[RFC v02 0/5] Power Capping Framework and RAPL Driver

[RFC v02 0/5] Power Capping Framework and RAPL Driver

[Srinivas Pandruvada]

Overview
With the evolution of technologies, which enables power monitoring and limiting,
more and more devices are able to constrain their power consumption under certain
limits. There are several use cases for such technologies:
- Power monitoring: Each device can report its power consumption.
- Power Limiting: Setting power limits on the devices allows users to guard against
platform reaching max system power level.
- Maximize performance: While staying below a power limit, it allows devices to
automatically adjust performance to meet demands
- Dynamic control and re-budgeting: If each device can be constrained to some power,
extra power can redistributed to other devices, which needs additional performance.

One such example of technology is RAPL (Running Average Power Limit) mechanism
available in the latest Intel Processors. Intel is slowly adding many devices under
RAPL control. Also there are other technologies available, for power capping various
devices. Soon it is very likely that other vendors are also adding or considering
such implementation.

Power Capping framework is an effort to have a uniform interface available to Linux
drivers, which will enable
- A uniform sysfs interface for all devices which can offer power capping
- A common API for drivers, which will avoid code duplication and easy
implementation of client drivers.

Also submitting Intel RAPL driver using power capping framework.


Revisions:
v02:
Sign-offs and reviewed-by tags
Stylistic issues suggested by Joe Perches
Removed "counter" from power_uw documentation as pointed by Jonathan Corbet
Submitting Intel RAPL driver using power capping framework

v01:
Use device model only to register zones and controllers.

v00:
Presented options

Jacob Pan (2):
  x86/msr: add 64bit _on_cpu access functions
  Introduce Intel RAPL power capping driver

Srinivas Pandruvada (3):
  PowerCap: Documentation
  PowerCap: Add class driver
  PowerCap: Added to drivers build

 Documentation/ABI/testing/sysfs-class-powercap   |  165 +++
 Documentation/powercap/PowerCappingFramework.txt |  686 ++++++++++++
 arch/x86/include/asm/msr.h                       |   22 +
 arch/x86/lib/msr-smp.c                           |   62 +
 drivers/Kconfig                                  |    2 +
 drivers/Makefile                                 |    1 +
 drivers/powercap/Kconfig                         |   24 +
 drivers/powercap/Makefile                        |    6 +
 drivers/powercap/intel_rapl.c                    | 1305 ++++++++++++++++++++++
 drivers/powercap/powercap_sys.c                  |  995 +++++++++++++++++
 include/linux/powercap.h                         |  300 +++++
 11 files changed, 3568 insertions(+)
 create mode 100644 Documentation/ABI/testing/sysfs-class-powercap
 create mode 100644 Documentation/powercap/PowerCappingFramework.txt
 create mode 100644 drivers/powercap/Kconfig
 create mode 100644 drivers/powercap/Makefile
 create mode 100644 drivers/powercap/intel_rapl.c
 create mode 100644 drivers/powercap/powercap_sys.c
 create mode 100644 include/linux/powercap.h

-- 
1.8.3.1

[RFC v02 1/5] PowerCap: Documentation

[Srinivas Pandruvada]

Added power cap framework documentation. This explains the use of power capping
framework, sysfs and programming interface.
There are two documents:
Documentation/powercap/PowerCappingFramework.txt: Explains use case and API in
details.
Documentation/ABI/testing/sysfs-class-powercap: Explains ABIs.

Reviewed-by: Len Brown <len.brown@intel.com>
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
---
 Documentation/ABI/testing/sysfs-class-powercap   | 165 ++++++
 Documentation/powercap/PowerCappingFramework.txt | 686 +++++++++++++++++++++++
 2 files changed, 851 insertions(+)
 create mode 100644 Documentation/ABI/testing/sysfs-class-powercap
 create mode 100644 Documentation/powercap/PowerCappingFramework.txt

diff --git a/Documentation/ABI/testing/sysfs-class-powercap b/Documentation/ABI/testing/sysfs-class-powercap
new file mode 100644
index 0000000..0e5d6e4
--- /dev/null
+++ b/Documentation/ABI/testing/sysfs-class-powercap
@@ -0,0 +1,165 @@
+What:		/sys/class/power_cap/
+Date:		August 2013
+KernelVersion:	3.12
+Contact:	linux-pm@vger.kernel.org
+Description:
+		The power_cap/ class sub directory belongs to the power cap
+		subsystem. Refer to
+		Documentation/powercap/PowerCappingFramework.txt for details.
+
+What:		/sys/class/power_cap/controller_name
+Date:		August 2013
+KernelVersion:	3.12
+Contact:	linux-pm@vger.kernel.org
+Description:
+		The /sys/class/power_cap/controller_name directories correspond
+		to each controller under power_cap control. Here controller_name
+		is a unique name under /sys/class_power_cap. Each
+		controller_name directory contains one or more power zones.
+
+What:		/sys/class/power_cap/controller_name/type
+Date:		August 2013
+KernelVersion:	3.12
+Contact:	linux-pm@vger.kernel.org
+Description:
+		For controller type is "controller". This allows user space
+		to differentiate between a controller device from a power zone
+		device.
+
+What:		/sys/class/power_cap/controller_name/power_zone
+Date:		August 2013
+KernelVersion:	3.12
+Contact:	linux-pm@vger.kernel.org
+Description:
+		A Controller can have one or more power zones. A power zone is
+		an abstraction of devices, which can be independently monitored
+		and controlled.
+
+What:		/sys/class/power_cap/controller_name/power_zone/power_zone
+Date:		August 2013
+KernelVersion:	3.12
+Contact:	linux-pm@vger.kernel.org
+Description:
+		A power zone can have one or more power zones as children.
+		This child power zone provides monitoring and control for
+		a subset of device under parent. E.g. if there is parent
+		power zone for whole CPU package, each CPU cores in it can be
+		a child power zone.
+
+What:		/sys/class/power_cap/controller_name/power_zone/name
+Date:		August 2013
+KernelVersion:	3.12
+Contact:	linux-pm@vger.kernel.org
+Description:
+		Specifies the name of this power zone.
+
+
+What:		/sys/class/power_cap/controller_name/power_zone/type
+Date:		August 2013
+KernelVersion:	3.12
+Contact:	linux-pm@vger.kernel.org
+Description:
+		For power zone type is "power-zone".
+
+
+What:		/sys/class/power_cap/controller_name/power_zone/energy_uj
+Date:		August 2013
+KernelVersion:	3.12
+Contact:	linux-pm@vger.kernel.org
+Description:
+		Current energy counter in micro-joules. Write "0" to reset.
+		If the counter can not be reset, then this attribute is
+		read-only.
+
+What:		/sys/class/power_cap/controller_name/power_zone/
+							max_energy_range_uj
+Date:		August 2013
+KernelVersion:	3.12
+Contact:	linux-pm@vger.kernel.org
+Description:
+		Range of the above energy counter in micro-joules.
+
+
+What:		/sys/class/power_cap/controller_name/power_zone/power_uw
+Date:		August 2013
+KernelVersion:	3.12
+Contact:	linux-pm@vger.kernel.org
+Description:
+		Current power in micro-watts. Write "0" to reset.
+		If the value can not be reset, then the attribute is read
+		only.
+
+What:		/sys/class/power_cap/controller_name/power_zone/
+							max_power_range_uw
+Date:		August 2013
+KernelVersion:	3.12
+Contact:	linux-pm@vger.kernel.org
+Description:
+		Range of the above power value in micro-watts.
+
+What:		/sys/class/power_cap/controller_name/power_zone/
+							constraint_X_name
+Date:		August 2013
+KernelVersion:	3.12
+Contact:	linux-pm@vger.kernel.org
+Description:
+		Each power zone can define one or more constraints. Each
+		constraint can have optional name. Here "X" can have values
+		from 0 to max integer.
+
+What:		/sys/class/power_cap/controller_name/power_zone/
+						constraint_X_power_limit_uw
+Date:		August 2013
+KernelVersion:	3.12
+Contact:	linux-pm@vger.kernel.org
+Description:
+		Power limit in micro-watts, which should be applicable for
+		the time window specified by "constraint_X_time_window_us".
+		Here "X" can have values from 0 to max integer.
+
+What:		/sys/class/power_cap/controller_name/power_zone/
+						constraint_X_time_window_us
+Date:		August 2013
+KernelVersion:	3.12
+Contact:	linux-pm@vger.kernel.org
+Description:
+		Time window in micro seconds. This is used along with
+		constraint_X_power_limit_uw to define a power constraint.
+		Here "X" can have values from 0 to max integer.
+
+
+What:		/sys/class/power_cap/controller_name/power_zone/
+						constraint_X_max_power_uw
+Date:		August 2013
+KernelVersion:	3.12
+Contact:	linux-pm@vger.kernel.org
+Description:
+		Maximum allowed power in micro watts for this constraint.
+		Here "X" can have values from 0 to max integer.
+
+What:		/sys/class/power_cap/controller_name/power_zone/
+						constraint_X_min_power_uw
+Date:		August 2013
+KernelVersion:	3.12
+Contact:	linux-pm@vger.kernel.org
+Description:
+		Minimum allowed power in micro watts for this constraint.
+		Here "X" can have values from 0 to max integer.
+
+What:		/sys/class/power_cap/controller_name/power_zone/
+						constraint_X_max_time_window_us
+Date:		August 2013
+KernelVersion:	3.12
+Contact:	linux-pm@vger.kernel.org
+Description:
+		Maximum allowed time window in micro seconds for this
+		constraint. Here "X" can have values from 0 to max integer.
+
+What:		/sys/class/power_cap/controller_name/power_zone/
+						constraint_X_min_time_window_us
+Date:		August 2013
+KernelVersion:	3.12
+Contact:	linux-pm@vger.kernel.org
+Description:
+		Minimum allowed time window in micro seconds for this
+		constraint. Here "X" can have values from 0 to max integer.
diff --git a/Documentation/powercap/PowerCappingFramework.txt b/Documentation/powercap/PowerCappingFramework.txt
new file mode 100644
index 0000000..8d9c02ee
--- /dev/null
+++ b/Documentation/powercap/PowerCappingFramework.txt
@@ -0,0 +1,686 @@
+Power Capping Framework
+==================================
+
+The Linux Power Capping Framework provides user-space with a common
+API to kernel-mode power-capping drivers.  At the same time,
+it provides the hardware-specific power-capping drivers with
+a uniform API to user-space.
+
+Terminology
+=========================
+The Power Capping framework organizes power capping devices under a tree structure.
+At the root level, each device is under some "controller", which is the enabler
+of technology. For example this can be "RAPL".
+Under each controllers, there are multiple power zones, which can be independently
+monitored and controlled.
+Each power zone can be organized as a tree with parent, children and siblings.
+Each power zone defines attributes to enable power monitoring and constraints.
+
+Example sysfs interface tree:
+
+/sys/devices/virtual/power_cap
+└── intel-rapl
+    ├── intel-rapl:0
+    │   ├── constraint_0_name
+    │   ├── constraint_0_power_limit_uw
+    │   ├── constraint_0_time_window_us
+    │   ├── constraint_1_name
+    │   ├── constraint_1_power_limit_uw
+    │   ├── constraint_1_time_window_us
+    │   ├── device -> ../../intel-rapl
+    │   ├── energy_uj
+    │   ├── intel-rapl:0:0
+    │   │   ├── constraint_0_name
+    │   │   ├── constraint_0_power_limit_uw
+    │   │   ├── constraint_0_time_window_us
+    │   │   ├── constraint_1_name
+    │   │   ├── constraint_1_power_limit_uw
+    │   │   ├── constraint_1_time_window_us
+    │   │   ├── device -> ../../intel-rapl:0
+    │   │   ├── energy_uj
+    │   │   ├── max_energy_range_uj
+    │   │   ├── name
+    │   │   ├── power
+    │   │   │   ├── async
+    │   │   │   ├── autosuspend_delay_ms
+    │   │   │   ├── control
+    │   │   │   ├── runtime_active_kids
+    │   │   │   ├── runtime_active_time
+    │   │   │   ├── runtime_enabled
+    │   │   │   ├── runtime_status
+    │   │   │   ├── runtime_suspended_time
+    │   │   │   └── runtime_usage
+    │   │   ├── subsystem -> ../../../../../../class/power_cap
+    │   │   ├── type
+    │   │   └── uevent
+    │   ├── intel-rapl:0:1
+    │   │   ├── constraint_0_name
+    │   │   ├── constraint_0_power_limit_uw
+    │   │   ├── constraint_0_time_window_us
+    │   │   ├── constraint_1_name
+    │   │   ├── constraint_1_power_limit_uw
+    │   │   ├── constraint_1_time_window_us
+    │   │   ├── device -> ../../intel-rapl:0
+    │   │   ├── energy_uj
+    │   │   ├── max_energy_range_uj
+    │   │   ├── name
+    │   │   ├── power
+    │   │   │   ├── async
+    │   │   │   ├── autosuspend_delay_ms
+    │   │   │   ├── control
+    │   │   │   ├── runtime_active_kids
+    │   │   │   ├── runtime_active_time
+    │   │   │   ├── runtime_enabled
+    │   │   │   ├── runtime_status
+    │   │   │   ├── runtime_suspended_time
+    │   │   │   └── runtime_usage
+    │   │   ├── subsystem -> ../../../../../../class/power_cap
+    │   │   ├── type
+    │   │   └── uevent
+    │   ├── max_energy_range_uj
+    │   ├── max_power_range_uw
+    │   ├── name
+    │   ├── power
+    │   │   ├── async
+    │   │   ├── autosuspend_delay_ms
+    │   │   ├── control
+    │   │   ├── runtime_active_kids
+    │   │   ├── runtime_active_time
+    │   │   ├── runtime_enabled
+    │   │   ├── runtime_status
+    │   │   ├── runtime_suspended_time
+    │   │   └── runtime_usage
+    │   ├── subsystem -> ../../../../../class/power_cap
+    │   ├── type
+    │   └── uevent
+    ├── intel-rapl:1
+    │   ├── constraint_0_name
+    │   ├── constraint_0_power_limit_uw
+    │   ├── constraint_0_time_window_us
+    │   ├── constraint_1_name
+    │   ├── constraint_1_power_limit_uw
+    │   ├── constraint_1_time_window_us
+    │   ├── device -> ../../intel-rapl
+    │   ├── energy_uj
+    │   ├── intel-rapl:1:0
+    │   │   ├── constraint_0_name
+    │   │   ├── constraint_0_power_limit_uw
+    │   │   ├── constraint_0_time_window_us
+    │   │   ├── constraint_1_name
+    │   │   ├── constraint_1_power_limit_uw
+    │   │   ├── constraint_1_time_window_us
+    │   │   ├── device -> ../../intel-rapl:1
+    │   │   ├── energy_uj
+    │   │   ├── max_energy_range_uj
+    │   │   ├── name
+    │   │   ├── power
+    │   │   │   ├── async
+    │   │   │   ├── autosuspend_delay_ms
+    │   │   │   ├── control
+    │   │   │   ├── runtime_active_kids
+    │   │   │   ├── runtime_active_time
+    │   │   │   ├── runtime_enabled
+    │   │   │   ├── runtime_status
+    │   │   │   ├── runtime_suspended_time
+    │   │   │   └── runtime_usage
+    │   │   ├── subsystem -> ../../../../../../class/power_cap
+    │   │   ├── type
+    │   │   └── uevent
+    │   ├── intel-rapl:1:1
+    │   │   ├── constraint_0_name
+    │   │   ├── constraint_0_power_limit_uw
+    │   │   ├── constraint_0_time_window_us
+    │   │   ├── constraint_1_name
+    │   │   ├── constraint_1_power_limit_uw
+    │   │   ├── constraint_1_time_window_us
+    │   │   ├── device -> ../../intel-rapl:1
+    │   │   ├── energy_uj
+    │   │   ├── max_energy_range_uj
+    │   │   ├── name
+    │   │   ├── power
+    │   │   │   ├── async
+    │   │   │   ├── autosuspend_delay_ms
+    │   │   │   ├── control
+    │   │   │   ├── runtime_active_kids
+    │   │   │   ├── runtime_active_time
+    │   │   │   ├── runtime_enabled
+    │   │   │   ├── runtime_status
+    │   │   │   ├── runtime_suspended_time
+    │   │   │   └── runtime_usage
+    │   │   ├── subsystem -> ../../../../../../class/power_cap
+    │   │   ├── type
+    │   │   └── uevent
+    │   ├── max_energy_range_uj
+    │   ├── max_power_range_uw
+    │   ├── name
+    │   ├── powerintel-rapl:0
+    │   │   ├── async
+    │   │   ├── autosuspend_delay_ms
+    │   │   ├── control
+    │   │   ├── runtime_active_kids
+    │   │   ├── runtime_active_time
+    │   │   ├── runtime_enabled
+    │   │   ├── runtime_status
+    │   │   ├── runtime_suspended_time
+    │   │   └── runtime_usage
+    │   ├── subsystem -> ../../../../../class/power_cap
+    │   ├── type
+    │   └── uevent
+    ├── power
+    │   ├── async
+    │   ├── autosuspend_delay_ms
+    │   ├── control
+    │   ├── runtime_active_kids
+    │   ├── runtime_active_time
+    │   ├── runtime_enabled
+    │   ├── runtime_status
+    │   ├── runtime_suspended_time
+    │   └── runtime_usage
+    ├── subsystem -> ../../../../class/power_cap
+    ├── type
+    └── uevent
+
+
+For example, above powercap sysfs tree represents RAPL(Running Average Power Limit)
+type controls available in the Intel® 64 and IA-32 Processor Architectures. Here
+under controller "intel-rapl" there are two CPU packages (package-0/1), which can
+provide power monitoring and controls (intel-rapl:0 and intel-rapl:1). Each power
+zone has a name.
+For example:
+cat /sys/class/power_cap/intel-rapl/intel-rapl:0/name
+package-0
+
+In addition to providing monitoring and control at package level, each package
+is further divided into child power zones (called domains in the RAPL specifications).
+Here zones represent controls for core and dram  parts. These zones can be represented
+as children of package.
+For example:
+cat /sys/class/power_cap/intel-rapl/intel-rapl:0/intel-rapl:0:1/name
+dram
+
+Under RAPL framework there are two constraints, one for
+short term and one for long term, with two different time windows. These can be
+represented as two constraints, with different time windows, power limits and names.
+For example:
+	constraint_0_name
+	constraint_0_power_limit_uw
+	constraint_0_time_window_us
+	constraint_1_name
+	constraint_1_power_limit_uw
+	constraint_1_time_window_us
+
+Power Zone Attributes
+=================================
+Monitoring attributes
+----------------------
+
+energy_uj (rw): Current energy counter in micro joules. Write "0" to reset.
+If the counter can not be reset, then this attribute is read only.
+
+max_energy_range_uj (ro): Range of the above energy counter in micro-joules.
+
+power_uw (rw): Current power in micro watts. Write "0" to resets the value.
+If the value can not be reset, then this attribute is read only.
+
+max_power_range_uw (ro): Range of the above power value in micro-watts.
+
+name (ro): Name of this power zone.
+
+It is possible that some domains can have both power and energy counters and
+ranges, but at least one is mandatory.
+
+Constraints
+----------------
+constraint_X_power_limit_uw (rw): Power limit in micro watts, which should be
+applicable for the time window specified by "constraint_X_time_window_us".
+
+constraint_X_time_window_us (rw): Time window in micro seconds.
+
+constraint_X_name (ro): An optional name of the constraint
+
+constraint_X_max_power_uw(ro): Maximum allowed power in micro watts.
+
+constraint_X_min_power_uw(ro): Minimum allowed power in micro watts.
+
+constraint_X_max_time_window_us(ro): Maximum allowed time window in micro seconds.
+
+constraint_X_min_time_window_us(ro): Minimum allowed time window in micro seconds.
+
+In addition each node has an attribute "type", which shows, whether is a controller
+or power zone. Except power_limit_uw and time_window_us other fields are optional.
+
+Power Cap Client Driver Interface
+==================================
+The API summary:
+
+Call powercap_allocate_controller to define a controller with a name.
+Call powercap_zone_register for each power zone for this controller.
+power zones can have other power zone as a parent or don't have a
+parent.
+During powercap_zone_register defines number of constraints and callbacks.
+
+To Free a power zone call powercap_zone_unregister.
+To free a controller call powercap_deallocate_controller.
+
+Rest of this document is generated by using kernel-doc on
+powercap.h
+
+struct powercap_zone_constraint_ops - Define constraint callbacks
+
+struct powercap_zone_constraint_ops {
+	int (* set_power_limit_uw) (struct powercap_zone_device *, int, u64);
+	int (* get_power_limit_uw) (struct powercap_zone_device *, int, u64 *);
+	int (* set_time_window_us) (struct powercap_zone_device *, int, u64);
+	int (* get_time_window_us) (struct powercap_zone_device *, int, u64 *);
+	int (* get_max_power_uw) (struct powercap_zone_device *, int, u64 *);
+	int (* get_min_power_uw) (struct powercap_zone_device *, int, u64 *);
+	int (* get_max_time_window_us) (struct powercap_zone_device *, int, u64 *);
+	int (* get_min_time_window_us) (struct powercap_zone_device *, int, u64 *);
+	const char *(* get_name) (struct powercap_zone_device *, int);
+	void (* cleanup) (struct powercap_zone_device *, int);
+};
+
+Members:
+
+set_power_limit_uw
+	Set power limit in micro-watts.
+
+get_power_limit_uw
+	Get power limit in micro-watts.
+
+set_time_window_us
+	Set time window in micro-seconds.
+
+get_time_window_us
+	Get time window in micro-seconds.
+
+get_max_power_uw
+	Get max power allowed in micro-watts.
+
+get_min_power_uw
+	Get min power allowed in micro-watts.
+
+get_max_time_window_us
+	Get max time window allowed in micro-seconds.
+
+get_min_time_window_us
+	Get min time window allowed in micro-seconds.
+
+get_name
+	Get the name of constraint
+
+cleanup
+	Do any clean up before the constraint is freed
+This structure is used to define the constraint callbacks for the client
+drivers. The following callbacks are mandatory and can't be NULL:
+ set_power_limit_uw
+ get_power_limit_uw
+ set_time_window_us
+ get_time_window_us
+ get_name
+
+
+
+
+
+struct powercap_controller - Defines a powercap controller
+
+struct powercap_controller {
+	char name[POWERCAP_CTRL_NAME_LENGTH + 1];
+	struct device device;
+	struct idr idr;
+	void * root_node;
+	struct mutex node_lock;
+	struct list_head ctrl_inst;
+};
+
+Members:
+
+name[POWERCAP_CTRL_NAME_LENGTH + 1]
+	name of controller
+
+device
+	device for this controller
+
+idr
+	idr to have unique id for its child
+
+root_node
+	Root holding power zones for this controller
+
+node_lock
+	mutex for node
+
+ctrl_inst
+	link to the controller list
+
+
+
+Description:
+
+Defines powercap controller instance
+
+
+struct powercap_zone_ops - Define power zone callbacks
+
+struct powercap_zone_ops {
+	int (* get_max_energy_range_uj) (struct powercap_zone_device *, u64 *);
+	int (* get_energy_uj) (struct powercap_zone_device *, u64 *);
+	int (* reset_energy_uj) (struct powercap_zone_device *);
+	int (* get_max_power_range_uw) (struct powercap_zone_device *, u64 *);
+	int (* get_power_uw) (struct powercap_zone_device *, u64 *);
+	int (* reset_power_uw) (struct powercap_zone_device *);
+	void (* cleanup) (struct powercap_zone_device *);
+};
+
+Members:
+
+get_max_energy_range_uj
+	Get maximum range of energy counter in
+				micro-joules.
+
+get_energy_uj
+	Get current energy counter in micro-joules.
+
+reset_energy_uj
+	Reset micro-joules energy counter.
+
+get_max_power_range_uw
+	Get maximum range of power counter in
+				micro-watts.
+
+get_power_uw
+	Get current power counter in micro-watts.
+
+reset_power_uw
+	Reset micro-watts power counter.
+
+cleanup
+	Do any clean up before the zone is freed
+
+
+
+Description:
+
+This structure defines zone callbacks to be implemented by client drivers.
+Client drives can define both energy and power related callbacks. But at
+the least one type (either power or energy) is mandatory.
+
+
+struct powercap_zone_attr - Defines a per zone attribute group
+
+struct powercap_zone_attr {
+	struct attribute * zone_dev_attrs[POWERCAP_ZONE_MAX_ATTRS];
+	int zone_attr_count;
+	struct attribute * const_dev_attrs[POWERCAP_CONSTRAINTS_MAX_ATTRS];
+	int const_attr_count;
+	struct attribute_group dev_zone_attr_group;
+	struct attribute_group dev_const_attr_group;
+	int attr_grp_cnt;
+	const struct attribute_group * dev_attr_groups[POWERCAP_MAX_ATTR_GROUPS + 1];
+};
+
+Members:
+
+zone_dev_attrs[POWERCAP_ZONE_MAX_ATTRS]
+	Device attribute list for power zone.
+
+zone_attr_count
+	Number of power zone attributes.
+
+const_dev_attrs[POWERCAP_CONSTRAINTS_MAX_ATTRS]
+	Constraint attributes.
+
+const_attr_count
+	Number of constraint related attributes
+
+dev_zone_attr_group
+	Attribute group for power zone attributes
+
+dev_const_attr_group
+	Attribute group for constraints
+
+attr_grp_cnt
+	Number of attribute groups
+
+dev_attr_groups[POWERCAP_MAX_ATTR_GROUPS + 1]
+	Used to assign to dev->group
+
+
+
+Description:
+
+Used to add an attribute group unique to a zone based on registry.
+
+
+struct powercap_zone_device - Defines instance of a power cap zone
+
+struct powercap_zone_device {
+	int id;
+	char * zone_dev_name;
+	char name[POWERCAP_ZONE_NAME_LENGTH + 1];
+	void * controller_inst;
+	const struct powercap_zone_ops * ops;
+	struct device device;
+	struct powercap_zone_attr attrs;
+	void * node;
+	int const_id_cnt;
+	struct mutex lock;
+	struct idr idr;
+	struct idr * par_idr;
+	void * drv_data;
+	struct list_head constraint_list;
+};
+
+Members:
+
+id
+	Unique id
+
+zone_dev_name
+	Zone device sysfs node name
+
+name[POWERCAP_ZONE_NAME_LENGTH + 1]
+	Power zone name.
+
+controller_inst
+	Controller instance for this zone
+
+ops
+	Pointer to the zone operation structure.
+
+device
+	Instance of a device.
+
+attrs
+	Attributes associated with this device
+
+node
+	Node pointer to insert to a tree data structure.
+
+const_id_cnt
+	Constraint id count
+
+lock
+	Mutex to protect zone related operations.
+
+idr
+	Instance to an idr entry for children zones.
+
+par_idr
+	To remove reference from the parent idr
+
+drv_data
+	Driver private data
+
+constraint_list
+	Link to the next power zone for this controller.
+
+
+
+Description:
+
+This defines a power zone instance. The fields of this structure are
+private, and should not be used by client drivers.
+
+
+Name:
+
+powercap_set_zone_data - Set private data for a zone
+
+Synopsis:
+
+void powercap_set_zone_data (struct powercap_zone_device * pcd_dev,
+                             void * pdata);
+
+Arguments:
+
+pcd_dev
+	A pointer to the valid zone instance.
+
+pdata
+	A pointer to the user private data.
+
+
+Description:
+
+Allows client drivers to associate some private data to zone instance.
+
+
+Name:
+
+powercap_get_zone_data - Get private data for a zone
+
+Synopsis:
+
+void * powercap_get_zone_data (struct powercap_zone_device * pcd_dev);
+
+Arguments:
+
+pcd_dev
+	A pointer to the valid zone instance.
+
+
+Description:
+
+Allows client drivers to get private data associate with a zone,
+using call to powercap_set_zone_data.
+
+
+Name:
+
+powercap_allocate_controller - Allocates a controller
+
+Synopsis:
+
+struct powercap_controller * powercap_allocate_controller (const char * controller_name);
+
+Arguments:
+
+controller_name
+	The Name of this controller, which will be shown
+			in the sysfs Interface.
+
+
+Description:
+
+Used to create a controller with the power capping class. Here controller
+can represent a type of technology, which can control a range of power zones.
+For example a controller can be RAPL (Running Average Power Limit)
+Intel® 64 and IA-32 Processor Architectures. The name can be any string
+which must be unique, otherwise this function returns NULL.
+On successful allocation, this API returns a pointer to the
+controller instance.
+
+
+Name:
+
+powercap_deallocate_controller - Deallocates a controller
+
+Synopsis:
+
+void powercap_deallocate_controller (struct powercap_controller * instance);
+
+Arguments:
+
+instance
+	A pointer to the valid controller instance.
+
+
+Description:
+
+Used to deallocate a controller with the power capping class. This
+takes only one argument, which is the pointer to the instance returned
+by a call to powercap_allocate_controller.
+When a controller is deallocated, all zones and associated constraints
+are freed.
+
+
+Name:
+
+powercap_zone_register - Register a power zone
+
+Synopsis:
+
+struct powercap_zone_device * powercap_zone_register (struct powercap_controller * controller,
+                                                      const char * name,
+                                                      struct powercap_zone_device * parent,
+                                                      const struct powercap_zone_ops * ops,
+                                                      int no_constraints,
+                                                      struct powercap_zone_constraint_ops * const_ops);
+
+Arguments:
+
+controller
+	A controller instance under which this zone operates.
+
+name
+	A name for this zone.
+
+parent
+	A pointer to the parent power zone instance if any or NULL
+
+ops
+	Pointer to zone operation callback structure.
+
+no_constraints
+	Number of constraints for this zone
+
+const_ops
+	Pointer to constraint callback structure
+
+
+Description:
+
+Used to register a power zone for a controller. Zones are organized in
+a tree like structure in sysfs under a controller.
+A power zone must a register a pointer to a structure representing zone
+callbacks.
+A power zone can have a some other power zone as a parent or it can be
+NULL to appear as a direct descendant of a controller.
+Each power zone can have number of constraints. Constraints appears
+under zones as attributes with unique id.
+
+
+Name:
+
+powercap_zone_unregister - Unregister a zone device
+
+Synopsis:
+
+int powercap_zone_unregister (struct powercap_controller * controller,
+                              struct powercap_zone_device * pcd_dev);
+
+Arguments:
+
+controller
+	A pointer to the valid instance of a controller.
+
+pcd_dev
+	A pointer to the valid zone instance for a controller
+
+
+Description:
+
+Used to unregister a zone device for a controller. Once a zone is
+unregistered then all its children and associated constraints are freed.
-- 
1.8.3.1

