slabmalloc

Simple slab based malloc implementation in rust, in order to provide the necessary interface to rusts liballoc library. slabmalloc only relies on libcore and is designed to be used in kernel level code as the only interface a client needs to provide is the necessary mechanism to allocate and free 4KiB frames (or any other default page-size on non-x86 hardware).

Build

By default this library should compile with nightly and stable versions of the Rust compiler.

Add the following line to the Cargo.toml dependencies:

slabmalloc = <version>

Usage

The slabmalloc API is designed to satisfy the rust liballoc low-level memory allocation interface:

#[global_allocator]
static MEM_PROVIDER: slabmalloc::SafeZoneAllocator = slabmalloc::SafeZoneAllocator::new();

• Use the ZoneAllocator to allocate arbitrary sized objects:

let object_size = 12;
let alignment = 4;
let layout = Layout::from_size_align(object_size, alignment).unwrap();

let mut mmap = MmapPageProvider::new();
let page = mmap.allocate_page().expect("Can't allocate a page");

let mut zone: ZoneAllocator = Default::default();
unsafe { zone.refill(layout, page)? };

let allocated = zone.allocate(layout)?;
zone.deallocate(allocated, layout)?;

• Use the SCAllocator to allocate fixed sized objects:

let object_size = 10;
let alignment = 8;
let layout = Layout::from_size_align(object_size, alignment).unwrap();

let mut mmap = MmapPageProvider::new();
let page = mmap.allocate_page().expect("Can't allocate a page");

let mut sa: SCAllocator = SCAllocator::new(object_size);
unsafe { sa.refill(page) };

sa.allocate(layout)?;

Performance

slabmalloc is optimized for single-threaded, fixed-size object allocations. For anything else it will probably perform poorly (for example if your workload does lots of reallocations, or if the allocator needs to scale to many cores).

At least on my system, it outperforms jemalloc in (silly) benchmarks:

test tests::jemalloc_allocate_deallocate       ... bench:          76 ns/iter (+/- 5)
test tests::jemalloc_allocate_deallocate_big   ... bench:         119 ns/iter (+/- 24)
test tests::slabmalloc_allocate_deallocate     ... bench:          38 ns/iter (+/- 8)
test tests::slabmalloc_allocate_deallocate_big ... bench:          38 ns/iter (+/- 11)

Documentation

• API Documentation