Re: backtrace not working on mainline x86_64

[Hien Nguyen <hien at us dot ibm dot com>]

Jose R. Santos wrote:
> bibo,mao wrote:
>
> > In recent x86_64, process kernel stack is different from trap stack, 
> > I modify runtime/stack.c _stp_stack_sprint() function, the original is
> >      __stp_stack_sprint (str, (unsigned long *)&REG_SP(regs), 
> > verbose, 0);
> >      __stp_stack_sprint (str, (unsigned long *)REG_SP(regs), verbose, 
> > 0);
> >
> > It works well for x86_86 in recent kernel version, this is only 
> > temporary method.
> >  
> That works for me.
>
> What would be a good permanent fix for this?  Should check for X86_64 
> and Kernel version be added to _stp_stack_sprint() function?
>
> -JRS
The permanent fix for this would be implement the dump_stack_to_buffer() 
function and push it to the main line (LKML). This is an exported 
function in the kernel.  Systemtap runtime  just invokes this function 
and pass to it a preallocated buffer, it will fill up the buffer with 
stack info (either verbose or non-verbose).

I and Anil worked on a prototype for i386 and ppc64 last year, but did 
not push it to the mainline yet, since could get away with what we are 
doing in the runtime now.

Hien.