[RFC v02 2/5] PowerCap: Add class driver

[Srinivas Pandruvada]

Added power cap class driver, which provides an API for client drivers
to use and provide a consistant sysfs interface to user mode.
For details on API refer to PowerCappingFramework.txt under
Documentation/powercap.

Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
---
 drivers/powercap/Kconfig        |  16 +
 drivers/powercap/Makefile       |   5 +
 drivers/powercap/powercap_sys.c | 995 ++++++++++++++++++++++++++++++++++++++++
 include/linux/powercap.h        | 300 ++++++++++++
 4 files changed, 1316 insertions(+)
 create mode 100644 drivers/powercap/Kconfig
 create mode 100644 drivers/powercap/Makefile
 create mode 100644 drivers/powercap/powercap_sys.c
 create mode 100644 include/linux/powercap.h

diff --git a/drivers/powercap/Kconfig b/drivers/powercap/Kconfig
new file mode 100644
index 0000000..f70b7b9
--- /dev/null
+++ b/drivers/powercap/Kconfig
@@ -0,0 +1,16 @@
+#
+# Generic powercap sysfs drivers configuration
+#
+
+menuconfig POWERCAP_SUPPORT
+	tristate "Generic powercap sysfs driver"
+	help
+	  A Power Capping Sysfs driver offers a generic mechanism for
+	  power capping. Usually it's made up of one or more controllers,
+	  power zones and constraints.
+	  If you want this support, you should say Y or M here.
+
+if POWERCAP_SUPPORT
+# Add client driver config here.
+
+endif
diff --git a/drivers/powercap/Makefile b/drivers/powercap/Makefile
new file mode 100644
index 0000000..f2acfed
--- /dev/null
+++ b/drivers/powercap/Makefile
@@ -0,0 +1,5 @@
+#
+# Makefile for powercap drivers
+#
+
+obj-$(CONFIG_POWERCAP_SUPPORT)	+= powercap_sys.o
diff --git a/drivers/powercap/powercap_sys.c b/drivers/powercap/powercap_sys.c
new file mode 100644
index 0000000..c15c8d4
--- /dev/null
+++ b/drivers/powercap/powercap_sys.c
@@ -0,0 +1,995 @@
+/*
+ * powercap sysfs class driver
+ * Copyright (c) 2013, Intel Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc.
+ *
+ */
+
+#include <linux/module.h>
+#include <linux/device.h>
+#include <linux/err.h>
+#include <linux/slab.h>
+#include <linux/powercap.h>
+
+#define POWERCAP_CONSTRAINT_NAME_LEN		20
+#define POWERCAP_ZONE_CONSTRAINT_DEV_NAME_LEN	30
+
+/**
+ * struct powercap_zone_node- Defines a node containing power zones
+ * @next:		Pointer to sibling
+ * @children_count:	Number of children for this node
+ * @child:		Pointer to first child
+ * @parent:		Pointer to the parent
+ * @pcd_dev:		pointer to power zone device in this node
+ *
+ * A power zone node to be part of a tree.
+ */
+struct powercap_zone_node {
+	struct powercap_zone_node *next;
+	int children_count;
+	struct powercap_zone_node *child;
+	struct powercap_zone_node *parent;
+	struct powercap_zone_device *pcd_dev;
+};
+
+/**
+ * struct powercap_zone_constraint_attrs - Define constraint attribute
+ * @name:		Constraint attribute name.
+ * @attr:		Device attribute
+ *
+ * Define each attribute, with a name, based on the constraint id.
+ */
+struct powercap_constraint_attr {
+	char name[POWERCAP_ZONE_CONSTRAINT_DEV_NAME_LEN + 1];
+	struct device_attribute attr;
+};
+
+/**
+ * struct powercap_zone_constraint- Defines instance of a constraint
+ * @id:			Instance Id of this constraint.
+ * @pcd_dev:		Pointer to the power zone for this constraint.
+ * @ops:		Pointer to the constraint callbacks.
+ * @priv_data:		Constaint private data
+ * @list:		Link to other constraints for this power zone.
+ *
+ * This defines a constraint instance.
+ */
+struct powercap_zone_constraint {
+	int id;
+	struct powercap_zone_device *pcd_dev;
+	struct powercap_zone_constraint_ops *ops;
+	struct powercap_constraint_attr power_limit_attr;
+	struct powercap_constraint_attr time_window_attr;
+	struct powercap_constraint_attr max_power_attr;
+	struct powercap_constraint_attr min_power_attr;
+	struct powercap_constraint_attr max_time_window_attr;
+	struct powercap_constraint_attr min_time_window_attr;
+	struct powercap_constraint_attr name_attr;
+	struct list_head list;
+};
+
+/* A list of powercap controllers */
+static LIST_HEAD(powercap_cntrl_list);
+/* Mutex to protect list of powercap controllers */
+static DEFINE_MUTEX(powercap_cntrl_list_lock);
+
+/*
+ * Power Zone attributes: Each power zone registered with this framework
+ * contains two types of attributes:
+ * One with fixed name: E.g. energy_uj
+ * One with variable name: E.g. constraint_0_power_limit_uw
+ * Using two attribute groups, which will be used during device_register.
+ * The fixed name attributes are using static DEVICE_ATTR.
+ * For variable name device_attribute fields are initialized by assigning
+ * name (constraint_X_power_limit_uw, here X can be 0 to max integer).
+ */
+
+/* Power zone ro attributes define */
+#define powercap_attr_ro(_name)		\
+	static DEVICE_ATTR(_name, 0444, show_##_name, NULL)
+
+/* Power zone rw attributes define */
+#define powercap_attr_rw(_name)		\
+	static DEVICE_ATTR(_name, 0644, show_##_name, store_##_name)
+
+/* constraint attributes define rw */
+#define powercap_const_attr_rw(_name)		\
+	static DEVICE_ATTR(_name, 0644, show_constraint_##_name, \
+				store_constraint_##_name)
+/* constraint attributes define ro */
+#define powercap_const_attr_ro(_name)		\
+	static DEVICE_ATTR(_name, 0644, show_constraint_##_name, NULL)
+
+/* Power zone show function */
+#define define_device_show(_attr)		\
+static ssize_t show_##_attr(struct device *dev, \
+					struct device_attribute *dev_attr,\
+					char *buf) \
+{ \
+	u64 value; \
+	ssize_t len = -EINVAL; \
+	struct powercap_zone_device *pcd_dev = dev_get_drvdata(dev); \
+	\
+	if (pcd_dev && pcd_dev->ops && pcd_dev->ops->get_##_attr) { \
+		mutex_lock(&pcd_dev->lock); \
+		if (!pcd_dev->ops->get_##_attr(pcd_dev, &value)) \
+			len = sprintf(buf, "%lld\n", value); \
+		mutex_unlock(&pcd_dev->lock); \
+	} \
+	\
+	return len; \
+}
+
+/* Power zone store function; only reset is possible */
+#define define_device_store(_attr)		\
+static ssize_t store_##_attr(struct device *dev,\
+				struct device_attribute *dev_attr, \
+				const char *buf, size_t count) \
+{ \
+	int err; \
+	struct powercap_zone_device *pcd_dev = dev_get_drvdata(dev); \
+	u64 value; \
+	\
+	err = kstrtoull(buf, 10, &value); \
+	if (err) \
+		return -EINVAL; \
+	if (value) \
+		return -EINVAL; \
+	if (pcd_dev && pcd_dev->ops && pcd_dev->ops->reset_##_attr) { \
+		mutex_lock(&pcd_dev->lock); \
+		if (!pcd_dev->ops->reset_##_attr(pcd_dev)) { \
+			mutex_unlock(&pcd_dev->lock); \
+			return count; \
+		} \
+		mutex_unlock(&pcd_dev->lock); \
+	} \
+	\
+	return -EINVAL; \
+}
+
+/* Find constraint pointer from an ID */
+static struct powercap_zone_constraint *find_constraint(
+			struct powercap_zone_device *pcd_dev, int id)
+{
+	struct powercap_zone_constraint *pconst = NULL;
+
+	list_for_each_entry(pconst, &pcd_dev->constraint_list, list) {
+		if (pconst->id == id) {
+			return pconst;
+			break;
+		}
+	}
+
+	return pconst;
+}
+
+/* Power zone constraint show function */
+#define define_device_constraint_show(_attr) \
+static ssize_t show_constraint_##_attr(struct device *dev, \
+				struct device_attribute *dev_attr,\
+				char *buf) \
+{ \
+	u64 value; \
+	ssize_t len = -ENODATA; \
+	struct powercap_zone_device *pcd_dev = dev_get_drvdata(dev); \
+	int id; \
+	struct powercap_zone_constraint *pconst;\
+	\
+	if (!pcd_dev) \
+		return -EINVAL; \
+	if (!sscanf(dev_attr->attr.name, "constraint_%d_", &id)) \
+		return -EINVAL; \
+	mutex_lock(&pcd_dev->lock); \
+	pconst = find_constraint(pcd_dev, id); \
+	if (pconst && pconst->ops && pconst->ops->get_##_attr) { \
+		if (!pconst->ops->get_##_attr(pcd_dev, id, &value)) \
+			len = sprintf(buf, "%lld\n", value); \
+	} \
+	mutex_unlock(&pcd_dev->lock); \
+	\
+	return len; \
+}
+
+/* Power zone constraint store function */
+#define define_device_constraint_store(_attr) \
+static ssize_t store_constraint_##_attr(struct device *dev,\
+				struct device_attribute *dev_attr, \
+				const char *buf, size_t count) \
+{ \
+	int err; \
+	u64 value; \
+	struct powercap_zone_device *pcd_dev = dev_get_drvdata(dev); \
+	int id; \
+	struct powercap_zone_constraint *pconst;\
+	\
+	if (!pcd_dev) \
+		return -EINVAL; \
+	if (!sscanf(dev_attr->attr.name, "constraint_%d_", &id)) \
+		return -EINVAL; \
+	err = kstrtoull(buf, 10, &value); \
+	if (err) \
+		return -EINVAL; \
+	mutex_lock(&pcd_dev->lock); \
+	pconst = find_constraint(pcd_dev, id); \
+	if (pconst && pconst->ops && pconst->ops->set_##_attr) { \
+		if (!pconst->ops->set_##_attr(pcd_dev, id, value)) { \
+			mutex_unlock(&pcd_dev->lock); \
+			return count; \
+		} \
+	} \
+	mutex_unlock(&pcd_dev->lock); \
+	\
+	return -ENODATA; \
+}
+
+/* Power zone information callbacks */
+define_device_show(power_uw);
+define_device_store(power_uw);
+define_device_show(max_power_range_uw);
+define_device_show(energy_uj);
+define_device_store(energy_uj);
+define_device_show(max_energy_range_uj);
+
+/* Power zone attributes */
+powercap_attr_ro(max_power_range_uw);
+powercap_attr_rw(power_uw);
+powercap_attr_ro(max_energy_range_uj);
+powercap_attr_rw(energy_uj);
+
+/* Power zone constraint attributes callbacks */
+define_device_constraint_show(power_limit_uw);
+define_device_constraint_store(power_limit_uw);
+define_device_constraint_show(time_window_us);
+define_device_constraint_store(time_window_us);
+define_device_constraint_show(max_power_uw);
+define_device_constraint_show(min_power_uw);
+define_device_constraint_show(max_time_window_us);
+define_device_constraint_show(min_time_window_us);
+
+static ssize_t show_constraint_name(struct device *dev,
+				struct device_attribute *dev_attr,
+				char *buf)
+{
+	const char *name;
+	struct powercap_zone_device *pcd_dev = dev_get_drvdata(dev);
+	int id;
+	ssize_t len = -ENODATA;
+	struct powercap_zone_constraint *pconst;
+
+	if (!pcd_dev)
+		return -EINVAL;
+	if (!sscanf(dev_attr->attr.name, "constraint_%d_", &id))
+		return -EINVAL;
+	mutex_lock(&pcd_dev->lock);
+	pconst = find_constraint(pcd_dev, id);
+	if (pconst && pconst->ops && pconst->ops->get_name) {
+		name = pconst->ops->get_name(pcd_dev, id);
+		if (name) {
+			snprintf(buf, POWERCAP_CONSTRAINT_NAME_LEN,
+								"%s\n", name);
+			buf[POWERCAP_CONSTRAINT_NAME_LEN] = '\0';
+			len = strlen(buf);
+		}
+	}
+	mutex_unlock(&pcd_dev->lock);
+
+	return len;
+}
+
+static void create_constraint_attribute(struct powercap_zone_constraint *pconst,
+				const char *name,
+				int mode,
+				struct powercap_constraint_attr *attr,
+				ssize_t (*show)(struct device *,
+					struct device_attribute *, char *),
+				ssize_t (*store)(struct device *,
+					struct device_attribute *,
+				const char *, size_t)
+				)
+{
+	snprintf(attr->name, POWERCAP_ZONE_CONSTRAINT_DEV_NAME_LEN,
+					"constraint_%d_%s", pconst->id, name);
+	attr->name[POWERCAP_ZONE_CONSTRAINT_DEV_NAME_LEN] = '\0';
+	attr->attr.attr.name = attr->name;
+	attr->attr.attr.mode = mode;
+	attr->attr.show = show;
+	attr->attr.store = store;
+}
+
+/* Create a constraint attribute, if it has required call backs */
+static int create_constraints(struct powercap_zone_constraint *pconst)
+{
+	int count;
+	struct powercap_zone_device *pcd_dev;
+
+	if (!pconst->ops)
+		return -EINVAL;
+	if (!pconst->ops->get_power_limit_uw ||
+		!pconst->ops->set_power_limit_uw ||
+		!pconst->ops->get_time_window_us ||
+		!pconst->ops->set_time_window_us) {
+		return -EINVAL;
+	}
+	pcd_dev = pconst->pcd_dev;
+	count = pcd_dev->attrs.const_attr_count;
+	if (count >=
+		(POWERCAP_CONSTRAINTS_MAX_ATTRS - POWERCAP_CONSTRAINTS_ATTRS))
+		return -EINVAL;
+
+	create_constraint_attribute(pconst, "power_limit_uw",
+					S_IWUSR | S_IRUGO,
+					&pconst->power_limit_attr,
+					show_constraint_power_limit_uw,
+					store_constraint_power_limit_uw);
+	pcd_dev->attrs.const_dev_attrs[count++] =
+				&pconst->power_limit_attr.attr.attr;
+
+	create_constraint_attribute(pconst, "time_window_us",
+					S_IWUSR | S_IRUGO,
+					&pconst->time_window_attr,
+					show_constraint_time_window_us,
+					store_constraint_time_window_us);
+	pcd_dev->attrs.const_dev_attrs[count++] =
+				&pconst->time_window_attr.attr.attr;
+
+	if (pconst->ops->get_name) {
+		create_constraint_attribute(pconst, "name", S_IRUGO,
+				&pconst->name_attr,
+				show_constraint_name,
+				NULL);
+		pcd_dev->attrs.const_dev_attrs[count++] =
+				&pconst->name_attr.attr.attr;
+	}
+	if (pconst->ops->get_max_power_uw) {
+		create_constraint_attribute(pconst, "max_power_uw", S_IRUGO,
+					&pconst->max_power_attr,
+					show_constraint_max_power_uw,
+					NULL);
+		pcd_dev->attrs.const_dev_attrs[count++] =
+					&pconst->max_power_attr.attr.attr;
+	}
+	if (pconst->ops->get_min_power_uw) {
+		create_constraint_attribute(pconst, "min_power_uw", S_IRUGO,
+					&pconst->min_power_attr,
+					show_constraint_min_power_uw,
+					NULL);
+		pcd_dev->attrs.const_dev_attrs[count++] =
+					&pconst->min_power_attr.attr.attr;
+	}
+	if (pconst->ops->get_max_time_window_us) {
+		create_constraint_attribute(pconst, "max_time_window_us",
+					S_IRUGO,
+					&pconst->max_time_window_attr,
+					show_constraint_max_time_window_us,
+					NULL);
+		pcd_dev->attrs.const_dev_attrs[count++] =
+				&pconst->max_time_window_attr.attr.attr;
+	}
+	if (pconst->ops->get_min_time_window_us) {
+		create_constraint_attribute(pconst, "min_time_window_us",
+					S_IRUGO,
+					&pconst->min_time_window_attr,
+					show_constraint_min_time_window_us,
+					NULL);
+		pcd_dev->attrs.const_dev_attrs[count++] =
+				&pconst->min_time_window_attr.attr.attr;
+	}
+
+	pcd_dev->attrs.const_attr_count = count;
+
+	return 0;
+}
+
+struct powercap_zone_constraint *powercap_zone_add_constraint(
+				struct powercap_zone_device *pcd_dev,
+				struct powercap_zone_constraint_ops *ops)
+{
+	struct powercap_zone_constraint *pconst;
+	int result;
+
+	if (!pcd_dev)
+		return ERR_PTR(-EINVAL);
+
+	pconst = kzalloc(sizeof(*pconst), GFP_KERNEL);
+	if (!pconst)
+		return ERR_PTR(-ENOMEM);
+	/*
+	 * No need to hold locks as this is called when device
+	 * is not created or power zone is not in existance
+	 */
+	pconst->pcd_dev = pcd_dev;
+	pconst->ops = ops;
+	pconst->id = pcd_dev->const_id_cnt;
+	result = create_constraints(pconst);
+	if (result) {
+		kfree(pconst);
+		return ERR_PTR(result);
+	}
+	pcd_dev->const_id_cnt++;
+	list_add_tail(&pconst->list, &pcd_dev->constraint_list);
+
+	return pconst;
+}
+
+static void delete_constraints(struct powercap_zone_device *pcd_dev)
+{
+	struct powercap_zone_constraint *p, *n;
+
+	/*
+	 * No need to hold locks as this is called either when device
+	 * is not created or from device_release callback
+	 */
+	list_for_each_entry_safe(p, n, &pcd_dev->constraint_list, list) {
+		list_del(&p->list);
+		kfree(p);
+	}
+}
+
+static int create_constraint_attributes(struct powercap_zone_device *pcd_dev,
+				int nr_constraints,
+				struct powercap_zone_constraint_ops *const_ops)
+{
+	int i;
+	int ret = 0;
+
+	if (pcd_dev->attrs.attr_grp_cnt >= POWERCAP_MAX_ATTR_GROUPS)
+		return -ENOMEM;
+
+	for (i = 0; i < nr_constraints; ++i) {
+		if (!powercap_zone_add_constraint(pcd_dev, const_ops)) {
+			ret = -ENOMEM;
+			break;
+		}
+	}
+	if (ret) {
+		delete_constraints(pcd_dev);
+		return ret;
+	}
+	pcd_dev->attrs.const_dev_attrs[pcd_dev->attrs.const_attr_count] = NULL;
+	pcd_dev->attrs.dev_const_attr_group.attrs =
+						pcd_dev->attrs.const_dev_attrs;
+	pcd_dev->attrs.dev_attr_groups[pcd_dev->attrs.attr_grp_cnt] =
+					&pcd_dev->attrs.dev_const_attr_group;
+
+	pcd_dev->attrs.attr_grp_cnt++;
+
+	return ret;
+}
+
+/* Allocate a node of a tree */
+static struct powercap_zone_node *create_node(
+					struct powercap_zone_device *pcd_dev)
+{
+	struct powercap_zone_node *node;
+
+	node = kzalloc(sizeof(*node), GFP_KERNEL);
+	if (!node)
+		return NULL;
+
+	node->pcd_dev = pcd_dev;
+
+	return node;
+}
+
+/* Insert a node into a tree */
+static void insert_node(struct powercap_controller *ctrl,
+				struct powercap_zone_node *elem,
+				struct powercap_zone_node *parent)
+{
+	struct powercap_zone_node *node;
+
+	mutex_lock(&ctrl->node_lock);
+	if (!ctrl->root_node) {
+		ctrl->root_node = elem;
+		mutex_unlock(&ctrl->node_lock);
+		return;
+	}
+	if (!parent)
+		parent = ctrl->root_node;
+	elem->parent = parent;
+	if (!parent->child)
+		parent->child = elem;
+	else {
+		/* Not a first child */
+		node = parent->child;
+		while (node->next)
+			node = node->next;
+		node->next = elem;
+	}
+	parent->children_count++;
+	mutex_unlock(&ctrl->node_lock);
+}
+
+/*
+ * Delete a node, once node is deleted, its children nodes will be deleted.
+ * This is only deleting node, not freeing any memory for the zone.
+ * For each zone, del_callback is called, which unregister the device
+ * for the power zone. The memory is freed in the device release callback.
+ */
+static void delete_node(struct powercap_controller *ctrl,
+			struct powercap_zone_node *node,
+			void (*del_callback)(struct powercap_zone_device *))
+{
+	struct powercap_zone_node *node_store;
+	struct powercap_zone_node *node_limit = node;
+	bool root_node_delete = false;
+
+
+	mutex_lock(&ctrl->node_lock);
+
+	if (node == ctrl->root_node)
+		root_node_delete = true;
+
+	while (node) {
+		node_store = node;
+		if (node->child) {
+			node = node->child;
+		} else {
+			/* reached leaf node */
+			struct powercap_zone_node *_tnode;
+			if (node_store->pcd_dev) {
+				dev_dbg(&node_store->pcd_dev->device,
+				"Delete child %s of parent %s\n",
+				node_store->pcd_dev->name,
+				node_store->parent ?
+				node_store->parent->pcd_dev->name : "ROOT");
+			}
+			/* Point node to next sibling */
+			node = node_store->next;
+			if (!node)
+				node = node_store->parent; /* back to root */
+			/*
+			 *Before the leaf is deleted, remove references from
+			 *parent and siblings
+			 */
+			_tnode = node_store->parent;
+			if (_tnode) {
+				_tnode->children_count--;
+				if (_tnode->child == node_store) {
+					/*very first child*/
+					_tnode->child = node_store->next;
+				} else {
+					/*Not a first child*/
+					struct powercap_zone_node *_node =
+								_tnode->child;
+					struct powercap_zone_node *_pnode =
+								_node;
+
+					while (_node != node_store) {
+						_pnode = _node;
+						_node = _node->next;
+					}
+					_pnode->next = node_store->next;
+				}
+			}
+			if (node_store->pcd_dev) {
+				/* Ready to delete */
+				(*del_callback)(node_store->pcd_dev);
+			}
+			if (node_store == node_limit) {
+				kfree(node_store);
+				break;
+			}
+			kfree(node_store); /* Leaf node is freed */
+			/* zone memory is freed in the device_release */
+		}
+	}
+	/*
+	 * If the request was for root node,
+	 * then whole tree is deleted
+	 */
+	if (root_node_delete)
+		ctrl->root_node = NULL;
+
+	mutex_unlock(&ctrl->node_lock);
+}
+
+/* Search a tree for a controller for a zone instance */
+static bool search_node(struct powercap_controller *ctrl,
+			struct powercap_zone_device *pcd_dev)
+{
+	bool valid = true;
+	bool found = false;
+	struct powercap_zone_node *node;
+
+	mutex_lock(&ctrl->node_lock);
+
+	node = ctrl->root_node;
+
+	while (node) {
+		if (valid) {
+			if (node->pcd_dev == pcd_dev) {
+				found = true;
+				break;
+			}
+		}
+		/* First check if is node has children, then siblings */
+		if (node->child && valid) {
+			node = node->child;
+			valid = true;
+		} else if (node->next) {
+			node = node->next;
+			valid = true;
+		} else {
+			/* Reached leaf, go back to parent and traverse */
+			node = node->parent;
+			valid = false;
+		}
+	}
+	mutex_unlock(&ctrl->node_lock);
+
+	return found;
+}
+
+/* Check the presence of a controller in the controller list */
+static bool check_controller_validity(void *controller)
+{
+	struct powercap_controller *pos = NULL;
+	bool found = false;
+
+	mutex_lock(&powercap_cntrl_list_lock);
+
+	list_for_each_entry(pos, &powercap_cntrl_list, ctrl_inst) {
+		if (pos == controller) {
+			found = true;
+			break;
+		}
+	}
+	mutex_unlock(&powercap_cntrl_list_lock);
+
+	return found;
+}
+
+static ssize_t show_name(struct device *dev,
+				struct device_attribute *attr,
+				char *buf)
