BTstack implements a set of Bluetooth protocols and profiles. To connect to other Bluetooth devices or to provide a Bluetooth services, BTstack has to be properly configured.

The configuration of BTstack is done both at compile time as well as at run time:

  • compile time configuration:

    • adjust btstack_config.h - this file describes the system configuration, used functionality, and also the memory configuration
    • add necessary source code files to your project
  • run time configuration of:

    • Bluetooth chipset
    • run loop
    • HCI transport layer
    • provided services
    • packet handlers

In the following, we provide an overview of the configuration that is necessary to setup BTstack. From the point when the run loop is executed, the application runs as a finite state machine, which processes events received from BTstack. BTstack groups events logically and provides them via packet handlers. We provide their overview here. For the case that there is a need to inspect the data exchanged between BTstack and the Bluetooth chipset, we describe how to configure packet logging mechanism. Finally, we provide an overview on power management in Bluetooth in general and how to save energy in BTstack.

Configuration in btstack_config.h

The file btstack_config.h contains three parts:

  • #define HAVE_* directives listed here. These directives describe available system properties, similar to config.h in a autoconf setup.
  • #define ENABLE_* directives listed here. These directives list enabled properties, most importantly ENABLE_CLASSIC and ENABLE_BLE.
  • other #define directives for BTstack configuration, most notably static memory, see next section and NVM configuration.

HAVE_* directives

System properties:

#define Description
HAVE_MALLOC Use dynamic memory
HAVE_AES128 Use platform AES128 engine - not needed usually
HAVE_BTSTACK_STDIN STDIN is available for CLI interface
HAVE_MBEDTLS_ECC_P256 mbedTLS provides NIST P-256 operations e.g. for LE Secure Connections

Embedded platform properties:

#define Description
HAVE_EMBEDDED_TIME_MS System provides time in milliseconds
HAVE_EMBEDDED_TICK System provides tick interrupt

POSIX platform properties:

#define Description
HAVE_POSIX_B300_MAPPED_TO_2000000 Workaround to use serial port with 2 mbps
HAVE_POSIX_B600_MAPPED_TO_3000000 Workaround to use serial port with 3 mpbs
HAVE_POSIX_FILE_IO POSIX File i/o used for hci dump
HAVE_POSIX_TIME System provides time function
LINK_KEY_PATH Path to stored link keys
LE_DEVICE_DB_PATH Path to stored LE device information

ENABLE_* directives

BTstack properties:

#define Description
ENABLE_CLASSIC Enable Classic related code in HCI and L2CAP
ENABLE_BLE Enable BLE related code in HCI and L2CAP
ENABLE_EHCILL Enable eHCILL low power mode on TI CC256x/WL18xx chipsets
ENABLE_LOG_DEBUG Enable log_debug messages
ENABLE_LOG_ERROR Enable log_error messages
ENABLE_LOG_INFO Enable log_info messages
ENABLE_SCO_OVER_HCI Enable SCO over HCI for chipsets (only TI CC256x/WL18xx, CSR + Broadcom H2/USB))
ENABLE_HFP_WIDE_BAND_SPEECH Enable support for mSBC codec used in HFP profile for Wide-Band Speech
ENBALE_LE_PERIPHERAL Enable support for LE Peripheral Role in HCI and Security Manager
ENBALE_LE_CENTRAL Enable support for LE Central Role in HCI and Security Manager
ENABLE_MICRO_ECC_FOR_LE_SECURE_CONNECTIONS Use micro-ecc library for ECC operations
ENABLE_LE_DATA_CHANNELS Enable LE Data Channels in credit-based flow control mode
ENABLE_LE_DATA_LENGTH_EXTENSION Enable LE Data Length Extension support
ENABLE_L2CAP_ENHANCED_RETRANSMISSION_MODE Enable L2CAP Enhanced Retransmission Mode. Mandatory for AVRCP Browsing
ENABLE_HCI_CONTROLLER_TO_HOST_FLOW_CONTROL Enable HCI Controller to Host Flow Control, see below
ENABLE_CC256X_BAUDRATE_CHANGE_FLOWCONTROL_BUG_WORKAROUND Enable workaround for bug in CC256x Flow Control during baud rate change, see chipset docs.

