Talc Allocator

_{If you find Talc useful, please consider leaving tip via Paypal}

What is this for?

• Embedded systems, OS kernels, and other no_std environments
• WebAssembly apps, as a drop-in replacement for the default allocator
• Subsystems in normal programs that need especially quick arena allocation

Why Talc?

• Performance is the primary focus, while retaining generality
• Custom Out-Of-Memory handlers for just-in-time heap management and recovery
• Supports creating and resizing arbitrarily many heaps
• Optional allocation statistics
• Partial validation with debug assertions enabled
• Verified with MIRI

Why not Talc?

• Doesn't integrate with operating systems' dynamic memory facilities out-of-the-box yet
• Doesn't scale well to allocation/deallocation-heavy concurrent processing
  • Though it's particularly good at concurrent reallocation.

Table of Contents

Targeting WebAssembly? You can find WASM-specific usage and benchmarks here.

• Setup
• Benchmarks
• General Usage
• Advanced Usage
• Conditional Features
• Stable Rust and MSRV
• Algorithm
• Future Development
• Changelog

Setup

As a global allocator:

use talc::*;

static mut ARENA: [u8; 10000] = [0; 10000];

#[global_allocator]
static ALLOCATOR: Talck<spin::Mutex<()>, ClaimOnOom> = Talc::new(unsafe {
    // if we're in a hosted environment, the Rust runtime may allocate before
    // main() is called, so we need to initialize the arena automatically
    ClaimOnOom::new(Span::from_const_array(core::ptr::addr_of!(ARENA)))
}).lock();

fn main() {
    let mut vec = Vec::with_capacity(100);
    vec.extend(0..300usize);
}

Or use it as an arena allocator via the Allocator API with spin as follows:

#![feature(allocator_api)]
use talc::*;
use core::alloc::{Allocator, Layout};

static mut ARENA: [u8; 10000] = [0; 10000];

fn main () {
    let talck = Talc::new(ErrOnOom).lock::<spin::Mutex<()>>();
    unsafe { talck.lock().claim(ARENA.as_mut().into()); }
    
    talck.allocate(Layout::new::<[u32; 16]>());
}

Note that while the spin crate's mutexes are used here, any lock implementing lock_api works.

See General Usage and Advanced Usage for more details.

Benchmarks

Heap Efficiency Benchmark Results

The average occupied capacity upon first allocation failure when randomly allocating/deallocating/reallocating.

Random Actions Benchmark

The number of successful allocations, deallocations, and reallocations within the allotted time.

1 Thread

4 Threads

Allocations & Deallocations Microbenchmark

Label indicates the maximum within 50 standard deviations from the median. Max allocation size is 0x10000.

General Usage

Here is the list of important Talc methods:

• Constructors:
  • new
• Information:
  • get_allocated_span - returns the minimum heap span containing all allocated memory in an established heap
  • get_counters - if feature "counters" is enabled, this returns a struct with allocation statistics
• Management:
  • claim - claim memory to establishing a new heap
  • extend - extend an established heap
  • truncate - reduce the extent of an established heap
  • lock - wraps the Talc in a Talck, which supports the GlobalAlloc and Allocator APIs
• Allocation:
  • malloc
  • free
  • grow
  • grow_in_place
  • shrink

Read their documentation for more info.

Span is a handy little type for describing memory regions, as trying to manipulate Range<*mut u8> or *mut [u8] or base_ptr-size pairs tends to be inconvenient or annoying.

Advanced Usage

The most powerful feature of the allocator is that it has a modular OOM handling system, allowing you to fail out of or recover from allocation failure easily.

Provided OomHandler implementations include:

• ErrOnOom: allocations fail on OOM
• ClaimOnOom: claims a heap upon first OOM, useful for initialization
• WasmHandler: itegrate with WebAssembly's memory module for automatic memory heap management

As an example of a custom implementation, recovering by extending the heap is implemented below.

use talc::*;

struct MyOomHandler {
    heap: Span,
}

impl OomHandler for MyOomHandler {
    fn handle_oom(talc: &mut Talc<Self>, layout: core::alloc::Layout) -> Result<(), ()> {
        // Talc doesn't have enough memory, and we just got called!
        // We'll go through an example of how to handle this situation.
    
        // We can inspect `layout` to estimate how much we should free up for this allocation
        // or we can extend by any amount (increasing powers of two has good time complexity).
        // (Creating another heap with `claim` will also work.)
    
