| Crates.io | portable-dlmalloc |
| lib.rs | portable-dlmalloc |
| version | 0.4.0 |
| created_at | 2024-11-20 16:40:47.753034+00 |
| updated_at | 2025-11-01 06:59:00.819751+00 |
| description | Portable Fork of Doug Lea's malloc Implementation. |
| homepage | https://github.com/Zero-Tang/portable-dlmalloc |
| repository | https://github.com/Zero-Tang/portable-dlmalloc |
| max_upload_size | |
| id | 1455022 |
| size | 270,433 |
Portable Fork of Doug Lea's malloc Implementation for Rust.
This code is originally implemented by Doug Lea. The original source code is no longer available from the FTP URL listed in the website, but you can still find it through Wayback Machine.
You may use this crate to help you make a portable global allocator.
You will have to implement the eight C functions as described in Port To Your Platform chapter.
dlmalloc guarantees the alignment of allocation in comparison to just some wrappers on malloc functions (e.g.: wrapping HeapAlloc in Windows). If your structure is defined to be aligned on a big alignment (e.g.: 1024 bytes), this allocator guarantees the returned pointer if aligned on your specific boundary. The minimum alignment of dlmalloc is four times of the pointer size. (e.g.: 32 bytes on 64-bit platform.)
To use this crate as your global allocator:
use portable_dlmalloc::DLMalloc;
#[global_alloactor] static GLOBAL_ALLOCATOR:DLMalloc=DLMalloc;
Then you will be able to use alloc crate.
extern crate alloc;
The default alignment of alloc trait method automatically determines the required alignment.
Your custom_mmap implementation must track all allocated pages so that you can release pages in shared address-space (e.g.: DLL in Windows, SO in Linux).
The final_lock routine will not be called for the global allocator! Your init_lock implementation must track all initialized locks so that you can finalize all locks, unless it is trivial to finalize the locks.
You can also use MspaceAlloc as your global allocator:
use portable_dlmalloc:MspaceAlloc;
#[global_allocator] static GLOBAL_ALLOCATOR:MspaceAlloc=MspaceAlloc::new(0);
To destroy this allocator:
unsafe
{
GLOBAL_ALLOCATOR.destroy();
}
Use this allocator only if:
The Allocator Trait is currently nightly-only. You are required to use a nightly rust compiler in order to use this feature.
To use alternate alloactor, you will have to declare that your crate uses allocator_api:
#![feature(allocator_api)]
You also need to enable alt-alloc in Cargo.toml section:
[dependencies.portable-dlmalloc]
version = "0.3.3"
features = ["alt-alloc"]
To use alternate allocator, you need to create an allocator using a reference to AltAlloc:
use portable_dlmalloc::AltAlloc;
let a=AltAlloc::new(0,false);
let ap:Box::<u32,&AltAlloc>=Box::new_in(123,&a);
println!("{:p} | {}",&raw const *ap,ap);
Please note that alternate allocator is a nightly-only API. If Rust removes this feature in future, this feature will be removed from this crate as well.
Caveat: You must use the allocator in references, as the AltAlloc does not implement Copy trait!
The MspaceAlloc allocator can also be used for custom allocator.
Since the new method is const fn, you can declare a custom allocator as a static variable without using something like lazy_static. However, the Drop trait of MspaceAlloc won't be called if such allocator is declared as static variable!
Caveat: You must use the allocator in references, as the MspaceAlloc does not implement Copy trait!
The raw module from this crate exports FFI bindings for dlmalloc library.
use portable_dlmalloc::raw::*;
For example, you may use dlmallopt to adjust mmap granularity (default is 2MiB in Rust crate):
dlmallopt(M_GRANULARITY,0x20000); // Change `mmap` granularity to 128KiB.
You may use dlpvalloc to allocate memory on page-granularity.
let p=dlpvalloc(12345);
assert_eq!(p as usize & 0xfff,0);
Warning: dlpvalloc - as well as other routines that allocate memories with higher granularities - may cause serious memory fragmentation if you overrely on them.
// Assume 4096 is page size.
let p=dlpvalloc(4096) as usize;
let q=dlpvalloc(4096) as usize;
// Consecutive allocations do not guarantee them to be adjacent.
assert_eq!(q-p,8192);
To port dlmalloc to your platform, implement the following procedures:
custom_mmap/custom_munmap: Allocate and free pages from the system. mmap should return (void*)-1 to indicate failure instead of NULL. munmap should return 0 to indicate success, and -1 to indicate failure.
#[no_mangle] unsafe extern "C" custom_mmap(length:usize)->*mut c_void;
#[no_mangle] unsafe extern "C" custom_munmap(ptr:*mut c_void,length:usize)->i32;
Hint: If you are in baremetal environment without mmap-like services like outdated embedded systems, you can just return the pointer to the free memory. sbrk, even though original implementation of dlmalloc supports it. Just emulate the behavior of mmap with sbrk.init_lock/final_lock/acquire_lock/release_lock: Implement thread-safety for dlmalloc. The minimal implementation can be a simple spinlock. You can leave the implementations empty for this set of routines if you do not need thread-safety. lock depends on your implementation. It can be *mut T where T can be anything that has the size of a pointer.
#[no_mangle] unsafe extern "C" init_lock(lock:*mut *mut c_void); // Initialize the mutex.
#[no_mangle] unsafe extern "C" final_lock(lock:*mut *mut c_void); // Finalize the mutex.
#[no_mangle] unsafe extern "C" acquire_lock(lock:*mut *mut c_void); // Acquire the mutex.
#[no_mangle] unsafe extern "C" release_lock(lock:*mut *mut c_void); // Release the mutex.
custom_abort: Implement abort() routine. dlmalloc calls custom_abort() when internal assertion fails. You may use panic here.
#[no_mangle] unsafe extern "C" custom_abort(message:*const u8,src_file:*const u8,src_line:u32)->!;
Note that message and src_file are null-terminated strings. The encodings of the string is implementation-specific of the C compiler. It is your duty to convert them into UTF-8 encoding. memcpy/memset: I suppose no explanations are needed for these two. dlmalloc uses these two routines, but they can be easily implemented anyway. You do not need to implement these two routines in Rust if your linker can find libraries that implement these two routines. Note that MSVC SDK provides source code of high-performance memcpy and memset implementations in Assembly!Note: If you are using Rust 2024 or higher, you must use #[unsafe(no_mangle)] as prefix! See Rust unsafe attributes for more details.
If, for some reasons, these procedure names must be reserved in your project, you may use the export_name attribute. Note that export_name attribute requires unsafe in Rust 2024 as well as no_mangle!
Since the core of the dlmalloc library is written in C, a working C compiler is required.
If your target is somewhat unorthodox, you need to set environment variables before executing cargo build:
CC: This environment variable specifies which compiler executable should be used to compile malloc.c.AR: This environment variable specifies which archiver executable should be used to archive this crate into a static library.CFLAGS: This environment variable specifies additional flags to the compiler. You might need this flag to add debug information (e.g.: -g). In kernel-mode with MSVC toolchain, you might need /GS- flag.If cc crate does not know how to invoke your compiler and/or archiver, you should write a script to emulate cc and/or ar.
In most circumstances, setting CC to clang and AR to llvm-ar should work well.
This crate is under the MIT license.