+{
+	struct powercap_zone_device *pcd_dev = dev_get_drvdata(dev);
+
+	return sprintf(buf, "%s\n", pcd_dev->name);
+}
+
+static DEVICE_ATTR(name, 0444, show_name, NULL);
+
+/* Create zone and attributes in sysfs */
+static int create_power_zone_attributes(struct powercap_zone_device *pcd_dev)
+{
+
+	int count = 0;
+
+	pcd_dev->attrs.attr_grp_cnt = 0;
+	/**
+	* Limit is POWERCAP_ZONE_MAX_ATTRS = 10, Only adding 5 attrs.
+	* So not checking range after each addition
+	*/
+	pcd_dev->attrs.zone_dev_attrs[count++] = &dev_attr_name.attr;
+	if (pcd_dev->ops->get_max_energy_range_uj)
+		pcd_dev->attrs.zone_dev_attrs[count++] =
+					&dev_attr_max_energy_range_uj.attr;
+	if (pcd_dev->ops->get_energy_uj)
+		pcd_dev->attrs.zone_dev_attrs[count++] =
+					&dev_attr_energy_uj.attr;
+	if (pcd_dev->ops->get_power_uw)
+		pcd_dev->attrs.zone_dev_attrs[count++] =
+					&dev_attr_power_uw.attr;
+	if (pcd_dev->ops->get_max_power_range_uw)
+		pcd_dev->attrs.zone_dev_attrs[count++] =
+					&dev_attr_max_power_range_uw.attr;
+	pcd_dev->attrs.zone_dev_attrs[count] = NULL;
+	pcd_dev->attrs.zone_attr_count = count;
+	pcd_dev->attrs.dev_zone_attr_group.attrs =
+					pcd_dev->attrs.zone_dev_attrs;
+	pcd_dev->attrs.dev_attr_groups[0] =
+					&pcd_dev->attrs.dev_zone_attr_group;
+	pcd_dev->attrs.attr_grp_cnt++;
+
+	return 0;
+}
+
+static void delete_zone(struct powercap_zone_device *pcd_dev)
+{
+	struct powercap_zone_constraint *p;
+
+	dev_dbg(&pcd_dev->device, "deleting %s\n", pcd_dev->name);
+	mutex_lock(&pcd_dev->lock);
+
+	list_for_each_entry(p, &pcd_dev->constraint_list, list) {
+		if (p->ops->cleanup)
+			p->ops->cleanup(pcd_dev, p->id);
+	}
+	if (pcd_dev->ops->cleanup)
+		pcd_dev->ops->cleanup(pcd_dev);
+
+	mutex_unlock(&pcd_dev->lock);
+
+	device_unregister(&pcd_dev->device);
+}
+
+static void powercap_release(struct device *dev)
+{
+	if (dev->parent) {
+		struct powercap_zone_device *pcd_dev = dev_get_drvdata(dev);
+
+		dev_dbg(dev, "powercap_release zone %s\n", pcd_dev->name);
+
+		delete_constraints(pcd_dev);
+		/* Remove id from parent idr struct */
+		idr_remove(pcd_dev->par_idr, pcd_dev->id);
+		/* Destroy idrs allocated for this zone */
+		idr_destroy(&pcd_dev->idr);
+		kfree(pcd_dev->zone_dev_name);
+		mutex_destroy(&pcd_dev->lock);
+		kfree(pcd_dev);
+	} else {
+		struct powercap_controller *instance = dev_get_drvdata(dev);
+
+		dev_dbg(dev, "powercap_release controller %s\n",
+							instance->name);
+		idr_destroy(&instance->idr);
+		mutex_destroy(&instance->node_lock);
+		kfree(instance);
+	}
+}
+
+static ssize_t type_show(struct device *dev,
+				struct device_attribute *attr,
+				char *buf)
+{
+	if (dev->parent)
+		strcpy(buf, "power-zone\n");
+	else
+		strcpy(buf, "controller\n");
+
+	return strlen(buf);
+}
+
+static struct device_attribute powercap_def_attrs[] = {
+		__ATTR_RO(type),
+		__ATTR_NULL
+};
+
+static struct class powercap_class = {
+	.name = "power_cap",
+	.dev_release = powercap_release,
+	.dev_attrs = powercap_def_attrs,
+};
+
+struct powercap_zone_device *powercap_zone_register(
+				struct powercap_controller *controller,
+				const char *name,
+				struct powercap_zone_device *parent,
+				const struct powercap_zone_ops *ops,
+				int nr_constraints,
+				struct powercap_zone_constraint_ops *const_ops)
+{
+	int result;
+	struct powercap_zone_device *pcd_dev;
+	struct device *dev_ptr;
+	struct idr *idr_ptr;
+	char *parent_name;
+	int name_sz;
+
+	if (!name || !controller)
+		return ERR_PTR(-EINVAL);
+	if (strlen(name) > POWERCAP_ZONE_NAME_LENGTH)
+		return ERR_PTR(-EINVAL);
+	if (!ops)
+		return ERR_PTR(-EINVAL);
+	if (!ops->get_energy_uj && !ops->get_power_uw)
+		return ERR_PTR(-EINVAL);
+	if (!check_controller_validity(controller))
+		return ERR_PTR(-EINVAL);
+	if (parent && !search_node(controller, parent))
+		return ERR_PTR(-EINVAL);
+	pcd_dev = kzalloc(sizeof(*pcd_dev), GFP_KERNEL);
+	if (!pcd_dev)
+		return ERR_PTR(-ENOMEM);
+
+	INIT_LIST_HEAD(&pcd_dev->constraint_list);
+	mutex_init(&pcd_dev->lock);
+	pcd_dev->ops = ops;
+	pcd_dev->controller_inst = controller;
+	strncpy(pcd_dev->name, name, POWERCAP_ZONE_NAME_LENGTH);
+	pcd_dev->name[POWERCAP_ZONE_NAME_LENGTH] = '\0';
+	if (!parent) {
+		dev_ptr = &controller->device;
+		idr_ptr = &controller->idr;
+		parent_name = controller->name;
+	} else {
+		dev_ptr = &parent->device;
+		idr_ptr = &parent->idr;
+		parent_name = parent->zone_dev_name;
+	}
+	pcd_dev->par_idr = idr_ptr;
+	pcd_dev->device.class = &powercap_class;
+
+	/* allocate enough which can accommodate parent + ":" + int value */
+	name_sz = strlen(parent_name) +	sizeof(int)*2 + 2;
+	pcd_dev->zone_dev_name =  kmalloc(name_sz + 1, GFP_KERNEL);
+	if (!pcd_dev->zone_dev_name) {
+		result = -ENOMEM;
+		goto err_name_alloc;
+	}
+	mutex_lock(&controller->node_lock);
+	/* Using idr to get the unique id */
+	result = idr_alloc(pcd_dev->par_idr, NULL, 0, 0, GFP_KERNEL);
+	if (result < 0) {
+		mutex_unlock(&controller->node_lock);
+		goto err_idr_alloc;
+	}
+	pcd_dev->id = result;
+	idr_init(&pcd_dev->idr);
+
+	snprintf(pcd_dev->zone_dev_name, name_sz - 1, "%s:%x",
+						parent_name, pcd_dev->id);
+	pcd_dev->zone_dev_name[name_sz] = '\0';
+	dev_set_name(&pcd_dev->device, pcd_dev->zone_dev_name);
+	pcd_dev->device.parent = dev_ptr;
+
+	create_power_zone_attributes(pcd_dev);
+	result = create_constraint_attributes(pcd_dev, nr_constraints,
+								const_ops);
+	if (result) {
+		idr_remove(pcd_dev->par_idr, pcd_dev->id);
+		mutex_unlock(&controller->node_lock);
+		goto err_dev_reg;
+	}
+	pcd_dev->attrs.dev_attr_groups[pcd_dev->attrs.attr_grp_cnt] = NULL;
+	pcd_dev->device.groups = pcd_dev->attrs.dev_attr_groups;
+
+	result = device_register(&pcd_dev->device);
+	if (result) {
+		delete_constraints(pcd_dev);
+		idr_remove(pcd_dev->par_idr, pcd_dev->id);
+		mutex_unlock(&controller->node_lock);
+		goto err_dev_reg;
+	}
+	mutex_unlock(&controller->node_lock);
+
+	dev_set_drvdata(&pcd_dev->device, pcd_dev);
+	pcd_dev->node = create_node(pcd_dev);
+	if (!pcd_dev->node) {
+		result = -ENOMEM;
+		goto err_dev_reg_done;
+	}
+	if (parent)
+		insert_node(controller, pcd_dev->node, parent->node);
+	else
+		insert_node(controller, pcd_dev->node, NULL);
+
+	return pcd_dev;
+
+err_dev_reg:
+	idr_destroy(&pcd_dev->idr);
+err_idr_alloc:
+	kfree(pcd_dev->zone_dev_name);
+err_name_alloc:
+	mutex_destroy(&pcd_dev->lock);
+	kfree(pcd_dev);
+	return ERR_PTR(result);
+
+err_dev_reg_done:
+	device_unregister(&pcd_dev->device);
+	return ERR_PTR(result);
+}
+EXPORT_SYMBOL_GPL(powercap_zone_register);
+
+int powercap_zone_unregister(struct powercap_controller *controller,
+				struct powercap_zone_device *pcd_dev)
+{
+	if (!pcd_dev || !controller)
+		return -EINVAL;
+
+	if (!search_node(controller, pcd_dev))
+		return -EINVAL;
+
+	delete_node(controller, pcd_dev->node, delete_zone);
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(powercap_zone_unregister);
+
+struct powercap_controller *powercap_allocate_controller(
+					const char *controller_name)
+{
+	struct powercap_controller *cntrl;
+	int result;
+
+	if (!controller_name)
+		return ERR_PTR(-EINVAL);
+
+	if (strlen(controller_name) > POWERCAP_CTRL_NAME_LENGTH)
+		return ERR_PTR(-EINVAL);
+
+	cntrl = kzalloc(sizeof(struct powercap_controller), GFP_KERNEL);
+	if (!cntrl)
+		return ERR_PTR(-ENOMEM);
+
+	mutex_init(&cntrl->node_lock);
+	INIT_LIST_HEAD(&cntrl->ctrl_inst);
+	strncpy(cntrl->name, controller_name, POWERCAP_CTRL_NAME_LENGTH);
+	cntrl->name[POWERCAP_CTRL_NAME_LENGTH] = '\0';
+	cntrl->device.class = &powercap_class;
+	dev_set_name(&cntrl->device, cntrl->name);
+	result = device_register(&cntrl->device);
+	if (result) {
+		kfree(cntrl);
+		return ERR_PTR(result);
+	}
+	dev_set_drvdata(&cntrl->device, cntrl);
+
+	cntrl->root_node = create_node(NULL);
+	if (!cntrl->root_node) {
+		result = -ENOMEM;
+		goto unregister;
+	}
+	idr_init(&cntrl->idr);
+	mutex_lock(&powercap_cntrl_list_lock);
+	list_add_tail(&cntrl->ctrl_inst, &powercap_cntrl_list);
+	mutex_unlock(&powercap_cntrl_list_lock);
+
+	return cntrl;
+
+unregister:
+	device_unregister(&cntrl->device);
+	return ERR_PTR(result);
+}
+EXPORT_SYMBOL_GPL(powercap_allocate_controller);
+
+void powercap_deallocate_controller(struct powercap_controller *instance)
+{
+	struct powercap_controller *pos = NULL;
+
+	mutex_lock(&powercap_cntrl_list_lock);
+
+	list_for_each_entry(pos, &powercap_cntrl_list, ctrl_inst) {
+		if (pos == instance)
+			break;
+	}
+	if (pos != instance) {
+		/* instance not found */
+		mutex_unlock(&powercap_cntrl_list_lock);
+		return;
+	}
+	list_del(&instance->ctrl_inst);
+	delete_node(instance, instance->root_node, delete_zone);
+
+	mutex_unlock(&powercap_cntrl_list_lock);
+
+	device_unregister(&instance->device);
+}
+EXPORT_SYMBOL_GPL(powercap_deallocate_controller);
+
+static int __init powercap_init(void)
+{
+	int result = 0;
+
+	result = class_register(&powercap_class);
+	if (result)
+		return result;
+
+	return result;
+}
+
+static void __exit powercap_exit(void)
+{
+	class_unregister(&powercap_class);
+}
+
+fs_initcall(powercap_init);
+module_exit(powercap_exit);
+
+MODULE_DESCRIPTION("PowerCap sysfs Driver");
+MODULE_AUTHOR("Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>");
+MODULE_LICENSE("GPL v2");
diff --git a/include/linux/powercap.h b/include/linux/powercap.h
new file mode 100644
index 0000000..9967bd0
--- /dev/null
+++ b/include/linux/powercap.h
@@ -0,0 +1,300 @@
+/*
+ * powercap.h : Exports all power class sysfs interface
+ * Copyright (c) 2013, Intel Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc.
+ *
+ */
+
+#ifndef __POWERCAP_H__
+#define __POWERCAP_H__
+
+#include <linux/device.h>
+#include <linux/idr.h>
+
+/*
+ * A power cap class device can contain multiple powercap controllers.
+ * Each controller can have multiple power zones, which can be independently
+ * controlled. Each power zone can have one or more constraints.
+ */
+
+#define POWERCAP_CTRL_NAME_LENGTH			30
+#define POWERCAP_ZONE_NAME_LENGTH			30
+
+struct powercap_zone_device;
+struct powercap_zone_constraint;
+
+/**
+ * struct powercap_zone_constraint_ops - Define constraint callbacks
+ * @set_power_limit_uw:		Set power limit in micro-watts.
+ * @get_power_limit_uw:		Get power limit in micro-watts.
+ * @set_time_window_us:		Set time window in micro-seconds.
+ * @get_time_window_us:		Get time window in micro-seconds.
+ * @get_max_power_uw:		Get max power allowed in micro-watts.
+ * @get_min_power_uw:		Get min power allowed in micro-watts.
+ * @get_max_time_window_us:	Get max time window allowed in micro-seconds.
+ * @get_min_time_window_us:	Get min time window allowed in micro-seconds.
+ * @get_name:			Get the name of constraint
+ * @cleanup:			Do any clean up before the constraint is freed
+ * This structure is used to define the constraint callbacks for the client
+ * drivers. The following callbacks are mandatory and can't be NULL:
+ *  set_power_limit_uw
+ *  get_power_limit_uw
+ *  set_time_window_us
+ *  get_time_window_us
+ *  get_name
+ */
+struct powercap_zone_constraint_ops {
+	int (*set_power_limit_uw)
+			(struct powercap_zone_device *, int, u64);
+	int (*get_power_limit_uw)
+			(struct powercap_zone_device *, int, u64 *);
+
+	int (*set_time_window_us)
+			(struct powercap_zone_device *, int, u64);
+	int (*get_time_window_us)
+			(struct powercap_zone_device *, int, u64 *);
+
+	int (*get_max_power_uw)
+			(struct powercap_zone_device *, int, u64 *);
+	int (*get_min_power_uw)
+			(struct powercap_zone_device *, int, u64 *);
+
+	int (*get_max_time_window_us)
+			(struct powercap_zone_device *, int, u64 *);
+	int (*get_min_time_window_us)
+			(struct powercap_zone_device *, int, u64 *);
+	const char *(*get_name) (struct powercap_zone_device *, int);
+	void (*cleanup) (struct powercap_zone_device *, int);
+};
+
+/**
+ * struct powercap_controller- Defines a powercap controller
+ * @name:		name of controller
+ * @device:		device for this controller
+ * @idr	:		idr to have unique id for its child
+ * @root_node:		Root holding power zones for this controller
+ * @node_lock:		mutex for node
+ * @ctrl_inst:		link to the controller list
+ *
+ * Defines powercap controller instance
+ */
+struct powercap_controller {
+	char name[POWERCAP_CTRL_NAME_LENGTH + 1];
+	struct device device;
+	struct idr idr;
+	void *root_node;
+	struct mutex node_lock;
+	struct list_head ctrl_inst;
+};
+
+/**
+ * struct powercap_zone_ops - Define power zone callbacks
+ * @get_max_energy_range_uj:	Get maximum range of energy counter in
+ *				micro-joules.
+ * @get_energy_uj:		Get current energy counter in micro-joules.
+ * @reset_energy_uj:		Reset micro-joules energy counter.
+ * @get_max_power_range_uw:	Get maximum range of power counter in
+ *				micro-watts.
+ * @get_power_uw:		Get current power counter in micro-watts.
+ * @reset_power_uw:		Reset micro-watts power counter.
+ * @cleanup:			Do any clean up before the zone is freed
+ *
+ * This structure defines zone callbacks to be implemented by client drivers.
+ * Client drives can define both energy and power related callbacks. But at
+ * the least one type (either power or energy) is mandatory.
+ */
+struct powercap_zone_ops {
+	int (*get_max_energy_range_uj)
+			(struct powercap_zone_device *, u64 *);
+	int (*get_energy_uj)
+			(struct powercap_zone_device *, u64 *);
+	int (*reset_energy_uj)
+			(struct powercap_zone_device *);
+
+	int (*get_max_power_range_uw)
+			(struct powercap_zone_device *, u64 *);
+	int (*get_power_uw)
+			(struct powercap_zone_device *, u64 *);
+	int (*reset_power_uw) (struct powercap_zone_device *);
+
+	void (*cleanup) (struct powercap_zone_device *);
+};
+
+#define	POWERCAP_ZONE_MAX_ATTRS		10 /* Currently only max 5 */
+#define	POWERCAP_CONSTRAINTS_ATTRS	8  /*  5 attrs/constraints */
+#define	POWERCAP_CONSTRAINTS_MAX_ATTRS	10 * POWERCAP_CONSTRAINTS_ATTRS
+					/* For 10 constraints per zone */
+#define POWERCAP_MAX_ATTR_GROUPS	2 /* One for zone and constraint */
+/**
+ * struct powercap_zone_attr- Defines a per zone attribute group
+ * @zone_dev_attrs:	Device attribute list for power zone.
+ * @zone_attr_count:	Number of power zone attributes.
+ * @const_dev_attrs:	Constraint attributes.
+ * @const_attr_count:	Number of constraint related attributes
+ * @dev_zone_attr_group: Attribute group for power zone attributes
+ * @dev_const_attr_group: Attribute group for constraints
+ * @attr_grp_cnt:	Number of attribute groups
+ * @dev_attr_groups:	Used to assign to dev->group
+ *
+ * Used to add an attribute group unique to a zone based on registry.
+ */
+struct powercap_zone_attr {
+	struct attribute *zone_dev_attrs[POWERCAP_ZONE_MAX_ATTRS];
+	int zone_attr_count;
+	struct attribute *const_dev_attrs[POWERCAP_CONSTRAINTS_MAX_ATTRS];
+	int const_attr_count;
+	struct attribute_group dev_zone_attr_group;
+	struct attribute_group dev_const_attr_group;
+	int attr_grp_cnt;
+	const struct attribute_group
+			*dev_attr_groups[POWERCAP_MAX_ATTR_GROUPS + 1];
+};
+
+/**
+ * struct powercap_zone_device- Defines instance of a power cap zone
+ * @id:			Unique id
+ * @zone_dev_name:	Zone device sysfs node name
+ * @name:		Power zone name.
+ * @controller_inst:	Controller instance for this zone
+ * @ops:		Pointer to the zone operation structure.
+ * @device:		Instance of a device.
+ * @attrs:		Attributes associated with this device
+ * @node:		Node pointer to insert to a tree data structure.
+ * @const_id_cnt:	Constraint id count
+ * @lock:		Mutex to protect zone related operations.
+ * @idr:		Instance to an idr entry for children zones.
+ * @par_idr:		To remove reference from the parent idr
+ * @drv_data:		Driver private data
+ * @constraint_list:	Link to the next power zone for this controller.
+ *
+ * This defines a power zone instance. The fields of this structure are
+ * private, and should not be used by client drivers.
+ */
+struct powercap_zone_device {
+	int id;
+	char *zone_dev_name;
+	char name[POWERCAP_ZONE_NAME_LENGTH + 1];
+	void *controller_inst;
+	const struct powercap_zone_ops *ops;
+	struct device device;
+	struct powercap_zone_attr attrs;
+	void *node;
+	int const_id_cnt;
+	struct mutex lock;
+	struct idr idr;
+	struct idr *par_idr;
+	void *drv_data;
+	struct list_head constraint_list;
+};
+
+/* For clients to get their device pointer, may be used for dev_dbgs */
+#define POWERCAP_GET_DEV(p_zone)	(&pzone->device)
+
+/**
+* powercap_set_zone_data() - Set private data for a zone
+* @pcd_dev:	A pointer to the valid zone instance.
+* @pdata:	A pointer to the user private data.
+*
+* Allows client drivers to associate some private data to zone instance.
+*/
+static inline void powercap_set_zone_data(struct powercap_zone_device *pcd_dev,
+						void *pdata)
+{
+	if (pcd_dev)
+		pcd_dev->drv_data = pdata;
+}
+
+/**
+* powercap_get_zone_data() - Get private data for a zone
+* @pcd_dev:	A pointer to the valid zone instance.
+*
+* Allows client drivers to get private data associate with a zone,
+* using call to powercap_set_zone_data.
+*/
+static inline void *powercap_get_zone_data(struct powercap_zone_device *pcd_dev)
+{
+	if (pcd_dev)
+		return pcd_dev->drv_data;
+	return NULL;
+}
+
+/* Controller allocate/deallocate API */
+
+/**
+* powercap_allocate_controller() - Allocates a controller
+* @controller_name:	The Name of this controller, which will be shown
+*			in the sysfs Interface.
+*
+* Used to create a controller with the power capping class. Here controller
+* can represent a type of technology, which can control a range of power zones.
+* For example a controller can be RAPL (Running Average Power Limit)
+* Intel® 64 and IA-32 Processor Architectures. The name can be any string
+* which must be unique, otherwise this function returns NULL.
+* On successful allocation, this API returns a pointer to the
+* controller instance.
+*/
+struct powercap_controller *powercap_allocate_controller(
+						const char *controller_name);
+
+/**
+* powercap_deallocate_controller() - Deallocates a controller
+* @instance:	A pointer to the valid controller instance.
+*
+* Used to deallocate a controller with the power capping class. This
+* takes only one argument, which is the pointer to the instance returned
+* by a call to powercap_allocate_controller.
+* When a controller is deallocated, all zones and associated constraints
+* are freed.
+*/
+void powercap_deallocate_controller(struct powercap_controller *instance);
+
+/* Zone register/unregister API */
+
+/**
+* powercap_zone_register() - Register a power zone
+* @controller:	A controller instance under which this zone operates.
+* @name:	A name for this zone.
+* @parent:	A pointer to the parent power zone instance if any or NULL
+* @ops:		Pointer to zone operation callback structure.
+* @no_constraints: Number of constraints for this zone
+* @const_ops:	Pointer to constraint callback structure
+*
+* Used to register a power zone for a controller. Zones are organized in
+* a tree like structure in sysfs under a controller.
+* A power zone must a register a pointer to a structure representing zone
+* callbacks.
+* A power zone can have a some other power zone as a parent or it can be
+* NULL to appear as a direct descendant of a controller.
+* Each power zone can have number of constraints. Constraints appears
+* under zones as attributes with unique id.
+*/
+struct powercap_zone_device *powercap_zone_register(
+			struct powercap_controller *controller,
+			const char *name,
+			struct powercap_zone_device *parent,
+			const struct powercap_zone_ops *ops,
+			int no_constraints,
+			struct powercap_zone_constraint_ops *const_ops);
+/**
+* powercap_zone_unregister() - Unregister a zone device
+* @controller:	A pointer to the valid instance of a controller.
+* @pcd_dev:	A pointer to the valid zone instance for a controller
+*
+* Used to unregister a zone device for a controller. Once a zone is
+* unregistered then all its children and associated constraints are freed.
+*/
+int powercap_zone_unregister(struct powercap_controller *controller,
+				struct powercap_zone_device *pcd_dev);
+
+#endif
-- 
1.8.3.1

