[PATCH v7 5/6] arm64: ftrace: add arch-specific stack tracer

Will Deacon will.deacon at arm.com
Tue Dec 22 01:48:44 PST 2015

On Tue, Dec 22, 2015 at 03:41:03PM +0900, AKASHI Takahiro wrote:
> On 12/21/2015 09:04 PM, Will Deacon wrote:
> >On Tue, Dec 15, 2015 at 05:33:43PM +0900, AKASHI Takahiro wrote:
> >>Regarding a function prologue analyzer, there is no guarantee that we can
> >>handle all the possible patterns of function prologue as gcc does not use
> >>any fixed templates to generate them. 'Instruction scheduling' is another
> >>issue here.
> >
> >Have you run this past any of the GCC folks?  It would be good to at least
> >make them aware of the heuristics you're using and the types of prologue
> >that we can handle. They even have suggestions to improve on your approach
> >(e.g. using -fstack-usage).
> Yeah, I can, but do you mind my including you in CC?
> 'cause I don't know what kind of comments you are expecting.

Sure, I'd be interested to be on Cc. I suspect they will say "we don't
guarantee frame layout, why can't you use -fstack-usage?", to which I
don't have a good answer.

Basically, I don't think a heuristic-based unwinder is supportable in
the long-term, so we need a plan to have unwinding support when building
under future compilers without having to pile more heuristics into this
code. If we have a plan that the compiler guys sign up to, then I'm ok
merging something like you have already as a stop-gap.

Make sense?


More information about the linux-arm-kernel mailing list