Marshal sched_getaffinity masks across process memory

[brandonpayton]

Purpose

Make sched_getaffinity(2) return its CPU mask through the normal guest/host/kernel memory contract instead of dereferencing a guest address in kernel memory.

This surfaced while validating the Kandelo Homebrew xz formula: liblzma probes the available CPU count before using its multithreaded encoder.

Root Cause

Syscall 238 was implemented in the Rust kernel, but absent from the shared host syscall-argument descriptors. The host therefore forwarded the process-memory pointer unchanged. The kernel interpreted that address in its own Wasm memory when constructing the output slice, trapped out of bounds, and the host converted the trap to EIO.

The change:

• declares SchedGetaffinity in the shared syscall inventory;
• marshals argument 2 as an output buffer sized by argument 1;
• returns the written byte count from the raw syscall, as musl's wrapper requires before it can preserve the mask;
• adds descriptor, generated-ABI, runtime, and libc fixture coverage.

ABI

The branch is rebased unchanged onto ABI 18 at 02e63bc46; its patch ID remains ee5f0f65460ab68429029f020def965b93fc4dac.

scripts/check-abi-version.sh classifies the two snapshot entries as additive-compatible:

• syscall-argument descriptor 238 added;
• syscall number 238 named.

Existing ABI entries are unchanged, so ABI_VERSION remains 18. The ABI snapshot and generated TypeScript bindings are committed in sync.

Validation

Exact ABI 18 rebase validation through scripts/dev-shell.sh:

• bash scripts/check-abi-version.sh: passed; snapshot, libc header, and TypeScript bindings are in sync;
• focused shared descriptor test sched_getaffinity_marshals_the_requested_cpuset: passed;
• git diff --check: passed;
• original versus rebased patch ID: identical.

The pre-rebase branch also passed kernel/shared tests, generated ABI and focused host tests, Node and Chromium runtime fixtures, POSIX, libc-test, and Sortix validation as recorded in the earlier PR run. CI is rebuilding the rebased branch against ABI 18.

cargo fmt is not available in the repository dev shell; the Rust changes retain the surrounding formatter style.

This main-repository PR is intentionally left for Brandon's review and must not be auto-merged.

[brandonpayton]

CI note: the failed Vitest job was rerun and reproduced only the same setup-timing assertion at host/test/teardown-reclaim.test.ts:71; every other suite remains green. The race is root-caused and fixed independently in #889, whose complete kernel/fork/Vitest/browser/libc/POSIX/Sortix gate is green. The unchanged pre-#889 merge base cannot make that test deterministic, so further reruns would not add evidence; merge or rebase after #889 is the correct path.

[brandonpayton]

Merged PR #907 supersedes this work through the strengthened affinity marshalling implementation now on main at 74339d68f. This head was not merged verbatim; the behavior was batch-validated.