BusyBox gets a real PTY

P80 could run BusyBox, but the shell was not really sitting on a tty. That showed up exactly where expected: no job control, awkward erase behavior, and live screen input that was too easy to confuse with the helper’s pipe-fed path.

P81 changes the shape. The initramfs now carries /dev/ptmx, mounts devpts, and runs console_sh. That helper opens a PTY master, makes the slave the controlling terminal for /bin/sh -i, then bridges the chip’s host-input MMIO FIFO into the PTY master. Output comes back from the PTY master to /dev/console/hvc0.

The bring-up found three useful bugs:

• /init used mkdir before BusyBox applet symlinks existed. It now calls /bin/busybox mkdir and /bin/busybox mount until busybox --install -s /bin has run.
• The Verilator UART path escaped ANSI bytes for readable logs. That was wrong for a live terminal; tty-screen and screen-smoke now run the harness with +uart_raw.
• The host-input handshake popped bytes based on host_input_ready after the clock edge. When the chip FIFO was full and the guest read one byte, that post-edge value could make the host discard a byte the chip never accepted. The harness now samples the handshake before the edge.

The passing tty-smoke waits for the prompt, then proves uname, applet symlinks, Backspace erase behavior, file creation, and file readback. The passing screen-smoke does the same through a detached screen session attached to /tmp/p81-busybox.

Passing runs:

• make tty-smoke ...: PASS, stopped after P81-FILE-OK at 179,472,594 post-load cycles with 268 input bytes read and delivered.
• make screen-smoke ...: PASS, proved P81-SCREEN-INPUT-OK, P81-SCREEN-UNAME-OK, P81-SCREEN-BACKSPACE-OK, P81-SCREEN-TOUCH-OK, screen-file-ok, and P81-SCREEN-FILE-OK.

The remaining terminal work is quality, not proof: better interactive latency, less noisy harness progress during live sessions, and a longer timeout or explicit “run forever” mode for manual shell sessions.