This module serves as an introduction to IO hardware, and its interactions with the CPU. We will introduce different types of hardware devices, how they communicate with the CPU, and different implementations for how a CPU handles IO requests from different hardware.

IO Hardware

Congratulations on reaching the end of this lesson! Let's review what we have learned so far about IO hardware:

- There are different types of IO devices that can be categorized into three general categories: _human-readable, machine-readable, and communication_. 
- There are two types of device drivers: _kernel-mode and user-mode_. Device drivers support the communication between IO devices and the CPU.
- There are three methods that IO devices use to read/write data: _character, block, and network_. 
- In IO systems, _blocking_ is a method in which when an IO makes a request, an application typically cannot continue executing other requests until it has the necessary information changes from the IO. 
- In contrast, _non-blocking_ requests get placed into a queue while waiting so that the CPU resources can be used to complete other tasks from the event pool of an application.
- The _interrupt handler_ is like a pool or queue of _interrupts_ being sent to the CPU. It handles the execution of interrupt signals as they are received from IO devices.
- _Memory-mapped IO_ refers to a system that is designed to allow both an IO device that is connected to a computer and the memory of the computer to share address space to the interface.
- _Direct memory access (DMA)_ refers to a method in which IO devices have direct access to the main memory of a computer. For DMA, a CPU will trigger the execution of data to/from an IO device to a computer, but then will continue to complete other tasks while the data transfer executes.


A desk is supporting a computer monitor and speakers. The speakers are playing music. The monitors show a video of planets moving. A printer is printing images.

Review

_Memory-mapped IO_ refers to a system that is designed to allow both an IO device that is connected to a computer, and the memory of the computer to share address space to the interface. 

There are a few advantages to using memory-mapped IO. It allows for similar sets of instructions to be used over multiple hardware components. IO devices are not treated differently from other kinds of memory devices. A separate set of instructions is not required for IO devices to communicate with a computer.

_Direct memory access (DMA)_ refers to a method in which IO devices have direct access to the main memory of a computer without too much involvement of the CPU. For DMA, a CPU will trigger the execution of data to/from an IO device to a computer, but then will continue to complete other tasks while the data transfer executes. In order to implement the DMA method, computers use hardware devices known as _Direct-memory access controllers_. 

One advantage of using DMAs is that it removes overhead for the CPU so that the CPU may process other tasks. However, using DMA may result in _cache coherency discrepancies_. This means that if a computer has a cache, and a DMA only has access to the main memory, it may update the main memory and so the cache, which has not been updated, will not match. 

A GIF showing four boxes labeled "IO Device", "DMA", "CPU" and "Main Memory". An arrow labeled "New Data" begins at the “IO device” box and goes to the “DMA” box. After that, two arrows starting at the “DMA” box move simultaneously. One of the arrows, labeled "Interrupt" will move to “CPU”, and the other, labeled "Data", moves to “Main Memory” at the same time.

Memory-mapped IO vs Direct-Memory Access

Device drivers and device controllers are important components of IO systems. _Device drivers_ exist as software programs that the OS uses to communicate with device controllers. _Device controllers_ are hardware units that work as an interface between physical IO devices, and device drivers. An _interface_ can be thought of as a bridge that brings the software side and hardware side together.

Device drivers and controllers are crucial for different IO devices to communicate with the OS. Consider them to be like a translating service. If one person speaks fluent Spanish, and another person only speaks English, a translator, who understands both languages is necessary for meaning making in the conversation. Similarly, device controllers work as a translator between an operating system that may understand code, and hardware that uses signals. The device drivers can be thought of as the service the translator provides.

A diagram with text boxes from top to bottom: applications, user drivers, kernel drivers, hardware. A printer is an example of a user driver.

Drivers & Controllers

An _interrupt_ is a signal that is sent from the hardware of an IO device to a computer to get its immediate attention. Because interrupts are handled on the hardware’s end, they decrease overhead on the software side. When a device sends an interrupt signal, the CPU is notified via some trigger and will immediately halt the task at hand. It will send the interrupt over to an interrupt handler. The interrupt handler is like a pool or queue of interrupts being sent to the CPU. Once the interrupt has been responded to, the CPU can go back to resuming its task.
 
