I/O System (12)

Question 1

Typical pc bus structure

Answer 1

Question 2

Type of I/O devices

Answer 2

• Storage
• transmission
• human-interaction

Question 3

I/O Port

Answer 3

Connection point

Question 4

I/O: Bus

Answer 4

• Could be daisy chain or shared direct acces
• PCI bus common in PCs and servers
• Exansion bus connects relatively slow devices
• Serial-attached SCSI: common disk interface

Question 5

I/O Controller

Answer 5

• AKA host adapter
• Sometimes it is integrated, sometimes it is in a separate circuit board
• Contains processor, microcode, private memory and bus controller

Question 6

Polling

Answer 6

For each byte of I/O:

1. Read busy bit from status register until 0
   
   • This is busy-waiting
     
     • Ok if device is fast
     • inefficient if device is slow
     • 2. Hosts sets read or write bit and copies data into data-out register if write is true
2. Host sets command-ready bit
3. controller sets busy bit, executes transfer
4. controller clears busy bit, error bit, command-ready bit when transfer done

Question 7

Interrupts

Answer 7

• Interrupts are a better choice if the busy bit is often 1;
• CPU interrupt-request line triggered by I/O device (processor checks it every epoch)
• Interrupt handler: handles it
  
  • Can be masked
• Interrupt vector:
  
  • to call the right interrupt handler
  • some handlers are not maskable
  • can be prioritized
  • Interrupt chaining if more than one device at same interrupt number
• Used also for exceptions
  
  • Page fault when memory access error
  • Terminate process, crash system due to hw error
• Multi-cpu sys. can handle interrupt concurrently

Question 8

DMA

Answer 8

• Direct Memory Access
• Used to avoid programmed I/O (one word at a time) for large data movement
• Requires DMA controller
  
  • Bypasses to CPU to transfer data directly between I/O and memory
• OS writes DMA command into memory block
  
  • source and dest location
  • read/write mode
  • bytes count
  • writes location of command block to DMA controller
  • Steals cycles (uses bus) from CPU, but still more efficient
  • Interrupt on completion
• DVMA: aware of virtual addresses.

Question 9

Application I/O interface

Answer 9

• Abstract interface
• Device layer hides differences among I/O controller from kernel
• I/O sys calls encapsule device behaviours in different classes
  
  • Behaviours
    
    • Character stream vs. block
    • sequential vs random access
    • Sync vs Async vs both
    • Sharable vs dedicated
    • speed of operation
    • read-write vs read only vs write only
• Each OS has its own I/O subsystem that handles this stuff

Question 10

I/O devices distinctive traits

Answer 10

• Device driver handle in a different way devices based on the traits
• Usually can be grouped by OS into:
  
  • block i/o
  • character i/o (stream)
  • memory-mapped file access
  • network sockets
• Usually there is a way to have raw access
  
  • UNIX: ioctl() -> sends arbitrary bits to device control register
    *

Question 11

Block and character devices

Answer 11

• Block devs:
  
  • disk drives
  • commands
    
    • read
    • write
    • seek
  • raw i/o, direct i/o, filesystem access
  • memory-mapped file possible
  • DMA
• Character devs:
  
  • keyboard, mouse, serial ports
  • commands
    
    • put
    • get

Question 12

Network devices

Answer 12

• Can vary:
  
  • block
  • character (stream)
• Linux, Unix, Windows include socket interface
  
  • separates network protocol from network op.
• many different approaches
  
  • pipes
  • fifo
  • streams
  • queues

Question 13

Clocks and timers

Answer 13

• Used to get time, elapsed time, timer
• Used to generate recurring waveforms
• ioctl() covers aspects of these

Question 14

Nonblocking and async I/O

Answer 14

• Blocking: process suspended until I/O completed
• Nonblocking: I/O returns as much as it is available
  
  • UI, data copy
  • done via multithreading
  • returns quickly with count of bytes read or written
• Async: process runs while I/O executes
  
  • Difficult to use
  • I/O subsystem signals completion

Question 15

Vectored I/O

Answer 15

• Allows one system call to do I/O in parallel on multiple devices
• Example: readve
  
  • accepts a vector of multiple buffers to read/write from/to
• Better than individuals calls
  
