Although interrupt driven I/O is much more efficient than program controlled I/O, all data is still transferred through the CPU. This will be inefficient if large quantities of data are being transferred between the peripheral and memory. The transfer will be slower than necessary, and the CPU will be unable to perform any other actions while it is taking place.
Many systems therefore use an additional strategy, known as direct memory access (DMA). DMA uses an additional piece of hardware - a DMA controller. The DMA controller can take over the system bus and transfer data between an I/O module and main memory without the intervention of the CPU. Whenever the CPU wants to transfer data, it tells the DMA controller the direction of the transfer, the I/O module involved, the location of the data in memory, and the size of the block of data to be transferred. It can then continue with other instructions and the DMA controller will interrupt it when the transfer is complete.
The CPU and the DMA controller cannot use the system bus at the same time, so some way must be found to share the bus between them. One of two methods is normally used.
Burst mode
The DMA controller transfers blocks of data by halting the CPU and controlling the system bus for the duration of the transfer. The transfer will be as quick as the weakest link in the I/O module/bus/memory chain, as data does not pass through the CPU, but the CPU must still be halted while the transfer takes place.
Cycle stealing
The DMA controller transfers data one word at a time, by using the bus during a part of an instruction cycle when the CPU is not using it, or by pausing the CPU for a single clock cycle on each instruction. This may slow the CPU down slightly overall, but will still be very efficient.
NEXT STOP: Interfacing Concepts: Serial
