Re: PIII.patch

dledford@redhat.com
Fri, 11 Jun 1999 06:38:34 -0400

I Lee Hetherington wrote:
>
> Setting CONFIG_X86_XMM seems to be the culprit. With this enabled, the
> kernel uses the XMM registers for memset, memcpy, etc., but it appears
> that their state isn't being properly saved and restored. Actually, I
> believe it is being saved properly, but not always restored properly.
> My program would show corruption of the registers, but they would
> eventually be restored by the kernel.
>
> With CONFIG_X86_XMM disabled, user-level XMM registers seem okay.
>
> --Lee Hetherington

OK, this would indicate that there is a problem. However, we use a scheme
that fundamentally should keep the registers consistent as the user space
level just like the MMX routines in the raid5 xor code of Ingo's. Evidently
though, that scheme fails under certain conditions, so I'll have to revisit it
and see what I can do (personally, my machine never gets into swap and as well
I don't have any userlevel XMM register using code so I didn't see this
behaviour, but that's why I released a test patch...)

-- 
  Doug Ledford   <dledford@redhat.com>
   Opinions expressed are my own, but
      they should be everybody's.


-
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/