On Thu, Sep 28, 2023, Christoph Hellwig wrote:
> On Mon, Sep 11, 2023 at 11:16:30AM +0900, David Stevens wrote:
> > From: David Stevens <stevensd@xxxxxxxxxxxx>
> >
> > This patch series adds support for mapping VM_IO and VM_PFNMAP memory
> > that is backed by struct pages that aren't currently being refcounted
> > (e.g. tail pages of non-compound higher order allocations) into the
> > guest.
> >
> > Our use case is virtio-gpu blob resources [1], which directly map host
> > graphics buffers into the guest as "vram" for the virtio-gpu device.
> > This feature currently does not work on systems using the amdgpu driver,
> > as that driver allocates non-compound higher order pages via
> > ttm_pool_alloc_page.
> >
> > First, this series replaces the __gfn_to_pfn_memslot API with a more
> > extensible __kvm_faultin_pfn API. The updated API rearranges
> > __gfn_to_pfn_memslot's args into a struct and where possible packs the
> > bool arguments into a FOLL_ flags argument. The refactoring changes do
> > not change any behavior.
>
> Instead of adding hacks to kvm you really should fix the driver / TTM
> to not do weird memory allocations.
I agree that the driver/TTM behavior is nasty, but from a KVM perspective the vast
majority of this series is long-overdue cleanups (at least, IMO). All of those
cleanups were my requirement for adding support for the behavior David and friends
actually care about.
KVM needs to be aware of non-refcounted struct page memory no matter what; see
CVE-2021-22543 and, commit f8be156be163 ("KVM: do not allow mapping valid but
non-reference-counted pages"). I don't think it makes any sense whatsoever to
remove that code and assume every driver in existence will do the right thing.
With the cleanups done, playing nice with non-refcounted paged instead of outright
rejecting them is a wash in terms of lines of code, complexity, and ongoing
maintenance cost.
