Fix compiling C demo apps

[macpacheco]

I was getting errors like:
crt0.c:(.text+0x27): relocation truncated to fit: R_X86_64_32S against `__bss_start'
When the C demos were being linked.
This goes away by changing CFLAGS in src/BareMetal-Demo/build.sh to include -fPIC
However gavare.app freezes with this change. It never returns to the Monitor prompt (waited 5 minutes). And no graphics are drawn.
I can't build gavare.app without -fPIC so I don't know if this is the culprit or not.
I'm running openSUSE Leap 15.3.
GCC 7.5.0
ld 2.39.0.20220810-150100.7.40
QEMU 5.2.0
On an AMD Ryzen 5 3500U with Radeon Vega Mobile Gfx (AuthenticAMD family 23 model 24)

[IanSeyler]

I cannot replicate the linking issue on my test linux system. I have updated the build script to not link gavare with crt0.o as it wasn't needed for that program and it was causing the image to be garbled.

xubuntu 24.04
GCC 13
LD 2.42
QEMU 8.2.2

[macpacheco]

Installed gcc-12 from my distro and built binutils-2.42 from source.
Now gavare returns to the Monitor prompt in a flash without any visible output.

crt0.c:(.text+0x27): relocation truncated to fit: R_X86_64_32S against `__bss_start'
Hmm, this suggests something is doing X32 instead of pure 32 bit or pure 64 bit.
Upon closer inspection I find that .o files created with gcc -m64 and regular linux 64 bit elf executable also reference to .data, .text, .bss using R_X86_64_32S.

If I change . = 0xFFFF800000000000;
To an address that fits in 32 bits, build.sh runs without errors or warnings, even without -fPIC.
I tried: . = 0x80000;
But gavare still doesn't produce any visible output.
Can you dump crt0.o with objdump -Srt ?

objdump -Srt src/BareMetal-Demo/src/crt0.o

src/BareMetal-Demo/src/crt0.o:     file format elf64-x86-64

SYMBOL TABLE:
0000000000000000 l    df *ABS*  0000000000000000 crt0.c
0000000000000000 l    d  .text  0000000000000000 .text
0000000000000000 g     F .text  000000000000002e _start
0000000000000000         *UND*  0000000000000000 __bss_start
0000000000000000         *UND*  0000000000000000 __bss_stop
0000000000000000         *UND*  0000000000000000 main



Disassembly of section .text:

0000000000000000 <_start>:
   0:   48 83 ec 08             sub    $0x8,%rsp
   4:   b8 00 00 00 00          mov    $0x0,%eax
                        5: R_X86_64_32  __bss_start
   9:   48 3d 00 00 00 00       cmp    $0x0,%rax
                        b: R_X86_64_32S __bss_stop
   f:   73 0e                   jae    1f <_start+0x1f>
  11:   c6 00 00                movb   $0x0,(%rax)
  14:   48 ff c0                inc    %rax
  17:   48 3d 00 00 00 00       cmp    $0x0,%rax
                        19: R_X86_64_32S        __bss_stop
  1d:   75 f2                   jne    11 <_start+0x11>
  1f:   b8 00 00 00 00          mov    $0x0,%eax
  24:   e8 00 00 00 00          call   29 <_start+0x29>
                        25: R_X86_64_PC32       main-0x4
  29:   48 83 c4 08             add    $0x8,%rsp
  2d:   c3                      ret

[macpacheco]

Added -mcmodel=large to build.sh for Demo. No I can build without -fPIC and the original c.ld (.=0xFFFF800000000000) and zero errors.
Perhaps your gcc assumes -mcmodel=large, while mine -mcmodel=small (man pages say -mcmodel=small is default).
Now objdump -Srt on crt0.o shows all references to __bss_start / __bss_stop show R_X86_64_64.
Still gavare does nothing and returns instantly.

[IanSeyler]

Here is my crt0.o:

src/BareMetal-Demo/src/crt0.o:     file format elf64-x86-64

SYMBOL TABLE:
0000000000000000 l    df *ABS*	0000000000000000 crt0.c
0000000000000000 l    d  .text	0000000000000000 .text
0000000000000030 l     F .text	0000000000000032 zero_bss
0000000000000000 g     F .text	0000000000000023 _start
0000000000000000         *UND*	0000000000000000 main
0000000000000000         *UND*	0000000000000000 __bss_start
0000000000000000         *UND*	0000000000000000 __bss_stop



Disassembly of section .text:

0000000000000000 <_start>:
   0:	f3 0f 1e fa          	endbr64
   4:	55                   	push   %rbp
   5:	48 89 e5             	mov    %rsp,%rbp
   8:	48 83 ec 10          	sub    $0x10,%rsp
   c:	e8 1f 00 00 00       	call   30 <zero_bss>
  11:	b8 00 00 00 00       	mov    $0x0,%eax
  16:	e8 00 00 00 00       	call   1b <_start+0x1b>
			17: R_X86_64_PLT32	main-0x4
  1b:	89 45 fc             	mov    %eax,-0x4(%rbp)
  1e:	8b 45 fc             	mov    -0x4(%rbp),%eax
  21:	c9                   	leave
  22:	c3                   	ret
  23:	66 66 2e 0f 1f 84 00 	data16 cs nopw 0x0(%rax,%rax,1)
  2a:	00 00 00 00 
  2e:	66 90                	xchg   %ax,%ax

0000000000000030 <zero_bss>:
  30:	f3 0f 1e fa          	endbr64
  34:	55                   	push   %rbp
  35:	48 89 e5             	mov    %rsp,%rbp
  38:	48 8d 05 00 00 00 00 	lea    0x0(%rip),%rax        # 3f <zero_bss+0xf>
			3b: R_X86_64_PC32	__bss_start-0x4
  3f:	48 89 45 f8          	mov    %rax,-0x8(%rbp)
  43:	eb 0c                	jmp    51 <zero_bss+0x21>
  45:	48 8b 45 f8          	mov    -0x8(%rbp),%rax
  49:	c6 00 00             	movb   $0x0,(%rax)
  4c:	48 83 45 f8 01       	addq   $0x1,-0x8(%rbp)
  51:	48 8d 05 00 00 00 00 	lea    0x0(%rip),%rax        # 58 <zero_bss+0x28>
			54: R_X86_64_PC32	__bss_stop-0x4
  58:	48 39 45 f8          	cmp    %rax,-0x8(%rbp)
  5c:	72 e7                	jb     45 <zero_bss+0x15>
  5e:	90                   	nop
  5f:	90                   	nop
  60:	5d                   	pop    %rbp
  61:	c3                   	ret

Perhaps I should add '-mcmodel=large' to the build script then.

[macpacheco]

Nope, your setup work, mine doesn't. Just trying to figure it out.