[RFC] thermal: Move trip point handling code from ACPI to generic code

From: Matthew Garrett
Date: Wed Oct 01 2008 - 08:14:01 EST

Having trip point handling code in a single central location makes more
sense than implementing it per thermal driver. This is a slightly rough
prototype that should be pretty much equivalent in functionality to the
existing code (at least, my laptop hasn't melted yet). Only thing I can
think of off-hand that's missing are the updates for crossing trip
points (another callback into the specific code?) and hooking up the
polling configuration that's currently in /proc. Any comments?

Motivation for this is wanting to start hooking DRM drivers into the
thermal layer as we get better information on handing their PM
functionality. I suspect lack of thermal control in GPUs is one of the
things that's causing us to hit thermal limits on some laptops.

diff --git a/drivers/acpi/thermal.c b/drivers/acpi/thermal.c
index 754967f..c106a02 100644
--- a/drivers/acpi/thermal.c
+++ b/drivers/acpi/thermal.c
@@ -549,145 +549,6 @@ static int acpi_thermal_get_trip_points(struct acpi_thermal *tz)
return acpi_thermal_trips_update(tz, ACPI_TRIPS_INIT);
}

-static int acpi_thermal_critical(struct acpi_thermal *tz)
-{
- if (!tz || !tz->trips.critical.flags.valid)
- return -EINVAL;
-
- if (tz->temperature >= tz->trips.critical.temperature) {
- printk(KERN_WARNING PREFIX "Critical trip point\n");
- tz->trips.critical.flags.enabled = 1;
- } else if (tz->trips.critical.flags.enabled)
- tz->trips.critical.flags.enabled = 0;
-
- acpi_bus_generate_proc_event(tz->device, ACPI_THERMAL_NOTIFY_CRITICAL,
- tz->trips.critical.flags.enabled);
- acpi_bus_generate_netlink_event(tz->device->pnp.device_class,
- tz->device->dev.bus_id,
- ACPI_THERMAL_NOTIFY_CRITICAL,
- tz->trips.critical.flags.enabled);
-
- /* take no action if nocrt is set */
- if(!nocrt) {
- printk(KERN_EMERG
- "Critical temperature reached (%ld C), shutting down.\n",
- KELVIN_TO_CELSIUS(tz->temperature));
- orderly_poweroff(true);
- }
-
- return 0;
-}
-
-static int acpi_thermal_hot(struct acpi_thermal *tz)
-{
- if (!tz || !tz->trips.hot.flags.valid)
- return -EINVAL;
-
- if (tz->temperature >= tz->trips.hot.temperature) {
- printk(KERN_WARNING PREFIX "Hot trip point\n");
- tz->trips.hot.flags.enabled = 1;
- } else if (tz->trips.hot.flags.enabled)
- tz->trips.hot.flags.enabled = 0;
-
- acpi_bus_generate_proc_event(tz->device, ACPI_THERMAL_NOTIFY_HOT,
- tz->trips.hot.flags.enabled);
- acpi_bus_generate_netlink_event(tz->device->pnp.device_class,
- tz->device->dev.bus_id,
- ACPI_THERMAL_NOTIFY_HOT,
- tz->trips.hot.flags.enabled);
-
- /* TBD: Call user-mode "sleep(S4)" function if nocrt is cleared */
-
- return 0;
-}
-
-static void acpi_thermal_passive(struct acpi_thermal *tz)
-{
- int result = 1;
- struct acpi_thermal_passive *passive = NULL;
- int trend = 0;
- int i = 0;
-
-
- if (!tz || !tz->trips.passive.flags.valid)
- return;
-
- passive = &(tz->trips.passive);
-
- /*
- * Above Trip?
- * -----------
- * Calculate the thermal trend (using the passive cooling equation)
- * and modify the performance limit for all passive cooling devices
- * accordingly. Note that we assume symmetry.
- */
- if (tz->temperature >= passive->temperature) {
- trend =
- (passive->tc1 * (tz->temperature - tz->last_temperature)) +
- (passive->tc2 * (tz->temperature - passive->temperature));
- ACPI_DEBUG_PRINT((ACPI_DB_INFO,
- "trend[%d]=(tc1[%lu]*(tmp[%lu]-last[%lu]))+(tc2[%lu]*(tmp[%lu]-psv[%lu]))\n",
- trend, passive->tc1, tz->temperature,
- tz->last_temperature, passive->tc2,
- tz->temperature, passive->temperature));
- passive->flags.enabled = 1;
- /* Heating up? */
- if (trend > 0)
- for (i = 0; i < passive->devices.count; i++)
- acpi_processor_set_thermal_limit(passive->
- devices.
- handles[i],
- ACPI_PROCESSOR_LIMIT_INCREMENT);
- /* Cooling off? */
- else if (trend < 0) {
- for (i = 0; i < passive->devices.count; i++)
- /*
- * assume that we are on highest
- * freq/lowest thrott and can leave
- * passive mode, even in error case
- */
- if (!acpi_processor_set_thermal_limit
- (passive->devices.handles[i],
- ACPI_PROCESSOR_LIMIT_DECREMENT))
- result = 0;
- /*
- * Leave cooling mode, even if the temp might
- * higher than trip point This is because some
- * machines might have long thermal polling
- * frequencies (tsp) defined. We will fall back
- * into passive mode in next cycle (probably quicker)
- */
- if (result) {
- passive->flags.enabled = 0;
- ACPI_DEBUG_PRINT((ACPI_DB_INFO,
- "Disabling passive cooling, still above threshold,"
- " but we are cooling down\n"));
- }
- }
- return;
- }
-
- /*
- * Below Trip?
- * -----------
- * Implement passive cooling hysteresis to slowly increase performance
- * and avoid thrashing around the passive trip point. Note that we
- * assume symmetry.
- */
- if (!passive->flags.enabled)
- return;
- for (i = 0; i < passive->devices.count; i++)
- if (!acpi_processor_set_thermal_limit
- (passive->devices.handles[i],
- ACPI_PROCESSOR_LIMIT_DECREMENT))
- result = 0;
- if (result) {
- passive->flags.enabled = 0;
- ACPI_DEBUG_PRINT((ACPI_DB_INFO,
- "Disabling passive cooling (zone is cool)\n"));
- }
-}
-
static void acpi_thermal_active(struct acpi_thermal *tz)
{
int result = 0;
@@ -762,13 +623,6 @@ static void acpi_thermal_active(struct acpi_thermal *tz)

