Linux Memory Management

Question 1

What memory segments conform a linux process, and what are their purposes

Answer 1

The text segment: program executable code (instructions), the data segment: storage for program variables, and the stack segment: stores the call stack of the current program.

Question 2

What system call allows a process to set the size of its data segment?

Answer 2

The brk system call, and it’s used by the malloc C library.

Question 3

How does the stack of a process grow?

Answer 3

If the stack pointer is moved below the stack segment of a process a hardware fault occurs and the operating system lowers the bottom of the stack segment by one page

Question 4

What does the stackc of a process contain initially?

Answer 4

It contains all the environment variables as well as the command line typed to the shell to invoke it.

Question 5

How does linux manage the text segment of the same program running in different processes?

Answer 5

It supports shared text segments, were two process share the same text segment.

Question 6

What is the purpose of memory-mapped files?

Answer 6

It maps a file onto a portion of a process’ address space so that the file can be read and written to as if it were a byte array in memory, making accesses much easier than I/O system calls.

Question 7

What API allows creating memory-mapped files and what are the parameters?

Answer 7

Memory-mapped files are created using the mmap and unmmap syscalls, the parameters are addr: the address at which the file is mapped , len: nof bytes to map, prot: protection mode and flags: whether it’s sharable or not.

Question 8

What’s the typical size of a linux process’ address space?

Answer 8

Each process typically gets 3GB for its virtual address space and 1GB reserved for its page tables adn other kernel data.

Question 9

How is physical memory split in a linux system?

Answer 9

It’s split into three zones, DMA (Direct Memory Access), NORMAL which is regularly paged, and HIGH_MEM, not permanently mapped.

Question 10

What information does the zone descriptor contain?

Answer 10

It contains memory utilization for each zone, number of active or inactive pages, low and high watermarks, and an array of free areas.

Question 11

Describe the paging scheme used by linux

Answer 11

It’s a four-level paging scheme. Each virtual address is broken up into five fields: directory fields are used as indexes into the appropiate page directories of which one is a private for each process.

Question 12

When and what parts of the kernel paged out?

Answer 12

No part of the kernel is paged out ever.

Question 13

What’s the name of the algorithm used for memory allocation?

Answer 13

Buddy algorithm

Question 14

Describe how the memory allocation algorithm works

Answer 14

Initially memory consists of a single contingous piece. When a request for memory comes in, it’s rounded up to a power of 2, and the available memory memory is split half by half until an appropriate size is matched

Question 15

What’s a problem with the buddy algorithm and how is it solved?

Answer 15

It leads to considerable internal fragmentation because allocation requests off by just a little above the closest power of two are doubled up. To alliviate the slab allocator is used.

Question 16

Describe how the slab allocator works

Answer 16

It takes chunks using the buddy algorithm but carves slabs (smaller units) from the allocated halves and manages them separately.

Question 17

What data structure describes the properties of a virtual memory area?

Answer 17

The vm_area_struct

Question 18

How is the copy on write mechanism implemented under the hood?

Answer 18

Memory areas (vm_area) are marked as read/write but copied memory pages are only marked as read. When the page is written to a page, a protection fault occurs and the kernel gives the process a copy of the page and marks it as read/write.

Question 19

What data structure holds information about all vm areas?

Answer 19

It’s called the mm_struct

Question 20

What is a memory page frame?

Answer 20

It’s a physical memory location the same size of a page, where a page is loaded into when it’s paged in.

Question 21

Where does the paging logic reside?

Answer 21

Partly in the kernel and partly in the page daemon (process 2). The page daemon runs periodically.

Question 22

Where is the text, stack, data and heap of a program paged out into?

Answer 22

They are paged out into their respective files on disk. Everything else is paged to the paging partition if present or one of the fixed-length paging files (swap area).

Question 23

Why is paging to a separate partition more efficient than paging to a paging file?

Answer 23

1. There is no mapping between file blocks and disk blocks, physical writes can be of any size, not just the file block size, and the page is always written contigously to disk; with a paging file that may or may not be the case.

Question 24

What types of pages does the PFRA distinguish? Explain their differences.

Answer 24

Unreclaimable: locked or reserved pages (never paged out). Swappable: must be wrotten bacck to swap area before being reclaimed. Syncable must be written back to disk if marked as dirty. Discardable: pages that can be reclaimed immediately.

Question 25

What challenge do shared pages impose in the memory management system?

Answer 25

All page tables of all address spaces sharing the page must be updated synchronously.

Question 26

Define the term swappiness of the system.

Answer 26

The ratio of pages with backing store vs pages which need to be swapped out selected during PFRA.

Question 27

How many states can a page be in for the PFR Algorithm and which flags encode these states?

Answer 27

4 states, encoded by PG_active, and PG_referenced