Skip to contentSkip to conversation

Add thermal thresholds support

PATCH Applied Daniel Lezcano 3 participants 27 messages Aug 16, 2024 – Sep 4, 2024 Thermal

v2 of 6 Applied Latest is v6 →

lore
Share & tools
lore
#0 lore raw

v2 cover Add thermal thresholds support

The trip points are a firmware description of the temperature limits of a specific thermal zone where we associate an action which is done by the kernel. The time resolution is low.

The userspace has to deal with a more complex thermal management based on heuristics from different information coming from different places. The logic is much more complex but based on a bigger time resolution, usually one second based.

The purpose of the userspace is to monitor the temperatures from different places and take actions. However, it can not be constantly reading the temperature to detect when a temperature threshold has been reached. This is especially bad for mobile or embedded system as that will lead to an unacceptable number of wakeup to check the temperature with nothing to do.

On the other side, the sensors are now most of the time interrupt driven. That means the thermal framework will use the temperature trip points to program the sensor to trigger an interrupt when a temperature limit is crossed.

Unfortunately, the userspace can not benefit this feature and current solutions found here and there, iow out-of-tree, are to add fake trip points in the firmware and enable the writable trip points.

This is bad for different reasons, the trip points are for in-kernel actions, the semantic of their types is used by the thermal framework and by adding trip points in the device tree is a way to overcome the current limitation but tampering with how the thermal framework is supposed to work. The writable trip points is a way to adjust a temperature limit given a specific platform if the firmware is not accurate enough and TBH it is more a debug feature from my POV.

The thresholds mechanism is a way to have the userspace to tell thermal framework to send a notification when a temperature limit is crossed. There is no id, no hysteresis, just the temperature and the direction of the limit crossing. That means we can be notified when a threshold is crossed the way up only, or the way down only or both ways. That allows to create hysteresis values if it is needed.

A threshold can be added, deleted or flushed. The latter means all thresholds belonging to a thermal zone will be deleted.

When a threshold is added:

- if the same threshold (temperature and direction) exists, an error is returned

- if a threshold is specified with the same temperature but a different direction, the specified direction is added

- if there is no threshold with the same temperature then it is created

When a threshold is deleted:

- if the same threshold (temperature and direction) exists, it is deleted

- if a threshold is specified with the same temperature but a different direction, the specified direction is removed

- if there is no threshold with the same temperature, then an error is returned

When the threshold are flushed:

- All thresholds related to a thermal zone are deleted

When a threshold is crossed:

- the userspace does not need to know which threshold(s) have been crossed, it will be notified with the thermal zone identifier, the current temperature and the previous temperature

- if multiple thresholds have been crossed between two updates only one notification will be send to the userspace, it is pointless to send a notification per thresholds crossed as the userspace can handle that easily when it has the temperature delta information

All aforementioned actions and events lead to a notification to the userspace. A threshold change (add, delete and flush) is notified to the userspace with the process id responsible of the action.

Along with the kernel changes, the thermal library has been extended to provide the different API to deal with the new threshold netlink events and commands.

In addition, the thermal-engine skeleton uses these new API by flushing and adding thresholds as well as getting the notification about these actions.

Overall the series has been tested with the thermal-engine skeleton and some selftests which are not part of this series.

Changelog:
V2:
- Compute min and max in thermal_zone_device_update() but keep the loop as it is (Rafael)

- Include slab.h to fix compilation warnings on some architectures with kmalloc and kfree (kernel test robot)

Daniel Lezcano (7):
thermal/core: Compute low and high boundaries in
thermal_zone_device_update()
thermal/core: Add thresholds support
thermal/core: Connect the threshold with the core thermal/netlink: Add the commands and the events for the thresholds tools/lib/thermal: Make more generic the command encoding function tools/lib/thermal: Add the threshold netlink ABI tools/thermal/thermal-engine: Take into account the thresholds API

 drivers/thermal/Kconfig                       |  15 ++
 drivers/thermal/Makefile                      |   3 +
 drivers/thermal/thermal_core.c                |  21 +-
 drivers/thermal/thermal_core.h                |   6 +-
 drivers/thermal/thermal_netlink.c             | 239 ++++++++++++++++-
 drivers/thermal/thermal_netlink.h             |   7 +
 drivers/thermal/thermal_thresholds.c          | 247 ++++++++++++++++++
 drivers/thermal/thermal_thresholds.h          |  54 ++++
 drivers/thermal/thermal_trip.c                |  27 +-
 include/linux/thermal.h                       |   3 +
 include/uapi/linux/thermal.h                  |  30 ++-
 tools/lib/thermal/commands.c                  | 167 +++++++++++-
 tools/lib/thermal/events.c                    |  58 +++-
 tools/lib/thermal/include/thermal.h           |  40 +++
 tools/lib/thermal/libthermal.map              |   5 +
 tools/lib/thermal/thermal.c                   |  17 ++
 tools/thermal/lib/Makefile                    |   2 +-
 tools/thermal/thermal-engine/thermal-engine.c | 109 +++++++-
 18 files changed, 972 insertions(+), 78 deletions(-)
 create mode 100644 drivers/thermal/thermal_thresholds.c
 create mode 100644 drivers/thermal/thermal_thresholds.h
Signature · 1 line
2.43.0
#1 lore raw

PATCH v2 1/7 thermal/core: Compute low and high boundaries in thermal_zone_device_update()

In order to set the scene for the thresholds support which have to manipulate the low and high temperature boundaries for the interrupt support, we must pass the low and high values to the incoming thresholds routine.

The variables are set from the thermal_zone_set_trips() where the function loops the thermal trips to figure out the next and the previous temperatures to set the interrupt to be triggered when they are crossed.

These variables will be needed by the function in charge of handling the thresholds in the incoming changes but they are local to the aforementioned function thermal_zone_set_trips().

Move the low and high boundaries computation out of the function in thermal_zone_device_update() so they are accessible from there.

The positive side effect is they are computed in the same loop as handle_thermal_trip(), so we remove one loop.

---
 drivers/thermal/thermal_core.c | 12 ++++++++++--
 drivers/thermal/thermal_core.h |  2 +-
 drivers/thermal/thermal_trip.c | 27 +--------------------------
 3 files changed, 12 insertions(+), 29 deletions(-)
drivers/thermal/thermal_core.c +10 -2
index 95c399f94744..166f48071487 100644
@@ -547,6 +547,7 @@ void __thermal_zone_device_update(struct thermal_zone_device *tz,
 	struct thermal_trip_desc *td;
 	LIST_HEAD(way_down_list);
 	LIST_HEAD(way_up_list);
+	int low = -INT_MAX, high = INT_MAX;
 	int temp, ret;
 
 	if (tz->suspended)
@@ -580,10 +581,17 @@ void __thermal_zone_device_update(struct thermal_zone_device *tz,
 
 	tz->notify_event = event;
 
-	for_each_trip_desc(tz, td)
+	for_each_trip_desc(tz, td) {
 		handle_thermal_trip(tz, td, &way_up_list, &way_down_list);
 
-	thermal_zone_set_trips(tz);
+		if (td->threshold <= tz->temperature && td->threshold > low)
+			low = td->threshold;
+
+		if (td->threshold >= tz->temperature && td->threshold < high)
+			high = td->threshold;
+	}
+
+	thermal_zone_set_trips(tz, low, high);
 
 	list_sort(NULL, &way_up_list, thermal_trip_notify_cmp);
 	list_for_each_entry(td, &way_up_list, notify_list_node)
drivers/thermal/thermal_core.h +1 -1
index 4cf2b7230d04..67a09f90eb95 100644
@@ -259,7 +259,7 @@ void thermal_governor_update_tz(struct thermal_zone_device *tz,
 
 const char *thermal_trip_type_name(enum thermal_trip_type trip_type);
 
-void thermal_zone_set_trips(struct thermal_zone_device *tz);
+void thermal_zone_set_trips(struct thermal_zone_device *tz, int low, int high);
 int thermal_zone_trip_id(const struct thermal_zone_device *tz,
 			 const struct thermal_trip *trip);
 void thermal_zone_trip_updated(struct thermal_zone_device *tz,
drivers/thermal/thermal_trip.c +1 -26
index 06a0554ddc38..1d43ab52e86a 100644
@@ -61,25 +61,8 @@ int thermal_zone_get_num_trips(struct thermal_zone_device *tz)
 }
 EXPORT_SYMBOL_GPL(thermal_zone_get_num_trips);
 
-/**
- * thermal_zone_set_trips - Computes the next trip points for the driver
- * @tz: a pointer to a thermal zone device structure
- *
- * The function computes the next temperature boundaries by browsing
- * the trip points. The result is the closer low and high trip points
- * to the current temperature. These values are passed to the backend
- * driver to let it set its own notification mechanism (usually an
- * interrupt).
- *
- * This function must be called with tz->lock held. Both tz and tz->ops
- * must be valid pointers.
- *
- * It does not return a value
- */
-void thermal_zone_set_trips(struct thermal_zone_device *tz)
+void thermal_zone_set_trips(struct thermal_zone_device *tz, int low, int high)
 {
-	const struct thermal_trip_desc *td;
-	int low = -INT_MAX, high = INT_MAX;
 	int ret;
 
 	lockdep_assert_held(&tz->lock);
@@ -87,14 +70,6 @@ void thermal_zone_set_trips(struct thermal_zone_device *tz)
 	if (!tz->ops.set_trips)
 		return;
 
-	for_each_trip_desc(tz, td) {
-		if (td->threshold <= tz->temperature && td->threshold > low)
-			low = td->threshold;
-
-		if (td->threshold >= tz->temperature && td->threshold < high)
-			high = td->threshold;
-	}
-
 	/* No need to change trip points */
 	if (tz->prev_low_trip == low && tz->prev_high_trip == high)
 		return;
Signature · 1 line
2.43.0
#2 lore raw

Re: [PATCH v2 1/7] thermal/core: Compute low and high boundaries in thermal_zone_device_update()

Quoted 15 lines · Daniel Lezcano

In order to set the scene for the thresholds support which have to manipulate the low and high temperature boundaries for the interrupt support, we must pass the low and high values to the incoming thresholds routine.

The variables are set from the thermal_zone_set_trips() where the function loops the thermal trips to figure out the next and the previous temperatures to set the interrupt to be triggered when they are crossed.

These variables will be needed by the function in charge of handling the thresholds in the incoming changes but they are local to the aforementioned function thermal_zone_set_trips().

Quoted 8 lines

Move the low and high boundaries computation out of the function in thermal_zone_device_update() so they are accessible from there.

The positive side effect is they are computed in the same loop as handle_thermal_trip(), so we remove one loop.

Looks good to me and I'd like to apply it earlier separately as I have material depending on it in the works.

Quoted 100 lines
---
 drivers/thermal/thermal_core.c | 12 ++++++++++--
 drivers/thermal/thermal_core.h |  2 +-
 drivers/thermal/thermal_trip.c | 27 +--------------------------
 3 files changed, 12 insertions(+), 29 deletions(-)
drivers/thermal/thermal_core.c +10 -2
index 95c399f94744..166f48071487 100644
@@ -547,6 +547,7 @@ void __thermal_zone_device_update(struct thermal_zone_device *tz,
        struct thermal_trip_desc *td;
        LIST_HEAD(way_down_list);
        LIST_HEAD(way_up_list);
+       int low = -INT_MAX, high = INT_MAX;
        int temp, ret;
 
        if (tz->suspended)
@@ -580,10 +581,17 @@ void __thermal_zone_device_update(struct thermal_zone_device *tz,
 
        tz->notify_event = event;
 
-       for_each_trip_desc(tz, td)
+       for_each_trip_desc(tz, td) {
                handle_thermal_trip(tz, td, &way_up_list, &way_down_list);
 
-       thermal_zone_set_trips(tz);
+               if (td->threshold <= tz->temperature && td->threshold > low)
+                       low = td->threshold;
+
+               if (td->threshold >= tz->temperature && td->threshold < high)
+                       high = td->threshold;
+       }
+
+       thermal_zone_set_trips(tz, low, high);
 
        list_sort(NULL, &way_up_list, thermal_trip_notify_cmp);
        list_for_each_entry(td, &way_up_list, notify_list_node)
drivers/thermal/thermal_core.h +1 -1
index 4cf2b7230d04..67a09f90eb95 100644
@@ -259,7 +259,7 @@ void thermal_governor_update_tz(struct thermal_zone_device *tz,
 
 const char *thermal_trip_type_name(enum thermal_trip_type trip_type);
 
-void thermal_zone_set_trips(struct thermal_zone_device *tz);
+void thermal_zone_set_trips(struct thermal_zone_device *tz, int low, int high);
 int thermal_zone_trip_id(const struct thermal_zone_device *tz,
                         const struct thermal_trip *trip);
 void thermal_zone_trip_updated(struct thermal_zone_device *tz,
drivers/thermal/thermal_trip.c +1 -27
index 06a0554ddc38..1d43ab52e86a 100644
@@ -61,25 +61,8 @@ int thermal_zone_get_num_trips(struct thermal_zone_device *tz)
 }
 EXPORT_SYMBOL_GPL(thermal_zone_get_num_trips);
 
-/**
- * thermal_zone_set_trips - Computes the next trip points for the driver
- * @tz: a pointer to a thermal zone device structure
- *
- * The function computes the next temperature boundaries by browsing
- * the trip points. The result is the closer low and high trip points
- * to the current temperature. These values are passed to the backend
- * driver to let it set its own notification mechanism (usually an
- * interrupt).
- *
- * This function must be called with tz->lock held. Both tz and tz->ops
- * must be valid pointers.
- *
- * It does not return a value
- */
-void thermal_zone_set_trips(struct thermal_zone_device *tz)
+void thermal_zone_set_trips(struct thermal_zone_device *tz, int low, int high)
 {
-       const struct thermal_trip_desc *td;
-       int low = -INT_MAX, high = INT_MAX;
        int ret;
 
        lockdep_assert_held(&tz->lock);
@@ -87,14 +70,6 @@ void thermal_zone_set_trips(struct thermal_zone_device *tz)
        if (!tz->ops.set_trips)
                return;
 
-       for_each_trip_desc(tz, td) {
-               if (td->threshold <= tz->temperature && td->threshold > low)
-                       low = td->threshold;
-
-               if (td->threshold >= tz->temperature && td->threshold < high)
-                       high = td->threshold;
-       }
-
        /* No need to change trip points */
        if (tz->prev_low_trip == low && tz->prev_high_trip == high)
                return;
--

2.43.0

#3 lore raw

Re: [PATCH v2 1/7] thermal/core: Compute low and high boundaries in thermal_zone_device_update()

Quoted 20 lines · Rafael J. Wysocki
Quoted 18 lines · Daniel Lezcano

In order to set the scene for the thresholds support which have to manipulate the low and high temperature boundaries for the interrupt support, we must pass the low and high values to the incoming thresholds routine.

The variables are set from the thermal_zone_set_trips() where the function loops the thermal trips to figure out the next and the previous temperatures to set the interrupt to be triggered when they are crossed.

These variables will be needed by the function in charge of handling the thresholds in the incoming changes but they are local to the aforementioned function thermal_zone_set_trips().

Move the low and high boundaries computation out of the function in thermal_zone_device_update() so they are accessible from there.

Quoted 8 lines
Quoted 5 lines

The positive side effect is they are computed in the same loop as handle_thermal_trip(), so we remove one loop.

Looks good to me and I'd like to apply it earlier separately as I have material depending on it in the works.

Sure

Signature · 5 lines
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs

Follow Linaro:  <http://www.facebook.com/pages/Linaro> Facebook |
<http://twitter.com/#!/linaroorg> Twitter |
<http://www.linaro.org/linaro-blog/> Blog
maintainer Applied
#4 lore raw

Re: [PATCH v2 1/7] thermal/core: Compute low and high boundaries in thermal_zone_device_update()

Quoted 25 lines · Daniel Lezcano
Quoted 23 lines · Rafael J. Wysocki
Quoted 21 lines · Daniel Lezcano

In order to set the scene for the thresholds support which have to manipulate the low and high temperature boundaries for the interrupt support, we must pass the low and high values to the incoming thresholds routine.

The variables are set from the thermal_zone_set_trips() where the function loops the thermal trips to figure out the next and the previous temperatures to set the interrupt to be triggered when they are crossed.

These variables will be needed by the function in charge of handling the thresholds in the incoming changes but they are local to the aforementioned function thermal_zone_set_trips().

Move the low and high boundaries computation out of the function in thermal_zone_device_update() so they are accessible from there.

The positive side effect is they are computed in the same loop as handle_thermal_trip(), so we remove one loop.

Quoted 7 lines
Quoted 5 lines
Quoted 2 lines

Looks good to me and I'd like to apply it earlier separately as I have material depending on it in the works.

Sure

Applied as 6.12 material, thanks!

#5 lore raw

PATCH v2 2/7 thermal/core: Add thresholds support

In reply to Daniel Lezcano

The trip points are a firmware description of the temperature limits of a specific thermal zone where we associate an action which is done by the kernel. The time resolution is low.

The userspace has to deal with a more complex thermal management based on heuristics from different information coming from different places. The logic is much more complex but based on a bigger time resolution, usually one second based.

The purpose of the userspace is to monitor the temperatures from different places and take actions. However, it can not be constantly reading the temperature to detect when a temperature threshold has been reached. This is especially bad for mobile or embedded system as that will lead to an unacceptable number of wakeup to check the temperature with nothing to do.

On the other side, the sensors are now most of the time interrupt driven. That means the thermal framework will use the temperature trip points to program the sensor to trigger an interrupt when a temperature limit is crossed.

Unfortunately, the userspace can not benefit this feature and current solutions found here and there, iow out-of-tree, are to add fake trip points in the firmware and enable the writable trip points.

This is bad for different reasons, the trip points are for in-kernel actions, the semantic of their types is used by the thermal framework and by adding trip points in the device tree is a way to overcome the current limitation but tampering with how the thermal framework is supposed to work. The writable trip points is a way to adjust a temperature limit given a specific platform if the firmware is not accurate enough and TBH it is more a debug feature from my POV.

The thresholds mechanism is a way to have the userspace to tell thermal framework to send a notification when a temperature limit is crossed. There is no id, no hysteresis, just the temperature and the direction of the limit crossing. That means we can be notified when a threshold is crossed the way up only, or the way down only or both ways. That allows to create hysteresis values if it is needed.

A threshold can be added, deleted or flushed. The latter means all thresholds belonging to a thermal zone will be deleted.

When a threshold is added:

- if the same threshold (temperature and direction) exists, an error is returned

- if a threshold is specified with the same temperature but a different direction, the specified direction is added

- if there is no threshold with the same temperature then it is created

When a threshold is deleted:

- if the same threshold (temperature and direction) exists, it is deleted

- if a threshold is specified with the same temperature but a different direction, the specified direction is removed

- if there is no threshold with the same temperature, then an error is returned

When the threshold are flushed:

- All thresholds related to a thermal zone are deleted

When a threshold is crossed:

- the userspace does not need to know which threshold(s) have been crossed, it will be notified with the current temperature and the previous temperature

- if multiple thresholds have been crossed between two updates only one notification will be send to the userspace, it is pointless to send a notification per thresholds crossed as the userspace can handle that easily when it has the temperature delta information

---
 drivers/thermal/Kconfig              |  15 ++
 drivers/thermal/Makefile             |   3 +
 drivers/thermal/thermal_core.h       |   4 +
 drivers/thermal/thermal_thresholds.c | 241 +++++++++++++++++++++++++++
 drivers/thermal/thermal_thresholds.h |  57 +++++++
 include/linux/thermal.h              |   3 +
 6 files changed, 323 insertions(+)
 create mode 100644 drivers/thermal/thermal_thresholds.c
 create mode 100644 drivers/thermal/thermal_thresholds.h
drivers/thermal/Kconfig +15 -0
index ed16897584b4..84f9643678d6 100644
@@ -40,6 +40,21 @@ config THERMAL_DEBUGFS
 	  Say Y to allow the thermal subsystem to collect diagnostic
 	  information that can be accessed via debugfs.
 
+config THERMAL_THRESHOLDS
+	bool "Thermal thresholds notification mechanism"
+	depends on THERMAL_NETLINK
+	help
+	  The userspace implements thermal engines which needs to get
+	  notified when temperature thresholds are crossed the way up
+	  and down. These notification allow them to analyze the
+	  thermal situation of the platform and take decision to
+	  fulfill specific thermal profile like 'balanced',
+	  'performance' or 'power saving'. In addition, the
+	  temperature of the skin sensor is very important in this
+	  case and must be monitored as well.
+
+	  If in doubt, say Y
+
 config THERMAL_EMERGENCY_POWEROFF_DELAY_MS
 	int "Emergency poweroff delay in milli-seconds"
 	default 0
drivers/thermal/Makefile +3 -0
index ce7a4752ef52..3b991b1a7db4 100644
@@ -7,6 +7,9 @@ obj-$(CONFIG_THERMAL)		+= thermal_sys.o
 thermal_sys-y			+= thermal_core.o thermal_sysfs.o
 thermal_sys-y			+= thermal_trip.o thermal_helpers.o
 
+# thermal thresholds
+thermal_sys-$(CONFIG_THERMAL_THRESHOLDS)	+= thermal_thresholds.o
+
 # netlink interface to manage the thermal framework
 thermal_sys-$(CONFIG_THERMAL_NETLINK)		+= thermal_netlink.o
 
drivers/thermal/thermal_core.h +4 -0
index 67a09f90eb95..0742c0f03d46 100644
@@ -13,6 +13,7 @@
 #include <linux/thermal.h>
 
 #include "thermal_netlink.h"
+#include "thermal_thresholds.h"
 #include "thermal_debugfs.h"
 
 struct thermal_trip_desc {
@@ -132,6 +133,9 @@ struct thermal_zone_device {
 	bool resuming;
 #ifdef CONFIG_THERMAL_DEBUGFS
 	struct thermal_debugfs *debugfs;
+#endif
+#ifdef CONFIG_THERMAL_THRESHOLDS
+	struct thresholds *thresholds;
 #endif
 	struct thermal_trip_desc trips[] __counted_by(num_trips);
 };
drivers/thermal/thermal_thresholds.c new file +241 -0

241 lines not on this page · raw message

drivers/thermal/thermal_thresholds.h new file +57 -0

57 lines not on this page · raw message

include/linux/thermal.h +3 -0
index 25fbf960b474..bf0b4a8218f6 100644
@@ -55,6 +55,9 @@ enum thermal_notify_event {
 	THERMAL_TZ_BIND_CDEV, /* Cooling dev is bind to the thermal zone */
 	THERMAL_TZ_UNBIND_CDEV, /* Cooling dev is unbind from the thermal zone */
 	THERMAL_INSTANCE_WEIGHT_CHANGED, /* Thermal instance weight changed */
+	THERMAL_THRESHOLD_ADDED, /* Threshold added */
+	THERMAL_THRESHOLD_DELETED, /* Threshold deleted */
+	THERMAL_THRESHOLD_FLUSHED, /* All thresholds deleted */
 };
 
 /**
Signature · 1 line
2.43.0
#6 lore raw

Re: [PATCH v2 2/7] thermal/core: Add thresholds support

Quoted 72 lines · Daniel Lezcano

The trip points are a firmware description of the temperature limits of a specific thermal zone where we associate an action which is done by the kernel. The time resolution is low.