static void acpi_thermal_check(void *context);

-static void acpi_thermal_run(unsigned long data)
-{
- struct acpi_thermal *tz = (struct acpi_thermal *)data;
- if (!tz->zombie)
- acpi_os_execute(OSL_GPE_HANDLER, acpi_thermal_check, (void *)data);
-}
-
static void acpi_thermal_active_off(void *data)
{
int result = 0;
@@ -812,123 +666,9 @@ static void acpi_thermal_active_off(void *data)

static void acpi_thermal_check(void *data)
{
- int result = 0;
struct acpi_thermal *tz = data;
- unsigned long sleep_time = 0;
- unsigned long timeout_jiffies = 0;
- int i = 0;
- struct acpi_thermal_state state;
-
-
- if (!tz) {
- printk(KERN_ERR PREFIX "Invalid (NULL) context\n");
- return;
- }
-
- /* Check if someone else is already running */
- if (!mutex_trylock(&tz->lock))
- return;
-
- state = tz->state;
-
- result = acpi_thermal_get_temperature(tz);
- if (result)
- goto unlock;
-
- if (!tz->tz_enabled)
- goto unlock;
-
- memset(&tz->state, 0, sizeof(tz->state));
-
- /*
- * Check Trip Points
- * -----------------
- * Compare the current temperature to the trip point values to see
- * if we've entered one of the thermal policy states. Note that
- * this function determines when a state is entered, but the
- * individual policy decides when it is exited (e.g. hysteresis).
- */
- if (tz->trips.critical.flags.valid)
- state.critical |=
- (tz->temperature >= tz->trips.critical.temperature);
- if (tz->trips.hot.flags.valid)
- state.hot |= (tz->temperature >= tz->trips.hot.temperature);
- if (tz->trips.passive.flags.valid)
- state.passive |=
- (tz->temperature >= tz->trips.passive.temperature);
- for (i = 0; i < ACPI_THERMAL_MAX_ACTIVE; i++)
- if (tz->trips.active[i].flags.valid)
- state.active |=
- (tz->temperature >=
- tz->trips.active[i].temperature);
-
- /*
- * Invoke Policy
- * -------------
- * Separated from the above check to allow individual policy to
- * determine when to exit a given state.
- */
- if (state.critical)
- acpi_thermal_critical(tz);
- if (state.hot)
- acpi_thermal_hot(tz);
- if (state.passive)
- acpi_thermal_passive(tz);
- if (state.active)
- acpi_thermal_active(tz);
-
- /*
- * Calculate State
- * ---------------
- * Again, separated from the above two to allow independent policy
- * decisions.
- */
- tz->state.critical = tz->trips.critical.flags.enabled;
- tz->state.hot = tz->trips.hot.flags.enabled;
- tz->state.passive = tz->trips.passive.flags.enabled;
- tz->state.active = 0;
- for (i = 0; i < ACPI_THERMAL_MAX_ACTIVE; i++)
- tz->state.active |= tz->trips.active[i].flags.enabled;

