Re: [Qemu-devel] [PATCH v5 15/24] hw: i386: Export the i386 ACPI SRAT build method

[Igor Mammedov <imammedo@redhat.com>]

On Thu, 22 Nov 2018 00:27:33 +0100
Samuel Ortiz <sameo@linux.intel.com> wrote:

> On Thu, Nov 15, 2018 at 02:28:54PM +0100, Igor Mammedov wrote:
> > On Mon,  5 Nov 2018 02:40:38 +0100
> > Samuel Ortiz <sameo@linux.intel.com> wrote:
> >   
> > > This is the standard way of building SRAT on x86 platfoms. But future
> > > machine types could decide to define their own custom SRAT build method
> > > through the ACPI builder methods.
> > > Moreover, we will also need to reach build_srat() from outside of
> > > acpi-build in order to use it as the ACPI builder SRAT build method.  
> > SRAT is usually highly machine specific (memory holes, layout, guest OS
> > specific quirks) so it's hard to generalize it.  
> Hence the need for an SRAT builder interface.
so far builder interface (trying to generalize acpi_build()) looks
not necessary.
I'd suggest to drop and call build_start() directly.

> > I'd  drop SRAT related patches from this series and introduce
> > i386/virt specific SRAT when you post patches for it.  
> virt uses the existing i386 build_srat() routine, there's nothing
> special about it. So this would be purely duplicated code.
Looking at build_srat(), it has a bunch of code to handle legacy
PC layout. You probably don't need any of it for new is86/virt
machine and can make simpler version of it.

In addition (probably repeating question I've asked elsewhere),
Do you have to use split initial memory model for new machine?
Is it possible to use only pc-dimms both for initial and hotplugged memory
at some address (4Gb?) without cutting out PCI hole or any toher holes in RAM layout?

> Cheers,
> Samuel.
>