LinuxBIOS & binutils evilness

[Jeff Noxon]

2.14.90.0.5 (Debian unstable) works for mkelfImage 2.5 but won't link
LinuxBIOS.

2.12.90.0.1 (Debian stable) works for LinuxBIOS but mkelfimage 2.5
complains about no a5 magic.

Am I stuck with two versions of binutils?  Or maybe you know how to make
LB link with the latest binutils?  It looks like a binutils bug.

Here is my link error:

[snip]
cp linuxbios_payload.nrv2b linuxbios_payload
echo "INCLUDE ldoptions" > ldscript.ld ; for file in  /usr/src/linuxbios/linuxbios/src/arch/i386/config/ldscript.base /usr/src/linuxbios/linuxbios/src/cpu/i386/entry16.lds /usr/src/linuxbios/linuxbios/src/cpu/i386/entry32.lds /usr/src/linuxbios/linuxbios/src/cpu/i386/reset16.lds /usr/src/linuxbios/linuxbios/src/arch/i386/lib/id.lds /usr/src/linuxbios/linuxbios/src/arch/i386/lib/failover-noz.lds ; do echo "INCLUDE $file" >> ldscript.ld ; done
gcc -nostdlib -nostartfiles -static -o linuxbios -T ldscript.ld crt0.o
crt0.o(.reset+0x1): In function `reset_vector':
: relocation truncated to fit: R_386_PC16 _start_offset
collect2: ld returned 1 exit status
make[1]: *** [linuxbios] Error 1
make[1]: Leaving directory `/usr/src/linuxbios/tyan'
make: *** [xtyan] Error 2

Jeff Noxon  writes:
> 2.14.90.0.5 (Debian unstable) works for mkelfImage 2.5 but won't link
> LinuxBIOS.

This one is change in behavior of binutils.
> 2.12.90.0.1 (Debian stable) works for LinuxBIOS but mkelfimage 2.5
> complains about no a5 magic.

That is a binutils bug.
> Am I stuck with two versions of binutils?  Or maybe you know how to make
> LB link with the latest binutils?  It looks like a binutils bug.

The code in the freebios2:src/cpu/i386/reset16.inc should avoid this
problem, but oit does not generate a far jump only a near jump.

The problem binutils does not do a good job of supporting 16bit
offsets.  If you would like to pull the code back and generate a patch
for the freebios1 tree it would be appreciated.

Eric

[Jeff Noxon]

Is there something wrong with the near jump approach?

I will attempt to pull freebios2 from cvs.  Sourceforge CVS seems not
to work most of the time I try to use it.  It would be nice to have CVS
hosted somewhere reliable, maybe BerliOS.

Is freebios2 booting yet on any platforms?

On Thu, Aug 14, 2003 at 12:11:37PM -0600, Eric W. Biederman wrote:
> The code in the freebios2:src/cpu/i386/reset16.inc should avoid this
> problem, but oit does not generate a far jump only a near jump.
>
> The problem binutils does not do a good job of supporting 16bit
> offsets.  If you would like to pull the code back and generate a patch
> for the freebios1 tree it would be appreciated.

[ron minnich]

On Thu, 14 Aug 2003, Jeff Noxon wrote:
> Is freebios2 booting yet on any platforms?

Both PPC and K8 platforms. We need to try to cut a 386 machine over, I am 
going to try with the AMD Elan boards.

ron

[Stefan Reinauer]

* Jeff Noxon  [030814 19:55]:
> 2.14.90.0.5 (Debian unstable) works for mkelfImage 2.5 but won't link
> LinuxBIOS.

do you have x86-64 support enabled in these binutils? This breaks
relocations 32->16bit.
> Here is my link error:
>
> crt0.o(.reset+0x1): In function `reset_vector':
> : relocation truncated to fit: R_386_PC16 _start_offset
> collect2: ld returned 1 exit status

src/cpu/i386/reset16.inc is the culprit. See the freebios2 tree
for a wild but working implementation of reset16.inc. The 
"jmp _start" is there implemented as:
reset_vector:
        /* jmp _start */
        .byte  0xe9
        .int   _start - ( . + 2 )

This should also be changed in the old LinuxBIOS tree. It works fine.

Stefan