x86 boot code build

[Garrett Cooper]

On Fri, Oct 5, 2012 at 12:40 AM, Garrett Cooper <yanegomi at gmail.com> wrote:
> On Fri, Oct 5, 2012 at 12:08 AM, Dag-Erling Smørgrav <des at des.no> wrote:
>> Konstantin Belousov <kostikbel at gmail.com> writes:
>>> Dag-Erling Smørgrav <des at des.no> writes:
>>> > Dimitry Andric <dimitry at andric.com> writes:
>>> > > Well, do we still officially support any real i386 machines?
>>> > No, 486 and up only.  Personally, I think we should ship 586
>>> > binaries (pentium-mmx) by default.
>>> There is absolutely no architectural difference between usermode ISA
>>> between i386 and pentiums, ignoring SMP-support instructions, which
>>> are usually not emited by the compiler anyway.
>>
>> By "binaries" I mean ISOs and freebsd-update, including the kernel.
>>
>> (actually, it's the kernel I care the most about)
>>
>>> Really interesting stuff started appearing with pentium pro, like CMOV
>>> instructions. Even more important, -march=pentiumpro generates much
>>> better -fPIC code (probably could be activated by -mcpu=pentiumpro).
>>
>> Which is why most Linux distributions target 686, but we can't if we
>> want to support small systems like the AMD Geode-based soekris net4xxx
>> and net5xxx out of the box.
>
>      I would target the appropriate architecture (amd64) where it
> matters (amd64), and target the lowest sane common denominator on
> i386. In reality, what does a couple MB mean on amd64 vs i386?

To clarify...

s/couple MB/couple MB in the loader program/

Thanks,
-Garrett