0

I will try to use the value analysis on a project in C, but this project contains some .c file where we can find assembler code. When I try to launch frama-C on these files, I get an error on the assembly code.

The assembler code is desgned for Motorola 68040, I have seen in the documentation, that I need to use the option -machdep to change the analysis platform for the module, but this platform isn't defined, so do I need to contact the support or can I configure the module to ignore the assembler code?

And a second question, on SpareCode module. Can we configure the module, to just see the dead code and conserve the spare code (in the case of a procedure)?

The code of otherfile.c (without comment) :

#pragma asm

     XDEF _CONVERSION_INTEL

MESSAGE     SET 20
NB_CARAC    SET 26

     SECTION mc3_sys_code

_CONVERSION_INTEL
    movem.l d1-d3/a0,-(sp)
    move.l  MESSAGE(sp),a0
    moveq   #0,d1 
    move.w  NB_CARAC(sp),d1
    moveq   #0,d3

 PERMUTE:
    move.w  (a0),d2
    rol.w   #8,d2
    move.w  d2,(a0)
    addq.l  #2,a0
    addq.l  #2,d3
    cmp.l   d3,d1
    bgt PERMUTE test
    movem.l (sp)+,d1-d3/a0
    rts

#pragma endasm
user3675417
  • 13
  • 3
  • "I get an error on the assembly code" is a bit vague. What is the actual message, and which command line does trigger it? – Virgile May 26 '14 at 09:22
  • My command line is : `frama-c-gui -absolute-valid-range 0x00000000-0xffffffff -val -slevel 200 -plevel 200 mainfile.c -main entryfct otherfile.c &`, the error message is : `otherfile.c:25:[kernel] user error: syntax error` and at this line we can find `_kill_warning();`. If I delete this line, the new error is `otherfile.c:29:[kernel] user error: syntax error` on ` XDEF _CONVERSION_INTEL`. when I compile the `otherfile.c` with `gcc -C -E -I. otherfile.c`, I have not error. – user3675417 May 26 '14 at 11:38
  • `gcc -C` indicates that `gcc` should stop after the pre-processing phase, so the fact that you don't have any error is not an indication that the file correctly typechecks. You should check whether `gcc -c -I. otherfile.c` gets correclty compiled. In addition, none of the lines you quote do contain assembly code, as the title suggests. This makes it difficult to help you without seeing the file `otherfile.c`. – Virgile May 26 '14 at 13:39
  • I add code on first message. `otherfile.c` doesn't compile with `gcc -c -I.`. I think that `gcc` doesn't know all keyword (I don't see `asm` option for `#pragma` on `gcc` documentation), but I know this code is executed. (I'm in internship, it's not my code) They use Microtec for compile their code, I had thought it was the same at `gcc`, so I need to configure another compiler ? And for `-machdep`, the Motorola processor is in big-endian, I need to change the platform too ? – user3675417 May 26 '14 at 14:42
  • Or maybe, I must use the instruction: `asm("assembler code");` ? Because some file have this instruction and Frama-c don't get a error on this instruction. – user3675417 May 26 '14 at 15:06

1 Answers1

4

Frama-C does not handle assembly code. It can parse some inline assembly (gcc's asm(...) instruction), but not a whole file. You should try to figure out what the PERMUTE procedure is doing and provide a replacement, either with a plain C definition or as a prototype + ACSL specification (the former being preferable if you intend to use Value Analysis).

-machdep will not let Frama-C interpret any assembly code. This option mainly fixes the size of standard integer types (e.g. 32 bit for int) and their representation (big or little endian). If you need support for a specific architecture that is not included in the ones that are currently supported (as shown with frama-c -machdep help), you should indeed contact Frama-C's support to see under which condition this can be added.

In the graphical user interface, Value Analysis will show you the code that is unreachable from the entry point of the analysis as stroken out and on a red background. It is also possible to write a script/plug-in that would extract this information programmatically, but there's nothing in the distribution for exactly that.

Virgile
  • 9,724
  • 18
  • 42