Re: [RFC][PATCH 3/3] x86: make MP a required-feature on 64-bit

[Dave Jones <davej@redhat.com>]

On Fri, Jun 20, 2014 at 08:05:23PM +0200, Borislav Petkov wrote:
 > On Fri, Jun 20, 2014 at 10:47:22AM -0700, H. Peter Anvin wrote:
 > > This is run before static_cpu_has().
 > 
 > static_cpu_has_safe() then - I didn't do it for no reason :-)
 > 
 > > The point, though, was that we "enforce" (taint) on 32 bits but not on
 > > 64 bits, which is clearly wrong.
 > 
 > Yeah, K7 is 32-bit only.
 > 
 > > My inclination is to completely kill amd_k7_smp_check() entirely,
 > > since noone seems to know when it actually matters and it is clearly
 > > historic.
 > 
 > I think DaveJ should know something about it - he gave that impression
 > last time when we were discussing 8c90487cdc64 ("Rename TAINT_UNSAFE_SMP
 > to TAINT_CPU_OUT_OF_SPEC").

AMD sold two separate SKUs: the Athlon XP and the Athlon MP.
Only the latter was supposedly "certified" for use in multi-processor
boards.  People found out however that sometimes the XP's 'worked'
if you modded them (see http://www.hardwaresecrets.com/article/How-to-Transform-an-Athlon-XP-into-an-Athlon-MP/24)

There was belief that AMD had reason beyond "more price mark-up for MP's"
and that those fuses had been blown for good reason (failing validation
in some conditions for eg).

I doubt anyone is actually even running such a system any more on
a modern kernel, and any weird crashes would be written off more by
"you're running 10+ year old hardware, it's probably broken" than
"it was never meant to do that".

	Dave