Re: [PATCH v2 1/1] KVM: arm64: allow the VM to select DEVICE_* and NORMAL_NC for IO memory

[Jason Gunthorpe <jgg@nvidia.com>]

On Wed, Dec 06, 2023 at 06:58:44PM +0000, Catalin Marinas wrote:

> -------------8<----------------------------
> diff --git a/drivers/vfio/pci/vfio_pci_core.c b/drivers/vfio/pci/vfio_pci_core.c
> index 1929103ee59a..b89d2dfcd534 100644
> --- a/drivers/vfio/pci/vfio_pci_core.c
> +++ b/drivers/vfio/pci/vfio_pci_core.c
> @@ -1863,7 +1863,7 @@ int vfio_pci_core_mmap(struct vfio_device *core_vdev, struct vm_area_struct *vma
>  	 * See remap_pfn_range(), called from vfio_pci_fault() but we can't
>  	 * change vm_flags within the fault handler.  Set them now.
>  	 */
> -	vm_flags_set(vma, VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP);
> +	vm_flags_set(vma, VM_VFIO | VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP);
>  	vma->vm_ops = &vfio_pci_mmap_ops;
> 
>  	return 0;
> diff --git a/include/linux/mm.h b/include/linux/mm.h
> index 418d26608ece..6df46fd7836a 100644
> --- a/include/linux/mm.h
> +++ b/include/linux/mm.h
> @@ -391,6 +391,13 @@ extern unsigned int kobjsize(const void *objp);
>  # define VM_UFFD_MINOR		VM_NONE
>  #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
> 
> +#ifdef CONFIG_64BIT
> +#define VM_VFIO_BIT		39
> +#define VM_VFIO			BIT(VM_VFIO_BIT)
> +#else
> +#define VM_VFIO			VM_NONE
> +#endif
> +
>  /* Bits set in the VMA until the stack is in its final location */
>  #define VM_STACK_INCOMPLETE_SETUP (VM_RAND_READ | VM_SEQ_READ | VM_STACK_EARLY)
> -------------8<----------------------------
> 
> In KVM, Akita's patch would take this into account, not just rely on
> "device==true".

Yes, Ankit let's try this please. I would not call it VM_VFIO though

VM_VFIO_ALLOW_WC ?

Introduce it in a separate patch and summarize this thread, with a
suggested-by from Catalin :)

Cc Sean Christopherson <seanjc@google.com> too as x86 kvm might
support the idea

> I think that's a key argument. The VMM cannot, on its own, configure the
> BAR and figure a way to communicate this to the guest. We could invent
> some para-virtualisation/trapping mechanism but that's unnecessarily
> complicated. In the DPDK case, DPDK both configures and interacts with
> the device. In the VMM/VM case, we need the VM to do this, we can't
> split the configuration in VMM and interaction with the device in the
> VM.

Yes

Thanks,
Jason