[Mips}Using DT tags for handling local ifuncs

[Jack Carter]

For the record, If it doesn't make a noticeable performance hit
I would like all of the GOT fixups to be explicit with relocations. Any gains made
by making ifunc symbols implicit are quickly lost invoking the resolver. 

We wouldn't even have to have the magic dynamic symbol table indexed to 
the order of the globals in the GOT anymore. If the relocation needs a symbol
it should have the index into the dynamic symbol table for that symbol.

Suddenly there are no more DTs for local and global GOT. There would possible
still be  DTs for special entries GOT[0-n].

I'm probably forgetting something fundamental here. It seems too easy.

But for now I will advance on the latest mips ifunc agreement.

Jack
________________________________________
From: Maciej W. Rozycki [macro@codesourcery.com]
Sent: Friday, January 10, 2014 10:32 AM
To: Richard Sandiford
Cc: Jack Carter; binutils@sourceware.org; Doug Gilmore
Subject: Re: [Mips}Using DT tags for handling local ifuncs

On Thu, 19 Dec 2013, Richard Sandiford wrote:

> > Dealing with the ifunc "local" entries implicitly will save a
> > relocation lookup, a tiny blip of time in relation to the other costs
> > of calling the resolver. So I am arguing about how many angels can
> > dance on a pin.
>
> Yeah, maybe this is one we'll have to agree to disagree on.  I think the
> benefit of having an implicitly-relocated irelative region is small at best.
> I like the generality of including the GOT R_MIPS_IRELATIVE GOT
> relocations in the general .rel.dyn pool and sorting them accordingly,
> because it feels more future-proof.  I also think an implicit region is
> harder to handle in a backward-compatible way, since if we just add new
> tags, older ld.sos would ignore them and not flag an error.
>
> If someone else has any opinions about implicit irelative relocs vs.
> explicit irelative relocs though, please shout.

 I do, I think that while looking smart at first the concept of implicit
GOT relocation on the MIPS target proved inflexible and difficult to
maintain over the years (see e.g. the troubles with the __ehdr_start
special symbol we had recently).  By using explicit relocs the
inflexibility is removed and the likelihood of the need to add another
special region in the GOT in the future minimised.  I haven't analysed any
program startup performance implications this may have though.

 NB as far as ld.so's backwards compatiblity is concerned I think we can
sort it out in either case, e.g. we could poison binaries with a dummy
relocation old ld.so would choke upon.  Or there might be another,
prettier way possible too, to be found by examining ld.so and seeing its
abnormal termination conditions.

  Maciej