Re: Stack traces

[Andi Kleen]

On Tue, Aug 08, 2000 at 11:36:31AM +1000, Dave Cecil wrote:
> I have a thread that appears to be hanging in my kernel module (or 
> something called by it) and I want to determine where the problem is 
> occurring.  Unfortunately, printks change the timing of the system, and the 
> problem no longer occurs.
> 
> What I really want is to get a stack trace of the offending thread.  I have 
> used alt-sysrq-t to get some thread information, but it only provides the 
> thread's saved PC (which isn't in my module, according to its address).  Is 
> there a way to get the thread's call stack?  Are there any routines I could 
> use or is there some documentation/code I could look at?
> 
> How do others go about solving such problems?  The only debugger I've found 
> for 2.2.12 kernels that would fit the bill is SGI's kgdb, and there's 
> obviously a bit of overhead involved in setting up the serial debugger.

kdb (no 'g') from SGI can do the same without a serial line.  Then it is 
as simple as btp <pid>

The other option is to compile a vmlinux with debugging symbols (-g), 
do gdb vmlinux /proc/kcore (note that on 2.2 /proc/kcore is a.out so you
may need a gdb that can read a.out or a ELF kcore patch), use the appended
ps macro to find the task structure of your thread, display its tss.esp 
member and backtrace from that.

-Andi

define ps 
set $inittask = &(init_task_union.task)
set $t = $inittask->next_task
while $t != $inittask
	output $t
	output "\t"
        output $t->pid
        output "\t"
        output $t->comm
        echo
        set  $t = $t->next_task
end
end