Chapter 13: I/O Systems- 6th ed I/O Hardware Application I/O Interface Kernel I/O Subsystem презентация

I/O Subsystem
Transforming I/O Requests to Hardware Operations
Streams
Performance

Review Chapters 2 and 3, and instructors notes on:
“Interrupt schemes and DMA”
This chapter gives more focus to these chapters and topics.

Instructor’s annotations in blue
Updated 12/5/03

interface to CPU - status, control, data
Bus (daisy chain or shared direct access) - main and specialized local (ex: PCI for main and SCSI for disks)
Controller (host adapter) - HW interface between Device and Bus - an adapter card or mother board module Controller has special purposes registers (commands, etc.) which when written to causes actions to take place - may be memory mapped
I/O instructions control devices - ex: in, out for Intel
Devices have addresses, used by
Direct I/O instructions - uses I/O instructions
Memory-mapped I/O - uses memory instructions

a device includes both control and data ports

I/O from device When not busy - set data in data port, set command in control port and let ‘er rip
Not desirable if excessive - since it is a busy wait which ties up CPU & interferes with productive work

Remember CS220 LABs

handler receives interrupts
Maskable to ignore or delay some interrupts
Interrupt vector to dispatch interrupt to correct handler
Based on priority
Some unmaskable
Interrupt mechanism also used for exceptions
Application can go away after I/O request, but is til responsible for transferring data to memory when it becomes available from the device.
Can have “nested” interrupts (with Priorities)
See Instructors notes: “Use of Interrupts and DMA”
Soft interrupts or “traps” generated from OS in system calls.

cannot be disabled

responsible for transferring data from controller to memory (on interrupt) when device mad it available.
Now DMA will do this - all CPU has to do is set up DMA and user the data when the DMA-complete interrupt arrives. … Interrupts still used - but only to signal DMA Complete.

Used to avoid programmed I/O for large data movement
Requires DMA controller
Bypasses CPU to transfer data directly between I/O device and memory
Cycle stealing: interference with CPU memory instructions during DMA transfer. - DMA takes priority - CPU pauses on memory part of word.

devices (an API to the programmer)
Attempts to abstract the characteristics of the many I/o devices into a few general classes.
I/O “system calls” encapsulate device behaviors in generic classes
Device-driver layer hides differences among I/O controllers from kernel
Devices vary in many dimensions
Character-stream or block
units for data transfer bytes vs blocks
Sequential or random-access - access methods
Synchronous (predictable response times) vs asynchronous (unpredictable response times)
Sharable or dedicated - implications on deadlock
Speed of operation - device/software issue
read-write, read only, or write only - permissions

==>

Example: ioctl(…) generic call
(roll your own)
in UNIX (p. 468), and other more specific commands or calls
open, read, ...

Fig. 13.6

at a low level

Use of I/O buffering

or sector clusters on a disk
Commands/calls include read, write, seek
Access is typically through a file-system interface
Raw I/O or file-system access - “binary xfr” of file data - interpretation is in application (personality of file lost)
Memory-mapped (to VM) file access possible - use memory instructions rather than I/O instructions - very efficient (ex: swap space for disk).
Device driver xfr’s blocks at a time - as in paging
DMA transfer is block oriented
Character devices include keyboards, mice, serial ports
Device driver xfr’s byte at a time
Commands include get, put - character at a time
Libraries layered on top allow line editing - ex: keyboard input
could be beefed up to use a line at a time (buffering)
Block & character devices also determine the two general device driver catagories

own interface - OS makes network device interface distinct from disk interface - due to significant differences between the two
Unix and Windows NT/9i/2000 include socket interface
Separates network protocol from network operation
Encapsulates details of various network devices for application … analogous to a file and the disk???
Includes select functionality - used to manage and access sockets - returns info on packets waiting or ability to accept packets - avoids polling
Approaches vary widely (pipes, FIFOs, streams, queues, mailboxes) … you saw some of these!

interval time used for timings, periodic interrupts
ioctl (on UNIX) covers odd aspects of I/O such as clocks and timers - a back door for device driver writers (roll your own). Can implement “secret” calls which may not be documented in a users or programming manual

blocks - lets other process execute) suspended until I/O completed
Easy to use and understand
Insufficient for some needs
multi-threading - depends on role of OS in thread management
Nonblocking - I/O call returns as much as available
User interface, data copy (buffered I/O)
Implemented via multi-threading
Returns quickly with count of bytes read or written - ex: read a “small” portion of a file very quickly, use it, and go back for more, ex: displaying video “continuously from a disk”
Asynchronous - process (making the asynch request) runs while I/O executes
Difficult to use - can it continue without the results of the I/O?
I/O subsystem signals process when I/O completed - via interrupt (soft), or setting of shared variable which is periodically tasted.

Fig 13.6
Scheduling
Some I/O request ordering via per-device queue
Some OSs try fairness
Buffering - store data in memory while transferring between devices
To cope with device speed mismatch - de-couples application from device action
To cope with device transfer size mismatch
To maintain “copy semantics” - guarantee that the version of data written to device from a buffer is identical to that which was there at the time of the “write call” - even if on return of the system call, the user modifies buffer - OS copies data to kernel buffer before returning control to user.
Double or “ping-pong” buffers - write in one and read from another - decouples devices and applications … idea can be extended to multiple buffers accesses in a circular fashion

copy of data
Always just a copy
Key to performance
How does this differ from a buffer?
Spooling - a buffer holding output/(input too) for a device
If device can serve only one request at a time
Avoids queuing applications making requests.
Data from an application is saved in a unique file associated with the application AND the particular request. Could be saved in files on a disk, or in memory.
Example: Printing
Device reservation - provides exclusive access to a device
System calls for allocation and deallocation
Watch out for deadlock - why?

transient write failures
Most return an error number or code when I/O request fails
System error logs hold problem reports

CRC checks - especially over network transfers of a lot of data, for example video in real time.

including open file tables, network connections, character device state
used by device drivers in manipulating devices and data transfer, and in for error recovery
data that has images on the disk must be kept in synch with disk copy.
Many, many complex data structures to track buffers, memory allocation, “dirty” blocks
Some use object-oriented methods and message passing to implement I/O
Make data structures object oriented classes to encapsulate the low level nature of the “device” - UNIX provides a seamless interface such as this.

and 12 on files

from disk for a process: How is connection made from file-name to disk controller:
Determine device holding file
Translate name to device representation
Physically read data from disk into buffer
Make data available to requesting process
Return control to process

See the 10 step scenario on pp. 479-481 (Silberschatz, 6th ed.) for a clear description.

read ahead

user-level process and a device

A STREAM consists of:
- STREAM head interfaces with the user process
- driver end interfaces with the device - zero or more STREAM modules between them.

Each module contains a read queue and a write queue

Message passing is used to communicate between queues

on CPU to execute device driver, kernel I/O code
resulting in context switching
interrupt overhead
Data copying - loads down memory bus
Network traffic especially stressful
See bulleted list on page 485 (Silberschatz, 6th ed.)
Improving Performance See bulleted list on page 485 (Silberschatz, 6th ed.)
Reduce number of context switches
Reduce data copying
Reduce interrupts by using large transfers, smart controllers, polling
Use DMA
Move proccessing primitives to hardware
Balance CPU, memory, bus, and I/O performance for highest throughput

… device hardware

Decision depends on trade-offs in the design layers: