How does posix_memalign differ from mmap

Question

How does posix_memalign with alignment size of 4096 bytes differ from mmap? Does it internally use mmap or some other mechanism?

Answer 1

From the standards (POSIX) point of view, posix_memalign() doesn’t really have any connection to mmap():

posix_memalign - aligned memory allocation

The posix_memalign() function shall allocate size bytes aligned on a boundary specified by alignment, and shall return a pointer to the allocated memory in memptr.

mmap - map pages of memory

The mmap() function shall establish a mapping between an address space of a process and a memory object.

The mmap() function shall be supported for the following memory objects:

• Regular files
• [SHM] > Shared memory objects <
• [TYM] > Typed memory objects <

Support for any other type of file is unspecified.

In particular, POSIX does not specify the MAP_ANONYMOUS flag that can be used (e. g. on Linux) to use mmap() to allocate memory without a corresponding file. So, a priori, the two functions perform unrelated tasks: posix_memalign() allocates dynamic memory (which POSIX mmap() cannot do), while mmap() maps files into the process address space (which posix_memalign() cannot do, as stated in jørgensen’s answer).

When talking about systems which support MAP_ANONYMOUS, such as Linux (which was tagged in the question), mmap() gains a capability similar to posix_memalign(), although mmap() will return a pointer aligned to the system page size, while posix_memalign() can be used to obtain memory with a different alignment as well.

Finally, memory allocated with posix_memalign() is freed using free(), while mmap() is cleaned up using munmap().

Answer 2

How does posix_memalign with alignment size of 4096 bytes differ from mmap?

The obvious answer: posix_memalign cannot mmap arbitrary files by fd. :)

As for allocation behavior of posix_memalign vs. mmap(MAP_ANOYMOUS): I see no requirement that posix_memalign(size) has to use mmap. It could as well share the sbrk mechanism with/from malloc and return you a properly-aligned pointer to part of the brk area.

Answer 3

posix_memalign is a higher-level API than mmap, designed to interoperate with malloc, free and realloc. mmap usage is more complicated because it offers more functionality than posix_memalign (mapping files into a process's address space). How it is implemented (in terms of mmap or otherwise) is left unspecified by the POSIX standard.

Use posix_memalign where you'd use malloc if you didn't have alignment restrictions.

Answer 4

Where mmap is available, posix_memalign is usually implemented using mmap. The main difference is that posix_memalign is in stdlib.h, where as mmap is a system call which might not be available and have different semantics on different platforms.