HAL interface

The HAL interface, declared in sensors.h, represents the interface between the Android framework and the hardware-specific software. A HAL implementation must define each function declared in sensors.h. The main functions are:

  • get_sensors_list - Returns the list of all sensors.
  • activate - Starts or stops a sensor.
  • batch - Sets a sensor’s parameters such as sampling frequency and maximum reporting latency.
  • setDelay - Used only in HAL version 1.0. Sets the sampling frequency for a given sensor.
  • flush - Flushes the FIFO of the specified sensor and reports a flush complete event when this is done.
  • poll - Returns available sensor events.

The implementation must be thread safe and allow these functions to be called from different threads.

The interface also defines several types used by those functions. The main types are:

  • sensors_module_t
  • sensors_poll_device_t
  • sensor_t
  • sensors_event_t

In addition to the sections below, see sensors.h for more information on those types.


int (*get_sensors_list)(struct sensors_module_t* module, struct sensor_t
  const** list);

Provides the list of sensors implemented by the HAL. See sensor_t for details on how the sensors are defined.

The order in which the sensors appear in the list is the order in which the sensors will be reported to the applications. Usually, the base sensors appear first, followed by the composite sensors.

If several sensors share the same sensor type and wake-up property, the first one in the list is called the “default” sensor. It is the one returned by getDefaultSensor(int sensorType, bool wakeUp).

This function returns the number of sensors in the list.

activate(sensor, true/false)

int (*activate)(struct sensors_poll_device_t *dev, int sensor_handle, int

Activates or deactivates a sensor.

sensor_handle is the handle of the sensor to activate/deactivate. A sensor’s handle is defined by the handle field of its sensor_t structure.

enabled is set to 1 to enable or 0 to disable the sensor.

One-shot sensors deactivate themselves automatically upon receiving an event, and they must still accept to be deactivated through a call to activate(..., enabled=0).

Non-wake-up sensors never prevent the SoC from going into suspend mode; that is, the HAL shall not hold a partial wake-lock on behalf of applications.

Wake-up sensors, when delivering events continuously, can prevent the SoC from going into suspend mode, but if no event needs to be delivered, the partial wake-lock must be released.

If enabled is 1 and the sensor is already activated, this function is a no-op and succeeds.

If enabled is 0 and the sensor is already deactivated, this function is a no-op and succeeds.

This function returns 0 on success and a negative error number otherwise.

batch(sensor, flags, sampling period, maximum report latency)

int (*batch)(
     struct sensors_poll_device_1* dev,
     int sensor_handle,
     int flags,
     int64_t sampling_period_ns,
     int64_t max_report_latency_ns);

Sets a sensor’s parameters, including sampling frequency and maximum report latency. This function can be called while the sensor is activated, in which case it must not cause any sensor measurements to be lost: Transitioning from one sampling rate to the other cannot cause lost events, nor can transitioning from a high maximum report latency to a low maximum report latency.

sensor_handle is the handle of the sensor to configure.

flags is currently unused.

sampling_period_ns is the sampling period at which the sensor should run, in nanoseconds. See sampling_period_ns for more details.

max_report_latency_ns is the maximum time by which events can be delayed before being reported through the HAL, in nanoseconds. See the max_report_latency_ns paragraph for more details.

This function returns 0 on success and a negative error number otherwise.


What the sampling_period_ns parameter means depends on the specified sensor's reporting mode:

  • Continuous: sampling_period_ns is the sampling rate, meaning the rate at which events are generated.
  • On-change: sampling_period_ns limits the sampling rate of events, meaning events are generated no faster than every sampling_period_ns nanoseconds. There might be periods longer than sampling_period_ns where no event is generated if the measured values do not change for long periods. See on-change reporting mode for more details.
  • One-shot: sampling_period_ns is ignored. It has no effect.
  • Special: See the specific sensor type descriptions for details on how sampling_period_ns is used for special sensors.

See Reporting modes for more information about the impact of sampling_period_ns in the different modes.

For continuous and on-change sensors,

  • if sampling_period_ns is less than sensor_t.minDelay, then the HAL implementation must silently clamp it to max(sensor_t.minDelay, 1ms). Android does not support the generation of events at more than 1000Hz.
  • if sampling_period_ns is greater than sensor_t.maxDelay, then the HAL implementation must silently truncate it to sensor_t.maxDelay.

Physical sensors sometimes have limitations on the rates at which they can run and the accuracy of their clocks. To account for this, we allow the actual sampling frequency to differ from the requested frequency, as long as it satisfies the requirements in the table below.

If the requested frequency is

Then the actual frequency must be

below min frequency (<1/maxDelay)

between 90% and 110% of the min frequency

between min and max frequency

between 90% and 220% of the requested frequency

above max frequency (>1/minDelay)

between 90% and 110% of the max frequency

and below 1100Hz

Note that this contract is valid only at the HAL level, where there is always a single client. At the SDK level, applications might get different rates, due to the multiplexing happening in the Framework. See Framework for more details.


max_report_latency_ns sets the maximum time in nanoseconds, by which events can be delayed and stored in the hardware FIFO before being reported through the HAL while the SoC is awake.