As an example, consider going to the grocery store looking for bananas. We see an employee stocking a shelf and decide to ask for their help. Once we have signaled for their attention, they acknowledge us, stop their task, and walk over to help us find the item. After they have found the bananas and let us know, they return to their previous task. In this example we are an IO device, sending an interrupt (our search for bananas) to the CPU (the employee).

_Polling_ provides similar functionality to interrupts; however, it is not a hardware mechanism. Polling is a CPU protocol, in which there are regular intervals set up for the CPU to take some time to check on whether any IO device requires its attention. Just as an interrupt handler takes care of interrupt requests, the CPU is responsible for handling IO requests in polling.


A diagram of a customer and employee at a grocery store. A customer signals an employee for help. The employee acknowledges the signal, stops their task, helps the customer ffind the item. After the employee completes the task and alerts the customer, they return to their previous task. 

Interrupts & Polling

A diagram showing an example of block, character, and network devices. A USB is a character device. A disk is a block device. An ethernet card is a network device.

In order for data to be successfully transferred between IO devices and the CPU, it is important that the data can be written and read over some medium. Devices are designed to read or write data in one of the following three ways:

_Character devices_ are represented as a sequential series of bytes. They are accessed one byte at a time. The operating system interacts with these devices with read/write system calls. One example of a character device is a USB. A USB has data written on it in a sequential manner.

_Block devices_ have memory stored in blocks of a fixed size. They allow for system calls where memory does not need to be read sequentially. Block devices allow for “random access”, meaning we can read or write to any place within the device. Most devices have blocks of the size 512 bytes or greater. Hard disks are a perfect example of block devices. 

_Network devices_ are different from character and block devices because they require a different interface (such as a socket interface) for access to other devices. An example of a network device is an ethernet card which is used to send and receive data over multiple devices.

Transferring Data

IO devices and operating systems communicate via data packets known as IO requests. When an IO device makes a request an application can respond in one of two ways: _blocking_, and _non-blocking_. 
 
##### Blocking Requests
Most IO requests are considered blocking requests. When the IO makes a request, an application typically cannot continue executing other requests until it has the necessary information changes from the IO. Therefore, blocking calls requires a process to stop and wait for input/output. Consider a word processing software - when we have a new document open, the application halts while waiting for the user to type some words.
 
##### Non-blocking Requests
Non-blocking requests get placed into a queue while waiting so that the CPU resources can be used to complete other tasks in an event pool. The event pool is the queue mentioned earlier. IO device responses can be handled later, as long as the request has been acknowledged by the OS. Non-blocking is also commonly referred to as asynchronous. Think about a collaborative application in which there are multiple requests being made by different users simultaneously. The application does not halt for every user each time it is waiting for some input from a single user.

Two diagrams are displayed showing how non-blocking and blocking can be applied in real-life at a store selling coffee.

In blocking, Customer 1 orders coffee while Customer 2 waits in line. Then Employee 1 takes the order. Next Employee 1 makes the coffee order. Finally, Employee 1 delivers the coffee.

In non-blocking, Customer 1 and 2 order coffee. Employee 1 takes the order. Employee 2 makes the order. Finally, Employee 2 delivers the coffee.

Blocking vs. Non-blocking

This image is a table which organizes IO devices into three categories: human-readable, machine-readable, and communication.  

IO, or Input/Output, devices refer to any physical devices that interact with a CPU. Input devices send signals to a computer and output devices allow for computers to send information out from a computer.  

When it comes to devices, we mainly think of examples that require direct human interaction; however, the large range of IO devices can be categorized into three categories: human-readable, machine-readable, and communication.

_Human readable devices_ are devices that can be interpreted/understood in a natural language structure by humans. Some examples include printers, keyboards, and a mouse.

_Machine readable devices_ are devices that are formatted to allow communication between different hardware, without the need for human interpretation. Some examples include hard drives/disks, controllers, and SD cards. 

_Communication devices_ are devices that allow devices to interact over a network. A network is a set of devices that are linked to share some resources over a shared medium. Some examples of communication devices include modems and Bluetooth adapters.