Linux Kernel Architecture

Chapter 1: Introduction and Overview

TypeDefinitions

The kernel usestypedefto define various data types in order to make itself independent of architecture-

specific features because of the different bit lengths for standard data types on individual processors.

The definitions have names such assector_t(to specify a sector number on a block device),pid_t(to

indicate a process identifier), and so on, and are defined by the kernel in architecture-specific code in such

a way as to ensure that they represent the applicable value range. Because it is not usually important to

know on which fundamental data types the definitions are based, and for simplicity’s sake, I do not

always discuss the exact definitions of data types in the following chapters. Instead, I use them without

further explanation — after all, they are simply non-compound standard data types under a different

name.

typedef’d variables must not be accessed directly, but only via auxiliary functions

that I introduce when we encounter the type. This ensures that they are properly

manipulated, although the type definition is transparent to the user.

At certain points, the kernel must make use of variables with an exact, clearly defined number of bits —

for example, when data structures need to be stored on hard disk. To allow data to be exchanged between

various systems (e.g., on USB sticks), the same external format must always be used, regardless of how

data are represented internally in the computer.

To this end, the kernel defines several integer data types that not only indicate explicitly whether they

are signed or unsigned, but also specify theexactnumber of bits they comprise.__s8and__u8are, for

example, 8-bit integers that are either signed (__s8) or unsigned (__u8).__u16and__s16,__u32and

__s32,and__u64and__s64are defined in the same way.

Byte Order

To represent numbers, modern computers use either thebig endianorlittle endianformat. The format

indicates how multibyte data types are stored. With big endian ordering, the most significant byte is

stored at the lowest address and the significance of the bytes decreases as the addresses increase. With

little endian ordering, the least significant byte is stored at the lowest address and the significance of

the bytes increases as the addresses increase (somearchitectures such as MIPS support both variants).

Figure 1-13 illustrates the issue.

0–7

0–7

0–7 8–15

8–15

16–23 24–31

Little

endian

char

short

int

char

short

int

Byte 0 1 2 3

0–7

8–15

24–31 16–23

0–7

8–15 0–7

Big

endian

Figure 1-13: Composition of elementary data

types depending on the endianness of the

underlying architecture.