Addressing Page Tables

Physical Page Tables: Set a predefined portion of physical ram for kernel page tables

• Inflexible - Page tables cannot be swapped
• Simple - Hardware support easy to impl

Kernel Virtual Memory: Use page tables to store kernel page tables

• Need to keep special code & data (e.g. outer PT, fault handlers) in physical memory

Managing Swap Space

1. Use a raw disk partition: Faster, require disk formatting to resize
2. Use a file in fs: Slower, more flexible to resize

Virtual Memory Area

VMA: A description of each address space region (e.g. shared libraries)

• Example Info: read/write/execute/shared/locked, name
• Linked list
• mmap/malloc creates a new VMA

MMap: Map a file to a memory region, synchronous read/write

Segfault: On page faults, check VMA first to verify if the address is valid

Ensuring Fairness

Space Fairness

• Fixed-Space Algos: Each process has a limit of pages

• Variable-Space Algos: Processes’ page limit grows/shrinks dynamically

  • Working Set Model: WS(t, D) = {pages referenced in [t, …, t-D]} The page limit need to be > working set size (problem: hard to determine D, and WS often changes)

  

Replacement

• Local Replacement: Evict each process’ own pages when full
• Global Replacement: One process can evict other processes’ pages

Sharing

Sharing pages: Put the same PTE in another process’ PT (with different permissions)

• Must update both PTE during page swapping
• Process shouldn’t store their pointers in the shared area (invalid for other processes)

Copy on Write: Avoid copying as long as possible, create shared mapping instead

• Read-only for both parent and child in fork(), copy when protection fault is triggered