dwfl_attach_state alternative taking Ebl?

[Milian Wolff]

On Thursday, March 30, 2017 12:57:33 PM CEST Mark Wielaard wrote:
> On Wed, 2017-03-29 at 23:57 +0200, Milian Wolff wrote:
> > On Mittwoch, 29. März 2017 21:48:08 CEST Mark Wielaard wrote:
> > > Would it help your use case if there was a dwfl_init_state (Dwfl *dwfl,
> > > int e_machine, unsigned char ei_class, unsigned char ei_data, ...)?
> > 
> > What magic values do I pass to e_machine, ei_class, ei_data?
> 
> That would be one of the EM e_machine architecture constants, ELFCLASS32
> or ELFCLASS64 for ei_class and ELFDATA2LSB or ELFDATA2MSB for ei_data.
> (e_machine could arguably be a GElf_Half).
> 
> > I guess the ebl  API that takes the Elf architecture or archicture
> > name would be better.
> 
> I think we should extend the ebl_openbackend calls with a variant that
> takes all three machine/class/data constants. If you look at the
> machines table in libebl/eblopenbackend.c you see that given just the EM
> architecture constant or (emulation) name without an Elf handle given we
> cannot distinguish between e.g. ppc64 (EM_PPC64/ELFCLASS64/ELFDATA2MSB)
> and ppc64le (EM_PPC64/ELFCLASS64/ELFDATA2LSB). You may obviously counter
> that just means that table isn't complete. But then we have to document
> (and maybe export?) the emulation names that people can rely on. Which
> is why I was suggesting we rely on the machine/class/data triple to
> uniquely identify the architecture. Maybe that is inconvenient though?

Ah ok, such a triple would be OK as well, I guess. But see below.

> > > And what exactly is your use case? Maybe we can come up with a better
> > > interface.
> > 
> > The use-case is parsing profiler data, e.g. in perfparser by Ulf / TQC. We
> > don't mess with Elf* anywhere, but need it to let dwfl_attach_state figure
> > out the target architecture. We do know the architecture already so this
> > is a lot of jumping through hoops, to find a fitting Elf* that can be
> > used for dwfl then...
> 
> OK. How do you know the Elf architecture in that case? How and by what
> is it given? Is that an EM constant or some architecture string?

In our case we either get it from perf, or the user specifies it directly on 
the command line. In both cases it is a string like "x86_64", "arm" or 
"aarch64". We map that to a list of architectures we know about, see e.g.:

http://code.qt.io/cgit/qt-creator/perfparser.git/tree/app/
perfregisterinfo.h#n29

So, any API that would allow us to map these architectures directly to a dwfl/
elf counterpart would simplify our code, or at least would make it easier to 
understand, as we wouldn't have to wait for an Elf file we can open before 
calling dwfl_attach_state.

Bye

-- 
Milian Wolff
mail@milianw.de
http://milianw.de