- /*
- * Calculate Sleep Time
- * --------------------
- * If we're in the passive state, use _TSP's value. Otherwise
- * use the default polling frequency (e.g. _TZP). If no polling
- * frequency is specified then we'll wait forever (at least until
- * a thermal event occurs). Note that _TSP and _TZD values are
- * given in 1/10th seconds (we must covert to milliseconds).
- */
- if (tz->state.passive) {
- sleep_time = tz->trips.passive.tsp * 100;
- timeout_jiffies = jiffies + (HZ * sleep_time) / 1000;
- } else if (tz->polling_frequency > 0) {
- sleep_time = tz->polling_frequency * 100;
- timeout_jiffies = round_jiffies(jiffies + (HZ * sleep_time) / 1000);
- }
-
- ACPI_DEBUG_PRINT((ACPI_DB_INFO, "%s: temperature[%lu] sleep[%lu]\n",
- tz->name, tz->temperature, sleep_time));
-
- /*
- * Schedule Next Poll
- * ------------------
- */
- if (!sleep_time) {
- if (timer_pending(&(tz->timer)))
- del_timer(&(tz->timer));
- } else {
- if (timer_pending(&(tz->timer)))
- mod_timer(&(tz->timer), timeout_jiffies);
- else {
- tz->timer.data = (unsigned long)tz;
- tz->timer.function = acpi_thermal_run;
- tz->timer.expires = timeout_jiffies;
- add_timer(&(tz->timer));
- }
- }
- unlock:
- mutex_unlock(&tz->lock);
+ thermal_zone_device_update(tz->thermal_zone);
}

/* sys I/F for generic thermal sysfs support */
@@ -1213,8 +953,21 @@ static int acpi_thermal_register_thermal_zone(struct acpi_thermal *tz)

for (i = 0; i < ACPI_THERMAL_MAX_ACTIVE &&
tz->trips.active[i].flags.valid; i++, trips++);
- tz->thermal_zone = thermal_zone_device_register("acpitz",
- trips, tz, &acpi_thermal_zone_ops);
+
+ if (tz->trips.passive.flags.valid)
+ tz->thermal_zone =
+ thermal_zone_device_register("acpitz", trips, tz,
+ &acpi_thermal_zone_ops,
+ tz->trips.passive.tc1,
+ tz->trips.passive.tc2,
+ tz->trips.passive.tsp*100,
+ tz->polling_frequency*100);
+ else
+ tz->thermal_zone =
+ thermal_zone_device_register("acpitz", trips, tz,
+ &acpi_thermal_zone_ops,
+ 0, 0, 0,
+ tz->polling_frequency);
if (IS_ERR(tz->thermal_zone))
return -ENODEV;

@@ -1736,10 +1489,6 @@ static int acpi_thermal_remove(struct acpi_device *device, int type)
acpi_thermal_notify);

/* Terminate policy */
- if (tz->trips.passive.flags.valid && tz->trips.passive.flags.enabled) {
- tz->trips.passive.flags.enabled = 0;
- acpi_thermal_passive(tz);
- }
if (tz->trips.active[0].flags.valid
&& tz->trips.active[0].flags.enabled) {
tz->trips.active[0].flags.enabled = 0;
diff --git a/drivers/thermal/thermal_sys.c b/drivers/thermal/thermal_sys.c
index 8ca2f59..7eff12f 100644
--- a/drivers/thermal/thermal_sys.c
+++ b/drivers/thermal/thermal_sys.c
@@ -30,6 +30,7 @@
#include <linux/idr.h>
#include <linux/thermal.h>
#include <linux/spinlock.h>
+#include <linux/reboot.h>

MODULE_AUTHOR("Zhang Rui");
MODULE_DESCRIPTION("Generic thermal management sysfs support");
@@ -161,7 +162,6 @@ trip_point_type_show(struct device *dev, struct device_attribute *attr,
if (!sscanf(attr->attr.name, "trip_point_%d_type", &trip))
return -EINVAL;

