I/O Hardware: The CPU's View of Devices

The CPU does not interact directly with devices like keyboards, disks, or printers. Instead, each device is managed by a Device Controller — a specialized electronic circuit (on the motherboard or the device itself) that handles low-level hardware details.

Device Controllers

A device controller is responsible for:

• Converting the byte stream from the OS into the electrical signals required by the device.
• Managing the device's internal buffer (e.g., the 512-byte sector buffer on a hard drive).
• Reporting status (busy, ready, error) via status registers.
• Accepting commands (read, write, format) via command registers.

Memory-Mapped I/O vs Port-Mapped I/O

Three I/O Techniques: From Polling to DMA

Historically, three techniques have been used to perform I/O. Each trades simplicity for efficiency.

1. Programmed I/O (Polling)

The CPU does all the work — it continuously checks the device's status register to see if it is ready.

1// Example: Reading a byte from a serial port via polling
2while ((inb(STATUS_PORT) & BUSY) != 0) {
3    // Spin — wait for device to become ready
4}
5byte data = inb(DATA_PORT);   // Read the data byte

2. Interrupt-Driven I/O

The CPU issues a command and then continues executing other processes. The device sends an interrupt when the operation is complete.

1// Example: Reading a sector from disk via interrupts
2// Step 1: Issue the command
3outb(COMMAND_PORT, READ_SECTOR);
4outb(LBA_PORT, sector_number);   // Specify which sector
5
6// Step 2: CPU is free to do other work!
7// ... process scheduling, computation, etc. ...
8
9// Step 3: When disk finishes, it raises an interrupt.
10// The Interrupt Handler executes:
11void disk_interrupt_handler() {
12    byte data[512];
13    for (int i = 0; i < 512; i++) {
14        data[i] = inb(DATA_PORT);
15    }
16    // Signal to the waiting process that data is ready
17}

3. Direct Memory Access (DMA)

For high-speed devices, even interrupt-driven I/O is too slow because the CPU must still copy data between the device buffer and memory byte-by-byte. DMA solves this by letting the device controller transfer data directly to/from RAM without CPU involvement.

How DMA Works (Step by Step):

1. The CPU programs the DMA controller with: source address (device buffer), destination address (RAM buffer), and transfer size.
2. The CPU is free to execute other processes while the DMA controller manages the transfer.
3. The DMA controller takes control of the system bus and transfers data word-by-word from device to RAM (or RAM to device).
4. When the full transfer is done, the DMA controller raises a single interrupt to signal completion.

Comparison of I/O Techniques:

Interrupt Handlers: The Kernel's Urgent Response

An interrupt is a hardware signal that tells the CPU that an event requiring immediate attention has occurred. When a device raises an interrupt, the CPU must stop its current execution, handle the device's needs, and then resume.

Interrupt Vector Table (IVT) / Interrupt Descriptor Table (IDT)

Every interrupt type is assigned a unique number (the interrupt vector). The CPU uses this number as an index into a table of handler addresses:

• x86 Real Mode: IVT at address 0x0000:0000 (256 entries × 4 bytes = 1024 bytes).
• x86 Protected Mode: IDT — up to 256 entries, each 8 bytes, location specified by the IDTR register.

Maskable vs Non-Maskable Interrupts

Top Half vs Bottom Half

Modern OS kernels split interrupt handling into two parts to minimize the time interrupts are disabled:

Device Drivers: The Kernel's Device Abstraction

A device driver is a kernel module that understands the specific protocol of a device controller and presents a uniform interface to the rest of the OS.

The Layered I/O Architecture

Key Insight: The VFS allows user applications to issue read() and write() calls without knowing whether the underlying device is a hard disk, SSD, USB drive, or network file system. The driver handles the conversion.

Driver Characteristics

SPOOLING: Sharing Exclusive Devices [2023 Q1c]

SPOOLING stands for Simultaneous Peripheral Operation Online. It is a technique that makes an exclusive (non-sharable) device appear sharable by buffering its output to a high-speed storage device (usually disk).

RAID: Redundant Array of Independent Disks [2024 Q1c]

RAID stands for Redundant Array of Independent Disks. It is a technique that combines multiple physical disk drives into a single logical unit to improve performance and/or reliability.

Key Insight: RAID 0 provides performance (parallel I/O) but zero redundancy. RAID 1 provides redundancy by duplication. RAID 5/6 provide redundancy with less space overhead by using parity calculations.

CLV vs CAV: Disk Rotation Strategies [2023 Q8a]

I/O Buffering and Caching

To improve I/O performance, the OS uses multiple levels of buffering and caching.