On Tue, Feb 15, 2022 at 10:16:40AM +0100, Eric Auger wrote: > Hi Connie, > > On 2/14/22 6:34 PM, Cornelia Huck wrote: > > On Mon, Feb 14 2022, Jean-Philippe Brucker <jean-phili...@linaro.org> wrote: > > > >> Replace the VIRTIO_IOMMU_F_BYPASS feature with > >> VIRTIO_IOMMU_F_BYPASS_CONFIG, which enables a config space bit to switch > >> global bypass on and off. > >> > >> Add a boot-bypass option, which defaults to 'on' to be in line with > >> other vIOMMUs and to allow running firmware/bootloader that are unaware > >> of the IOMMU. x86 doesn't need a workaround to boot with virtio-iommu > >> anymore. > >> > >> Since v2 [1]: > >> * Added the new bypass bits to the migration stream. > >> As discussed on the v2 thread, we assume that cross-version > >> compatibility is not required for live migration at the moment, so we > >> only increase the version number. Patch 2 says: "We add the bypass > >> field to the migration stream without introducing subsections, based > >> on the assumption that this virtio-iommu device isn't being used in > >> production enough to require cross-version migration at the moment > >> (all previous version required workarounds since they didn't support > >> ACPI and boot-bypass)." > >> > >> [1] > >> https://lore.kernel.org/qemu-devel/20220127142940.671333-1-jean-phili...@linaro.org/ > > One thing that we could do to avoid surprises in the unlikely case that > > somebody has a virtio-iommu device and wants to migrate to an older > > machine version is to add a migration blocker for the virtio-iommu > > device for all compat machines for versions 6.2 or older (i.e. only 7.0 > > or newer machine types can have a migratable virtio-iommu device > > starting with QEMU 7.0.) Not too complicated to implement, but I'm not > > sure whether we'd add too much code to prevent something very unlikely > > to happen anyway. I would not insist on it :) > As nobody has shout and we are not aware of anybody using the device in > production mode yet due to the missing boot bypass feature this series > brings, I would be personally in favour of leaving things as is. Now, up > to Jean if he wants to go and implement your suggestion.
I agree, it seems too unlikely that someone would want to migrate it back to 6.2 where it wasn't really useable except for experiments Thanks, Jean