-
ret = tz->ops->get_trip_type(tz, trip, &trip_type);
if (ret)
return ret;
@@ -778,12 +778,166 @@ void thermal_cooling_device_unregister(struct

EXPORT_SYMBOL(thermal_cooling_device_unregister);

+static void thermal_zone_device_set_polling(struct thermal_zone_device *tz,
+ int delay)
+{
+ cancel_delayed_work_sync(&(tz->poll_queue));
+
+ if (!delay)
+ return;
+
+ if (delay > 1000)
+ schedule_delayed_work(&(tz->poll_queue),
+ round_jiffies(delay * HZ));
+ else
+ schedule_delayed_work(&(tz->poll_queue), delay * HZ);
+}
+
+static void thermal_zone_device_passive(struct thermal_zone_device *tz,
+ int temp, int trip_temp, int trip)
+{
+ int trend = 0;
+ struct thermal_cooling_device_instance *instance;
+ struct thermal_cooling_device *cdev;
+ int state, max_state;
+
+ if (tz->passive == false)
+ return;
+
+ /*
+ * Above Trip?
+ * -----------
+ * Calculate the thermal trend (using the passive cooling equation)
+ * and modify the performance limit for all passive cooling devices
+ * accordingly. Note that we assume symmetry.
+ */
+ if (temp >= trip_temp) {
+ tz->passive = true;
+
+ trend = (tz->tc1 * (temp - tz->last_temperature)) +
+ (tz->tc2 * (temp - trip_temp));
+
+ /* Heating up? */
+ if (trend > 0) {
+ list_for_each_entry(instance, &tz->cooling_devices,
+ node) {
+ if (instance->trip != trip)
+ continue;
+ cdev = instance->cdev;
+ cdev->ops->get_cur_state(cdev, &state);
+ cdev->ops->get_max_state(cdev, &max_state);
+ if (state++ < max_state)
+ cdev->ops->set_cur_state(cdev, state);
+ }
+ } else if (trend < 0) { /* Cooling off? */
+ list_for_each_entry(instance, &tz->cooling_devices,
+ node) {
+ if (instance->trip != trip)
+ continue;
+ cdev = instance->cdev;
+ cdev->ops->get_cur_state(cdev, &state);
+ cdev->ops->get_max_state(cdev, &max_state);
+ if (state > 0)
+ cdev->ops->set_cur_state(cdev, --state);
+ }
+ return;
+ }
+ }
+
+ /*
+ * Below Trip?
+ * -----------
+ * Implement passive cooling hysteresis to slowly increase performance
+ * and avoid thrashing around the passive trip point. Note that we
+ * assume symmetry.
+ */
+ list_for_each_entry(instance, &tz->cooling_devices, node) {
+ if (instance->trip != trip)
+ continue;
+ cdev = instance->cdev;
+ cdev->ops->get_cur_state(cdev, &state);
+ cdev->ops->get_max_state(cdev, &max_state);
+ if (state > 0)
+ cdev->ops->set_cur_state(cdev, --state);
+ if (state == 0)
+ tz->passive = false;
+ }
+}
+
+void thermal_zone_device_update(struct thermal_zone_device *tz)
+{
+ int temp, trip_temp;
+ int count;
+ enum thermal_trip_t trip_type;
+ struct thermal_cooling_device_instance *instance;
+ struct thermal_cooling_device *cdev;
+
+ tz->ops->get_temp(tz, &temp);
+
+ for (count = 0; count < tz->trips; count++) {
+ tz->ops->get_trip_type(tz, count, &trip_type);
+ tz->ops->get_trip_temp(tz, count, &trip_temp);
+
+ switch (trip_type) {
+ case THERMAL_TRIP_CRITICAL:
+ if (temp > trip_temp)
+ /* FIXME: send notification */
+ orderly_poweroff(true);
+ break;
+ case THERMAL_TRIP_HOT:
+ if (temp > trip_temp)
+ /* FIXME: send notification */
+ printk(KERN_WARNING "Hot trip point\n");
+ break;
+ case THERMAL_TRIP_ACTIVE:
+ list_for_each_entry(instance, &tz->cooling_devices,
+ node) {
+ if (instance->trip != count)
+ continue;
+
+ cdev = instance->cdev;
+
+ if (temp > trip_temp)
+ cdev->ops->set_cur_state(cdev, 1);
+ else
+ cdev->ops->set_cur_state(cdev, 0);
+ }
+ break;
+ case THERMAL_TRIP_PASSIVE:
+ thermal_zone_device_passive(tz, temp, trip_temp, count);
+ break;
+ }
+ }
+ tz->last_temperature = temp;
+ if (tz->passive)
+ thermal_zone_device_set_polling(tz, tz->passive_delay);
+ else if (tz->polling_delay)
+ thermal_zone_device_set_polling(tz, tz->polling_delay);
+}
+
+EXPORT_SYMBOL(thermal_zone_device_update);
+
+static void thermal_zone_device_check(struct work_struct *work)
+{
+ struct thermal_zone_device *tz = container_of(work, struct
+ thermal_zone_device,
+ poll_queue.work);
+ thermal_zone_device_update(tz);
+}
+
/**
* thermal_zone_device_register - register a new thermal zone device
* @type: the thermal zone device type
* @trips: the number of trip points the thermal zone support
* @devdata: private device data
* @ops: standard thermal zone device callbacks
+ * @tc1: thermal coefficient 1 for passive calculations
+ * @tc2: thermal coefficient 2 for passive calculations
+ * @passive_delay: number of milliseconds to wait between polls when
+ * performing passive cooling
+ * @polling_delay: number of milliseconds to wait between polls when checking
+ * whether trip points have been crossed (0 for interrupt
+ * driven systems)
*
* thermal_zone_device_unregister() must be called when the device is no
* longer needed.
@@ -792,7 +946,10 @@ struct thermal_zone_device *thermal_zone_device_register(char *type,
int trips,
void *devdata, struct
thermal_zone_device_ops
- *ops)
+ *ops, int tc1, int
+ tc2,
+ int passive_delay,
+ int polling_delay)
{
struct thermal_zone_device *tz;
struct thermal_cooling_device *pos;
@@ -826,6 +983,11 @@ struct thermal_zone_device *thermal_zone_device_register(char *type,
tz->device.class = &thermal_class;
tz->devdata = devdata;
tz->trips = trips;
+ tz->tc1 = tc1;
+ tz->tc2 = tc2;
+ tz->passive_delay = passive_delay;
+ tz->polling_delay = polling_delay;
+
sprintf(tz->device.bus_id, "thermal_zone%d", tz->id);
result = device_register(&tz->device);
if (result) {
@@ -871,6 +1033,10 @@ struct thermal_zone_device *thermal_zone_device_register(char *type,
}
mutex_unlock(&thermal_list_lock);

