[fedora-arm] F13 glibc patches for ARM?

[Gordan Bobic]

 On Mon, 15 Aug 2011 11:30:36 +0100, Andrew Haley <aph at redhat.com> 
 wrote:
> On 08/15/2011 11:11 AM, Gordan Bobic wrote:
>>  On Mon, 15 Aug 2011 09:57:44 +0100, Andrew Haley <aph at redhat.com>
>>  wrote:
>>> On 08/10/2011 02:43 PM, Gordan Bobic wrote:
>>>>>>  On Sat, 06 Aug 2011 10:24:03 +0100, Andrew Haley 
>>>>>> <aph at redhat.com>
>>>>>>  wrote:
>>>>>>> On 08/05/2011 09:07 PM, Gordan Bobic wrote:
>>>>>>>> Thanks for pointing out, it does look like the same bug. So
>>>>>>>> what's
>>>>>>>> the
>>>>>>>> fix/workaround?
>>>>>>>>
>>>>>>>> On 08/05/2011 08:59 PM, Niels de Vos wrote:
>>>>>>>>> On 5 Aug 2011 19:15, "Gordan Bobic" <gordan at bobich.net
>>>>>>>>> <mailto:gordan at bobich.net>> wrote:
>>>>>>>>>  >
>>>>>>>>>  > How does one pick the correct "ports" part of glibc for 
>>>>>>>>> the
>>>>>>>>> correct core?
>>>>>>>>>  >
>>>>>>>>>  > I tried to build the RHEL6 glibc with the F13 ports tar
>>>>>>>>> ball,
>>>>>>>>> but the
>>>>>>>>>  > build eventually fails:
>>>>>>>>>  >
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> 
>>>>>>>>> /usr/lib/gcc/armv5tel-redhat-linux-gnueabi/4.4.5/libgcc_eh.a(unwind-arm.o):
>>>>>>>>>  > In function `__gnu_Unwind_Backtrace':
>>>>>>>>>  > (.text+0x8b8): undefined reference to `__stack_chk_guard'
>>>>>>>>>  > collect2: ld returned 1 exit status
>>>>>>>>>
>>>>>>>>> Looks very much like the error in
>>>>>>>>> https://bugzilla.redhat.com/show_bug.cgi?id=726495
>>>>>>>
>>>>>>> What is it linking against?  __stack_chk_guard should be 
>>>>>>> defined
>>>>>>> in
>>>>>>> libc.a.
>>>>
>>>>  What I would really like to know is what fix the bugzilla ticket
>>>> above
>>>>  refers to. The last response from Andrew reads:
>>>>  "I suspect this is just another manifestation of the bug, now
>>>> fixed,
>>>>  that causes
>>>>  gcj programs not to link,  I certainly had no such problem when I
>>>> built
>>>>  libc
>>>>  yesterday."
>>>>
>>>>  There's no reference to another bug. Can anyone point me in the
>>>>  direction of the relevant bugzilla ticket that fixes the said gcj
>>>>  linking issue? I just searched on RH bugzilla and couldn't find
>>>> anything
>>>>  of relevance.
>>>
>>> Sorry, I gave up on F13 when it became EOL and moved to F15.  No 
>>> more
>>> bugs were being accepted.
>>>
>>> Is /usr/lib/gcc/armv5tel-redhat-linux-gnueabi/4.4.4/libgcc_s.so (or
>>> whatever)
>>> a symlink or a text file on your system?
>>>
>>> It should look like this:
>>>
>>> /* GNU ld script
>>>    Use the shared library, but some functions are only in
>>>    the static library.  */
>>> GROUP ( /lib/libgcc_s.so.1 libgcc.a )
>>
>>  It's a symlink to libgcc_s.so.1. Should it be a text file??
>
> Yes.  This is a bug in the Fedora gcc specfile.  Upstream gcc is
> correct.

 Got a link to the relevant spec file patch handy?

>>  In the meantime I was able to get glibc to build using Niels' hack
>>  here:
>>  https://bugzilla.redhat.com/show_bug.cgi?id=726495
>>
>>  This seems to work, but can you think of a better way to do this?
>
> My builds of unwind-arm.o do not call `__stack_chk_guard'.
>
> I think we need to look at why there is a call to `__stack_chk_guard'
> in your unwind-arm.o.  I suspect that it may be compiled with
> -fstack-protector, and it shouldn't be.
>
> Can you have a look and see where that call is coming from?

 Well, glibc's rebuild doesn't seem output the things it runs as you 
 would expect "make" to, so it's a bit hard to tell. I'll un-apply Niels' 
 patch and see if I can track it down when it occurs. It may be a while 
 because the build time is non-trivial on a SheevaPlug.

 Gordan