[YUZHEthefool]

A PID-targeted wait4/waitid briefly published waiting_child = Some(0) before narrowing the filter after its child scan. A sibling exiting in that window could wake the parent, and each retry reopened the window. Remote diagnostics reproduced two extra wakes while waiting for PID 55: sibling PID 54 woke it once at exit notification and again at the final switch-out handoff. This resolves the residual selective-wait issue recorded in #49.

Resolve the target's global PID before publishing Blocked, using the child's retained namespace identity, then scan again for reapable children. An exit during resolution is therefore observed by the scan or wakes the already-published wait. Signal/kill checks remain under the parent PCB lock, and an unresolved target stays runnable for stale/incomplete child-list recovery.

• Share the unchanged PID-match wake operation between exit notification, group notification and the scheduler handoff.
• Keep PID-targeted group waits selective while retaining their child-event counter check.
• Add five hosted interleaving/cleanup tests. Repeat the Ring-3 sibling check 16 times and also assert zero extra wakes across the nested reaper. The existing counter and zero-wake requirement are retained.
• Keep the rebuilt userspace ELF in a separate commit.

Fixes: 0a42bd8 (feat(sched): selective wait wake, guard-free SCHED-TRAP and a switch sentinel)

Validation

Executed on the requested remote host with Ubuntu 22.04.5, QEMU 6.2.0, OVMF and the pinned Rust toolchain.

Check	Result
Diagnostic baseline, repeated sibling waits	Reproduced two sibling wakes on the 24th pair; guest finished 50 passed / 1 failed. Stopped after PID 1 exited once the failure evidence was complete.
Negative control restoring only the broad publication filter	New sibling-interleaving regression fails at the unexpected wake. All five focused tests pass with the fix.
Hosted subcrate gate	PASS: 613 unit tests, CpuLocal doctests and 3 test-code compile checks; default test parallelism.
make fmt-check lint clippy build, syscall and musl feature builds	PASS; ABI oracle: 11 structs / 100 values / 17 tripwires. Existing warnings retained.
Ring-3 oracle, four CPUs, 300 seconds	PASS: 51 passed / 0 failed; PID 1 exit 0.
Ring-3 oracle, one CPU, 300 seconds	PASS: 50 passed / 0 failed; PID 1 exit 0. The dual-CPU COW leg is explicitly skipped on one CPU.
Boot / runtime, 120 seconds each	QUALIFIED (exit 3): 28/46/0 and 35/39/0 passed/deferred/failed.
SMP two and four CPUs, 120 seconds each	QUALIFIED (exit 3): each 37/37/0; required topology and three cross-CPU witnesses present.
musl libc/ABI gate, one CPU, 300 seconds	PASS (exit 0): required libc/ABI markers, namespace wait, blocked-signal and pthread checks; PID 1 exit 0 and zero NX faults. SMP-only musl probes are explicitly skipped on this lane.

Final guest images were built in a fresh source/build directory and frozen separately for each feature. Modified source files and the syscall-fixture hashes match the submitted tree. An earlier incremental-image run reused stale objects because archive timestamps predated the build; that result is excluded. A musl launch initially selected the default ESP and is also excluded; the corrected run uses the isolated musl artifact with its packaged SHA-256 verified. Diagnostic/Ring-3 serial evidence, gate verdicts and source/image hashes are retained privately. Qualified boot/runtime gates remain qualified; no hardware qualification is claimed.

Source/regressions: d1dcc62; generated syscall fixture: 6683ae9. All selected remote checks are complete with the qualified limits above. CI for 6683ae9 is running.