Re: [PATCH/RFT] finish i386 and x86-64 sysdata conversion

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

On 8/7/07, Jeff Garzik <jeff@garzik.org> wrote:
> Yinghai Lu wrote:
> > the sysdata stuff try to fix:
> > when calgary iommu code was introduced, it is trying to share sysdata
> > that orginally is initialized k8_bus.c and used by pcibus_to_node. but
> > that only happen with AMD K8 platform. and have nothing to do with
> > calgary.
>
> incorrect:  sysdata is initialized when the structure is created, which
> clearly not only in k8_bus.c

before his first patch, numa juse use sysdata pointer for node.

>
>
> > Then my patch get_mp_bus_to_node_as_early will use that sysdata as
> > node as early as possible before the pci_scan bus is called.
>
> that's incorrect, since it assumes NUMA owns ->sysdata

At that time, only it use that.

>
>
> > then Mul came out his patch ...to split the sharing that will never happen.
>
> that's incorrect, ->sysdata will be shared among Calgary, NUMA, and
> PCI-domain support.

but not the same time on run time system with Calgary, NUMA.

>
>
> > Also i don't think anyone test muli sysdata patch with amd64 platform
> > with acpi=off. even muli himself.
>
> I did.  I did all PCI domain support development on AMD64 platform.

really, but his first patch is broken with acpi=off.
I guess you tested his patch plus your patch. instead of test his
patch barely.  I tested his patch + my patch, and it works well, and
even test his patch only.

>
>
> > calgary support bus numa at this time? i mean mult peer root bus on
> > different node..
>
> My PCI domains patch add support for multiple peer root buses.  That's
> the definition of PCI domain support.
>
> Branch 'pciseg' of
> git://git.kernel.org/pub/scm/linux/kernel/git/jgarzik/misc-2.6.git

good, i will look at it.

YH