Why does my backtrace contain a mixture of function call chains?

So your kernel oopsed and gave you a stack trace that mostly makes sense, but has a function or two on it that do not get called at all in this code path?

This is a normal occurence in Linux. Unless CONFIG_FRAME_POINTER (or later, CONFIG_UNWINDER_FRAME_POINTER is enabled, the function print_context_stack() simply walks the whole stack and looks for any value that might be the address of a function in the kernel. It has no way of knowing whether that address is a stack frame return address from the current code path, a left-over return address from a previous code path or just a random value that was left on the stack.

If you want to always get reliable stack traces when an oops happens, make sure you enable CONFIG_FRAME_POINTER or CONFIG_UNWINDER_FRAME_POINTER, as well as CONFIG_STACKTRACE, CONFIG_UNWIND_INFO and CONFIG_STACK_UNWIND. For x86_64, CONFIG_UNWINDER_ORC is also a reliable stack trace dumper.

With these configuration options switched on, the kernel knows exactly which addresses are part of the current call chain, and which are not.

How do I manually get a stack trace?

You can manually try to create the stack trace from an oops message or from the stack dump of a task. See the ABI page for details on the stack convention for your architecture.


KernelNewbies: FAQ/StackTrace (last edited 2021-01-12 06:14:08 by RandyDunlap)