Re: usability issues with inlining and backtraces

[Milian Wolff <milian.wolff@kdab.com>]

[-- Attachment #1: Type: text/plain, Size: 2376 bytes --]

On Mittwoch, 17. August 2016 00:13:10 CEST Namhyung Kim wrote:
> On Mon, Aug 15, 2016 at 10:13:24AM -0700, Andi Kleen wrote:
> > Milian Wolff <milian.wolff@kdab.com> writes:
> > > On Monday, August 15, 2016 11:32:33 AM CEST Milian Wolff wrote:
> > >> On Monday, June 13, 2016 6:07:13 PM CEST Milian Wolff wrote:
> > >> > Ping?
> > >> > 
> > >> > Andi, I'd be very interested in your answers on my questions below.
> > >> > Most
> > >> > notably, `srcline` with `--call-graph ...,address` seems to be highly
> > >> > buggy
> > >> > for me.
> > >> 
> > >> Reping again.
> > >> 
> > >> Andi, or anyone else - I would really appreciate some more feedback on
> > >> my
> > >> questions below.
> > >> 
> > >> I esp. wonder how to leverage srcline and addresses in the call-graph,
> > >> as
> > >> that has the potential to greatly increase perf's value to me.
> > > 
> > > OK, I have found at least two culprits that partially explain the
> > > behavior.
> > > 
> > > On one hand, I'm simply missing debug information for some libraries,
> > > such as libm, libc etc. pp. Due to that, the addr2line step fails and
> > > sym+offset is displayed.
> > > 
> > > On the other hand, there seems to be a bug in perf:
> > > 
> > > When hist_entry__get_srcline calls get_srcline I see from debug output
> > > 
> > > he->ip=307e, sym=main -> rip_obj2dump=40307e -> mandelbrot.h:39
> > > 
> > > But in callchain_list__sym_name, I see:
> > > 
> > > cl->ip=40373d, sym=main -> rip_obj2dump=80373d-> main+8390413
> > > 
> > > So it seems like the callchain_list is missing a normalization step that
> > > is
> > > applied to the IP addresses stored in hist_entry. Can someone point me
> > > into
> > > the right direction?
> > > 
> > > If I change callchain_list__sym_name to pass cl->ip - cl->ms.map->start
> > > to
> > > map__rip_2objdump this feature works much better for me, but I doubt
> > > that this is the correct fix.
> > 
> > Yes it's something like that. Namhyung or Jiri probably know
> > what the right fix would be.
> 
> I think the unwind libs should set entry->ip to al.addr instead of
> ip.  The __hists__add_entry() sets hist_entry->ip using al->addr.

Thank you, that works. I've submitted a patch accordingly.

Cheers

-- 
Milian Wolff | milian.wolff@kdab.com | Software Engineer
KDAB (Deutschland) GmbH&Co KG, a KDAB Group company
Tel: +49-30-521325470
KDAB - The Qt Experts

[-- Attachment #2: smime.p7s --]
[-- Type: application/pkcs7-signature, Size: 5903 bytes --]