% CVSId: $Id: kernel_oopsing.tex,v 1.4 2003/03/03 08:23:10 mulix Exp $ \documentclass[final, total, pdf, colorBG, slideColor, azure]{prosper} \title{Linux Kernel Debugging} \subtitle{Your kernel just oopsed - What do you do, hotshot?} \author{Muli Ben-Yehuda} \email{mulix@mulix.org} \institution{IBM Haifa Research Lab} \slideCaption{Kernel Debugging, IBM HRL LKDSG 2003} \begin{document} \maketitle \begin{slide}{Kernel Debugging - Why?} \begin{itemize} \item Why would we want to debug the kernel? after all, it's the one part of the system that we never have to worry about, because it always works. \item Well, no. \end{itemize} \end{slide} \begin{slide}{Kernel Debugging - Why?(cont)} \begin{itemize} \item Because a driver is not working as well as it should, or is not working at all. \item Because we have a school or work project. \item Because the kernel is crashing, and we don't know why. \item Because we want to learn how the kernel works. \item Because it's fun! Real men hack kernels ;-) \end{itemize} \end{slide} \begin{slide}{Broad Overview of the Kernel} \begin{itemize} \item Over a million lines of code. \begin{itemize} \item Documentation/ \item drivers/ \item kernel/ \item arch/ \item fs/ \item lib/ \item mm/ \item net/ \item Others: security/ include/ sound/ init/ usr/ crypto/ ipc/ \end{itemize} \end{itemize} \end{slide} \begin{slide}{Broad Kernel Overview (cont)} \begin{itemize} \item Supports runtime loading and unloading of additional code (kernel modules). \item Configured using Kconfig, a domain specifc configuration language. \item Built using kbuild, a collection of complex Makefiles. \item Heavily dependant on gcc and gccisms. Does {\em not} use or link with user space libraries, although supplies many of them - sprintf, memcpy, strlen, printk (not printf!). \end{itemize} \end{slide} \begin{slide}{Read the Source, Luke} \begin{itemize} \item The source is there - use it to figure out what's going on. \item Linux kernel developers frown upon binary only modules, because they don't have the source and thus cannot debug them. \item Later kernels include facilities to mark when a binary only module has been loaded (``tainted kernels''). Kernel developers will kindly refuse to help debug a problem when a kernel has been tainted. \end{itemize} \end{slide} \begin{slide}{Read the Source, Luke (cont)} Use the right tools for the job. Tools to navigate the source include: \begin{itemize} \item lxr - \href{http://www.iglu.org.il/lxr/}{http://www.iglu.org.il/lxr/} \item find and grep \item ctags, etags, gtags and their ilk. \end{itemize} Use a good IDE \begin{itemize} \item emacs \item vi \item One brave soul I heard about used MS Visual Studio! \end{itemize} \end{slide} \begin{slide}{Use the source} The two oldest and most useful debugging aids are \begin{itemize} \item Your brain. \item printf. \end{itemize} Use them! the kernel gives you printk, which \begin{itemize} \item Can be called from interrupt context. \item Behaves mostly like printf, except that it doesn't support floating point. \end{itemize} \end{slide} \begin{slide}{Use the Source (cont)} Use something like this snippet to turn printks on and off depending on whether you're building a debug or relase build. \begin{tiny} \begin{verbatim} #ifdef DEBUG_FOO #define CDBG(msg, args...) do { \ printk(KERN_DEBUG "[%s] " msg , __func__ , ##args );\ } while (0) #else /* !defined(DEBUG_FOO) */ #define CDBG(msg, args...) do {} while (0) #endif /* !defined(DEBUG_FOO) */ \end{verbatim} \end{tiny} \end{slide} \begin{slide}{Use the Source (cont)} \begin{itemize} \item For really tough bugs, write code to solve bugs. Don't be afraid to insert new kernel modules to monitor or affect your primary development focus. \item {\bf Code defensively}. Whenever you suspect memory overwrites or use after free, use memory poisoning. \item Enable all of the kernel debug options - they will find your bugs for you! \item \begin{tiny} \begin{verbatim} #define assert(x) do { if (!