Notes: - ENABLE_MICRO_ECC_FOR_LE_SECURE_CONNECTIONS: Only some Bluetooth 4.2+ controllers (e.g., EM9304, ESP32) support the necessary HCI commands. Others reasons to enable the ECC software implementations are if the Host is much faster or if the micro-ecc library is already provided (e.g., ESP32, WICED)

HCI Controller to Host Flow Control

In general, BTstack relies on flow control of the HCI transport, either via Hardware CTS/RTS flow control for UART or regular USB flow control. If this is not possible, e.g on an SoC, BTstack can use HCI Controller to Host Flow Control by defining ENABLE_HCI_CONTROLLER_TO_HOST_FLOW_CONTROL. If enabled, the HCI Transport implementation must be able to buffer the specified packets. In addition, it also need to be able to buffer a few HCI Events. Using a low number of host buffers might result in less throughput.

Host buffer configuration for HCI Controller to Host Flow Control:

#define Description
HCI_HOST_ACL_PACKET_NUM Max number of ACL packets
HCI_HOST_ACL_PACKET_LEN Max size of HCI Host ACL packets
HCI_HOST_SCO_PACKET_NUM Max number of ACL packets
HCI_HOST_SCO_PACKET_LEN Max size of HCI Host SCO packets

Memory configuration directives

The structs for services, active connections and remote devices can be allocated in two different manners:

  • statically from an individual memory pool, whose maximal number of elements is defined in the btstack_config.h file. To initialize the static pools, you need to call at runtime btstack_memory_init function. An example of memory configuration for a single SPP service with a minimal L2CAP MTU is shown in Listing {@lst:memoryConfigurationSPP}.

  • dynamically using the malloc/free functions, if HAVE_MALLOC is defined in btstack_config.h file.

For each HCI connection, a buffer of size HCI_ACL_PAYLOAD_SIZE is reserved. For fast data transfer, however, a large ACL buffer of 1021 bytes is recommend. The large ACL buffer is required for 3-DH5 packets to be used.

#define Description
HCI_ACL_PAYLOAD_SIZE Max size of HCI ACL payloads
MAX_NR_BNEP_CHANNELS Max number of BNEP channels
MAX_NR_BNEP_SERVICES Max number of BNEP services
MAX_NR_BTSTACK_LINK_KEY_DB_MEMORY_ENTRIES Max number of link key entries cached in RAM
MAX_NR_GATT_CLIENTS Max number of GATT clients
MAX_NR_HCI_CONNECTIONS Max number of HCI connections
MAX_NR_HFP_CONNECTIONS Max number of HFP connections
MAX_NR_L2CAP_CHANNELS Max number of L2CAP connections
MAX_NR_L2CAP_SERVICES Max number of L2CAP services
MAX_NR_RFCOMM_CHANNELS Max number of RFOMMM connections
MAX_NR_RFCOMM_MULTIPLEXERS Max number of RFCOMM multiplexers, with one multiplexer per HCI connection
MAX_NR_RFCOMM_SERVICES Max number of RFCOMM services
MAX_NR_SERVICE_RECORD_ITEMS Max number of SDP service records
MAX_NR_SM_LOOKUP_ENTRIES Max number of items in Security Manager lookup queue
MAX_NR_WHITELIST_ENTRIES Max number of items in GAP LE Whitelist to connect to
MAX_NR_LE_DEVICE_DB_ENTRIES Max number of items in LE Device DB

The memory is set up by calling btstack_memory_init function:


Here's the memory configuration for a basic SPP server.


