Transcript IOSystems
I/O Hardware Incredible variety of I/O devices Operating System Concepts 13.1 Silberschatz, Galvin and Gagne 2002 I/O Hardware Devices vary in many dimensions Direction Read, Write, Read-Write Character, Block Speed Latency, Transfer rate, Delay between operations Access Sequential, Random Sharing Sharable, Dedicated IO subsystem reduces perceived differences for apps, and optimizes performances for apps Operating System Concepts 13.2 Silberschatz, Galvin and Gagne 2002 I/O Hardware Common concepts (provide abstraction) Port (serial, parallel, ethernet) Bus (daisy chain or shared direct access) Controller (host adapter operates ports/bus/device) See my picture Operating System Concepts 13.3 Silberschatz, Galvin and Gagne 2002 A Kernel I/O Structure Operating System Concepts 13.4 Silberschatz, Galvin and Gagne 2002 Application I/O Interface I/O system calls encapsulate device behaviors in generic classes Device-driver layer hides differences among I/O controllers from kernel Makes OS independent of the IO hardware Provided by the device/controller manufacturers DDs are part of the OS (not processes) usually OS defines interface to DDs Non-standard across OSs => device manufacturers have to provide a DD for each OS (bugger) Applications normally reach the DDs via the OS Escape entry (e.g., ioctl) allows more direct access Operating System Concepts 13.5 Silberschatz, Galvin and Gagne 2002 Kernel I/O Subsystem Scheduling To maximize performance I/O request ordering via per-device queue Some OSs try fairness Device reservation - provides exclusive access to a device (e.g., in VMS) System calls for allocation and deallocation Wait for device on call, e.g., NT Watch out for deadlock Operating System Concepts 13.6 Silberschatz, Galvin and Gagne 2002 Kernel I/O Subsystem Buffering - store data in memory while transferring between devices To cope with device speed mismatch E.g., modem to disk (x1000) Double buffering To cope with device transfer size mismatch E.g., keyboard to disk Collating network packets To maintain “copy semantics” Caching - fast memory holding copy of data Always just a copy (as opposed to a buffer) Often implemented in buffer system Key to performance Operating System Concepts 13.7 Silberschatz, Galvin and Gagne 2002 Error Handling OS can recover from transient failures E.g., disk read, device unavailable, network failures Permanent failures OS can make devices unavailable Need operator intervention System calls return an error code when I/O fails System error logs hold problem reports More detailed information than return values HW diagnostic information, e.g., from SCSI controllers Operating System Concepts 13.8 Silberschatz, Galvin and Gagne 2002 Blocking and Non-blocking I/O Blocking - process suspended until I/O completed Easy to use and understand Insufficient for some needs Non-blocking - I/O call returns as much as available E.g., user interface, data copy (buffered I/O) Can be implemented via multi-threading Returns quickly with count of bytes read or written Asynchronous - process runs while I/O executes Difficult to use I/O subsystem signals process when I/O completed Operating System Concepts 13.9 Silberschatz, Galvin and Gagne 2002 Life Cycle of An I/O Request Request may be satisfied immediately, e.g., in cache The request for the DD may have to be queued CPU runs async with device DD waits in sync with device It’s blocking I/O. Non-blocking I/O always “can satisfy request” Asynchronous I/O does not block the process Direct I/O instructions Placed in registers Status, control, data-in, data-out Memory-mapped I/O Maps registers onto RAM Can be faster than I/O instructions Operating System Concepts 13.10 Silberschatz, Galvin and Gagne 2002 An Alternative - Polling Synchronous communication Controller waits for command-ready bit in controller status register bit to be set Host waits for busy bit in controller status register to be clear (initially it is clear) Host places command in command register, and any required data in data register Host sets command-ready bit, and waits for it to clear Controller notices command-ready bit, sets busy bit, clears command-ready bit. Host loops waiting for busy bit to clear, while controller does IO Controller clears busy and command-ready bits, and loops Vantages Done once for each byte Wasteful of CPU time if IO takes long Can use offboard CPU, e.g., in SCSI controller Useful for fast data streams Operating System Concepts 13.11 Silberschatz, Galvin and Gagne 2002 Direct Memory Access Used to avoid programmed I/O for large data movement If device transmits close to memory speeds, little time is left for processing Requires DMA controller Bypasses CPU to transfer data directly between I/O device and memory DMA controller steals RAM cycles from CPU Operating System Concepts 13.12 Silberschatz, Galvin and Gagne 2002 Six Step Process to Perform DMA Transfer Operating System Concepts 13.13 Silberschatz, Galvin and Gagne 2002 Spooling Spooling - hold output for a device If device can serve only one request at a time Provides asynchronous I/O i.e., Printing and the lpd Operating System Concepts 13.14 Silberschatz, Galvin and Gagne 2002 Network Devices Approaches vary widely Pipes FIFOs Streams Queues Mailboxes Socket interface Separates network protocol from network operation Includes select functionality Operating System Concepts 13.15 Silberschatz, Galvin and Gagne 2002 Kernel Data Structures Kernel keeps state info for I/O components, including open file tables, network connections, character device state Many, many complex data structures to track buffers, memory allocation, “dirty” blocks Some use object-oriented methods and message passing to implement I/O, e.g., NT, Nachos, nu Operating System Concepts 13.16 Silberschatz, Galvin and Gagne 2002 4.3 BSD Kernel I/O Structure Cooked interfaces are buffered Block buffers C-lists Raw interfaces are unbuffered Devices have major and minor device numbers. Major number used as index into array of DD entry points Direct access via ioctl() and /dev files Operating System Concepts 13.17 Silberschatz, Galvin and Gagne 2002 Block Buffer Cache Consist of buffer headers, each of which can point to a piece of physical memory, and contain a device number and a block number on the device. The buffer headers for blocks not currently in use are kept in several linked lists: Buffers not recently used, or with invalid contents (AGE list). Buffers recently used, linked in LRU order (LRU list). Buffers with no associated physical memory (EMPTY list). On read the cache is searched. If the block is found it is used - no I/O transfer is necessary. If it is not found, a buffer is chosen from the AGE list, or the LRU list if AGE is empty. On write, if the block is in the cache, then write and set dirty bit Dirty blocks are output by regular sync() Blocks may be fragmented, and headers are taken from EMPTY Operating System Concepts 13.18 Silberschatz, Galvin and Gagne 2002 Raw Device Interfaces The raw device interface — unlike the block interface, it bypasses the block buffer cache. Each disk driver maintains a queue of pending transfers: whether it is a read or a write a main memory address for the transfer a device address for the transfer a transfer size Can use user memory for a transfer record Operating System Concepts 13.19 Silberschatz, Galvin and Gagne 2002 C-Lists Terminal drivers use a character buffering system which involves keeping small blocks of characters in linked lists. A write system call to a terminal enqueues characters on a list for the device. An initial transfer is started, and interrupts cause dequeueing of characters and further transfers, i.e., it’s asynchronous Input is similarly interrupt driven. It is also possible to have the device driver bypass the canonical queue and return characters directly from the raw queue — raw mode (used by full-screen editors and other programs that need to react to every keystroke). Operating System Concepts 13.20 Silberschatz, Galvin and Gagne 2002