Re: [PATCH v5 0/5] Disable Deprecated Features by Default on s390 CPU Models

On Thu, Jul 24, 2025 at 14:39:02 -0400, Collin Walling wrote: > On 7/24/25 03:35, Thomas Huth via Devel wrote: >> On 30/06/2025 05.19, Collin Walling wrote: >>> Changelog >>> >>> v5 >>> - dropped the "none" test in qemuxmlactivetest (see commit for >>> details) >>> - reordered patches to introduce some tests first, then add >>> qemu.conf changes >>> >>> v4 >>> - added qemu.conf option to dictate the default behavior for the >>> deprecated_features attribute (Boris) >>> - added qemuxmlactivetests (Boris) >>> - snuck in missing documentation for deprecated_features in >>> formatdomain.rst >>> >>> v3 >>> - added qemu caps check to avoid breaking s390 guests trying to >>> default deprecated_features='off' on QEMU versions that >>> do not support reporting these features >>> >>> v2 >>> - changed behavior from disabling features on the host model to >>> instead flagging the guest CPU to disable deprecated features >>> - removed disabling deprecated features on host model in >>> virQEMUCapsInitCPUModelS390 >>> - added flagging deprecated_feats in qemuProcessUpdateGuestCPU >>> - added tests for deprecated_features='on' >>> - split virQEMUCapsUpdateCPUDeprecatedFeatures update and >>> qemuProcessUpdateGuestCPU changes >>> >>> The intention of reporting deprecated features and modifying the guest >>> CPU model was to alleviate the user from the burden of preparing a guest >>> with the necessary amendments to assure migration to newer hardware. >>> While that goal was met by way of the "deprecated_features='on|off'" >>> attribute, it still adds an extra step that the user must be aware to >>> prepare a guest for migration and the errors that stem from an >>> unsuccessful migration (due to feature incompatibility) is not always >>> clear how to resolve. >>> >>> These patches make s390 CPU *host models* migration ready from the get-go >>> by introducing a qemu.conf option for disabling deprecated features by >>> default. They may still be disabled for other model types via the >>> respective attribute, or reenabled if desired. The configured behavior >>> may be overridden by explicitly providing the attribute within the >>> guest XML. >> >> The patch series sounds reasonable to me, so FWIW: >> >> Series >> Acked-by: Thomas Huth <thuth(a)redhat.com&gt; >> >> Peter, Michal, could we still get this merged for libvirt 11.6 ? >> >> Thomas > > Thanks for the Ack and helping to move this along. > > I've posted a patch to the list for NEWS mentioning this features in > case this series is accepted in time for 11.6. Both this series and the NEWS patch are pushed now. Jirka