Listing: Memory configuration for a basic SPP server. {#lst:memoryConfigurationSPP}

In this example, the size of ACL packets is limited to the minimum of 52 bytes, resulting in an L2CAP MTU of 48 bytes. Only a singleHCI connection can be established at any time. On it, two L2CAP services are provided, which can be active at the same time. Here, these two can be RFCOMM and SDP. Then, memory for one RFCOMM multiplexer is reserved over which one connection can be active. Finally, up to three link keys can be cached in RAM.

Non-volatile memory (NVM) directives

If implemented, bonding information is stored in Non-volatile memory. For Classic, a single link keys and its type is stored. For LE, the bonding information contains various values (long term key, random number, EDIV, signing counter, identity, ...)Often, this implemented using Flash memory. Then, the number of stored entries are limited by:

#define Description
NVM_NUM_LINK_KEYS Max number of Classic Link Keys that can be stored
NVM_NUM_DEVICE_DB_ENTRIES Max number of LE Device DB entries that can be stored

Source tree structure

The source tree has been organized to easily setup new projects.

Path Description
chipset Support for individual Bluetooth chipsets
doc Sources for BTstack documentation
example Example applications available for all ports
platform Support for special OSs and/or MCU architectures
port Complete port for a MCU + Chipset combinations
src Bluetooth stack implementation
test Unit and PTS tests
tool Helper tools for BTstack

The core of BTstack, including all protocol and profiles, is in src/.

Support for a particular platform is provided by the platform/ subfolder. For most embedded ports, platform/embedded/ provides btstack_run_loop_embedded and the hci_transport_h4_embedded implementation that require hal_cpu.h, hal_led.h, and hal_uart_dma.h plus hal_tick.h or hal_time_ms to be implemented by the user.

To accommodate a particular Bluetooth chipset, the chipset/ subfolders provide various btstack_chipset_ implementations. Please have a look at the existing ports in port/*.

Run loop configuration

To initialize BTstack you need to initialize the memory and the run loop respectively, then setup HCI and all needed higher level protocols.

BTstack uses the concept of a run loop to handle incoming data and to schedule work. The run loop handles events from two different types of sources: data sources and timers. Data sources represent communication interfaces like an UART or an USB driver. Timers are used by BTstack to implement various Bluetooth-related timeouts. They can also be used to handle periodic events.

Data sources and timers are represented by the btstack_data_source_t and btstack_timer_source_t structs respectively. Each of these structs contain at least a linked list node and a pointer to a callback function. All active timers and data sources are kept in link lists. While the list of data sources is unsorted, the timers are sorted by expiration timeout for efficient processing.

Timers are single shot: a timer will be removed from the timer list before its event handler callback is executed. If you need a periodic timer, you can re-register the same timer source in the callback function, as shown in Listing [PeriodicTimerHandler]. Note that BTstack expects to get called periodically to keep its time, see Section on time abstraction for more on the tick hardware abstraction.

BTstack provides different run loop implementations that implement the btstack_run_loop_t interface:

  • Embedded: the main implementation for embedded systems, especially without an RTOS.
  • POSIX: implementation for POSIX systems based on the select() call.
  • CoreFoundation: implementation for iOS and OS X applications
  • WICED: implementation for the Broadcom WICED SDK RTOS abstraction that wraps FreeRTOS or ThreadX.
  • Windows: implementation for Windows based on Event objects and WaitForMultipleObjects() call.

Depending on the platform, data sources are either polled (embedded), or the platform provides a way to wait for a data source to become ready for read or write (POSIX, CoreFoundation, Windows), or, are not used as the HCI transport driver and the run loop is implemented in a different way (WICED). In any case, the callbacks must be to explicitly enabled with the btstack_run_loop_enable_data_source_callbacks(..) function.

In your code, you'll have to configure the run loop before you start it as shown in Listing [listing:btstackInit]. The application can register data sources as well as timers, e.g., for periodical sampling of sensors, or for communication over the UART.

