V1 trouble with gcc 3.2.3

[Dave Ashley]

I'm using a fairly old version of V1 (July or so) and when I have an
etherboot ebi payload and build linuxbios with gcc 3.2.3 linuxbios fails,
whereas an older gcc 2.96 works.

It gets to the part where it is going to load/launch the payload. Here
is what it does when it works, linuxbios built with gcc 2.96:
Wrote linuxbios table at: 00000500 - 00000674  checksum ea53

Welcome to elfboot, the open sourced starter.
January 2002, Eric Biederman.
Version 1.2

    34:init_bytes() - zkernel_start:0xfff00000  zkernel_mask:0x0000ffff
Found ELF candiate at offset 0
New segment addr 0x94000 size 0x7d48 offset 0x60 filesize 0x3538
(cleaned up) New segment addr 0x94000 size 0x7d48 offset 0x60 filesize 0x3538
Loading Segment: addr: 0x0000000000094000 memsz: 0x0000000000007d48 filesz: 0x0000000000003538
Clearing Segment: addr: 0x0000000000097538 memsz: 0x0000000000004810
Jumping to boot code at 0x94000
ROM segment 0x5f5e length 0x86ba reloc 0x9400
etc.

--------------- Here is what it does when linuxbios is built with gcc 3.2.3:
Wrote linuxbios table at: 00000500 - 00000634  checksum eeae

Welcome to elfboot, the open sourced starter.
January 2002, Eric Biederman.
Version 1.2

    37:init_bytes() - zkernel_start:0xfff00000  zkernel_mask:0x0000ffff
Found ELF candiate at offset 0
New segment addr 0x94000 size 0x7b28 offset 0x60 filesize 0x3338
(cleaned up) New segment addr 0x94000 size 0x7b28 offset 0x60 filesize 0x3338
Loading Segment: addr: 0x0000000000094000 memsz: 0x0000000000007b28 filesz: 0x0000000000003338
Clearing Segment: addr: 0x0000000000097338 memsz: 0x00000000000047f0
Jumping to boot code at 0x94000
0


---------------
The payload is the same in both cases. The filesizes are different by
0x200 bytes. My guess is it is something in elfboot. I remember
reading something about the -fno-merge-constants option which might
apply--older XFree86 servers have an ELF loader that doesn't like the
new default behaviour of gcc and you have to turn this option to get
their elf loading to work again.

This is probably something that has been hashed out already, sorry
about that.

Any advice would be welcome.

-Dave