_BUFFER buf;
ACPI_OBJECT *pkg, *res;
ACPI_STATUS status;
int count, error, i, j;
static int once = 1;
uint32_t *p;
/* Get the control values and parameters for each state. */
error = ENXIO;
sc = device_get_softc(dev);
buf.Pointer = NULL;
buf.Length = ACPI_ALLOCATE_BUFFER;
status = AcpiEvaluateObject(sc->handle, "_PSS", NULL, &buf);
if (ACPI_FAILURE(status))
return (ENXIO);
pkg = (ACPI_OBJECT *)buf.Pointer;
if (!ACPI_PKG_VALID(pkg, 1)) {
device_printf(dev, "invalid top level _PSS package\n");
goto out;
}
sc->px_count = pkg->Package.Count;
sc->px_states = malloc(sc->px_count * sizeof(struct acpi_px),
M_ACPIPERF, M_WAITOK | M_ZERO);
/*
* Each state is a package of {CoreFreq, Power, TransitionLatency,
* BusMasterLatency, ControlVal, StatusVal}, sorted from highest
* performance to lowest.
*/
count = 0;
for (i = 0; i < sc->px_count; i++) {
res = &pkg->Package.Elements[i];
if (!ACPI_PKG_VALID(res, 6)) {
if (once) {
once = 0;
device_printf(dev, "invalid _PSS package\n");
}
continue;
}
/* Parse the rest of the package into the struct. */
p = &sc->px_states[count].core_freq;
for (j = 0; j < 6; j++, p++)
acpi_PkgInt32(res, j, p);
/*
* Check for some impossible frequencies that some systems
* use to indicate they don't actually support this Px state.
*/
if (sc->px_states[count].core_freq == 0 ||
sc->px_states[count].core_freq == 9999 ||
sc->px_states[count].core_freq == 0x9999 ||
sc->px_states[count].core_freq >= 0xffff)
continue;
/* Check for duplicate entries */
if (count > 0 &&
sc->px_states[count - 1].core_freq ==
sc->px_states[count].core_freq)
continue;
count++;
}
sc->px_count = count;
/* No valid Px state found so give up. */
if (count == 0)
goto out;
AcpiOsFree(buf.Pointer);
/* Get the control and status registers (one of each). */
buf.Pointer = NULL;
buf.Length = ACPI_ALLOCATE_BUFFER;
status = AcpiEvaluateObject(sc->handle, "_PCT", NULL, &buf);
if (ACPI_FAILURE(status))
goto out;
/* Check the package of two registers, each a Buffer in GAS format. */
pkg = (ACPI_OBJECT *)buf.Pointer;
if (!ACPI_PKG_VALID(pkg, 2)) {
device_printf(dev, "invalid perf register package\n");
goto out;
}
error = acpi_PkgGas(sc->dev, pkg, 0, &sc->perf_ctrl_type, &sc->px_rid,
Re-work Cx handling to be per-cpu and asymmetrical, fixing support on modern dual-core systems as well. - Parse the _CST packages for each cpu and track all the states individually, on a per-cpu basis. - Revert to generic FADT/P_BLK based Cx control if the _CST package is not present on all cpus. In that case, the new driver will still support per-cpu Cx state handling. The driver will determine the highest Cx level that can be supported by all the cpus and configure the available Cx state based on that. - Fixed the case where multiple cpus in the system share the same registers for Cx state handling. To do that, added a new flag parameter to the acpi_PkgGas and acpi_bus_alloc_gas functions that enable the caller to add the RF_SHAREABLE flag. This flag could also be useful to other callers (acpi_throttle?) in the tree but this change is not yet made. - For Core Duo cpus, both cores seems to be taken out of C3 state when any one of the cores need to transition out. This broke the short sleep detection logic. It is disabled now if there is more than one cpu in the system for now as it fixed it in my case. This quirk may need to be re-enabled later differently. - Added support to control cx_lowest on a per-cpu basis. There is still a generic cx_lowest to enable changing cx_lowest for all cpus with a single sysctl and for ease of use. Sample output for the new sysctl: dev.cpu.0.cx_supported: C1/1 C2/1 C3/57 dev.cpu.0.cx_lowest: C3 dev.cpu.0.cx_usage: 0.00% 43.16% 56.83% dev.cpu.1.cx_supported: C1/1 C2/1 C3/57 dev.cpu.1.cx_lowest: C3 dev.cpu.1.cx_usage: 0.00% 45.65% 54.34% hw.acpi.cpu.cx_lowest: C3 This work was done by Stephane E. Potvin with some simple reworking by myself. Thank you. Submitted by: Stephane E. Potvin <sepotvin / videotron.ca> MFC after: 2 weeks
2007-01-07 21:53:42 +00:00
&sc->perf_ctrl, 0);
if (error) {
/*
* If the register is of type FFixedHW, we can only return
* info, we can't get or set new settings.
