Re: [Qemu-devel] [PATCH 13/27] acpi: memory hotplug ACPI hardware implementation

[Igor Mammedov <imammedo@redhat.com>]

On Mon, 25 Nov 2013 14:45:15 +0100
Paolo Bonzini <pbonzini@redhat.com> wrote:

> Il 25/11/2013 08:27, Markus Armbruster ha scritto:
> > "Michael S. Tsirkin" <mst@redhat.com> writes:
> > 
> > [...]
> >> Here this made me think of the following question:
> >> should not memory available on boot be listed in CMOS?
> >> Let's assume new BIOS gets interrupt telling it there's new memory.
> >> It clears the interrupt and then system is reset.
> >>
> >> Will system after reset detect the new hotplugged memory?
> 
> Can it just scan all slots from MHPD._INI?
> 
> For example on reset all slots could be disabled by the
> ACPIHotpluggableDimmBus (even coldplugged ones), and scanned + enabled
> by ASL.
memory devices are scanned by OSPM during ACPI tables parsing and 
guest (can) use present devices right away.

Currently I don't see necessity for enabling individual memory devices from
ASL since it adds one more round trip for QEMU<->guest protocol for no
particular benefit. The only use case Michael suggested would be running
OLD BIOS (without memory hotplug support) on machine with actively used
memory hotplug (if OLD BIOS runs on machine with memory hotplug but there is
no hotplugged memory devices, it runs just fine), so case looks like not worth
an effort and complexity it would bring.

I believe having completely functional DIMM device after its realize() method
is completed is more strait-forward and easier to understand/support than
having guest to complete it's initialization.
That would allow later to put present at boot DIMMs into E820 (way before ASL
is executed) if that would be necessary and also help to convert initial
memory into hotpluggable one as well.

> 
> >> How?
> > 
> > Do physical systems with hot-pluggable memory exist?  How do they
> > behave?
> 
> I guess they assume BIOS is updated in lockstep with the introduction of
> the feature.
> 
> Paolo
>