Re: [RFC] pb unwinding from pthread_cond_wait on ppc-linux (RFA?)

[Joel Brobecker <brobecker at adacore dot com>]

Hi Daniel,

> > On Thu, Dec 09, 2004 at 05:00:17PM +0100, Joel Brobecker wrote:
> > > > Precisely!  That's what I thought it would be.  It's trying to load lr
> > > > with the address of @+16, so that the function can access PIC data
> > > > using PC-relative displacement.
> > > 
> > > Daniel, you never stop to impress me.
> > > 
> > > > (Does this obsolete the "branch in first three insns" check?  I'm not
> > > > sure if there are other possible reasons for that.)
> > > 
> > > Here is a new patch that implements your suggestion. Indeed, I could
> > > then remove the "branch in first three insns" check...
> 
> FYI, rethinking this, this is not such a good idea (removing the check,
> I mean).  While the check itself is pretty bogus, the comment above
> says:
>         {                       /* bl foo, 
>                                    to save fprs??? */
> 
> I know at least Darwin does this.
> 
> So maybe the new check should be additional instead of a replacement.

I am really sorry for dropping the ball on this one. I have reorganized
my working schedule in an attempt to allow me to get back to some tasks
that I left, so hopefully it'll work out.

If Darwin does this, then you are probably right that the test should
not be a replacement but an addition.

Unfortunately, I no longer have access to a ppc-linux machine. I did
the work back then as an evaluation of this port, but we ended up
not adding this platform to our list of supported ones. Would you
be able to do the testing?

I can test on ppc-aix, though, but I'm thinking that this will likely
not have much impact.

-- 
Joel