|
| 1 | +% Custom Allocators |
| 2 | + |
| 3 | +Allocating memory isn't always the easiest thing to do, and while Rust generally |
| 4 | +takes care of this by default it often becomes necessary to customize how |
| 5 | +allocation occurs. The compiler and standard library currently allow switching |
| 6 | +out the default global allocator in use at compile time. The design is currently |
| 7 | +spelled out in [RFC 1183][rfc] but this will walk you through how to get your |
| 8 | +own allocator up and running. |
| 9 | + |
| 10 | +[rfc]: https://github.com/rust-lang/rfcs/blob/master/text/1183-swap-out-jemalloc.md |
| 11 | + |
| 12 | +# Default Allocator |
| 13 | + |
| 14 | +The compiler currently ships two default allocators: `alloc_system` and |
| 15 | +`alloc_jemalloc` (some targets don't have jemalloc, however). These allocators |
| 16 | +are just normal Rust crates and contain an implementation of the routines to |
| 17 | +allocate and deallocate memory. The standard library is not compiled assuming |
| 18 | +either one, and the compiler will decide which allocator is in use at |
| 19 | +compile-time depending on the type of output artifact being produced. |
| 20 | + |
| 21 | +Binaries generated by the compiler will use `alloc_jemalloc` by default (where |
| 22 | +available). In this situation the compiler "controls the world" in the sense of |
| 23 | +it has power over the final link. Primarily this means that the allocator |
| 24 | +decision can be left up the compiler. |
| 25 | + |
| 26 | +Dynamic and static libraries, however, will use `alloc_system` by default. Here |
| 27 | +Rust is typically a 'guest' in another application or another world where it |
| 28 | +cannot authoritatively decide what allocator is in use. As a result it resorts |
| 29 | +back to the standard APIs (e.g. `malloc` and `free`) for acquiring and releasing |
| 30 | +memory. |
| 31 | + |
| 32 | +# Switching Allocators |
| 33 | + |
| 34 | +Although the compiler's default choices may work most of the time, it's often |
| 35 | +necessary to tweak certain aspects. Overriding the compiler's decision about |
| 36 | +which allocator is in use is done simply by linking to the desired allocator: |
| 37 | + |
| 38 | +```rust,no_run |
| 39 | +#![feature(alloc_system)] |
| 40 | +
|
| 41 | +extern crate alloc_system; |
| 42 | +
|
| 43 | +fn main() { |
| 44 | + let a = Box::new(4); // allocates from the system allocator |
| 45 | + println!("{}", a); |
| 46 | +} |
| 47 | +``` |
| 48 | + |
| 49 | +In this example the binary generated will not link to jemalloc by default but |
| 50 | +instead use the system allocator. Conversely to generate a dynamic library which |
| 51 | +uses jemalloc by default one would write: |
| 52 | + |
| 53 | +```rust,no_run |
| 54 | +#![feature(alloc_jemalloc)] |
| 55 | +#![crate_type = "dylib"] |
| 56 | +
|
| 57 | +extern crate alloc_jemalloc; |
| 58 | +
|
| 59 | +pub fn foo() { |
| 60 | + let a = Box::new(4); // allocates from jemalloc |
| 61 | + println!("{}", a); |
| 62 | +} |
| 63 | +# fn main() {} |
| 64 | +``` |
| 65 | + |
| 66 | +# Writing a custom allocator |
| 67 | + |
| 68 | +Sometimes even the choices of jemalloc vs the system allocator aren't enough and |
| 69 | +an entirely new custom allocator is required. In this you'll write your own |
| 70 | +crate which implements the allocator API (e.g. the same as `alloc_system` or |
| 71 | +`alloc_jemalloc`). As an example, let's take a look at a simplified and |
| 72 | +annotated version of `alloc_system` |
| 73 | + |
| 74 | +```rust,no_run |
| 75 | +# // only needed for rustdoc --test down below |
| 76 | +# #![feature(lang_items)] |
| 77 | +// The compiler needs to be instructed that this crate is an allocator in order |
| 78 | +// to realize that when this is linked in another allocator like jemalloc should |
| 79 | +// not be linked in |
| 80 | +#![feature(allocator)] |
| 81 | +#![allocator] |
| 82 | +
|
