On Wed, Jun 29, 2016 at 08:00:22AM -0600, Alex Williamson wrote:
On Wed, 29 Jun 2016 09:47:55 +0100 "Daniel P. Berrange" <berrange@redhat.com> wrote:
On Tue, Jun 28, 2016 at 02:30:47PM -0400, Bandan Das wrote:
"Daniel P. Berrange" <berrange@redhat.com> writes:
Adding Alex & Bandan, since they signed off the kernel patch which I'm thinking either pci-back should be made to work more like vfio, or the kernel patch should be reverted or fixed to take account of the way pci-back works.
Whichever way, I don't consider this a libvirt problem to solve. As Linus' always says - the kernel must never break existing userspace
Agreed, but in this specific case, the usage is unsafe since unknown indexes are potentially being passed to the driver operations. It should always have been 3. to begin with.
Whether the userspace usage is good or not is irrelevant - this kernel change has broken existing userspace apps and that is not acceptable and must be fixed.
I'm fine with suggestions to change future libvirt to work in a better way, but we need to fix the regressions seen by *current* libvirt releases
I don't think this is a reasonable demand. For one, the change was made 2yrs ago and nobody noticed until now, I don't think there are stable kernel releases to cover all those kernels. There must be some sort of statute of limitations.
Oh sorry, I totally missed the date on that. I was thinking this was a recent kernel regression. I agree that if 2+ years have passed, this ship has sailed. Regards, Daniel -- |: http://berrange.com -o- http://www.flickr.com/photos/dberrange/ :| |: http://libvirt.org -o- http://virt-manager.org :| |: http://autobuild.org -o- http://search.cpan.org/~danberr/ :| |: http://entangle-photo.org -o- http://live.gnome.org/gtk-vnc :|