It depends on the specifics of how the bus and (processor) caches are implemented. In a simple case the CPU would stall on load/store instruction or instruction fetch waiting for the bus to become inactive.
However, since most CPUs have a cache between the main memory bus and the cores this is often avoided. If the access hits the cache then no access may go out onto the bus. Alternatively, the cache may allow writes to main memory to be deferred until the bus is inactive, or a write may never go out to main memory at all. Some caching hardware "snoops" the memory bus, effectively listening for memory access that would hit a cache line in the cache, and then updates the cache automatically. This saves the CPU from having master a separate transaction to get the DMA'd data from RAM into the cache.
Additionally, some buses support a priority system for hardware, which would allow the DMA to be suspended if higher priority hardware needs to use the bus, such as the CPU.