| 83 | +// Allocators are not allowed to depend on the standard library which in turn |
| 84 | +// requires an allocator in order to avoid circular dependencies. This crate, |
| 85 | +// however, can use all of libcore. |
| 86 | +#![feature(no_std)] |
| 87 | +#![no_std] |
| 88 | +
|
| 89 | +// Let's give a unique name to our custom allocator |
| 90 | +#![crate_name = "my_allocator"] |
| 91 | +#![crate_type = "rlib"] |
| 92 | +
|
| 93 | +// Our system allocator will use the in-tree libc crate for FFI bindings. Note |
| 94 | +// that currently the external (crates.io) libc cannot be used because it links |
| 95 | +// to the standard library (e.g. `#![no_std]` isn't stable yet), so that's why |
| 96 | +// this specifically requires the in-tree version. |
| 97 | +#![feature(libc)] |
| 98 | +extern crate libc; |
| 99 | +
|
| 100 | +// Listed below are the five allocation functions currently required by custom |
| 101 | +// allocators. Their signatures and symbol names are not currently typechecked |
| 102 | +// by the compiler, but this is a future extension and are required to match |
| 103 | +// what is found below. |
| 104 | +// |
| 105 | +// Note that the standard `malloc` and `realloc` functions do not provide a way |
| 106 | +// to communicate alignment so this implementation would need to be improved |
| 107 | +// with respect to alignment in that aspect. |
| 108 | +
|
| 109 | +#[no_mangle] |
| 110 | +pub extern fn __rust_allocate(size: usize, _align: usize) -> *mut u8 { |
| 111 | + unsafe { libc::malloc(size as libc::size_t) as *mut u8 } |
| 112 | +} |
| 113 | +
|
| 114 | +#[no_mangle] |
| 115 | +pub extern fn __rust_deallocate(ptr: *mut u8, _old_size: usize, _align: usize) { |
| 116 | + unsafe { libc::free(ptr as *mut libc::c_void) } |
| 117 | +} |
| 118 | +
|
| 119 | +#[no_mangle] |
| 120 | +pub extern fn __rust_reallocate(ptr: *mut u8, _old_size: usize, size: usize, |
| 121 | + _align: usize) -> *mut u8 { |
| 122 | + unsafe { |
| 123 | + libc::realloc(ptr as *mut libc::c_void, size as libc::size_t) as *mut u8 |
| 124 | + } |
| 125 | +} |
| 126 | +
|
| 127 | +#[no_mangle] |
| 128 | +pub extern fn __rust_reallocate_inplace(_ptr: *mut u8, old_size: usize, |
| 129 | + _size: usize, _align: usize) -> usize { |
| 130 | + old_size // this api is not supported by libc |
| 131 | +} |
| 132 | +
|
| 133 | +#[no_mangle] |
| 134 | +pub extern fn __rust_usable_size(size: usize, _align: usize) -> usize { |
| 135 | + size |
| 136 | +} |
| 137 | +
|
| 138 | +# // just needed to get rustdoc to test this |
| 139 | +# fn main() {} |
| 140 | +# #[lang = "panic_fmt"] fn panic_fmt() {} |
| 141 | +# #[lang = "eh_personality"] fn eh_personality() {} |
| 142 | +``` |
| 143 | + |
| 144 | +After we compile this crate, it can be used as follows: |
| 145 | + |
| 146 | +```rust,ignore |
| 147 | +extern crate my_allocator; |
| 148 | +
|
| 149 | +fn main() { |
| 150 | + let a = Box::new(8); // allocates memory via our custom allocator crate |
| 151 | + println!("{}", a); |
| 152 | +} |
| 153 | +``` |
| 154 | + |
| 155 | +# Custom allocator limitations |
| 156 | + |
| 157 | +There are a few restrictions when working with custom allocators which may cause |
| 158 | +compiler errors: |
| 159 | + |
| 160 | +* Any one artifact may only be linked to at most one allocator. Binaries, |
| 161 | + dylibs, and staticlibs must link to exactly one allocator, and if none have |
| 162 | + been explicitly chosen the compiler will choose one. On the other than rlibs |
| 163 | + do not need to link to an allocator (but still can). |
| 164 | + |
| 165 | +* A consumer of an allocator is tagged with `#![needs_allocator]` (e.g. the |
| 166 | + `liballoc` crate currently) and an `#[allocator]` crate cannot transitively |
| 167 | + depend on a crate which needs an allocator (e.g. circular dependencies are not |
| 168 | + allowed). This basically means that allocators must restrict themselves to |
| 169 | + libcore currently. |
0 commit comments