wrong retf

[radare]

0x00000000 cafeba retf 0xffffffffffffbafe

ollydis shows retf 47870

[aquynh]

Is this regression, or a new issue?

[radare]

i didnt upgraded to the latest capstone because this test breaks.. but new output looks good.. so maybe this issues is fixed in recent next.

-0x004019da   # 3: ror ebp, 0xffffffffffffffe5
+0x004019da   # 3: ror ebp, 0xe5

[radare]

didnt verified this behaviour with other disassemblers or intel documentation either, you can check yourself

[aquynh]

So problem is solved?

[radare]

well, right now in capstone-next i get this:

$ ./cstool x32 cafeba
 0  ca fe ba                                         retf   -0x4502

ollydbg shows that:

$ rasm2 -a x86.olly -b 32 -d  cafeba
retf 47870

Zyan shows this:

$ rasm2 -a x86.zyan -b 32 -d  cafeba
ret 0xBAFE

so imho, old and new outputs are wrong, because the argument of retf should be unsigned.

In fact, i have written an r2 script to debug this, and i can confirm, the argument is unsigned when executing this in real hardware. same goes for the retn instruction.

This is the r2pipe script to verify this:

import r2pipe
r2 = r2pipe.open("./a.out", ["-d"])
r2.cmd("dcu main;3ds;.dr*")
print(r2.cmd("?v rsp-rbp"))
r2.quit()

and the stupid hello world program to run

main() {
 //  asm("retn 0xff00");
  asm("retn 0xff");
}

[aquynh]

Fixed, please confirmed