Chapter 14. The Process Address Space

Chapter 11, "Memory Management," looked at how the kernel manages physical memory. In addition to managing its own memory, the kernel also has to manage the process address spacethe representation of memory given to each user-space process on the system. Linux is a virtual memory operating system, and thus the resource of memory is virtualized among the processes on the system. To an individual process, the view is as if it alone has full access to the system's physical memory. More importantly, the address space of even a single process can be much larger than physical memory. This chapter discusses how the kernel manages the process address space.

The process address space consists of the linear address range presented to each process and, more importantly, the addresses within this space that the process is allowed to use. Each process is given a flat 32- or 64-bit address space, with the size depending on the architecture. The term "flat" describes the fact that the address space exists in a single range. (As an example, a 32-bit address space extends from the address 0 to 429496729.) Some operating systems provide a segmented address space, with addresses existing not in a single linear range, but instead in multiple segments. Modern virtual memory operating systems generally have a flat memory model and not a segmented one. Normally, this flat address space is unique to each process. A memory address in one process's address space tells nothing of that memory address in another process's address space. Both processes can have different data at the same address in their respective address spaces. Alternatively, processes can elect to share their address space with other processes. We know these processes as threads.

A memory address is a given value within the address space, such as 4021f000. This particular value identifies a specific byte in a process's 32-bit address space. The interesting part of the address space is the intervals of memory addresses, such as 08048000-0804c000, that the process has permission to access. These intervals of legal addresses are called memory areas. The process, through the kernel, can dynamically add and remove memory areas to its address space.

The process can access a memory address only in a valid memory area. Memory areas have associated permissions, such as readable, writable, and executable, that the associated process must respect. If a process accesses a memory address not in a valid memory area, or if it accesses a valid area in an invalid manner, the kernel kills the process with the dreaded "Segmentation Fault" message.

Memory areas can contain all sorts of goodies, such as

• A memory map of the executable file's code, called the text section

• A memory map of the executable file's initialized global variables, called the data section

• A memory map of the zero page (a page consisting of all zeros, used for purposes such as this) containing uninitialized global variables, called the bss section^[1]

  ^[1] The term "BSS" is historical and quite old. It stands for block started by symbol. Uninitialized variables are not stored in the executable object because they do not have any associated value. But the C standard decrees that uninitialized global variables are assigned certain default values (basically, all zeros), so the kernel loads the variables (without value) from the executable into memory and maps the zero page over the area, thereby giving the variables the value zero, without having to waste space in the object file with explicit initializations.

• A memory map of the zero page used for the process's user-space stack (do not confuse this with the process's kernel stack, which is separate and maintained and used by the kernel)

• An additional text, data, and bss section for each shared library, such as the C library and dynamic linker, loaded into the process's address space

• Any memory mapped files

• Any shared memory segments

• Any anonymous memory mappings, such as those associated with malloc()^[2]

  ^[2] Newer versions of glibc implement malloc() via mmap(), in addition to brk().

All valid addresses in the process address space exist in exactly one area; memory areas do not overlap. As you can see, there is a separate memory area for each different chunk of memory in a running process: the stack, the object code, global variables, mapped file, and so on.