Re: [PATCH] [4/7] Don't use large pages to map the first 2/4MB of memory

[Andi Kleen <andi@firstfloor.org>]

On Fri, Mar 21, 2008 at 07:03:19PM +0100, Thomas Gleixner wrote:
> On Fri, 21 Mar 2008, Andi Kleen wrote:
> > > Also we split the first GB mapping anyway due to the various regions
> > > (NX, RO, UC) in there.
> > 
> > I didn't think so unless you have DEBUG_RODATA enabled? 
> 
> NX is independent of DEBUG_RODATA 

Sure, but it is still not split by default. At least I don't see any code
for that anywhere except in my own patchkit.

> and the RODATA protection should be
> made unconditional on anyway.

Requiring hundreds instead of two TLB entries for the kernel text? 

I must say I personally cannot ever remember any bug caught by RODATA
anyways, so I am a bit dubious on its value.

> > Also there
> > should be no UC region there as known by the kernel. There might
> > be a WC region there from the frame buffer code, but that is an MTRR,
> > not a pageattr.
> 
> The first ioremap of the PCI space splits the GB page as well.

PCI space is normally in the fourth or sometimes third GB page,
not in the first.  I am not aware of any system that has the PCI hole 
in the first GB.

-Andi