[RFC v02 3/5] PowerCap: Added to drivers build

[Srinivas Pandruvada]

Added changes to Makefile and Kconfig to include in driver build.

Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
---
 drivers/Kconfig  | 2 ++
 drivers/Makefile | 1 +
 2 files changed, 3 insertions(+)

diff --git a/drivers/Kconfig b/drivers/Kconfig
index 9953a42..89ff2f8 100644
--- a/drivers/Kconfig
+++ b/drivers/Kconfig
@@ -166,4 +166,6 @@ source "drivers/ipack/Kconfig"
 
 source "drivers/reset/Kconfig"
 
+source "drivers/powercap/Kconfig"
+
 endmenu
diff --git a/drivers/Makefile b/drivers/Makefile
index 130abc1..e207b2d 100644
--- a/drivers/Makefile
+++ b/drivers/Makefile
@@ -152,3 +152,4 @@ obj-$(CONFIG_IIO)		+= iio/
 obj-$(CONFIG_VME_BUS)		+= vme/
 obj-$(CONFIG_IPACK_BUS)		+= ipack/
 obj-$(CONFIG_NTB)		+= ntb/
+obj-$(CONFIG_POWERCAP_SUPPORT)	+= powercap/
-- 
1.8.3.1

[RFC v02 4/5] x86/msr: add 64bit _on_cpu access functions

[Srinivas Pandruvada]

From: Jacob Pan <jacob.jun.pan@linux.intel.com>

Having 64-bit MSR access methods on given CPU can avoid shifting and
simplify MSR content manipulation. We already have other combinations
of rdmsrl_xxx and wrmsrl_xxx but missing the _on_cpu version.

Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
---
 arch/x86/include/asm/msr.h | 22 ++++++++++++++++
 arch/x86/lib/msr-smp.c     | 62 ++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 84 insertions(+)

diff --git a/arch/x86/include/asm/msr.h b/arch/x86/include/asm/msr.h
index cb75028..e139b13 100644
--- a/arch/x86/include/asm/msr.h
+++ b/arch/x86/include/asm/msr.h
@@ -218,10 +218,14 @@ void msrs_free(struct msr *msrs);
 #ifdef CONFIG_SMP
 int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h);
 int wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h);
+int rdmsrl_on_cpu(unsigned int cpu, u32 msr_no, u64 *q);
+int wrmsrl_on_cpu(unsigned int cpu, u32 msr_no, u64 q);
 void rdmsr_on_cpus(const struct cpumask *mask, u32 msr_no, struct msr *msrs);
 void wrmsr_on_cpus(const struct cpumask *mask, u32 msr_no, struct msr *msrs);
 int rdmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h);
 int wrmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h);
+int rdmsrl_safe_on_cpu(unsigned int cpu, u32 msr_no, u64 *q);
+int wrmsrl_safe_on_cpu(unsigned int cpu, u32 msr_no, u64 q);
 int rdmsr_safe_regs_on_cpu(unsigned int cpu, u32 regs[8]);
 int wrmsr_safe_regs_on_cpu(unsigned int cpu, u32 regs[8]);
 #else  /*  CONFIG_SMP  */
@@ -235,6 +239,16 @@ static inline int wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
 	wrmsr(msr_no, l, h);
 	return 0;
 }
+static inline int rdmsrl_on_cpu(unsigned int cpu, u32 msr_no, u64 *q)
+{
+	rdmsrl(msr_no, *q);
+	return 0;
+}
+static inline int wrmsrl_on_cpu(unsigned int cpu, u32 msr_no, u64 q)
+{
+	wrmsrl(msr_no, q);
+	return 0;
+}
 static inline void rdmsr_on_cpus(const struct cpumask *m, u32 msr_no,
 				struct msr *msrs)
 {
@@ -254,6 +268,14 @@ static inline int wrmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
 {
 	return wrmsr_safe(msr_no, l, h);
 }
+static inline int rdmsrl_safe_on_cpu(unsigned int cpu, u32 msr_no, u64 *q)
+{
+	return rdmsrl_safe(msr_no, q);
+}
+static inline int wrmsrl_safe_on_cpu(unsigned int cpu, u32 msr_no, u64 q)
+{
+	return wrmsrl_safe(msr_no, q);
+}
 static inline int rdmsr_safe_regs_on_cpu(unsigned int cpu, u32 regs[8])
 {
 	return rdmsr_safe_regs(regs);
diff --git a/arch/x86/lib/msr-smp.c b/arch/x86/lib/msr-smp.c
index a6b1b86..518532e 100644
--- a/arch/x86/lib/msr-smp.c
+++ b/arch/x86/lib/msr-smp.c
@@ -47,6 +47,21 @@ int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h)
 }
 EXPORT_SYMBOL(rdmsr_on_cpu);
 
+int rdmsrl_on_cpu(unsigned int cpu, u32 msr_no, u64 *q)
+{
+	int err;
+	struct msr_info rv;
+
+	memset(&rv, 0, sizeof(rv));
+
+	rv.msr_no = msr_no;
+	err = smp_call_function_single(cpu, __rdmsr_on_cpu, &rv, 1);
+	*q = rv.reg.q;
+
+	return err;
+}
+EXPORT_SYMBOL(rdmsrl_on_cpu);
+
 int wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
 {
 	int err;
@@ -63,6 +78,22 @@ int wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
 }
 EXPORT_SYMBOL(wrmsr_on_cpu);
 
+int wrmsrl_on_cpu(unsigned int cpu, u32 msr_no, u64 q)
+{
+	int err;
+	struct msr_info rv;
+
+	memset(&rv, 0, sizeof(rv));
+
+	rv.msr_no = msr_no;
+	rv.reg.q = q;
+
+	err = smp_call_function_single(cpu, __wrmsr_on_cpu, &rv, 1);
+
+	return err;
+}
+EXPORT_SYMBOL(wrmsrl_on_cpu);
+
 static void __rwmsr_on_cpus(const struct cpumask *mask, u32 msr_no,
 			    struct msr *msrs,
 			    void (*msr_func) (void *info))
@@ -159,6 +190,37 @@ int wrmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
 }
 EXPORT_SYMBOL(wrmsr_safe_on_cpu);
 
+int wrmsrl_safe_on_cpu(unsigned int cpu, u32 msr_no, u64 q)
+{
+	int err;
+	struct msr_info rv;
+
+	memset(&rv, 0, sizeof(rv));
+
+	rv.msr_no = msr_no;
+	rv.reg.q = q;
+
+	err = smp_call_function_single(cpu, __wrmsr_safe_on_cpu, &rv, 1);
+
+	return err ? err : rv.err;
+}
+EXPORT_SYMBOL(wrmsrl_safe_on_cpu);
+
+int rdmsrl_safe_on_cpu(unsigned int cpu, u32 msr_no, u64 *q)
+{
+	int err;
+	struct msr_info rv;
+
+	memset(&rv, 0, sizeof(rv));
+
+	rv.msr_no = msr_no;
+	err = smp_call_function_single(cpu, __rdmsr_safe_on_cpu, &rv, 1);
+	*q = rv.reg.q;
+
+	return err ? err : rv.err;
+}
+EXPORT_SYMBOL(rdmsrl_safe_on_cpu);
+
 /*
  * These variants are significantly slower, but allows control over
  * the entire 32-bit GPR set.
-- 
1.8.3.1

[RFC v02 5/5] Introduce Intel RAPL power capping driver

[Srinivas Pandruvada]

From: Jacob Pan <jacob.jun.pan@linux.intel.com>

RAPL(Running Average Power Limit) interface provides platform software
with the ability to monitor, control, and get notifications on SOC
power consumptions. Since its first appearance on Sandy Bridge, more
features have being added to extend its usage. In RAPL, platforms are
divided into domains for fine grained control. These domains include
package, DRAM controller, CPU core (Power Plane 0), graphics uncore
(power plane 1), etc.

The purpose of this driver is to expose RAPL for userspace
consumption. Overall, RAPL fits in the new powercap class driver in
that platform level power capping controls are exposed via this
generic interface.

Zhang, Rui's initial RAPL driver was used as a reference and starting
point. Many thanks.
https://lkml.org/lkml/2011/5/26/93

Unlike the patch above, which is mainly for monitoring, this driver
focus on the control and usability by user applications.

Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
---
 drivers/powercap/Kconfig      |    8 +
 drivers/powercap/Makefile     |    1 +
 drivers/powercap/intel_rapl.c | 1319 +++++++++++++++++++++++++++++++++++++++++
 3 files changed, 1328 insertions(+)
 create mode 100644 drivers/powercap/intel_rapl.c

diff --git a/drivers/powercap/Kconfig b/drivers/powercap/Kconfig
index f70b7b9..7fd9d48 100644
--- a/drivers/powercap/Kconfig
+++ b/drivers/powercap/Kconfig
@@ -13,4 +13,12 @@ menuconfig POWERCAP_SUPPORT
 if POWERCAP_SUPPORT
 # Add client driver config here.
 
+config INTEL_RAPL
+	tristate "Intel RAPL Support"
+	default n
+	---help---
+	  RAPL (Running Average Power Limit) provides mechanisms to enforce
+	  and monitor power limits of supported Intel CPUs. Limits can be
+	  set by RAPL domains such as package, core, graphics, etc.
+
 endif
diff --git a/drivers/powercap/Makefile b/drivers/powercap/Makefile
index f2acfed..a08b8b0 100644
--- a/drivers/powercap/Makefile
+++ b/drivers/powercap/Makefile
@@ -3,3 +3,4 @@
 #
 
