WEBVTT 00:00.000 --> 00:01.088 Reset to Rust: 00:01.688 --> 00:08.067 an explainer page for exercise 31k_boot, from power-on to your first line of Rust. 00:08.667 --> 00:14.555 A script steps the page, it was built with Claude, and its addresses are examples, not the lecture's. 00:15.080 --> 00:18.195 This comes before Tuesday's lecture, October 6. 00:18.795 --> 00:23.275 On Thursday, October 8, you write the first code your kernel runs. 00:23.875 --> 00:27.821 No solution is shown: only the state the machine must reach. 00:28.320 --> 00:32.053 The machine QEMU emulates has one address space. 00:32.653 --> 00:35.555 RAM starts at 0x8000_0000. 00:36.155 --> 00:41.808 Below it are devices: a store to a device's address is a message to that device. 00:42.320 --> 00:46.331 At reset there is one hart, RISC-V's name for a processor core. 00:46.931 --> 00:50.301 Its pc, the program counter, points at the boot ROM: 00:50.901 --> 00:54.805 built-in code whose last instruction jumps to the base of RAM. 00:55.320 --> 00:59.928 The linker script, given to you, assigns every byte of your kernel an address. 01:00.528 --> 01:02.171 It starts at the base of RAM, 01:02.771 --> 01:06.355 and the first piece placed there is the section named .entry. 01:06.880 --> 01:13.877 The only code in that section is the stub, _entry: a few assembly instructions, yours to write. 01:14.477 --> 01:17.720 Listed first, so placed first: where the ROM jumps. 01:18.320 --> 01:22.053 Placed last, the boot stack: 16 KiB of zeros. 01:22.560 --> 01:24.203 Now the boot, step by step: 01:24.803 --> 01:26.168 the hart's registers, 01:26.768 --> 01:28.752 the base of RAM, holding your kernel, 01:29.352 --> 01:30.461 and the address map. 01:31.960 --> 01:38.893 Reset. pc is in the boot ROM, and sp, the stack pointer, holds zero in this machine: not a stack. 01:39.893 --> 01:43.755 The ROM's jump lands on your kernel: nothing else runs first. 01:45.280 --> 01:50.976 Now pc is at the stub: four numbered slots, one for each instruction you will write. 01:51.576 --> 01:56.525 They stay empty. Watch sp instead: Rust cannot start without a stack. 01:58.040 --> 02:03.800 After slot 1, sp holds the address of the stack array: its first byte, the low end. 02:04.800 --> 02:09.472 After slot 2, t0 holds the array's size, 16 KiB. 02:10.072 --> 02:11.203 Is the stack ready? 02:11.720 --> 02:12.531 Not yet. 02:14.031 --> 02:19.769 After slot 3, sp holds the array's top: 16 KiB above the low end. 02:20.369 --> 02:24.593 A stack grows down, so from here every push lands inside the array. 02:25.193 --> 02:26.345 Now it is a stack. 02:27.880 --> 02:35.603 After slot 4, pc is in kmain: your Rust code. No slot stored anything: the stub needed no stack. 02:36.603 --> 02:40.848 Rust's first store goes through sp, and lands inside the array. 02:42.360 --> 02:47.821 Printing is a store too: each byte of the banner is stored to the UART, the serial port. 02:48.821 --> 02:50.677 Then the marker, OSLINGS:PASS: 02:51.277 --> 02:53.197 the one line oslings looks for. 02:54.720 --> 02:57.557 The last store, to the test finisher, ends the run. 02:58.157 --> 03:00.952 oslings finds the marker, and reports a pass. 03:01.480 --> 03:05.491 Now a broken case, no stack: a stub with only slot 4. 03:06.091 --> 03:09.739 pc reaches kmain, and sp still holds zero. 03:11.240 --> 03:17.341 Rust's first store goes through sp, and faults: nothing is at that address, so it cannot finish. 03:18.341 --> 03:24.763 With no handler installed for a fault, pc goes to address zero, and faults there without end. 03:26.280 --> 03:29.565 Nothing is printed, and oslings reports a timeout. 03:30.165 --> 03:33.771 Its message is a guess: it sees silence, not the reason. 03:34.771 --> 03:36.947 This is why the stack pointer comes first. 03:37.480 --> 03:42.259 Another broken case leaves out slot 3: sp stays at the low end. 03:43.259 --> 03:45.520 Rust's first store lands below the array, 03:46.120 --> 03:49.683 and in this build, the strings the kernel prints are in its path. 03:50.200 --> 03:52.675 Nothing faults: this is ordinary RAM. 03:53.675 --> 04:00.160 But the stack overwrote the strings, so what the UART sent is not text, and oslings shows none of it. 04:00.760 --> 04:04.451 It reports: kernel exited without printing OSLINGS:PASS. 04:04.960 --> 04:06.197 You write two things. 04:06.797 --> 04:14.243 The stub's four instruction lines, which leave sp at the top of the array, STACK0, and pc at kmain. 04:14.843 --> 04:17.552 And one line in kmain that prints the marker. 04:18.080 --> 04:23.157 If you get a timeout with no output, the hart likely faulted before it could print: 04:23.757 --> 04:27.384 Rust ran without a stack, or the skeleton is still untouched. 04:27.880 --> 04:30.056 If the kernel exits without the marker: 04:30.656 --> 04:34.197 the banner as output means only the marker line is missing. 04:34.797 --> 04:37.891 No output means sp was likely at the low end. 04:38.491 --> 04:42.480 But OSLINGS:FAIL means Rust ran and panicked: the stack works. 04:43.000 --> 04:50.061 A hart at reset can run instructions and has nothing else: your stub gives it a stack before any Rust runs. 04:50.661 --> 04:54.907 Next: the free list explainer, Tuesday's lecture, and Thursday's session.