Re: [Caml-list] Compiler bug?

[Dmitry Bely <dmitry.bely@gmail.com>]

On Mon, Aug 6, 2012 at 2:34 PM, Alain Frisch <alain@frisch.fr> wrote:
> On 08/06/2012 12:20 PM, Dmitry Bely wrote:
>>
>> I always thought that local roots are memory-based. How GC knows that
>> in this case [ebx] should be updated? (remember, the parameter was
>> passed in [eax]) Could you point me to the relevant part of Ocaml
>> compiler sources?
>
>
> The compiler generates 'frame descriptors', which associate to each possible
> GC call site the set of registers and stack slots which hold values.
>
> Look for caml_frame_descriptors.  The logic to scan frame descriptors is in
> asmrun/roots.c.  The implementation of caml_call_gc (e.g. in amd64.S)
> saves/restored machine registers to/from caml_gc_regs.

Aha. _caml_call_gc is

_caml_call_gc:
    ; Record lowest stack address and return address
        mov	eax, [esp]
        mov     _caml_last_return_address, eax
        lea     eax, [esp+4]
        mov     _caml_bottom_of_stack, eax
    ; Save all regs used by the code generator
L105:   push    ebp
        push    edi
        push    esi
        push    edx
        push    ecx
        push    ebx
        push    eax
        mov     _caml_gc_regs, esp
    ; Call the garbage collector
        call	_caml_garbage_collection
    ; Restore all regs used by the code generator
	pop     eax
        pop     ebx
        pop     ecx
        pop     edx
        pop     esi
        pop     edi
        pop     ebp
    ; Return to caller
        ret

it preserves registers and so it cannot update them, like it does with
memory-based local roots. The initial question if [ebx] contents is
invalidated by GC remains open, at least for me.

- Dmitry Bely