Using HCI connection to Linux kernel about python-bluezero HOT 16 CLOSED

I think in my next experiment I'm going to Pythonise an example and supporting subset of the NodeJS BLE library BluetoothHCISocket by @sandeepmistry

from python-bluezero.

Good work @WayneKeenan!

There are two performance issues that are a concern with DBus

1. continuously scanning for devices. E.g. if you want to constantly scan for nearby beacons. This was discussed on the Bluez mailing list
   http://marc.info/?l=linux-bluetooth&m=148123125327634&w=2
   There is no way to do this with the DBus API currently
2. Throughput with read/write/notify of GATT characteristics. This was the issue raised on the Adafruit BLE library that was linked to on the Bluezero retrospective.
   New DBus API of AcquireNotify and AcquireWrite seems to be trying to address this.
   https://git.kernel.org/pub/scm/bluetooth/bluez.git/tree/doc/gatt-api.txt#n117

Does that help?

from python-bluezero.

added a noddy class in a gist that the test just simply bounces the hci0 interface.

Needs to be run with sudo.

from python-bluezero.

I have implemented the experiment and have LE Scanning working, please see https://github.com/TheBubbleworks/python-hcipy

from python-bluezero.

There is now a Connection and a PhysicalWeb Beacon example.

Scripts can also can run as a non-root user, please check out the README for more info.

I'm keen to understand the performance benefits of the approach, @ukBaz please could you outline the structure of a specific test scenario you had in mind? (it was beacon related I think)

from python-bluezero.

Thanks @kaidokert .

I've started a wiki to track these libraries and make it easier for people to update directly:
https://github.com/ukBaz/python-bluezero/wiki

from python-bluezero.

think the way to go is to use struct.pack and fcntl.ioctl, e.g.

buffer = struct.pack("I", dev_id)
fcntl.ioctl(sock.fileno(), HCIDEVUP, buffer)

My test case gets past the data format but is hitting now hitting complaint from the kernel:

IOError: [Errno 25] Inappropriate ioctl for device

My experiment is here

I think running as root might be required. To circumvent that requirement Noble/Bleno use an external binary which has a ACL applied to the binary, that is something not possible for a 'script', AFAIK.

aversion to running a python bluezero daemon as root?

from python-bluezero.

The reason for the failure is because the HCI commands passed to ioctl need to be converted from the 'bare' number to a Linux ioctl encoding, e.g. http://code.activestate.com/recipes/578225-linux-ioctl-numbers-in-python/

I've had a quick try, doesn't work as-is, needs a bit more massaging, will take a look later.

from python-bluezero.

This is the 'call stack' with links to the functions:

1. Python enable_le_scan() (example from @colin-guyon)
2. bt_hci_send_cmd() (the underlying C source for import bluetooth._bluetooth as bluez found in many Python BLE libs)
3. hci_send_cmd() (BlueZ userland C code)
4. writev() (POSIX/GLIBC)
5. The Magic Happens(tm)

So, all that's needs is to compress the bits in-between 1 and 5 into 1 :)

EDIT: I should caveat the above by saying this was a manual trawl thru the filesystem and online repositories, not from a runtime or IDE introspection, so please beware it maybe inaccurate.

from python-bluezero.

I meant to add that the prototype demonstrates 3 key aspects of using socket io for HCI:

1. sending HCI Commands (e.g. HCIDEVUP)
2. decoding HCI responses (e.g. HCIDEVINFO, EVT_LE_META_EVENT)
3. handling unsolicited HCI events (e.g. EVT_LE_META_EVENT for LE Scanning)

I've also kept in mind the need to do automated testing so the core class has been written with future automated tests in mind, comments in the code.

from python-bluezero.

That's a great demonstration - I spent quite some time banging my head against a brick wall with a prototype similar to Barry's opener, without knowing about the IOCTL numbers (I was getting a "file descriptor in a bad state" type error from fcntl.ioctl). I'm glad Barry asked the question and that there is a purely pythonic way to do HCI!

from python-bluezero.

I have just added the port of the LE advertisement test, https://github.com/TheBubbleworks/python-hcipy/blob/master/le_advertisement_test.py

from python-bluezero.

Just as a comment, here is yet another half finished pile of code that kind of speaks HCI ( written for TI BLE embedded configurations )

https://github.com/mzxu/pyble/tree/master/ble

Assumes a serial port, but i've verified it to somewhat work with plain character device too for basic functions. At least the constant tables might be useful for anyone trying direct HCI interface implementation

from python-bluezero.

Thanks @kaidokert .

We should add this to the list of libraries on #126

Maybe we should start a wiki with all these libraries listed.

from python-bluezero.

Not to flood, but i came across this too https://github.com/Jumperr-labs/hci-protocol . Built on top of https://en.wikipedia.org/wiki/Construct_(python_library) which provides nice self-documenting code.

Jumperr-labs also has python-gatt-server and jumper-ble-logger repos that use this implementation.

from python-bluezero.

Closing this and following the Python HCI sockets work happening elsewhere

from python-bluezero.