Apple's HomeKit protocol supports both IP and BLE devices. While there appear to be a few open source implementations of IP stacks around (notably https://github.com/KhaosT/HAP-NodeJS), I couldn't find any BLE stacks. So, here's one for you to play with.

This stack runs on the Nordic BLE nRF51 series of chips. The projects compiles using the "standard" Eclipse/GCC toolchain setup.

The code provides all the services required to pair iOS with a BLE device and to operate that device once paired. It runs on the Nordic nRF51 PCA10028 development board.

Here are some perliminary timings. Note that most of the crypto code is in C but with some assembly used to speed the 256-bit integer multiplies. Also note that these timings do not include the time to send and receive payloads. But, anyway, numbers:

Pairing is dominated by the SRP algorithm which is very slow and expensive. Fortunately this only happens once when the iOS device is being associated with the HomeKit device:

Time: 40 seconds

Verify happens everytime an iOS device reconnected to the HomeKit device. Ideally this should be as fast as possible. I've included the C-only and assembly timings here - the difference is dramatic:

Time (C code): 4 seconds

Time (with assembly multiply): 1.2 seconds

Recently Nordic released their next generation BLE platform. I though it might be interesting to see how well it performs Pairing and Verify. As you can see, this chip is much, much faster:

Time: 3.7 seconds

Time (with assembly multiply): 0.26 seconds

The HomeKit code is approximately 28K (compiled with -Os) and uses 1K of RAM. During Pairing an additional 11K of stack RAM is used.

I want to thank a number of projects which made this possible:

1. https://github.com/KhaosT/HAP-NodeJS - which documents the HomeKit protocols for IP and allowed me to guess how they were implemented over Bluetooth.

2. https://tweetnacl.cr.yp.to - which provides the compact eliptical curve implementations, as well as the sha512 hash.

3. https://tls.mbed.org - which provides the core multi-precission math routines used in the SRP implementation.

4. https://munacl.cryptojedi.org/ - which provides the ARM Cortex-M0 optimized Curve25519 implementation and fast multiply routines.

Please note that this software was produced without any reference to any propriatery documentation or information. I am not a MFi licensee, nor do I have access to any related information.