Re: Anyone working on ftrace function graph support on ARM?

[Ingo Molnar <mingo@elte.hu>]

* Russell King - ARM Linux <linux@arm.linux.org.uk> wrote:

> On Tue, Mar 24, 2009 at 11:48:58PM +0100, Frederic Weisbecker wrote:
> > On Tue, Mar 24, 2009 at 06:29:03PM -0400, Abhishek Sagar wrote:
> > > Instead of just restoring the old backed-up args, lr can be fixed up
> > > inside the entry hook to point to the return hook. So when the traced
> > > function returns, it actually returns to the return hook (where we can
> > > restore the original return address). This means that
> > > -finstrument-functions is not required at all. This is analogous to how
> > > kretprobes work. The only difference here is that instead of planting a
> > > kprobe at the function entry and redirecting the function return to the
> > > profiling exit routine, we can use mcount. This is slightly more
> > > complicated to implement but can be a very efficient alternative to
> > > kretprobes.
> > > --
> > > Abhishek
> > > 
> > 
> > Indeed, you need to override lr, that even the only solution.
> > I was still thinking in an x86 way with its on stack return address.
> 
> As pointed out in my previous mail, identifying where on the stack 
> the return address is stored is only possible for OABI with frame 
> pointers.
> 
> EABI will probably be possible with the stack unwinding code, but 
> it probably won't be cheap.  The EABI unwinder is scheduled for 
> merging during the present now-open merge window.

Unwinding is not realistic or desired for the function tracer - it 
runs in every kernel function so performance is paramount.

So, if i understood you correctly, an OABI_COMPAT and FRAME_POINTERS 
dependency has to be added to the ARM HAVE_FUNCTION_GRAPH_TRACER 
Kconfig rule.

	Ingo