Input/Output (CS 273 (OS), Fall 2020)

Layered architecture of I/O system

User level software.
- System calls ("unbuffered I/O") such as open, read
- Higher-level "buffered I/O" libraries such as stdio.h, with functions printf, scanf, etc.
- Layers of buffering
- Spooling
Device-independent OS software.
- Naming; e.g., major/minor device numbers, file paths
- Protection
- Uniform block size
- Low-level buffering
- Identifying free blocks for file system
- File locks (dedicated)
- Higher error processing.
Device driver.
- Interact concretely with obscure, cranky device
- Most error handling
Interrupt handler.
- An interrupt is a hardware signal notifying the CPU of a hardware event.
- Interrupt handler code is executed when an interrupt occurs; different handlers for different interrupt numbers.
- Interrupt vector
- Precise interrupt:
  - PC is saved in a known place
  - All instructions before *PC have been fully executed
  - No instructions beyond *PC have been executed
  - Execution state of *PC is known
  Relevance: In a superscalar machine (e.g., Pentium), instructions (or parts of instructions) may be completed out of order.
  Note that any instructions that had been executed beyond *PC would have to be "undone."

Device controller.
- Attached to bus; provides an interface for a specific device with all its technical details.
- E.g.: Disk controller reads disk blocks, performs error correction, performs DMA (using more buffering)
- E.g.: Video controller governs the CRT, accesses video memory
- E.g.: Network controller implements low level protocols (e.g., Ethernet)
Device
- Block devices, e.g., disks (including CD-ROM, DVD)
- Character devices, e.g., keyboard, serial lines, mouse
- Memory-mapped I/O (video RAM memory is another example of a buffer)
- DMA
  1. CPU programs DMA controller
  2. DMA controller requests data transfer from disk controller
  3. Disk controller transfers data from disk to main memory
  4. Disk controller sends ACK to DMA controller
  5. DMA controller sends interrupt to CPU
- Other, e.g., clocks

Programmed I/O
- OS polls (busy wait until device is ready) for each I/O action, using an OS-level buffer to store the data being input or output.
- Advantages: Simple
- Disadvantages: Busy wait
Interrupt-driven I/O
- Block the process requesting I/O until the operation can be performed without polling.
- Advantages: Better CPU utilization
- Disadvantages: Handling an interrupt for each character
DMA
- Move large quantities of data in a single operation, asynchronously with CPU
- Advantages: Considerably less overhead per amount of data moved, for large quantities
- Disadvantages: Slower (greater overhead) in the case of small amounts of data