> >Would it be sufficient if we
> >compiled using the CMPXCHG instruction and then did a run-time test for the
> >386 and never executed this instruction?
>
> Yes. That would be fine. I know that your code compiled just
> fine when I got rid of the ifdef that checked for 386, so I know
> the assembler will not choke on the instruction, even though the
> compiler is being called with -m386.
>
> Richard's solution sounds OK too. As you are probably aware,
> there have been similiar discussions about this type of approach
> to MMX and other x86 instruction additions. I guess it's a function
> of how much overhead you see as associated with the run time check
> for the 386 CPU.
>
> If the run time check for a 386 is a small overhead, but
> something you would still like to avoid when possible, and if
> do not want to implement Richard's proposed emulation for whatever
> reason, then you might want to consider arranging things so that
> you only compile in the checks if CONFIG_M386 is set, since not
> setting CONFIG_M386 will already generate a kernel that cannot run
> on a 386 (because it will compile in bswap instructions which do not exist
> on a 386). So, if CONFIG_M386 is not set, you know this kernel already
> can only run on a processor that supports cmpxchg, so you can skip the
> run time tests. The best style for doing this would be to define a
> symbol like CONFIG_X86_CMPXCHG in arch/i386/config.in, and test
> for that symbol.
I get the feeling that we are adding more and more cruft to the M386 path,
that is only used for processors >386. While I can see the reasoning
behind this, maybe it's time to introduce a new compile-option.
CONFIG_M386_GENERIC, which is the crufted version with compability for all
processors
and
CONFIG_M386, which is cruftfree and known to work on 386's, without any
optimisations for other processors.
In my opinion, Linux is getting a bit too slow on 386's, without any real
reason.
/David Weinehall
_ _
// David Weinehall <tao@acc.umu.se> /> Northern lights wander \\
// Project MCA Linux hacker // Dance across the winter sky //
\> http://www.acc.umu.se/~tao/ </ Full colour fire </
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/