Cortex-R doesn't have MMU, is this has some advantages?

That's a common way of doing things in hard real time systems. A memory management system can put in too much latency for an interrupt because of the table walking required. In general in real time systems they are willing to sacrifice a little performance or flexibility provided they can remove jitter in latency - they want to be sure that everything will actually be done on time..

Very good answer!

laoniu_c - remember it's still possible to make your own 'virtual memory'; it just won't be handled by hardware, but you can still swap out memory blocks to for instance an SD/MMC card (manually). While this sounds as a loss, you still have very fine control of latencies and jitter. A 'swap' operation can be interrupted if necessary, or you can calculate the number of bytes to swap in advance, so that your timing is controlled down to a single clock cycle.