 obj-$(CONFIG_POWERCAP_SUPPORT)	+= powercap_sys.o
+obj-$(CONFIG_INTEL_RAPL)       += intel_rapl.o
diff --git a/drivers/powercap/intel_rapl.c b/drivers/powercap/intel_rapl.c
new file mode 100644
index 0000000..46642b8
--- /dev/null
+++ b/drivers/powercap/intel_rapl.c
@@ -0,0 +1,1319 @@
+/*
+ * Intel Running Average Power Limit (RAPL) Driver
+ * Copyright (c) 2013, Intel Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc.
+ *
+ */
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/list.h>
+#include <linux/init.h>
+#include <linux/types.h>
+#include <linux/device.h>
+#include <linux/thermal.h>
+#include <linux/slab.h>
+#include <linux/log2.h>
+#include <linux/bitmap.h>
+#include <linux/delay.h>
+#include <linux/sysfs.h>
+#include <linux/cpu.h>
+#include <linux/powercap.h>
+
+#include <asm/processor.h>
+#include <asm/cpu_device_id.h>
+
+
+/* bitmasks for RAPL MSRs, used by primitive access functions */
+#define ENERGY_STATUS_MASK      0xffffffff
+
+#define POWER_LIMIT1_MASK       0x7FFF
+#define POWER_LIMIT1_ENABLE     BIT(15)
+#define POWER_LIMIT1_CLAMP      BIT(16)
+
+#define POWER_LIMIT2_MASK       (0x7FFFULL<<32)
+#define POWER_LIMIT2_ENABLE     (0x1ULL<<47)
+#define POWER_LIMIT2_CLAMP      (0x1ULL<<48)
+#define POWER_PKG_LOCK          (0x1ULL<<63)
+#define POWER_PP_LOCK           BIT(31)
+
+#define TIME_WINDOW1_MASK       (0x7F<<17)
+#define TIME_WINDOW2_MASK       (0x7FULL<<49)
+
+#define POWER_UNIT_OFFSET	0
+#define POWER_UNIT_MASK		0x0F
+
+#define ENERGY_UNIT_OFFSET	0x08
+#define ENERGY_UNIT_MASK	0x1F00
+
+#define TIME_UNIT_OFFSET	0x10
+#define TIME_UNIT_MASK		0xF0000
+
+#define POWER_INFO_MAX_MASK     (0x7fffULL<<32)
+#define POWER_INFO_MIN_MASK     (0x7fffULL<<16)
+#define POWER_INFO_MAX_TIME_WIN_MASK     (0x3fULL<<48)
+#define POWER_INFO_THERMAL_SPEC_MASK     0x7fff
+
+#define PERF_STATUS_THROTTLE_TIME_MASK 0xffffffff
+#define PP_POLICY_MASK         0x1F
+
+/* Non HW constants */
+#define RAPL_PRIMITIVE_DERIVED       BIT(1) /* not from raw data */
+#define RAPL_PRIMITIVE_DUMMY         BIT(2)
+
+/* scale RAPL units to avoid floating point math inside kernel */
+#define POWER_UNIT_SCALE     (1000000)
+#define ENERGY_UNIT_SCALE    (1000000)
+#define TIME_UNIT_SCALE      (1000000)
+
+#define TIME_WINDOW_MAX_MSEC 40000
+#define TIME_WINDOW_MIN_MSEC 250
+
+enum unit_type {
+	NA_UNIT, /* no translation */
+	POWER_UNIT,
+	ENERGY_UNIT,
+	TIME_UNIT,
+};
+
+enum rapl_domain_type {
+	RAPL_DOMAIN_PKG, /* entire package/socket */
+	RAPL_DOMAIN_PP0, /* core power plane */
+	RAPL_DOMAIN_PP1, /* graphics uncore */
+	RAPL_DOMAIN_DRAM,/* DRAM controller */
+	RAPL_DOMAIN_MAX,
+};
+
+enum rapl_domain_msr_id {
+	RAPL_DOMAIN_MSR_LIMIT,
+	RAPL_DOMAIN_MSR_STATUS,
+	RAPL_DOMAIN_MSR_PERF,
+	RAPL_DOMAIN_MSR_POLICY,
+	RAPL_DOMAIN_MSR_INFO,
+	RAPL_DOMAIN_MSR_MAX,
+};
+
+/* per domain data, some are optional */
+enum rapl_primitives {
+	energy,
+	power_limit1,
+	power_limit2,
+	lock,
+
+	pl1_enable,
+	pl1_clamp,
+	pl2_enable,
+	pl2_clamp,
+
+	time_window1,
+	time_window2,
+	thermal_spec_power,
+	max_power,
+
+	min_power,
+	max_window,
+	throttle_time,
+	prio_level,
+
+	/* below are not raw primitive data */
+	average_power,
+	nr_rapl_primitives,
+};
+
+#define NR_RAW_PRIMITIVES (nr_rapl_primitives - 2)
+
+/* Can be expanded to include events, etc.*/
+struct rapl_domain_data {
+	u64 primitives[nr_rapl_primitives];
+	unsigned long timestamp;
+};
+
+
+#define	DOMAIN_STATE_INACTIVE           BIT(0)
+#define	DOMAIN_STATE_POWER_LIMIT_SET    BIT(1)
+#define DOMAIN_STATE_BIOS_LOCKED        BIT(2)
+
+#define NR_POWER_LIMITS (2)
+struct rapl_power_limit {
+	struct powercap_zone_constraint *constraint;
+	int prim_id; /* primitive ID used to enable */
+	struct rapl_domain *domain;
+	const char *name;
+};
+
+static const char pl1_name[] = "long_term";
+static const char pl2_name[] = "short_term";
+
+struct rapl_domain {
+	const char *name;
+	enum rapl_domain_type id;
+	int msrs[RAPL_DOMAIN_MSR_MAX];
+	struct powercap_zone_device *pzone;
+	struct powercap_zone_ops *ops;
+	struct rapl_domain_data rdd;
+	struct rapl_power_limit rpl[NR_POWER_LIMITS];
+	u64 attr_map; /* track capabilities */
+	unsigned int state;
+	int pkg_id;
+};
+
+/* Each physical package contains multiple domains, these are the common
+ * data across RAPL domains within a package.
+ */
+struct rapl_pkg {
+	unsigned int id; /* physical package/socket id */
+	unsigned int nr_domains;
+	unsigned long domain_map; /* bit map of active domains */
+	unsigned int power_unit_divisor;
+	unsigned int energy_unit_divisor;
+	unsigned int time_unit_divisor;
+	struct rapl_domain *domains;
+	struct powercap_zone_device *pzone; /* keep track of parent zone */
+	int nr_cpus; /* active cpus on the pkg, topology info is lost during
+		      * cpu hotplug. so we have to track ourselves.
+		      */
+	unsigned long power_limit_irq; /* keep track of pkg power limit
+					* notify interrupt enable status.
+					*/
+	struct list_head plist;
+};
+#define PACKAGE_PLN_INT_SAVED   BIT(0)
+#define MAX_PRIM_NAME (32)
+
+/* per domain data. used to describe individual knobs such that access function
+ * can be consolidated into one instead of many inline functions.
+ */
+struct rapl_primitive_info {
+	const char *name;
+	u64 mask;
+	int shift;
+	enum rapl_domain_msr_id id;
+	enum unit_type unit;
+	u32 flag;
+};
+
+#define PRIMITIVE_INFO_INIT(p, m, s, i, u, f) {	\
+		.name = #p,			\
+		.mask = m,			\
+		.shift = s,			\
+		.id = i,			\
+		.unit = u,			\
+		.flag = f			\
+	}
+
+static void rapl_init_domains(struct rapl_pkg *rp);
+static int rapl_read_data_raw(struct rapl_domain *rd,
+			enum rapl_primitives prim,
+			bool xlate, u64 *data);
+static int rapl_write_data_raw(struct rapl_domain *rd,
+			enum rapl_primitives prim,
+			unsigned long long value);
+static u64 rapl_unit_xlate(int pkg, enum unit_type type, u64 value, int to_raw);
+static void pkg_power_limit_irq_save(int pkg_id);
+
+static LIST_HEAD(rpkgs); /* guarded by CPU hotplug lock */
+
+static const char * const rapl_domain_names[] = {
+	"package",
+	"core",
+	"uncore",
+	"dram",
+};
+
+static struct powercap_controller *pccr; /* PowerCap Controller */
+
+/* caller to ensure CPU hotplug lock is held */
+static struct rapl_pkg *find_package_by_id(int id)
+{
+	struct rapl_pkg *rp;
+
+	list_for_each_entry(rp, &rpkgs, plist) {
+		if (rp->id == id)
+			return rp;
+	}
+
+	return NULL;
+}
+
+/* caller to ensure CPU hotplug lock is held */
+static int find_active_cpu_on_pkg(int pkg_id)
+{
+	int i;
+
+	for_each_online_cpu(i) {
+		if (topology_physical_package_id(i) == pkg_id)
+			return i;
+	}
+	/* all CPUs on this package are offline */
+
+	return -ENODEV;
+}
+
+/* caller must hold cpu hotplug lock */
+static void rapl_cleanup_data(void)
+{
+	struct rapl_pkg *p, *tmp;
+
+	if (list_empty(&rpkgs))
+		return;
+	list_for_each_entry_safe(p, tmp, &rpkgs, plist) {
+		kfree(p->domains);
+		list_del(&p->plist);
+		kfree(p);
+	}
+}
+
+
+int get_energy_counter(struct powercap_zone_device *pzdev, u64 *energy_raw)
+{
+	struct rapl_domain *rd;
+	u64 energy_now;
+
+	get_online_cpus();
+	rd = (struct rapl_domain *)powercap_get_zone_data(pzdev);
+	if (!rd)
+		return -ENODEV;
+
+	if (!rapl_read_data_raw(rd, energy, true, &energy_now)) {
+		*energy_raw = energy_now;
+		put_online_cpus();
+
+		return 0;
+	}
+	put_online_cpus();
+
+	return -EIO;
+}
+
+int get_max_power_counter(struct powercap_zone_device *pzdev, u64 *pwr)
+{
+	struct rapl_domain *rd;
+	u64 val64;
+
+	get_online_cpus();
+	rd = (struct rapl_domain *)powercap_get_zone_data(pzdev);
+	if (!rd)
+		return -ENODEV;
+
+	if (!rapl_read_data_raw(rd, thermal_spec_power, true, &val64)) {
+		*pwr = val64;
+		put_online_cpus();
+		return 0;
+	}
+	put_online_cpus();
+	return -EIO;
+}
+
+int get_max_energy_counter(struct powercap_zone_device *pcd_dev, u64 *energy)
+{
+	*energy = rapl_unit_xlate(0, ENERGY_UNIT, ENERGY_STATUS_MASK, 0);
+	return 0;
+}
+
+
+static struct powercap_zone_ops zone_ops_pkg = {
+	.get_energy_uj = get_energy_counter,
+	.get_max_power_range_uw = get_max_power_counter,
+	.get_max_energy_range_uj = get_max_energy_counter,
+};
+
+static struct powercap_zone_ops zone_ops_pp0 = {
+	.get_energy_uj = get_energy_counter,
+	.get_max_energy_range_uj = get_max_energy_counter,
+};
+
+static struct powercap_zone_ops zone_ops_pp1 = {
+	.get_energy_uj = get_energy_counter,
+	.get_max_energy_range_uj = get_max_energy_counter,
+
+};
+static struct powercap_zone_ops zone_ops_dram = {
+	.get_energy_uj = get_energy_counter,
+	.get_max_energy_range_uj = get_max_energy_counter,
+};
+
+static int set_power_limit(struct powercap_zone_device *pzdev, int id,
+			u64 power_limit)
+{
+	struct rapl_domain *rd;
+	struct rapl_pkg *rp;
+	int ret = 0;
+
+	get_online_cpus();
+
+	rd = (struct rapl_domain *)powercap_get_zone_data(pzdev);
+	if (!rd)
+		return -ENODEV;
+	rp = find_package_by_id(rd->pkg_id);
+	if (!rp) {
+		ret = -ENODEV;
+		goto set_exit;
+	}
+
+	if (rd->state & DOMAIN_STATE_BIOS_LOCKED) {
+		dev_warn(&pzdev->device, "%s locked by BIOS, monitoring only\n",
+			rd->name);
+		ret = -EACCES;
+		goto set_exit;
+	}
+
+	switch (rd->rpl[id].prim_id) {
+	case pl1_enable:
+		rapl_write_data_raw(rd, pl1_enable, 1);
+		rapl_write_data_raw(rd, pl1_clamp, 1);
+		rapl_write_data_raw(rd, power_limit1, power_limit);
+		break;
+	case pl2_enable:
+		rapl_write_data_raw(rd, pl2_enable, 1);
+		rapl_write_data_raw(rd, pl2_clamp, 1);
+		rapl_write_data_raw(rd, power_limit2, power_limit);
+		break;
+	default:
+		ret = -EINVAL;
+	}
+	if (!ret)
+		pkg_power_limit_irq_save(rd->pkg_id);
+set_exit:
+	put_online_cpus();
+	return ret;
+}
+
+static int get_current_power_limit(struct powercap_zone_device *pzdev, int id,
+					u64 *data)
+{
+	struct rapl_domain *rd;
+	u64 val;
+	int prim;
+	int ret = 0;
+
+	get_online_cpus();
+	rd = (struct rapl_domain *)powercap_get_zone_data(pzdev);
+	if (!rd)
+		return -ENODEV;
+
+	switch (rd->rpl[id].prim_id) {
+	case pl1_enable:
+		prim = power_limit1;
+		break;
+	case pl2_enable:
+		prim = power_limit2;
+		break;
+	default:
+		put_online_cpus();
+		return -EINVAL;
+	}
+	if (rapl_read_data_raw(rd, prim, true, &val))
+		ret = -EIO;
+	else
+		*data = val;
+
+	put_online_cpus();
+
+	return ret;
+}
+
+static int set_time_window(struct powercap_zone_device *pzdev, int id,
+								u64 window)
+{
+	struct rapl_domain *rd;
+	int ret = 0;
+
+	get_online_cpus();
+	rd = (struct rapl_domain *)powercap_get_zone_data(pzdev);
+	if (!rd)
+		return -ENODEV;
+
+	switch (rd->rpl[id].prim_id) {
+	case pl1_enable:
+		rapl_write_data_raw(rd, time_window1, window);
+		break;
+	case pl2_enable:
+		rapl_write_data_raw(rd, time_window2, window);
+		break;
+	default:
+		ret = -EINVAL;
+	}
+	put_online_cpus();
+	return ret;
+}
+
+int get_time_window(struct powercap_zone_device *pzdev, int id, u64 *data)
+{
+	struct rapl_domain *rd;
+	u64 val;
+	int ret = 0;
+
+	get_online_cpus();
+	rd = (struct rapl_domain *)powercap_get_zone_data(pzdev);
+	if (!rd)
+		return -ENODEV;
+
+	switch (rd->rpl[id].prim_id) {
+	case pl1_enable:
+		ret = rapl_read_data_raw(rd, time_window1, true, &val);
+		break;
+	case pl2_enable:
+		ret = rapl_read_data_raw(rd, time_window2, true, &val);
+		break;
+	default:
+		put_online_cpus();
+		return -EINVAL;
+	}
+	if (!ret)
+		*data = val;
+	put_online_cpus();
+
+	return ret;
+}
+
+const char *get_constraint_name(struct powercap_zone_device *pzdev, int id)
+{
+	struct rapl_power_limit *rpl;
+	struct rapl_domain *rd;
+
+	rd = (struct rapl_domain *)powercap_get_zone_data(pzdev);
+	if (!rd)
+		return NULL;
+
+	rpl = (struct rapl_power_limit *) &rd->rpl[id];
+
+	return rpl->name;
+}
+
+struct powercap_zone_constraint_ops const_ops = {
+	.set_power_limit_uw = set_power_limit,
+	.get_power_limit_uw = get_current_power_limit,
+	.set_time_window_us = set_time_window,
+	.get_time_window_us = get_time_window,
+	.get_name = get_constraint_name,
+};
+
+static int find_nr_power_limit(struct rapl_domain *rd)
+{
+	int i;
+
+	for (i = 0; i < NR_POWER_LIMITS; i++) {
+		if (rd->rpl[i].name == NULL)
+			break;
+	}
+
+	return i;
+}
+
+/* called after domain detection and package level data are set */
+static void rapl_init_domains(struct rapl_pkg *rp)
+{
+	int i;
+	struct rapl_domain *rd = rp->domains;
+
+	for (i = 0; i < RAPL_DOMAIN_MAX; i++) {
+		unsigned int mask = rp->domain_map & (1 << i);
+		switch (mask) {
+		case 1 << RAPL_DOMAIN_PKG:
+			rd->name = rapl_domain_names[RAPL_DOMAIN_PKG];
+			rd->id = RAPL_DOMAIN_PKG;
+			rd->msrs[0] = MSR_PKG_POWER_LIMIT;
+			rd->msrs[1] = MSR_PKG_ENERGY_STATUS;
+			rd->msrs[2] = MSR_PKG_PERF_STATUS;
+			rd->msrs[3] = 0;
+			rd->msrs[4] = MSR_PKG_POWER_INFO;
+			rd->ops = &zone_ops_pkg;
+			rd->rpl[0].prim_id = pl1_enable;
+			rd->rpl[0].name = pl1_name;
+			rd->rpl[1].prim_id = pl2_enable;
+			rd->rpl[1].name = pl2_name;
+			break;
+		case 1 << RAPL_DOMAIN_PP0:
+			rd->name = rapl_domain_names[RAPL_DOMAIN_PP0];
+			rd->id = RAPL_DOMAIN_PP0;
+			rd->msrs[0] = MSR_PP0_POWER_LIMIT;
+			rd->msrs[1] = MSR_PP0_ENERGY_STATUS;
+			rd->msrs[2] = 0;
+			rd->msrs[3] = MSR_PP0_POLICY;
+			rd->msrs[4] = 0;
+			rd->ops = &zone_ops_pp0;
+			rd->rpl[0].prim_id = pl1_enable;
+			rd->rpl[0].name = pl1_name;
+			break;
+		case 1 << RAPL_DOMAIN_PP1:
+			rd->name = rapl_domain_names[RAPL_DOMAIN_PP1];
+			rd->id = RAPL_DOMAIN_PP1;
+			rd->msrs[0] = MSR_PP1_POWER_LIMIT;
+			rd->msrs[1] = MSR_PP1_ENERGY_STATUS;
+			rd->msrs[2] = 0;
+			rd->msrs[3] = MSR_PP1_POLICY;
+			rd->msrs[4] = 0;
+			rd->ops = &zone_ops_pp1;
+			rd->rpl[0].prim_id = pl1_enable;
+			rd->rpl[0].name = pl1_name;
+			break;
+		case 1 << RAPL_DOMAIN_DRAM:
+			rd->name = rapl_domain_names[RAPL_DOMAIN_DRAM];
+			rd->id = RAPL_DOMAIN_DRAM;
+			rd->msrs[0] = MSR_DRAM_POWER_LIMIT;
+			rd->msrs[1] = MSR_DRAM_ENERGY_STATUS;
+			rd->msrs[2] = MSR_DRAM_PERF_STATUS;
+			rd->msrs[3] = 0;
+			rd->msrs[4] = MSR_DRAM_POWER_INFO;
+			rd->ops = &zone_ops_dram;
+			rd->rpl[0].prim_id = pl1_enable;
+			rd->rpl[0].name = pl1_name;
+			break;
+		}
+		if (mask) {
+			rd->pkg_id = rp->id;
+			rd++;
+		}
+	}
+}
+
+static u64 rapl_unit_xlate(int pkg, enum unit_type type, u64 value, int to_raw)
+{
+	u64 divisor = 1;
+	int scale = 1; /* scale to user friendly data without floating point */
+	u64 f, y; /* fraction and exp. used for time unit */
+	struct rapl_pkg *rp;
+
+	rp = find_package_by_id(pkg);
+	if (!rp)
+		return value;
+
+	switch (type) {
+	case POWER_UNIT:
+		divisor = rp->power_unit_divisor;
+		scale = POWER_UNIT_SCALE;
+		break;
+	case ENERGY_UNIT:
+		scale = ENERGY_UNIT_SCALE;
+		divisor = rp->energy_unit_divisor;
+		break;
+	case TIME_UNIT:
+		divisor = rp->time_unit_divisor;
+		scale = TIME_UNIT_SCALE;
+		/* special processing based on 2^Y*(1+F)/4 = val/divisor */
+		if (!to_raw) {
+			f = (value & 0x60) >> 5;
+			y = value & 0x1f;
+			value = (1<<y)*(4+f)*scale/4;
+			return div64_u64(value, divisor);
+		} else {
+			do_div(value, scale);
+			value *= divisor;
+			y = ilog2(value);
+			f = div64_u64(4 * (value-(1<<y)), 1<<y);
+			value = (y & 0x1f) | ((f&0x3)<<5);
+			return value;
+		}
+		break;
+	case NA_UNIT:
+	default:
+		return value;
+	};
+
+	if (to_raw)
+		return div64_u64(value * divisor, scale);
+	else
+		return div64_u64(value * scale, divisor);
+}
+
+/* in the order of enum rapl_primitives */
+static struct rapl_primitive_info rpi[] = {
+	/* name, mask, shift, msr index, unit divisor */
+	PRIMITIVE_INFO_INIT(energy, ENERGY_STATUS_MASK, 0,
+				RAPL_DOMAIN_MSR_STATUS, ENERGY_UNIT, 0),
+	PRIMITIVE_INFO_INIT(power_limit1, POWER_LIMIT1_MASK, 0,
+				RAPL_DOMAIN_MSR_LIMIT, POWER_UNIT, 0),
+	PRIMITIVE_INFO_INIT(power_limit2, POWER_LIMIT2_MASK, 32,
+				RAPL_DOMAIN_MSR_LIMIT, POWER_UNIT, 0),
+	PRIMITIVE_INFO_INIT(lock, POWER_PP_LOCK, 31,
+				RAPL_DOMAIN_MSR_LIMIT, NA_UNIT, 0),
+	PRIMITIVE_INFO_INIT(pl1_enable, POWER_LIMIT1_ENABLE, 15,
+				RAPL_DOMAIN_MSR_LIMIT, NA_UNIT, 0),
+	PRIMITIVE_INFO_INIT(pl1_clamp, POWER_LIMIT1_CLAMP, 16,
+				RAPL_DOMAIN_MSR_LIMIT, NA_UNIT, 0),
+	PRIMITIVE_INFO_INIT(pl2_enable, POWER_LIMIT2_ENABLE, 47,
+				RAPL_DOMAIN_MSR_LIMIT, NA_UNIT, 0),
+	PRIMITIVE_INFO_INIT(pl2_clamp, POWER_LIMIT2_CLAMP, 48,
+				RAPL_DOMAIN_MSR_LIMIT, NA_UNIT, 0),
+	PRIMITIVE_INFO_INIT(time_window1, TIME_WINDOW1_MASK, 17,
+				RAPL_DOMAIN_MSR_LIMIT, TIME_UNIT, 0),
+	PRIMITIVE_INFO_INIT(time_window2, TIME_WINDOW2_MASK, 49,
+				RAPL_DOMAIN_MSR_LIMIT, TIME_UNIT, 0),
+	PRIMITIVE_INFO_INIT(thermal_spec_power, POWER_INFO_THERMAL_SPEC_MASK,
+				0, RAPL_DOMAIN_MSR_INFO, POWER_UNIT, 0),
+	PRIMITIVE_INFO_INIT(max_power, POWER_INFO_MAX_MASK, 32,
+				RAPL_DOMAIN_MSR_INFO, POWER_UNIT, 0),
+	PRIMITIVE_INFO_INIT(min_power, POWER_INFO_MIN_MASK, 16,
+				RAPL_DOMAIN_MSR_INFO, POWER_UNIT, 0),
+	PRIMITIVE_INFO_INIT(max_window, POWER_INFO_MAX_TIME_WIN_MASK, 48,
+				RAPL_DOMAIN_MSR_INFO, TIME_UNIT, 0),
+	PRIMITIVE_INFO_INIT(throttle_time, PERF_STATUS_THROTTLE_TIME_MASK, 0,
+				RAPL_DOMAIN_MSR_PERF, TIME_UNIT, 0),
+	PRIMITIVE_INFO_INIT(prio_level, PP_POLICY_MASK, 0,
+				RAPL_DOMAIN_MSR_POLICY, NA_UNIT, 0),
+	/* non-hardware */
+	PRIMITIVE_INFO_INIT(average_power, 0, 0, 0, POWER_UNIT,
+				RAPL_PRIMITIVE_DERIVED),
+	{NULL, 0, 0, 0},
+};
+
+/* Read primitive data based on its related struct rapl_primitive_info.
+ * if xlate flag is set, return translated data based on data units, i.e.
+ * time, energy, and power.
