Prep: The Console and the Kernel Shell — 45k · 46k

Session: Thu Nov 12, 1h45 · Exercises: 45k_console, 46k_shell · Prep time: ~55 min · Lecture: Device Interrupts, the PLIC, and the Console · Shells, and the Reference Kernel

What you will build

A keypress makes the UART raise source 10 and the PLIC deliver a supervisor external interrupt; your handler asks the PLIC which source fired, drains the waiting bytes into the ring buffer, and tells the PLIC it is done. Then the evaluate step of a kernel-mode REPL: split a line into words and send the first to one of four given commands (pwd, ls, cd, mkdir), or report it as not found. The tests check that a simulated interrupt lands its byte in the buffer and that mkdir docs then ls lists docs; then cargo run boots to a rv6$ prompt.

Concepts you need

• Claim, service, complete; two ways a console dies — Device Interrupts §2.3 · rv6 Architecture § Path 2
• Top half, bottom half, the lock-free ring, wfi — Device Interrupts §4.1, §4.2, §4.3
• A shell is a REPL, and here also the line discipline — Shells §2.1, §2.3 · Device Interrupts §5.2
• Tokens are borrowed views; split_whitespace allocates nothing — Shells §3.1
• Dispatch by match; output through the Out trait — Shells §4.1, §4.3 · rv6 Architecture § Two shells
• Reading to extend, not to rebuild — Shells §1.3, §6.1

Read before class

What	Time
Device Interrupts §2.3, §3, §4, §6 (claim, service, complete; the ring; the whole path)	25 min
Shells §2–§4 (REPL, tokens, dispatch, Out)	20 min
Shells §1.3, §6.1 · rv6 Architecture: Path 2, Two shells	10 min

Mental model

Type ls and Enter into a four-byte ring. The counters only grow; % 4 picks the slot:

                                BUF      HEAD  TAIL
'l', 's' -> two handler pushes  [ls..]    0     2    in the trap, SIE off
shell getc -> 'l', echo it      [ls..]    1     2    normal priority
shell getc -> 's', echo it      [ls..]    2     2    empty: HEAD == TAIL
shell getc -> wfi               [ls..]    2     2    halted, zero cycles
Enter -> handler push           [ls\r.]   2     3    wfi returns; "ls" is a line

The handler writes only BUF and TAIL, the reader only HEAD; a stale read errs safely, so one hart needs no lock, and a lock in the handler would deadlock with whatever it interrupted. The ring holds no echo, no erasing, no line: those are the shell's job, off the interrupt path.

Check yourself

1. Your handler buffers the byte but never writes the source number back to the PLIC. What do you see?
   Answer

   The first keypress works; later ones vanish. The PLIC still considers source 10 claimed and never delivers it again; nothing panics.
2. The opposite: it completes without reading the UART. Why is that different?
   Answer

   An interrupt storm: the byte still sits in the receive register, the level-triggered line stays high, and the PLIC re-raises it the moment the gateway re-arms; the kernel spins at 100% CPU.
3. The tokenizer yields &str slices into the line buffer. What stops the REPL from clearing the line too early?
   Answer

   The tokens borrow the line, so the borrow checker refuses line.clear() until the evaluate step returns.

What "done" looks like

oslings run is green, then oslings submit before you leave. Not green? Submit anyway (substantial credit), then finish by Thursday 11:59 pm and submit again. Midterm 2 is next Thursday, Nov 19, and covers processes through user mode.

Extra credit today

47k_file_commands (+0.5): touch, cat, echo TEXT > FILE, rm, and rmdir. The redirect and rmdir are given as worked examples; the other three stitch together filesystem promises: name to inode number, bytes into a fixed buffer, core::str::from_utf8, unlink. See Shells §1.3 and §3.4.

If you finish early

Work Shells Practice Problem 5, the deadlock in the kernel shell; it comes back on Midterm 2. Then read chapter 5 of the xv6 book, "Interrupts and device drivers," or start Friday's prep page, Prep: User Mode.