Armt: add some instructions

[b14aa178]

im not ensure it right, but dis is ok.

[serpilliere]

I am restarting CI here, as it seems the faulty test is due to bad travis.

[b14aa178]

hi @serpilliere, i add test code for instruction, but why test/arch/arm/sem.py seems do not include armt semantic test? do you forget it or other reason? if i write some armt instruction semantic test, it' s a right place?

[serpilliere]

Hi @wzrybuddy !
Currently in miasm, we have tests for the assembler/disassebmler in test/arch/arm/arch.py in which we try to assemble/disassemble each supported instruction (ok, maybe not every form of every instructions...)
It's a bit different for semantic: we only add regression tests for complex semantic or if the semantic is really prone to error.

To test the semantic, instead of testing the semantic directly, our preference goes to implement a little code and emulate it with special case, then check the registers/flags.

For example, here is the test for an instruction in aarch64, in miasm/test/arch/aarch64/unit/mn_ubfm.py:

#! /usr/bin/env python2

import sys

from asm_test import Asm_Test
from pdb import pm


class Test_UBFM1(Asm_Test):
    TXT = '''
main:
       MOVZ    X0, 0x5600
       UBFM    X0, X0, 8, 15
       RET     LR
    '''
    def check(self):
        assert(self.myjit.cpu.X0 == 0x56)
        pass

class Test_UBFM2(Asm_Test):
    TXT = '''
main:
       MOVZ    X0, 0x56
       UBFM    X0, X0, 4, 55
       RET     LR
    '''
    def check(self):
        assert(self.myjit.cpu.X0 == 0x5)
        pass


if __name__ == "__main__":
    [test(*sys.argv[1:])() for test in [Test_UBFM1, Test_UBFM2 ]]

You have a bunch of these for x86 in miasm/test/arch/x86/unit/` if you want more examples!

[serpilliere]

Ho, by the way, I will fix the tipo and relaunch your tests: it seems broken tests here are not your fault: it's only due to our bad english...

[serpilliere]

I fix tipo with #1041
Once it's ok and merged, I will hint you here, then you will be able to rebase your PR and fully pass regression tests.

[serpilliere]

It seems some of your instructions have a length of 31 bits.

[serpilliere]

Hi @327135569 ,
Thank you for the PR update! (and sorry for the delay)
The PR seems ok. The only thing that may a problem is the semantic.
Maybe we could either:

• remove the semantic for the todo (which will create an error on run time if we encounter it =
• or explicitly add a raise` NotimplementedError, which will also raise an error at run time

I think this is necessary because if we let code empty, it will silently give an IR which will be wrong (as mnemonic will miss) and it is quite disturbing during analysis.

Are you ok to fix this or do you want me to do a pr on your pr ?

[b14aa178]

Thank you! I will try to add semantic later.

[serpilliere]

Hey, you are close to a good semantic.
Just a remark here, as you are doing a [0:32] (which will extract the first 32 bits) you don't need to mask it before with the & 0xFFFFFFFF

[serpilliere]

Same here, the bit truncation will do the mask job

[serpilliere]

Hey @327135569
It seems your are very close to the good semantic, I just added 2 remarks here are I think everything will be ok!

[b14aa178]

Gosh, i am too slow to add semantic.. but seems OK, can you merge ?