Overview

• Virtual Memory and Memory Management:
  • virtual memory hardware
  • segmentation and paging
• Block Devices
• Drivers:
  • RAMdisk
  • hard disk
  • terminal
  • display
• Fork, exit, exec, and modules implementations
• Signals and signal handling
• File Systems
  • Posix API
  • implementation

Virtual Memory

• each process should have its own address space
• and be unable (except in unusual circumstances) to read or write other processe's address spaces
• exception: it would be good to share our text (code) segments
• this means program addresses (virtual addresses) are not the same as machine addresses (physical addresses)
• virtual memory hardware makes this reasonably fast:
  • the memory management unit (MMU) does this translation, including
  • the translation lookaside buffer (TLB) is an associative memory that keeps recent translations of virtual to physical addresses
  • the TLB may be refilled in hardware or in software
• the memory management hardware also provides page protection, and page faults for pages that are not mapped

Memory Management

• segmentation vs. paging
• should a process's virtual memory be contiguous in physical memory?
  • yes: segmentation. Fewer segment registers are needed, but management of physical memory is more complex
  • no: paging. More page table entries are needed, but management of blocks is simpler
• what should be swapped to disk?
  • entire processes
  • unused pages that are not likely to be referred to again soon
• how to find free memory?
  • first fit, best fit, etc. Or, move segments around in physical memory
  • keep a linked list of free blocks, with the pointers stored in the nodes themselves
• segmentation also refers to the different parts of a computer program (on disk) or memory areas of a process (in memory) -- e.g. the stack segment
• paging may refer to bringing in blocks of a memory-mapped file in from disk, or writing dirty pages out to disk

Block Devices

• floppy, hard disk are physically block oriented
• for convenience, CD-ROM, RAMdisk, Flash storage, etc, which are not intrinsically block oriented, can be used in a block-oriented fashion
• I/O is done one or more blocks at a time (no transfer is less than one block)
• in many cases, accessing successive blocks is much faster than accessing blocks in random order

Device Drivers

• block device drivers all have a lot in common
• open and close the device
• read and write blocks

Fork, exit, exec, modules

• implementation, esp. in Minix
• fork
  • copy the address space, perhaps using Copy on Write, CoW
  • copy file descriptors, do the right thing with signals
• exit: deallocate resources, inform parent
• zombie processes
• exec
  • replace text (code)
  • initialize stack, including arguments
  • jump to the initial location
• overlays: load and execute parts of a program, under program control
• modules
  • load and execute parts of the kernel
  • module gets control on interrupt, and when system calls want to do I/O

Signals

• an "interrupt" for user processes
• OS signals a process whenever there is an illegal operation: bad memory accesses (bad pointers), arithmetic errors (divide by zero), bad I/O, and illegal instructions
• OS signals a process also when told to: kill, alarm, ^C, ^Z
• a process can tell the OS to ignore (most) signals, use the default behavior, or call a signal handler
• the default behavior might be to ignore the signal or kill the process, perhaps with a core dump

File Systems

• Posix API: open, creat, close, fsync, read, write
• also system calls: mmap, munmap
• file system implementation:
  • contiguous allocation is a simple strategy and very fast for sequential access, but deletion requires defragmenting, and only one file can be growing at any given time
  • block allocation is much more common. It is harder to make accesses fast, but there is no need for defragmentation, and it is very flexible
  • block allocation requires keeping track of blocks
  • an index node (inode) can directly keep track of the first few blocks, and indirectly of the remaining blocks as well
  • given that the file system uses inodes, that is a good place to store file attributes as well
• directories are handled like normal files with special content (and the file system must know they are directories)
• variable length file names in a directory require some form of storage management of the space in the directory

Projects

• Project 3: create a system call, modify the scheduler to be real-time
• Project 4: create a new RAMdisk
• Project 5: Linux and sbrk
• Project 6: how do you make a file system?

In this class so far

• basics: shells, system calls, user-level I/O
• processes, synchronization, deadlocks, scheduling, interrupts, context switching
• virtual machines
• operating system structure, Minix organization: message passing, kernel/drivers/servers/user programs
• virtual memory
• kernel implementation of I/O
• file systems and persistent storage