Badari Pulavarty wrote:
On Tue, 2005-11-08 at 17:17 -0500, Dave Anderson wrote:
> Badari Pulavarty wrote:
>
> > On Tue, 2005-11-08 at 14:52 -0500, Dave Anderson wrote:
> > > BTW-- what kernel version introduced this change for ppc64?
> >
> > 4-level pagetable went into 2.6.14 for ppc64.
> > Yuck. That reminds me that "pud_clear_bad()" is not a good
> > choice for detecting new layout (its a generic routine - not
> > ppc64 specific and is in 2.6.11 onwards).
> >
> > Funny thing is, I am not sure original crash is working
> > on pre-2.6.14 kernel either. I see PTE_SHIFT changed from
> > 16 to 17 in 2.6.10. I guess crash is broken from 2.6.10
> > onwards for ppc64.
> >
> > 2.6.10 onwards:
> > #define PTE_SHIFT (17)
> >
> > < 2.6.10: (which crash understands)
> > #define PTE_SHIFT (16)
>
> The PTE_SHIFT change was orthogonal to the intro of 4-level page tables?
> You would think that they would be tied together, no?
PTE_SHIFT is orthogonal to 4-level changes and it happened in 2.6.10.
My point is, "crash" wouldn't work for kernels between 2.6.10-2.6.14
(even with my changes).
Right -- that's what I thought you were implying. Does that bother
you guys (IBM)? I'm looking at throught Red-Hat-colored glasses,
presuming that RHEL5 will be based on 2.6.14 or later. It seems to
be a rather large hole, but I'll let you guys decide on how to best
handle it (or not) .
Dave