        // This function will be repeatedly called until we free up enough memory or 
        // we return Err(()) causing allocation failure. Be careful to avoid conditions where 
        // the heap isn't sufficiently extended indefinitely, causing an infinite loop.
    
        // an arbitrary address limit for the sake of example
        const HEAP_TOP_LIMIT: *mut u8 = 0x80000000 as *mut u8;
    
        let old_heap: Span = talc.oom_handler.heap;
    
        // we're going to extend the heap upward, doubling its size
        // but we'll be sure not to extend past the limit
        let new_heap: Span = old_heap.extend(0, old_heap.size()).below(HEAP_TOP_LIMIT);
    
        if new_heap == old_heap {
            // we won't be extending the heap, so we should return Err
            return Err(());
        }
    
        unsafe {
            // we're assuming the new memory up to HEAP_TOP_LIMIT is unused and allocatable
            talc.oom_handler.heap = talc.extend(old_heap, new_heap);
        }
    
        Ok(())
    }
}

Conditional Features

• "lock_api" (default): Provides the Talck locking wrapper type that implements GlobalAlloc.
• "allocator" (default, requires nightly): Provides an Allocator trait implementation via Talck.
• "nightly_api" (default, requires nightly): Provides the Span::from(*mut [T]) and Span::from_slice functions.
• "counters": Talc will track heap and allocation metrics. Use Talc::get_counters to access them.
• "allocator-api2": Talck will implement allocator_api2::alloc::Allocator if "allocator" is not active.

Stable Rust and MSRV

Talc can be built on stable Rust by disabling "allocator" and "nightly_api". The MSRV is 1.67.1.

Disabling "nightly_api" disables Span::from(*mut [T]), Span::from(*const [T]), Span::from_const_slice and Span::from_slice.

Algorithm

This is a dlmalloc-style linked list allocator with boundary tagging and bucketing, aimed at general-purpose use cases. Allocation is O(n) worst case (but in practice its near-constant time, see microbenchmarks), while in-place reallocations and deallocations are O(1).

Additionally, the layout of chunk metadata is rearranged to allow for smaller minimum-size chunks to reduce memory overhead of small allocations. The minimum chunk size is 3 * usize, with a single usize being reserved per allocation. This is more efficient than dlmalloc and galloc, despite using a similar algorithm.

Future Development

• Support better concurrency, as it's the main deficit of the allocator
• Change the default features to be stable by default
• Add out-of-the-box support for more systems
• Allow for integrating with a backing allocator & (deferred) freeing of unused memory (e.g. better integration with mmap/paging)

Update: All of this is currently in the works. No guarantees on when it will be done, but significant progress has been made.

Changelog

v4.4.2

• polarathene: Replace README relative links with fully-qualified links.

• polarathene: Improve docs for stable_examples/examples/std_global_allocator.rs.

• Improved docs for stable_examples/examples/stable_allocator_api.rs and stable_examples/examples/std_global_allocator.rs.

• Deprecated the Span::from* function for converting from shared references and const pointers, as they make committing UB easy. These will be removed in v5.

• Fixed up a bunch of warnings all over the project.

v4.4.1

• Added utility function except to Span, which takes the set difference, potentially splitting the Span. Thanks bjorn3 for the suggestion!

v4.4.0

• Added feature allocator-api2 which allows using the Allocator trait on stable via the allocator-api2 crate. Thanks jess-sol!

v4.3.1

• Updated the README a little

v4.3.0

• Added an implementation for Display for the counters. Hopefully this makes your logs a bit prettier.

  • Bug me if you have opinions about the current layout, I'm open to changing it.

• Added Frusa and RLSF to the benchmarks.

  • Good showing by RLSF all around, and Frusa has particular workloads it excels at.

• Changed random actions benchmark to measure over various allocation sizes.

v4.2.0

• Optimized reallocation to allows other allocation operations to occur while memcopy-ing if an in-place reallocation failed.

  • As a side effect Talc now has a grow_in_place function that returns Err if growing the memory in-place isn't possible.
  • A graph of the random actions benchmark with a workload that benefits from this has been included in the benchmarks section.

• Added Span::from_* and From<> functions for const pointers and shared references.

  • This makes creating a span in static contexts on stable much easier: Span::from_const_array(addr_of!(MEMORY))

• Fix: Made Talck derive Debug again.

• Contribution by Ken Hoover: add Talc arena-style allocation size and perf WASM benchmarks

  • This might be a great option if you have a known dynamic memory requirement and would like to reduce your WASM size a little more.

