Hello.
[Added Erik here to be part of the discussion]
This mostly goes to Koen and Erik. Please coordinate how you are wanting
to work on this. And be pro-active. Waiting for the other one to start
will just lead to starving :-)
I started to hack together some parts as I would do them. Really not
working at all right now and only compile tested. I will give more
details below. I am not sure if I will find time to work more on this
next week before I go on vacation, so here is what I have:
https://git.kernel.org/pub/scm/linux/kernel/git/sschmidt/wpan-next.git/log/?h=wpanusb-original
Feel free to use any of it or ignore it if you have something better.
Really just my take in a really rushed break.
On 05.06.20 13:07, Koen Zandberg wrote:
Hello
On 03-06-2020 20:18, Stefan Schmidt wrote:
Hello.
Happy to see that we finally have the critical mass to get this moved
forward. :-)
Here is my take on what I think needs to be done.
On a first review I found nothing wrong with the design. It needs
further extending and flexibility in my opinion, though.
I would suggest using USB bulk endpoints for both the tx and rx paths.
An interrupt IN endpoint might be useful for events from the radio back
to the host such as ack information from a transmit. This way we can
keep the control messages to configuration only. This is similar to how
USB ethernet devices are using different USB endpoints. I also see
issues with transferring large 802.15.4g frames over control endpoints.
Something similar like CDC-ECM would be my preference here: Split the
frame in multiple bulk transfers and detect the end of the frame by a
transfer size not equal to the endpoint size.
For the above I completely ignore the move to bulk. This still needs doing.
o Add a GET_EXTENDED_ADDR command to receive the extended address
provided by the transceiver itself, or firmware in some way.
+1
My take on this is in the above branch.
o Adding a GET_CAPABILITIES command to query device capabilities
during init to enable and set needed ieee802154_ops based on the
device. Given that we aim to support as many transceivers as possible
we can't rely on static device knowledge to configure wpanusb correctly.
I started to work on this. The real questions is are we passing all the
frequency bands and others settings through the USB spec or are we going
with tables for these inside the firmware and driver and only reference
them? The later is more efficient the first one is more flexible and
likely more extensible for future changes.
Does it make sense to include also a "protocol" version here, to allow
extending the feature set of the driver later without causing
compatibility issues between the firmware and the kernel?
How about we are using the USB version part of the descriptor for this?
We would identify a newer version device not only the spec. The device
handling could work on both.
o Add opcode for set_lbt in USB spec
Added a first take on this.
This requires some clarification for me how the radio should be
configured. Is this just a CSMA/CCA switch?
o Add opcode for set_frame_retries USB spec. (If a transceiver does
not support AutoACK in hardware do Zephyr and RIOT support a software
fallback to handle AutoACK?)
Added a first take on this. Needs USB helper functions to handle the
config data being passed to the device.
This can be implemented in RIOT. I don't think there is something in
place at the moment, most of our radios support this in hardware, but I
see no technical reason why not to support this.
o How are we going to handle transceiver which allow MTUs > 127? Not a
high priority as the kernel part does not support this either right now.
There is some preliminary support for 802.15.4g radios in RIOT. I know
some developers that would prefer not to have to have the MTU limited to
127 bytes :)
This is completely ignored for now.
o Do Zephyr or RIOT expose additional functionality we should support
here?
o Koen, you offered to look into implementing the firmware support for
the USB spec in RIOT. Does the spec fit what RIOT has as abstraction
for ieee802154?
Yes, implementing configuration settings as USB control messages makes
glueing them to the radio abstraction layer very easy. For now RIOT has
configuration for:
- reading and writing channel/page settings
- read/write to addresses, both long and short
- PAN ID
- TX power settings
- reading the max PSDU size
- Ack config settings
- CCA and CSMA configuration, enabling/disabling, retries and backoff
exponent (max/min)
- CCA threshold and mode
Furthermore, it is possible to get frame metadata such as the received
signal strength and the number of retries required for the frame
transmit. All these settings depend a bit on the radio hardware features
of course, but thats what we have the GET_CAPABILITIES for.
If I have time to do more work on this I will update you folks. As I
wrote above. Read over it and feel free to use any of the above or
implement it yourself.
Looking forward to the progress on this when back from my vacation. :-)
regards
Stefan Schmidt
