Transferring Data To and From Devices - Introduction

OK - we know when a device needs service, and we know (in general terms) how we communicate commands and status to and from the device. Now there is one more question before we see some examples:

How is data transferred to and from devices?

There are, again, two possibilities:

• Programmed I/O and
• DMA

In looking at these, let's take input from, say, a disk or network controller as an example.

Typically a device will collect input data in the controller. When there is a certain amount of data in the controller the device will generate an interrupt (or set a flag in a status register in the case of polled I/O) to inform the CPU. Now we need to somehow get the data from the device to main memory.

• With programmed I/O the CPU does the actual tranfer itself, probably a word at a time (possibly in larger chunks depending on the controller and instruction set of the CPU). The transfer may proceed quite quickly, but the CPU is tied up the entire time doing the transfer itself. This is not an issue if there is no other process to run, but is definitely an issue if there is some other process that could be making use of the CPU. In that case the ready process will have to wait.
• With DMA (direct memory access), the controller is capable of writing the buffer contents to main memory itself. During the write, the CPU is free to execute another process. When the transfer is complete, the controller can infor the CPU by way of an interrupt.

Notice that even though the CPU is freed up to execute another process, there is still additional contention for another critical hardware component - the bus. Here the bus is being used both by the DMA device to transfer the buffer contents to memory, and by the CPU to fetch instructions and manipulate data. This can cause slower process execution during DMA, but is still a big win overall.