[BUG] Thread-exit diagnostics fail a syscall-result assertion #54
Labels
No labels
alpha-1
bug
ci
docs
duplicate
enhancement
help wanted
invalid
performance
phase-6
question
refactor
security
wontfix
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set
Reference
erix/kernel#54
Loading…
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Summary
The thread-exit and new execution-watch guest cases panic on an exact native syscall-result assertion. The older running-members case passes on the same source graph.
Reproduction
Use Integration
42b90e2420c9df22cac37c288c1f402bebf3558dwith its originalcomponents.kernel-lifetime.tomland the repository boot-image test signing fixture. Run:Each separate attempt reaches the owned invocation workload and fails.
scripts/run-native-kernel-vm.sh memberspasses separately.Expected and actual behavior
Each original scenario should complete all ordered, unique markers and return QEMU status 33. Instead, the four affected cases return 37 with
ERIX_KERNEL:PANIC; the runner returns 1. The panic points to the exact return-register assertion ininvocation_vm.rs, line 266. Factory completion has not been reported.Environment and identity
5f646535d996aa02e39a321a4fc289a3799e84bf.42b90e2420c9df22cac37c288c1f402bebf3558d; unmodified signed catalog and scenarios.Evidence and prior checks
All five original serial logs, selected revisions, runner outputs and statuses are retained. Each runner confirms cleanup without quarantine. The passing members case independently exercises the common factory, context, enrollment and attenuation sequence. The affected syscall and cause need a failure-only diagnostic; neither runtime nor fixture fault is assumed.
Tracking
Related: execution observation and cleanup, Kernel PR #3, and Phase 6. Preserve exact assertions, original markers, capability roles and deadlines while diagnosing and repairing the failure.
The numeric failure diagnostics identify two fixture defects. Later Notify installation overwrites the earlier observation packet, causing
INVALID_ARGUMENTbefore the intendedDENIEDcheck. Separately, directed receives may run before their senders and legitimately returnRETRY.The repair gives each actor separate counted packet storage and uses genuine blocking receive custody. Resumed user code still checks the exact successful result and payload. The design is recorded in #52; the affected scenarios and older member/TLS cases will run on the repaired signed graph before this issue closes.
The fresh signed graph passes both initial-execution watch cases, including zero results, and the older members and TLS regressions. Thread-exit and member-watch still fail the same unchanged SUBMIT/NOT_FOUND assertion because their supervisor’s original receive registration is live at that point. All six original attempts clean up without quarantine.
The diagnostic design now extends positive original receiver retirement to those two cases and places the unchanged absence probe after their genuine creator acknowledgement. The implementation keeps all native admission rules, oracles and deadlines. This regression remains open.
Fixed in Kernel 4ddaa480, selected by Integration 73c7a647. Fresh thread-exit, member-watch, initial full/zero, running-members and TLS VMs all pass. Each retains its original ordered unique markers, QEMU status 33 gate, release/TCG/one-CPU configuration and 60-second guest deadline. All six runs finish without warnings and confirm original cleanup without quarantine.
The corrections give result packets separate storage, register watches after real Start, and retire the original receiver before its acknowledged absence probe. All earlier failures remain recorded. Full CI remains independently affected by the runner descriptor backend; ordinary Posixd/libc and self-hosting acceptance remain open.