This is the mail archive of the
mailing list for the binutils project.
Re: Allow copy relocations with pie links
- From: Cary Coutant <ccoutant at google dot com>
- To: Sriraman Tallam <tmsriram at google dot com>
- Cc: "H.J. Lu" <hjl dot tools at gmail dot com>, binutils <binutils at sourceware dot org>, Ian Lance Taylor <iant at google dot com>, David Li <davidxl at google dot com>
- Date: Thu, 8 May 2014 14:30:16 -0700
- Subject: Re: Allow copy relocations with pie links
- Authentication-results: sourceware.org; auth=none
- References: <CAAs8HmxwPuH7_s1yLxeW_6qkMCB+_kYSzB6yYG85QZS6yVHpZg at mail dot gmail dot com> <CAHACq4q9qP_h=PuwhJkE6KMt_xf6_Ne3unGRdQpvFqaXL=OxPw at mail dot gmail dot com> <CAAs8HmxkGVYAyq4wf1FChqowmQRTO9POve8Ve5f4kb3DXPfywQ at mail dot gmail dot com> <CAMe9rOqFiwXw3G=zFvWq4o27OGXss2RiyZMUKOqzxE8_eFxnaw at mail dot gmail dot com> <CAHACq4rdUe4_Rhj8vwA0=NU6O3WTOR84HQJKqB8AiSi25N4pgA at mail dot gmail dot com> <CAAs8HmwtGtnGaK2oyoUoDAZ=R1FnGkNFte-9A5sMD5qHS0R3tw at mail dot gmail dot com>
>> Right. With Sri's test case, we get an absolute R_386_32 relocation
>> for the access, and the DT_TEXTREL flag is set in the PIE binary
>> because of the text relocation. There is no COPY relocation. Since we
>> had a path in the code to check for copy relocs for the PC-relative
>> relocations, I think it makes sense to adjust the test there, though.
> It took me a while to understand this discussion. I think I get it
> now, it looks like there is no need for me to change i386.cc as no
> data access can produce a pc relative relocation. Cary, how do I
> think I should adjust the test? The test itself passes on i386 even
> without the change.
I think your patch to i386.cc is fine. If it's possible to reach the
call to may_need_copy_reloc in the case that handles R_386_PCxx
relocations, it's correct to check for output_is_executable() instead
of !output_is_position_independent(). If the ABI forbids a compiler
from ever using one of those relocations to access global data, then
we won't ever reach the code anyway. To me, the choice is either to
patch it as you've done, or remove that whole if statement. Given that
the code is there and always has been, let's leave the patch as is.
There's always the chance that someone might write a bit of assembly
code that uses that relocation.