*/
if (error == EOPNOTSUPP) {
sc->info_only = TRUE;
error = 0;
} else
device_printf(dev, "failed in PERF_CTL attach\n");
goto out;
}
sc->px_rid++;
error = acpi_PkgGas(sc->dev, pkg, 1, &sc->perf_sts_type, &sc->px_rid,
Re-work Cx handling to be per-cpu and asymmetrical, fixing support on modern dual-core systems as well. - Parse the _CST packages for each cpu and track all the states individually, on a per-cpu basis. - Revert to generic FADT/P_BLK based Cx control if the _CST package is not present on all cpus. In that case, the new driver will still support per-cpu Cx state handling. The driver will determine the highest Cx level that can be supported by all the cpus and configure the available Cx state based on that. - Fixed the case where multiple cpus in the system share the same registers for Cx state handling. To do that, added a new flag parameter to the acpi_PkgGas and acpi_bus_alloc_gas functions that enable the caller to add the RF_SHAREABLE flag. This flag could also be useful to other callers (acpi_throttle?) in the tree but this change is not yet made. - For Core Duo cpus, both cores seems to be taken out of C3 state when any one of the cores need to transition out. This broke the short sleep detection logic. It is disabled now if there is more than one cpu in the system for now as it fixed it in my case. This quirk may need to be re-enabled later differently. - Added support to control cx_lowest on a per-cpu basis. There is still a generic cx_lowest to enable changing cx_lowest for all cpus with a single sysctl and for ease of use. Sample output for the new sysctl: dev.cpu.0.cx_supported: C1/1 C2/1 C3/57 dev.cpu.0.cx_lowest: C3 dev.cpu.0.cx_usage: 0.00% 43.16% 56.83% dev.cpu.1.cx_supported: C1/1 C2/1 C3/57 dev.cpu.1.cx_lowest: C3 dev.cpu.1.cx_usage: 0.00% 45.65% 54.34% hw.acpi.cpu.cx_lowest: C3 This work was done by Stephane E. Potvin with some simple reworking by myself. Thank you. Submitted by: Stephane E. Potvin <sepotvin / videotron.ca> MFC after: 2 weeks
2007-01-07 21:53:42 +00:00
&sc->perf_status, 0);
if (error) {
if (error == EOPNOTSUPP) {
sc->info_only = TRUE;
error = 0;
} else
device_printf(dev, "failed in PERF_STATUS attach\n");
goto out;
}
sc->px_rid++;
/* Get our current limit and register for notifies. */
acpi_px_available(sc);
AcpiInstallNotifyHandler(sc->handle, ACPI_DEVICE_NOTIFY,
acpi_px_notify, sc);
error = 0;
out:
if (error) {
if (sc->px_states) {
free(sc->px_states, M_ACPIPERF);
sc->px_states = NULL;
}
if (sc->perf_ctrl) {
bus_release_resource(sc->dev, sc->perf_ctrl_type, 0,
sc->perf_ctrl);
bus_delete_resource(sc->dev, sc->perf_ctrl_type, 0);
sc->perf_ctrl = NULL;
}
if (sc->perf_status) {
bus_release_resource(sc->dev, sc->perf_sts_type, 1,
sc->perf_status);
bus_delete_resource(sc->dev, sc->perf_sts_type, 1);
sc->perf_status = NULL;
}
sc->px_rid = 0;
sc->px_count = 0;
}
if (buf.Pointer)
AcpiOsFree(buf.Pointer);
return (error);
}
static void
acpi_px_startup(void *arg)
{
/* Signal to the platform that we are taking over CPU control. */
2007-03-22 18:16:43 +00:00
if (AcpiGbl_FADT.PstateControl == 0)
return;
ACPI_LOCK(acpi);
2007-03-22 18:16:43 +00:00
AcpiOsWritePort(AcpiGbl_FADT.SmiCommand, AcpiGbl_FADT.PstateControl, 8);
ACPI_UNLOCK(acpi);
}
static void
acpi_px_notify(ACPI_HANDLE h, UINT32 notify, void *context)
{
struct acpi_perf_softc *sc;
sc = context;
if (notify != ACPI_NOTIFY_PERF_STATES)
return;
acpi_px_available(sc);
/* TODO: Implement notification when frequency changes. */
}
/*
* Find the highest currently-supported performance state.
* This can be called at runtime (e.g., due to a docking event) at
* the request of a Notify on the processor object.
