Re: [RFC PATCH 2/3] vfio/hisilicon: register the driver to vfio

[Jason Gunthorpe]

On Tue, Apr 13, 2021 at 11:36:22AM +0800, Longfang Liu wrote:
> Register the live migration driver of the accelerator module to vfio
> 
> Signed-off-by: Longfang Liu <liulongfang@xxxxxxxxxx>
>  drivers/vfio/pci/vfio_pci.c         | 11 +++++++++++
>  drivers/vfio/pci/vfio_pci_private.h |  9 +++++++++
>  2 files changed, 20 insertions(+)
> 
> diff --git a/drivers/vfio/pci/vfio_pci.c b/drivers/vfio/pci/vfio_pci.c
> index 65e7e6b..e1b0e37 100644
> +++ b/drivers/vfio/pci/vfio_pci.c
> @@ -407,6 +407,17 @@ static int vfio_pci_enable(struct vfio_pci_device *vdev)
>  		}
>  	}
>  
> +	if (pdev->vendor == PCI_VENDOR_ID_HUAWEI &&
> +	    IS_ENABLED(CONFIG_VFIO_PCI_HISI_MIGRATION)) {
> +		ret = vfio_pci_hisilicon_acc_init(vdev);

This is exactly what we want to avoid with the work we are doing on
the vfio-pci modularity.

It is a complete mess to just keep adding more stuff here, and as
we've discussed to death really ugly to have such a ridiculously wide
ID match.

You should be working with us on that project and base your work on
top of Max's series.. Alex given the interest here I think we should
find some way forward while we work on completed version of the mlx5
pci vfio driver..

I'm also confused how this works securely at all, as a general rule a
VFIO PCI driver cannot access the MMIO memory of the function it is
planning to assign to the guest. There is a lot of danger that the
guest could access that MMIO space one way or another.

Here I see the driver obtaining a devm_ioremap() of the same pdev it
is going to assign (and I really wonder why pci_release_mem_regions()
exists at all..)

This is why the mlx5 RFC was showing how to juggle two PCI devices via
the aux device connector.

Jason