  • decreases context switching
  • some versions provide atomiciy
    
    • preventing threads to change data while read/writes are being performed

Question 16

Kernel I/O subsystem Scheduling

Answer 16

• Scheduling
  
  • can be done via queue per device
  • can be done using fairness as a metric
  • can be done using Quality of Service as a metric (IPQOS)

Question 17

Kernel I/O subsystem Buffering

Answer 17

• Buffering: store data in memory while transferring between devs
  
  • must be done because of:
    
    • speed mismatch
    • transfer size mismatch
  • To maintain copy-semantics: the version of the data writte to disk is the correct one
    
    • simple way: copy application data into kernel buffer before returning control to application
      
      • disk wirte performed from kernel buffer, so subsequent changes do not matter.
  • double buffering: two copies of data
    
    • kernel and user
    • varying sized
    • full/being processed and not-full/being used
    • copy-on-write can be used for max efficiency in some cases

Question 18

Kernel I/O Subsystem caching

Answer 18

• The faster device should cache data
• should be a copy
• impactful on performance
• can be combined with buffering sometimes

Question 19

Kernel I/O Subsystem Spooling

Answer 19

• Hold data to output until the device completes the I/O request.
• Used in printing, because the printer is very very very slow

Question 20

Kernel I/O Subsystem Device reservation

Answer 20

• Provides exclusive access to device
• there are sys calls for alloc/dealloc
• be aware of deadlock

Question 21

Error Handling

Answer 21

• OS can recover from
  
  • disk read
  • unavailable device
  • write failures
• How:
  
  • retrying
  • more advanced system stop using the device after frequency of errors gets high
• operations return error codes
• Errors are reported in log

Question 22

I/O protection

Answer 22

• User may disrupt normal operation via illegal I/O instr.
  
  • all i/o instructions must need privileges
  • I/O must be done via sys calls
  • Memorymapped i/o and i/o port memory location must be protected too.

Question 23

Kernel Data Structures

Answer 23

• Kernel keeps state of IO comp:
  
  • open files
  • network connection
  • character device state
• Kernel has also complex data structures to track buffers, memory allocation, dirty blocks
• Some use object oriented methods and message passing to handle I/O
  
  • windows uses message passing: message flows from user to kernel

Question 24

Power management

Answer 24

• Related to I/O
• Computers use electricity, generate head -> require cooling
• OSes can help manage and improve use:
  
  • Virtual machine orchestration: move VMs around server to balance power use
• Mobile computing has power management as one of the main modules in OS
• modern sysstem use ACPI firmware that has routines for dev discovery, management, error, and power management that can be called by the kernel.

Question 25

Kernel I/O subsystem general view

Answer 25

* Coordinates extensive collection of services, made available to applications and kernel: * access control * operation control (can't seek on a modem) * fs allocation * buffering, caching, spooling * i/o scheduling * dev status monitoring * dev-driver config * power management * Upper level of the I/O subsystem access devices via the standard device driver interface.

Question 26

Life Cycle of I/O req.

Answer 26

Question 27

STREAMS

Answer 27

* Implemented from Unix sys. V on * Full duplex comm. channel between user process and a device * STREAM: * stream head interfaces with user proc. * driver end interface with dev. * 0 or more modules between the above points * Modules contain read and write queue * Message passing is used to communicate between queues * Async inside, sync communication with stream head.

Question 28

Performance and how to improve it

Answer 28

* Factors: * context switch between interrupts * data copying * network traffic * Request to CPU to execute device driver and kernel I/O code * Improving * reduce context switch * reduce data copying * reduce interrupts by * making large transfers using dma * using smart controllers * polling * move user process / daemons to kernel threads * balance CPU, memory, bus and I/O for max. throughput

Question 29

Device I/O code impact on performance

Answer 29

Question 30

Advantages of kernel buffering

Answer 30

* In a system with page and/or swapping it is possible to enable swapping out a process while waitining on I/O, while the kernel buffers receives the data that the process is waitinig for. Simply put: we can decouple work from the user memory to disk space. * The advantage over a single buffer is that the single buffe is also having a pipelined behaviour, in which we can transfer from kernel to user and from disk to kernel at the same time. * Kernel buffer can act as a cache, the process could find the kernel buffer filled without having to wait.