*/
static void
acpi_px_available(struct acpi_perf_softc *sc)
{
ACPI_STATUS status;
struct cf_setting set;
status = acpi_GetInteger(sc->handle, "_PPC", &sc->px_max_avail);
/* If the old state is too high, set current state to the new max. */
if (ACPI_SUCCESS(status)) {
if (sc->px_curr_state != CPUFREQ_VAL_UNKNOWN &&
sc->px_curr_state > sc->px_max_avail) {
acpi_px_to_set(sc->dev,
&sc->px_states[sc->px_max_avail], &set);
acpi_px_set(sc->dev, &set);
}
} else
sc->px_max_avail = 0;
}
static int
acpi_px_to_set(device_t dev, struct acpi_px *px, struct cf_setting *set)
{
if (px == NULL || set == NULL)
return (EINVAL);
set->freq = px->core_freq;
set->power = px->power;
/* XXX Include BM latency too? */
set->lat = px->trans_lat;
set->volts = CPUFREQ_VAL_UNKNOWN;
set->dev = dev;
set->spec[PX_SPEC_CONTROL] = px->ctrl_val;
set->spec[PX_SPEC_STATUS] = px->sts_val;
return (0);
}
static int
acpi_px_settings(device_t dev, struct cf_setting *sets, int *count)
{
struct acpi_perf_softc *sc;
int x, y;
sc = device_get_softc(dev);
if (sets == NULL || count == NULL)
return (EINVAL);
if (*count < sc->px_count - sc->px_max_avail)
return (E2BIG);
/* Return a list of settings that are currently valid. */
y = 0;
for (x = sc->px_max_avail; x < sc->px_count; x++, y++)
acpi_px_to_set(dev, &sc->px_states[x], &sets[y]);
*count = sc->px_count - sc->px_max_avail;
return (0);
}
static int
acpi_px_set(device_t dev, const struct cf_setting *set)
{
struct acpi_perf_softc *sc;
int i, status, sts_val, tries;
if (set == NULL)
return (EINVAL);
sc = device_get_softc(dev);
/* If we can't set new states, return immediately. */
if (sc->info_only)
return (ENXIO);
/* Look up appropriate state, based on frequency. */
for (i = sc->px_max_avail; i < sc->px_count; i++) {
if (CPUFREQ_CMP(set->freq, sc->px_states[i].core_freq))
break;
}
if (i == sc->px_count)
return (EINVAL);
/* Write the appropriate value to the register. */
PX_SET_REG(sc->perf_ctrl, sc->px_states[i].ctrl_val);
/*
* Try for up to 10 ms to verify the desired state was selected.
* This is longer than the standard says (1 ms) but in some modes,
* systems may take longer to respond.
*/
sts_val = sc->px_states[i].sts_val;
for (tries = 0; tries < 1000; tries++) {
status = PX_GET_REG(sc->perf_status);
/*
* If we match the status or the desired status is 8 bits
* and matches the relevant bits, assume we succeeded. It
* appears some systems (IBM R32) expect byte-wide access
* even though the standard says the register is 32-bit.
*/
if (status == sts_val ||
((sts_val & ~0xff) == 0 && (status & 0xff) == sts_val))
break;
DELAY(10);
}
if (tries == 1000) {
device_printf(dev, "Px transition to %d failed\n",
sc->px_states[i].core_freq);
return (ENXIO);
}
sc->px_curr_state = i;
return (0);
}
static int
acpi_px_get(device_t dev, struct cf_setting *set)
{
struct acpi_perf_softc *sc;
uint64_t rate;
int i;
struct pcpu *pc;
if (set == NULL)
return (EINVAL);
sc = device_get_softc(dev);
/* If we can't get new states, return immediately. */
if (sc->info_only)
return (ENXIO);
/* If we've set the rate before, use the cached value. */
if (sc->px_curr_state != CPUFREQ_VAL_UNKNOWN) {
acpi_px_to_set(dev, &sc->px_states[sc->px_curr_state], set);
return (0);
}
/* Otherwise, estimate and try to match against our settings. */
pc = cpu_get_pcpu(dev);
if (pc == NULL)
return (ENXIO);
cpu_est_clockrate(pc->pc_cpuid, &rate);
rate /= 1000000;
for (i = 0; i < sc->px_count; i++) {
if (CPUFREQ_CMP(sc->px_states[i].core_freq, rate)) {
sc->px_curr_state = i;
acpi_px_to_set(dev, &sc->px_states[i], set);
break;
}
}
/* No match, give up. */
if (i == sc->px_count) {
sc->px_curr_state = CPUFREQ_VAL_UNKNOWN;
set->freq = CPUFREQ_VAL_UNKNOWN;
}
return (0);
}
static int
acpi_px_type(device_t dev, int *type)
{
struct acpi_perf_softc *sc;
if (type == NULL)
return (EINVAL);
sc = device_get_softc(dev);
*type = CPUFREQ_TYPE_ABSOLUTE;
if (sc->info_only)
*type |= CPUFREQ_FLAG_INFO_ONLY;
return (0);
}