smolvm v1.9.2
What's Changed
- Relocate libkrun.dylib's Homebrew dylib references to @loader_path so macOS packed launchers boot by @BinSquare in #1013
- Point the install.sh sync note at the real website path so the website copy is kept in sync by @BinSquare in #1017
- Widen the read window for the whole flatten so a large pack does not fail with a spurious EAGAIN by @Bnjoroge1 in #957
- Name machine options that were written after
--instead of silently ignoring them by @Bnjoroge1 in #958 - Skip the /workspace fallback for volumes mounted below it, not only at it by @NickyHeC in #1005
- Support automatic CUDA replay and distributed multi-GPU execution by @BinSquare in #998
- Use as_chunks instead of chunks_exact with a constant size in the gpu_loopback example so clippy passes by @BinSquare in #1019
- Reject unknown fields on the exec and run API so a mis-cased safety field can't be silently dropped, and clarify that --storage bounds the writable disk by @BinSquare in #1018
- Grow an existing storage/overlay disk to the requested size instead of booting the stale one by @Bnjoroge1 in #956
- Bump libkrun to the build with aarch64-KVM snapshot and fork support so machine fork works on Linux aarch64 by @BinSquare in #1021
- Stage archive-flatten scratch on the storage disk instead of the guest tmpfs by @BinSquare in #1014
- Bump the workspace to 1.9.2 by @BinSquare in #1022
Full Changelog: v1.9.1...v1.9.2