On Tue, Oct 06, 2020 at 11:56:16PM -0500, Rob Landley wrote: > We've been implementing a new USB device (starting by adapting the Joris example > CDC-ACM VHDL to a new I/O chip, although we're more interested in EEM), and > attempting to talk to the result makes Linux VERY unhappy. > > On one machine it panics the kernel, on another it makes the USB subsystem fork > off multiple kernel threads stuck eating 100% CPU until a reboot, and also not > notice when we physically unplug the device afterwards. > > I'm aware our device is wrong (haven't figured out HOW it's wrong yet), but... > Linux's USB stack should not be doing that? MacOS instead times something out > and disables the device. (Which again doesn't help us debug it, but at least > doesn't require a reboot afterwards.) > > We finally got a good packet capture, alas from a windows GUI tool we had to > take screenshots of, which totals 6 megabytes so I threw the files in a > temporary directory on my web server. (I apologize in advance for dreamhost: > it's cheap.) > > https://landley.net/isb_usb_weallsb > > Here are two emails about it, the first is an earlier message with the kernel > panic dump and the second is yesterday's packet capture. Does anybody understand > what's going on here? 5.3.0 is pretty old, we have fixed a number of issues like this in newer kernels (or at the very least, the stable kernel trees). So if you could test there, that would be great. Also, do you have the output of 'lsusb -v' of your device? Perhaps your descriptors are incorrect? thanks, greg k-h
