I am trying to use the UART async API on stm32f7. It works nicely up to about 460.8kbps, but above that I start loosing characters during buffer switch. I immediately provide new buffers when requested, and increasing the buffer size doesn't help. So this appears to happen when new data arrives before my RX handler is done.

Describe the solution you'd like
I currently have my own implementation of this (without UART async API) where I manually configure the STM32 DMA to use circular mode and set up interrupts on Half-transfer, Transfer Complete and UART RX IDLE. It can sustain data rates far above 1Mbps even with very small buffers.

So I would like the STM32 implementation of the UART async API to use either the circular mode or the double-buffer mode of the DMA peripheral (if available). Then DMA hardware would just continue to receive UART characters regardless of how long time I spend in my RX event handler.

Additional context
I'm on Zephyr v2.5.0