[PATCH v2] bfd: fill in PE load config directory entry.

[Jeremy Drake]

On Wed, 12 Mar 2025, LIU Hao wrote:

> 在 2025-3-12 16:55, Jan Beulich 写道:
> > On 11.03.2025 20:28, Jeremy Drake wrote:
> > > I'm sorry, I was having a hard time understanding what you were
> > > suggesting.  I *think* a light bulb just went on, that you are suggesting
> > > issuing an error if h1 is NULL (the outer if).  This is not what I
> > > intended: it is not an error if there is no symbol _load_config_used, it
> > > should just not fill in the load config directory.  However, if the symbol
> > > does exist, it should be of the correct type and have sufficient data
> > > available at it to read the size.  Does this make sense, or am I
> > > misunderstanding either the code or your feedback?
> >
> > Hmm, yes, makes sense. But then why does the diagnostic say "is missing"?
> >
> > Jan
>
> I suspect that message was copied from `_tls_used`.

Yes, exactly.  I just copied the _tls_used block and adjusted to
_load_config_used and the load config directory.  Actual
new code is getting the size out of the section, instead of being a
hardcoded size based on 32/64 bit.

>
> There are two issues in that message: First is that it should have said
> `_tls_used` which is what should be defined in source code. Second is that it
> looks like the error is triggered when the symbol is defined in an unexpected
> way, either it's a common (tentative?) definition or it has no output section.
> So it should be
>
>    (_("%pB: unable to fill in DataDictionary[9] because _tls_used is not
> defined correctly"),
>
> and likewise for `_load_config_used`.
>
> There's another minor issue I think, which is these symbols cannot be weak.

If there's agreement on a different message I can update and send a new
version of the patch.