[PATCH] D32887: [ELF] Emit __ehdr_start when there is a segment containing program headers

[Rui Ueyama via llvm-commits]

On Tue, May 9, 2017 at 2:20 PM, Petr Hosek <phosek at google.com> wrote:

> On Tue, May 9, 2017 at 1:56 PM Rafael Avila de Espindola <
> rafael.espindola at gmail.com> wrote:
>
>> * Add an extra walk, but only if we know the link will fail. I *think*
>>   this was what the previous patch did. IMHO so far this is the best
>>   option.
>>
>
> That's the one I implemented initially.
>
>
>> * Report this particular undefined symbol with less information. I would
>>   save this as an easy last resort.
>>
>
> I considered this option, it'd be easy to implement but it's undesirable.
>
>
>> * Always define the symbol.
>>
>
> I'm fine implementing this as an initial solution.
>

Then let's go with that for now.


> * Report undefined symbols in InputSection::relocateNonAlloc and
>>   InputSectionBase::relocate instead of the current location (requires a
>>   benchmark).
>>
>
> I considered this but haven't implemented it as it's a bigger change. I
> can give it a try to see what would be the impact.
>

This is an interesting approach, I've not thought about that. But it moves
the error checking at very end of the entire link process, so the time to
get an missing symbol error will be longer than it is now. Since missing
symbol is a common error, I think we want to report that as soon as
possible.

* Remember more information during the first scan, that seems the least
>>   desirable.
>>
>
> That's the current implementation.
>
>
>> * Figure out earlier if we are going to allocate the headers. This is
>>   hard and just a conservative heuristic. For example, it is possible to
>>   construct a case where a PT_GNU_RELRO is not needed, but we don't know
>>   that until we known that a .got is not needed.
>>
>
> I tried this one, but it felt too complicated, there was a lot of
> duplication of code and I always managed to come up with a counter-example
> where it didn't work.
>
>
>> We already use the strategy of tentatively creating things and deleting
>> them if we find out we don't need them, so if we can make that work in
>> here that would be my preference.
>
>
> IIUC this would work if we move the reporting to
> InputSection::relocateNonAlloc and InputSectionBase::relocate; we can
> define __ehdr_start unconditionally and later change it to undefined symbol
> if we find out it's needed.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170509/04480b1f/attachment.html>