On 07/11/2022 10.53, Cédric Le Goater wrote:
Hello,
On 9/2/22 19:27, Matthew Rosato wrote:
The zpcii-disable machine property can be used to force-disable the use
of zPCI interpretation facilities for a VM. By default, this setting
will be off for machine 7.2 and newer.
KVM will tell if the zPCI interpretation feature is available for
the VM depending on the host capabilities and activation can be
handled with the "interpret" property at the device level.
For migration compatibility, zPCI interpretation can be globally
deactivated with a compat property. So, I don't understand how the
"zpcii-disable" machine option is useful.
The patch could possibly be reverted for 7.2 and replaced with :
@@ -921,9 +921,13 @@ static void ccw_machine_7_1_instance_opt
static void ccw_machine_7_1_class_options(MachineClass *mc)
{
S390CcwMachineClass *s390mc = S390_CCW_MACHINE_CLASS(mc);
+ static GlobalProperty compat[] = {
+ { TYPE_S390_PCI_DEVICE, "interpret", "off", },
+ };
ccw_machine_7_2_class_options(mc);
compat_props_add(mc->compat_props, hw_compat_7_1, hw_compat_7_1_len);
+ compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
s390mc->max_threads = S390_MAX_CPUS;
s390mc->topology_capable = false;
}
Thinking about this twice, I'm not sure whether we would need it at all
since zpci cannot be migrated at all (see the "unmigratable = 1" in
hw/s390x/s390-pci-bus.c) ... OTOH, doing it via the compat_props also does
not really hurt.
Anyway, the zpcii-disable switch really does not seem to be necessary...
Matthew, do you think it's ok if we revert "zpcii-disable" patch?
Thomas