The userspace has to deal with a more complex thermal management based on heuristics from different information coming from different places. The logic is much more complex but based on a bigger time resolution, usually one second based.

The purpose of the userspace is to monitor the temperatures from different places and take actions. However, it can not be constantly reading the temperature to detect when a temperature threshold has been reached. This is especially bad for mobile or embedded system as that will lead to an unacceptable number of wakeup to check the temperature with nothing to do.

On the other side, the sensors are now most of the time interrupt driven. That means the thermal framework will use the temperature trip points to program the sensor to trigger an interrupt when a temperature limit is crossed.

Unfortunately, the userspace can not benefit this feature and current solutions found here and there, iow out-of-tree, are to add fake trip points in the firmware and enable the writable trip points.

This is bad for different reasons, the trip points are for in-kernel actions, the semantic of their types is used by the thermal framework and by adding trip points in the device tree is a way to overcome the current limitation but tampering with how the thermal framework is supposed to work. The writable trip points is a way to adjust a temperature limit given a specific platform if the firmware is not accurate enough and TBH it is more a debug feature from my POV.

The thresholds mechanism is a way to have the userspace to tell thermal framework to send a notification when a temperature limit is crossed. There is no id, no hysteresis, just the temperature and the direction of the limit crossing. That means we can be notified when a threshold is crossed the way up only, or the way down only or both ways. That allows to create hysteresis values if it is needed.

A threshold can be added, deleted or flushed. The latter means all thresholds belonging to a thermal zone will be deleted.

When a threshold is added:

- if the same threshold (temperature and direction) exists, an error is returned

- if a threshold is specified with the same temperature but a different direction, the specified direction is added

- if there is no threshold with the same temperature then it is created

When a threshold is deleted:

- if the same threshold (temperature and direction) exists, it is deleted

- if a threshold is specified with the same temperature but a different direction, the specified direction is removed

- if there is no threshold with the same temperature, then an error is returned

When the threshold are flushed:

- All thresholds related to a thermal zone are deleted

When a threshold is crossed:

Quoted 8 lines

- the userspace does not need to know which threshold(s) have been crossed, it will be notified with the current temperature and the previous temperature

- if multiple thresholds have been crossed between two updates only one notification will be send to the userspace, it is pointless to send a notification per thresholds crossed as the userspace can handle that easily when it has the temperature delta information

The above seems to be an exact copy of the first part of the cover letter.

Quoted 25 lines
---
 drivers/thermal/Kconfig              |  15 ++
 drivers/thermal/Makefile             |   3 +
 drivers/thermal/thermal_core.h       |   4 +
 drivers/thermal/thermal_thresholds.c | 241 +++++++++++++++++++++++++++
 drivers/thermal/thermal_thresholds.h |  57 +++++++
 include/linux/thermal.h              |   3 +
 6 files changed, 323 insertions(+)
 create mode 100644 drivers/thermal/thermal_thresholds.c
 create mode 100644 drivers/thermal/thermal_thresholds.h
drivers/thermal/Kconfig +5 -0
index ed16897584b4..84f9643678d6 100644
@@ -40,6 +40,21 @@ config THERMAL_DEBUGFS
          Say Y to allow the thermal subsystem to collect diagnostic
          information that can be accessed via debugfs.
 
+config THERMAL_THRESHOLDS
+       bool "Thermal thresholds notification mechanism"
+       depends on THERMAL_NETLINK
+       help
+         The userspace implements thermal engines which needs to get

drivers/thermal/Kconfig config THERMAL_DEBUGFS · line 57

Quoted 10 lines
drivers/thermal/Kconfig +10 -0
+         notified when temperature thresholds are crossed the way up
+         and down. These notification allow them to analyze the
+         thermal situation of the platform and take decision to
+         fulfill specific thermal profile like 'balanced',
+         'performance' or 'power saving'. In addition, the
+         temperature of the skin sensor is very important in this
+         case and must be monitored as well.
+
+         If in doubt, say Y
+

I'm not sure if this needs an additional user-selectable Kconfig option. It is not modular anyway and not so big, and distros don't like user-selectable Kconfig options.

Quoted 30 lines
drivers/thermal/Kconfig +0 -0
 config THERMAL_EMERGENCY_POWEROFF_DELAY_MS
        int "Emergency poweroff delay in milli-seconds"
        default 0
drivers/thermal/Makefile +3 -0
index ce7a4752ef52..3b991b1a7db4 100644
@@ -7,6 +7,9 @@ obj-$(CONFIG_THERMAL)           += thermal_sys.o
 thermal_sys-y                  += thermal_core.o thermal_sysfs.o
 thermal_sys-y                  += thermal_trip.o thermal_helpers.o
 
+# thermal thresholds
+thermal_sys-$(CONFIG_THERMAL_THRESHOLDS)       += thermal_thresholds.o
+
 # netlink interface to manage the thermal framework
 thermal_sys-$(CONFIG_THERMAL_NETLINK)          += thermal_netlink.o
 
drivers/thermal/thermal_core.h +1 -0
index 67a09f90eb95..0742c0f03d46 100644
@@ -13,6 +13,7 @@
 #include <linux/thermal.h>
 
 #include "thermal_netlink.h"
+#include "thermal_thresholds.h"
 #include "thermal_debugfs.h"
 
 struct thermal_trip_desc {

drivers/thermal/thermal_core.h struct thermal_zone_device { · line 138

Quoted 6 lines
drivers/thermal/thermal_core.h +3 -0
@@ -132,6 +133,9 @@ struct thermal_zone_device {
        bool resuming;
 #ifdef CONFIG_THERMAL_DEBUGFS
        struct thermal_debugfs *debugfs;
+#endif
+#ifdef CONFIG_THERMAL_THRESHOLDS
+       struct thresholds *thresholds;

Why does it need to be a pointer?

I would just use a plain struct list_head for it anyway.

Also, as stated in my reply to the cover letter, I would prefer to clearly distinguish these thresholds from trip thresholds, so I would call this user_thresholds.

Quoted 12 lines
drivers/thermal/thermal_core.h +0 -0
 #endif
        struct thermal_trip_desc trips[] __counted_by(num_trips);
 };
drivers/thermal/thermal_thresholds.c new file +3 -0
new file mode 100644
index 000000000000..0241b468cfbd
@@ -0,0 +1,241 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright 2024 Linaro Limited

drivers/thermal/thermal_thresholds.c line 13

Quoted 10 lines
drivers/thermal/thermal_thresholds.c new file +10 -0
+ *
+ * Author: Daniel Lezcano <[email protected]>
+ *
+ * Thermal thresholds
+ */
+#include <linux/list.h>
+#include <linux/list_sort.h>
+#include <linux/slab.h>
+
+#include "thermal_core.h"

+#include "thermal_thresholds.h"

drivers/thermal/thermal_thresholds.c line 17

Quoted 4 lines
drivers/thermal/thermal_thresholds.c new file +4 -0
+
+struct thresholds {
+       struct list_head list;
+};

This duplicates the definition in the header file.

Besides, why is the wrapper struct needed?

Quoted 4 lines
drivers/thermal/thermal_thresholds.c new file +4 -0
+
+int thermal_thresholds_init(struct thermal_zone_device *tz)
+{
+       struct thresholds *thresholds;

drivers/thermal/thermal_thresholds.c line 31

Quoted 10 lines
drivers/thermal/thermal_thresholds.c new file +10 -0
+
+       thresholds = kmalloc(sizeof(*thresholds), GFP_KERNEL);
+       if (!thresholds)
+               return -ENOMEM;
+
+       INIT_LIST_HEAD(&thresholds->list);
+       tz->thresholds = thresholds;
+
+       return 0;
+}

I'd rather embed "thresholds" in struct thermal_zone_device and avoid allocating memory separately for it. Less code, less complexity.

Quoted 40 lines
drivers/thermal/thermal_thresholds.c new file +40 -0
+
+void thermal_thresholds_exit(struct thermal_zone_device *tz)
+{
+       thermal_thresholds_flush(tz);
+       kfree(tz->thresholds);
+       tz->thresholds = NULL;
+}
+
+static int __thermal_thresholds_cmp(void *data,
+                                   const struct list_head *l1,
+                                   const struct list_head *l2)
+{
+       struct threshold *t1 = container_of(l1, struct threshold, list);
+       struct threshold *t2 = container_of(l2, struct threshold, list);
+
+       return t1->temperature - t2->temperature;
+}
+
+static struct threshold *__thermal_thresholds_find(const struct thresholds *thresholds, int temperature)
+{
+       struct threshold *t;
+
+       list_for_each_entry(t, &thresholds->list, list)
+               if (t->temperature == temperature)
+                       return t;
+
+       return NULL;
+}
+
+static bool __thermal_threshold_is_crossed(struct threshold *threshold, int temperature,
+                                          int last_temperature, int direction,
+                                          int *low, int *high)
+{
+       if (temperature > threshold->temperature && threshold->temperature > *low &&
+           (THERMAL_THRESHOLD_WAY_DOWN & threshold->direction))
+               *low = threshold->temperature;
+
+       if (temperature < threshold->temperature && threshold->temperature < *high &&
+           (THERMAL_THRESHOLD_WAY_UP & threshold->direction))
+               *high = threshold->temperature;

drivers/thermal/thermal_thresholds.c line 81

Quoted 10 lines
drivers/thermal/thermal_thresholds.c new file +10 -0
+
+       if (temperature < threshold->temperature &&
+           last_temperature >= threshold->temperature &&
+           (threshold->direction & direction))
+               return true;
+
+       if (temperature >= threshold->temperature &&
+           last_temperature < threshold->temperature &&
+           (threshold->direction & direction))
+               return true;

I would combine the checks, so something like this

if (temperature >= threshold->temperature) {
        if (threshold->temperature > *low &&
THERMAL_THRESHOLD_WAY_DOWN & threshold->direction)
                *low = threshold->temperature;

        if (last_temperature < threshold->temperature &&
threshold->direction & direction)
                return true;
} else {
        if (threshold->temperature < *high && THERMAL_THRESHOLD_WAY_UP
& threshold->direction)
              *high = threshold->temperature;

        if (last_temperature >= threshold->temperature &&
threshold->direction & direction)
                return true;
}
Quoted 36 lines
drivers/thermal/thermal_thresholds.c new file +36 -0
+
+       return false;
+}
+
+static bool thermal_thresholds_handle_raising(struct thresholds *thresholds, int temperature,
+                                             int last_temperature, int *low, int *high)
+{
+       struct threshold *t;
+
+       list_for_each_entry(t, &thresholds->list, list) {
+               if (__thermal_threshold_is_crossed(t, temperature, last_temperature,
+                                                  THERMAL_THRESHOLD_WAY_UP, low, high))
+                       return true;
+       }
+
+       return false;
+}
+
+static bool thermal_thresholds_handle_dropping(struct thresholds *thresholds, int temperature,
+                                              int last_temperature, int *low, int *high)
+{
+       struct threshold *t;
+
+       list_for_each_entry_reverse(t, &thresholds->list, list) {
+               if (__thermal_threshold_is_crossed(t, temperature, last_temperature,
+                                                  THERMAL_THRESHOLD_WAY_DOWN, low, high))
+                       return true;
+       }
+
+       return false;
+}
+
+void thermal_thresholds_flush(struct thermal_zone_device *tz)
+{
+       struct thresholds *thresholds = tz->thresholds;
+       struct threshold *entry, *tmp;

drivers/thermal/thermal_thresholds.c line 127

Quoted 10 lines
drivers/thermal/thermal_thresholds.c new file +10 -0
+
+       lockdep_assert_held(&tz->lock);
+
+       list_for_each_entry_safe(entry, tmp, &thresholds->list, list) {
+               list_del(&entry->list);
+               kfree(entry);
+       }
+
+       __thermal_zone_device_update(tz, THERMAL_THRESHOLD_FLUSHED);
+}

I'd move the function above before thermal_thresholds_exit() which uses it. Having it here is somewhat confusing.

drivers/thermal/thermal_thresholds.c line 129

Quoted 2 lines
drivers/thermal/thermal_thresholds.c new file +2 -0
+
+int thermal_thresholds_handle(struct thermal_zone_device *tz, int *low, int *high)

This function doesn't return anything other than 0 AFAICS. Make it void?

Quoted 4 lines
drivers/thermal/thermal_thresholds.c new file +4 -0
+{
+       struct thresholds *thresholds = tz->thresholds;
+
+       int temperature = tz->temperature;

drivers/thermal/thermal_thresholds.c line 143

Quoted 10 lines
drivers/thermal/thermal_thresholds.c new file +10 -0
+       int last_temperature = tz->last_temperature;
+       bool notify;
+
+       lockdep_assert_held(&tz->lock);
+
+       /*
+        * We need a second update in order to detect a threshold being crossed
+        */
+       if (last_temperature == THERMAL_TEMP_INVALID)
+               return 0;

So user space won't get notified when tz->temperature is above some thresholds the first time this runs. Fair enough, but won't they be confused by subsequent notifications that will not cover some thresholds?

Quoted 20 lines
drivers/thermal/thermal_thresholds.c new file +20 -0
+
+       /*
+        * The temperature is stable, so obviously we can not have
+        * crossed a threshold.
+        */
+       if (last_temperature == temperature)
+               return 0;
+
+       /*
+        * Since last update the temperature:
+        * - increased : thresholds are crossed the way up
+        * - decreased : thresholds are crossed the way down
+        */
+       if (temperature > last_temperature)
+               notify = thermal_thresholds_handle_raising(thresholds, temperature,
+                                                          last_temperature, low, high);
+       else
+               notify = thermal_thresholds_handle_dropping(thresholds, temperature,
+                                                           last_temperature, low, high);
+

drivers/thermal/thermal_thresholds.c line 173

Quoted 10 lines
drivers/thermal/thermal_thresholds.c new file +10 -0
+       if (notify)
+               pr_debug("A threshold has been crossed the way %s, with a temperature=%d, last_temperature=%d\n",
+                        temperature > last_temperature ? "up" : "down", temperature, last_temperature);
+
+       return 0;
+}
+
+int thermal_thresholds_add(struct thermal_zone_device *tz, int temperature, int direction)
+{
+       struct thresholds *thresholds = tz->thresholds;

So IMO it would be cleaner to just put "thresholds" into struct thermal_zone_device directly as a struct list_head because the above wouldn't be needed then.

drivers/thermal/thermal_thresholds.c line 181

Quoted 8 lines
drivers/thermal/thermal_thresholds.c new file +8 -0
+       struct threshold *t;
+
+       lockdep_assert_held(&tz->lock);
+
+       t = __thermal_thresholds_find(thresholds, temperature);
+       if (t) {
+               if (t->direction == direction)
+                       return -EEXIST;

Why is it useful to return an error here?

drivers/thermal/thermal_thresholds.c line 194

Quoted 13 lines
drivers/thermal/thermal_thresholds.c new file +13 -0
+
+               t->direction |= direction;
+       } else {
+
+               t = kmalloc(sizeof(*t), GFP_KERNEL);
+               if (!t)
+                       return -ENOMEM;
+
+               INIT_LIST_HEAD(&t->list);
+               t->temperature = temperature;
+               t->direction = direction;
+               list_add(&t->list, &thresholds->list);
+               list_sort(NULL, &thresholds->list, __thermal_thresholds_cmp);

And the above would become

+ list_add(&t->list, &tz->thresholds);
+ list_sort(NULL, &tz->thresholdst, __thermal_thresholds_cmp);

And analogously below.

Quoted 53 lines
drivers/thermal/thermal_thresholds.c new file +47 -0
+       }
+
+       __thermal_zone_device_update(tz, THERMAL_THRESHOLD_ADDED);
+
+       return 0;
+}
+
+int thermal_thresholds_delete(struct thermal_zone_device *tz, int temperature, int direction)
+{
+       struct thresholds *thresholds = tz->thresholds;
+       struct threshold *t;
+
+       lockdep_assert_held(&tz->lock);
+
+       t = __thermal_thresholds_find(thresholds, temperature);
+       if (!t)
+               return -ENOENT;
+
+       if (t->direction == direction) {
+               list_del(&t->list);
+               kfree(t);
+       } else {
+               t->direction &= ~direction;
+       }
+
+       __thermal_zone_device_update(tz, THERMAL_THRESHOLD_DELETED);
+
+       return 0;
+}
+
+int thermal_thresholds_for_each(struct thermal_zone_device *tz,
+                               int (*cb)(struct threshold *, void *arg), void *arg)
+{
+       struct thresholds *thresholds = tz->thresholds;
+       struct threshold *entry;
+       int ret;
+
+       lockdep_assert_held(&tz->lock);
+
+       list_for_each_entry(entry, &thresholds->list, list) {
+               ret = cb(entry, arg);
+               if (ret)
+                       return ret;
+       }
+
+       return 0;
+}
drivers/thermal/thermal_thresholds.h new file +0 -0
new file mode 100644
index 000000000000..7c8ce150d6d0

drivers/thermal/thermal_thresholds.h line 10

Quoted 10 lines
drivers/thermal/thermal_thresholds.h new file +10 -0
@@ -0,0 +1,57 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+
+#define THERMAL_THRESHOLD_WAY_UP   BIT(0)
+#define THERMAL_THRESHOLD_WAY_DOWN BIT(1)
+
+struct threshold {
+       int temperature;
+       int direction;
+       struct list_head list;
+};

IMO it would be better to put the list field at the top for better alignment and such. I would also call it something like list_node.

And I'd call this struct user_threshold as a whole (as per the previous remarks about the naming).

Quoted 52 lines
drivers/thermal/thermal_thresholds.h new file +47 -0
+
+#ifdef CONFIG_THERMAL_THRESHOLDS
+int thermal_thresholds_init(struct thermal_zone_device *tz);
+void thermal_thresholds_exit(struct thermal_zone_device *tz);
+void thermal_thresholds_flush(struct thermal_zone_device *tz);
+int thermal_thresholds_add(struct thermal_zone_device *tz, int temperature, int direction);
+int thermal_thresholds_delete(struct thermal_zone_device *tz, int temperature, int direction);
+int thermal_thresholds_handle(struct thermal_zone_device *tz, int *low, int *high);
+int thermal_thresholds_for_each(struct thermal_zone_device *tz,
+                               int (*cb)(struct threshold *, void *arg), void *arg);
+#else
+static inline int thermal_thresholds_init(struct thermal_zone_device *tz)
+{
+       return 0;
+}
+
+static inline void thermal_thresholds_exit(struct thermal_zone_device *tz)
+{
+       ;
+}
+
+static inline void thermal_thresholds_flush(struct thermal_zone_device *tz)
+{
+       ;
+}
+
+static inline int thermal_thresholds_add(struct thermal_zone_device *tz, int temperature, int direction)
+{
+       return 0;
+}
+
+static inline int thermal_thresholds_delete(struct thermal_zone_device *tz, int temperature, int direction)
+{
+       return 0;
+}
+
+static inline int thermal_thresholds_handle(struct thermal_zone_device *tz, int *low, int *high)
+{
+       return 0;
+}
+
+static inline int thermal_thresholds_for_each(struct thermal_zone_device *tz,
+                                             int (*cb)(struct threshold *, void *arg), void *arg)
+{
+       return 0;
+}
+#endif
include/linux/thermal.h +0 -0
index 25fbf960b474..bf0b4a8218f6 100644

include/linux/thermal.h enum thermal_notify_event { · line 57

Quoted 3 lines
include/linux/thermal.h +0 -0
@@ -55,6 +55,9 @@ enum thermal_notify_event {
        THERMAL_TZ_BIND_CDEV, /* Cooling dev is bind to the thermal zone */
        THERMAL_TZ_UNBIND_CDEV, /* Cooling dev is unbind from the thermal zone */
        THERMAL_INSTANCE_WEIGHT_CHANGED, /* Thermal instance weight changed */

There is an additional item here in the mainline (THERMAL_TZ_RESUME).

include/linux/thermal.h enum thermal_notify_event { · line 60

Quoted 3 lines
include/linux/thermal.h +3 -0
+       THERMAL_THRESHOLD_ADDED, /* Threshold added */
+       THERMAL_THRESHOLD_DELETED, /* Threshold deleted */
+       THERMAL_THRESHOLD_FLUSHED, /* All thresholds deleted */

I'd add "TZ" to these names, eg. THERMAL_TZ_THRESHOLD_ADDED, or even THERMAL_TZ_ADD_THRESHOLD in analogy with the cdev events above.

And THERMAL_TZ_FLUSH_THRESHOLDS sounds more like proper English to me. ;-)

Quoted 4 lines
include/linux/thermal.h +0 -1
 };
 
 /**
--
#7 lore raw

Re: [PATCH v2 2/7] thermal/core: Add thresholds support

Quoted 3 lines · Rafael J. Wysocki

On Fri, Aug 16, 2024 at 10:12 AM Daniel Lezcano
<[email protected]> wrote:

[cut]

Quoted 19 lines
Quoted 6 lines
drivers/thermal/thermal_thresholds.c new file +3 -0
@@ -0,0 +1,241 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright 2024 Linaro Limited

drivers/thermal/thermal_thresholds.c line 13

Quoted 10 lines
drivers/thermal/thermal_thresholds.c new file +10 -0
+ *
+ * Author: Daniel Lezcano <[email protected]>
+ *
+ * Thermal thresholds
+ */
+#include <linux/list.h>
+#include <linux/list_sort.h>
+#include <linux/slab.h>
+
+#include "thermal_core.h"

+#include "thermal_thresholds.h"

Quoted 8 lines

drivers/thermal/thermal_thresholds.c line 17

Quoted 4 lines
drivers/thermal/thermal_thresholds.c new file +4 -0
+
+struct thresholds {
+       struct list_head list;
+};

This duplicates the definition in the header file.

Besides, why is the wrapper struct needed?

