12: Memory Management

Transcription

1 12: Memory Management Mark Handley Address Binding Program goes through multiple steps from compilation to execution. At some stage, addresses in the program must be bound to physical memory addresses: Compile Time Load Time Execution Time Object file System Library Dynamically Loaded Lib Source program Compiler Object file Linker Load module Loader Running Program 1

2 Address Binding Compile Time: Know at compile time where program will go in memory. Compiler generates absolute code. Load Time: Compiler generates relocatable code. Loader does binding to absolute addresses. Execution Time: Process can be moved during execution. Binding must happen at run time - requires use of special memory management hardware. Binding at Execution Time: Logical vs Physical Addresses Simplest memory management unit: adds base register to all logical addresses to generate physical memory addresses Limit CPU logical address < + physical address Memory trap Base MMU 2

3 Static Linking System libraries are treated like any other object files, and linked into the binary executable by the linker. Advantages: Simple. No possibility for version mismatch between binary program and library. Disadvantages: Large binaries on disk and in memory. Every program using a library needs its own copy of the library code. Dynamic Linking and Shared Libraries Linking with dynamic libraries is delayed until execution time. Program only includes a stub for each library routine. When stub is first executed, the routine is loaded into memory, and the stub is replaced with the address of the routine. Advantages: Only one copy of library needed on disk. OS can allow processes to share text segment of libraries, so only one copy need in memory at a time. Disadvantages: Versioning can be an issue - handled reasonably well on Unix using version numbering. 3

4 Memory Management Ideally programmers want memory that is Large Fast Non volatile Memory hierarchy Small amount of fast, expensive memory cache Some medium-speed, medium price main memory Gigabytes of slow, cheap disk storage Memory manager handles the memory hierarchy Memory Management What can the OS do if there isn t enough main memory to hold all the processes that want to execute? Two approaches: Swapping Virtual Memory 4

5 Swapping A process must be in memory to be executed. When it s not being executed, it can be temporarily swapped out to disk. Goal is to maintain enough processes in memory to keep the CPU utilized effectively. Not to swap too often - execution time must be long relative to swap time. Need to be certain no I/O operations in progress at time swap out occurs. Swapping and Relocation Memory allocation changes as processes move into and out of memory. Shaded regions are unused memory 5

6 Swapping: How much space to allocate? Allocating space for growing data segment Allocating space for growing stack & data segment Keeping Track of Memory (a) Part of memory with 5 processes, 3 holes tick marks show allocation units shaded regions are free (b) Corresponding bit map (c) Same information as a list 6

7 Memory Management with Linked Lists Merging free-list segments is easy with a linked list: Four neighbor combinations for the terminating process X Allocation Algorithms First Fit Scan the free list until the first hole is found that is large enough to fit the process to be swapped in. Best Fit Scan the entire list and find the smallest hole that will fit the process to be swapped in. Problem: leaves lots of small unusable holes. Worst Fit Use the largest available hole. Shown to be worse than best fit or first fit. 7

8 Fragmentation All allocation algorithms result in holes being left that are not usable because they re not big enough for any process. Typically as much as 1/3 of memory may be unusable. Solutions: Compaction Only possibly with dynamic relocation Slow, expensive. Non-contiguous memory allocation Paging, Segmentation. Virtual Memory Combined size of a program, its stack, and data may exceed available physical memory. Only the parts currently needed are kept in real memory The rest are kept on disk. On a multiprogramming system, parts of many processes are in memory at once. When part of one process is being fetched from disk, another is running on the CPU. 8

9 Virtual Memory: Role of the MMU Virtual address space is larger than the physical memory. When an address with no mapping to physical memory is accessed, the MMU generates a trap. OS handles the trap to load the relevant virtual memory from disk Paging Virtual Memory is divided up into pages. Physical Memory is divided into page frames, which hold pages when they re loaded into memory. The page table holds the mapping between the two. 9

10 Paging When an instruction is executed that references memory: MOV REG, 0 Virtual address 0 is sent to MMU. MMU figures out which page this maps to, and then which page frame holds that page, and adds the appropriate offset to create a physical memory address. If the page is not currently in memory, a page fault trap is generated, and the OS must: Pick a little-used frame, Save its contents to disk, load the missing page into it. Inside the MMU Example: 16 bit virtual addresses 4-bit page number 12-bit offset => 16 pages 4Kbytes per page Present/absent bit indicates if a page is in memory 10

11 Summary Relocation: Mapping program addresses to physical memory addresses. Dynamic Libraries Swapping Moving processes to and from disk. Fragmentation Virtual Memory Paging Next Time: Paging Continued. 11