On Tue, May 13, 2008 at 06:02:25PM -0700, Andrew Morton wrote:
> > > In case this is a genuine bug, somebody else more familiar
> > > with that stuff should evaluate it's effects (I just found it
> > > by some shell pipeline and it seems suspicious looking).
> > Should be fairly as proven by practice; it's there since day of of 64-bit
> > pagetable for 32-bit hw support which was November 29, 2004.
> It's unlikely that anyone would notice an error in pte_mkyoung(). It
> will affect page reclaim behaviour and _might_ be demonstrable with a
> carefully set up test. But an error in here won't cause crashes or
> lockups or anything.
It's even more subtle than that. Only a special variant of the pagetables
used for the MIPS equivalent of PAE is affected. For performance reason
this format contains two copies of the some of the bits. The result of
the bug was one being maintained corrected, the other one not.
> What this needs is someone who understands the architecture (ie: you
> ;)) to take a look, please.
I've applied the patch yesterday; will go to Linus in the next merge.