The run loop is set up by calling btstack_run_loop_init function and providing an instance of the actual run loop. E.g. for the embedded platform, it is:


The complete Run loop API is provided here.

Run loop embedded

In the embedded run loop implementation, data sources are constantly polled and the system is put to sleep if no IRQ happens during the poll of all data sources.

The complete run loop cycle looks like this: first, the callback function of all registered data sources are called in a round robin way. Then, the callback functions of timers that are ready are executed. Finally, it will be checked if another run loop iteration has been requested by an interrupt handler. If not, the run loop will put the MCU into sleep mode.

Incoming data over the UART, USB, or timer ticks will generate an interrupt and wake up the microcontroller. In order to avoid the situation where a data source becomes ready just before the run loop enters sleep mode, an interrupt-driven data source has to call the btstack_run_loop_embedded_trigger function. The call to btstack_run_loop_embedded_trigger sets an internal flag that is checked in the critical section just before entering sleep mode causing another run loop cycle.

To enable the use of timers, make sure that you defined HAVE_EMBEDDED_TICK or HAVE_EMBEDDED_TIME_MS in the config file.

Run loop POSIX

The data sources are standard File Descriptors. In the run loop execute implementation, select() call is used to wait for file descriptors to become ready to read or write, while waiting for the next timeout.

To enable the use of timers, make sure that you defined HAVE_POSIX_TIME in the config file.

Run loop CoreFoundation (OS X/iOS)

This run loop directly maps BTstack's data source and timer source with CoreFoundation objects. It supports ready to read and write similar to the POSIX implementation. The call to btstack_run_loop_execute() then just calls CFRunLoopRun().

To enable the use of timers, make sure that you defined HAVE_POSIX_TIME in the config file.

Run loop Windows

The data sources are Event objects. In the run loop implementation WaitForMultipleObjects() call is all is used to wait for the Event object to become ready while waiting for the next timeout.

Run loop WICED

WICED SDK API does not provide asynchronous read and write to the UART and no direct way to wait for one or more peripherals to become ready. Therefore, BTstack does not provide direct support for data sources. Instead, the run loop provides a message queue that allows to schedule functions calls on its thread via btstack_run_loop_wiced_execute_code_on_main_thread().

The HCI transport H4 implementation then uses two lightweight threads to do the blocking read and write operations. When a read or write is complete on the helper threads, a callback to BTstack is scheduled.

HCI Transport configuration

The HCI initialization has to adapt BTstack to the used platform. The first call is to hci_init() and requires information about the HCI Transport to use. The arguments are:

  • HCI Transport implementation: On embedded systems, a Bluetooth module can be connected via USB or an UART port. On embedded, BTstack implements HCI UART Transport Layer (H4) and H4 with eHCILL support, a lightweight low-power variant by Texas Instruments. For POSIX, there is an implementation for HCI H4, HCI H5 and H2 libUSB, and for WICED HCI H4 WICED. These are accessed by linking the appropriate file, e.g., platform/embedded/hci_transport_h4_embedded.c and then getting a pointer to HCI Transport implementation. For more information on adapting HCI Transport to different environments, see here.
hci_transport_t * transport = hci_transport_h4_instance();
  • HCI Transport configuration: As the configuration of the UART used in the H4 transport interface are not standardized, it has to be provided by the main application to BTstack. In addition to the initial UART baud rate, the main baud rate can be specified. The HCI layer of BTstack will change the init baud rate to the main one after the basic setup of the Bluetooth module. A baud rate change has to be done in a coordinated way at both HCI and hardware level. For example, on the CC256x, the HCI command to change the baud rate is sent first, then it is necessary to wait for the confirmation event from the Bluetooth module. Only now, can the UART baud rate changed.
hci_uart_config_t* config = &hci_uart_config;

After these are ready, HCI is initialized like this:

hci_init(transport, config);

In addition to these, most UART-based Bluetooth chipset require some special logic for correct initialization that is not covered by the Bluetooth specification. In particular, this covers:

  • setting the baudrate
  • setting the BD ADDR for devices without an internal persistent storage
  • upload of some firmware patches.

