index.md (1953B)
1 +++ 2 title = 'Principles of IO software' 3 +++ 4 # Principles of IO software 5 the concepts: 6 7 - device independence: 8 - IO software provides abstraction over actual hardware 9 - programs shouldn't have to care about device particularities 10 - uniform naming: 11 - because if you have to type a unique 20 letter id to access a device you’ll give up 12 - error handling: 13 - errors should be handled closest to their source so we don’t have to give a crap 14 - synchronous vs asynchronous IO: 15 - programs don't want to deal with interrupts. 16 - so OS turns async operations into blocking operations 17 - then lower levels have to deal with interrupts 18 - buffering: 19 - networking: incoming packets have to be buffered 20 - audio: buffering to avoid clicks 21 22 the layers in practice: 23 24  25 26 interrupt handler 27 28 - device driver doesn't handle low-level interrupt directly 29 - lowe-level handler is relatively generic 30 - on linux, calls device driver-specific interrupt handler if registered 31 - on minix, sends message to device driver that registered for interrupt 32 33 device driver: 34 35 - accepts abstract reqs from device-independent layer and transforms them into commands for the device 36 - can queue new requests as necessary 37 - usually needs to wait for completion of the request (IRQ) 38 - checks for errors and answers requests from device-independent layer 39 40 device independent IO software: 41 42 - uniform interfacing for device drivers (driver nterface, naming (major/minor numbers), and protection 43 - buffering: necessary for both character and block devices 44 - error reporting: hardware-level, driver-level, etc. 45 - allocing/deallocing dedicated devices (e.g. printers) 46 - device-independent block size: unify blocks/characters across devices 47 48 user-level IO software 49 50 - single interface allowing user to access devices 51 - in C: fopen, close, fflush, frpintf, and so on 52 - safely multiplexes access to exclusive devices