On second thought, it can hold a pointer to the first threshold that was strictly above the zone temperature when thermal_thresholds_handle() ran last time, so something like:

struct thresholds {
      struct list_head list;
      struct user_threhold *first_above;
};

and first_above == NULL would mean that the zone temperature was above all of the threshold or at the highest one.

Then thermal_thresholds_handle() could do something like:

tr = tz->user_thresholds.first_above;

if (tr && tz->temperature >= tr->temperature) {
    do {
        if (tr->direction & THERMAL_THRESHOLD_WAY_UP)
            notify = true;

        if (tr->list_node.next != &tz->user_thresholds.list) {
            tr = list_next_entry(tr, list_node);
        else
            tr = NULL;
    } while (tr && tz->temperature >= tr->temperature);
} else {
    if (!tr)
        tr = list_last_entry(&tz->user_thresholds.list,
                     struct user_threshold, list_node);

    while (tz->temperature < tr->temperature)
        if (tr->direction & THERMAL_THRESHOLD_WAY_DOWN)
            notify = true;

        if (tr->list_node.prev != &tz->user_thresholds.list) {
            tr = list_prev_entry(tr, list_node);
        else
            break;
    }
    if (tz->temperature >= tr->temperature)
        tr = NULL;
}

tz->user_thresholds.first_above = tr;

which is a bit simpler than the code in the current patch.

Quoted 80 lines
Quoted 4 lines
drivers/thermal/thermal_thresholds.c new file +4 -0
+
+int thermal_thresholds_init(struct thermal_zone_device *tz)
+{
+       struct thresholds *thresholds;

drivers/thermal/thermal_thresholds.c line 31

Quoted 10 lines
drivers/thermal/thermal_thresholds.c new file +10 -0
+
+       thresholds = kmalloc(sizeof(*thresholds), GFP_KERNEL);
+       if (!thresholds)
+               return -ENOMEM;
+
+       INIT_LIST_HEAD(&thresholds->list);
+       tz->thresholds = thresholds;
+
+       return 0;
+}

I'd rather embed "thresholds" in struct thermal_zone_device and avoid allocating memory separately for it. Less code, less complexity.

Quoted 40 lines
drivers/thermal/thermal_thresholds.c new file +40 -0
+
+void thermal_thresholds_exit(struct thermal_zone_device *tz)
+{
+       thermal_thresholds_flush(tz);
+       kfree(tz->thresholds);
+       tz->thresholds = NULL;
+}
+
+static int __thermal_thresholds_cmp(void *data,
+                                   const struct list_head *l1,
+                                   const struct list_head *l2)
+{
+       struct threshold *t1 = container_of(l1, struct threshold, list);
+       struct threshold *t2 = container_of(l2, struct threshold, list);
+
+       return t1->temperature - t2->temperature;
+}
+
+static struct threshold *__thermal_thresholds_find(const struct thresholds *thresholds, int temperature)
+{
+       struct threshold *t;
+
+       list_for_each_entry(t, &thresholds->list, list)
+               if (t->temperature == temperature)
+                       return t;
+
+       return NULL;
+}
+
+static bool __thermal_threshold_is_crossed(struct threshold *threshold, int temperature,
+                                          int last_temperature, int direction,
+                                          int *low, int *high)
+{
+       if (temperature > threshold->temperature && threshold->temperature > *low &&
+           (THERMAL_THRESHOLD_WAY_DOWN & threshold->direction))
+               *low = threshold->temperature;
+
+       if (temperature < threshold->temperature && threshold->temperature < *high &&
+           (THERMAL_THRESHOLD_WAY_UP & threshold->direction))
+               *high = threshold->temperature;

drivers/thermal/thermal_thresholds.c line 81

Quoted 10 lines
drivers/thermal/thermal_thresholds.c new file +10 -0
+
+       if (temperature < threshold->temperature &&
+           last_temperature >= threshold->temperature &&
+           (threshold->direction & direction))
+               return true;
+
+       if (temperature >= threshold->temperature &&
+           last_temperature < threshold->temperature &&
+           (threshold->direction & direction))
+               return true;

I would combine the checks, so something like this

if (temperature >= threshold->temperature) {
        if (threshold->temperature > *low &&
THERMAL_THRESHOLD_WAY_DOWN & threshold->direction)
                *low = threshold->temperature;

        if (last_temperature < threshold->temperature &&
threshold->direction & direction)
                return true;
} else {
Quoted 8 lines
        if (threshold->temperature < *high && THERMAL_THRESHOLD_WAY_UP
& threshold->direction)
              *high = threshold->temperature;

        if (last_temperature >= threshold->temperature &&
threshold->direction & direction)
                return true;
}

Also, I'm not sure why "high" and "low" are needed at all.

The current and last zone temperature could be included in the notification just fine and user space should be able to figure out which thresholds are affected.

Quoted 4 lines
Quoted 4 lines
drivers/thermal/thermal_thresholds.c new file +4 -0
+
+       return false;
+}
+
#8 lore raw

Re: [PATCH v2 2/7] thermal/core: Add thresholds support

Quoted 78 lines · Rafael J. Wysocki
Quoted 3 lines · Rafael J. Wysocki

On Fri, Aug 16, 2024 at 10:12 AM Daniel Lezcano
<[email protected]> wrote:

[cut]

Quoted 19 lines
Quoted 6 lines
drivers/thermal/thermal_thresholds.c new file +3 -0
@@ -0,0 +1,241 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright 2024 Linaro Limited

drivers/thermal/thermal_thresholds.c line 13

Quoted 10 lines
drivers/thermal/thermal_thresholds.c new file +10 -0
+ *
+ * Author: Daniel Lezcano <[email protected]>
+ *
+ * Thermal thresholds
+ */
+#include <linux/list.h>
+#include <linux/list_sort.h>
+#include <linux/slab.h>
+
+#include "thermal_core.h"

+#include "thermal_thresholds.h"

Quoted 8 lines

drivers/thermal/thermal_thresholds.c line 17

Quoted 4 lines
drivers/thermal/thermal_thresholds.c new file +4 -0
+
+struct thresholds {
+       struct list_head list;
+};

This duplicates the definition in the header file.

Besides, why is the wrapper struct needed?

On second thought, it can hold a pointer to the first threshold that was strictly above the zone temperature when thermal_thresholds_handle() ran last time, so something like:

struct thresholds {
       struct list_head list;
       struct user_threhold *first_above;
};

and first_above == NULL would mean that the zone temperature was above all of the threshold or at the highest one.

Then thermal_thresholds_handle() could do something like:

tr = tz->user_thresholds.first_above;

if (tr && tz->temperature >= tr->temperature) {
     do {
         if (tr->direction & THERMAL_THRESHOLD_WAY_UP)
             notify = true;

         if (tr->list_node.next != &tz->user_thresholds.list) {
             tr = list_next_entry(tr, list_node);
         else
             tr = NULL;
     } while (tr && tz->temperature >= tr->temperature);
} else {
     if (!tr)
         tr = list_last_entry(&tz->user_thresholds.list,
                      struct user_threshold, list_node);

     while (tz->temperature < tr->temperature)
         if (tr->direction & THERMAL_THRESHOLD_WAY_DOWN)
             notify = true;

         if (tr->list_node.prev != &tz->user_thresholds.list) {
             tr = list_prev_entry(tr, list_node);
         else
             break;
     }
     if (tz->temperature >= tr->temperature)
         tr = NULL;
}
Quoted 3 lines

tz->user_thresholds.first_above = tr;

which is a bit simpler than the code in the current patch.

TBH, it is probably a matter of taste but I don't see how it is simpler, especially if someone wants to to understand or add something in the code. I would prefer to keep the initial routine which was already tested.

[ ... ]

Quoted 6 lines

Also, I'm not sure why "high" and "low" are needed at all.

The current and last zone temperature could be included in the notification just fine and user space should be able to figure out which thresholds are affected.

I'm not sure to get the question. high and low are needed because the thermal core will call set_trips(). The trips and the thresholds are combined to find out the high and low temperature to program the driver to trigger an interrupt.

Signature · 5 lines
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs

Follow Linaro:  <http://www.facebook.com/pages/Linaro> Facebook |
<http://twitter.com/#!/linaroorg> Twitter |
<http://www.linaro.org/linaro-blog/> Blog
#9 lore raw

Re: [PATCH v2 2/7] thermal/core: Add thresholds support

In reply to Rafael J. Wysocki
Quoted 78 lines · Rafael J. Wysocki
Quoted 76 lines · Daniel Lezcano

The trip points are a firmware description of the temperature limits of a specific thermal zone where we associate an action which is done by the kernel. The time resolution is low.

The userspace has to deal with a more complex thermal management based on heuristics from different information coming from different places. The logic is much more complex but based on a bigger time resolution, usually one second based.

The purpose of the userspace is to monitor the temperatures from different places and take actions. However, it can not be constantly reading the temperature to detect when a temperature threshold has been reached. This is especially bad for mobile or embedded system as that will lead to an unacceptable number of wakeup to check the temperature with nothing to do.

On the other side, the sensors are now most of the time interrupt driven. That means the thermal framework will use the temperature trip points to program the sensor to trigger an interrupt when a temperature limit is crossed.

Unfortunately, the userspace can not benefit this feature and current solutions found here and there, iow out-of-tree, are to add fake trip points in the firmware and enable the writable trip points.

This is bad for different reasons, the trip points are for in-kernel actions, the semantic of their types is used by the thermal framework and by adding trip points in the device tree is a way to overcome the current limitation but tampering with how the thermal framework is supposed to work. The writable trip points is a way to adjust a temperature limit given a specific platform if the firmware is not accurate enough and TBH it is more a debug feature from my POV.

The thresholds mechanism is a way to have the userspace to tell thermal framework to send a notification when a temperature limit is crossed. There is no id, no hysteresis, just the temperature and the direction of the limit crossing. That means we can be notified when a threshold is crossed the way up only, or the way down only or both ways. That allows to create hysteresis values if it is needed.

A threshold can be added, deleted or flushed. The latter means all thresholds belonging to a thermal zone will be deleted.

When a threshold is added:

- if the same threshold (temperature and direction) exists, an error is returned

- if a threshold is specified with the same temperature but a different direction, the specified direction is added

- if there is no threshold with the same temperature then it is created

When a threshold is deleted:

- if the same threshold (temperature and direction) exists, it is deleted

- if a threshold is specified with the same temperature but a different direction, the specified direction is removed

- if there is no threshold with the same temperature, then an error is returned

When the threshold are flushed:

- All thresholds related to a thermal zone are deleted

When a threshold is crossed:

- the userspace does not need to know which threshold(s) have been crossed, it will be notified with the current temperature and the previous temperature

Quoted 6 lines
Quoted 4 lines

- if multiple thresholds have been crossed between two updates only one notification will be send to the userspace, it is pointless to send a notification per thresholds crossed as the userspace can handle that easily when it has the temperature delta information

The above seems to be an exact copy of the first part of the cover letter.

Yes, it is done on purpose as it is the commit bringing the feature. I thought it is convenient for the developer to read the description of the commit introducing the feature.

[ ... ]

Quoted 11 lines
Quoted 11 lines
drivers/thermal/Kconfig +11 -0
+config THERMAL_THRESHOLDS
+       bool "Thermal thresholds notification mechanism"
+       depends on THERMAL_NETLINK
+       help
+         The userspace implements thermal engines which needs to get
+         notified when temperature thresholds are crossed the way up
+         and down. These notification allow them to analyze the
+         thermal situation of the platform and take decision to
+         fulfill specific thermal profile like 'balanced',
+         'performance' or 'power saving'. In addition, the
+         temperature of the skin sensor is very important in this
Quoted 8 lines

drivers/thermal/Kconfig config THERMAL_DEBUGFS · line 57

Quoted 4 lines
drivers/thermal/Kconfig +4 -0
+         case and must be monitored as well.
+
+         If in doubt, say Y
+

I'm not sure if this needs an additional user-selectable Kconfig option. It is not modular anyway and not so big, and distros don't like user-selectable Kconfig options.

Ok, I can drop the user selectable option.

[ ... ]

Quoted 37 lines

drivers/thermal/thermal_core.h struct thermal_zone_device { · line 138

Quoted 3 lines
drivers/thermal/thermal_core.h +3 -0
+#endif
+#ifdef CONFIG_THERMAL_THRESHOLDS
+       struct thresholds *thresholds;

Why does it need to be a pointer?

I would just use a plain struct list_head for it anyway.

Also, as stated in my reply to the cover letter, I would prefer to clearly distinguish these thresholds from trip thresholds, so I would call this user_thresholds.

Quoted 12 lines
  #endif
         struct thermal_trip_desc trips[] __counted_by(num_trips);
  };
drivers/thermal/thermal_thresholds.c new file +3 -0
new file mode 100644
index 000000000000..0241b468cfbd
@@ -0,0 +1,241 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright 2024 Linaro Limited

drivers/thermal/thermal_thresholds.c line 13

Quoted 10 lines
drivers/thermal/thermal_thresholds.c new file +10 -0
+ *
+ * Author: Daniel Lezcano <[email protected]>
+ *
+ * Thermal thresholds
+ */
+#include <linux/list.h>
+#include <linux/list_sort.h>
+#include <linux/slab.h>
+
+#include "thermal_core.h"

+#include "thermal_thresholds.h"

Quoted 6 lines

drivers/thermal/thermal_thresholds.c line 17

Quoted 4 lines
drivers/thermal/thermal_thresholds.c new file +4 -0
+
+struct thresholds {
+       struct list_head list;
+};

This duplicates the definition in the header file.

No actually it is plural.

struct threshold;
struct thresholds;

Quoted 1 line

Besides, why is the wrapper struct needed?

Because I think the threshold will continue to evolve. So instead of dealing with lists, we use a dedicated structure. In case we add new fields or we change the list by something else the functions prototypes are untouched. It is my way of coding in order to always set the scene for future changes.

Quoted 8 lines
Quoted 8 lines
drivers/thermal/thermal_thresholds.c new file +8 -0
+int thermal_thresholds_init(struct thermal_zone_device *tz)
+{
+       struct thresholds *thresholds;
+
+       thresholds = kmalloc(sizeof(*thresholds), GFP_KERNEL);
+       if (!thresholds)
+               return -ENOMEM;
+
Quoted 8 lines

drivers/thermal/thermal_thresholds.c line 31

Quoted 5 lines
drivers/thermal/thermal_thresholds.c new file +5 -0
+       INIT_LIST_HEAD(&thresholds->list);
+       tz->thresholds = thresholds;
+
+       return 0;
+}

I'd rather embed "thresholds" in struct thermal_zone_device and avoid allocating memory separately for it. Less code, less complexity.

Ok

Quoted 24 lines
Quoted 11 lines
drivers/thermal/thermal_thresholds.c new file +11 -0
+static bool __thermal_threshold_is_crossed(struct threshold *threshold, int temperature,
+                                          int last_temperature, int direction,
+                                          int *low, int *high)
+{
+       if (temperature > threshold->temperature && threshold->temperature > *low &&
+           (THERMAL_THRESHOLD_WAY_DOWN & threshold->direction))
+               *low = threshold->temperature;
+
+       if (temperature < threshold->temperature && threshold->temperature < *high &&
+           (THERMAL_THRESHOLD_WAY_UP & threshold->direction))
+               *high = threshold->temperature;

drivers/thermal/thermal_thresholds.c line 81

Quoted 10 lines
drivers/thermal/thermal_thresholds.c new file +10 -0
+
+       if (temperature < threshold->temperature &&
+           last_temperature >= threshold->temperature &&
+           (threshold->direction & direction))
+               return true;
+
+       if (temperature >= threshold->temperature &&
+           last_temperature < threshold->temperature &&
+           (threshold->direction & direction))
+               return true;

I would combine the checks, so something like this

Quoted 16 lines
if (temperature >= threshold->temperature) {
         if (threshold->temperature > *low &&
THERMAL_THRESHOLD_WAY_DOWN & threshold->direction)
                 *low = threshold->temperature;
         if (last_temperature < threshold->temperature &&
threshold->direction & direction)
                 return true;
} else {
         if (threshold->temperature < *high && THERMAL_THRESHOLD_WAY_UP
& threshold->direction)
               *high = threshold->temperature;

         if (last_temperature >= threshold->temperature &&
threshold->direction & direction)
                 return true;
}

Ok

[ ... ]

Quoted 9 lines
Quoted 9 lines
drivers/thermal/thermal_thresholds.c new file +9 -0
+void thermal_thresholds_flush(struct thermal_zone_device *tz)
+{
+       struct thresholds *thresholds = tz->thresholds;
+       struct threshold *entry, *tmp;
+
+       lockdep_assert_held(&tz->lock);
+
+       list_for_each_entry_safe(entry, tmp, &thresholds->list, list) {
+               list_del(&entry->list);
Quoted 8 lines

drivers/thermal/thermal_thresholds.c line 127

Quoted 5 lines
drivers/thermal/thermal_thresholds.c new file +5 -0
+               kfree(entry);
+       }
+
+       __thermal_zone_device_update(tz, THERMAL_THRESHOLD_FLUSHED);
+}

I'd move the function above before thermal_thresholds_exit() which uses it. Having it here is somewhat confusing.

Sure

Quoted 4 lines

drivers/thermal/thermal_thresholds.c line 67

Quoted 2 lines
drivers/thermal/thermal_thresholds.c new file +2 -0
+
+int thermal_thresholds_handle(struct thermal_zone_device *tz, int *low, int *high)

This function doesn't return anything other than 0 AFAICS. Make it void?

Ok

Quoted 11 lines
Quoted 11 lines
drivers/thermal/thermal_thresholds.c new file +11 -0
+{
+       struct thresholds *thresholds = tz->thresholds;
+
+       int temperature = tz->temperature;
+       int last_temperature = tz->last_temperature;
+       bool notify;
+
+       lockdep_assert_held(&tz->lock);
+
+       /*
+        * We need a second update in order to detect a threshold being crossed
Quoted 8 lines

drivers/thermal/thermal_thresholds.c line 143

Quoted 3 lines
drivers/thermal/thermal_thresholds.c new file +3 -0
+        */
+       if (last_temperature == THERMAL_TEMP_INVALID)
+               return 0;

So user space won't get notified when tz->temperature is above some thresholds the first time this runs. Fair enough, but won't they be confused by subsequent notifications that will not cover some thresholds?

It is unlikely to happen.

When the thermal zone is created and enabled, it updates the temperature of the thermal zone so tz->temperature is no longer THERMAL_TEMP_INVALID.

At this step, there is no userspace set yet.

Assuming there is zero update until we create a threshold. When this one is created then the thermal zone is updated, the temperature is read, the tz->last_temperature = tz->temperature and tz->temperature has the new value. Then handle_thresholds is called and finally set_trips

If the temperature is above or below a threshold then it is detected and notified because tz->last_temperature is not equal to THERMAL_TEMP_INVALID.

Others situations are IMO resulting from a bogus driver/sensor and should be handled at the sensor level, not the core code.

[ ... ]

Quoted 7 lines

drivers/thermal/thermal_thresholds.c line 173

Quoted 3 lines
drivers/thermal/thermal_thresholds.c new file +3 -0
+int thermal_thresholds_add(struct thermal_zone_device *tz, int temperature, int direction)
+{
+       struct thresholds *thresholds = tz->thresholds;

So IMO it would be cleaner to just put "thresholds" into struct thermal_zone_device directly as a struct list_head because the above wouldn't be needed then.

I can change that to list directly but if we add anything (which can probably happen) then the API change as well as anything below again.

Quoted 10 lines

drivers/thermal/thermal_thresholds.c line 181

Quoted 8 lines
drivers/thermal/thermal_thresholds.c new file +8 -0
+       struct threshold *t;
+
+       lockdep_assert_held(&tz->lock);
+
+       t = __thermal_thresholds_find(thresholds, temperature);
+       if (t) {
+               if (t->direction == direction)
+                       return -EEXIST;

Why is it useful to return an error here?

I was expecting this comment :)

We have the choice between :
* we do nothing
* we return an error

Let's assume the userspace is misbehaving and because of an internal bug of a thermal engine it creates multiple times the same threshold (eg. index not incremented, etc ...). If the kernel reports nothing, then the user space will never detect this problem. If it reports the error the user space can choose to ignore it or follow it up.

The kernel is strict and it is up to the user space to ignore it or not.

Quoted 27 lines

drivers/thermal/thermal_thresholds.c line 194

Quoted 13 lines
drivers/thermal/thermal_thresholds.c new file +13 -0
+
+               t->direction |= direction;
+       } else {
+
+               t = kmalloc(sizeof(*t), GFP_KERNEL);
+               if (!t)
+                       return -ENOMEM;
+
+               INIT_LIST_HEAD(&t->list);
+               t->temperature = temperature;
+               t->direction = direction;
+               list_add(&t->list, &thresholds->list);
+               list_sort(NULL, &thresholds->list, __thermal_thresholds_cmp);

And the above would become

(quoted diff) +2 -0
+        list_add(&t->list, &tz->thresholds);
+        list_sort(NULL, &tz->thresholdst, __thermal_thresholds_cmp);

And analogously below.

Quoted 6 lines
drivers/thermal/thermal_thresholds.c new file +6 -0
+       }
+
+       __thermal_zone_device_update(tz, THERMAL_THRESHOLD_ADDED);
+
+       return 0;
+}

[ ... ]

Quoted 11 lines

drivers/thermal/thermal_thresholds.c

Quoted 5 lines
drivers/thermal/thermal_thresholds.c new file +5 -0
+struct threshold {
+       int temperature;
+       int direction;
+       struct list_head list;
+};

IMO it would be better to put the list field at the top for better alignment and such. I would also call it something like list_node.

And I'd call this struct user_threshold as a whole (as per the previous remarks about the naming).

Ok

[ ... ]

Quoted 7 lines

drivers/thermal/thermal_thresholds.c enum thermal_notify_event { · line 57

Quoted 4 lines
drivers/thermal/thermal_thresholds.c new file +0 -0
@@ -55,6 +55,9 @@ enum thermal_notify_event {
         THERMAL_TZ_BIND_CDEV, /* Cooling dev is bind to the thermal zone */
         THERMAL_TZ_UNBIND_CDEV, /* Cooling dev is unbind from the thermal zone */
         THERMAL_INSTANCE_WEIGHT_CHANGED, /* Thermal instance weight changed */

There is an additional item here in the mainline (THERMAL_TZ_RESUME).

