Created attachment 282539 [details]
btmon logs

When rfkill un-blocking this controller using the Bluetooth settings panel on gnome-control-center, about 1 out of 3 attempts fail. We can see in the kernel logs (with dynamic debug enabled) and in btmon that a HCI reset is getting a timeout.

Interestingly, the problem is only reproducible if there was a LE scanning procedure running before the rfkill switch was blocked. So blocking and unblocking the adapter in a loop like the one bellow does not hit the problem:

  while true ; do echo 1 > /sys/class/rfkill/rfkill2/soft ; sleep 1 ; do echo 0 > /sys/class/rfkill/rfkill2/soft ; sleep 3 ; done

But if I also run another loop in parallel trying to start a LE scan (which fails most of the time because the adapter is OFF, but things align and a scan starts quite a few times), the problem reproduces very easily:

  while true ; do btmgmt find -l ; done