+ INIT_DELAYED_WORK(&(tz->poll_queue), thermal_zone_device_check);
+
+ thermal_zone_device_set_polling(tz, tz->polling_delay);
+
if (!result)
return tz;

@@ -910,6 +1076,8 @@ void thermal_zone_device_unregister(struct thermal_zone_device *tz)
tz->ops->unbind(tz, cdev);
mutex_unlock(&thermal_list_lock);

+ thermal_zone_device_set_polling(tz, 0);
+
if (tz->type[0])
device_remove_file(&tz->device, &dev_attr_type);
device_remove_file(&tz->device, &dev_attr_temp);
diff --git a/include/linux/thermal.h b/include/linux/thermal.h
index 9e3475a..38ac33d 100644
--- a/include/linux/thermal.h
+++ b/include/linux/thermal.h
@@ -94,11 +94,18 @@ struct thermal_zone_device {
struct device device;
void *devdata;
int trips;
+ int tc1;
+ int tc2;
+ int passive_delay;
+ int polling_delay;
+ int last_temperature;
+ bool passive;
struct thermal_zone_device_ops *ops;
struct list_head cooling_devices;
struct idr idr;
struct mutex lock; /* protect cooling devices list */
struct list_head node;
+ struct delayed_work poll_queue;
#if defined(CONFIG_THERMAL_HWMON)
struct list_head hwmon_node;
struct thermal_hwmon_device *hwmon;
@@ -110,13 +117,16 @@ struct thermal_zone_device {
struct thermal_zone_device *thermal_zone_device_register(char *, int, void *,
struct
thermal_zone_device_ops
- *);
+ *, int tc1, int tc2,
+ int passive_freq,
+ int polling_freq);
void thermal_zone_device_unregister(struct thermal_zone_device *);

int thermal_zone_bind_cooling_device(struct thermal_zone_device *, int,
struct thermal_cooling_device *);
int thermal_zone_unbind_cooling_device(struct thermal_zone_device *, int,
struct thermal_cooling_device *);
+void thermal_zone_device_update(struct thermal_zone_device *);
struct thermal_cooling_device *thermal_cooling_device_register(char *, void *,
struct
thermal_cooling_device_ops

--
Matthew Garrett | mjg59@xxxxxxxxxxxxx
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/