Re: [RFC PATCH 1/2] x86: Fix rbp saving in pt_regs on irq entry

[Frederic Weisbecker <fweisbec@gmail.com>]

On Fri, Jan 07, 2011 at 01:31:30PM +0100, Ingo Molnar wrote:
> 
> * Jan Beulich <JBeulich@novell.com> wrote:
> 
> > > Now I don't understand how this is all useful as this is not a normal proc but 
> > > an interruption. We can't get back the return address from the CFA. Or am I 
> > > missing something?
> > 
> > Unwind annotations, when written correctly, allow unwinding through all kinds of 
> > execution flows, including interrupts or exceptions as well as including stack 
> > switches.
> 
> Yeah and that's rather useful, as exception contexts can nest in very weird ways, 
> especially with NMIs involved. For example a 7-context combination is possible:
> 
>   user-space -> syscall -> pagefault -> softirq -> hardirq -> debug trap -> NMI
> 
> And the call frame walking logic needs to be able to get all the way back to 
> user-space ...
> 
> For that every transition needs to work flawlessly, for debugging (and CFI based 
> profiling) to work fine.
> 
> Most of those transitions can happen at any instruction boundary that a given 
> context executes, so the total number of possible combinations is virtually endless.
> 
> Unfortunately we dont seem to have a good way to test any of this automatically. 
> Putting a perf probe on every assembly instruction perhaps, and checking whether the 
> frame manages to go back all the way to user-space?

May be.

Once I'll have perf callchain based on CFI ready, we'll perhaps find some issues
there. Although I guess there are already tools that can make use of that.

> 
> Thanks,
> 
> 	Ingo