linking error while trying to add print_stack_trace

Question

I am getting a error that looks like a linking problem, after adding a call to "print_stack_trace" in one of the exception handlers inside my embedded Linux kernel code.

Details are as below:

1. My objective & context: I am trying to debug an "imprecise external abort" memory fault during my kernel's boot-up on my A9 board. I am getting the imprecise abort error as below:

   [    1.248680] Unhandled fault: imprecise external abort (0x406) at 0xc397ffec
   

My objective is to try and gather more details of the stack and registers when this abort is occurring. Once the abort occurs, I noted that it is getting directed to a handler called asmlinkage void __exception do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs).

I have called dump_stack() inside this function, which gives me the following stack and register data:

[    1.245755] CPU: 0 PID: 240 Comm: hotplug Not tainted 3.14.26-ts-armv7l #3
[    1.245845] [<c0011fe0>] (unwind_backtrace) from [<c0010934>] (show_stack+0x10/0x14)
[    1.245869] [<c0010934>] (show_stack) from [<c00083c0>] (do_DataAbort+0x40/0x9c)
[    1.245906] [<c00083c0>] (do_DataAbort) from [<c0277f74>] (__dabt_usr+0x34/0x40)
[    1.245919] Exception stack(0xc397ffb0 to 0xc397fff8)
[    1.245935] ffa0:                                     00000000 00000000 00000000 00000000
[    1.245952] ffc0: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
[    1.245969] ffe0: 00000000 be95fe80 00000000 b6f98eec 00000010 ffffffff
[    1.245985] Unhandled fault: imprecise external abort (0x406) at 0xc397ffec

I can note from the above data that address ffec corresponds to b6f98eec. I have no idea what value is this that causes the abort. I ran a nm and objdump against my Linux image (vmlinux file) and could not trace this to any symbol address. Now, I am trying to call save_stack_trace() and print_stack_trace() from inside the same handler, to obtain more details about this abort.

2. Changes I have done: I added save_stack_trace() and print_stack_trace() as follows: At the top of the file, I have the following: Header file inclusions =>

   #include<linux/stacktrace.h>
   #include<asm/stacktrace.h>
   
   #ifdef CONFIG_STACKTRACE
   static unsigned long stack_entries[32];
   
   static struct stack_trace my_trace = {
       .nr_entries = 0,
       .entries = &stack_entries[0],
       .max_entries = 32,
       .skip = 0
   };
   #endif
   

Inside the function do_DataAbort, I have added the following code:

#ifdef CONFIG_STACKTRACE
    save_stack_trace(&my_trace);
    printk(KERN_ALERT "calling print_stack \n");
    print_stack_trace(&my_trace, 32);
#endif

At the top of the header file include/Linux/stacktrace.h, I have tried to define CONFIG_STACKTRACE.

#define CONFIG_STACKTRACE

3. My observation and problem: When I compile this code, I am getting an error that looks like a linker error:

   arch/arm/mm/built-in.o: In function 'do_DataAbort':
   :(.exception.text+0x58): undefined reference to 'print_stack_trace'
   make[1]: *** [vmlinux] Error 1
   

Request your help to resolve this. Please let me know if any information is required.

Answer 1

Never define CONFIG_* macros manually:

// DO NOT DO THAT!
#define CONFIG_STACKTRACE

They are intended to be set (or not set) during kernel's configuration (make menuconfig, make defconfig or other *config targets) and to be stored into .config file.

A configuration parameter may depends on target machine(architecture) or other parameters, without which functionality, defined by given parameter, will not work correctly (or even will not compile, like in your case).

---

If you need some configuration parameter to be set, one way could be set it via make menuconfig:

1. Press key /, a dialog Search Configuration Parameter will appear.
2. Write the name of the configuration parameter (case insensitive, CONFIG_ prefix is not required). E.g., enter stacktrace. Press OK.
3. In the search ouput find exact name of your parameter: Symbol: STACKTRACE.
4. Current value of the parameter will be given in []. If it is [=y] or [=m], then the symbol is already defined in the kernel. If it is [=n], then check, how to set given symbol.
5. Symbols, which can be set by the user, have Prompt: line, which describes location of the symbol in the menu. Note, that symbol is shown in the menu only when expression under Dependens on: is saticfied. E.g., STACKTRACE symbol can be set only when STACKTRACE_SUPPORT symbol is set. Value of dependent symbols is already shown in the expression.
6. Symbols, which have no Prompt: line, cannot be set by the user and can only be set automatically: in the arch-specific way, or being selected (Selected by:) by other symbols. On x86 symbol STACKTRACE_SUPPORT is set, arm also defines this symbol.
7. If any of configuration has been changed, kernel needs to be rebuilt.

Answer 2

You should not just add a #define CONFIG_STACKTRACE. You have to enable this functionality in your kernel config instead. This will ensure the needed functionality is built into your kernel.