Re: [rfa] Fix software-watchpoint failures by adding epilogue detection

["Ulrich Weigand" <uweigand@de.ibm.com>]

Dan Jacobowitz wrote:

> One of the patches we haven't had a chance to submit took the same
> approach to solve these failures.  I also did ARM mode, as well as
> some additional sequences - probably those used by RealView.  I've
> attached it.

Thanks!  Unfortunately it doesn't fix the problem with our current
ARM compiler, because your heuristics doesn't recognize the sequence
we get as epilogue:

    83c8:       f107 0710       add.w   r7, r7, #16
    83cc:       46bd            mov     sp, r7
    83ce:       bd80            pop     {r7, pc}

On the other hand, your patch does indeed accept a couple of additonal
instructions compared to mine.  I'll come up with a merged approach ...

> We have at least one other patch related to ARM testsuite failures,
> specifically an ERROR (probably infinite backtrace) in gdb1250.exp.
> I can extract it for you if you'd like, but I'm not sure it's in
> condition to be merged.

Hmm, this test is passing now.  We did have problems with backtraces
out of abort, but this was due to a compiler problem -- it simply
would not save the return address at all for no-return functions.
This is now fixed in the compiler ...

Bye,
Ulrich

-- 
  Dr. Ulrich Weigand
  GNU Toolchain for Linux on System z and Cell BE
  Ulrich.Weigand@de.ibm.com