Debugging Embedded Linux Platforms with GDB and Python
If you write code for Linux systems, chances are you will have used the venerable GNU Debugger (GDB). Acting as a back end for many GUIs and the interface to various JTAG debugging tools in the embedded world, GDB is the foremost debugger for Linux. As of release 7.0, GDB gained a compelling new capability: support for scripting debugging actions using a Python interpreter. In this article, I take a look at how to drive GDB using Python and apply this knowledge to the vexatious issue of debugging an embedded Linux platform.
Debugging Linux programs on an x86 PC platform, although not necessarily easy, at least is well supported by a variety of tools. Most Linux distributions package development and debugging tools to assist with anything from profiling runtime performance through tracing memory leaks and detecting deadlocks.
Embedded platforms are rarely so well served. Although a number of projects seek to provide the kind of polish and integration for embedded development that is taken for granted on the desktop, these are not yet widely adopted in all areas of embedded Linux development. Many embedded devices are developed using what effectively is a handcrafted Linux distribution, closely tied to the specific goals of that device. The time required to integrate a wide range of handy debugging tools into that environment, especially in the fast-paced world of consumer electronics, is an overhead few teams can meet.
Many embedded platforms seek to save on resource overhead through the use of “low-fat” system libraries (such as uClibc in the place of glibc), which may make the integration of some debugging tools more difficult. Indeed, in some cases, the architecture of the CPU used by the target platform will prevent the use of certain tools altogether. The excellent Valgrind instrumentation framework, for example, has limited support for the ARM architecture and no support at all for MIPS or SuperH.
The nature of embedded devices often means that CPU cycles and memory are scarce. Any debugging tool that weighs too heavily on either may make its use impractical on an embedded device, especially when attempting to debug race conditions and the like.
The net result of this inconsistent provision of debugging tools across the embedded Linux world is that most developers have to makeshift as best they can with the tools that are available. Happily, GDB is widely available for embedded devices because it is easy to cross-compile and supports a wide range of target architectures. And with the recent integration of Python scripting support, GDB can extended beyond the typical debugging tasks of single stepping and variable examination.
GDB has long supported extension via pre-canned sequences of debugger commands. This ability makes it possible to automate certain parts of a debugging work flow and even implement new debugger functions.
Integrating Python into GDB adds an extra dimension to the possibilities of GDB scripting and extension. In addition to the simple functions and flow control of GDB's native scripting language, the full power of the Python language is made available.
The Python GDB API is presented as a Python module called gdb, which provides access to GDB's internal representation of a process under debugging. The module includes interfaces to process information, threads, stack frames, values and types, functions, symbols and break points. In addition, a mechanism is provided to inject commands into the GDB command-line interface.
The result is that the internals of GDB are now available as a rich set of libraries for programmatic driving of the debugger. This creates a whole range of new opportunities for extension and automation. For example, let's imagine you want to debug calls to malloc() in a large application, but you're really interested only in calls from a certain module. Ideally, you want to be able to break execution only when one of the module's functions is in the backtrace at the point that malloc() is called. The Python API gives you that flexibility.
To explore the use of Python within GDB, let's debug a small C program, the code for which is shown in Listing 1. It performs the simple task of printing the phrase “Hello World!” in a rather convoluted manner, and it has at least one obvious bug. Besides being over-engineered for the task at hand, hello_world.c makes use of two mutexes for serializing access to different data structures, and not all users of these mutexes agree on the order in which locks should be acquired. This quickly yields a runtime deadlock.
- A Little GUI for Your CLI
- Android Candy: Google Keep
- Readers' Choice Awards 2014
- Handling the workloads of the Future
- How Can We Get Business to Care about Freedom, Openness and Interoperability?
- diff -u: What's New in Kernel Development
- Synchronize Your Life with ownCloud
- Days Between Dates?
- December 2014 Issue of Linux Journal: Readers' Choice