Not at all clear from your presentation how you've implemented any of this forward / tunnelling, or how you might expect this to be debugged / diagnosed.
The Rx and Tx can occur concurrently. If I were in a polling loop, I'd probably implement it with a FIFO / Circular buffer, allowing me to monitor the respective TXE and RXNE of the USARTs I wanted forwarding to each other, and allowing for elasticity where the rates or bursts of data are not identical.
