Index
> Linux > simple sdl game segfaults on 64bit linux, why? |
| Author |
|
|
Furs 23 Oct 2017, 15:08
I don't think you can use 'ret' on Linux to exit from the entry point main, never seen it that way. Call exit instead:
Code: mov eax, 60 syscall |
|||
23 Oct 2017, 15:08 |
|
|
ginere 23 Oct 2017, 16:39
Furs wrote: Also you might have to align the stack to 16 bytes before calling SDL_Init. Thank you. That did the trick. Stack was aligned to $08 so I did sub rsp, $08 before SDL_Init, and after SDL_Init I put add rsp, $08 to get old rsp back. |
|||
|
23 Oct 2017, 16:39 |
|
|
Furs 23 Oct 2017, 20:17
You should use a dummy push instead (e.g. push rax), it's much smaller (1 byte vs 4) and it's entry point anyway.
I know that Linux is probably going to always give this (mis)alignment on entry point just to be compatible, but I'm just saying that libc (the library that hijacks the entry point of all C apps that use it on Linux) realigns the stack on entry point (as if it's unreliable). Paranoia?  But I'd err on the side of caution and just use a "and rsp, -16". |
|||
|
23 Oct 2017, 20:17 |
|
|
system error 28 Nov 2017, 15:13
Furs wrote: I don't think you can use 'ret' on Linux to exit from the entry point main, never seen it that way. Call exit instead: "ret" IS the standard exit point for every main function of gcc binaries both on Windows and Linux. You been sending "patches" to GCC people and you didn't know this?? :O |
|||
|
28 Nov 2017, 15:13 |
|
|
Furs 28 Nov 2017, 16:30
main function is not the entry point in C programs, ret works because the C runtime (which hijacks the entry point) calls it, so it then calls exit itself.
Of course I did say "entry point main" even in your quote but it's not like you can understand English. |
|||
|
28 Nov 2017, 16:30 |
|
|
system error 28 Nov 2017, 17:58
Quote: main function is not the entry point in C programs Quote: ret works because the C runtime (which hijacks the entry point) calls it, so it then calls exit itself. "ret" will always work because the return address is in RSP. Doesn't need extra "hijacking" and all those "expert" bullshits xD Simple hello world's binaries; Code: #include <stdio.h> int main() //Try "int Expert_Begin()" here xD { printf("Furs keeps sending patches to GCC..."); return 0; } output from GDB Code: 0x00000000004015b0 <+0>: push rbp 0x00000000004015b1 <+1>: mov rbp,rsp 0x00000000004015b4 <+4>: sub rsp,0x20 0x00000000004015b8 <+8>: call 0x401750 <__main> 0x00000000004015bd <+13>: lea rcx,[rip+0x2a6c] # 0x404030 0x00000000004015c4 <+20>: call 0x402b60 <printf> 0x00000000004015c9 <+25>: mov eax,0x0 0x00000000004015ce <+30>: add rsp,0x20 0x00000000004015d2 <+34>: pop rbp 0x00000000004015d3 <+35>: ret |
|||
|
28 Nov 2017, 17:58 |
|
|
system error 28 Nov 2017, 18:20
Expert advise from Furs
Quote: I don't think you can use 'ret' on Linux to exit from the entry point main, never seen it that way Oopppsss (From C's main()) Code: 0x00000000004015b0 <+0>: push rbp 0x00000000004015b1 <+1>: mov rbp,rsp 0x00000000004015b4 <+4>: sub rsp,0x20 0x00000000004015b8 <+8>: call 0x401750 <__main> 0x00000000004015bd <+13>: lea rcx,[rip+0x2a6c] # 0x404030 0x00000000004015c4 <+20>: call 0x402b60 <printf> 0x00000000004015c9 <+25>: mov eax,0x0 0x00000000004015ce <+30>: add rsp,0x20 0x00000000004015d2 <+34>: pop rbp 0x00000000004015d3 <+35>: ret ;==> gcc does it ALL THE TIME but Furs never "seen" it that way?? Baloney expert :O |
|||
|
28 Nov 2017, 18:20 |
|
|
revolution 28 Nov 2017, 22:52
So what is the answer? Can we, or can we not, simply use ret to return to the loader in Linux? Is it "official" or just a coincidence? Do the docs say anything on the matter? Do the docs even exist?
|
|||
|
28 Nov 2017, 22:52 |
|
|
evk1 29 Nov 2017, 00:33
revolution wrote: So what is the answer? Can we, or can we not, simply use ret to return to the loader in Linux? Is it "official" or just a coincidence? Do the docs say anything on the matter? Do the docs even exist? You can use ret, if you are going to link your object file with gcc (or another compiler). In this case gcc automatically includes precompiled startup code which calls the main function to you program. However if you specify entry point manually (--entry=main with gcc, or entry main with fasm) you should not use ret, since at the time when program starts [rsp] contains not a return address, but argc. The program startup process is documented in the ELF specification for x86_64. _________________ Sorry for my English |
|||
|
29 Nov 2017, 00:33 |
|
|
Melissa 29 Nov 2017, 06:04
revolution wrote: So what is the answer? Can we, or can we not, simply use ret to return to the loader in Linux? Is it "official" or just a coincidence? Do the docs say anything on the matter? Do the docs even exist? You use ret if your code is called from C runtime. If it is not you specify entry point, usually _start and call sysexit. If making standalone fasm executable sysexit is arbitrary and entry point wahtever. You should be aware that in case of main called from C runtime rdi contains argc,rsi points to argv array. In case not, those values are on stack. |
|||
|
29 Nov 2017, 06:04 |
|
|
revolution 29 Nov 2017, 10:54
Thanks for the answers everyone. I think system error owes Furs an apology.
|
|||
|
29 Nov 2017, 10:54 |
|
|
system error 29 Nov 2017, 11:30
Edit by revolution: removed trolling
Last edited by system error on 30 Nov 2017, 06:05; edited 2 times in total |
|||
|
29 Nov 2017, 11:30 |
|
|
Furs 29 Nov 2017, 13:04
@system error: In respect to the output from gdb, that's not his code, it's the C runtime's code which calls his main, so thanks for proving my point?
Usually, when people code "main" function in asm, I don't assume them to use the C runtime (why wouldn't they use C otherwise?). Don't get me wrong, I understand people who want total control over the code in their app (obviously excluding external library calls), but linking the C runtime automatically makes your program not contain only your own code, since it pulls it, so I don't see much point in coding "main" in asm then (the app will be semi-bloated anyway; I mean if you want main in asm, minimal bloat should be a goal no? not like main is a hot path). (btw I don't use the C runtime even in C/C++ code, just pass -nostdlib to GCC you can still use some of the standard library as normal imported dynamic libs, just not baked into your own app; some do require an "init" function call __init_libc_main tho) |
|||
|
29 Nov 2017, 13:04 |
|
< Last Thread | Next Thread > |
Forum Rules:
|