+ * RAPL MSRs are non-architectual and are laided out not consistantly across
+ * domains. Here we use primitive info to allow writing consolidated access
+ * functions.
+ * For a given primitive, it is processed by MSR mask and shift. Unit conversion
+ * is pre-assigned based on RAPL unit MSRs read at init time.
+ * 63-------------------------- 31--------------------------- 0
+ * |                           xxxxx (mask)                   |
+ * |                                |<- shift ----------------|
+ * 63-------------------------- 31--------------------------- 0
+ */
+static int rapl_read_data_raw(struct rapl_domain *rd,
+			enum rapl_primitives prim,
+			bool xlate, u64 *data)
+{
+	u64 value, final;
+	u32 msr;
+	struct rapl_primitive_info *rp = &rpi[prim];
+	int cpu;
+
+	if (!rp->name || rp->flag & RAPL_PRIMITIVE_DUMMY)
+		return -EINVAL;
+
+	msr = rd->msrs[rp->id];
+	if (!msr)
+		return -EINVAL;
+	/* use physical package id to look up active cpus */
+	cpu = find_active_cpu_on_pkg(rd->pkg_id);
+	if (cpu < 0)
+		return cpu;
+
+	/* special-case pkg lock bit since pkg domain uses a different bit */
+	if (prim == lock && rd->id == RAPL_DOMAIN_PKG) {
+		rp->mask = POWER_PKG_LOCK;
+		rp->shift = 63;
+	}
+	/* non-hardware data are collected by the polling thread */
+	if (rp->flag & RAPL_PRIMITIVE_DERIVED) {
+		*data = rd->rdd.primitives[prim];
+		return 0;
+	}
+
+	if (rdmsrl_safe_on_cpu(cpu, msr, &value)) {
+		pr_debug("failed to read msr 0x%x on cpu %d\n", msr, cpu);
+		return -EIO;
+	}
+
+	final = value & rp->mask;
+	final = final >> rp->shift;
+	if (xlate)
+		*data = rapl_unit_xlate(rd->pkg_id, rp->unit, final, 0);
+	else
+		*data = final;
+
+	return 0;
+}
+
+/* Similar use of primitive info in the read counterpart */
+static int rapl_write_data_raw(struct rapl_domain *rd,
+			enum rapl_primitives prim,
+			unsigned long long value)
+{
+	u64 msr_val;
+	u32 msr;
+	struct rapl_primitive_info *rp = &rpi[prim];
+	int cpu;
+
+	cpu = find_active_cpu_on_pkg(rd->pkg_id);
+	if (cpu < 0)
+		return cpu;
+	msr = rd->msrs[rp->id];
+	if (rdmsrl_safe_on_cpu(cpu, msr, &msr_val)) {
+		dev_dbg(&rd->pzone->device,
+			"failed to read msr 0x%x on cpu %d\n", msr, cpu);
+		return -EIO;
+	}
+	value = rapl_unit_xlate(rd->pkg_id, rp->unit, value, 1);
+	msr_val &= ~rp->mask;
+	msr_val |= value << rp->shift;
+	if (wrmsrl_safe_on_cpu(cpu, msr, msr_val)) {
+		dev_dbg(&rd->pzone->device,
+			"failed to write msr 0x%x on cpu %d\n", msr, cpu);
+		return -EIO;
+	}
+
+	return 0;
+}
+
+/* Raw RAPL data stored in MSRs are in certain scales. We need to
+ * convert them into standard units based on the divisors reported in
+ * the RAPL unit MSRs.
+ */
+static int rapl_check_unit(struct rapl_pkg *rp, int cpu)
+{
+	u64 msr_val;
+	u32 value;
+
+	if (rdmsrl_safe_on_cpu(cpu, MSR_RAPL_POWER_UNIT, &msr_val)) {
+		pr_err("Failed to read power unit MSR 0x%x on CPU %d, exit.\n",
+			MSR_RAPL_POWER_UNIT, cpu);
+		return -ENODEV;
+	}
+
+	/* energy unit: 1/enery_unit_divisor Joules */
+	value = (msr_val & ENERGY_UNIT_MASK) >> ENERGY_UNIT_OFFSET;
+	rp->energy_unit_divisor = 1 << value;
+
+	/* power unit: 1/power_unit_divisor Watts */
+	value = (msr_val & POWER_UNIT_MASK) >> POWER_UNIT_OFFSET;
+	rp->power_unit_divisor = 1 << value;
+
+	/* time unit: 1/time_unit_divisor Seconds */
+	value = (msr_val & TIME_UNIT_MASK) >> TIME_UNIT_OFFSET;
+	rp->time_unit_divisor = 1 << value;
+
+	pr_debug("Physical package %d units: energy=%d, time=%d, power=%d\n",
+		rp->id,
+		rp->energy_unit_divisor,
+		rp->time_unit_divisor,
+		rp->power_unit_divisor);
+
+	return 0;
+}
+
+/* REVISIT:
+ * When package power limit is set artificially low by RAPL, LVT
+ * thermal interrupt for package power limit should be ignored
+ * since we are not really exceeding the real limit. The intention
+ * is to avoid excessive interrupts while we are trying to save power.
+ * A useful feature might be routing the pkg_power_limit interrupt
+ * to userspace via eventfd. once we have a usecase, this is simple
+ * to do by adding an atomic notifier.
+ */
+
+/* disable and save the original per package power limit interrupt state */
+static void pkg_power_limit_irq_save(int pkg_id)
+{
+	u32 l, h = 0;
+	int cpu;
+	struct rapl_pkg *rp;
+
+	rp = find_package_by_id(pkg_id);
+	if (!rp)
+		return;
+
+	if (!boot_cpu_has(X86_FEATURE_PTS) || !boot_cpu_has(X86_FEATURE_PLN))
+		return;
+
+	cpu = find_active_cpu_on_pkg(pkg_id);
+	if (cpu < 0)
+		return;
+	/* save the state of PLN irq mask bit before disabling it */
+	rdmsr_safe_on_cpu(cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT, &l, &h);
+	if (!(rp->power_limit_irq & PACKAGE_PLN_INT_SAVED)) {
+		rp->power_limit_irq = l & PACKAGE_THERM_INT_PLN_ENABLE;
+		rp->power_limit_irq |= PACKAGE_PLN_INT_SAVED;
+	}
+	l &= ~PACKAGE_THERM_INT_PLN_ENABLE;
+	wrmsr_on_cpu(cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
+}
+
+/* restore per package power limit interrupt enable state */
+static void pkg_power_limit_irq_restore(int pkg_id)
+{
+	u32 l, h;
+	int cpu;
+	struct rapl_pkg *rp;
+
+	rp = find_package_by_id(pkg_id);
+	if (!rp)
+		return;
+
+	if (!boot_cpu_has(X86_FEATURE_PTS) || !boot_cpu_has(X86_FEATURE_PLN))
+		return;
+
+	cpu = find_active_cpu_on_pkg(pkg_id);
+	if (cpu < 0)
+		return;
+
+	/* irq enable state not saved, nothing to restore */
+	if (!(rp->power_limit_irq & PACKAGE_PLN_INT_SAVED))
+		return;
+	rdmsr_safe_on_cpu(cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT, &l, &h);
+
+	if (rp->power_limit_irq & PACKAGE_THERM_INT_PLN_ENABLE)
+		l |= PACKAGE_THERM_INT_PLN_ENABLE;
+	else
+		l &= ~PACKAGE_THERM_INT_PLN_ENABLE;
+
+	wrmsr_on_cpu(cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
+}
+
+static const struct x86_cpu_id rapl_ids[] = {
+	{ X86_VENDOR_INTEL, 6, 0x2a},/* SNB */
+	{ X86_VENDOR_INTEL, 6, 0x2d},/* SNB EP */
+	{ X86_VENDOR_INTEL, 6, 0x3a},/* IVB */
+	{ X86_VENDOR_INTEL, 6, 0x45},/* HSW */
+	/* TODO: Add more CPU IDs after testing */
+	{}
+};
+MODULE_DEVICE_TABLE(x86cpu, rapl_ids);
+
+/* read once for all raw primitive data for all pkgs, domains */
+static void rapl_update_domain_data(void)
+{
+	int dmn, prim;
+	u64 val;
+	struct rapl_pkg *rp;
+
+	list_for_each_entry(rp, &rpkgs, plist) {
+		for (dmn = 0; dmn < rp->nr_domains; dmn++) {
+			pr_debug("update pkg %d domain %s data\n", rp->id,
+				rp->domains[dmn].name);
+			/* exclude non-raw primitives */
+			for (prim = 0; prim < NR_RAW_PRIMITIVES; prim++)
+				if (!rapl_read_data_raw(&rp->domains[dmn], prim,
+								rpi[prim].unit,
+								&val))
+					rp->domains[dmn].rdd.primitives[prim] =
+									val;
+		}
+	}
+
+}
+
+/* unregister all active rapl pkgs from the thermal layer, hotplug lock held */
+static int rapl_unregister_powercap(void)
+{
+	int i;
+	struct rapl_pkg *rp;
+	struct rapl_domain *rd, *rd_pkg = NULL;
+
+	list_for_each_entry(rp, &rpkgs, plist) {
+		pkg_power_limit_irq_restore(rp->id);
+		/* might only need to delete parent */
+		for (i = 0; i < rp->nr_domains; i++) {
+			rd = &rp->domains[i];
+
+			pr_debug("remove pkg %d, %s domain, undo power limit\n",
+				rp->id, rd->name);
+			rapl_write_data_raw(rd, pl1_enable, 0);
+			rapl_write_data_raw(rd, pl2_enable, 0);
+			rapl_write_data_raw(rd, pl1_clamp, 0);
+			rapl_write_data_raw(rd, pl2_clamp, 0);
+			if (rd->id == RAPL_DOMAIN_PKG) {
+				rd_pkg = rd;
+				continue;
+			}
+			powercap_zone_unregister(pccr, rd->pzone);
+		}
+		/* do the package zone last */
+		if (rd_pkg)
+			powercap_zone_unregister(pccr, rd_pkg->pzone);
+	}
+	powercap_deallocate_controller(pccr);
+
+	return 0;
+}
+
+static int rapl_pkg_register_powercap(struct rapl_pkg *rp)
+{
+	int i;
+	struct rapl_domain *rd;
+	int ret = 0;
+	char dev_name[POWERCAP_ZONE_NAME_LENGTH];
+	struct powercap_zone_device *pzdev = NULL;
+	int nr_pl;
+
+	/* first we register package domain as the parent zone*/
+	for (i = 0; i < rp->nr_domains; i++) {
+		rd = &rp->domains[i];
+		if (rd->id == RAPL_DOMAIN_PKG) {
+			nr_pl = find_nr_power_limit(rd);
+			pr_debug("register socket %d pkg domain %d\n",
+				rp->id, i);
+			memset(dev_name, 0, sizeof(dev_name));
+			snprintf(dev_name, sizeof(dev_name), "%s-%d",
+				rd->name, rp->id);
+			pzdev = powercap_zone_register(pccr, dev_name,
+						NULL, rd->ops, nr_pl, &const_ops);
+			if (IS_ERR(pzdev)) {
+				pr_debug("failed to register pkg, %d\n",
+					rp->id);
+				ret = PTR_ERR(pzdev);
+				goto exit_pkg;
+			}
+			powercap_set_zone_data(pzdev, rd);
+			rd->pzone = pzdev;
+			rp->pzone = pzdev;
+			/* done, only one package domain per socket */
+			break;
+		}
+	}
+	if (!pzdev) {
+		pr_err("no package domain found, unknown topology!\n");
+		ret = -ENODEV;
+		goto exit_pkg;
+	}
+	/* now register domains as children of the socket/package*/
+	for (i = 0; i < rp->nr_domains; i++) {
+		rd = &rp->domains[i];
+		if (rd->id == RAPL_DOMAIN_PKG)
+			continue;
+		/* number of power limits per domain varies */
+		nr_pl = find_nr_power_limit(rd);
+		pzdev = powercap_zone_register(pccr, rd->name,
+					rp->pzone, rd->ops, nr_pl, &const_ops);
+
+		if (IS_ERR(pzdev)) {
+			pr_debug("failed to register pzdev, %d:%s:%s\n",
+				rp->id, rd->name, dev_name);
+			ret = PTR_ERR(pzdev);
+			goto err_cleanup;
+		}
+		powercap_set_zone_data(pzdev, rd);
+		rd->pzone = pzdev;
+	}
+
+exit_pkg:
+	return ret;
+err_cleanup:
+	/* clean up previously initialized domains within the package if we
+	 * failed after the first domain setup.
+	 */
+	while (--i >= 0) {
+		pr_debug("unregister zone/pkg %d domain %d\n", rp->id, i);
+		rd = &rp->domains[i];
+		powercap_zone_unregister(pccr, rd->pzone);
+	}
+
+	return ret;
+}
+
+static int rapl_register_powercap(void)
+{
+	int i = 0;
+	struct rapl_domain *rd;
+	int ret = 0;
+	struct rapl_pkg *rp;
+
+	pccr = powercap_allocate_controller("intel-rapl");
+	if (IS_ERR(pccr)) {
+		pr_debug("failed to register powercap controller.\n");
+		return PTR_ERR(pccr);
+	}
+	/* read the initial data */
+	rapl_update_domain_data();
+	list_for_each_entry(rp, &rpkgs, plist)
+		if (rapl_pkg_register_powercap(rp))
+			goto err_cleanup_pkg;
+	return ret;
+
+err_cleanup_pkg:
+	/* clean up previously initialized pkgs */
+	list_for_each_entry_continue_reverse(rp, &rpkgs, plist) {
+		for (i = 0; i < rp->nr_domains; i++) {
+			rd = &rp->domains[i];
+			pr_debug("unregister zone/pkg %d, %s domain\n",
+				rp->id, rd->name);
+			powercap_zone_unregister(pccr, rd->pzone);
+		}
+	}
+
+	return ret;
+}
+
+static int rapl_check_domain(int cpu, int domain)
+{
+	unsigned msr;
+	u64 val1, val2 = 0;
+	int retry = 0;
+
+	switch (domain) {
+	case RAPL_DOMAIN_PKG:
+		msr = MSR_PKG_ENERGY_STATUS;
+		break;
+	case RAPL_DOMAIN_PP0:
+		msr = MSR_PP0_ENERGY_STATUS;
+		break;
+	case RAPL_DOMAIN_PP1:
+		msr = MSR_PP1_ENERGY_STATUS;
+		break;
+	case RAPL_DOMAIN_DRAM:
+		msr = MSR_DRAM_ENERGY_STATUS;
+		break;
+	default:
+		pr_err("invalid domain id %d\n", domain);
+		return -EINVAL;
+	}
+	if (rdmsrl_safe_on_cpu(cpu, msr, &val1))
+		return -ENODEV;
+
+	/* energy counters roll slowly on some domains */
+	while (++retry < 10) {
+		usleep_range(10000, 15000);
+		rdmsrl_safe_on_cpu(cpu, msr, &val2);
+		if ((val1 & ENERGY_STATUS_MASK) != (val2 & ENERGY_STATUS_MASK))
+			return 0;
+	}
+	/* if energy counter does not change, report as bad domain */
+	pr_info("domain %s energy ctr %llu:%llu not working, skip\n",
+		rapl_domain_names[domain], val1, val2);
+
+	return -ENODEV;
+}
+
+/* Detect active and valid domains for the given CPU, caller must
+ * ensure the CPU belongs to the targeted package and CPU hotlug is disabled.
+ */
+static int rapl_detect_domains(struct rapl_pkg *rp, int cpu)
+{
+	int i;
+	int ret = 0;
+	struct rapl_domain *rd;
+	u64 locked;
+
+	for (i = 0; i < RAPL_DOMAIN_MAX; i++) {
+		/* use physical package id to read counters */
+		if (!rapl_check_domain(cpu, i))
+			rp->domain_map |= 1 << i;
+	}
+	rp->nr_domains = bitmap_weight(&rp->domain_map,	RAPL_DOMAIN_MAX);
+	if (!rp->nr_domains) {
+		pr_err("no valid rapl domains found in pkg %d\n", rp->id);
+		ret = -ENODEV;
+		goto done;
+	}
+	pr_debug("found %d domains on package %d\n", rp->nr_domains, rp->id);
+
+	rp->domains = kcalloc(rp->nr_domains + 1, sizeof(struct rapl_domain),
+			GFP_KERNEL);
+	if (!rp->domains) {
+		ret = -ENOMEM;
+		goto done;
+	}
+	rapl_init_domains(rp);
+
+	for (i = 0; i < rp->nr_domains; i++) {
+		rd = &rp->domains[i];
+		/* check if the domain is locked by BIOS */
+		if (rapl_read_data_raw(rd, lock, false, &locked)) {
+			pr_info("RAPL pkg %d domain %s locked by BIOS\n",
+				rp->id, rd->name);
+				rd->state |= DOMAIN_STATE_BIOS_LOCKED;
+		}
+	}
+
+
+done:
+	return ret;
+}
+
+static bool is_package_new(int pkg)
+{
+	struct rapl_pkg *rp;
+
+	/* caller prevents cpu hotplug, there will be no new packages added
+	 * or deleted while traversing the package list, no need for locking.
+	 */
+	list_for_each_entry(rp, &rpkgs, plist) {
+		if (pkg == rp->id)
+			return false;
+	}
+
+	return true;
+}
+
+/* RAPL interface can be made of a two-level hierarchy: package level and domain
+ * level. We first detect the number of packages then domains of each package.
+ * We have to consider the possiblity of CPU online/offline due to hotplug and
+ * other scenarios.
+ */
+static int rapl_detect_topology(void)
+{
+	int i;
+	int phy_pkg_id;
+	struct rapl_pkg *new_pkg, *rp;
+
+	for_each_online_cpu(i) {
+		phy_pkg_id = topology_physical_package_id(i);
+		if (is_package_new(phy_pkg_id)) {
+			new_pkg = kzalloc(sizeof(struct rapl_pkg), GFP_KERNEL);
+			if (!new_pkg) {
+				rapl_cleanup_data();
+				return -ENOMEM;
+			}
+			/* add the new package to the list */
+			new_pkg->id = phy_pkg_id;
+			new_pkg->nr_cpus = 1;
+
+			/* check if the package contains valid domains */
+			if (rapl_detect_domains(new_pkg, i) ||
+				rapl_check_unit(new_pkg, i)) {
+				kfree(new_pkg->domains);
+				kfree(new_pkg);
+				/* free up the pkgs already initialized */
+				rapl_cleanup_data();
+				return -ENODEV;
+			}
+			INIT_LIST_HEAD(&new_pkg->plist);
+			list_add(&new_pkg->plist, &rpkgs);
+		} else {
+			rp = find_package_by_id(phy_pkg_id);
+			if (rp)
+				++rp->nr_cpus;
+		}
+	}
+
+	return 0;
+}
+
+/* called from CPU hotplug notifier, hotplug lock held */
+static void rapl_remove_pkg(struct rapl_pkg *rp)
+{
+	struct rapl_domain *rd;
+	int i;
+
+	for (i = 0; i < rp->nr_domains; i++) {
+		rd = &rp->domains[i];
+		pr_debug("remove pkg %d, %s domain\n", rp->id, rd->name);
+		powercap_zone_unregister(pccr, rd->pzone);
+	}
+	kfree(rp->domains);
+	list_del(&rp->plist);
+	kfree(rp);
+}
+
+/* called from CPU hotplug notifier, hotplug lock held */
+static int rapl_add_pkg(int cpu)
+{
+	int ret = 0;
+	int phy_pkg_id;
+	struct rapl_pkg *rp;
+
+	phy_pkg_id = topology_physical_package_id(cpu);
+	rp = kzalloc(sizeof(struct rapl_pkg), GFP_KERNEL);
+	if (!rp)
+		return -ENOMEM;
+
+	/* add the new package to the list */
+	rp->id = phy_pkg_id;
+	rp->nr_cpus = 1;
+	/* check if the package contains valid domains */
+	if (rapl_detect_domains(rp, cpu) ||
+		rapl_check_unit(rp, cpu)) {
+		ret = -ENODEV;
+		goto err_free_pkg;
+	}
+	if (!rapl_pkg_register_powercap(rp)) {
+		INIT_LIST_HEAD(&rp->plist);
+		list_add(&rp->plist, &rpkgs);
+		return ret;
+	}
+
+err_free_pkg:
+	kfree(rp->domains);
+	kfree(rp);
+
+	return ret;
+}
+
+/* Handles CPU hotplug on multi-socket systems.
+ * If a CPU goes online as the first CPU of the physical package
+ * we add the RAPL package to the system. Similarly, when the last
+ * CPU of the package is removed, we remove the RAPL package and its
+ * associated domains. Cooling devices are handled accordingly at
+ * per-domain level.
+ */
+static int rapl_cpu_callback(struct notifier_block *nfb,
+				unsigned long action, void *hcpu)
+{
+	unsigned long cpu = (unsigned long)hcpu;
+	int phy_pkg_id;
+	struct rapl_pkg *rp;
+
+	phy_pkg_id = topology_physical_package_id(cpu);
+	switch (action) {
+	case CPU_ONLINE:
+	case CPU_ONLINE_FROZEN:
+	case CPU_DOWN_FAILED:
+	case CPU_DOWN_FAILED_FROZEN:
+		rp = find_package_by_id(phy_pkg_id);
+		if (rp)
+			++rp->nr_cpus;
+		else
+			rapl_add_pkg(cpu);
+		break;
+	case CPU_DOWN_PREPARE:
+	case CPU_DOWN_PREPARE_FROZEN:
+		rp = find_package_by_id(phy_pkg_id);
+		if (!rp)
+			break;
+		if (--rp->nr_cpus == 0)
+			rapl_remove_pkg(rp);
+	}
+
+	return NOTIFY_OK;
+}
+
+static struct notifier_block rapl_cpu_notifier = {
+	.notifier_call = rapl_cpu_callback,
+};
+
+static int __init rapl_init(void)
+{
+	int ret = 0;
+
+	if (!x86_match_cpu(rapl_ids)) {
+		pr_err("driver does not support CPU family %d model %d\n",
+			boot_cpu_data.x86, boot_cpu_data.x86_model);
+
+		return -ENODEV;
+	}
+	/* prevent CPU hotplug during detection */
+	get_online_cpus();
+	ret = rapl_detect_topology();
+	if (ret)
+		goto done;
+
+	if (rapl_register_powercap()) {
+		rapl_cleanup_data();
+		ret = -ENODEV;
+		goto done;
+	}
+	register_hotcpu_notifier(&rapl_cpu_notifier);
+done:
+	put_online_cpus();
+
+	return ret;
+}
+
+static void __exit rapl_exit(void)
+{
+	get_online_cpus();
+	unregister_hotcpu_notifier(&rapl_cpu_notifier);
+	rapl_unregister_powercap();
+	rapl_cleanup_data();
+	put_online_cpus();
+}
+
+module_init(rapl_init);
+module_exit(rapl_exit);
+
+MODULE_DESCRIPTION("Driver for Intel RAPL (Running Average Power Limit)");
+MODULE_AUTHOR("Jacob Pan <jacob.jun.pan@intel.com>");
+MODULE_LICENSE("GPL v2");
-- 
1.8.3.1

