Hopefully Simple Question on Debugging Kernel Modules

[Scott Long]

John Baldwin wrote:
> On Monday 23 February 2009 12:10:07 pm Scott Long wrote:
>> John Baldwin wrote:
>>> On Friday 20 February 2009 6:40:56 pm David Christensen wrote:
>>>> I'm sure this is a simple question but the answer is alluding my Google
>>>> search capabilities.  My driver is being loaded as a kernel module and
>>>> is failing with the following error:
>>>>
>>>> Fatal trap 12: page fault while in kernel mode
>>>> cpuid = 0; apic id = 00
>>>> fault virtual address   = 0xfffffffe40abe9dc
>>>> fault code              = supervisor write data, page not present
>>>> instruction pointer     = 0x8:0xffffffff920b638f
>>>> stack pointer           = 0x10:0xffffffff9212bb10
>>>> frame pointer           = 0x10:0xffffffff9212bbb0
>>>> code segment            = base 0x0, limit 0xfffff, type 0x1b
>>>>                         = DPL 0, pres 1, long 1, def32 0, gran 1
>>>> processor eflags        = interrupt enabled, resume, IOPL = 0
>>>> current process         = 12 (irq268: bce0)
>>>> [thread pid 12 tid 100166 ]
>>>> Stopped at      bce_intr+0x8df: addl    $0x1,0x2c854(%r12,%rax,4)
>>>> db>
>>>>
>>>> I simply need to find the offending source line in my driver.  Not sure 
>>>> how I've managed to get the driver running at all without this but it's 
>>>> time to do things the right way.  I have KDB/DDB/GDB built into my 
>>>> -CURRENT kernel already.  It'd be great to find the source line while in
>>>> the kernel debugger but I'm also fine with rebooting the system to 
>>>> identify the line number.
>>> Just use gdb on bce.ko (built with debug symbols):
>>>
>>> gdb /path/to/if_bce.ko
>>> (gdb) l *bce_intr+0x8df
>>>
>>> If you get a crashdump you can run kgdb on it and just walk up to the 
> relevant 
>>> stack frame and use 'l' there to get a listing.
>>>
>> One thing that I've never figured out is how debugging symbols are 
>> handled in module builds these days.  If I go to /sys/modules/bce and
>> do 'make', it generates a .ko and explicitly strips it.  I wind up
>> having to re-run the link command by hand so I get symbols.  What is
>> the correct way to do this?  Note that I'm not interested in answers
>> that involve "go to /usr/src and run make buildkernel" =-)
> 
> make DEBUG_FLAGS=-g is what I use.  The same thing works for userland tools 
> and the kernel (usually we put 'makeoptions DEBUG_FLAGS=-g' in a kernel 
> config so it is "automatic" for kernels though).
> 

Ah, I was still using 'CFLAGS+= -g".  Thanks to you and Mr. Campbell for
the tip.

Scott