Re: [RFC] Fix float128 IFUNC relocations on ppc64le [BZ #21707]

[Florian Weimer <fweimer at redhat dot com>]

On 07/06/2017 04:04 PM, Carlos O'Donell wrote:
> On 07/06/2017 02:12 AM, Florian Weimer wrote:
>> On 07/05/2017 10:11 PM, Joseph Myers wrote:
>>> The installed compiler does -lgcc -lc -lgcc (roughly; -lgcc may actually 
>>> use -lgcc_s or -lgcc_eh, and -lc may vary for e.g. profiling).  -lc is a 
>>> linker script doing e.g. GROUP ( /lib64/libc.so.6 
>>> /usr/lib64/libc_nonshared.a  AS_NEEDED ( /lib64/ld-linux-x86-64.so.2 ) ).  
>>> The shared libgcc is also a linker script on some platforms.
>>>
>>> The glibc build uses libc.so --as-needed ld.so --no-as-needed -lgcc 
>>> (again, roughly) (but libc.a -lgcc libc.a -lgcc in the static build case).
>>>
>>> I suppose the suggestion is that we should both use -lgcc before linking 
>>> with libc, not just after (this being the issue involved in the present 
>>> case), and use -Wl,--start-group -Wl,--end-group around libc.so and ld.so, 
>>> both to be more similar to how things work with the installed compiler?  
>>> Which makes sense, but is also rather risky for this development stage.
>>
>> Okay, then let's postpone the more general approach and use whatever
>> works on POWER for the release.
> 
> Right, for now only POWER needs the link command fixed.
> 
>> I assumed that the general fix would still be POWER-specific, but that
>> does not seem to be the case.
> 
> Right, the fix is in general code, but it fixes a general problem for
> all machines. Every machine, including x86_64, used to segfault upon
> first use of __thread in an IFUNC resolver function, and the change
> fixes that.

Note that I was discussing the Makefile changes, not the apply_urel
reordering (which affects only static binaries anyway).

Florian