Re: Oops in 2.6.23-rc1-git9, arch/x86_64/pci/k8-bus.c::fill_mp_bus_to_cpumask()

["Yinghai Lu" <yhlu.kernel@gmail.com>]

On 8/4/07, Muli Ben-Yehuda <muli@il.ibm.com> wrote:
> On Sat, Aug 04, 2007 at 10:52:22PM -0700, Andrew Morton wrote:
> > On Sat, 4 Aug 2007 12:02:35 -0700 "Yinghai Lu" <yhlu.kernel@gmail.com> wrote:
> >
> > > On 8/4/07, Andrew Morton <akpm@linux-foundation.org> wrote:
> > > > On Sat, 4 Aug 2007 10:45:31 -0700 "Yinghai Lu" <yhlu.kernel@gmail.com> wrote:
> > > >
> > > > > Andrew,
> > > > >
> > > > > still need
> > > x86_64-get-mp_bus_to_node-as-early-v2.patch is already in the -mm
> > >
> > > it fixs
> > > pirq_peer_trick(void) in irq.c
> > > pcibios_fixup_peer_bridges in legacy.c
> > >
> > > by allocate sd.
> > >
> >
> > I don't understand you much, sorry.  Are you saying that
> > x86_64-get-mp_bus_to_node-as-early-v2.patch should be merged in
> > 2.6.23?
>
> If I may try to elaborate, yhlu is pointing out that the ->sysdata
> conversion missed a couple of spots in i386 where we need to allocate
> a pci_sysdata object as well. Unfortunately I don't think
> x86_64-get-mp_bus_to_node-as-early-v2 is a suitable fix as it does a
> whole bunch of other stuff too. I am now preparing a minimal patch
> (based on x86_64-get-mp_bus_to_node-as-early-v2) that only takes care
> of the missing ->sysdata bits. If that patch cures the original bug
> and doesn't cause any more regressions, it's definitely 2.6.23
> material.

You just use my patch plus Andy's patch together...

YH