A value of zero signifies that the events must be reported as soon as they are measured, either skipping the FIFO altogether, or emptying the FIFO as soon as one event from this sensor is present in it.

For example, an accelerometer activated at 50Hz with max_report_latency_ns=0 will trigger interrupts 50 times per second when the SoC is awake.

When max_report_latency_ns>0, sensor events do not need to be reported as soon as they are detected. They can be temporarily stored in the hardware FIFO and reported in batches, as long as no event is delayed by more than max_report_latency_ns nanoseconds. That is, all events since the previous batch are recorded and returned at once. This reduces the amount of interrupts sent to the SoC and allows the SoC to switch to a lower power mode (idle) while the sensor is capturing and batching data.

Each event has a timestamp associated with it. Delaying the time at which an event is reported does not impact the event timestamp. The timestamp must be accurate and correspond to the time at which the event physically happened, not the time it is being reported.

Allowing sensor events to be stored temporarily in the hardware FIFO does not modify the behavior of poll: events from different sensors can be interleaved, and as usual, all events from the same sensor are time-ordered.

See Batching for more details on sensor batching, including behaviors in suspend mode and out of suspend mode.

setDelay(sensor, sampling period)

int (*setDelay)(
     struct sensors_poll_device_t *dev,
     int sensor_handle,
     int64_t sampling_period_ns);

After HAL version 1.0, this function is deprecated and is never called. Instead, the batch function is called to set the sampling_period_ns parameter.

In HAL version 1.0, setDelay was used instead of batch to set sampling_period_ns.


int (*flush)(struct sensors_poll_device_1* dev, int sensor_handle);

Add a flush complete event to the end of the hardware FIFO for the specified sensor and flushes the FIFO; those events are delivered as usual (i.e.: as if the maximum reporting latency had expired) and removed from the FIFO.

The flush happens asynchronously (i.e.: this function must return immediately). If the implementation uses a single FIFO for several sensors, that FIFO is flushed and the flush complete event is added only for the specified sensor.

If the specified sensor has no FIFO (no buffering possible), or if the FIFO, was empty at the time of the call, flush must still succeed and send a flush complete event for that sensor. This applies to all sensors other than one-shot sensors.

When flush is called, even if a flush event is already in the FIFO for that sensor, an additional one must be created and added to the end of the FIFO, and the FIFO must be flushed. The number of flush calls must be equal to the number of flush complete events created.

flush does not apply to one-shot sensors; if sensor_handle refers to a one-shot sensor, flush must return -EINVAL and not generate any flush complete metadata event.

This function returns 0 on success, -EINVAL if the specified sensor is a one-shot sensor or wasn’t enabled, and a negative error number otherwise.


int (*poll)(struct sensors_poll_device_t *dev, sensors_event_t* data, int

Returns an array of sensor data by filling the data argument. This function must block until events are available. It will return the number of events read on success, or a negative error number in case of an error.

The number of events returned in data must be less or equal to the count argument. This function shall never return 0 (no event).

Sequence of calls

When the device boots, get_sensors_list is called.

When a sensor gets activated, the batch function will be called with the requested parameters, followed by activate(..., enable=1).

Note that in HAL version 1_0, the order was the opposite: activate was called first, followed by set_delay.

When the requested characteristics of a sensor are changing while it is activated, the batch function is called.

flush can be called at any time, even on non-activated sensors (in which case it must return -EINVAL)

When a sensor gets deactivated, activate(..., enable=0) will be called.

In parallel to those calls, the poll function will be called repeatedly to request data. poll can be called even when no sensors are activated.


sensors_module_t is the type used to create the Android hardware module for the sensors. The implementation of the HAL must define an object HAL_MODULE_INFO_SYM of this type to expose the get_sensors_list function. See the definition of sensors_module_t in sensors.h and the definition of hw_module_t for more information.

sensors_poll_device_t / sensors_poll_device_1_t

sensors_poll_device_1_t contains the rest of the methods defined above: activate, batch, flush and poll. Its common field (of type hw_device_t) defines the version number of the HAL.


sensor_t represents an Android sensor. Here are some of its important fields:

name: A user-visible string that represents the sensor. This string often contains the part name of the underlying sensor, the type of the sensor, and whether it is a wake-up sensor. For example, “LIS2HH12 Accelerometer”, “MAX21000 Uncalibrated Gyroscope”, “BMP280 Wake-up Barometer”, “MPU6515 Game Rotation Vector”

handle: The integer used to refer to the sensor when registering to it or generating events from it.

type: The type of the sensor. See the explanation of sensor type in What are Android sensors? for more details, and see Sensor types for official sensor types. For non-official sensor types, type must start with SENSOR_TYPE_DEVICE_PRIVATE_BASE

stringType: The type of the sensor as a string. When the sensor has an official type, set to SENSOR_STRING_TYPE_*. When the sensor has a manufacturer specific type, stringType must start with the manufacturer reverse domain name. For example, a sensor (say a unicorn detector) defined by the Cool-product team at Fictional-Company could use stringType=”com.fictional_company.cool_product.unicorn_detector”. The stringType is used to uniquely identify non-official sensors types. See sensors.h for more information on types and string types.

