On Tue, Apr 28, 2015 at 04:08:42PM +0200, Peter Zijlstra wrote:
> On Tue, Apr 28, 2015 at 09:04:54AM -0500, Josh Poimboeuf wrote:
> > On Tue, Apr 28, 2015 at 02:16:06PM +0200, Peter Zijlstra wrote:
> > > On Mon, Apr 27, 2015 at 08:56:27AM -0500, Josh Poimboeuf wrote:
> > > > Frame pointer based stack traces aren't always reliable. One big reason
> > > > is that most asm functions don't set up the frame pointer.
> > > >
> > > > Fix that by enforcing that all asm functions honor CONFIG_FRAME_POINTER.
> > > > This is done with a new stackvalidate host tool which is automatically
> > > > run for every compiled .S file and which validates that every asm
> > > > function does the proper frame pointer setup.
> > >
> > > Would it make sense (maybe as an additional CONFIG_*_DEBUG thing) to
> > > also process the output of GCC with this tool? To both double check GCC
> > > and to give the tool more input?
> >
> > I tried that, but I discovered that gcc's usage of frame pointers would
> > be a lot harder to validate. It only sets up the frame pointer in code
> > paths which have call instructions. There are a lot of functions which
> > have conditional jumps at the beginning which can jump straight to a
> > return instruction without first doing the frame pointer setup.
>
> Hmm, would not such code break your patching?
No, because we'll also do some runtime stack validation (which will be a
future patch set). If we detect preemption or an irq frame on the
stack, we'll assume the stack is unreliable and delay the patching of
the task (*). Otherwise the stack will only consist of calls down to
schedule() which will be guaranteed to have frame pointers.
(*) This can be even further improved by making _all_ stacks reliable if
we can ensure that dwarf call frame information is reliable (more
future patch sets).
--
Josh
--
To unsubscribe from this list: send the line "unsubscribe live-patching" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
