Re: linux-next: sparc32/gcc-3.4.5/ld-2.15 build failure

[Steven Whitehouse <swhiteho@redhat.com>]

Hi,

On Thu, 2008-07-31 at 00:35 -0700, Andrew Morton wrote:
> On Thu, 31 Jul 2008 17:16:03 +1000 Stephen Rothwell <sfr@canb.auug.org.au> wrote:
> 
> > Hi Dave,
> > 
> > linux-next builds (sparc32 defconfig) have been failing for some time
> > like this:
> > 
> > .tmp_kallsyms2.o(.rodata+0x0): In function `kallsyms_addresses':
> > : relocation truncated to fit: R_SPARC_32 _text
> > .tmp_kallsyms2.o(.rodata+0x4): In function `kallsyms_addresses':
> > : relocation truncated to fit: R_SPARC_32 _text
> > .tmp_kallsyms2.o(.rodata+0x8): In function `kallsyms_addresses':
> > : relocation truncated to fit: R_SPARC_32 _text
> > 
> > (several more)
> > 
> > I finally bisected it down to this commit:
> > 
> > commit f9247273cb69ba101877e946d2d83044409cc8c5
> > Author: Steven Whitehouse <swhiteho@redhat.com>
> > Date:   Thu Jul 24 17:22:13 2008 +0100
> > 
> >     UFS: add const to parser token table
> >     
> > Reverting this commit and the followup commit:
> > 
> > commit fb2e405fc1fc8b20d9c78eaa1c7fd5a297efde43
> > Author: Adrian Bunk <bunk@kernel.org>
> > Date:   Fri Jul 25 02:55:49 2008 +0300
> > 
> >     fix fs/nfs/nfsroot.c compilation
> > 
> > allows the sparc32 build to succeed.  This is toolchain specific as a
> > different cross toolchain I have does not get this error.
> > 
> > Failing toolchain:
> > 
> > $ gcc-3.4.5-glibc-2.3.6/sparc64-unknown-linux-gnu/bin/sparc64-unknown-linux-gnu-gcc --version
> > sparc64-unknown-linux-gnu-gcc (GCC) 3.4.5
> > $ gcc-3.4.5-glibc-2.3.6/sparc64-unknown-linux-gnu/bin/sparc64-unknown-linux-gnu-ld --version
> > GNU ld version 2.15
> > 
> > An OK toolchain:
> > $ cross/bin/sparc64-linux-gcc --version
> > sparc64-linux-gcc (GCC) 4.2.4
> > $ cross/bin/sparc64-linux-ld --version
> > GNU ld (GNU Binutils) 2.18
> > 
> > Both these commits are actually in Linus' tree, now.
> 
> gee.  nfsroot.c is now effectively doing `const __initconst' which
> might be upsetting the compiler.  And perhaps one of the forty-odd
> other users of match_table_t needs __initconst or somesuch.
> 
According to my find & grep there is only the one place which has marked
out the match_table_t for a different section.

> Yes, I'd second a revert-and-try-again-later on that one.
It looks like there is no simple solution, so I guess we'll have to do
that :( The only other alternative would be to remove the __initconst
which isn't ideal either,

Steve.