Re: [PATCH v8 8/9] hisi_acc_vfio_pci: Add support for VFIO live migration

[Cornelia Huck <cohuck@redhat.com>]

On Thu, Mar 10 2022, Alex Williamson <alex.williamson@redhat.com> wrote:

> Do you think we should go so far as to formalize this via a MAINTAINERS
> entry, for example:
>
> diff --git a/Documentation/vfio/vfio-pci-vendor-driver-acceptance.rst b/Documentation/vfio/vfio-pci-vendor-driver-acceptance.rst
> new file mode 100644
> index 000000000000..54ebafcdd735
> --- /dev/null
> +++ b/Documentation/vfio/vfio-pci-vendor-driver-acceptance.rst
> @@ -0,0 +1,35 @@
> +.. SPDX-License-Identifier: GPL-2.0
> +
> +Acceptance criteria for vfio-pci device specific driver variants
> +================================================================
> +
> +Overview
> +--------
> +The vfio-pci driver exists as a device agnostic driver using the
> +system IOMMU and relying on the robustness of platform fault
> +handling to provide isolated device access to userspace.  While the
> +vfio-pci driver does include some device specific support, further
> +extensions for yet more advanced device specific features are not
> +sustainable.  The vfio-pci driver has therefore split out
> +vfio-pci-core as a library that may be reused to implement features
> +requiring device specific knowledge, ex. saving and loading device
> +state for the purposes of supporting migration.
> +
> +In support of such features, it's expected that some device specific
> +variants may interact with parent devices (ex. SR-IOV PF in support of
> +a user assigned VF) or other extensions that may not be otherwise
> +accessible via the vfio-pci base driver.  Authors of such drivers
> +should be diligent not to create exploitable interfaces via such
> +interactions or allow unchecked userspace data to have an effect
> +beyond the scope of the assigned device.
> +
> +New driver submissions are therefore requested to have approval via
> +Sign-off for any interactions with parent drivers.  Additionally,
> +drivers should make an attempt to provide sufficient documentation
> +for reviewers to understand the device specific extensions, for
> +example in the case of migration data, how is the device state
> +composed and consumed, which portions are not otherwise available to
> +the user via vfio-pci, what safeguards exist to validate the data,
> +etc.  To that extent, authors should additionally expect to require
> +reviews from at least one of the listed reviewers, in addition to the
> +overall vfio maintainer.
> diff --git a/MAINTAINERS b/MAINTAINERS
> index 4322b5321891..4f7d26f9aac6 100644
> --- a/MAINTAINERS
> +++ b/MAINTAINERS
> @@ -20314,6 +20314,13 @@ F:	drivers/vfio/mdev/
>  F:	include/linux/mdev.h
>  F:	samples/vfio-mdev/
>  
> +VFIO PCI VENDOR DRIVERS
> +R:	Your Name <your.name@here.com>
> +L:	kvm@vger.kernel.org
> +S:	Maintained
> +P:	Documentation/vfio/vfio-pci-vendor-driver-acceptance.rst
> +F:	drivers/vfio/pci/*/

This works as long as the only subdirectories are for vendor drivers;
should something else come up, we'd need to add an exclude statement, so
no biggie.

> +
>  VFIO PLATFORM DRIVER
>  M:	Eric Auger <eric.auger@redhat.com>
>  L:	kvm@vger.kernel.org
>
> Ideally we'd have at least Yishai, Shameer, Jason, and yourself listed
> as reviewers (Connie and I are included via the higher level entry).
> Thoughts from anyone?  Volunteers for reviewers if we want to press
> forward with this as formal acceptance criteria?  Thanks,
>
> Alex

I like having this formalized. More eyeballs are good (especially as
getting good review is one of the worst bottlenecks), and I'd trust
people having worked on other vendor drivers having a better grip on
issues that have not been my priority.