Managing a simultaneous bi-directional device communication.

[bjgillet]

Hi,

I have a USB device that I can control from host, writing/reading to and from it. No issue with that.
Issue is that at any time, the device by itself may generate events and send information without host being aware of it.
i.e. : at any time, a button pressed on device will trigger an IRQ with data incoming to host without advice.

• How can I check for this from host ?
• Supposing : Host is currently having IRQ OUT/IN communications with device. At same time, device send data through IRQ IN channel. How to dissociate those input flows eventually buffering or pausing one of them ?
• Is pyusb thread safe so some way of priorization and dedication may be set up ?
  I am somewhat lost here.

Obviously, i looked at docs and so, perhaps missed a key point, but could not find an answer.
Thanks in advance for any pointer, hint or help.

[bjgillet]

Hi again,

Yesterday evening I did some basic tests.
I think I have found a possible way to achieve what I am searching for.
Here it is so you don't lost time doing a full answer from scratch.
However, if in it you an obvious issue I could not have seen please let me know.
On my side I will give a status after full coding and testing.

Idea is :
Having a Thread (ie. dev_reader) that will read() on end-point with a timeout=0.
Each frame captured will be written in a thread safe Queue, with another thread (ie. dev_data_manage) that may be started or stopped from either dev_reader or from fact Queue is empty.

By default, dev_reader will be active.
dev_reader will be stopped.

From main thread, when a host to dev exchange will be done : stop dev_reader. Check/wait for dev_data to be stopped (with watchdog timeout). Then do the exchange host->dev. As exchange done, restart dev_reader Thread.Should restart dev_data.

This should avoid any useful traffic loss or data mangling (from few tests done so far).

So, do you see any key points I missed or hided issues I could hit ?

If not, I will let you know about result, so it could eventually help others.

[bjgillet]

As committed I confirm previous did the thing.
Due to limitations in python multi-threading, I had to implement an in-class pause/resume/stop mechanism through Event instances. (discovering python for something else than just basic scripting + total newbie vs. USB... Cool.)
Also, setting read timeout to 0 was not good option. Just catch timeout error and pass, check events, and act accordingly.
Don't forget to flush input streams between two cmd in/out thread switches for coherency when the paused will be waken up.
Then it works pretty well.