requiredPermission: A string representing the permission that applications must possess to see the sensor, register to it and receive its data. An empty string means applications do not require any permission to access this sensor. Some sensor types like the heart rate monitor have a mandatory requiredPermission. All sensors providing sensitive user information (such as the heart rate) must be protected by a permission.

flags: Flags for this sensor, defining the sensor’s reporting mode and whether the sensor is a wake-up sensor or not. For example, a one-shot wake-up sensor will have flags = SENSOR_FLAG_ONE_SHOT_MODE | SENSOR_FLAG_WAKE_UP. The bits of the flag that are not used in the current HAL version must be left equal to 0.

maxRange: The maximum value the sensor can report, in the same unit as the reported values. The sensor must be able to report values without saturating within [-maxRange; maxRange]. Note that this means the total range of the sensor in the generic sense is 2*maxRange. When the sensor reports values over several axes, the range applies to each axis. For example, a “+/- 2g” accelerometer will report maxRange = 2*9.81 = 2g.

resolution: The smallest difference in value that the sensor can measure. Usually computed based on maxRange and the number of bits in the measurement.

power: The power cost of enabling the sensor, in milliAmps. This is nearly always more that the power consumption reported in the datasheet of the underlying sensor. See Base sensors != physical sensors for more details and see Power measurement process for details on how to measure the power consumption of a sensor. If the sensor’s power consumption depends on whether the device is moving, the power consumption while moving is the one reported in the power field.

minDelay: For continuous sensors, the sampling period, in microseconds, corresponding to the fastest rate the sensor supports. See sampling_period_ns for details on how this value is used. Beware that minDelay is expressed in microseconds while sampling_period_ns is in nanoseconds. For on-change and special reporting mode sensors, unless otherwise specified, minDelay must be 0. For one-shot sensors, it must be -1.

maxDelay: For continuous and on-change sensors, the sampling period, in microseconds, corresponding to the slowest rate the sensor supports. See sampling_period_ns for details on how this value is used. Beware that maxDelay is expressed in microseconds while sampling_period_ns is in nanoseconds. For special and one-shot sensors, maxDelay must be 0.

fifoReservedEventCount: The number of events reserved for this sensor in the hardware FIFO. If there is a dedicated FIFO for this sensor, then fifoReservedEventCount is the size of this dedicated FIFO. If the FIFO is shared with other sensors, fifoReservedEventCount is the size of the part of the FIFO that is reserved for that sensor. On most shared-FIFO systems, and on systems that do not have a hardware FIFO this value is 0.

fifoMaxEventCount: The maximum number of events that could be stored in the FIFOs for this sensor. This is always greater or equal to fifoReservedEventCount. This value is used to estimate how quickly the FIFO will get full when registering to the sensor at a specific rate, supposing no other sensors are activated. On systems that do not have a hardware FIFO, fifoMaxEventCount is 0. See Batching for more details.

For sensors with an official sensor type, some of the fields are overwritten by the framework. For example, accelerometer sensors are forced to have a continuous reporting mode, and heart rate monitors are forced to be protected by the SENSOR_PERMISSION_BODY_SENSORS permission.


Sensor events generated by Android sensors and reported through the poll function are of type sensors_event_t. Here are some important fields of sensors_event_t:

version: Must be sizeof(struct sensors_event_t)

sensor: The handle of the sensor that generated the event, as defined by sensor_t.handle.

type: The sensor type of the sensor that generated the event, as defined by sensor_t.type.

timestamp: The timestamp of the event in nanoseconds. This is the time the event happened (a step was taken, or an accelerometer measurement was made), not the time the event was reported. timestamp must be synchronized with the elapsedRealtimeNano clock, and in the case of continuous sensors, the jitter must be small. Timestamp filtering is sometimes necessary to satisfy the CDD requirements, as using only the SoC interrupt time to set the timestamps causes too high jitter, and using only the sensor chip time to set the timestamps can cause de-synchronization from the elapsedRealtimeNano clock, as the sensor clock drifts.

data and overlapping fields: The values measured by the sensor. The meaning and units of those fields are specific to each sensor type. See sensors.h and the definition of the different Sensor types for a description of the data fields. For some sensors, the accuracy of the readings is also reported as part of the data, through a status field. This field is only piped through for those select sensor types, appearing at the SDK layer as an accuracy value. For those sensors, the fact that the status field must be set is mentioned in their sensor type definition.

Metadata flush complete events

Metadata events have the same type as normal sensor events: sensors_event_meta_data_t = sensors_event_t. They are returned together with other sensor events through poll. They possess the following fields:

version: Must be META_DATA_VERSION


sensor, reserved, and timestamp: Must be 0

meta_data.what: Contains the metadata type for this event. There is currently a single valid metadata type: META_DATA_FLUSH_COMPLETE.

META_DATA_FLUSH_COMPLETE events represent the completion of the flush of a sensor FIFO. When meta_data.what=META_DATA_FLUSH_COMPLETE, meta_data.sensor must be set to the handle of the sensor that has been flushed. They are generated when and only when flush is called on a sensor. See the section on the flush function for more information.