nx_nishi_usb_os.nx

nx_nishi_usb_os.nx -- x86 ladder R-KERN-4: INTEGRATION -- one booted kernel that runs a userland. The kernel-internals rungs proved GDT/IDT (R-KERN-1), paging (R-KERN-2), and preemption (R-KERN-3) SEPARATELY. This fuses them into ONE kernel, loaded off the persisted image, that performs the full OS bring-up and then runs userland tasks: boot (MBR -> INT 13h load -> kernel) -> LONG mode -> lgdt (GDT) -> build the IDT in RAM with a TIMER gate (0x20) AND a SYSCALL gate (0x80) -> lidt -> build page tables + load CR3 -> STI -> launch. Then two userland tasks run preemptively (timer-driven, round-robin); each loops incrementing its own counter and issues `int 0x80`, which VECTORS THROUGH THE KERNEL'S IDT to the syscall handler the kernel installed -- the handler services it (increments a kernel-side syscall counter) and IRETs back. KAT: ONE kernel did it all -- (T1) booted off disk to LONG mode; (T2) lgdt loaded the GDT; (T3) lidt loaded the IDT and the syscall gate points at the kernel's handler; (T4) CR3 + the page walk works; (T5) BOTH userland tasks ran (preemptive multitasking); (T6) the tasks' syscalls were SERVICED by the kernel's handler (kernel-side count > 0). NEG/liar-kill (T7): with the timer masked, task B never runs. HONEST SCOPE: like the other rungs, the context switch + the IF-gated timer are modeled (as nx_kernel_sched), and the IDT here uses 8-byte handler slots (R-KERN-1 already proved the full split-field gate); the syscall path (task -> IDT[0x80] -> kernel handler -> iret) is genuinely vectored. The shared syscall counter is intentionally unlocked, so a preempt mid-increment can race (a lower bound, and an honest demonstration of why kernels CLI/lock in handlers). NEVER-BRICK (Rule 26): writes a FILE; INT 13h *reads* only; no /dev. expect_exit: 0 license_tier: ORIGINAL