(x)) BUG(); } while (0) \end{verbatim} \end{tiny} \item Linux 2.5 has BUG\_ON(). \end{itemize} \end{slide} \begin{slide}{Kernel Debuggers} Linux has several kernel debuggers, none of which are in the main tree (for the time being). The two most common are \begin{itemize} \item kdb - \href{http://oss.sgi.com/projects/kdb}{http://oss.sgi.com/projects/kdb} \item kgdb - \href{http://kgdb.sourceforge.net/}{http://kgdb.sourceforge.net/} \end{itemize} \end{slide} \begin{slide}{KGDB} \begin{itemize} \item Requires two machines, a slave and a master. \item gdb runs on the master, controlling a gdb stub in the slave kernel via the serial port. \item When an OOPS or a panic occurs, you drop into the debugger. \item Very very useful for the situations where you dump core in an interrupt handler and no oops data makes it to disk - you drop into the debugger with the correct backtrace. \end{itemize} \end{slide} \begin{slide}{ksymoops} \begin{itemize} \item Read Documentation/oops-tracing.txt \item Install ksymoops, available from \href{ftp://ftp.il.kernel.org}{ftp://ftp.il.kernel.org} \item Run it on the oops (get it from the logs, serial console, or copy from the screen). \item ksymoops gives you a human readable back trace. \item Sometimes the oops data can be trusted ("easy" bugs like a NULL pointer dereference) and sometimes it's no more than a general hint to what is going wrong (memory corruption overwrite EIP). \end{itemize} \end{slide} \begin{slide}{ksymoops(cont)} \begin{itemize} \item Linux 2.5 includes an "in kernel" oops tracer, called kksymoops. Don't forget to enable it when compiling your new 2.5 kernel! \item It can be found under Kernel Hacking -> Load all symbols for debugging/kksymoops (CONFIG\_KALLSYMS). \end{itemize} \end{slide} \begin{slide}{ksymoops(cont)} \begin{tiny} \begin{verbatim} Unable to handle kernel NULL pointer dereference at virtual address 00000000 printing eip: c014a9cc *pde = 00000000 Oops: 0002 CPU: 0 EIP: 0060:[] Not tainted EFLAGS: 00010202 EIP is at sys_open+0x2c/0x90 eax: 00000001 ebx: 00000001 ecx: ffffffff edx: 00000000 esi: bffffaec edi: ce07e000 ebp: cdbcffbc esp: cdbcffb0 ds: 007b es: 007b ss: 0068 Process cat (pid: 862, threadinfo=cdbce000 task=cdcf7380) Stack: bffffaec 40013020 bffff9b4 cdbce000 c010adc7 bffffaec 00008000 00000000 40013020 bffff9b4 bffff868 00000005 0000007b 0000007b 00000005 420dabd4 00000073 00000246 bffff848 0000007b Call Trace: [] syscall_call+0x7/0xb Code: 89 1d 00 00 00 00 e8 59 fc ff ff 89 c6 85 f6 78 2f 8b 4d 10 \end{verbatim} \end{tiny} \end{slide} \begin{slide}{LKCD} \begin{itemize} \item LKCD - Linux Kernel Crash Dump \item \href{http://lkcd.sf.net}{http://lkcd.sf.net} \item Saves a dump of the system's state at the time the dump occurs. \item A dump occurs when the kernel panics or oopses, or when requested by the administrator. \item Must be configured before the crash occurs! \end{itemize} \end{slide} \begin{slide}{Making sense of kernel data} \begin{itemize} \item System.map - kernel function addresses \item /proc/kcore - image of system memory \item vmlinux - the uncompressed kernel, can be disassembled using objdump(1). \end{itemize} \end{slide} \begin{slide}{User Mode Linux} \begin{itemize} \item For some kinds of kernel development (architecture independent, file systems, memory management), using UML is a life saver. \item Allows you to run the Linux kernel in user space, and debug it with gdb. \item Work is underway at making valgrind work on UML, which is expected to find many bugs. \end{itemize} \end{slide} \begin{slide}{Magic SysRq} \begin{itemize} \item More info at Documentation/sysrq.txt. \item a 'magical' key combo you can hit which the kernel will respond to regardless of whatever else it is doing, unless it is completely locked up. \item CONFIG\_MAGIC\_SYSRQ, echo ``1'' > /proc/sys/kernel/sysrq \item On x86, press 'ALT-SysRq-'. The sysrq key is also known as the 'Print Screen' key. \end{itemize} \end{slide} \begin{slide}{Magic SysRq(cont)} \begin{itemize} \item 'b' - Will immediately reboot the system without syncing or unmounting your disks. \item 'o' - Will shut your system off (if configured and supported). \item 's' - Will attempt to sync all mounted filesystems. \item 'p' - Will dump the current registers and flags to your console. \item 't' - Will dump a list of current tasks and their information to your console. \item 'm' - Will dump current memory info to your console. \item 'h' - The most important key - will display help ;-) \end{itemize} \end{slide} \begin{slide}{Happy Hacking!} \center{Questions? Comments?} \center{Happy Oopsing!} \end{slide} \end{document} %%% Local Variables: %%% mode: latex %%% TeX-master: t %%% End: