Stream: git-wasmtime

Topic: wasmtime / PR #7884 Allocate thread IDs for wasi_thread_s...

view this post on Zulip Wasmtime GitHub notifications bot (Feb 07 2024 at 13:52):

Milek7 opened PR #7884 from Milek7:threadid to bytecodealliance:main:

TID returned by wasi_thread_spawn must be non-zero and upper three bits must always be zero: https://github.com/WebAssembly/wasi-threads?tab=readme-ov-file#design-choice-thread-ids

Current implementation didn't fulfill this requirements, only clearing the top bit from random value. Additionally, from birthday paradox probability of collision in allowed TID range nears 50% after spawning 27000 threads. This is large but not totally impossible, thus replace random generation with sequential allocation.

view this post on Zulip Wasmtime GitHub notifications bot (Feb 07 2024 at 13:52):

Milek7 requested pchickey for a review on PR #7884.

view this post on Zulip Wasmtime GitHub notifications bot (Feb 07 2024 at 13:52):

Milek7 requested wasmtime-core-reviewers for a review on PR #7884.

view this post on Zulip Wasmtime GitHub notifications bot (Feb 07 2024 at 13:56):

Milek7 edited PR #7884:

TID returned by wasi_thread_spawn must be non-zero and upper three bits must always be zero: https://github.com/WebAssembly/wasi-threads?tab=readme-ov-file#design-choice-thread-ids

Current implementation didn't fulfill this requirements, only clearing the top bit from random value. This breaks wasi-libc mutexes when random TID is larger than 0x1FFFFFFF.

Additionally, from birthday paradox probability of collision in allowed TID range nears 50% after spawning 27000 threads. This is large but not totally impossible, thus replace random generation with sequential allocation.

view this post on Zulip Wasmtime GitHub notifications bot (Feb 07 2024 at 17:47):

pchickey requested abrown for a review on PR #7884.

view this post on Zulip Wasmtime GitHub notifications bot (Feb 07 2024 at 18:45):

Milek7 commented on PR #7884:

This PR makes 0x1FFFFFFF the last TID that will be returned, but it is incorrect on wasm-libc because it reserves that for main thread ID. But wasi-threads proposal as written allows for this, so I'm not sure whether assume that's wasm-libc problem or limit it there 0x1FFFFFFE.

view this post on Zulip Wasmtime GitHub notifications bot (Feb 07 2024 at 18:45):

Milek7 edited a comment on PR #7884:

This PR makes 0x1FFFFFFF the last TID that will be returned, but it is incorrect on wasm-libc because it reserves that for main thread ID. But wasi-threads proposal as written allows for this, so I'm not sure whether assume that's wasm-libc problem or limit it here to 0x1FFFFFFE.

view this post on Zulip Wasmtime GitHub notifications bot (Feb 07 2024 at 20:52):

Milek7 deleted a comment on PR #7884:

This PR makes 0x1FFFFFFF the last TID that will be returned, but it is incorrect on wasm-libc because it reserves that for main thread ID. But wasi-threads proposal as written allows for this, so I'm not sure whether assume that's wasm-libc problem or limit it here to 0x1FFFFFFE.

view this post on Zulip Wasmtime GitHub notifications bot (Feb 07 2024 at 20:56):

Milek7 edited PR #7884:

TID returned by wasi_thread_spawn must be non-zero and upper three bits must always be zero: https://github.com/WebAssembly/wasi-threads?tab=readme-ov-file#design-choice-thread-ids

Current implementation didn't fulfill this requirements, only clearing the top bit from random value. This breaks wasi-libc mutexes when random TID is larger or equal than 0x3FFFFFFF.

Additionally, from birthday paradox probability of collision in allowed TID range nears 50% after spawning 27000 threads. This is large but not totally impossible, thus replace random generation with sequential allocation.

view this post on Zulip Wasmtime GitHub notifications bot (Feb 07 2024 at 20:57):

Milek7 edited PR #7884:

TID returned by wasi_thread_spawn must be non-zero and upper three bits must always be zero: https://github.com/WebAssembly/wasi-threads?tab=readme-ov-file#design-choice-thread-ids

Current implementation didn't fulfill this requirements, only clearing the top bit from random value. This breaks wasi-libc mutexes when random TID is larger or equal to 0x3FFFFFFF.

Additionally, from birthday paradox probability of collision in allowed TID range nears 50% after spawning 27000 threads. This is large but not totally impossible, thus replace random generation with sequential allocation.

view this post on Zulip Wasmtime GitHub notifications bot (Feb 12 2024 at 16:06):

abrown submitted PR review.

view this post on Zulip Wasmtime GitHub notifications bot (Feb 12 2024 at 16:07):

abrown commented on PR #7884:

@Milek7, thanks for the PR!

view this post on Zulip Wasmtime GitHub notifications bot (Feb 12 2024 at 16:41):

abrown merged PR #7884.

Last updated: Nov 22 2024 at 17:03 UTC