Quoted 8 lines

drivers/thermal/thermal_thresholds.c enum thermal_notify_event { · line 60

Quoted 3 lines
drivers/thermal/thermal_thresholds.c new file +3 -0
+       THERMAL_THRESHOLD_ADDED, /* Threshold added */
+       THERMAL_THRESHOLD_DELETED, /* Threshold deleted */
+       THERMAL_THRESHOLD_FLUSHED, /* All thresholds deleted */

I'd add "TZ" to these names, eg. THERMAL_TZ_THRESHOLD_ADDED, or even THERMAL_TZ_ADD_THRESHOLD in analogy with the cdev events above.

And THERMAL_TZ_FLUSH_THRESHOLDS sounds more like proper English to me. ;-)

Ok, will do the change

Signature · 5 lines
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs

Follow Linaro:  <http://www.facebook.com/pages/Linaro> Facebook |
<http://twitter.com/#!/linaroorg> Twitter |
<http://www.linaro.org/linaro-blog/> Blog
#10 lore raw

Re: [PATCH v2 2/7] thermal/core: Add thresholds support

Quoted 85 lines · Daniel Lezcano
Quoted 83 lines · Rafael J. Wysocki
Quoted 80 lines · Daniel Lezcano

The trip points are a firmware description of the temperature limits of a specific thermal zone where we associate an action which is done by the kernel. The time resolution is low.

The userspace has to deal with a more complex thermal management based on heuristics from different information coming from different places. The logic is much more complex but based on a bigger time resolution, usually one second based.

The purpose of the userspace is to monitor the temperatures from different places and take actions. However, it can not be constantly reading the temperature to detect when a temperature threshold has been reached. This is especially bad for mobile or embedded system as that will lead to an unacceptable number of wakeup to check the temperature with nothing to do.

On the other side, the sensors are now most of the time interrupt driven. That means the thermal framework will use the temperature trip points to program the sensor to trigger an interrupt when a temperature limit is crossed.

Unfortunately, the userspace can not benefit this feature and current solutions found here and there, iow out-of-tree, are to add fake trip points in the firmware and enable the writable trip points.

This is bad for different reasons, the trip points are for in-kernel actions, the semantic of their types is used by the thermal framework and by adding trip points in the device tree is a way to overcome the current limitation but tampering with how the thermal framework is supposed to work. The writable trip points is a way to adjust a temperature limit given a specific platform if the firmware is not accurate enough and TBH it is more a debug feature from my POV.

The thresholds mechanism is a way to have the userspace to tell thermal framework to send a notification when a temperature limit is crossed. There is no id, no hysteresis, just the temperature and the direction of the limit crossing. That means we can be notified when a threshold is crossed the way up only, or the way down only or both ways. That allows to create hysteresis values if it is needed.

A threshold can be added, deleted or flushed. The latter means all thresholds belonging to a thermal zone will be deleted.

When a threshold is added:

- if the same threshold (temperature and direction) exists, an error is returned

- if a threshold is specified with the same temperature but a different direction, the specified direction is added

- if there is no threshold with the same temperature then it is created

When a threshold is deleted:

- if the same threshold (temperature and direction) exists, it is deleted

- if a threshold is specified with the same temperature but a different direction, the specified direction is removed

- if there is no threshold with the same temperature, then an error is returned

When the threshold are flushed:

- All thresholds related to a thermal zone are deleted

When a threshold is crossed:

- the userspace does not need to know which threshold(s) have been crossed, it will be notified with the current temperature and the previous temperature

- if multiple thresholds have been crossed between two updates only one notification will be send to the userspace, it is pointless to send a notification per thresholds crossed as the userspace can handle that easily when it has the temperature delta information

Quoted 5 lines
Quoted 1 line

The above seems to be an exact copy of the first part of the cover letter.

Yes, it is done on purpose as it is the commit bringing the feature. I thought it is convenient for the developer to read the description of the commit introducing the feature.

I agree, but then you could just say "refer to the changelog of the first patch for details" in the cover letter.

Quoted 70 lines

[ ... ]

Quoted 11 lines
Quoted 11 lines
drivers/thermal/Kconfig +11 -0
+config THERMAL_THRESHOLDS
+       bool "Thermal thresholds notification mechanism"
+       depends on THERMAL_NETLINK
+       help
+         The userspace implements thermal engines which needs to get
+         notified when temperature thresholds are crossed the way up
+         and down. These notification allow them to analyze the
+         thermal situation of the platform and take decision to
+         fulfill specific thermal profile like 'balanced',
+         'performance' or 'power saving'. In addition, the
+         temperature of the skin sensor is very important in this
Quoted 8 lines

drivers/thermal/Kconfig config THERMAL_DEBUGFS · line 57

Quoted 4 lines
drivers/thermal/Kconfig +4 -0
+         case and must be monitored as well.
+
+         If in doubt, say Y
+

I'm not sure if this needs an additional user-selectable Kconfig option. It is not modular anyway and not so big, and distros don't like user-selectable Kconfig options.

Ok, I can drop the user selectable option.

[ ... ]

Quoted 37 lines

drivers/thermal/thermal_core.h struct thermal_zone_device { · line 138

Quoted 3 lines
drivers/thermal/thermal_core.h +3 -0
+#endif
+#ifdef CONFIG_THERMAL_THRESHOLDS
+       struct thresholds *thresholds;

Why does it need to be a pointer?

I would just use a plain struct list_head for it anyway.

Also, as stated in my reply to the cover letter, I would prefer to clearly distinguish these thresholds from trip thresholds, so I would call this user_thresholds.

Quoted 12 lines
  #endif
         struct thermal_trip_desc trips[] __counted_by(num_trips);
  };
drivers/thermal/thermal_thresholds.c new file +3 -0
new file mode 100644
index 000000000000..0241b468cfbd
@@ -0,0 +1,241 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright 2024 Linaro Limited

drivers/thermal/thermal_thresholds.c line 13

Quoted 10 lines
drivers/thermal/thermal_thresholds.c new file +10 -0
+ *
+ * Author: Daniel Lezcano <[email protected]>
+ *
+ * Thermal thresholds
+ */
+#include <linux/list.h>
+#include <linux/list_sort.h>
+#include <linux/slab.h>
+
+#include "thermal_core.h"

+#include "thermal_thresholds.h"

Quoted 6 lines

drivers/thermal/thermal_thresholds.c line 17

Quoted 4 lines
drivers/thermal/thermal_thresholds.c new file +4 -0
+
+struct thresholds {
+       struct list_head list;
+};

This duplicates the definition in the header file.

Quoted 4 lines

No actually it is plural.

struct threshold;
struct thresholds;

I saw the first one, but for some reason I thought the other was there too, sorry.

But this means that it is missing from thermal_core.h where it is used in the struct thermal_zone_device definition.

Quoted 3 lines
Quoted 1 line

Besides, why is the wrapper struct needed?

Because I think the threshold will continue to evolve.

Fair enough, like I said here:

https://lore.kernel.org/linux-pm/CAJZ5v0h=DgBSiFbdmnzSFjEJd6sdBffCODspxmM-G92FN2HGiA@mail.gmail.com/

Quoted 145 lines

So instead of
dealing with lists, we use a dedicated structure. In case we add new fields or we change the list by something else the functions prototypes are untouched. It is my way of coding in order to always set the scene for future changes.

Quoted 8 lines
Quoted 8 lines
drivers/thermal/thermal_thresholds.c new file +8 -0
+int thermal_thresholds_init(struct thermal_zone_device *tz)
+{
+       struct thresholds *thresholds;
+
+       thresholds = kmalloc(sizeof(*thresholds), GFP_KERNEL);
+       if (!thresholds)
+               return -ENOMEM;
+
Quoted 8 lines

drivers/thermal/thermal_thresholds.c line 31

Quoted 5 lines
drivers/thermal/thermal_thresholds.c new file +5 -0
+       INIT_LIST_HEAD(&thresholds->list);
+       tz->thresholds = thresholds;
+
+       return 0;
+}

I'd rather embed "thresholds" in struct thermal_zone_device and avoid allocating memory separately for it. Less code, less complexity.

Ok

Quoted 24 lines
Quoted 11 lines
drivers/thermal/thermal_thresholds.c new file +11 -0
+static bool __thermal_threshold_is_crossed(struct threshold *threshold, int temperature,
+                                          int last_temperature, int direction,
+                                          int *low, int *high)
+{
+       if (temperature > threshold->temperature && threshold->temperature > *low &&
+           (THERMAL_THRESHOLD_WAY_DOWN & threshold->direction))
+               *low = threshold->temperature;
+
+       if (temperature < threshold->temperature && threshold->temperature < *high &&
+           (THERMAL_THRESHOLD_WAY_UP & threshold->direction))
+               *high = threshold->temperature;

drivers/thermal/thermal_thresholds.c line 81

Quoted 10 lines
drivers/thermal/thermal_thresholds.c new file +10 -0
+
+       if (temperature < threshold->temperature &&
+           last_temperature >= threshold->temperature &&
+           (threshold->direction & direction))
+               return true;
+
+       if (temperature >= threshold->temperature &&
+           last_temperature < threshold->temperature &&
+           (threshold->direction & direction))
+               return true;

I would combine the checks, so something like this

Quoted 16 lines
if (temperature >= threshold->temperature) {
         if (threshold->temperature > *low &&
THERMAL_THRESHOLD_WAY_DOWN & threshold->direction)
                 *low = threshold->temperature;
         if (last_temperature < threshold->temperature &&
threshold->direction & direction)
                 return true;
} else {
         if (threshold->temperature < *high && THERMAL_THRESHOLD_WAY_UP
& threshold->direction)
               *high = threshold->temperature;

         if (last_temperature >= threshold->temperature &&
threshold->direction & direction)
                 return true;
}

Ok

[ ... ]

Quoted 9 lines
Quoted 9 lines
drivers/thermal/thermal_thresholds.c new file +9 -0
+void thermal_thresholds_flush(struct thermal_zone_device *tz)
+{
+       struct thresholds *thresholds = tz->thresholds;
+       struct threshold *entry, *tmp;
+
+       lockdep_assert_held(&tz->lock);
+
+       list_for_each_entry_safe(entry, tmp, &thresholds->list, list) {
+               list_del(&entry->list);
Quoted 8 lines

drivers/thermal/thermal_thresholds.c line 127

Quoted 5 lines
drivers/thermal/thermal_thresholds.c new file +5 -0
+               kfree(entry);
+       }
+
+       __thermal_zone_device_update(tz, THERMAL_THRESHOLD_FLUSHED);
+}

I'd move the function above before thermal_thresholds_exit() which uses it. Having it here is somewhat confusing.

Sure

Quoted 4 lines

drivers/thermal/thermal_thresholds.c line 67

Quoted 2 lines
drivers/thermal/thermal_thresholds.c new file +2 -0
+
+int thermal_thresholds_handle(struct thermal_zone_device *tz, int *low, int *high)

This function doesn't return anything other than 0 AFAICS. Make it void?

Ok

Quoted 11 lines
Quoted 11 lines
drivers/thermal/thermal_thresholds.c new file +11 -0
+{
+       struct thresholds *thresholds = tz->thresholds;
+
+       int temperature = tz->temperature;
+       int last_temperature = tz->last_temperature;
+       bool notify;
+
+       lockdep_assert_held(&tz->lock);
+
+       /*
+        * We need a second update in order to detect a threshold being crossed
Quoted 8 lines

drivers/thermal/thermal_thresholds.c line 143

Quoted 3 lines
drivers/thermal/thermal_thresholds.c new file +3 -0
+        */
+       if (last_temperature == THERMAL_TEMP_INVALID)
+               return 0;

So user space won't get notified when tz->temperature is above some thresholds the first time this runs. Fair enough, but won't they be confused by subsequent notifications that will not cover some thresholds?

It is unlikely to happen.

When the thermal zone is created and enabled, it updates the temperature of the thermal zone so tz->temperature is no longer THERMAL_TEMP_INVALID.

At this step, there is no userspace set yet.

Assuming there is zero update until we create a threshold. When this one is created then the thermal zone is updated, the temperature is read, the tz->last_temperature = tz->temperature and tz->temperature has the new value. Then handle_thresholds is called and finally set_trips

If the temperature is above or below a threshold then it is detected and notified because tz->last_temperature is not equal to THERMAL_TEMP_INVALID.

Others situations are IMO resulting from a bogus driver/sensor and should be handled at the sensor level, not the core code.

[ ... ]

Quoted 7 lines

drivers/thermal/thermal_thresholds.c line 173

Quoted 3 lines
drivers/thermal/thermal_thresholds.c new file +3 -0
+int thermal_thresholds_add(struct thermal_zone_device *tz, int temperature, int direction)
+{
+       struct thresholds *thresholds = tz->thresholds;

So IMO it would be cleaner to just put "thresholds" into struct thermal_zone_device directly as a struct list_head because the above wouldn't be needed then.

Quoted 2 lines

I can change that to list directly but if we add anything (which can probably happen) then the API change as well as anything below again.

But it can be

 #ifdef CONFIG_THERMAL_DEBUGFS
     struct thermal_debugfs *debugfs;
 #endif
+   struct thresholds thresholds;
     struct thermal_trip_desc trips[] __counted_by(num_trips);

};

Quoted 17 lines
Quoted 10 lines

drivers/thermal/thermal_thresholds.c line 181

Quoted 8 lines
drivers/thermal/thermal_thresholds.c new file +8 -0
+       struct threshold *t;
+
+       lockdep_assert_held(&tz->lock);
+
+       t = __thermal_thresholds_find(thresholds, temperature);
+       if (t) {
+               if (t->direction == direction)
+                       return -EEXIST;

Why is it useful to return an error here?

I was expecting this comment :)

We have the choice between :
* we do nothing
* we return an error

Quoted 7 lines

Let's assume the userspace is misbehaving and because of an internal bug of a thermal engine it creates multiple times the same threshold (eg. index not incremented, etc ...). If the kernel reports nothing, then the user space will never detect this problem. If it reports the error the user space can choose to ignore it or follow it up.

The kernel is strict and it is up to the user space to ignore it or not.

Well, I'm not sure.

IMO it will just cause user space code to be more complex in some valid use cases, but whatever.

#11 lore raw

PATCH v2 3/7 thermal/core: Connect the threshold with the core

In reply to Daniel Lezcano

Initialize, de-initialize and handle the threshold in the same place than the trip points.

---
 drivers/thermal/thermal_core.c | 9 +++++++++
 1 file changed, 9 insertions(+)
drivers/thermal/thermal_core.c +9 -0
index 166f48071487..0a4d7f6e5cf5 100644
@@ -591,6 +591,8 @@ void __thermal_zone_device_update(struct thermal_zone_device *tz,
 			high = td->threshold;
 	}
 
+	thermal_thresholds_handle(tz, &low, &high);
+
 	thermal_zone_set_trips(tz, low, high);
 
 	list_sort(NULL, &way_up_list, thermal_trip_notify_cmp);
@@ -1561,6 +1563,10 @@ thermal_zone_device_register_with_trips(const char *type,
 			goto unregister;
 	}
 
+	result = thermal_thresholds_init(tz);
+	if (result)
+		goto remove_hwmon;
+
 	mutex_lock(&thermal_list_lock);
 	mutex_lock(&tz->lock);
 	list_add_tail(&tz->node, &thermal_tz_list);
@@ -1581,6 +1587,8 @@ thermal_zone_device_register_with_trips(const char *type,
 
 	return tz;
 
+remove_hwmon:
+	thermal_remove_hwmon_sysfs(tz);
 unregister:
 	device_del(&tz->device);
 release_device:
@@ -1669,6 +1677,7 @@ void thermal_zone_device_unregister(struct thermal_zone_device *tz)
 
 	thermal_set_governor(tz, NULL);
 
+	thermal_thresholds_exit(tz);
 	thermal_remove_hwmon_sysfs(tz);
 	ida_free(&thermal_tz_ida, tz->id);
 	ida_destroy(&tz->ida);
Signature · 1 line
2.43.0
#12 lore raw

PATCH v2 4/7 thermal/netlink: Add the commands and the events for the thresholds

In reply to Daniel Lezcano

The thresholds exist but there is no notification neither action code related to them yet.

These changes implement the netlink for the notifications when the thresholds are crossed, added, deleted or flushed as well as the commands which allows to get the list of the thresholds, flush them, add and delete.

As different processes in userspace can interact with the thresholds, the process id responsible of the action (add, delete or flush) will be added in the notification. This way a thermal engine is able to detect if another process is interfering with the thresholds. A process id of zero is the kernel as it is by convention usually.

---
 drivers/thermal/thermal_netlink.c    | 239 ++++++++++++++++++++++++++-
 drivers/thermal/thermal_netlink.h    |   7 +
 drivers/thermal/thermal_thresholds.c |  36 ++--
 drivers/thermal/thermal_thresholds.h |  15 +-
 include/uapi/linux/thermal.h         |  30 +++-
 5 files changed, 290 insertions(+), 37 deletions(-)
drivers/thermal/thermal_netlink.c +233 -6
index 97157c453630..3febb119a7f0 100644
@@ -9,6 +9,7 @@
 #include <linux/module.h>
 #include <linux/notifier.h>
 #include <linux/kernel.h>
+#include <net/sock.h>
 #include <net/genetlink.h>
 #include <uapi/linux/thermal.h>
 
@@ -49,12 +50,19 @@ static const struct nla_policy thermal_genl_policy[THERMAL_GENL_ATTR_MAX + 1] =
 	[THERMAL_GENL_ATTR_CPU_CAPABILITY_ID]		= { .type = NLA_U32 },
 	[THERMAL_GENL_ATTR_CPU_CAPABILITY_PERFORMANCE]	= { .type = NLA_U32 },
 	[THERMAL_GENL_ATTR_CPU_CAPABILITY_EFFICIENCY]	= { .type = NLA_U32 },
+
+	/* Thresholds */
+	[THERMAL_GENL_ATTR_THRESHOLD]		= { .type = NLA_NESTED },
+	[THERMAL_GENL_ATTR_THRESHOLD_TEMP]	= { .type = NLA_U32 },
+	[THERMAL_GENL_ATTR_THRESHOLD_WAY]	= { .type = NLA_U32 },
+	[THERMAL_GENL_ATTR_THRESHOLD_PID]	= { .type = NLA_U32 },
 };
 
 struct param {
 	struct nlattr **attrs;
 	struct sk_buff *msg;
 	const char *name;
+	pid_t pid;
 	int tz_id;
 	int cdev_id;
 	int trip_id;
@@ -62,6 +70,8 @@ struct param {
 	int trip_type;
 	int trip_hyst;
 	int temp;
+	int last_temp;
+	int direction;
 	int cdev_state;
 	int cdev_max_state;
 	struct thermal_genl_cpu_caps *cpu_capabilities;
@@ -234,6 +244,36 @@ static int thermal_genl_event_cpu_capability_change(struct param *p)
 	return -EMSGSIZE;
 }
 
+static int thermal_genl_event_threshold_add(struct param *p)
+{
+	if (nla_put_u32(p->msg, THERMAL_GENL_ATTR_TZ_ID, p->tz_id) ||
+	    nla_put_u32(p->msg, THERMAL_GENL_ATTR_THRESHOLD_TEMP, p->temp) ||
+	    nla_put_u32(p->msg, THERMAL_GENL_ATTR_THRESHOLD_WAY, p->direction) ||
+	    nla_put_u32(p->msg, THERMAL_GENL_ATTR_THRESHOLD_PID, p->pid))
+		return -EMSGSIZE;
+
+	return 0;
+}
+
+static int thermal_genl_event_threshold_flush(struct param *p)
+{
+	if (nla_put_u32(p->msg, THERMAL_GENL_ATTR_TZ_ID, p->tz_id),
+	    nla_put_u32(p->msg, THERMAL_GENL_ATTR_THRESHOLD_PID, p->pid))
+		return -EMSGSIZE;
+
+	return 0;
+}
+
+static int thermal_genl_event_threshold_up(struct param *p)
+{
+	if (nla_put_u32(p->msg, THERMAL_GENL_ATTR_TZ_ID, p->tz_id) ||
+	    nla_put_u32(p->msg, THERMAL_GENL_ATTR_TZ_LAST_TEMP, p->last_temp) ||
+	    nla_put_u32(p->msg, THERMAL_GENL_ATTR_TZ_TEMP, p->temp))
+		return -EMSGSIZE;
+
+	return 0;
+}
+
 int thermal_genl_event_tz_delete(struct param *p)
 	__attribute__((alias("thermal_genl_event_tz")));
 
@@ -246,6 +286,12 @@ int thermal_genl_event_tz_disable(struct param *p)
 int thermal_genl_event_tz_trip_down(struct param *p)
 	__attribute__((alias("thermal_genl_event_tz_trip_up")));
 
+int thermal_genl_event_threshold_delete(struct param *p)
+	__attribute__((alias("thermal_genl_event_threshold_add")));
+
+int thermal_genl_event_threshold_down(struct param *p)
+	__attribute__((alias("thermal_genl_event_threshold_up")));
+
 static cb_t event_cb[] = {
 	[THERMAL_GENL_EVENT_TZ_CREATE]		= thermal_genl_event_tz_create,
 	[THERMAL_GENL_EVENT_TZ_DELETE]		= thermal_genl_event_tz_delete,
@@ -259,6 +305,11 @@ static cb_t event_cb[] = {
 	[THERMAL_GENL_EVENT_CDEV_STATE_UPDATE]	= thermal_genl_event_cdev_state_update,
 	[THERMAL_GENL_EVENT_TZ_GOV_CHANGE]	= thermal_genl_event_gov_change,
 	[THERMAL_GENL_EVENT_CPU_CAPABILITY_CHANGE] = thermal_genl_event_cpu_capability_change,
+	[THERMAL_GENL_EVENT_THRESHOLD_ADD]	= thermal_genl_event_threshold_add,
+	[THERMAL_GENL_EVENT_THRESHOLD_DELETE]	= thermal_genl_event_threshold_delete,
+	[THERMAL_GENL_EVENT_THRESHOLD_FLUSH]	= thermal_genl_event_threshold_flush,
+	[THERMAL_GENL_EVENT_THRESHOLD_DOWN]	= thermal_genl_event_threshold_down,
+	[THERMAL_GENL_EVENT_THRESHOLD_UP]	= thermal_genl_event_threshold_up,
 };
 
 /*
@@ -401,6 +452,43 @@ int thermal_genl_cpu_capability_event(int count,
 }
 EXPORT_SYMBOL_GPL(thermal_genl_cpu_capability_event);
 
