>>>> gcc for i386 can be used with the assembly prefix ".code16gcc" to
quoted text
>>>> generate
>>>> 16-bit (real-mode) code.  This header file provides the assembly
>>>> prefix.
>>> This only works correctly with newer GCCs if you pass the
>>> -fno-toplevel-reorder option (and it only works on older
>>> GCC versions by accident).
>>
>> And on older ones (gcc-3.3-hammer / 3.4 / 4.0 ) -fno-unit-at-a-time
>>
>> Still it seems quite fragile to me agreed.
>
> The alternative, of course, is to compile to an .s file and insert
> .code16gcc into the .s file.  This makes the Makefile uglier, but  
> would
> be more resilient against oddball gcc changes.

This would be even more fragile.  The exact format of GCC's
assembler code output isn't defined at all, so in principle
this is a hopeless task.  In practice just putting the
.code16gcc directive on the first line would likely work
though, GCC never generates a .code32 AFAIK, but it isn't
guaranteed that this will work (or will keep working).

quoted text
> I'm a bit surprised about the claim w.r.t. -fno-unit-at-a-time  
> (although
> I guess that is the default and one would thus typically not see  
> this.)

-fno-unit-at-a-time is the default on three year old compilers,
yes.  Newer compilers have unit-at-a-time enabled by default.

quoted text
>  Got any pointers why that would cause a global asm() to be scrambled
> around?

An asm() outside of a function is a top-level thing, just
like functions and file-scope variable definitions, and with
unit-at-a-time compilation all such blocks can be reordered
(or even omitted, or inlined, or whatever).  Note that even
before unit-at-a-time was introduced there was no guarantee
of emitting everything in the order it appears in the source
file.


Segher

-
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/