This is provided by the various btstack_chipset_t implementation in the chipset/ subfolders. As an example, the bstack_chipset_cc256x_instance function returns a pointer to a chipset struct suitable for the CC256x chipset.

btstack_chipset_t * chipset = btstack_chipset_cc256x_instance();

In some setups, the hardware setup provides explicit control of Bluetooth power and sleep modes. In this case, a btstack_control_t struct can be set with hci_set_control.

Finally, the HCI implementation requires some form of persistent storage for link keys generated during either legacy pairing or the Secure Simple Pairing (SSP). This commonly requires platform specific code to access the MCU’s EEPROM of Flash storage. For the first steps, BTstack provides a (non) persistent store in memory. For more see here.

btstack_link_key_db_t * link_key_db = &btstack_link_key_db_memory_instance();

The higher layers only rely on BTstack and are initialized by calling the respective *_init function. These init functions register themselves with the underlying layer. In addition, the application can register packet handlers to get events and data as explained in the following section.


One important construct of BTstack is service. A service represents a server side component that handles incoming connections. So far, BTstack provides L2CAP, BNEP, and RFCOMM services. An L2CAP service handles incoming connections for an L2CAP channel and is registered with its protocol service multiplexer ID (PSM). Similarly, an RFCOMM service handles incoming RFCOMM connections and is registered with the RFCOMM channel ID. Outgoing connections require no special registration, they are created by the application when needed.

Packet handlers configuration

After the hardware and BTstack are set up, the run loop is entered. From now on everything is event driven. The application calls BTstack functions, which in turn may send commands to the Bluetooth module. The resulting events are delivered back to the application. Instead of writing a single callback handler for each possible event (as it is done in some other Bluetooth stacks), BTstack groups events logically and provides them over a single generic interface. Appendix Events and Errors summarizes the parameters and event codes of L2CAP and RFCOMM events, as well as possible errors and the corresponding error codes.

Here is summarized list of packet handlers that an application might use:

  • HCI event handler - allows to observer HCI, GAP, and general BTstack events.

  • L2CAP packet handler - handles LE Connection parameter requeset updates

  • L2CAP service packet handler - handles incoming L2CAP connections, i.e., channels initiated by the remote.

  • L2CAP channel packet handler - handles outgoing L2CAP connections, i.e., channels initiated internally.

  • RFCOMM service packet handler - handles incoming RFCOMM connections, i.e., channels initiated by the remote.

  • RFCOMM channel packet handler - handles outgoing RFCOMM connections, i.e., channels initiated internally.

These handlers are registered with the functions listed in Table below.

Packet Handler Registering Function
HCI packet handler hci_add_event_handler
L2CAP packet handler l2cap_register_packet_handler
L2CAP service packet handler l2cap_register_service
L2CAP channel packet handler l2cap_create_channel
RFCOMM service packet handler rfcomm_register_service and rfcomm_register_service_with_initial_credits
RFCOMM channel packet handler rfcomm_create_channel and rfcomm_create_channel_with_initial_credits

Table: Functions for registering packet handlers.

HCI, GAP, and general BTstack events are delivered to the packet handler specified by hci_add_event_handler function. In L2CAP, BTstack discriminates incoming and outgoing connections, i.e., event and data packets are delivered to different packet handlers. Outgoing connections are used access remote services, incoming connections are used to provide services. For incoming connections, the packet handler specified by l2cap_register_service is used. For outgoing connections, the handler provided by l2cap_create_channel is used. RFCOMM and BNEP are similar.

The application can register a single shared packet handler for all protocols and services, or use separate packet handlers for each protocol layer and service. A shared packet handler is often used for stack initialization and connection management.

