ppc/spapr: cleanup -machine pseries,nvdimm=X handling

Since NVDIMM support was introduced on pseries machine,
it ignored machine's nvdimm=on|off option and effectively
was always enabled on machines that support NVDIMM.
Later on commit
  (28f5a716212 ppc/spapr_nvdimm: do not enable support with 'nvdimm=off')
makes QEMU error out in case user explicitly set 'nvdimm=off'
on CLI by peeking at machine_opts.

However that's a workaround and leaves 'nvdimms_state->is_enabled'
in inconsistent state (false) when it should be set true
by default.

Instead of using on machine_opts, set default to true for pseries
machine in initfn time. If user sets manually 'nvdimm=off'
it will overwrite default value to false and QEMU will error
as expected without need to peek into machine_opts.

That way pseries will have, nvdimm enabled by default and
will honor user provided 'nvdimm=on|off'.

Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Message-Id: <20201208164606.4109134-1-imammedo@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

1 files changed, 13 insertions, 0 deletions

diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
index dee48a0..489cefc 100644
--- a/hw/ppc/spapr.c
+++ b/hw/ppc/spapr.c
@@ -3275,6 +3275,19 @@ static void spapr_instance_init(Object *obj)
 {
     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
+    MachineState *ms = MACHINE(spapr);
+    MachineClass *mc = MACHINE_GET_CLASS(ms);
+
+    /*
+     * NVDIMM support went live in 5.1 without considering that, in
+     * other archs, the user needs to enable NVDIMM support with the
+     * 'nvdimm' machine option and the default behavior is NVDIMM
+     * support disabled. It is too late to roll back to the standard
+     * behavior without breaking 5.1 guests.
+     */
+    if (mc->nvdimm_supported) {
+        ms->nvdimms_state->is_enabled = true;
+    }
 
     spapr->htab_fd = -1;
     spapr->use_hotplug_event_source = true;