[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <ZAokiR4EC3gFAuJ1@ziepe.ca>
Date: Thu, 9 Mar 2023 14:25:13 -0400
From: Jason Gunthorpe <jgg@...pe.ca>
To: Dominik Behr <dbehr@...omium.org>
Cc: Alex Williamson <alex.williamson@...hat.com>,
Grzegorz Jaszczyk <jaz@...ihalf.com>,
linux-kernel@...r.kernel.org, dmy@...ihalf.com, tn@...ihalf.com,
upstream@...ihalf.com, dtor@...gle.com, kevin.tian@...el.com,
cohuck@...hat.com, abhsahu@...dia.com, yishaih@...dia.com,
yi.l.liu@...el.com, kvm@...r.kernel.org, libvir-list@...hat.com
Subject: Re: [PATCH] vfio/pci: Propagate ACPI notifications to the user-space
On Wed, Mar 08, 2023 at 05:51:32PM -0800, Dominik Behr wrote:
> All other ACPI events that are available to userspace are on netlink already.
> As for translation of ACPI paths. It is sort of a requirement for VMM
> to translate the PCI path from host to guest because the PCI device
> tree in the guest is totally different already. The same follows for
> ACPI paths.
>
> What would you propose instead of netlink?
> Sysfs entry for VFIO PCI device that accepts eventfd and signals the
> events via eventfd? Or moving it into ACPI layer entirely and adding
> eventfd sysfs interface for all ACPI devices?
I think Alex is asking why wouldn't you push it through the vfio
device FD? There is an unambiguous relationship between the QEMU vPCI
identity and the VFIO device, and we already have a good security
model for VMM access to the device FD.
Jason
Powered by blists - more mailing lists