Memos From the Cube

Well nuts.

Last week I completed my final goal for my XPLAIN replacement firmware – to merge the XPLAIN USART bridge and PDI programmer into the one single firmware, so that either can be used with the use of a jumper, much like the method used to access the DFU class bootloader. I was hoping that people would be jumping for it, but it seems not the case; not only has the interest been marginal at best, but it turns out I forgot to add a critical file to the SVN repository. That’s right, it would never have compiled for anyone anyway, yet I only find out almost a week later. Hear that? That’s the sound of my nerd heart breaking when I realized that no one cared about the work I did. Nevertheless, perhaps when Atmel begin to use it, it will gain a little more recognition — and the repository is now fixed.

Actually, I did have a big success last night; I managed to connect the free uIP TCP/IP stack up to my RNDIS host class driver, and make a very simple webserver work when connected to my ADSL modem via USB. That’s a little different to my own home-grown TCP/IP stack used in the regular LUFA demos, because uIP is designed to be far more robust (and can handle multiple simultaneous connections). The code is now in the SVN under Projects/Webserver, and I’ll be extending it to include FatFS so that it can serve files off of the dataflash.

I was having some real trouble getting it all working correctly, with some very odd behavior: while about 30% of the time my code would read in packets correctly, process them via uIP and respond, the other 70% of the time it seemed that something was getting severely corrupted (corrupt data being read in from the USB interface, random restarts, etc.). Initially I suspected a uIP misconfiguration causing stack corruption, but eventually I narrowed it down to the way I was handling bidirectional data endpoints in the RNDIS host driver. It seems a patch I made a while back broke things rather than fixed them. While the USB AVRs can’t really support bidirectional endpoints in device mode (it’s possible to do it, but only if the host expects half-duplex communications) host mode is another matter entirely. I was under the impression that a trick similar to device mode was needed to get bidirectional data flowing, but it turns out I can just allocate two separate pipes. Doh – when I switched the pipe token to change the data direction, data from the device being sent would corrupt the USB banks and cause all the random issues.

That means the RNDIS and CDC host drivers in the latest release are badly broken, and will be fixed in the next official release. I hate rolling back patches, but in this case I was wrong to change the code in the first place. For those wanting the code right now, you can either download it as an archive straight from the GIT repository host, or use the GIT/SVN mirrors.

I was planning on connecting my USBKEY up to the general internet to show off its new capabilities, but yet more heartache – something either in my ISP or my equipment is blocking outside connections, despite my fiddling with port forwarding, DMZ and Firewalls. For now, you’ll just have to test the source on your own network equipment.

One more bit of news; while no one seems to care about my XPLAIN work, LUFA in general is finding more new users. Each LUFA release typically gets around 450 downloads total – the latest releast is just about to hit the 1000 mark. Awesome!