[PATCH] fix clobbers in the K8 CAR

[Rudolf Marek]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hello,

This patch adds clobber regs into the play as assembly trashes random registers.
I was bitten by this while doing MSR I/O because ECX got corrupted. I attempted
to create nice barrier for GCC while changing the stack location too.

The memcpy code was taken from Linux kernel, dont know to whom attribute the
copyright on it. If it is OK with the note, otherwise:

Signed-off-by: Rudolf Marek <[e-mail]>

Thanks,
Rudolf
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.10 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAkuBtsgACgkQ3J9wPJqZRNXtXgCeJGvuA7dTdE5A+Tybw6zNIi2x
YbkAni1oBOZ9i6iqartAdKxaI8xJP76T
=ebvY
-----END PGP SIGNATURE-----

[ron minnich]

2010/2/21 Rudolf Marek <[e-mail]>:
> This patch adds clobber regs into the play as assembly trashes random registers.
> I was bitten by this while doing MSR I/O because ECX got corrupted. I attempted
> to create nice barrier for GCC while changing the stack location too.

The complex ways in which we have to convince gcc to do __asm__
correctly make me wonder if we shouldn't start
putting assembly into assembly into .s files, not .c files. If we have
this much trouble with __asm__, how much are we really
gaining when we use it? Why not just set up a .s?

ron

[Patrick Georgi]

Am 22.02.2010 17:31, schrieb ron minnich:
> The complex ways in which we have to convince gcc to do __asm__
> correctly make me wonder if we shouldn't start
> putting assembly into assembly into .s files, not .c files. If we have
> this much trouble with __asm__, how much are we really
> gaining when we use it? Why not just set up a .s?
> romcc in some cases.
> The other issue is that in an .s file, we're bound to function calling
> conventions. with __asm__, we can really include code.

Patrick

[ron minnich]

On Mon, Feb 22, 2010 at 8:34 AM, Patrick Georgi <[e-mail]> wrote:
> The other issue is that in an .s file, we're bound to function calling
> conventions. with __asm__, we can really include code.

I understand. But I work with a very nice operating system that has
never had a need for __asm__, so I question the need for it myself.

I'd still like to see the reason that memset is so slow.

Also, I realize we have dropped support for other architectures, but
coreboot is still in principle portable. This change will be the
beginning of the removal of the portability. Before we start to make
such a change we need to make a conscious decision that losing
portability is what we want.

Part of "portability", in this case, is eventually moving to native
64-bit mode for the x86_64. I think we'll have to do that when we have
machines with more memory than PAE can address.

thanks

ron

[Patrick Georgi]

Am 21.02.2010 23:42, schrieb Rudolf Marek:
> Hello,
>
> This patch adds clobber regs into the play as assembly trashes random registers.
> I was bitten by this while doing MSR I/O because ECX got corrupted. I attempted
> to create nice barrier for GCC while changing the stack location too.
>
> The memcpy code was taken from Linux kernel, dont know to whom attribute the
> copyright on it. If it is OK with the note, otherwise:
>
> Signed-off-by: Rudolf Marek <[e-mail]>
> Acked-by: Patrick Georgi <[e-mail]>

and committed as r5153