Re: [Libunwind-devel] [RFC PATCH 0/3] Add support for dwarf compat mode unwinding

[Jiri Olsa <jolsa@redhat.com>]

On Mon, Feb 03, 2014 at 04:28:06PM +0100, Jean Pihet wrote:
> Hi Arun,
> 
> On 22 January 2014 06:15, Arun Sharma <arun@sharma-home.net> wrote:
> > On Tue, Jan 21, 2014 at 12:16 AM, Jean Pihet <jean.pihet@linaro.org> wrote:
> >
> >>>> This is the case when e.g. profiling an ARMv7 binary that runs on an
> >>>> ARMv8 (aka AARCH64) platform.
> >>>
> >>> Why not configure libunwind for ARMv7 in that case?
> >>>
> >>> I'm trying to understand how is your use case different from using
> >>> x86_32 libunwind to do local unwinding on a x86_64 machine.
> >> Can you give more details on the use case and how to configure
> >> libunwind?
> >
> > You could either build libunwind on a ARMv7 machine and copy it over
> > or setup a 32 bit chroot on a 64 bit kernel and compile libunwind
> > inside that.
> >
> >> The provided patches dynamically detect the target binary
> >> address size and parse the debug info accordingly, all that in a
> >> single library.
> >
> > This is generally called cross-unwinding in libunwind lingo. Some
> > description here:
> >
> > http://www.nongnu.org/libunwind/man/libunwind%283%29.html#section_4
> >
> >>
> >> Note: my use case is to call libunwind from the perf utility in order
> >> to unwind from the dwarf info.
> >>
> >
> > You could link in two copies of libunwind into the perf binary:
> > * libunwind.a for local (host == target) unwinding
> > * libunwind-arm.a for 32 bit cross-unwinding
> >
> > Doing cross-unwinding requires you to write a bunch of "accessors" on
> > how to access the address space of a non-local thread.
> >
> > Something like ./configure --target=arm on aarch64.
> 
> Thanks for the link and info.
> 
> Is there a concrete example of cross-unwinding with multiple targets,
> for example on x86_64 using native and x86_32 libunwind libraries
> simultaneously?
> I am trying to assess the impact of multiple unwinding libs in the perf code.
> 
> Jiri, Arnaldo,
> How is that done on x86? I do not think this can be done with the
> current perf code, am I correct?

correct, but it sounds cool ;-)

jirka