Re: php73 build failure

[Martin Husemann <martin%duskware.de@localhost>]

On Tue, Oct 15, 2019 at 11:41:06AM +0000, maya%NetBSD.org@localhost wrote:
> As background:
> 
> relro:
> resolve all the relocations immediately, mark the table read-only
> 
> ifunc:
> on your first attempt to run this function, first run the resolver,
> decide which function it should resolve to*, then run that function.
> 
> * This means modifying the relocation table, which is read-only.
> This segfaults.
> 
> So, it's entirely in ld.so's interpretation of what bind now and ifunc
> mean, but what's the correct way to handle this scenario?

With both I would say the only reasonable working scenario is when the
first ifunc call *would* result in the same relocation that the early
binding already did (so no modifications needed).

So detect, verify, abort if different? Or just shortcut the ifunc relocation
completely?

Martin