Re: [RFC v02 5/5] Introduce Intel RAPL power capping driver

[Joe Perches]

On Wed, 2013-08-07 at 09:12 -0700, Srinivas Pandruvada wrote:
> RAPL(Running Average Power Limit) interface provides platform software
> with the ability to monitor, control, and get notifications on SOC
> power consumptions.

trivial notes:

> diff --git a/drivers/powercap/intel_rapl.c b/drivers/powercap/intel_rapl.c
[]
> +#define POWER_LIMIT1_MASK       0x7FFF
> +#define POWER_LIMIT1_ENABLE     BIT(15)
> +#define POWER_LIMIT1_CLAMP      BIT(16)
> +
> +#define POWER_LIMIT2_MASK       (0x7FFFULL<<32)
> +#define POWER_LIMIT2_ENABLE     (0x1ULL<<47)
> +#define POWER_LIMIT2_CLAMP      (0x1ULL<<48)
> +#define POWER_PKG_LOCK          (0x1ULL<<63)

Maybe it's time add a generic bitops.h macro

#define BIT_ULL(nr)	(1ULL << (nr))

[]

> +/* called after domain detection and package level data are set */
> +static void rapl_init_domains(struct rapl_pkg *rp)
> +{
> +	int i;
> +	struct rapl_domain *rd = rp->domains;
> +
> +	for (i = 0; i < RAPL_DOMAIN_MAX; i++) {
> +		unsigned int mask = rp->domain_map & (1 << i);
> +		switch (mask) {
> +		case 1 << RAPL_DOMAIN_PKG:

Maybe use defines or BIT(RAPL_DOMAIN_PKG)

Other than that the code seems to read pretty easily.

Re: [RFC v02 1/5] PowerCap: Documentation

[Rob Landley]

On 08/07/2013 11:12:41 AM, Srinivas Pandruvada wrote:
> Added power cap framework documentation. This explains the use of  
> power capping
> framework, sysfs and programming interface.
> There are two documents:
> Documentation/powercap/PowerCappingFramework.txt: Explains use case  
> and API in
> details.
> Documentation/ABI/testing/sysfs-class-powercap: Explains ABIs.
> 
> Reviewed-by: Len Brown <len.brown@intel.com>
> Signed-off-by: Srinivas Pandruvada  
> <srinivas.pandruvada@linux.intel.com>
> Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
> ---
>  Documentation/ABI/testing/sysfs-class-powercap   | 165 ++++++
>  Documentation/powercap/PowerCappingFramework.txt | 686  
> +++++++++++++++++++++++
...
> --- /dev/null
> +++ b/Documentation/powercap/PowerCappingFramework.txt
> @@ -0,0 +1,686 @@
> +Power Capping Framework
> +==================================
> +
> +The Linux Power Capping Framework provides user-space with a common
> +API to kernel-mode power-capping drivers.  At the same time,
> +it provides the hardware-specific power-capping drivers with
> +a uniform API to user-space.

s/.  At the same time, it provides/, and/

> +Terminology
> +=========================
> +The Power Capping framework organizes power capping devices under a  
> tree structure.
> +At the root level, each device is under some "controller", which is  
> the enabler
> +of technology.

A controller is the enabler of technology?

What does that mean?

> For example this can be "RAPL".

Ah, clears it right up.

> +Under each controllers,

each doesn't take a plural.