+int thermal_notify_threshold_add(const struct thermal_zone_device *tz,
+				 int temperature, int direction, int pid)
+{
+	struct param p = { .tz_id = tz->id, .temp = temperature, .direction = direction, .pid = pid };
+
+	return thermal_genl_send_event(THERMAL_GENL_EVENT_THRESHOLD_ADD, &p);
+}
+
+int thermal_notify_threshold_delete(const struct thermal_zone_device *tz,
+				    int temperature, int direction, int pid)
+{
+	struct param p = { .tz_id = tz->id, .temp = temperature, .direction = direction, .pid = pid };
+
+	return thermal_genl_send_event(THERMAL_GENL_EVENT_THRESHOLD_DELETE, &p);
+}
+
+int thermal_notify_threshold_flush(const struct thermal_zone_device *tz, int pid)
+{
+	struct param p = { .tz_id = tz->id, .pid = pid };
+
+	return thermal_genl_send_event(THERMAL_GENL_EVENT_THRESHOLD_FLUSH, &p);
+}
+
+int thermal_notify_threshold_down(const struct thermal_zone_device *tz)
+{
+	struct param p = { .tz_id = tz->id, .temp = tz->temperature, .last_temp = tz->last_temperature };
+
+	return thermal_genl_send_event(THERMAL_GENL_EVENT_THRESHOLD_DOWN, &p);
+}
+
+int thermal_notify_threshold_up(const struct thermal_zone_device *tz)
+{
+	struct param p = { .tz_id = tz->id, .temp = tz->temperature, .last_temp = tz->last_temperature };
+
+	return thermal_genl_send_event(THERMAL_GENL_EVENT_THRESHOLD_UP, &p);
+}
+
 /*************************** Command encoding ********************************/
 
 static int __thermal_genl_cmd_tz_get_id(struct thermal_zone_device *tz,
@@ -575,12 +663,130 @@ static int thermal_genl_cmd_cdev_get(struct param *p)
 	return ret;
 }
 
+static int __thermal_genl_cmd_threshold_get(struct threshold *threshold, void *arg)
+{
+	struct sk_buff *msg = arg;
+
+	if (nla_put_u32(msg, THERMAL_GENL_ATTR_THRESHOLD_TEMP, threshold->temperature) ||
+	    nla_put_u32(msg, THERMAL_GENL_ATTR_THRESHOLD_WAY, threshold->direction))
+		return -1;
+
+	return 0;
+}
+
+static int thermal_genl_cmd_threshold_get(struct param *p)
+{
+	struct thermal_zone_device *tz;
+	struct sk_buff *msg = p->msg;
+	struct nlattr *start_trip;
+	int id, ret;
+
+	if (!p->attrs[THERMAL_GENL_ATTR_TZ_ID])
+		return -EINVAL;
+
+	id = nla_get_u32(p->attrs[THERMAL_GENL_ATTR_TZ_ID]);
+
+	tz = thermal_zone_get_by_id(id);
+	if (!tz)
+		return -EINVAL;
+
+	start_trip = nla_nest_start(msg, THERMAL_GENL_ATTR_THRESHOLD);
+	if (!start_trip)
+		return -EMSGSIZE;
+
+	mutex_lock(&tz->lock);
+	ret = thermal_thresholds_for_each(tz, __thermal_genl_cmd_threshold_get, msg);
+	mutex_unlock(&tz->lock);
+
+	if (ret)
+		return -EMSGSIZE;	
+
+	nla_nest_end(msg, start_trip);
+
+	return 0;
+}
+
+static int thermal_genl_cmd_threshold_add(struct param *p)
+{
+	struct thermal_zone_device *tz;
+	int id, temp, direction, ret = 0;
+
+	if (!p->attrs[THERMAL_GENL_ATTR_TZ_ID] ||
+	    !p->attrs[THERMAL_GENL_ATTR_THRESHOLD_TEMP] ||
+	    !p->attrs[THERMAL_GENL_ATTR_THRESHOLD_WAY])
+		return -EINVAL;
+
+	id = nla_get_u32(p->attrs[THERMAL_GENL_ATTR_TZ_ID]);
+	temp = nla_get_u32(p->attrs[THERMAL_GENL_ATTR_THRESHOLD_TEMP]);
+	direction = nla_get_u32(p->attrs[THERMAL_GENL_ATTR_THRESHOLD_WAY]);
+
+	tz = thermal_zone_get_by_id(id);
+	if (!tz)
+		return -EINVAL;
+
+	mutex_lock(&tz->lock);
+	ret = thermal_thresholds_add(tz, temp, direction, p->pid);
+	mutex_unlock(&tz->lock);
+
+	return ret;
+}
+
+static int thermal_genl_cmd_threshold_delete(struct param *p)
+{
+	struct thermal_zone_device *tz;
+	int id, temp, direction, ret = 0;
+
+	if (!p->attrs[THERMAL_GENL_ATTR_TZ_ID] ||
+	    !p->attrs[THERMAL_GENL_ATTR_THRESHOLD_TEMP] ||
+	    !p->attrs[THERMAL_GENL_ATTR_THRESHOLD_WAY])
+		return -EINVAL;
+
+	id = nla_get_u32(p->attrs[THERMAL_GENL_ATTR_TZ_ID]);
+	temp = nla_get_u32(p->attrs[THERMAL_GENL_ATTR_THRESHOLD_TEMP]);
+	direction = nla_get_u32(p->attrs[THERMAL_GENL_ATTR_THRESHOLD_WAY]);
+
+	tz = thermal_zone_get_by_id(id);
+	if (!tz)
+		return -EINVAL;
+
+	mutex_lock(&tz->lock);
+	ret = thermal_thresholds_delete(tz, temp, direction, p->pid);
+	mutex_unlock(&tz->lock);
+
+	return ret;
+}
+
+static int thermal_genl_cmd_threshold_flush(struct param *p)
+{
+	struct thermal_zone_device *tz;
+	int id;
+
+	if (!p->attrs[THERMAL_GENL_ATTR_TZ_ID])
+		return -EINVAL;
+
+	id = nla_get_u32(p->attrs[THERMAL_GENL_ATTR_TZ_ID]);
+
+	tz = thermal_zone_get_by_id(id);
+	if (!tz)
+		return -EINVAL;
+
+	mutex_lock(&tz->lock);
+	thermal_thresholds_flush(tz, p->pid);
+	mutex_unlock(&tz->lock);
+
+	return 0;
+}
+
 static cb_t cmd_cb[] = {
-	[THERMAL_GENL_CMD_TZ_GET_ID]	= thermal_genl_cmd_tz_get_id,
-	[THERMAL_GENL_CMD_TZ_GET_TRIP]	= thermal_genl_cmd_tz_get_trip,
-	[THERMAL_GENL_CMD_TZ_GET_TEMP]	= thermal_genl_cmd_tz_get_temp,
-	[THERMAL_GENL_CMD_TZ_GET_GOV]	= thermal_genl_cmd_tz_get_gov,
-	[THERMAL_GENL_CMD_CDEV_GET]	= thermal_genl_cmd_cdev_get,
+	[THERMAL_GENL_CMD_TZ_GET_ID]		= thermal_genl_cmd_tz_get_id,
+	[THERMAL_GENL_CMD_TZ_GET_TRIP]		= thermal_genl_cmd_tz_get_trip,
+	[THERMAL_GENL_CMD_TZ_GET_TEMP]		= thermal_genl_cmd_tz_get_temp,
+	[THERMAL_GENL_CMD_TZ_GET_GOV]		= thermal_genl_cmd_tz_get_gov,
+	[THERMAL_GENL_CMD_CDEV_GET]		= thermal_genl_cmd_cdev_get,
+	[THERMAL_GENL_CMD_THRESHOLD_GET]	= thermal_genl_cmd_threshold_get,
+	[THERMAL_GENL_CMD_THRESHOLD_ADD]	= thermal_genl_cmd_threshold_add,
+	[THERMAL_GENL_CMD_THRESHOLD_DELETE]	= thermal_genl_cmd_threshold_delete,
+	[THERMAL_GENL_CMD_THRESHOLD_FLUSH]	= thermal_genl_cmd_threshold_flush,
 };
 
 static int thermal_genl_cmd_dumpit(struct sk_buff *skb,
@@ -623,6 +829,7 @@ static int thermal_genl_cmd_doit(struct sk_buff *skb,
 	if (!msg)
 		return -ENOMEM;
 	p.msg = msg;
+	p.pid = task_tgid_vnr(current);
 
 	hdr = genlmsg_put_reply(msg, info, &thermal_genl_family, 0, cmd);
 	if (!hdr)
@@ -691,6 +898,26 @@ static const struct genl_small_ops thermal_genl_ops[] = {
 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 		.dumpit = thermal_genl_cmd_dumpit,
 	},
+	{
+		.cmd = THERMAL_GENL_CMD_THRESHOLD_GET,
+		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
+		.doit = thermal_genl_cmd_doit,
+	},
+	{
+		.cmd = THERMAL_GENL_CMD_THRESHOLD_ADD,
+		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
+		.doit = thermal_genl_cmd_doit,
+	},
+	{
+		.cmd = THERMAL_GENL_CMD_THRESHOLD_DELETE,
+		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
+		.doit = thermal_genl_cmd_doit,
+	},
+	{
+		.cmd = THERMAL_GENL_CMD_THRESHOLD_FLUSH,
+		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
+		.doit = thermal_genl_cmd_doit,
+	},
 };
 
 static struct genl_family thermal_genl_family __ro_after_init = {
@@ -703,7 +930,7 @@ static struct genl_family thermal_genl_family __ro_after_init = {
 	.unbind		= thermal_genl_unbind,
 	.small_ops	= thermal_genl_ops,
 	.n_small_ops	= ARRAY_SIZE(thermal_genl_ops),
-	.resv_start_op	= THERMAL_GENL_CMD_CDEV_GET + 1,
+	.resv_start_op	= __THERMAL_GENL_CMD_MAX,
 	.mcgrps		= thermal_genl_mcgrps,
 	.n_mcgrps	= ARRAY_SIZE(thermal_genl_mcgrps),
 };
drivers/thermal/thermal_netlink.h +7 -0
index e01221e8816b..715acee04f31 100644
@@ -53,6 +53,13 @@ int thermal_notify_tz_gov_change(const struct thermal_zone_device *tz,
 int thermal_genl_sampling_temp(int id, int temp);
 int thermal_genl_cpu_capability_event(int count,
 				      struct thermal_genl_cpu_caps *caps);
+int thermal_notify_threshold_add(const struct thermal_zone_device *tz,
+				 int temperature, int direction, pid_t pid);
+int thermal_notify_threshold_delete(const struct thermal_zone_device *tz,
+				    int temperature, int direction, pid_t pid);
+int thermal_notify_threshold_flush(const struct thermal_zone_device *tz, pid_t pid);
+int thermal_notify_threshold_down(const struct thermal_zone_device *tz);
+int thermal_notify_threshold_up(const struct thermal_zone_device *tz);
 #else
 static inline int thermal_netlink_init(void)
 {
drivers/thermal/thermal_thresholds.c +21 -15
index 0241b468cfbd..e0cfb4969203 100644
@@ -32,7 +32,7 @@ int thermal_thresholds_init(struct thermal_zone_device *tz)
 
 void thermal_thresholds_exit(struct thermal_zone_device *tz)
 {
-	thermal_thresholds_flush(tz);
+	thermal_thresholds_flush(tz, 0);
 	kfree(tz->thresholds);
 	tz->thresholds = NULL;
 }
@@ -111,7 +111,7 @@ static bool thermal_thresholds_handle_dropping(struct thresholds *thresholds, in
 	return false;
 }
 
-void thermal_thresholds_flush(struct thermal_zone_device *tz)
+void thermal_thresholds_flush(struct thermal_zone_device *tz, pid_t pid)
 {
 	struct thresholds *thresholds = tz->thresholds;
 	struct threshold *entry, *tmp;
@@ -123,6 +123,8 @@ void thermal_thresholds_flush(struct thermal_zone_device *tz)
 		kfree(entry);
 	}
 
+	thermal_notify_threshold_flush(tz, pid);
+
 	__thermal_zone_device_update(tz, THERMAL_THRESHOLD_FLUSHED);
 }
 
@@ -132,7 +134,6 @@ int thermal_thresholds_handle(struct thermal_zone_device *tz, int *low, int *hig
 
 	int temperature = tz->temperature;
 	int last_temperature = tz->last_temperature;
-	bool notify;
 
 	lockdep_assert_held(&tz->lock);
 
@@ -154,21 +155,21 @@ int thermal_thresholds_handle(struct thermal_zone_device *tz, int *low, int *hig
 	 * - increased : thresholds are crossed the way up
 	 * - decreased : thresholds are crossed the way down
 	 */
-	if (temperature > last_temperature)
-		notify = thermal_thresholds_handle_raising(thresholds, temperature,
-							   last_temperature, low, high);
-	else
-		notify = thermal_thresholds_handle_dropping(thresholds, temperature,
-							    last_temperature, low, high);
-
-	if (notify)
-		pr_debug("A threshold has been crossed the way %s, with a temperature=%d, last_temperature=%d\n",
-			 temperature > last_temperature ? "up" : "down", temperature, last_temperature);
+	if (temperature > last_temperature) {
+		if (thermal_thresholds_handle_raising(thresholds, temperature,
+						      last_temperature, low, high))
+			thermal_notify_threshold_up(tz);
+	} else {
+		if (thermal_thresholds_handle_dropping(thresholds, temperature,
+						       last_temperature, low, high))
+			thermal_notify_threshold_down(tz);
+	}
 
 	return 0;
 }
 
-int thermal_thresholds_add(struct thermal_zone_device *tz, int temperature, int direction)
+int thermal_thresholds_add(struct thermal_zone_device *tz,
+			   int temperature, int direction, pid_t pid)
 {
 	struct thresholds *thresholds = tz->thresholds;
 	struct threshold *t;
@@ -194,12 +195,15 @@ int thermal_thresholds_add(struct thermal_zone_device *tz, int temperature, int
 		list_sort(NULL, &thresholds->list, __thermal_thresholds_cmp);
 	}
 
+	thermal_notify_threshold_add(tz, temperature, direction, pid);
+
 	__thermal_zone_device_update(tz, THERMAL_THRESHOLD_ADDED);
 
 	return 0;
 }
 
-int thermal_thresholds_delete(struct thermal_zone_device *tz, int temperature, int direction)
+int thermal_thresholds_delete(struct thermal_zone_device *tz,
+			      int temperature, int direction, pid_t pid)
 {
 	struct thresholds *thresholds = tz->thresholds;
 	struct threshold *t;
@@ -219,6 +223,8 @@ int thermal_thresholds_delete(struct thermal_zone_device *tz, int temperature, i
 
 	__thermal_zone_device_update(tz, THERMAL_THRESHOLD_DELETED);
 
+	thermal_notify_threshold_delete(tz, temperature, direction, pid);
+
 	return 0;
 }
 
drivers/thermal/thermal_thresholds.h +6 -9
index 7c8ce150d6d0..b7a149b3400c 100644
@@ -1,8 +1,5 @@
 /* SPDX-License-Identifier: GPL-2.0 */
 
-#define THERMAL_THRESHOLD_WAY_UP   BIT(0)
-#define THERMAL_THRESHOLD_WAY_DOWN BIT(1)
-
 struct threshold {
 	int temperature;
 	int direction;
@@ -12,9 +9,9 @@ struct threshold {
 #ifdef CONFIG_THERMAL_THRESHOLDS
 int thermal_thresholds_init(struct thermal_zone_device *tz);
 void thermal_thresholds_exit(struct thermal_zone_device *tz);
-void thermal_thresholds_flush(struct thermal_zone_device *tz);
-int thermal_thresholds_add(struct thermal_zone_device *tz, int temperature, int direction);
-int thermal_thresholds_delete(struct thermal_zone_device *tz, int temperature, int direction);
+void thermal_thresholds_flush(struct thermal_zone_device *tz, pid_t pid);
+int thermal_thresholds_add(struct thermal_zone_device *tz, int temperature, int direction, pid_t pid);
+int thermal_thresholds_delete(struct thermal_zone_device *tz, int temperature, int direction, pid_t pid);
 int thermal_thresholds_handle(struct thermal_zone_device *tz, int *low, int *high);
 int thermal_thresholds_for_each(struct thermal_zone_device *tz,
 				int (*cb)(struct threshold *, void *arg), void *arg);
@@ -29,17 +26,17 @@ static inline void thermal_thresholds_exit(struct thermal_zone_device *tz)
 	;
 }
 
-static inline void thermal_thresholds_flush(struct thermal_zone_device *tz)
+static inline void thermal_thresholds_flush(struct thermal_zone_device *tz, pid_t pid)
 {
 	;
 }
 
-static inline int thermal_thresholds_add(struct thermal_zone_device *tz, int temperature, int direction)
+static inline int thermal_thresholds_add(struct thermal_zone_device *tz, int temperature, int direction, pid_t pid)
 {
 	return 0;
 }
 
-static inline int thermal_thresholds_delete(struct thermal_zone_device *tz, int temperature, int direction)
+static inline int thermal_thresholds_delete(struct thermal_zone_device *tz, int temperature, int direction, pid_t pid)
 {
 	return 0;
 }
include/uapi/linux/thermal.h +23 -7
index fc78bf3aead7..bcbaf62a1727 100644
@@ -3,6 +3,8 @@
 #define _UAPI_LINUX_THERMAL_H
 
 #define THERMAL_NAME_LENGTH	20
+#define THERMAL_THRESHOLD_WAY_UP	0x1
+#define THERMAL_THRESHOLD_WAY_DOWN	0x2
 
 enum thermal_device_mode {
 	THERMAL_DEVICE_DISABLED = 0,
@@ -18,7 +20,7 @@ enum thermal_trip_type {
 
 /* Adding event notification support elements */
 #define THERMAL_GENL_FAMILY_NAME		"thermal"
-#define THERMAL_GENL_VERSION			0x01
+#define THERMAL_GENL_VERSION			0x02
 #define THERMAL_GENL_SAMPLING_GROUP_NAME	"sampling"
 #define THERMAL_GENL_EVENT_GROUP_NAME		"event"
 
@@ -28,6 +30,7 @@ enum thermal_genl_attr {
 	THERMAL_GENL_ATTR_TZ,
 	THERMAL_GENL_ATTR_TZ_ID,
 	THERMAL_GENL_ATTR_TZ_TEMP,
+	THERMAL_GENL_ATTR_TZ_LAST_TEMP,
 	THERMAL_GENL_ATTR_TZ_TRIP,
 	THERMAL_GENL_ATTR_TZ_TRIP_ID,
 	THERMAL_GENL_ATTR_TZ_TRIP_TYPE,
@@ -48,6 +51,10 @@ enum thermal_genl_attr {
 	THERMAL_GENL_ATTR_CPU_CAPABILITY_ID,
 	THERMAL_GENL_ATTR_CPU_CAPABILITY_PERFORMANCE,
 	THERMAL_GENL_ATTR_CPU_CAPABILITY_EFFICIENCY,
+	THERMAL_GENL_ATTR_THRESHOLD,
+	THERMAL_GENL_ATTR_THRESHOLD_TEMP,
+	THERMAL_GENL_ATTR_THRESHOLD_WAY,
+	THERMAL_GENL_ATTR_THRESHOLD_PID,
 	__THERMAL_GENL_ATTR_MAX,
 };
 #define THERMAL_GENL_ATTR_MAX (__THERMAL_GENL_ATTR_MAX - 1)
@@ -75,6 +82,11 @@ enum thermal_genl_event {
 	THERMAL_GENL_EVENT_CDEV_STATE_UPDATE,	/* Cdev state updated */
 	THERMAL_GENL_EVENT_TZ_GOV_CHANGE,	/* Governor policy changed  */
 	THERMAL_GENL_EVENT_CPU_CAPABILITY_CHANGE,	/* CPU capability changed */
+	THERMAL_GENL_EVENT_THRESHOLD_ADD,	/* A thresold has been added */
+	THERMAL_GENL_EVENT_THRESHOLD_DELETE,	/* A thresold has been deleted */
+	THERMAL_GENL_EVENT_THRESHOLD_FLUSH,	/* All thresolds have been deleted */
+	THERMAL_GENL_EVENT_THRESHOLD_UP,	/* A thresold has been crossed the way up */
+	THERMAL_GENL_EVENT_THRESHOLD_DOWN,	/* A thresold has been crossed the way down */
 	__THERMAL_GENL_EVENT_MAX,
 };
 #define THERMAL_GENL_EVENT_MAX (__THERMAL_GENL_EVENT_MAX - 1)
@@ -82,12 +94,16 @@ enum thermal_genl_event {
 /* Commands supported by the thermal_genl_family */
 enum thermal_genl_cmd {
 	THERMAL_GENL_CMD_UNSPEC,
-	THERMAL_GENL_CMD_TZ_GET_ID,	/* List of thermal zones id */
-	THERMAL_GENL_CMD_TZ_GET_TRIP,	/* List of thermal trips */
-	THERMAL_GENL_CMD_TZ_GET_TEMP,	/* Get the thermal zone temperature */
-	THERMAL_GENL_CMD_TZ_GET_GOV,	/* Get the thermal zone governor */
-	THERMAL_GENL_CMD_TZ_GET_MODE,	/* Get the thermal zone mode */
-	THERMAL_GENL_CMD_CDEV_GET,	/* List of cdev id */
+	THERMAL_GENL_CMD_TZ_GET_ID,		/* List of thermal zones id */
+	THERMAL_GENL_CMD_TZ_GET_TRIP,		/* List of thermal trips */
+	THERMAL_GENL_CMD_TZ_GET_TEMP,		/* Get the thermal zone temperature */
+	THERMAL_GENL_CMD_TZ_GET_GOV,		/* Get the thermal zone governor */
+	THERMAL_GENL_CMD_TZ_GET_MODE,		/* Get the thermal zone mode */
+	THERMAL_GENL_CMD_CDEV_GET,		/* List of cdev id */
+	THERMAL_GENL_CMD_THRESHOLD_GET,		/* List of thresholds */
+	THERMAL_GENL_CMD_THRESHOLD_ADD,		/* Add a threshold */
+	THERMAL_GENL_CMD_THRESHOLD_DELETE,	/* Delete a threshold */
+	THERMAL_GENL_CMD_THRESHOLD_FLUSH,	/* Flush all the thresholds */
 	__THERMAL_GENL_CMD_MAX,
 };
 #define THERMAL_GENL_CMD_MAX (__THERMAL_GENL_CMD_MAX - 1)
Signature · 1 line
2.43.0
#13 lore raw

PATCH v2 5/7 tools/lib/thermal: Make more generic the command encoding function

In reply to Daniel Lezcano

The thermal netlink has been extended with more commands which require an encoding with more information. The generic encoding function puts the thermal zone id with the command name. It is the unique parameters.

