Re: [PATCH] Use global TLB flushes in MTRR code

[Ingo Molnar <mingo@elte.hu>]

* Andi Kleen <andi@firstfloor.org> wrote:

> > > It changes behaviour in some minor ways but I don't think it makes 
> > > any difference. PGE only influences TLB flushes (according to its 
> > > specification) and all the TLB flushes still run with PGE 
> > > disabled.
> > 
> > now that i pointed out the difference, your position changed to 
> > "changes behavior in minor ways" ;-)
> 
> The instruction stream changes (more cr* accesses), but the actual 
> flushes do not. [...]

You are still dead wrong. You refuse to realize the _obvious_ and 
_fundamental_ thing that the PGE bit does: it enables global TLBs while 
we change MTRRs!

That would be a completely new and totally untested modus operandi for a 
large array of x86 CPUs. You should read the Intel manual about the 
recommended way to change MTRRs (Vol. 3A 10-41) - it describes disabling 
the PGE during MTRR changes if it's enabled. The primary purpose of that 
is of course to flush the TLB entries - but still it's documented that 
way and the current code does it that way. It would be rather stupid for 
us to do otherwise: we simply cannot guarantee that x86 CPUs that 
implement MTRRs and PGE have no undocumented or _unknown_ erratas in 
this area. It _might_ be fine, while what you say is that it _is_ fine - 
which we simply cannot know. (And yes, there are documented CPU erratas 
related to global TLBs and MTRR's, so this area is far from being an 
unwritten page.)

So i refuse to apply such a pointless and risky "cleanup" patch from you 
to arch/x86 [which increases code size as well], _especially_ given how 
many times i've explained this very issue to you already and especially 
because you continue to mislead about the true effects of the patch.

	Ingo