infrun: step through indirect branch thunks

  Hello Pedro,

Thanks for your review.

> > With version 7.3 GCC supports new options
> >
> >    -mindirect-branch=<choice>
> >    -mfunction-return=<choice>
> >
> > The choices are:
> >
> >     keep                behaves as before
> >     thunk               jumps through a thunk
> >     thunk-external      jumps through an external thunk
> >     thunk-inline        jumps through an inlined thunk
> >
> > For thunk and thunk-external, GDB would, on a call to the thunk, step into the
> > thunk and then resume to its caller assuming that this is an undebuggable
> > function.  On a return thunk, GDB would stop inside the thunk.
> 
> I was expecting to see the testscase looping over all possible
> combinations, but only "thunk" is tested, it seems.  Why is that?

Thunk and thunk-external result in the same code, except that the former
contains the thunk in a comdat section and the latter only contains a reference
to an external thunk that would need to be provided separately.

An inline thunk will be part of the caller and won't trigger GDB's
step-over-undebuggable heuristic.

And keep would not use a thunk so it should be covered by existing stepping
tests.

> > The tests assume a fixed number of instruction steps to reach a thunk.  This
> > depends on the compiler as well as the architecture.  They may need
> adjustments
> > when we add support for more architectures.  Or we can simply drop those
> tests
> > that cover being able to step into thunks using instruction stepping.
> 
> The tests sound useful, but isn't there some way we can make them more
> robust to compiler's whims?  Maybe an upper-bounded number of instruction
> steps
> until some pattern?

That might work.

> Guess all this code could be shared betwee 32-bit/64-bit,
> if you made this take the names array and a range as parameters:

So far, amd64-tdep.c was using functions from i386-tdep.c but it looks
cleaner if we put those into a common x86-tdep.c.  I'm adding one.

> > +/* Check whether PC lies inside an indirect branch thunk.  */
> > +
> > +static int
> > +in_indirect_branch_thunk (struct gdbarch *gdbarch, CORE_ADDR pc)
> > +{
> > +  if (!gdbarch_in_indirect_branch_thunk_p (gdbarch))
> > +    return 0;
> 
> Do we need to check the _p predicate elsewhere?  Why not just
> make the default return false, and always call the hook?

OK.

> Should the gdb.base/ testcase have tests for stepping into the thunks?

It is using the same code path as for the record tests.  But that holds for
all the new gdb.base/ tests I added.  I'll add it.

Below is the updated patch.  I ran the new tests on 64-bit and 32-bit IA.
I previously ran the full suite for v1 of this patch on 64-bit IA.

While regenerating gdbarch, I noticed that it generated the following
unrelated changes:

Intel Deutschland GmbH
Registered Address: Am Campeon 10-12, 85579 Neubiberg, Germany
Tel: +49 89 99 8853-0, www.intel.de
Managing Directors: Christin Eisenschmid, Christian Lamprechter
Chairperson of the Supervisory Board: Nicole Lau
Registered Office: Munich
Commercial Register: Amtsgericht Muenchen HRB 186928

infrun: step through indirect branch thunks

Commit Message

Comments

Patch