• wasm-size now uses wasm-opt, giving more realistic size differences for users of wasm-pack

• Improved shell scripts

• Overhauled microbenchmarks

  • No longer simulates high-heap pressure as tolerating allocation failure is rare
  • Data is now displayed using box-and-whisker plots

v4.1.1

• Fix: Reset MSRV to 1.67.1 and added a check to test.sh for it

v4.1.0 (yanked, use 4.1.1)

• Added optional tracking of allocation metrics. Thanks Ken Hoover for the suggestion!
  • Enable the "counters" feature. Access the data via talc.get_counters()
  • Metrics include allocation count, bytes available, fragmentation, overhead, and more.
• Improvements to documentation
• Improved and updated benchmarks
• Integrated the WASM performance benchmark into the project. Use wasm-bench.sh to run (requires wasm-pack and deno)
• Improved wasm-size and wasm-size.sh

v4.0.0

• Changed Talck's API to be more inline with Rust norms.
  • Talck now hides its internal structure (no more .0).
  • Talck::talc() has been replaced by Talck::lock().
  • Talck::new() and Talck::into_inner(self) have been added.
  • Removed TalckRef and implemented the Allocator trait on Talck directly. No need to call talck.allocator() anymore.
• Changed API for provided locking mechanism
  • Moved AssumeUnlockable into talc::locking::AssumeUnlockable
  • Removed Talc::lock_assume_single_threaded, use .lock::<talc::locking::AssumeUnlockable>() if necessary.
• Improvements to documentation here and there. Thanks polarathene for the contribution!

v3.1.2

• Some improvements to documentation.

v3.1.1

• Changed the WASM OOM handler's behavior to be more robust if other code calls memory.grow during the allocator's use.

v3.1.0

• Reduced use of nightly-only features, and feature-gated the remainder (Span::from(*mut [T]) and Span::from_slice) behind nightly_api.
• nightly_api feature is default-enabled
  • WARNING: use of default-features = false may cause unexpected errors if the gated functions are used. Consider adding nightly_api or using another function.

v3.0.1

• Improved documentation
• Improved and updated benchmarks
  • Increased the range of allocation sizes on Random Actions. (sorry Buddy Allocator!)
  • Increased the number of iterations the Heap Efficiency benchmark does to produce more accurate and stable values.

v3.0.0

• Added support for multiple discontinuous heaps! This required some major API changes
  • new_arena no longer exists (use new and then claim)
  • init has been replaced with claim
  • claim, extend and truncate now return the new heap extent
  • InitOnOom is now ClaimOnOom.
  • All of the above now have different behavior and documentation.
• Each heap now has a fixed overhead of one usize at the bottom.

To migrate from v2 to v3, keep in mind that you must keep track of the heaps if you want to resize them, by storing the returned Spans. Read claim, extend and truncate's documentation for all the details.

v2.2.1

• Rewrote the allocator internals to place allocation metadata above the allocation.
  • This will have the largest impact on avoiding false sharing, where previously, the allocation metadata for one allocation would infringe on the cache-line of the allocation before it, even if a sufficiently high alignment was demanded. Single-threaded performance marginally increased, too.
• Removed heap_exhaustion and replaced heap_efficiency benchmarks.
• Improved documentation and other resources.
• Changed the WASM size measurement to include slightly less overhead.

v2.2.0

• Added dlmalloc to the benchmarks.
• WASM should now be fully supported via TalckWasm. Let me know what breaks ;)
  • Find more details here.

v2.1.0

• Tests are now passing on 32 bit targets.
• Documentation fixes and improvements for various items.
• Fixed using lock_api without allocator.
• Experimental WASM support has been added via TalckWasm on WASM targets.

v2.0.0

• Removed dependency on spin and switched to using lock_api (thanks Stefan Lankes)
  • You can specify the lock you want to use with talc.lock::<spin::Mutex<()>>() for example.
• Removed the requirement that the Talc struct must not be moved, and removed the mov function.
  • The arena is now used to store metadata, so extremely small arenas will result in allocation failure.
• Made the OOM handling system use generics and traits instead of a function pointer.
  • Use ErrOnOom to do what it says on the tin. InitOnOom is similar but inits to the given span if completely uninitialized. Implement OomHandler on any struct to implement your own behaviour (the OOM handler state can be accessed from handle_oom via talc.oom_handler).
• Changed the API and internals of Span and other changes to pass miri's Stacked Borrows checks.
  • Span now uses pointers exclusively and carries provenance.
• Updated the benchmarks in a number of ways, notably adding buddy_alloc and removing simple_chunk_allocator.