Re: [PATCH] mem/x86: add processor address space check for VM memory

[Ani Sinha]

On Mon, Sep 18, 2023 at 4:28 PM David Hildenbrand <david@redhat.com> wrote:
>
> On 18.09.23 12:54, Ani Sinha wrote:
> > On Mon, Sep 18, 2023 at 3:49 PM David Hildenbrand <david@redhat.com> wrote:
> >>
> >> On 18.09.23 12:11, Ani Sinha wrote:
> >>
> >>>
> >>>      Ok hopefully my last question. I am still confused on something.
> >>>      Does the above mean that the hole64 will actually start from an
> >>>      address that is beyond maxram? Like basically if you added all of
> >>>      ram_below_4G, ram_above_4G, hot plug_mem and pci_hole64 then can it
> >>>      exceed maxram? I think it will. Does this not an issue?
> >>
> >> If you'd have a 2 GiB VM, the device memory region and hole64 would
> >> always be placed >= 4 GiB address, yes.
> >>
> >> As maxram is just a size, and not a PFN, I don't think there is any
> >> issue with that.
> >
> > So this is all just a scheme to decide what to place where with maxram
> > amount of memory available. When the processor needs to access the
>
> Yes. ram_size and maxram_size are only used to create the memory layout.
>
> > memory mapped PCI device, its simply dynamically mapped to the
> > available physical ram. Is my understanding correct here?
>
> I'm no expert on that, but from my understanding that's what the
> pci/pci64 hole is for -- mapping PCI BARs into these areas, such that
> they don't conflict with actual guest RAM. That's why we still account
> these "holes" as valid GFN that could be used+accessed by the VM once a
> PCI BAR gets mapped in there.

Yes that was my understanding too but since device drivers need to
access those BAR addresses, they need to be mapped to the actual
available physical ram.