I'll justify this message's presence in the rlug archives by adding that
Linus as also disgusted by Intel's treatment/denial of the "ia32e"
fiasco.  Linus' reaction can also be found, but he lacks Theo's, erm,
flare, I'll call it.

----- Forwarded message from Theo de Raadt <[EMAIL PROTECTED]> -----

> Sender: [EMAIL PROTECTED]
> From: Theo de Raadt <[EMAIL PROTECTED]>
> Subject: OpenBSD/amd64 on new Intel hardware
> Date: Fri, 27 Feb 2004 12:01:10 -0700
> To: [EMAIL PROTECTED]
> 
> So we've had OpenBSD/amd64 tested on not-yet-released Intel ia32e
> hardware.  That's the new Intel hardware that has the AMD64
> instructions added.
> 
> And the boot floppy works ...
> 
> Of course, it is working without W^X protection.  This is because
> Intel decided against adding the MMU page table bit for NXE -- per
> page non-execute.  On most processors this is about 180 logic gates to
> block loading of ptes which have NXE set into the ITLB (and cause
> a fault instead).
> 
> In 64 bit mode, of course, the processor architecture entirely lacks
> segmentation, so i386 trick with the CS limit is not possible.
> 
> Intel might have some hidden non-executable feature that we are not
> yet aware of, but that is the entire problem with Intel: Secrecy,
> secrecy, fucking bullshit secrecy.
> 
> Therefore, if anyone is unfortunate and gets their hands on one of
> these, I might urge you to run OpenBSD/i386 on it instead.
> OpenBSD/i386 has the standard CS-limit W^X protection.  But the way we
> see it now, Intel crippled their processor by making it IMPOSSIBLE to
> make tag memory as non-executable in 64 bit mode.
> 
> On the other hand, OpenBSD/amd64 on our AMD cpus are doing W^X just
> fine.
> 
> Apparently Intel cloned the entire AMD64 architecture but missed out
> on a very small but crucially important piece.

----- End forwarded message -----
_______________________________________________
RLUG mailing list
[EMAIL PROTECTED]
http://www.rlug.org/mailman/listinfo/rlug

Reply via email to