The next changes will provide more parameters to the command. Set the scene for those new parameters by making the encoding function more generic.

---
 tools/lib/thermal/commands.c | 41 ++++++++++++++++++++++++++++--------
 1 file changed, 32 insertions(+), 9 deletions(-)
tools/lib/thermal/commands.c +32 -9
index 73d4d4e8d6ec..a9223df91dcf 100644
@@ -261,8 +261,23 @@ static struct genl_ops thermal_cmd_ops = {
 	.o_ncmds	= ARRAY_SIZE(thermal_cmds),
 };
 
-static thermal_error_t thermal_genl_auto(struct thermal_handler *th, int id, int cmd,
-					 int flags, void *arg)
+struct cmd_param {
+	int tz_id;
+};
+
+typedef int (*cmd_cb_t)(struct nl_msg *, struct cmd_param *);
+
+static int thermal_genl_tz_id_encode(struct nl_msg *msg, struct cmd_param *p)
+{
+	if (p->tz_id >= 0 && nla_put_u32(msg, THERMAL_GENL_ATTR_TZ_ID, p->tz_id))
+		return -1;
+
+	return 0;
+}
+
+static thermal_error_t thermal_genl_auto(struct thermal_handler *th, cmd_cb_t cmd_cb,
+					 struct cmd_param *param,
+					 int cmd, int flags, void *arg)
 {
 	struct nl_msg *msg;
 	void *hdr;
@@ -276,7 +291,7 @@ static thermal_error_t thermal_genl_auto(struct thermal_handler *th, int id, int
 	if (!hdr)
 		return THERMAL_ERROR;
 
-	if (id >= 0 && nla_put_u32(msg, THERMAL_GENL_ATTR_TZ_ID, id))
+	if (cmd_cb && cmd_cb(msg, param))
 		return THERMAL_ERROR;
 
 	if (nl_send_msg(th->sk_cmd, th->cb_cmd, msg, genl_handle_msg, arg))
@@ -289,30 +304,38 @@ static thermal_error_t thermal_genl_auto(struct thermal_handler *th, int id, int
 
 thermal_error_t thermal_cmd_get_tz(struct thermal_handler *th, struct thermal_zone **tz)
 {
-	return thermal_genl_auto(th, -1, THERMAL_GENL_CMD_TZ_GET_ID,
+	return thermal_genl_auto(th, NULL, NULL, THERMAL_GENL_CMD_TZ_GET_ID,
 				 NLM_F_DUMP | NLM_F_ACK, tz);
 }
 
 thermal_error_t thermal_cmd_get_cdev(struct thermal_handler *th, struct thermal_cdev **tc)
 {
-	return thermal_genl_auto(th, -1, THERMAL_GENL_CMD_CDEV_GET,
+	return thermal_genl_auto(th, NULL, NULL, THERMAL_GENL_CMD_CDEV_GET,
 				 NLM_F_DUMP | NLM_F_ACK, tc);
 }
 
 thermal_error_t thermal_cmd_get_trip(struct thermal_handler *th, struct thermal_zone *tz)
 {
-	return thermal_genl_auto(th, tz->id, THERMAL_GENL_CMD_TZ_GET_TRIP,
-				 0, tz);
+	struct cmd_param p = { .tz_id = tz->id };
+
+	return thermal_genl_auto(th, thermal_genl_tz_id_encode, &p,
+				 THERMAL_GENL_CMD_TZ_GET_TRIP, 0, tz);
 }
 
 thermal_error_t thermal_cmd_get_governor(struct thermal_handler *th, struct thermal_zone *tz)
 {
-	return thermal_genl_auto(th, tz->id, THERMAL_GENL_CMD_TZ_GET_GOV, 0, tz);
+	struct cmd_param p = { .tz_id = tz->id };
+
+	return thermal_genl_auto(th, thermal_genl_tz_id_encode, &p,
+				 THERMAL_GENL_CMD_TZ_GET_GOV, 0, tz);
 }
 
 thermal_error_t thermal_cmd_get_temp(struct thermal_handler *th, struct thermal_zone *tz)
 {
-	return thermal_genl_auto(th, tz->id, THERMAL_GENL_CMD_TZ_GET_TEMP, 0, tz);
+	struct cmd_param p = { .tz_id = tz->id };
+
+	return thermal_genl_auto(th, thermal_genl_tz_id_encode, &p,
+				 THERMAL_GENL_CMD_TZ_GET_TEMP, 0, tz);
 }
 
 thermal_error_t thermal_cmd_exit(struct thermal_handler *th)
Signature · 1 line
2.43.0
#14 lore raw

PATCH v2 6/7 tools/lib/thermal: Add the threshold netlink ABI

In reply to Daniel Lezcano

The thermal framework supports the thresholds and allows the userspace to create, delete, flush, get the list of the thresholds as well as getting the list of the thresholds set for a specific thermal zone.

Add the netlink abstraction in the thermal library to take full advantage of thresholds for the userspace program.

---
 tools/lib/thermal/commands.c        | 128 +++++++++++++++++++++++++++-
 tools/lib/thermal/events.c          |  58 ++++++++++---
 tools/lib/thermal/include/thermal.h |  40 +++++++++
 tools/lib/thermal/libthermal.map    |   5 ++
 tools/lib/thermal/thermal.c         |  17 ++++
 tools/thermal/lib/Makefile          |   2 +-
 6 files changed, 235 insertions(+), 15 deletions(-)
tools/lib/thermal/commands.c +127 -1
index a9223df91dcf..12539a519b81 100644
@@ -5,6 +5,7 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <unistd.h>
+#include <limits.h>
 
 #include <thermal.h>
 #include "thermal_nl.h"
@@ -33,6 +34,11 @@ static struct nla_policy thermal_genl_policy[THERMAL_GENL_ATTR_MAX + 1] = {
 	[THERMAL_GENL_ATTR_CDEV_CUR_STATE]      = { .type = NLA_U32 },
 	[THERMAL_GENL_ATTR_CDEV_MAX_STATE]      = { .type = NLA_U32 },
 	[THERMAL_GENL_ATTR_CDEV_NAME]           = { .type = NLA_STRING },
+
+        /* Thresholds */
+        [THERMAL_GENL_ATTR_THRESHOLD]      	= { .type = NLA_NESTED },
+        [THERMAL_GENL_ATTR_THRESHOLD_TEMP]      = { .type = NLA_U32 },
+        [THERMAL_GENL_ATTR_THRESHOLD_WAY]       = { .type = NLA_U32 },
 };
 
 static int parse_tz_get(struct genl_info *info, struct thermal_zone **tz)
@@ -182,6 +188,38 @@ static int parse_tz_get_gov(struct genl_info *info, struct thermal_zone *tz)
 	return THERMAL_SUCCESS;
 }
 
+static int parse_threshold_get(struct genl_info *info, struct thermal_zone *tz)
+{
+	struct nlattr *attr;
+	struct thermal_threshold *__tt = NULL;
+	size_t size = 0;
+	int rem;
+
+	nla_for_each_nested(attr, info->attrs[THERMAL_GENL_ATTR_THRESHOLD], rem) {
+
+		if (nla_type(attr) == THERMAL_GENL_ATTR_THRESHOLD_TEMP) {
+
+			size++;
+
+			__tt = realloc(__tt, sizeof(*__tt) * (size + 2));
+			if (!__tt)
+				return THERMAL_ERROR;
+
+			__tt[size - 1].temperature = nla_get_u32(attr);
+		}
+
+		if (nla_type(attr) == THERMAL_GENL_ATTR_THRESHOLD_WAY)
+			__tt[size - 1].direction = nla_get_u32(attr);
+	}
+
+	if (__tt)
+		__tt[size].temperature = INT_MAX;
+
+	tz->thresholds = __tt;
+
+	return THERMAL_SUCCESS;
+}
+
 static int handle_netlink(struct nl_cache_ops *unused,
 			  struct genl_cmd *cmd,
 			  struct genl_info *info, void *arg)
@@ -210,6 +248,10 @@ static int handle_netlink(struct nl_cache_ops *unused,
 		ret = parse_tz_get_gov(info, arg);
 		break;
 
+	case THERMAL_GENL_CMD_THRESHOLD_GET:
+		ret = parse_threshold_get(info, arg);
+		break;
+
 	default:
 		return THERMAL_ERROR;
 	}
@@ -253,6 +295,34 @@ static struct genl_cmd thermal_cmds[] = {
 		.c_maxattr	= THERMAL_GENL_ATTR_MAX,
 		.c_attr_policy	= thermal_genl_policy,
 	},
+        {
+                .c_id           = THERMAL_GENL_CMD_THRESHOLD_GET,
+                .c_name         = (char *)"Get thresholds list",
+                .c_msg_parser   = handle_netlink,
+                .c_maxattr      = THERMAL_GENL_ATTR_MAX,
+                .c_attr_policy  = thermal_genl_policy,
+        },
+        {
+                .c_id           = THERMAL_GENL_CMD_THRESHOLD_ADD,
+                .c_name         = (char *)"Add a threshold",
+                .c_msg_parser   = handle_netlink,
+                .c_maxattr      = THERMAL_GENL_ATTR_MAX,
+                .c_attr_policy  = thermal_genl_policy,
+        },
+        {
+                .c_id           = THERMAL_GENL_CMD_THRESHOLD_DELETE,
+                .c_name         = (char *)"Delete a threshold",
+                .c_msg_parser   = handle_netlink,
+                .c_maxattr      = THERMAL_GENL_ATTR_MAX,
+                .c_attr_policy  = thermal_genl_policy,
+        },
+        {
+                .c_id           = THERMAL_GENL_CMD_THRESHOLD_FLUSH,
+                .c_name         = (char *)"Flush the thresholds",
+                .c_msg_parser   = handle_netlink,
+                .c_maxattr      = THERMAL_GENL_ATTR_MAX,
+                .c_attr_policy  = thermal_genl_policy,
+        },
 };
 
 static struct genl_ops thermal_cmd_ops = {
@@ -263,13 +333,29 @@ static struct genl_ops thermal_cmd_ops = {
 
 struct cmd_param {
 	int tz_id;
+	int temp;
+	int direction;
 };
 
 typedef int (*cmd_cb_t)(struct nl_msg *, struct cmd_param *);
 
 static int thermal_genl_tz_id_encode(struct nl_msg *msg, struct cmd_param *p)
 {
-	if (p->tz_id >= 0 && nla_put_u32(msg, THERMAL_GENL_ATTR_TZ_ID, p->tz_id))
+	if (nla_put_u32(msg, THERMAL_GENL_ATTR_TZ_ID, p->tz_id))
+		return -1;
+
+	return 0;
+}
+
+static int thermal_genl_threshold_encode(struct nl_msg *msg, struct cmd_param *p)
+{
+	if (thermal_genl_tz_id_encode(msg, p))
+		return -1;
+
+	if (nla_put_u32(msg, THERMAL_GENL_ATTR_THRESHOLD_TEMP, p->temp))
+		return -1;
+
+	if (nla_put_u32(msg, THERMAL_GENL_ATTR_THRESHOLD_WAY, p->direction))
 		return -1;
 
 	return 0;
@@ -338,6 +424,46 @@ thermal_error_t thermal_cmd_get_temp(struct thermal_handler *th, struct thermal_
 				 THERMAL_GENL_CMD_TZ_GET_TEMP, 0, tz);
 }
 
+thermal_error_t thermal_cmd_threshold_get(struct thermal_handler *th,
+                                          struct thermal_zone *tz)
+{
+	struct cmd_param p = { .tz_id = tz->id };
+
+        return thermal_genl_auto(th, thermal_genl_tz_id_encode, &p,
+				 THERMAL_GENL_CMD_THRESHOLD_GET, 0, tz);
+}
+
+thermal_error_t thermal_cmd_threshold_add(struct thermal_handler *th,
+                                          struct thermal_zone *tz,
+                                          int temperature,
+                                          int direction)
+{
+	struct cmd_param p = { .tz_id = tz->id, .temp = temperature, .direction = direction };
+
+        return thermal_genl_auto(th, thermal_genl_threshold_encode, &p,
+				 THERMAL_GENL_CMD_THRESHOLD_ADD, 0, tz);
+}
+
+thermal_error_t thermal_cmd_threshold_delete(struct thermal_handler *th,
+                                             struct thermal_zone *tz,
+                                             int temperature,
+                                             int direction)
+{
+	struct cmd_param p = { .tz_id = tz->id, .temp = temperature, .direction = direction };
+
+        return thermal_genl_auto(th, thermal_genl_threshold_encode, &p,
+				 THERMAL_GENL_CMD_THRESHOLD_DELETE, 0, tz);
+}
+
+thermal_error_t thermal_cmd_threshold_flush(struct thermal_handler *th,
+                                            struct thermal_zone *tz)
+{
+	struct cmd_param p = { .tz_id = tz->id };
+
+        return thermal_genl_auto(th, thermal_genl_tz_id_encode, &p,
+				 THERMAL_GENL_CMD_THRESHOLD_FLUSH, 0, tz);
+}
+
 thermal_error_t thermal_cmd_exit(struct thermal_handler *th)
 {
 	if (genl_unregister_family(&thermal_cmd_ops))
tools/lib/thermal/events.c +45 -13

70 lines not on this page · raw message

tools/lib/thermal/include/thermal.h +40 -0
index 1abc560602cf..0571ec0bf438 100644
@@ -4,11 +4,20 @@
 #define __LIBTHERMAL_H
 
 #include <linux/thermal.h>
+#include <sys/types.h>
 
 #ifndef LIBTHERMAL_API
 #define LIBTHERMAL_API __attribute__((visibility("default")))
 #endif
 
+#ifndef THERMAL_THRESHOLD_WAY_UP
+#define THERMAL_THRESHOLD_WAY_UP 0x1
+#endif
+
+#ifndef THERMAL_THRESHOLD_WAY_DOWN
+#define THERMAL_THRESHOLD_WAY_DOWN 0x2
+#endif
+
 #ifdef __cplusplus
 extern "C" {
 #endif
@@ -31,6 +40,11 @@ struct thermal_events_ops {
 	int (*cdev_delete)(int cdev_id, void *arg);
 	int (*cdev_update)(int cdev_id, int cur_state, void *arg);
 	int (*gov_change)(int tz_id, const char *gov_name, void *arg);
+	int (*threshold_add)(int tz_id, int temperature, int direction, pid_t pid, void *arg);
+	int (*threshold_delete)(int tz_id, int temperature, int direction, pid_t pid, void *arg);
+	int (*threshold_flush)(int tz_id, pid_t pid, void *arg);
+	int (*threshold_up)(int tz_id, int temp, int last_temp, void *arg);
+	int (*threshold_down)(int tz_id, int temp, int last_temp, void *arg);
 };
 
 struct thermal_ops {
@@ -45,12 +59,18 @@ struct thermal_trip {
 	int hyst;
 };
 
+struct thermal_threshold {
+	int temperature;
+	int direction;
+};
+	
 struct thermal_zone {
 	int id;
 	int temp;
 	char name[THERMAL_NAME_LENGTH];
 	char governor[THERMAL_NAME_LENGTH];
 	struct thermal_trip *trip;
+	struct thermal_threshold *thresholds;
 };
 
 struct thermal_cdev {
@@ -74,12 +94,16 @@ typedef int (*cb_tt_t)(struct thermal_trip *, void *);
 
 typedef int (*cb_tc_t)(struct thermal_cdev *, void *);
 
+typedef int (*cb_th_t)(struct thermal_threshold *, void *);
+
 LIBTHERMAL_API int for_each_thermal_zone(struct thermal_zone *tz, cb_tz_t cb, void *arg);
 
 LIBTHERMAL_API int for_each_thermal_trip(struct thermal_trip *tt, cb_tt_t cb, void *arg);
 
 LIBTHERMAL_API int for_each_thermal_cdev(struct thermal_cdev *cdev, cb_tc_t cb, void *arg);
 
+LIBTHERMAL_API int for_each_thermal_threshold(struct thermal_threshold *th, cb_th_t cb, void *arg);
+
 LIBTHERMAL_API struct thermal_zone *thermal_zone_find_by_name(struct thermal_zone *tz,
 							      const char *name);
 
@@ -124,6 +148,22 @@ LIBTHERMAL_API thermal_error_t thermal_cmd_get_governor(struct thermal_handler *
 LIBTHERMAL_API thermal_error_t thermal_cmd_get_temp(struct thermal_handler *th,
 						    struct thermal_zone *tz);
 
+LIBTHERMAL_API thermal_error_t thermal_cmd_threshold_get(struct thermal_handler *th,
+							 struct thermal_zone *tz);
+
+LIBTHERMAL_API thermal_error_t thermal_cmd_threshold_add(struct thermal_handler *th,
+                                                         struct thermal_zone *tz,
+                                                         int temperature,
+                                                         int direction);
+
+LIBTHERMAL_API thermal_error_t thermal_cmd_threshold_delete(struct thermal_handler *th,
+                                                            struct thermal_zone *tz,
+                                                            int temperature,
+                                                            int direction);
+
+LIBTHERMAL_API thermal_error_t thermal_cmd_threshold_flush(struct thermal_handler *th,
+                                                           struct thermal_zone *tz);
+
 /*
  * Netlink thermal samples
  */
tools/lib/thermal/libthermal.map +5 -0
index d5e77738c7a4..d657176aa47f 100644
@@ -4,6 +4,7 @@ LIBTHERMAL_0.0.1 {
 		for_each_thermal_zone;
 		for_each_thermal_trip;
 		for_each_thermal_cdev;
+		for_each_thermal_threshold;
 		thermal_zone_find_by_name;
 		thermal_zone_find_by_id;
 		thermal_zone_discover;
@@ -17,6 +18,10 @@ LIBTHERMAL_0.0.1 {
 		thermal_cmd_get_trip;
 		thermal_cmd_get_governor;
 		thermal_cmd_get_temp;
+		thermal_cmd_threshold_get;
+		thermal_cmd_threshold_add;
+		thermal_cmd_threshold_delete;
+		thermal_cmd_threshold_flush;
 		thermal_sampling_init;
 		thermal_sampling_handle;
 		thermal_sampling_fd;
tools/lib/thermal/thermal.c +17 -0
index 72a76dc205bc..4851744d482e 100644
@@ -1,10 +1,24 @@
 // SPDX-License-Identifier: LGPL-2.1+
 // Copyright (C) 2022, Linaro Ltd - Daniel Lezcano <[email protected]>
 #include <stdio.h>
+#include <limits.h>
 #include <thermal.h>
 
 #include "thermal_nl.h"
 
+int for_each_thermal_threshold(struct thermal_threshold *th, cb_th_t cb, void *arg)
+{
+	int i, ret = 0;
+
+	if (!th)
+		return 0;
+
+	for (i = 0; th[i].temperature != INT_MAX; i++)
+		ret |= cb(&th[i], arg);
+
+	return ret;
+}
+	
 int for_each_thermal_cdev(struct thermal_cdev *cdev, cb_tc_t cb, void *arg)
 {
 	int i, ret = 0;
@@ -80,6 +94,9 @@ static int __thermal_zone_discover(struct thermal_zone *tz, void *th)
 	if (thermal_cmd_get_trip(th, tz) < 0)
 		return -1;
 
+	if (thermal_cmd_threshold_get(th, tz) < 0)
+		return -1;
+	
 	if (thermal_cmd_get_governor(th, tz))
 		return -1;
 
tools/thermal/lib/Makefile +1 -1
index 82db451935c5..f2552f73a64c 100644
@@ -3,7 +3,7 @@
 
 LIBTHERMAL_TOOLS_VERSION = 0
 LIBTHERMAL_TOOLS_PATCHLEVEL = 0
-LIBTHERMAL_TOOLS_EXTRAVERSION = 1
+LIBTHERMAL_TOOLS_EXTRAVERSION = 2
 
 MAKEFLAGS += --no-print-directory
 
Signature · 1 line
2.43.0
#15 lore raw

PATCH v2 7/7 tools/thermal/thermal-engine: Take into account the thresholds API

In reply to Daniel Lezcano

Enhance the thermal-engine skeleton with the thresholds added in the kernel and use the API exported by the thermal library.

