This is only a rough guide ---- we are currently doing significant lab rewrites.
The first chunk of the class will be writing all the low-level code needed to run the pi and using modern techniques to validate it. Doing so will remove all magic from what is going on since all of the interesting code on both the pi and unix side will be written by you:
-
0-blink: get everyone up to speed and all necessary software installed. This should be a fast lab.
-
1-gpio: write your own code to control the r/pi
GPIO
pins using the Broadcom document GPIO description. You will use this to implement your own blink and touch sensor. -
2-bootloader: you will implement your own bootloader to transfer the code from your laptop to the pi. The most common r/pi bootloader out there uses the
xmodem
protocol. This approach is overly complicated. You will do a much simpler, more likely to be correct protocol. It has the advantage that it will later make it easier to do a network boot. -
3-cross-check: you will use read-write logging to verify that your GPIO code is equivalent to everyone else's. If one person got the code right, everyone will have it right.
A key part of this class is having you write all the low-level, fundamental code your OS will need. The good thing about this approach is that there is no magic. A bad thing is that a single mistake makes more a miserable quarter. Thus, we show you modern tricks for ensuring your code is correct.
-
4-uart: you write your first real device driver, for the UART, using only the Broadcom document. At this point, all key code on the pi is written by you. You will use the cross checking code from lab3 to verify your implementation matches everyone else's.
A great way to understand the UART hardware is to write your own software version of it and use this to communicate with your laptop. Doing so gives you a second source of output, which will turn out to be very useful later when we do networking.
-
5-replay: in a twist on lab3, you will use Unix system calls to interpose between your Unix and pi bootloader code, record all reads and writes, and test your bootloader implementation by replaying these back, both as seen and with systematic corruption.
This approach comes from the model-checking community, and I believe after you implement this lab and test (and fix) your bootloader you will be surprised if it breaks later. (In general, the approach we follow here applies well to other network protocols which have multi-step protocols and many potential failure modes, difficult to test in practice.)
At this point you have written low-level device driver code, a bootloader, and pretty much replaced all of our code. You've learned how to crush it under a barrage of automatic checks based on read-write equivalance and simple systematic exploration so that you will be surprised if the code ever breaks.
We are now going to switch gears to intensively implementing core OS functionality: threads, interrupts, virtual memory and file systems.
-
6-threads: we build a simple, but functional threads package. You will write the code for non-preemptive context switching: Most people don't understand such things so, once again, you'll leave lab knowing something many do not.
-
7-interrupts: you will walk through a simple, self-contained implementation of pi interrupts (for timer-interrupts), kicking each line until you understand what, how, why. You will use these to then implement a version of
gprof
(Unix statistical profiler) in about 30 lines.Perhaps the thing I love most about this course is that because we write all the code ourselves, we aren't constantly fighting some large, lumbering OS that can't get out of its own way. As a result, simple ideas only require simple code. This lab is a great example: a simple idea, about twenty minutes of code, an interesting result. If we did on Unix could spend weeks or more fighting various corner cases and have a result that is much much much slower and, worse, in terms of insight.
-
8-device-interrupts: we setup GPIO interrupts, then use this to build networking on the esp8266 wireless device. You will modify your previous labs so that you have an interrupt-based software UART implementation. You will then use this to connect and control the ESP8266 that we give out in class and ping other pi's.
Without interrupts, it's difficult to get networking working, since our GPIO pins (and our UART options) have limited space and, thus, unless our code checks them at exactly the right time, incoming messages will vaporize.
-
10-ESP: (note: do not do
9-esp8266
the code is not useful.) You use the $2 esp8266 to add networking. The initial lab just makes the esp work on your laptop using the provided library. The homework has you migrate it to your pi so you can communicate / control other pi's. The code depends on having an interrupt-based UART channel.
-
11-fuse.A: In this lab you will use the FUSE file system to wrap up your pi as a special file system and mount it on your laptop, where you can use standard utilities (and your normal shell) to interact with it. You can then control the pi by reading / writing to special files: e.g., echoing a
1
topi/reboot
to cause a reboot, echoing a program to/pi/run
to run it.This lab is a great example of the power of Unix's simple, powerful OO-interface that lets you package a variety of disparete things as files, directories, links and interact with them using a uniform set of verbs (e.g.,
open()-read()-write()-close()
). -
12-fuse.B: We continue fuse since it was a big lab :).
-
13-fat32: The SD and virtual memory are the biggest unknowns in our universe, so we'll bang out quick versions of each, and then circle back around and make your system more real.
You will write a simple read-only FAT32 file system that can read from your SD card. You will use this to do a very OS-style "hello world": use your FAT32 to read
hello-fixed.bin
from last lab from your SD card, jump to it and run it.
-
14-vm: You take a working VM system and, using the ARM documents, replace its page table manipulation with your own. You also add exception handling to catch protection and missing translation faults. The former gives you the tools to start doing user-level protection, the latter for demand paging and extending a stack as code accesses it.
-
15-vm-ops: The previous lab defined the main noun in the virtual memory universe (the page table); this lab does the main verbs used to set up the VM hardware, including how to synchronize hardware, translation, and page table state (more subtle than it sounds). At the end you should be able to delete all our starter code.
[HAVE NOT UPDATED THE LABS BELOW]
-
16-processes: today you will make user level processes. This will combine the virtual memory, system calls, and threads into a capstone implementation. You will use this ability to your fuse-FS support multiprocessing, so that you can have multiple programs running at the same time.
-
Additional labs that we can fit in to make the above more real and more integrated.
-
virtualization: this lab will show how to virtualize hardware. We will use simple tricks to transparently flip how your pi code is compiled so you can run it on Unix, only shipping the GPIO reads and writes to a small stub on the pi. As a result, you have full Unix debugging for pi code (address space protection, valgrind, etc) while getting complete fidelity in how the pi will behave (since we ship the reads and writes to it directly).
-
sonar-int: we take a bit of a fun break, and bang out a quick device driver for a simple sonar device. You will then get a feel for how interrupts can be used to simplify code structure (counter-intuitive!) by adapting the interrupt code from the previous lab to make this code better.