Andi Kleen wrote:

On Wednesday 23 August 2006 11:01, Zachary Amsden wrote:
  

quoted

Andi Kleen wrote:
    

quoted

quoted

Yes, after discussion with Rusty, it appears that beefing up 
stop_machine_run is the right way to go.  And it has benefits for 
non-paravirt code as well, such as allowing plug-in kprobes or oprofile 
extension modules to be loaded without having to deal with a debug 
exception or NMI during module load/unload.
    
        

I'm still unclear where you think those debug exceptions will come from
      

kprobes set in the stop_machine code - which is probably a really bad 
idea, but nothing today actively stops kprobes from doing that.
    

kprobes don't cause any debug exceptions. You mean int3?

Anyways this can be fixed by marking the stop machine code __kprobes

-Andi
  

I need to look at the kprobes code in more depth to answer completely.  
But in general, there could be a problem if DRs are set to fire on any 
EIP or memory address touched during the critical stop_machine region, 
or int3 breakpoints are set in that code or any code it calls.

Zach