Buddy Memory Allocator

This allocator hasn’t been benchmarked at all and has barely been tested. Use at your own risk!

The file buddy-malloc.c implements a buddy memory allocator, which is an allocator that allocates memory within a fixed linear address range. It spans the address range with a binary tree that tracks free space. Both “malloc” and “free” are O(log N) time where N is the maximum possible number of allocations.

The “buddy” term comes from how the tree is used. When memory is allocated, nodes in the tree are split recursively until a node of the appropriate size is reached. Every split results in two child nodes, each of which is the buddy of the other. When a node is freed, the node and its buddy can be merged if the buddy is also free. This makes the memory available for larger allocations again. Wikipedia has more information.

I wrote this because I needed a simple allocator for a side project (a compiler for WebAssembly) and I didn’t feel like attempting to port a standard memory allocator (they are many thousands of lines long and are really complicated). I found a few other smaller allocators but ended up not being able to use them. For example, one of the simplest allocators is the one from K&R C book, but it’s O(N) time which is unacceptably slow.

The code uses the Linux kernel as inspiration in a few places. One is the use of circular doubly-linked lists to track free memory blocks. Another trick is using a single bit per node to store the state of the node, which is described in detail here. This allocator uses the “brk” syscall to request more memory from the kernel. While archaic, this method is a good fit for WebAssembly’s linear memory model.

This code is available under the MIT license.