---
 tools/thermal/thermal-engine/thermal-engine.c | 109 +++++++++++++++---
 1 file changed, 96 insertions(+), 13 deletions(-)
tools/thermal/thermal-engine/thermal-engine.c +96 -13
index 9b1476a2680f..ddc30a27acda 100644
@@ -38,6 +38,14 @@ struct thermal_data {
 	struct thermal_handler *th;
 };
 
+static int show_threshold(struct thermal_threshold *th, __maybe_unused void *arg)
+{
+	INFO("threshold temp=%d, direction=%d\n",
+	     th->temperature, th->direction);
+
+	return 0;
+}
+
 static int show_trip(struct thermal_trip *tt, __maybe_unused void *arg)
 {
 	INFO("trip id=%d, type=%d, temp=%d, hyst=%d\n",
@@ -70,6 +78,8 @@ static int show_tz(struct thermal_zone *tz, __maybe_unused void *arg)
 
 	for_each_thermal_trip(tz->trip, show_trip, NULL);
 
+	for_each_thermal_threshold(tz->thresholds, show_threshold, NULL);
+	
 	show_temp(tz, arg);
 
 	show_governor(tz, arg);
@@ -77,6 +87,30 @@ static int show_tz(struct thermal_zone *tz, __maybe_unused void *arg)
 	return 0;
 }
 
+static int set_threshold(struct thermal_zone *tz, __maybe_unused void *arg)
+{
+	struct thermal_handler *th = arg;
+	int thresholds[] = { 43000, 65000, 49000, 55000, 57000 };
+	size_t i;
+
+	INFO("Setting threshold for thermal zone '%s', id=%d\n", tz->name, tz->id);
+
+	if (thermal_cmd_threshold_flush(th, tz)) {
+		ERROR("Failed to flush all previous thresholds\n");
+		return -1;
+	}
+
+	for (i = 0; i < sizeof(thresholds) / sizeof(thresholds[0]); i++)
+		if (thermal_cmd_threshold_add(th, tz, thresholds[i],
+					      THERMAL_THRESHOLD_WAY_UP |
+					      THERMAL_THRESHOLD_WAY_DOWN)) {
+			ERROR("Failed to set threshold\n");
+			return -1;
+		}
+
+	return 0;
+}
+
 static int tz_create(const char *name, int tz_id, __maybe_unused void *arg)
 {
 	INFO("Thermal zone '%s'/%d created\n", name, tz_id);
@@ -197,20 +231,66 @@ static int gov_change(int tz_id, const char *name, __maybe_unused void *arg)
 	return 0;
 }
 
+static int threshold_add(int tz_id, int temp, int direction,
+			 pid_t pid, __maybe_unused void *arg)
+{
+	INFO("Threshold added by pid=%d, tz_id=%d: temp=%d, direction=%d\n",
+	     pid, tz_id, temp, direction);
+
+	return 0;
+}
+
+static int threshold_delete(int tz_id, int temp, int direction,
+			    pid_t pid, __maybe_unused void *arg)
+{
+	INFO("Threshold deleted by pid=%d, tz_id=%d: temp=%d, direction=%d\n",
+	     pid, tz_id, temp, direction);
+
+	return 0;
+}
+
+static int threshold_flush(int tz_id, pid_t pid, __maybe_unused void *arg)
+{
+	INFO("Thresholds flushed by pid=%d, tz_id=%d\n", pid, tz_id);
+
+	return 0;
+}
+
+static int threshold_up(int tz_id, int temp, int last_temp, __maybe_unused void *arg)
+{
+	INFO("Threshold crossed way up tz_id=%d: temp=%d, last_temp=%d\n",
+	     tz_id, temp, last_temp);
+
+	return 0;
+}
+
+static int threshold_down(int tz_id, int temp, int last_temp, __maybe_unused void *arg)
+{
+	INFO("Threshold crossed way down tz_id=%d: temp=%d, last_temp=%d\n",
+	     tz_id, temp, last_temp);
+
+	return 0;
+}
+
 static struct thermal_ops ops = {
-	.events.tz_create	= tz_create,
-	.events.tz_delete	= tz_delete,
-	.events.tz_disable	= tz_disable,
-	.events.tz_enable	= tz_enable,
-	.events.trip_high	= trip_high,
-	.events.trip_low	= trip_low,
-	.events.trip_add	= trip_add,
-	.events.trip_delete	= trip_delete,
-	.events.trip_change	= trip_change,
-	.events.cdev_add	= cdev_add,
-	.events.cdev_delete	= cdev_delete,
-	.events.cdev_update	= cdev_update,
-	.events.gov_change	= gov_change
+	.events.tz_create		= tz_create,
+	.events.tz_delete		= tz_delete,
+	.events.tz_disable		= tz_disable,
+	.events.tz_enable		= tz_enable,
+	.events.trip_high		= trip_high,
+	.events.trip_low		= trip_low,
+	.events.trip_add		= trip_add,
+	.events.trip_delete		= trip_delete,
+	.events.trip_change		= trip_change,
+	.events.cdev_add		= cdev_add,
+	.events.cdev_delete		= cdev_delete,
+	.events.cdev_update		= cdev_update,
+	.events.gov_change		= gov_change,
+	.events.threshold_add		= threshold_add,
+	.events.threshold_delete	= threshold_delete,
+	.events.threshold_flush		= threshold_flush,
+	.events.threshold_up		= threshold_up,
+	.events.threshold_down		= threshold_down,
 };
 
 static int thermal_event(__maybe_unused int fd, __maybe_unused void *arg)
@@ -280,6 +360,7 @@ enum {
 	THERMAL_ENGINE_DAEMON_ERROR,
 	THERMAL_ENGINE_LOG_ERROR,
 	THERMAL_ENGINE_THERMAL_ERROR,
+	THERMAL_ENGINE_THRESHOLD_ERROR,	
 	THERMAL_ENGINE_MAINLOOP_ERROR,
 };
 
@@ -318,6 +399,8 @@ int main(int argc, char *argv[])
 		return THERMAL_ENGINE_THERMAL_ERROR;
 	}
 
+	for_each_thermal_zone(td.tz, set_threshold, td.th);
+
 	for_each_thermal_zone(td.tz, show_tz, td.th);
 
 	if (mainloop_init()) {
Signature · 1 line
2.43.0
#16 lore raw
In reply to Daniel Lezcano
Quoted 34 lines · Daniel Lezcano

The trip points are a firmware description of the temperature limits of a specific thermal zone where we associate an action which is done by the kernel. The time resolution is low.

The userspace has to deal with a more complex thermal management based on heuristics from different information coming from different places. The logic is much more complex but based on a bigger time resolution, usually one second based.

The purpose of the userspace is to monitor the temperatures from different places and take actions. However, it can not be constantly reading the temperature to detect when a temperature threshold has been reached. This is especially bad for mobile or embedded system as that will lead to an unacceptable number of wakeup to check the temperature with nothing to do.

On the other side, the sensors are now most of the time interrupt driven. That means the thermal framework will use the temperature trip points to program the sensor to trigger an interrupt when a temperature limit is crossed.

Unfortunately, the userspace can not benefit this feature and current solutions found here and there, iow out-of-tree, are to add fake trip points in the firmware and enable the writable trip points.

This is bad for different reasons, the trip points are for in-kernel actions, the semantic of their types is used by the thermal framework and by adding trip points in the device tree is a way to overcome the current limitation but tampering with how the thermal framework is supposed to work. The writable trip points is a way to adjust a temperature limit given a specific platform if the firmware is not accurate enough and TBH it is more a debug feature from my POV.

Quoted 6 lines

The thresholds mechanism is a way to have the userspace to tell thermal framework to send a notification when a temperature limit is crossed. There is no id, no hysteresis, just the temperature and the direction of the limit crossing. That means we can be notified when a threshold is crossed the way up only, or the way down only or both ways. That allows to create hysteresis values if it is needed.

First off, I'd prefer these things to be referred to as "user thresholds" to avoid confusion with trip thresholds.

Quoted 7 lines

A threshold can be added, deleted or flushed. The latter means all thresholds belonging to a thermal zone will be deleted.

When a threshold is added:

- if the same threshold (temperature and direction) exists, an error is returned

Why is it useful to return an error in this case? It appears that the existing threshold could be used just fine.

Quoted 19 lines

- if a threshold is specified with the same temperature but a different direction, the specified direction is added

- if there is no threshold with the same temperature then it is created

When a threshold is deleted:

- if the same threshold (temperature and direction) exists, it is deleted

- if a threshold is specified with the same temperature but a different direction, the specified direction is removed

- if there is no threshold with the same temperature, then an error is returned

When the threshold are flushed:

Quoted 7 lines

- All thresholds related to a thermal zone are deleted

When a threshold is crossed:

- the userspace does not need to know which threshold(s) have been crossed, it will be notified with the thermal zone identifier, the current temperature and the previous temperature

This has a bit of a warm-up issue when the zone temperature is above a threshold to start with.

Or is it not a problem because thresholds can only be added when the zone temperature is known?

Quoted 4 lines

- if multiple thresholds have been crossed between two updates only one notification will be send to the userspace, it is pointless to send a notification per thresholds crossed as the userspace can handle that easily when it has the temperature delta information

That's clever.

Quoted 14 lines

All aforementioned actions and events lead to a notification to the userspace. A threshold change (add, delete and flush) is notified to the userspace with the process id responsible of the action.

Along with the kernel changes, the thermal library has been extended to provide the different API to deal with the new threshold netlink events and commands.

In addition, the thermal-engine skeleton uses these new API by flushing and adding thresholds as well as getting the notification about these actions.

Overall the series has been tested with the thermal-engine skeleton and some selftests which are not part of this series.

#17 lore raw
In reply to Daniel Lezcano

Hi Daniel,

Quoted 42 lines · Daniel Lezcano

The trip points are a firmware description of the temperature limits of a specific thermal zone where we associate an action which is done by the kernel. The time resolution is low.

The userspace has to deal with a more complex thermal management based
on heuristics from different information coming from different places. The logic is much more complex but based on a bigger time resolution, usually one second based.

The purpose of the userspace is to monitor the temperatures from different places and take actions. However, it can not be constantly reading the temperature to detect when a temperature threshold has been reached. This is especially bad for mobile or embedded system as that will lead to an unacceptable number of wakeup to check the temperature with nothing to do.

On the other side, the sensors are now most of the time interrupt driven. That means the thermal framework will use the temperature trip
points to program the sensor to trigger an interrupt when a temperature limit is crossed.

Unfortunately, the userspace can not benefit this feature and current solutions found here and there, iow out-of-tree, are to add fake trip points in the firmware and enable the writable trip points.

This is bad for different reasons, the trip points are for in-kernel actions, the semantic of their types is used by the thermal framework and by adding trip points in the device tree is a way to overcome the current limitation but tampering with how the thermal framework is supposed to work. The writable trip points is a way to adjust a temperature limit given a specific platform if the firmware is not accurate enough and TBH it is more a debug feature from my POV.

The thresholds mechanism is a way to have the userspace to tell thermal framework to send a notification when a temperature limit is crossed. There is no id, no hysteresis, just the temperature and the direction of the limit crossing. That means we can be notified when a threshold is crossed the way up only, or the way down only or both ways. That allows to create hysteresis values if it is needed.

Quoted 3 lines

A threshold can be added, deleted or flushed. The latter means all thresholds belonging to a thermal zone will be deleted.

So you are proposing to add threshold via netlink, not adding any new sysfs attribute? That is not clear here.

I think you are adding"
THERMAL_GENL_CMD_THRESHOLD_GET
THERMAL_GENL_CMD_THRESHOLD_ADD
THERMAL_GENL_CMD_THRESHOLD_DELETE
THERMAL_GENL_CMD_THRESHOLD_FLUSH

We need to document our netlink messages including old ones.

Also we should add "MODIFY" as we tend to change them quite often.

Also no hysteresis, that is practically we can't use. Temperature changes so much that that will flood user space. You will get 100s of events on CPU temperature if you set temperature threshold in CPU. We have a whole filtering in driver to avoid this.
You need a rate limit here.

There are multiple user processes can add threshold and there is no ownership. So one process can cause too much noise to others as it is multicast.

We worked on a change to filer these as discussed during last LPC, but not posted yet. This will really need this as this will be too many messages.

Thanks,
Srinivas

Quoted 95 lines

When a threshold is added:

 - if the same threshold (temperature and direction) exists, an error is returned

 - if a threshold is specified with the same temperature but a different direction, the specified direction is added

 - if there is no threshold with the same temperature then it is created

When a threshold is deleted:

 - if the same threshold (temperature and direction) exists, it is deleted

 - if a threshold is specified with the same temperature but a different direction, the specified direction is removed

 - if there is no threshold with the same temperature, then an error is returned

When the threshold are flushed:

 - All thresholds related to a thermal zone are deleted

When a threshold is crossed:

 - the userspace does not need to know which threshold(s) have been crossed, it will be notified with the thermal zone identifier, the current temperature and the previous temperature

 - if multiple thresholds have been crossed between two updates only one notification will be send to the userspace, it is pointless to send a notification per thresholds crossed as the userspace can handle that easily when it has the temperature delta information

All aforementioned actions and events lead to a notification to the userspace. A threshold change (add, delete and flush) is notified to the userspace with the process id responsible of the action.

Along with the kernel changes, the thermal library has been extended to provide the different API to deal with the new threshold netlink events and commands.

In addition, the thermal-engine skeleton uses these new API by flushing and adding thresholds as well as getting the notification about these actions.

Overall the series has been tested with the thermal-engine skeleton and some selftests which are not part of this series.

Changelog:
  V2:
    - Compute min and max in thermal_zone_device_update() but keep the loop as it is (Rafael)

    - Include slab.h to fix compilation warnings on some architectures
      with kmalloc and kfree (kernel test robot)

Daniel Lezcano (7):
  thermal/core: Compute low and high boundaries in thermal_zone_device_update()
  thermal/core: Add thresholds support
  thermal/core: Connect the threshold with the core thermal/netlink: Add the commands and the events for the thresholds tools/lib/thermal: Make more generic the command encoding function tools/lib/thermal: Add the threshold netlink ABI tools/thermal/thermal-engine: Take into account the thresholds API

 drivers/thermal/Kconfig                       |  15 ++ drivers/thermal/Makefile                      |   3 + drivers/thermal/thermal_core.c                |  21 +- drivers/thermal/thermal_core.h                |   6 +- drivers/thermal/thermal_netlink.c             | 239
++++++++++++++++-
 drivers/thermal/thermal_netlink.h             |   7 + drivers/thermal/thermal_thresholds.c          | 247
++++++++++++++++++
 drivers/thermal/thermal_thresholds.h          |  54 ++++ drivers/thermal/thermal_trip.c                |  27 +- include/linux/thermal.h                       |   3 + include/uapi/linux/thermal.h                  |  30 ++- tools/lib/thermal/commands.c                  | 167 +++++++++++- tools/lib/thermal/events.c                    |  58 +++- tools/lib/thermal/include/thermal.h           |  40 +++ tools/lib/thermal/libthermal.map              |   5 + tools/lib/thermal/thermal.c                   |  17 ++ tools/thermal/lib/Makefile                    |   2 +- tools/thermal/thermal-engine/thermal-engine.c | 109 +++++++-

 18 files changed, 972 insertions(+), 78 deletions(-)

 create mode 100644 drivers/thermal/thermal_thresholds.c create mode 100644 drivers/thermal/thermal_thresholds.h

#18 lore raw

Hi Srinivas,

Quoted 59 lines · Srinivas

Hi Daniel,

Quoted 42 lines · Daniel Lezcano

The trip points are a firmware description of the temperature limits of a specific thermal zone where we associate an action which is done by the kernel. The time resolution is low.

The userspace has to deal with a more complex thermal management based
on heuristics from different information coming from different places. The logic is much more complex but based on a bigger time resolution, usually one second based.

The purpose of the userspace is to monitor the temperatures from different places and take actions. However, it can not be constantly reading the temperature to detect when a temperature threshold has been reached. This is especially bad for mobile or embedded system as that will lead to an unacceptable number of wakeup to check the temperature with nothing to do.

On the other side, the sensors are now most of the time interrupt driven. That means the thermal framework will use the temperature trip
points to program the sensor to trigger an interrupt when a temperature limit is crossed.

Unfortunately, the userspace can not benefit this feature and current solutions found here and there, iow out-of-tree, are to add fake trip points in the firmware and enable the writable trip points.

This is bad for different reasons, the trip points are for in-kernel actions, the semantic of their types is used by the thermal framework and by adding trip points in the device tree is a way to overcome the current limitation but tampering with how the thermal framework is supposed to work. The writable trip points is a way to adjust a temperature limit given a specific platform if the firmware is not accurate enough and TBH it is more a debug feature from my POV.

The thresholds mechanism is a way to have the userspace to tell thermal framework to send a notification when a temperature limit is crossed. There is no id, no hysteresis, just the temperature and the direction of the limit crossing. That means we can be notified when a threshold is crossed the way up only, or the way down only or both ways. That allows to create hysteresis values if it is needed.

Quoted 3 lines

A threshold can be added, deleted or flushed. The latter means all thresholds belonging to a thermal zone will be deleted.

So you are proposing to add threshold via netlink, not adding any new sysfs attribute? That is not clear here.

I think you are adding"
THERMAL_GENL_CMD_THRESHOLD_GET
THERMAL_GENL_CMD_THRESHOLD_ADD
THERMAL_GENL_CMD_THRESHOLD_DELETE
THERMAL_GENL_CMD_THRESHOLD_FLUSH

Quoted 5 lines

We need to document our netlink messages including old ones.

Also we should add "MODIFY" as we tend to change them quite often.

Also no hysteresis, that is practically we can't use.

The direction thing is equivalent to hysteresis though.

Instead of using one threshold with a given hysteresis value, use two of them with different temperature values and different directions.

Quoted 2 lines

Temperature changes so much that that will flood user space. You will get 100s of events on CPU temperature if you set temperature threshold in CPU.

Events only trigger when thresholds are crossed in a specific direction, but overall you have a point.

Quoted 10 lines

We have a whole filtering in driver to avoid this. You need a rate limit here.

There are multiple user processes can add threshold and there is no ownership. So one process can cause too much noise to others as it is multicast.

We worked on a change to filer these as discussed during last LPC, but not posted yet. This will really need this as this will be too many messages.

Unless there is a way to limit your subscription to events regarding a specific thermal zone, for instance.

Anyway, I have to admit ignorance regarding the user space usage model related to this. For example, is it expected that there will be one user space entity managing the thresholds or there can be many.

#19 lore raw
Quoted 92 lines · Rafael J. Wysocki

Hi Srinivas,

Quoted 77 lines · Srinivas

Hi Daniel,

Quoted 58 lines · Daniel Lezcano

The trip points are a firmware description of the temperature limits
of a specific thermal zone where we associate an action which is done
by the kernel. The time resolution is low.

The userspace has to deal with a more complex thermal management based
on heuristics from different information coming from different places. The logic is much more complex but based on a bigger time resolution, usually one second based.

The purpose of the userspace is to monitor the temperatures from different places and take actions. However, it can not be constantly
reading the temperature to detect when a temperature threshold has
been reached. This is especially bad for mobile or embedded system as
that will lead to an unacceptable number of wakeup to check the temperature with nothing to do.

On the other side, the sensors are now most of the time interrupt driven. That means the thermal framework will use the temperature trip
points to program the sensor to trigger an interrupt when a temperature limit is crossed.

Unfortunately, the userspace can not benefit this feature and current
solutions found here and there, iow out-of-tree, are to add fake trip
points in the firmware and enable the writable trip points.

This is bad for different reasons, the trip points are for in- kernel
actions, the semantic of their types is used by the thermal framework
and by adding trip points in the device tree is a way to overcome the
current limitation but tampering with how the thermal framework is
supposed to work. The writable trip points is a way to adjust a temperature limit given a specific platform if the firmware is not
accurate enough and TBH it is more a debug feature from my POV.

The thresholds mechanism is a way to have the userspace to tell thermal framework to send a notification when a temperature limit is
crossed. There is no id, no hysteresis, just the temperature and the
direction of the limit crossing. That means we can be notified when a
threshold is crossed the way up only, or the way down only or both
ways. That allows to create hysteresis values if it is needed.

Quoted 4 lines

A threshold can be added, deleted or flushed. The latter means all
thresholds belonging to a thermal zone will be deleted.

So you are proposing to add threshold via netlink, not adding any new
sysfs attribute? That is not clear here.

I think you are adding"
THERMAL_GENL_CMD_THRESHOLD_GET
THERMAL_GENL_CMD_THRESHOLD_ADD
THERMAL_GENL_CMD_THRESHOLD_DELETE
THERMAL_GENL_CMD_THRESHOLD_FLUSH

Quoted 5 lines

We need to document our netlink messages including old ones.

Also we should add "MODIFY" as we tend to change them quite often.

Also no hysteresis, that is practically we can't use.

The direction thing is equivalent to hysteresis though.

Instead of using one threshold with a given hysteresis value, use two of them with different temperature values and different directions.

Quoted 6 lines
Quoted 3 lines

Temperature changes so much that that will flood user space. You will get 100s of
events on CPU temperature if you set temperature threshold in CPU.

Events only trigger when thresholds are crossed in a specific direction, but overall you have a point.

This is good enough for some sensor which changes slowly like skin. But some sensors like CPU and board are not like that. We publish both raw and filtered event count in debugfs for x86 package temp. For example, I just ran a sample on a client for one threshold.

Total notification from hardware: 224
Notified to user space: 16

Quoted 9 lines
Quoted 9 lines

We have a whole filtering in driver to avoid this. You need a rate limit here.

There are multiple user processes can add threshold and there is no ownership. So one process can cause too much noise to others as it is
multicast.

Quoted 8 lines
Quoted 4 lines

We worked on a change to filer these as discussed during last LPC, but
not posted yet. This will really need this as this will be too many messages.

Unless there is a way to limit your subscription to events regarding a
specific thermal zone, for instance.

That's what this change does.

Quoted 5 lines

Anyway, I have to admit ignorance regarding the user space usage model
related to this.  For example, is it expected that there will be one user space entity managing the thresholds or there can be many.

Good question. That's why we didn't send this change. If there is one agent this is fine.

But in embedded space, there may be more than one. Hence they needed netlink multicast instead of just one char device.

Thanks,
Srinivas

#20 lore raw
Quoted 94 lines · Srinivas
Quoted 92 lines · Rafael J. Wysocki

Hi Srinivas,

Quoted 77 lines · Srinivas

Hi Daniel,

Quoted 58 lines · Daniel Lezcano

The trip points are a firmware description of the temperature limits
of a specific thermal zone where we associate an action which is done
by the kernel. The time resolution is low.

The userspace has to deal with a more complex thermal management based
on heuristics from different information coming from different places. The logic is much more complex but based on a bigger time resolution, usually one second based.

The purpose of the userspace is to monitor the temperatures from different places and take actions. However, it can not be constantly
reading the temperature to detect when a temperature threshold has
been reached. This is especially bad for mobile or embedded system as
that will lead to an unacceptable number of wakeup to check the temperature with nothing to do.

On the other side, the sensors are now most of the time interrupt driven. That means the thermal framework will use the temperature trip
points to program the sensor to trigger an interrupt when a temperature limit is crossed.

Unfortunately, the userspace can not benefit this feature and current
solutions found here and there, iow out-of-tree, are to add fake trip
points in the firmware and enable the writable trip points.

This is bad for different reasons, the trip points are for in- kernel
actions, the semantic of their types is used by the thermal framework
and by adding trip points in the device tree is a way to overcome the
current limitation but tampering with how the thermal framework is
supposed to work. The writable trip points is a way to adjust a temperature limit given a specific platform if the firmware is not
accurate enough and TBH it is more a debug feature from my POV.

The thresholds mechanism is a way to have the userspace to tell thermal framework to send a notification when a temperature limit is
crossed. There is no id, no hysteresis, just the temperature and the
direction of the limit crossing. That means we can be notified when a
threshold is crossed the way up only, or the way down only or both
ways. That allows to create hysteresis values if it is needed.

Quoted 4 lines

A threshold can be added, deleted or flushed. The latter means all
thresholds belonging to a thermal zone will be deleted.

So you are proposing to add threshold via netlink, not adding any new
sysfs attribute? That is not clear here.

I think you are adding"
THERMAL_GENL_CMD_THRESHOLD_GET
THERMAL_GENL_CMD_THRESHOLD_ADD
THERMAL_GENL_CMD_THRESHOLD_DELETE
THERMAL_GENL_CMD_THRESHOLD_FLUSH

Quoted 5 lines

We need to document our netlink messages including old ones.

Also we should add "MODIFY" as we tend to change them quite often.

Also no hysteresis, that is practically we can't use.

The direction thing is equivalent to hysteresis though.

Instead of using one threshold with a given hysteresis value, use two of them with different temperature values and different directions.

Quoted 8 lines
Quoted 6 lines
Quoted 3 lines

Temperature changes so much that that will flood user space. You will get 100s of
events on CPU temperature if you set temperature threshold in CPU.

