POSIX timer ABI checkpoint

Scope

This checkpoint covers native x86-64 Linux syscall numbers 222-226:

timer_create
timer_settime
timer_gettime
timer_getoverrun
timer_delete

It also covers the rt_sigtimedwait path used by musl's SIGEV_THREAD implementation.

Implementation

The kernel now keeps timer objects in the creating task, validates clock ids, notification modes, signal numbers, timespec ranges, timer flags, and user pointers, and supports one-shot and periodic relative or absolute timers. Timer expiry queues SIGEV_SIGNAL notifications in the shared process-pending queue, while SIGEV_THREAD_ID is delivered to the requested thread. Both paths wake a task parked in rt_sigtimedwait and account repeated expirations as overrun. Timer queries and deletion use the same task-owned object and return Linux-shaped timespec values.

musl's SIGEV_THREAD sequence is supported without an application port: musl's helper pthread creates a kernel SIGEV_THREAD_ID timer for signal 32, blocks that signal, and waits for it through rt_sigtimedwait. The kernel queues and wakes that exact thread.

Evidence

tools/tests/musl_guest_test.c includes a raw musl regression covering SIGEV_NONE, set/get/delete, and an actual SIGEV_THREAD callback. The test was built in both static and dynamic forms and run from build/images/leonos4-musl-probes.iso under QEMU/OVMF with 2 GiB RAM:


[musl-abi:static] END failed=0
[musl-abi:dynamic] END failed=0

The desktop Vim run used the unchanged static Linux Vim binary. Its serial log contains no E1286 or Could not set timeout message, and the test saved and read back /tmp/vimtest.txt.

Remaining boundary

The targeted behavior is implemented and QEMU-tested. Full Linux POSIX timer coverage is still pending for complete siginfo.si_timerid/si_overrun data, all clock sources, timer resource limits, and stress under SMP. VMware has not been run in this environment.