> there are multiple power zones, which can be independently
> +monitored and controlled.
> +Each power zone can be organized as a tree with parent, children and  
> siblings.
> +Each power zone defines attributes to enable power monitoring and  
> constraints.
> +
> +Example sysfs interface tree:
> +
> +/sys/devices/virtual/power_cap
> +└── intel-rapl
... intel intel intel intel...
> +
> +For example, above powercap sysfs tree represents RAPL(Running  
> Average Power Limit)
> +type controls available in the Intel® 64 and IA-32 Processor  
> Architectures. Here

What are the chances of this ever being applied to a non-intel  
processor? (Should it be under Documentation/x86, or is it presented as  
something with a nonzero chance of actually ever being generic?)

> +under controller "intel-rapl" there are two CPU packages  
> (package-0/1), which can
> +provide power monitoring and controls (intel-rapl:0 and  
> intel-rapl:1). Each power
> +zone has a name.
> +For example:
> +cat /sys/class/power_cap/intel-rapl/intel-rapl:0/name
> +package-0
> +
> +In addition to providing monitoring and control at package level,  
> each package
> +is further divided into child power zones (called domains in the RAPL
> specifications).

Where are the RAPL specifications, and is this framework just an  
implementation of them or is it more generic?

> +Here zones represent controls for core and dram  parts. These zones  
> can be represented
> +as children of package.
> +For example:
> +cat /sys/class/power_cap/intel-rapl/intel-rapl:0/intel-rapl:0:1/name
> +dram
> +
> +Under RAPL framework there are two constraints, one for
> +short term and one for long term, with two different time windows.  
> These can be
> +represented as two constraints, with different time windows, power  
> limits and names.
> +For example:
> +	constraint_0_name
> +	constraint_0_power_limit_uw
> +	constraint_0_time_window_us
> +	constraint_1_name
> +	constraint_1_power_limit_uw
> +	constraint_1_time_window_us
> +
> +Power Zone Attributes
> +=================================
> +Monitoring attributes
> +----------------------
> +
> +energy_uj (rw): Current energy counter in micro joules. Write "0" to  
> reset.
> +If the counter can not be reset, then this attribute is read only.
> +
> +max_energy_range_uj (ro): Range of the above energy counter in  
> micro-joules.
> +
> +power_uw (rw): Current power in micro watts. Write "0" to resets the  
> value.
> +If the value can not be reset, then this attribute is read only.
> +
> +max_power_range_uw (ro): Range of the above power value in  
> micro-watts.
> +
> +name (ro): Name of this power zone.
> +
> +It is possible that some domains can have both power and energy  
> counters and
> +ranges, but at least one is mandatory.
> +
> +Constraints
> +----------------
> +constraint_X_power_limit_uw (rw): Power limit in micro watts, which  
> should be
> +applicable for the time window specified by  
> "constraint_X_time_window_us".
> +
> +constraint_X_time_window_us (rw): Time window in micro seconds.
> +
> +constraint_X_name (ro): An optional name of the constraint
> +
> +constraint_X_max_power_uw(ro): Maximum allowed power in micro watts.
> +
> +constraint_X_min_power_uw(ro): Minimum allowed power in micro watts.
> +
> +constraint_X_max_time_window_us(ro): Maximum allowed time window in  
> micro seconds.
> +
> +constraint_X_min_time_window_us(ro): Minimum allowed time window in  
> micro seconds.
> +
> +In addition each node has an attribute "type", which shows, whether  
> is a controller
> +or power zone. Except power_limit_uw and time_window_us other fields  
> are optional.
> +
> +Power Cap Client Driver Interface
> +==================================
> +The API summary:
> +
> +Call powercap_allocate_controller to define a controller with a name.
> +Call powercap_zone_register for each power zone for this controller.
> +power zones can have other power zone as a parent or don't have a
> +parent.

Trying not to nitpick "english isn't a first language here", but...

Power zones can have another power zone as a parent or no parent.

> +During powercap_zone_register defines number of constraints and  
> callbacks.
> +
> +To Free a power zone call powercap_zone_unregister.
> +To free a controller call powercap_deallocate_controller.
> +
> +Rest of this document is generated by using kernel-doc on
> +powercap.h

Isn't that what Documentation/DocBook is for? (If powercap.h is  
modified the need to update this file is nonobvious...)

Rob

Re: [RFC v02 1/5] PowerCap: Documentation

[Srinivas Pandruvada]

On 08/08/2013 06:43 AM, Rob Landley wrote:
> On 08/07/2013 11:12:41 AM, Srinivas Pandruvada wrote:
>> Added power cap framework documentation. This explains the use of 
>> power capping
>> framework, sysfs and programming interface.
>> There are two documents:
>> Documentation/powercap/PowerCappingFramework.txt: Explains use case 
>> and API in
>> details.
>> Documentation/ABI/testing/sysfs-class-powercap: Explains ABIs.
>>
>> Reviewed-by: Len Brown <len.brown@intel.com>
>> Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
>> Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
>> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
>> ---
>>  Documentation/ABI/testing/sysfs-class-powercap   | 165 ++++++
>>  Documentation/powercap/PowerCappingFramework.txt | 686 
>> +++++++++++++++++++++++
> ...
>> --- /dev/null
>> +++ b/Documentation/powercap/PowerCappingFramework.txt
>> @@ -0,0 +1,686 @@
>> +Power Capping Framework
>> +==================================
>> +
>> +The Linux Power Capping Framework provides user-space with a common
>> +API to kernel-mode power-capping drivers.  At the same time,
>> +it provides the hardware-specific power-capping drivers with
>> +a uniform API to user-space.
>
> s/.  At the same time, it provides/, and/
>
>> +Terminology
>> +=========================
>> +The Power Capping framework organizes power capping devices under a 
>> tree structure.
>> +At the root level, each device is under some "controller", which is 
>> the enabler
>> +of technology.
>
> A controller is the enabler of technology?
>
> What does that mean?
>
>> For example this can be "RAPL".
>
> Ah, clears it right up.
>
>> +Under each controllers,
>
> each doesn't take a plural.
>
>> there are multiple power zones, which can be independently
>> +monitored and controlled.
>> +Each power zone can be organized as a tree with parent, children and 
>> siblings.
>> +Each power zone defines attributes to enable power monitoring and 
>> constraints.
>> +
>> +Example sysfs interface tree:
>> +
>> +/sys/devices/virtual/power_cap
>> +└── intel-rapl
> ... intel intel intel intel...
>> +
>> +For example, above powercap sysfs tree represents RAPL(Running 
>> Average Power Limit)
>> +type controls available in the Intel® 64 and IA-32 Processor 
>> Architectures. Here
>
> What are the chances of this ever being applied to a non-intel 
> processor? (Should it be under Documentation/x86, or is it presented 
> as something with a nonzero chance of actually ever being generic?)
>
This framework has nothing to do with Intel processors. This is not 
processor specific. It can be even used for non processor parts like a 
wireless module.
Any part which allows some configuration of its power budget, can use 
this framework.
>> +under controller "intel-rapl" there are two CPU packages 
>> (package-0/1), which can
>> +provide power monitoring and controls (intel-rapl:0 and 
>> intel-rapl:1). Each power
>> +zone has a name.
>> +For example:
>> +cat /sys/class/power_cap/intel-rapl/intel-rapl:0/name
>> +package-0
>> +
>> +In addition to providing monitoring and control at package level, 
>> each package
>> +is further divided into child power zones (called domains in the RAPL
>> specifications).
>
> Where are the RAPL specifications, and is this framework just an 
> implementation of them or is it more generic?
RAPL specifications are part of Intel Software Developer's manual. This 
can be downloaded from Intel website.
>> +Here zones represent controls for core and dram  parts. These zones 
>> can be represented
>> +as children of package.
>> +For example:
>> +cat /sys/class/power_cap/intel-rapl/intel-rapl:0/intel-rapl:0:1/name
>> +dram
>> +
>> +Under RAPL framework there are two constraints, one for
>> +short term and one for long term, with two different time windows. 
>> These can be
>> +represented as two constraints, with different time windows, power 
>> limits and names.
>> +For example:
>> +    constraint_0_name
>> +    constraint_0_power_limit_uw
>> +    constraint_0_time_window_us
>> +    constraint_1_name
>> +    constraint_1_power_limit_uw
>> +    constraint_1_time_window_us
>> +
>> +Power Zone Attributes
>> +=================================
>> +Monitoring attributes
>> +----------------------
>> +
>> +energy_uj (rw): Current energy counter in micro joules. Write "0" to 
>> reset.
>> +If the counter can not be reset, then this attribute is read only.
>> +
>> +max_energy_range_uj (ro): Range of the above energy counter in 
>> micro-joules.
>> +
>> +power_uw (rw): Current power in micro watts. Write "0" to resets the 
>> value.
>> +If the value can not be reset, then this attribute is read only.
>> +
>> +max_power_range_uw (ro): Range of the above power value in micro-watts.
>> +
>> +name (ro): Name of this power zone.
>> +
>> +It is possible that some domains can have both power and energy 
>> counters and
>> +ranges, but at least one is mandatory.
>> +
>> +Constraints
>> +----------------
>> +constraint_X_power_limit_uw (rw): Power limit in micro watts, which 
>> should be
>> +applicable for the time window specified by 
>> "constraint_X_time_window_us".
>> +
>> +constraint_X_time_window_us (rw): Time window in micro seconds.
>> +
>> +constraint_X_name (ro): An optional name of the constraint
>> +
>> +constraint_X_max_power_uw(ro): Maximum allowed power in micro watts.
>> +
>> +constraint_X_min_power_uw(ro): Minimum allowed power in micro watts.
>> +
>> +constraint_X_max_time_window_us(ro): Maximum allowed time window in 
>> micro seconds.
>> +
>> +constraint_X_min_time_window_us(ro): Minimum allowed time window in 
>> micro seconds.
>> +
>> +In addition each node has an attribute "type", which shows, whether 
>> is a controller
>> +or power zone. Except power_limit_uw and time_window_us other fields 
>> are optional.
>> +
>> +Power Cap Client Driver Interface
>> +==================================
>> +The API summary:
>> +
>> +Call powercap_allocate_controller to define a controller with a name.
>> +Call powercap_zone_register for each power zone for this controller.
>> +power zones can have other power zone as a parent or don't have a
>> +parent.
>
> Trying not to nitpick "english isn't a first language here", but...
>
> Power zones can have another power zone as a parent or no parent.
>
Thanks for pointing out grammatical errors. I have to let our technical 
writer reviews this.
>> +During powercap_zone_register defines number of constraints and 
>> callbacks.
>> +
>> +To Free a power zone call powercap_zone_unregister.
>> +To free a controller call powercap_deallocate_controller.
>> +
>> +Rest of this document is generated by using kernel-doc on
>> +powercap.h
>
> Isn't that what Documentation/DocBook is for? (If powercap.h is 
> modified the need to update this file is nonobvious...)
Yes. But at RFC stage, it is easy to review, if I paste here.
>
> Rob

Re: [RFC v02 0/5] Power Capping Framework and RAPL Driver

[Srinivas Pandruvada]

On 08/07/2013 09:12 AM, Srinivas Pandruvada wrote:
> Overview
> With the evolution of technologies, which enables power monitoring and limiting,
> more and more devices are able to constrain their power consumption under certain
> limits. There are several use cases for such technologies:
> - Power monitoring: Each device can report its power consumption.
> - Power Limiting: Setting power limits on the devices allows users to guard against
> platform reaching max system power level.
> - Maximize performance: While staying below a power limit, it allows devices to
> automatically adjust performance to meet demands
> - Dynamic control and re-budgeting: If each device can be constrained to some power,
> extra power can redistributed to other devices, which needs additional performance.
>
> One such example of technology is RAPL (Running Average Power Limit) mechanism
> available in the latest Intel Processors. Intel is slowly adding many devices under
> RAPL control. Also there are other technologies available, for power capping various
> devices. Soon it is very likely that other vendors are also adding or considering
> such implementation.
>
> Power Capping framework is an effort to have a uniform interface available to Linux
> drivers, which will enable
> - A uniform sysfs interface for all devices which can offer power capping
> - A common API for drivers, which will avoid code duplication and easy
> implementation of client drivers.
>
> Also submitting Intel RAPL driver using power capping framework.
>
>
> Revisions:
Thanks for everyone for review. I have received enough negative 
comments, which will require another version.
So please hold off your reviews till next version.

> v02:
> Sign-offs and reviewed-by tags
> Stylistic issues suggested by Joe Perches
> Removed "counter" from power_uw documentation as pointed by Jonathan Corbet
> Submitting Intel RAPL driver using power capping framework
>
> v01:
> Use device model only to register zones and controllers.
>
> v00:
> Presented options
>
> Jacob Pan (2):
>    x86/msr: add 64bit _on_cpu access functions
>    Introduce Intel RAPL power capping driver
>
> Srinivas Pandruvada (3):
>    PowerCap: Documentation
>    PowerCap: Add class driver
>    PowerCap: Added to drivers build
>
>   Documentation/ABI/testing/sysfs-class-powercap   |  165 +++
>   Documentation/powercap/PowerCappingFramework.txt |  686 ++++++++++++
>   arch/x86/include/asm/msr.h                       |   22 +
>   arch/x86/lib/msr-smp.c                           |   62 +
>   drivers/Kconfig                                  |    2 +
>   drivers/Makefile                                 |    1 +
>   drivers/powercap/Kconfig                         |   24 +
>   drivers/powercap/Makefile                        |    6 +
>   drivers/powercap/intel_rapl.c                    | 1305 ++++++++++++++++++++++
>   drivers/powercap/powercap_sys.c                  |  995 +++++++++++++++++
>   include/linux/powercap.h                         |  300 +++++
>   11 files changed, 3568 insertions(+)
>   create mode 100644 Documentation/ABI/testing/sysfs-class-powercap
>   create mode 100644 Documentation/powercap/PowerCappingFramework.txt
>   create mode 100644 drivers/powercap/Kconfig
>   create mode 100644 drivers/powercap/Makefile
>   create mode 100644 drivers/powercap/intel_rapl.c
>   create mode 100644 drivers/powercap/powercap_sys.c
>   create mode 100644 include/linux/powercap.h
>

[RFC v02 4/5] x86/msr: add 64bit _on_cpu access functions

[Srinivas Pandruvada]

From: Jacob Pan <jacob.jun.pan@linux.intel.com>

Having 64-bit MSR access methods on given CPU can avoid shifting and
simplify MSR content manipulation. We already have other combinations
of rdmsrl_xxx and wrmsrl_xxx but missing the _on_cpu version.

Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
---
 arch/x86/include/asm/msr.h | 22 ++++++++++++++++
 arch/x86/lib/msr-smp.c     | 62 ++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 84 insertions(+)

diff --git a/arch/x86/include/asm/msr.h b/arch/x86/include/asm/msr.h
index cb75028..e139b13 100644
--- a/arch/x86/include/asm/msr.h
+++ b/arch/x86/include/asm/msr.h
@@ -218,10 +218,14 @@ void msrs_free(struct msr *msrs);
 #ifdef CONFIG_SMP
 int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h);
 int wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h);
+int rdmsrl_on_cpu(unsigned int cpu, u32 msr_no, u64 *q);
+int wrmsrl_on_cpu(unsigned int cpu, u32 msr_no, u64 q);
 void rdmsr_on_cpus(const struct cpumask *mask, u32 msr_no, struct msr *msrs);
 void wrmsr_on_cpus(const struct cpumask *mask, u32 msr_no, struct msr *msrs);
 int rdmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h);
 int wrmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h);
+int rdmsrl_safe_on_cpu(unsigned int cpu, u32 msr_no, u64 *q);
+int wrmsrl_safe_on_cpu(unsigned int cpu, u32 msr_no, u64 q);
 int rdmsr_safe_regs_on_cpu(unsigned int cpu, u32 regs[8]);
 int wrmsr_safe_regs_on_cpu(unsigned int cpu, u32 regs[8]);
 #else  /*  CONFIG_SMP  */
@@ -235,6 +239,16 @@ static inline int wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
 	wrmsr(msr_no, l, h);
 	return 0;
 }
+static inline int rdmsrl_on_cpu(unsigned int cpu, u32 msr_no, u64 *q)
+{
+	rdmsrl(msr_no, *q);
+	return 0;
+}
+static inline int wrmsrl_on_cpu(unsigned int cpu, u32 msr_no, u64 q)
+{
+	wrmsrl(msr_no, q);
+	return 0;
+}
 static inline void rdmsr_on_cpus(const struct cpumask *m, u32 msr_no,
 				struct msr *msrs)
 {
@@ -254,6 +268,14 @@ static inline int wrmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
 {
 	return wrmsr_safe(msr_no, l, h);
 }
+static inline int rdmsrl_safe_on_cpu(unsigned int cpu, u32 msr_no, u64 *q)
+{
+	return rdmsrl_safe(msr_no, q);
+}
+static inline int wrmsrl_safe_on_cpu(unsigned int cpu, u32 msr_no, u64 q)
+{
+	return wrmsrl_safe(msr_no, q);
+}
 static inline int rdmsr_safe_regs_on_cpu(unsigned int cpu, u32 regs[8])
 {
 	return rdmsr_safe_regs(regs);
diff --git a/arch/x86/lib/msr-smp.c b/arch/x86/lib/msr-smp.c
index a6b1b86..518532e 100644
--- a/arch/x86/lib/msr-smp.c
+++ b/arch/x86/lib/msr-smp.c
@@ -47,6 +47,21 @@ int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h)
 }
 EXPORT_SYMBOL(rdmsr_on_cpu);
 
+int rdmsrl_on_cpu(unsigned int cpu, u32 msr_no, u64 *q)
+{
+	int err;
+	struct msr_info rv;
+
+	memset(&rv, 0, sizeof(rv));
+
+	rv.msr_no = msr_no;
+	err = smp_call_function_single(cpu, __rdmsr_on_cpu, &rv, 1);
+	*q = rv.reg.q;
+
+	return err;
+}
+EXPORT_SYMBOL(rdmsrl_on_cpu);
+
 int wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
 {
 	int err;
@@ -63,6 +78,22 @@ int wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
 }
 EXPORT_SYMBOL(wrmsr_on_cpu);
 
+int wrmsrl_on_cpu(unsigned int cpu, u32 msr_no, u64 q)
+{
+	int err;
+	struct msr_info rv;
+
+	memset(&rv, 0, sizeof(rv));
+
+	rv.msr_no = msr_no;
+	rv.reg.q = q;
+
+	err = smp_call_function_single(cpu, __wrmsr_on_cpu, &rv, 1);
+
+	return err;
+}
+EXPORT_SYMBOL(wrmsrl_on_cpu);
+
 static void __rwmsr_on_cpus(const struct cpumask *mask, u32 msr_no,
 			    struct msr *msrs,
 			    void (*msr_func) (void *info))
@@ -159,6 +190,37 @@ int wrmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
 }
 EXPORT_SYMBOL(wrmsr_safe_on_cpu);
 
+int wrmsrl_safe_on_cpu(unsigned int cpu, u32 msr_no, u64 q)
+{
+	int err;
+	struct msr_info rv;
+
+	memset(&rv, 0, sizeof(rv));
+
+	rv.msr_no = msr_no;
+	rv.reg.q = q;
+
+	err = smp_call_function_single(cpu, __wrmsr_safe_on_cpu, &rv, 1);
+
+	return err ? err : rv.err;
+}
+EXPORT_SYMBOL(wrmsrl_safe_on_cpu);
+
+int rdmsrl_safe_on_cpu(unsigned int cpu, u32 msr_no, u64 *q)
+{
+	int err;
+	struct msr_info rv;
+
+	memset(&rv, 0, sizeof(rv));
+
+	rv.msr_no = msr_no;
+	err = smp_call_function_single(cpu, __rdmsr_safe_on_cpu, &rv, 1);
+	*q = rv.reg.q;
+
+	return err ? err : rv.err;
+}
+EXPORT_SYMBOL(rdmsrl_safe_on_cpu);
+
 /*
  * These variants are significantly slower, but allows control over
  * the entire 32-bit GPR set.
-- 
1.8.3.1