Hi, > Hi, again, > > I haven't tested any patch or bisected, but I have another data point. > I built and tested Linux 5.8.18, with the same firmware, and it is > working correctly. I reduced the test case to just rfkilling the > connection, which showed the register dump immediately (before that I > was using the airplane toggle on the keyboard, which isn't working > correctly, it disables and immediately reenables the radio, for some > unfathomable reason). > So, now I'm inclined to believe this is some sort of race condition > between rfkill and pending transactions. Which also means it's not a regression. You can add a dump_stack() in the function that dumps the registers to get a clue.