Memory Management

Abstract

Memory management in the kernel is significantly more complex than it is for a user space program. A user space program typically deals with a flat linear address space and can allocate memory in more or less arbitrary blocks without worrying about the source or arrangement of this memory. It has simple interfaces that typically take a size in bytes as an argument and deliver a yay or nay result, depending on the availability of the requested memory. At worst, the consequence of a failed allocation or misuse of memory is the termination of the offending process. However, things are not as straightforward in the kernel. The kernel has to deal with multiple memory spaces, including its own, as well as the mapping of memory between those memory spaces and physical memory. While user space programs deal with virtual memory, where the underlying physical arrangement is irrelevant, the kernel often needs to know whether the memory is contiguous and where it is located. This is because some hardware devices are unable to read from certain memory addresses or have specific requirements regarding the alignment of the memory, for example, because it can read only from memory that has been aligned to a 16-byte boundary or because it cannot read from addresses higher than 32-bit. However, the most obvious challenge of kernel memory management is to use as little as possible because it is a scarce resource, especially for embedded devices such as the iPhone or iPad. Incorrect use of memory in the kernel can lead to subtle and not so subtle consequences.