Part I · C, and what the machine does with it¶
Part header
What this part is for¶
Four chapters. One of them teaches reading what a compiler produced, and the other three have one job between them: enough C to read a kernel and change it, and no more.
The limit is deliberate. C is a large language and most of it never appears in the code this book reads, so a part that set out to cover the language would spend most of its length on things you are never going to meet here.
What stops people reading kernel code is not C’s syntax. It is the memory model your language was built to hide: memory is one array of bytes, everything in it has an index, and a type is mostly a statement about how wide a step is. Every language has that underneath. Yours decided you did not need to see it; C decided you did.
This part is first because everything after it is written in C and half of it asks you to change some. Part II builds the machine’s primitives in C and its problems have you rewriting a trap handler; Part IV reads a kernel; Part V compiles C and reads what came out. None of those is an exercise in the language. All of them are impossible without the model.
What it leaves out¶
Control flow, operators and functions. You already have these, and C spells them much as your language does. Where it differs the chapters say so and move on.
The standard library, which is ch04’s subject from the opposite direction: a kernel has almost none of it, and that chapter asks what it does instead. Floating point, for the same reason — the kernel does not use it, and why it does not is a better question than how it works.
And the craft of C as a language you would build something in: ownership discipline, API design, build systems, the long list of ways to invoke undefined behaviour. The problems in this book ask you to change a program that exists and predict what the change does — which needs the model and almost none of the craft. That is not a soft option. It is everything the rest of the book needs, and three chapters can reach it, where the craft would take far longer.
Where to start¶
Read the four in order. ch02 is short and assumes no C, and every chapter after it puts a listing in front of you. If you already write C, ch03 is the memory model you already have — skim it rather than skip it, because ch04 is where it stops being the C you know.
ch03 does not open with pointers: it opens with one complete program that prints four numbers, and the pointers arrive afterwards as a way of explaining numbers you have already watched appear.
Which machine, and what it cannot tell you¶
Most of this part is compiled for RISC-V and run on the xv6 kernel under QEMU; ch02 also compiles for the board’s architecture and runs nothing at all. Three of the four chapters ask you to read the disassembly of code you wrote — ch02 for both of the book’s instruction sets, the other two for RISC-V.
Nothing in this part is timed, and nothing in it may be. QEMU models no cache, no branch predictor, no store buffer and no memory latency, so a duration measured inside it is a fact about the laptop running the emulator, not about the machine being emulated. The figures here are therefore all structural — what a program printed, what instructions the compiler emitted — and the question of what any of it costs is not asked until Part V, on hardware that can answer it.
Where this leaves you¶
Not a C programmer. Able to read a C program, and to change one and say what the change will do — which is what the rest of the book requires, and Part II starts requiring it immediately.