Re: [PATCH] Move assembly code out of the kernel

[Bean]

On Thu, Jul 31, 2008 at 7:22 PM, Robert Millan <address@hidden> wrote:
> On Thu, Jul 31, 2008 at 03:46:47PM +0800, Bean wrote:
>> diff --git a/include/grub/symbol.h b/include/grub/symbol.h
>> index e951490..eef966f 100644
>> --- a/include/grub/symbol.h
>> +++ b/include/grub/symbol.h
>> @@ -43,4 +43,13 @@
>>  # define EXPORT_VAR(x)       x
>>  #endif /* ! GRUB_SYMBOL_GENERATOR */
>>
>> +#define REAL_STUB_START(x)   x ## _stub: ; \
>> +     .long   x ## _start ; \
>> +     .long   x ## _end - x ## _start ; \
>> +x ## _start: ; \
>> +     .code16
>> +
>> +#define REAL_STUB_END(x)     .code32 ; \
>> +x ## _end:
>
> Shouldn't this be in an i386/pc/ subdir?

Oh, perhaps I can move them to include/i386/pc/kernel.h

>
>> + * On entry, %eax should points to a structure that specifies the code/data 
>> to
>> + * be copied:
>> + *
>> + *   offset 0: address of code/data
>> + *   offset 4: size of code/data
>> + *
>> + * grub_copy_real_stub fetches the address from offset 0, if it's not below 
>> 1M or
>> + * properly aligned, it copies it to a reserved area and updates the 
>> address at
>> + * offset 0, so that it don't need to do it again the next time.
>> + *
>> + * Currently, the reserved area for stub code/data is 0x80000 - 0x88000. It
>> + * should be big enough for assembly code.
>
> I wonder if this could be made simpler.  For example, by having a function
> that just calls "int" in real mode.  Then callers could use inline assembly
> to put parameters in appropiate registers and reading results from appropiate
> registers.  Would this be faster than copiing hooks?

INT sounds good, but it can't cover all cases. For example, it can't
pass parameter using stack, or call a far pointer. Besides, we still
need the stub area to store data.

-- 
Bean