Re: [PATCH 2/2] Add a testcase for copy reloc against protected data

["H.J. Lu" <hjl dot tools at gmail dot com>]

On Wed, Apr 1, 2015 at 12:51 PM, Carlos O'Donell <carlos@redhat.com> wrote:
> On 03/31/2015 08:26 AM, H.J. Lu wrote:
>> On Fri, Mar 6, 2015 at 3:24 PM, H.J. Lu <hjl.tools@gmail.com> wrote:
>>> On Fri, Mar 6, 2015 at 3:07 PM, Joseph Myers <joseph@codesourcery.com> wrote:
>>>> On Fri, 6 Mar 2015, H.J. Lu wrote:
>>>>
>>>>> On Fri, Mar 6, 2015 at 2:23 PM, Joseph Myers <joseph@codesourcery.com> wrote:
>>>>>> On Fri, 6 Mar 2015, H.J. Lu wrote:
>>>>>>
>>>>>>>> Suppose I configure / build with new binutils but a GCC version without
>>>>>>>> your patches.  What will the results be?  That the tests don't build /
>>>>>>>> run?  That they build / run but fail (best avoided if possible)?  In
>>>>>>>> either case, you need clear documentation for architecture maintainers on
>>>>>>>> what GCC versions (*not* requiring any uncommitted GCC patches) must be
>>>>>>>> used to identify whether architecture-specific changes are needed and to
>>>>>>>> test such changes.
>>>>>>>>
>>>>>>>
>>>>>>> I will submit a separate patch to address GCC issue.
>>>>>>
>>>>>> I don't think these tests should go in glibc until:
>>>>>
>>>>> Which tests were you talking about?  My copy relocation
>>>>> tests work with any GCC versions, with and without the fix
>>>>> for
>>>>>
>>>>> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65248
>>>>
>>>> Thanks for the explanation.  In
>>>> <https://sourceware.org/ml/libc-alpha/2015-03/msg00170.html> you
>>>> originally sent patches for all three of GCC, binutils, glibc, so giving
>>>> the impression there might be dependencies between these patches.  If the
>>>> tests work with any GCC version (i.e. fail if the architecture hasn't been
>>>> fixed, pass if it has been fixed) then GCC changes shouldn't need to block
>>>> them.
>>>>
>>>
>>> To get copy reloc against protected data symbol working, you
>>> need to fix glibc, binutils and GCC.  My glibc tests work around
>>> the GCC bug so that you don't need the fixed GCC to build/test
>>> glibc.
>>
>> I checked in my testcase patch.  I didn't close
>>
>> https://sourceware.org/bugzilla/show_bug.cgi?id=17711
>>
>> since only x86 is fixed.  To fix it on a target with copy reloc, you need to
>> properly handle ELF_RTYPE_CLASS_EXTERN_PROTECTED_DATA like
>> x86.  You also need to fix linker if tests fail.
>>
>
> I assume when you say x86 you mean all of the variants of that machine.
>
> Your test fail for me on F21 x86_64 with gcc 4.9.2 and ld 2.24 (Fedora
> versions). I haven't looked into this, but hopefully we'll get Fedora--
> specific build bots up to help with this kind of testing.
>
> XPASS: conform/ISO11/complex.h/conform
> XPASS: conform/ISO11/stdalign.h/conform
> XPASS: conform/ISO11/stdnoreturn.h/conform
> FAIL: elf/tst-protected1a
> FAIL: elf/tst-protected1b
> UNRESOLVED: timezone/testdata/UTC
> Summary of test results:
>       2 FAIL
>    2199 PASS
>       1 UNRESOLVED
>     189 XFAIL
>       3 XPASS
>
> [carlos@koi build]$ cat glibc/elf/tst-protected1a.out
> `protected1' in main and moda doesn't have same address
> `protected3' in main and moda doesn't have same address
> `protected1' in main doesn't have the updated value
> `protected1' in moda has the wrong value
> `protected3' in main doesn't have the updated value
> `protected3' in main doesn't have the updated value
> [carlos@koi build]$ cat glibc/elf/tst-protected1b.out
> `protected3' in main and modb doesn't have same address
> `protected3' in main doesn't have the updated value
>
> What's wrong?
>

Linker bug.  You need linker from master or 2.25 branch.

-- 
H.J.