Separate packet handlers can be used for each L2CAP service and outgoing connection. For example, to connect with a Bluetooth HID keyboard, your application could use three packet handlers: one to handle HCI events during discovery of a keyboard registered by l2cap_register_packet_handler; one that will be registered to an outgoing L2CAP channel to connect to keyboard and to receive keyboard data registered by l2cap_create_channel; after that keyboard can reconnect by itself. For this, you need to register L2CAP services for the HID Control and HID Interrupt PSMs using l2cap_register_service. In this call, you’ll also specify a packet handler to accept and receive keyboard data.

All events names have the form MODULE_EVENT_NAME now, e.g., gap_event_advertising_report. To facilitate working with events and get rid of manually calculating offsets into packets, BTstack provides auto-generated getters for all fields of all events in src/hci_event.h. All functions are defined as static inline, so they are not wasting any program memory if not used. If used, the memory footprint should be identical to accessing the field directly via offsets into the packet. For example, to access fields address_type and address from the gap_event_advertising_report event use following getters:

uint8_t address type = gap_event_advertising_report_get_address_type(event);
bd_addr_t address;
gap_event_advertising_report_get_address(event, address);

Bluetooth HCI Packet Logs

If things don't work as expected, having a look at the data exchanged between BTstack and the Bluetooth chipset often helps.

For this, BTstack provides a configurable packet logging mechanism via hci_dump.h:

void hci_dump_open(const char *filename, hci_dump_format_t format);

On POSIX systems, you can call hci_dump_open with a path and HCI_DUMP_BLUEZ or HCI_DUMP_PACKETLOGGER in the setup, i.e., before entering the run loop. The resulting file can be analyzed with Wireshark or the Apple's PacketLogger tool.

On embedded systems without a file system, you still can call hci_dump_open(NULL, HCI_DUMP_STDOUT). It will log all HCI packets to the console via printf. If you capture the console output, incl. your own debug messages, you can use the tool in the tools folder to convert a text output into a PacketLogger file.

In addition to the HCI packets, you can also enable BTstack's debug information by adding


to the btstack_config.h and recompiling your application.

Bluetooth Power Control

In most BTstack examples, the device is set to be discoverable and connectable. In this mode, even when there's no active connection, the Bluetooth Controller will periodically activate its receiver in order to listen for inquiries or connecting requests from another device. The ability to be discoverable requires more energy than the ability to be connected. Being discoverable also announces the device to anybody in the area. Therefore, it is a good idea to pause listening for inquiries when not needed. Other devices that have your Bluetooth address can still connect to your device.

To enable/disable discoverability, you can call:

 * @brief Allows to control if device is discoverable. OFF by default.
void gap_discoverable_control(uint8_t enable);

If you don't need to become connected from other devices for a longer period of time, you can also disable the listening to connection requests.

To enable/disable connectability, you can call:

 * @brief Override page scan mode. Page scan mode enabled by l2cap when services are registered
 * @note Might be used to reduce power consumption while Bluetooth module stays powered but no (new)
 *       connections are expected
void gap_connectable_control(uint8_t enable);

For Bluetooth Low Energy, the radio is periodically used to broadcast advertisements that are used for both discovery and connection establishment.

To enable/disable advertisements, you can call:

 * @brief Enable/Disable Advertisements. OFF by default.
 * @param enabled
void gap_advertisements_enable(int enabled);

If a Bluetooth Controller is neither discoverable nor connectable, it does not need to periodically turn on its radio and it only needs to respond to commands from the Host. In this case, the Bluetooth Controller is free to enter some kind of deep sleep where the power consumption is minimal.

Finally, if that's not sufficient for your application, you could request BTstack to shutdown the Bluetooth Controller. For this, the "on" and "off" functions in the btstack_control_t struct must be implemented. To shutdown the Bluetooth Controller, you can call:

 * @brief Requests the change of BTstack power mode.
int  hci_power_control(HCI_POWER_MODE mode);

with mode set to HCI_POWER_OFF. When needed later, Bluetooth can be started again via by calling it with mode HCI_POWER_ON, as seen in all examples.