[coLinux-users] PAE issue fixed (was: Re: colinux-0.6.2-pre1 crashed on athlon 64)

[Dan A. <da...@co...>]

On Mon, Sep 27, 2004 at 10:43:28PM +0100, Nuno Lucas wrote:
> Joe Wells, dando pulos de alegria, escreveu :
> >
> >Is this an issue only for AMD CPUs or does it also affect some Intel
> >CPUs?
> >
> >I've tried to follow this thread on the mailing lists and it is not
> >clear to me what conclusion to draw.
> >
> 
> To make it clearer, the problem can be triggered with any Pentium
> II/III/IV and the old Pentium Pro. And any compatible AMD that
> implements the PAE mechanism (Physical Address Extensions).
> 
> Windows doesn't enable PAE unless explicitely ordered to (with the /PAE
> switch), but in order to use DEP/NX, the processor needs to be run with
> PAE enabled (to have 64 bits page tables, instead of 32 bits).
> 
> So, on Amd64 processors (and maybe Intel Ithanium, not sure about this),
> the bug will be triggered as soon as one enables the SP2 DEP/NX feature,
> but because PAE is also enabled.
> 
> What is not tested yet is if, after correcting the PAE issue, it will
> have problems with the DEP/NX one. But the second should be easy enough
> to solve.

I've just finished committing a fix for the PAE issue. It took the whole day,
and there were more extensive changes than I thought. The context switch code 
is a bit larger now (and also works differently than what you read in the 
slides and OLS paper) but expect no notable overhead like before.

Anyway, it appears to be stable on my PAE enabled XP SP2 and also on
my Linux boxes and Windows 2000 in QEMU. Users can test it by compiling
the binaries or wait until I release a snapshot this weekend.

-- 
Dan Aloni
da...@co...