Re: ghci on arm

[Karel Gardas <karel.gardas@centrum.cz>]

Hello!

Camm, thanks a lot for the suggestion. I've thought it would be really 
nice if we can reuse some of GCL code. Unfortunately we don't live in a 
perfect engineering world, but here we hit (probably!) the issue of 
different licenses in use. Am I right assuming GCL is using some kind of 
GPL/LGPL license? GHC on the other hand is using BSD (or BSD-like) so if 
I'm right we're not able to mix the code here.

Anyway, idea of reusing some other project code looks really appealing. 
Do you know any project dealing with ARM linker and yet with BSD or BSD 
like license?

Thanks a lot!
Karel


On 01/26/12 04:12 AM, Camm Maguire wrote:
> Greetings!  GCL has a working native linker now on armel and armhf.  We
> have a generic elf relocation file, with machine specific headers
> implementing only those relocs output by gcc.  I can step you through
> the code if interested.
>
> For example, here is the core file for elf32_arm, which should be almost
> readable:
>
> #define R_ARM_THM_CALL        10
> #define R_ARM_CALL 28
> #define R_ARM_V4BX 40
> #define R_ARM_THM_MOVW_ABS_NC 47
> #define R_ARM_THM_MOVW_ABS    48
>      case R_ARM_THM_CALL:
>        s+=a;
>        if (ELF_ST_TYPE(sym->st_info)==STT_FUNC) s|=1;
>        s-=p+4; /*FIXME maybe drop 4 and add_val below*/
>        s=((long)s>>1);
>        massert(!(abs(s)&0xffc00000));
>        store_val(where,MASK(11),s>>11);
>        store_val(where,MASK(11)<<16,(s&0x7ff)<<16);
>        break;
>      case R_ARM_THM_MOVW_ABS_NC:
>        s+=a;
>        if (ELF_ST_TYPE(sym->st_info)==STT_FUNC) s|=1;
>        s&=0xffff;
>        s=((s>>12)&0xf)|(((s>>11)&0x1)<<10)|((s&0xff)<<16)|(((s>>8)&0x7)<<28);
>        add_vals(where,~0L,s);
>        break;
>      case R_ARM_THM_MOVW_ABS:
>        s+=a;
>        s>>=16;
>        s=((s>>12)&0xf)|(((s>>11)&0x1)<<10)|((s&0xff)<<16)|(((s>>8)&0x7)<<28);
>        add_vals(where,~0L,s);
>        break;
>      case R_ARM_CALL:
>        add_vals(where,MASK(24),((long)(s+a-p))>>2);
>        break;
>      case R_ARM_ABS32:
>        add_val(where,~0L,s+a);
>        break;
>      case R_ARM_V4BX:
>        add_val(where,~0L,s+a);
>        break;
>
>
> I've found the quickest way to implement these is to make a simple a.out
> file with ld -R<main program>  <foo.o>  and reverse engineer, looking at
> binutils and some platform docs as necessary.
>
> I'm missing ia64 only -- any help here?
>
> Take care,
>
>
>
> Clint Adams<clint@debian.org>  writes:
>
> > On Tue, Jan 10, 2012 at 03:17:41PM -0500, Ben Gamari wrote:
> > > If anyone knows people experienced in ARM linker implementation, then feel
> > > free to send them my way. At this point I'm at the difficult point of
> > > figuring out why some relocation (plural?) sends the CPU to the
> > > middle of nowhere. I've had next to no experience with linker
> > > implementation in general, much less the mess that seems to be ARM, so
> > > feedback from someone with some experience in this field would be
> > > definitely appreciated. As Karel said, there is no Haskell code here at
> > > all; strictly C.
> >
> > Maybe Camm has dealt with something similar already and can offer
> > some insight?