Book Image

Linux Kernel Debugging

By : Kaiwan N. Billimoria
Book Image

Linux Kernel Debugging

By: Kaiwan N. Billimoria

Overview of this book

The Linux kernel is at the very core of arguably the world’s best production-quality OS. Debugging it, though, can be a complex endeavor. Linux Kernel Debugging is a comprehensive guide to learning all about advanced kernel debugging. This book covers many areas in-depth, such as instrumentation-based debugging techniques (printk and the dynamic debug framework), and shows you how to use Kprobes. Memory-related bugs tend to be a nightmare – two chapters are packed with tools and techniques devoted to debugging them. When the kernel gifts you an Oops, how exactly do you interpret it to be able to debug the underlying issue? We’ve got you covered. Concurrency tends to be an inherently complex topic, so a chapter on lock debugging will help you to learn precisely what data races are, including using KCSAN to detect them. Some thorny issues, both debug- and performance-wise, require detailed kernel-level tracing; you’ll learn to wield the impressive power of Ftrace and its frontends. You’ll also discover how to handle kernel lockups, hangs, and the dreaded kernel panic, as well as leverage the venerable GDB tool within the kernel (KGDB), along with much more. By the end of this book, you will have at your disposal a wide range of powerful kernel debugging tools and techniques, along with a keen sense of when to use which.
Table of Contents (17 chapters)
1
Part 1: A General Introduction and Approaches to Kernel Debugging
4
Part 2: Kernel and Driver Debugging Tools and Techniques
11
Part 3: Additional Kernel Debugging Tools and Techniques

Using the trace-cmd, KernelShark, and perf-tools ftrace frontends

There's no doubt that the Linux kernel ftrace infrastructure is immensely powerful, enabling you to look deep inside the kernel, throwing light into the dark corners of the system, as it were. This power does come at the cost of a somewhat steep learning curve – lots of sysfs-based tuning and options knobs that you need to be intimately aware of, plus the burden of filtering a possibly huge amount of noise in the resulting traces (as you'd have already learned from the previous sections of this chapter!). Steven Rostedt thus built a powerful and elegant command-line-based frontend to ftrace, trace-cmd. What's more, there's a true GUI frontend to trace-cmd itself, the KernelShark program. It parses the trace data recorded (trace.dat by default) by trace-cmd and displays it in a more human-digestible GUI. In a similar manner, Brendan Gregg has built the perf-tools script-based frontend project...