Re: [patch] x86_64 specific function return probes

[Andi Kleen <ak@suse.de>]

On Fri, Jun 03, 2005 at 09:40:26AM -0700, Lynch, Rusty wrote:
> From: Andi Kleen [mailto:ak@suse.de]
> >On Thu, Jun 02, 2005 at 01:58:50PM -0700, Rusty Lynch wrote:
> >> The following patch adds the x86_64 architecture specific
> implementation
> >
> >[....]
> >
> >Thanks for the long description.
> >
> >but...
> >
> >> +struct task_struct  *arch_get_kprobe_task(void *ptr)
> >> +{
> >> +	return ((struct thread_info *) (((unsigned long) ptr) &
> >> +					(~(THREAD_SIZE -1))))->task;
> >> +}
> >
> >and
> >
> >
> >> +	tsk = arch_get_kprobe_task(sara);
> >
> >
> >This is still wrong when the code is not executing on the process
> >stack, but on a interrupt/Exception stack. Any reason you cannot
> >just use current here?
> >
> >-Andi
> 
> Ah... you are talking about if someone registers a return probe on
> something like an interrupt handler, right? 

Yes.

> 
> I was under the impression that I could not always count on the current
> I get from interrupt context to map to the current seen by the target
> function (that triggers the breakpoint.)  It sounds like an invalid
> assumption lead to some extra complexity that isn't correct for all
> cases.


It is an invalid assumption, current works in all contexts. Except
if your GS is broken, but that should only happen when the kernel
is already very crashed.

-Andi