DByte Native Kernel Blocking

v11.6.0 extends the single-core CPL0 scheduler with real sleeping and blocking. Normal tasks cross a kernel-internal int 0x81 boundary so the CPU and existing interrupt stub create the same canonical PUSHAD plus EIP/CS/EFLAGS resume image used by IRQ0. No copied register context, runtime shim, or textual assembly exists.

task_sleep_ticks accepts intervals below half the u32 tick space. IRQ0 performs a bounded eight-task scan and uses a signed wrapping delta to wake deadlines safely across tick rollover. Sleeping and blocked tasks remain mapped but are never selected.

An Event owns eight fixed FIFO waiter slots. Signals are non-sticky: an empty signal succeeds and is forgotten, while a non-empty signal makes exactly the oldest waiter READY. event_signal preserves incoming IF and is safe from normal task or ordinary maskable hardware IRQ context. It never schedules, acknowledges hardware, or sends PIC EOI; wait and sleep remain task-context-only operations.

The proof boots at both 32 and 64 MiB. One worker sleeps without spinning, two waiters prove FIFO order, another runnable task advances while peers are blocked, and separate probes prove empty task/IRQ pulses are not remembered. Idle reclaims all exited worker stacks and frames before a probe verifies deterministic reuse.

Run the deterministic objects, relocation/disassembly checks, and QEMU proof:

powershell -ExecutionPolicy Bypass -File scripts/verify_native_blocking_core.ps1

Ignored evidence is written under tmp/native_v11.6.0/.

SOURCE: DBYTE_NATIVE_KERNEL_BLOCKING.md