30.03.2016 10:50, Pavel Hrdina пишет:
On Tue, Mar 29, 2016 at 07:06:14PM +0300, Maxim Nestratov wrote:
29.03.2016 16:31, Pavel Hrdina пишет:
Commit 7068b56c introduced several hyperv features. Not all hyperv
features are supported by old enough kernels and we shouldn't allow to
start a guest if kernel doesn't support any of the hyperv feature.
There is one exception, for backward compatibility we cannot error out
if one of the RELAXED, VAPIC or SPINLOCKS isn't supported, for the same
reason we ignore invtsc, to not break restoring saved domains with older
libvirt.
[...]
Hmm, qemu already checks them and simply ignores most of them and
doesn't prevent guest from starting in case they are not supported and
optional. In case they are reqired it fails. Why should we check them
here? At least we should follow the logic qemu has.
Yes, that's true that QEMU do some checks and ignores most of the features
missing in kernel, but that's no reason why we should do the same. Libvirt
tries to present in domain XML only those features and devices that are actually
present in the guest. Thus if you tell libvirt that you want some hyperv
feature but your host kernel doesn't support it, we should let the user know
that this feature isn't supported instead of ignoring that fact and start the
guest anyway.
Pavel
I see your point. And what if a user wants to define some features just
to ask libvirt/qemu to do its best if possible and ignore if it isn't
without redefining domain xml? Does such case make sense?
Maxim
--
libvir-list mailing list
libvir-list@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/libvir-list