Il 21/08/2013 19:35, Andreas Färber ha scritto:
Am 21.08.2013 19:01, schrieb Michael S. Tsirkin:
On Wed, Aug 21, 2013 at 06:43:16PM +0200, Paolo Bonzini wrote:
The pvpanic situation is already messed up enough. Let us give our libvirt friends an easy indication that we have untied our side.
Not-yet-signed-off-by: Paolo Bonzini <pbonzini@redhat.com> ... because we first have to determine how to expose the device's existence in the ACPI tables or in fw_cfg.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
So it's isa-pvpanic meaning "I-am-sure-this-is-the-last-bug-pvpanic"
If we feel there's need to give libvirt a way to do introspection into QEMU bugs, let's architect one. Randomly renaming devices in the vain hope it's the last major bug is not it.
NACK
Seconded. While we shouldn't rule out renaming devices, doing so as a criteria for libvirt sounds utterly wrong.
Paolo, you are right that a single "NACK" cannot be a criteria, but a single convincing justification by a random reviewer should be sufficient to reconsider. :)
Adding some device property or obtaining the info via some existing query-* QMP command might be better alternatives.
Device properties are a good way to communicate changes to the guest, but here the guest ABI is unchanged. Anyhow, this patch is not strictly necessary. It's just a safety net to avoid that libvirt uses a buggy device on buggy versions of QEMU. Paolo