tst-prelink fails when built with -fPIC

tst-prelink fails when built with -fPIC

[Aurelien Jarno]

Hi all,

Debian builds gcc 6.2 with --enable-default-pie, which correspond to
always passing -pie -fPIE -fPIC. This causes the tst-prelink to fail on
many architectures (at least arm, aarch64, i386, mips*) due to the use
of -fPIC.

Would it make sense to build this test with -fno-PIC? In that case I'll
send a patch.

Thanks,
Aurelien

-- 
Aurelien Jarno                          GPG: 4096R/1DDD8C9B
aurelien@aurel32.net                 http://www.aurel32.net

Re: tst-prelink fails when built with -fPIC

[Florian Weimer]

On 11/25/2016 11:28 AM, Aurelien Jarno wrote:
> Hi all,
>
> Debian builds gcc 6.2 with --enable-default-pie, which correspond to
> always passing -pie -fPIE -fPIC. This causes the tst-prelink to fail on
> many architectures (at least arm, aarch64, i386, mips*) due to the use
> of -fPIC.

What does the failure look like?

Thanks,
Florian

Re: tst-prelink fails when built with -fPIC

[Aurelien Jarno]

On 2016-11-25 12:49, Florian Weimer wrote:
> On 11/25/2016 11:28 AM, Aurelien Jarno wrote:
> >Hi all,
> >
> >Debian builds gcc 6.2 with --enable-default-pie, which correspond to
> >always passing -pie -fPIE -fPIC. This causes the tst-prelink to fail on
> >many architectures (at least arm, aarch64, i386, mips*) due to the use
> >of -fPIC.
> 
> What does the failure look like?

Running tst-prelink with LD_TRACE_PRELINKING=1 doesn't output the
following line for the stdout symbol:

  conflict 0xf7532000 0x00008598 -> 0x56559000 0x00003124 x 0xf7532000 0x001b3dfc /0 stdout

Aurelien

-- 
Aurelien Jarno                          GPG: 4096R/1DDD8C9B
aurelien@aurel32.net                 http://www.aurel32.net

Re: tst-prelink fails when built with -fPIC

[Joseph Myers]

On Fri, 25 Nov 2016, Aurelien Jarno wrote:

> Hi all,
> 
> Debian builds gcc 6.2 with --enable-default-pie, which correspond to
> always passing -pie -fPIE -fPIC. This causes the tst-prelink to fail on

Why -fPIC?  It's documented as -fPIE -pie, *not* -fPIC.

Now, -fPIE ends up with glibc building with -DPIC, and in some cases that 
may break things if code isn't ready for PIC building outside shared 
libraries.  Those cases can be fixed individually.

-- 
Joseph S. Myers
joseph@codesourcery.com

Re: tst-prelink fails when built with -fPIC

[Aurelien Jarno]

On 2016-11-25 17:20, Joseph Myers wrote:
> On Fri, 25 Nov 2016, Aurelien Jarno wrote:
> 
> > Hi all,
> > 
> > Debian builds gcc 6.2 with --enable-default-pie, which correspond to
> > always passing -pie -fPIE -fPIC. This causes the tst-prelink to fail on
> 
> Why -fPIC?  It's documented as -fPIE -pie, *not* -fPIC.

You are correct it's only -fPIE. Now using either -fno-PIE or -fno-PIC
makes the tst-prelink to pass. But -fno-PIE is the correct option here.

> Now, -fPIE ends up with glibc building with -DPIC, and in some cases that 
> may break things if code isn't ready for PIC building outside shared 
> libraries.  Those cases can be fixed individually.

Indeed. I have already send a patch for x86-64, I am currently testing a
patch for s390.

Aurelien

-- 
Aurelien Jarno                          GPG: 4096R/1DDD8C9B
aurelien@aurel32.net                 http://www.aurel32.net

Re: tst-prelink fails when built with -fPIC

[Stefan Liebler]

On 11/25/2016 07:15 PM, Aurelien Jarno wrote:
> On 2016-11-25 17:20, Joseph Myers wrote:
>> On Fri, 25 Nov 2016, Aurelien Jarno wrote:
>>
>>> Hi all,
>>>
>>> Debian builds gcc 6.2 with --enable-default-pie, which correspond to
>>> always passing -pie -fPIE -fPIC. This causes the tst-prelink to fail on
>>
>> Why -fPIC?  It's documented as -fPIE -pie, *not* -fPIC.
> 
> You are correct it's only -fPIE. Now using either -fno-PIE or -fno-PIC
> makes the tst-prelink to pass. But -fno-PIE is the correct option here.
> 
>> Now, -fPIE ends up with glibc building with -DPIC, and in some cases that
>> may break things if code isn't ready for PIC building outside shared
>> libraries.  Those cases can be fixed individually.
> 
> Indeed. I have already send a patch for x86-64, I am currently testing a
> patch for s390.
> 
> Aurelien
> 

Are there any news for this?

Bye.
Stefan