Events only trigger when thresholds are crossed in a specific direction, but overall you have a point.

This is good enough for some sensor which changes slowly like skin.

Right, and I think that this is the driving use case.

Quoted 1 line

But some sensors like CPU and board are not like that.

Sure, so this interface will not be for them.

Quoted 5 lines

We publish both raw and filtered event count in debugfs for x86 package temp. For example, I just ran a sample on a client for one threshold.

Total notification from hardware: 224
Notified to user space: 16

So did you apply the Daniel's patches and run the test or did you do something else?

Quoted 20 lines
Quoted 9 lines
Quoted 9 lines

We have a whole filtering in driver to avoid this. You need a rate limit here.

There are multiple user processes can add threshold and there is no ownership. So one process can cause too much noise to others as it is
multicast.

Quoted 8 lines
Quoted 4 lines

We worked on a change to filer these as discussed during last LPC, but
not posted yet. This will really need this as this will be too many messages.

Unless there is a way to limit your subscription to events regarding a
specific thermal zone, for instance.

That's what this change does.

Quoted 1 line
Quoted 7 lines
Quoted 4 lines

Anyway, I have to admit ignorance regarding the user space usage model
related to this. For example, is it expected that there will be one user space entity managing the thresholds or there can be many.

Good question. That's why we didn't send this change. If there is one agent this is fine.

I think that the addition and deletion of a user threshold should be privileged operations.

If that is the case, it all boils down to proper coordination in user space.

Quoted 2 lines

But in embedded space, there may be more than one. Hence they needed netlink multicast instead of just one char device.

I see.

#21 lore raw
Quoted 131 lines · Rafael J. Wysocki
Quoted 129 lines · Srinivas
Quoted 126 lines · Rafael J. Wysocki

Hi Srinivas,

Quoted 100 lines · Srinivas

Hi Daniel,

Quoted 79 lines · Daniel Lezcano

The trip points are a firmware description of the temperature limits
of a specific thermal zone where we associate an action which is
done
by the kernel. The time resolution is low.

The userspace has to deal with a more complex thermal management
based
on heuristics from different information coming from different
places. The logic is much more complex but based on a bigger time
resolution, usually one second based.

The purpose of the userspace is to monitor the temperatures from
different places and take actions. However, it can not be constantly
reading the temperature to detect when a temperature threshold
has
been reached. This is especially bad for mobile or embedded system as
that will lead to an unacceptable number of wakeup to check the
temperature with nothing to do.

On the other side, the sensors are now most of the time interrupt
driven. That means the thermal framework will use the temperature
trip
points to program the sensor to trigger an interrupt when a temperature limit is crossed.

Unfortunately, the userspace can not benefit this feature and current
solutions found here and there, iow out-of-tree, are to add fake
trip
points in the firmware and enable the writable trip points.

This is bad for different reasons, the trip points are for in-
kernel
actions, the semantic of their types is used by the thermal framework
and by adding trip points in the device tree is a way to overcome
the
current limitation but tampering with how the thermal framework
is
supposed to work. The writable trip points is a way to adjust a
temperature limit given a specific platform if the firmware is
not
accurate enough and TBH it is more a debug feature from my POV.

The thresholds mechanism is a way to have the userspace to tell
thermal framework to send a notification when a temperature limit
is
crossed. There is no id, no hysteresis, just the temperature and
the
direction of the limit crossing. That means we can be notified
when a
threshold is crossed the way up only, or the way down only or both
ways. That allows to create hysteresis values if it is needed.

Quoted 5 lines

A threshold can be added, deleted or flushed. The latter means
all
thresholds belonging to a thermal zone will be deleted.

So you are proposing to add threshold via netlink, not adding any
new
sysfs attribute? That is not clear here.

I think you are adding"
THERMAL_GENL_CMD_THRESHOLD_GET
THERMAL_GENL_CMD_THRESHOLD_ADD
THERMAL_GENL_CMD_THRESHOLD_DELETE
THERMAL_GENL_CMD_THRESHOLD_FLUSH

Quoted 6 lines

We need to document our netlink messages including old ones.

Also we should add "MODIFY" as we tend to change them quite often.

Also no hysteresis, that is practically we can't use.

The direction thing is equivalent to hysteresis though.

Instead of using one threshold with a given hysteresis value, use two
of them with different temperature values and different directions.

Quoted 5 lines

Temperature changes so much that that will flood user space. You
will get 100s of
events on CPU temperature if you set temperature threshold in CPU.

Events only trigger when thresholds are crossed in a specific direction, but overall you have a point.

Quoted 7 lines
Quoted 1 line

This is good enough for some sensor which changes slowly like skin.

Right, and I think that this is the driving use case.

Quoted 1 line

But some sensors like CPU and board are not like that.

Sure, so this interface will not be for them.

But when we design an interface, I think this shouldn't be sensor dependent.

Quoted 10 lines
Quoted 7 lines

We publish both raw and filtered event count in debugfs for x86 package temp.
For example, I just ran a sample on a client for one threshold.

Total notification from hardware: 224
Notified to user space: 16

So did you apply the Daniel's patches and run the test or did you do something else?

We already use netlink to send notification to user space via writable trip. So didn't apply any patches. I don't see these patches will do any different.

Quoted 36 lines
Quoted 32 lines
Quoted 19 lines
Quoted 12 lines

We have a whole filtering in driver to avoid this. You need a rate limit here.

There are multiple user processes can add threshold and there is no
ownership. So one process can cause too much noise to others as it
is
multicast.

Quoted 6 lines

We worked on a change to filer these as discussed during last LPC,
but
not posted yet. This will really need this as this will be too many
messages.

Quoted 4 lines

Unless there is a way to limit your subscription to events regarding
a
specific thermal zone, for instance.

That's what this change does.

Quoted 6 lines

Anyway, I have to admit ignorance regarding the user space usage model
related to this.  For example, is it expected that there will be one
user space entity managing the thresholds or there can be many.

Quoted 3 lines

Good question. That's why we didn't send this change. If there is one
agent this is fine.

Quoted 6 lines

I think that the addition and deletion of a user threshold should be privileged operations.

If that is the case, it all boils down to proper coordination in user space.

Some kernel API use some cookie as parameter. Here also something like that can be used to match.

Quoted 5 lines
Quoted 3 lines

But in embedded space, there may be more than one. Hence they needed
netlink multicast instead of just one char device.

I see.

Thanks,
Srinivas

#22 lore raw
Quoted 159 lines · Srinivas
Quoted 139 lines · Rafael J. Wysocki
Quoted 137 lines · Srinivas
Quoted 134 lines · Rafael J. Wysocki

Hi Srinivas,

Quoted 107 lines · Srinivas

Hi Daniel,

Quoted 86 lines · Daniel Lezcano

The trip points are a firmware description of the temperature
limits
of a specific thermal zone where we associate an action whichcan't
is
done
by the kernel. The time resolution is low.

The userspace has to deal with a more complex thermal management
based
on heuristics from different information coming from different
places. The logic is much more complex but based on a bigger
time
resolution, usually one second based.

The purpose of the userspace is to monitor the temperatures from
different places and take actions. However, it can not be constantly
reading the temperature to detect when a temperature threshold
has
been reached. This is especially bad for mobile or embedded system as
that will lead to an unacceptable number of wakeup to check the
temperature with nothing to do.

On the other side, the sensors are now most of the time interrupt
driven. That means the thermal framework will use the temperature
trip
points to program the sensor to trigger an interrupt when a temperature limit is crossed.

Unfortunately, the userspace can not benefit this feature and
current
solutions found here and there, iow out-of-tree, are to add fake
trip
points in the firmware and enable the writable trip points.

This is bad for different reasons, the trip points are for in-
kernel
actions, the semantic of their types is used by the thermal framework
and by adding trip points in the device tree is a way to overcome
the
current limitation but tampering with how the thermal framework
is
supposed to work. The writable trip points is a way to adjust
a
temperature limit given a specific platform if the firmware is
not
accurate enough and TBH it is more a debug feature from my POV.

The thresholds mechanism is a way to have the userspace to tell
thermal framework to send a notification when a temperature limit
is
crossed. There is no id, no hysteresis, just the temperature
and
the
direction of the limit crossing. That means we can be notified
when a
threshold is crossed the way up only, or the way down only or
both
ways. That allows to create hysteresis values if it is needed.

Quoted 5 lines

A threshold can be added, deleted or flushed. The latter means
all
thresholds belonging to a thermal zone will be deleted.

So you are proposing to add threshold via netlink, not adding any
new
sysfs attribute? That is not clear here.

I think you are adding"
THERMAL_GENL_CMD_THRESHOLD_GET
THERMAL_GENL_CMD_THRESHOLD_ADD
THERMAL_GENL_CMD_THRESHOLD_DELETE
THERMAL_GENL_CMD_THRESHOLD_FLUSH

Quoted 6 lines

We need to document our netlink messages including old ones.

Also we should add "MODIFY" as we tend to change them quite often.

Also no hysteresis, that is practically we can't use.

The direction thing is equivalent to hysteresis though.

Instead of using one threshold with a given hysteresis value, use
two
of them with different temperature values and different directions.

Quoted 5 lines

Temperature changes so much that that will flood user space. You
will get 100s of
events on CPU temperature if you set temperature threshold in CPU.

Events only trigger when thresholds are crossed in a specific direction, but overall you have a point.

Quoted 8 lines
Quoted 2 lines

This is good enough for some sensor which changes slowly like skin.

Right, and I think that this is the driving use case.

Quoted 1 line

But some sensors like CPU and board are not like that.

Sure, so this interface will not be for them.

But when we design an interface, I think this shouldn't be sensor dependent.

Quoted 8 lines
Quoted 7 lines

We publish both raw and filtered event count in debugfs for x86 package temp.
For example, I just ran a sample on a client for one threshold.

Total notification from hardware: 224
Notified to user space: 16

Quoted 7 lines
Quoted 3 lines

So did you apply the Daniel's patches and run the test or did you do
something else?

We already use netlink to send notification to user space via writable
trip. So didn't apply any patches. I don't see these patches will do any different.

I mean without rate limit here all 224 notifications will be passed to user space (not 16). Temperature goes up and down over threshold for 100s of times before settling.

I can give a try with this series to confirm.

Thanks,
Srinivas

Quoted 63 lines
Quoted 42 lines
Quoted 38 lines
Quoted 23 lines
Quoted 15 lines

We have a whole filtering in driver to avoid this. You need a rate limit here.

There are multiple user processes can add threshold and there is no
ownership. So one process can cause too much noise to others as
it
is
multicast.

Quoted 7 lines

We worked on a change to filer these as discussed during last LPC,
but
not posted yet. This will really need this as this will be too
many
messages.

Quoted 4 lines

Unless there is a way to limit your subscription to events regarding
a
specific thermal zone, for instance.

That's what this change does.

Quoted 8 lines

Anyway, I have to admit ignorance regarding the user space usage
model
related to this.  For example, is it expected that there will be
one
user space entity managing the thresholds or there can be many.

Quoted 3 lines

Good question. That's why we didn't send this change. If there is one
agent this is fine.

Quoted 8 lines

I think that the addition and deletion of a user threshold should be
privileged operations.

If that is the case, it all boils down to proper coordination in user
space.

Some kernel API use some cookie as parameter. Here also something like
that can be used to match.

Quoted 5 lines
Quoted 3 lines

But in embedded space, there may be more than one. Hence they needed
netlink multicast instead of just one char device.

I see.

Thanks,
Srinivas

#23 lore raw
Quoted 164 lines · Srinivas
Quoted 162 lines · Srinivas
Quoted 139 lines · Rafael J. Wysocki
Quoted 137 lines · Srinivas
Quoted 134 lines · Rafael J. Wysocki

Hi Srinivas,

Quoted 107 lines · Srinivas

Hi Daniel,

Quoted 86 lines · Daniel Lezcano

The trip points are a firmware description of the temperature
limits
of a specific thermal zone where we associate an action whichcan't
is
done
by the kernel. The time resolution is low.

The userspace has to deal with a more complex thermal management
based
on heuristics from different information coming from different
places. The logic is much more complex but based on a bigger
time
resolution, usually one second based.

The purpose of the userspace is to monitor the temperatures from
different places and take actions. However, it can not be constantly
reading the temperature to detect when a temperature threshold
has
been reached. This is especially bad for mobile or embedded system as
that will lead to an unacceptable number of wakeup to check the
temperature with nothing to do.

On the other side, the sensors are now most of the time interrupt
driven. That means the thermal framework will use the temperature
trip
points to program the sensor to trigger an interrupt when a temperature limit is crossed.

Unfortunately, the userspace can not benefit this feature and
current
solutions found here and there, iow out-of-tree, are to add fake
trip
points in the firmware and enable the writable trip points.

This is bad for different reasons, the trip points are for in-
kernel
actions, the semantic of their types is used by the thermal framework
and by adding trip points in the device tree is a way to overcome
the
current limitation but tampering with how the thermal framework
is
supposed to work. The writable trip points is a way to adjust
a
temperature limit given a specific platform if the firmware is
not
accurate enough and TBH it is more a debug feature from my POV.

The thresholds mechanism is a way to have the userspace to tell
thermal framework to send a notification when a temperature limit
is
crossed. There is no id, no hysteresis, just the temperature
and
the
direction of the limit crossing. That means we can be notified
when a
threshold is crossed the way up only, or the way down only or
both
ways. That allows to create hysteresis values if it is needed.

Quoted 5 lines

A threshold can be added, deleted or flushed. The latter means
all
thresholds belonging to a thermal zone will be deleted.

So you are proposing to add threshold via netlink, not adding any
new
sysfs attribute? That is not clear here.

I think you are adding"
THERMAL_GENL_CMD_THRESHOLD_GET
THERMAL_GENL_CMD_THRESHOLD_ADD
THERMAL_GENL_CMD_THRESHOLD_DELETE
THERMAL_GENL_CMD_THRESHOLD_FLUSH

Quoted 6 lines

We need to document our netlink messages including old ones.

Also we should add "MODIFY" as we tend to change them quite often.

Also no hysteresis, that is practically we can't use.

The direction thing is equivalent to hysteresis though.

Instead of using one threshold with a given hysteresis value, use
two
of them with different temperature values and different directions.

Quoted 5 lines

Temperature changes so much that that will flood user space. You
will get 100s of
events on CPU temperature if you set temperature threshold in CPU.

Events only trigger when thresholds are crossed in a specific direction, but overall you have a point.

Quoted 8 lines
Quoted 2 lines

This is good enough for some sensor which changes slowly like skin.

Right, and I think that this is the driving use case.

Quoted 1 line

But some sensors like CPU and board are not like that.

Sure, so this interface will not be for them.

But when we design an interface, I think this shouldn't be sensor dependent.

Quoted 11 lines
Quoted 7 lines

We publish both raw and filtered event count in debugfs for x86 package temp.
For example, I just ran a sample on a client for one threshold.

Total notification from hardware: 224
Notified to user space: 16

So did you apply the Daniel's patches and run the test or did you do
something else?

Quoted 8 lines
Quoted 4 lines

We already use netlink to send notification to user space via writable
trip. So didn't apply any patches. I don't see these patches will do any different.

I mean without rate limit here all 224 notifications will be passed to user space (not 16). Temperature goes up and down over threshold for 100s of times before settling.

Yeah, that needs a real trip point with hysteresis.

Or the threshold could be removed once crossed and then added back when the temperature falls down sufficiently.

Quoted 1 line

I can give a try with this series to confirm.

That would be premature at this point.

#24 lore raw
In reply to Pandruvada, Srinivas
Quoted 3 lines · Srinivas
Quoted 2 lines · Rafael J. Wysocki

On Thu, Aug 22, 2024 at 12:16 AM Pandruvada, Srinivas <[email protected]> wrote:

[ ... ]

Quoted 5 lines
Quoted 2 lines

So did you apply the Daniel's patches and run the test or did you do something else?

We already use netlink to send notification to user space via writable trip. So didn't apply any patches. I don't see these patches will do any different.

Actually you are missing the point of the thresholds approach.

The goal is to track the temperature easily from userspace without constantly polling the temperatures in all the places.

The trip points are a firmware descriptions. Their number is fixed. They are designed for in-kernel thermal framework. They have a type. A governor is supposed to be tied with it. A cooling device also.

Writable trip points means you should be able to add trip points dedicated to the userspace to the firmware which is not possible. Then reuse them from userspace which is unrelated to the in-kernel thermal management. It is difficult to deal with because of the need of tracking the low and high limits from userspace.

The thresholds are there to allow the userspace to have the benefit of the interrupt driven temperature monitoring.

Obviously if you set the thresholds to a temperature equal to a mitigation trip point then when the limit is reached you will receive notifications for the trip point going back and forth as well as notifications for the thresholds.

Signature · 5 lines
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs

Follow Linaro:  <http://www.facebook.com/pages/Linaro> Facebook |
<http://twitter.com/#!/linaroorg> Twitter |
<http://www.linaro.org/linaro-blog/> Blog
#25 lore raw
Quoted 10 lines · Daniel Lezcano
Quoted 3 lines · Srinivas
Quoted 2 lines · Rafael J. Wysocki

On Thu, Aug 22, 2024 at 12:16 AM Pandruvada, Srinivas <[email protected]> wrote:

[ ... ]

Quoted 3 lines
Quoted 3 lines

So did you apply the Daniel's patches and run the test or did you do
something else?

Quoted 7 lines
Quoted 5 lines

We already use netlink to send notification to user space via writable
trip. So didn't apply any patches. I don't see these patches will do
any different.

Actually you are missing the point of the thresholds approach.

I very well understand. I proposed similar approach without netlink several years back. Also submitted patches to use IIO.

Quoted 4 lines

The goal is to track the temperature easily from userspace without constantly polling the temperatures in all the places.

Exactly.

Quoted 5 lines

The trip points are a firmware descriptions. Their number is fixed. They
are designed for in-kernel thermal framework. They have a type. A governor is supposed to be tied with it. A cooling device also.

Yes, I understand the whole approach. trips are trip where you want governors to take action.

Quoted 8 lines

Writable trip points means you should be able to add trip points dedicated to the userspace to the firmware which is not possible. Then
reuse them from userspace which is unrelated to the in-kernel thermal management. It is difficult to deal with because of the need of tracking
the low and high limits from userspace.

Yes.

Quoted 4 lines

The thresholds are there to allow the userspace to have the benefit of
the interrupt driven temperature monitoring.

Yes. I am not denying benefits.

Quoted 5 lines

Obviously if you set the thresholds to a temperature equal to a mitigation trip point then when the limit is reached you will receive notifications for the trip point going back and forth as well as notifications for the thresholds.

Not correct. It is not at mitigation point. This is how several sensors several sensors work. I submitted a range of sensor drivers to IIO framework, several have this issue.

To make it useful, you need to have some rate limiting. Netlink is not a low overhead user-kernel interface.

Looks like you finalized design and just looking for patch reviews!

Quoted 4 lines
#26 lore raw
In reply to Pandruvada, Srinivas

Hi Srinivas,

Quoted 52 lines · Srinivas

Hi Daniel,

Quoted 42 lines · Daniel Lezcano

The trip points are a firmware description of the temperature limits of a specific thermal zone where we associate an action which is done by the kernel. The time resolution is low.

The userspace has to deal with a more complex thermal management based
on heuristics from different information coming from different places. The logic is much more complex but based on a bigger time resolution, usually one second based.

The purpose of the userspace is to monitor the temperatures from different places and take actions. However, it can not be constantly reading the temperature to detect when a temperature threshold has been reached. This is especially bad for mobile or embedded system as that will lead to an unacceptable number of wakeup to check the temperature with nothing to do.

On the other side, the sensors are now most of the time interrupt driven. That means the thermal framework will use the temperature trip
points to program the sensor to trigger an interrupt when a temperature limit is crossed.

Unfortunately, the userspace can not benefit this feature and current solutions found here and there, iow out-of-tree, are to add fake trip points in the firmware and enable the writable trip points.

This is bad for different reasons, the trip points are for in-kernel actions, the semantic of their types is used by the thermal framework and by adding trip points in the device tree is a way to overcome the current limitation but tampering with how the thermal framework is supposed to work. The writable trip points is a way to adjust a temperature limit given a specific platform if the firmware is not accurate enough and TBH it is more a debug feature from my POV.

The thresholds mechanism is a way to have the userspace to tell thermal framework to send a notification when a temperature limit is crossed. There is no id, no hysteresis, just the temperature and the direction of the limit crossing. That means we can be notified when a threshold is crossed the way up only, or the way down only or both ways. That allows to create hysteresis values if it is needed.

Quoted 3 lines

A threshold can be added, deleted or flushed. The latter means all thresholds belonging to a thermal zone will be deleted.

So you are proposing to add threshold via netlink, not adding any new sysfs attribute? That is not clear here.

Quoted 7 lines

I think you are adding"
THERMAL_GENL_CMD_THRESHOLD_GET
THERMAL_GENL_CMD_THRESHOLD_ADD
THERMAL_GENL_CMD_THRESHOLD_DELETE
THERMAL_GENL_CMD_THRESHOLD_FLUSH

We need to document our netlink messages including old ones.

We can do that but I would prefer to do that in a separate series.

Quoted 1 line

Also we should add "MODIFY" as we tend to change them quite often.

We can not change a threshold, only delete or add it.

Quoted 2 lines

Also no hysteresis, that is practically we can't use. Temperature changes so much that that will flood user space.

A threshold with a direction allows to create an hysteresis. Instead of one threshold, you will have two with different directions and temperature.

Quoted 4 lines

You will get 100s of
events on CPU temperature if you set temperature threshold in CPU. We have a whole filtering in driver to avoid this.
You need a rate limit here.

I disagree, the user space is not supposed to monitor at a high rate a particular thermal zone. If the CPU is sending 100s notifications per second that means the threshold is set close to the mitigation trip point temperature which is a bad user space configuration.

Quoted 3 lines

There are multiple user processes can add threshold and there is no ownership. So one process can cause too much noise to others as it is multicast.

ownership is something we may want to handle later.

What about restricting the thermal netlink to the 'root' user only ATM?

Quoted 3 lines

We worked on a change to filer these as discussed during last LPC, but not posted yet. This will really need this as this will be too many messages.

With a correct configuration, that would not happen because it is different from receiving events from the trip points.

Signature · 5 lines
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs

Follow Linaro:  <http://www.facebook.com/pages/Linaro> Facebook |
<http://twitter.com/#!/linaroorg> Twitter |
<http://www.linaro.org/linaro-blog/> Blog