Skip to contentSkip to conversation

vmbus: Don't make QOM property registration conditional

PATCH In review Eduardo Habkost 2 participants 6 messages Oct 9, 2020 – Apr 27, 2021

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 vmbus: Don't make QOM property registration conditional

Having properties registered conditionally makes QOM type introspection difficult. Instead of skipping registration of the "instanceid" property, always register the property but validate its value against the instance id required by the class.

---
Note: due to the lack of concrete vmbus-dev subclasses in the QEMU tree, this patch couldn't be tested.

---
 hw/hyperv/vmbus.c | 20 +++++++++++++++-----
 1 file changed, 15 insertions(+), 5 deletions(-)
hw/hyperv/vmbus.c +15 -5
index 896e981f85..481d7e8d3c 100644
@@ -2371,6 +2371,14 @@ static void vmbus_dev_realize(DeviceState *dev, Error **errp)
 
     assert(!qemu_uuid_is_null(&vdev->instanceid));
 
+    if (!qemu_uuid_is_null(&vdc->instanceid)) {
+        /* Class wants to only have a single instance with a fixed UUID */
+        if (!qemu_uuid_is_equal(&vdev->instanceid, &vdc->instanceid)) {
+            error_setg(&err, "instance id can't be changed");
+            goto error_out;
+        }
+    }
+
     /* Check for instance id collision for this class id */
     QTAILQ_FOREACH(child, &BUS(vmbus)->children, sibling) {
         VMBusDevice *child_dev = VMBUS_DEVICE(child->child);
@@ -2437,18 +2445,22 @@ static void vmbus_dev_unrealize(DeviceState *dev)
     free_channels(vdev);
 }
 
+static Property vmbus_dev_props[] = {
+    DEFINE_PROP_UUID("instanceid", VMBusDevice, instanceid),
+    DEFINE_PROP_END_OF_LIST()
+};
+
+
 static void vmbus_dev_class_init(ObjectClass *klass, void *data)
 {
     DeviceClass *kdev = DEVICE_CLASS(klass);
+    device_class_set_props(kdev, vmbus_dev_props);
     kdev->bus_type = TYPE_VMBUS;
     kdev->realize = vmbus_dev_realize;
     kdev->unrealize = vmbus_dev_unrealize;
     kdev->reset = vmbus_dev_reset;
 }
 
-static Property vmbus_dev_instanceid =
-                        DEFINE_PROP_UUID("instanceid", VMBusDevice, instanceid);
-
 static void vmbus_dev_instance_init(Object *obj)
 {
     VMBusDevice *vdev = VMBUS_DEVICE(obj);
@@ -2457,8 +2469,6 @@ static void vmbus_dev_instance_init(Object *obj)
     if (!qemu_uuid_is_null(&vdc->instanceid)) {
         /* Class wants to only have a single instance with a fixed UUID */
         vdev->instanceid = vdc->instanceid;
-    } else {
-        qdev_property_add_static(DEVICE(vdev), &vmbus_dev_instanceid);
     }
 }
 
Signature · 1 line
2.26.2
#1 lore raw
Quoted 9 lines · Eduardo Habkost

Having properties registered conditionally makes QOM type introspection difficult. Instead of skipping registration of the "instanceid" property, always register the property but validate its value against the instance id required by the class.

---
Note: due to the lack of concrete vmbus-dev subclasses in the QEMU tree, this patch couldn't be tested.

Will test it tomorrow since I have a VMBus device implementation.

Maciej

#2 lore raw
Quoted 6 lines · Maciej S. Szmigiero
Quoted 5 lines · Eduardo Habkost

Having properties registered conditionally makes QOM type introspection difficult. Instead of skipping registration of the "instanceid" property, always register the property but validate its value against the instance id required by the class.

Quoted 6 lines
Quoted 4 lines

---
Note: due to the lack of concrete vmbus-dev subclasses in the QEMU tree, this patch couldn't be tested.

Will test it tomorrow since I have a VMBus device implementation.

Thanks!

Is anybody willing to volunteer to be listed as maintainer (or reviewer, at least) of hw/hyperv/vmbus.c?

Signature · 1 line
Eduardo
#3 lore raw
Quoted 7 lines · Eduardo Habkost
Quoted 6 lines · Maciej S. Szmigiero
Quoted 5 lines · Eduardo Habkost

Having properties registered conditionally makes QOM type introspection difficult. Instead of skipping registration of the "instanceid" property, always register the property but validate its value against the instance id required by the class.

Quoted 9 lines
Quoted 6 lines
Quoted 4 lines

---
Note: due to the lack of concrete vmbus-dev subclasses in the QEMU tree, this patch couldn't be tested.

Will test it tomorrow since I have a VMBus device implementation.

Thanks!

Tested the patch with a hv-balloon device and is seems to work okay, so:

Maciej

#4 lore raw
Quoted 13 lines · Maciej S. Szmigiero
Quoted 12 lines · Eduardo Habkost
Quoted 11 lines · Maciej S. Szmigiero
Quoted 9 lines · Eduardo Habkost

Having properties registered conditionally makes QOM type introspection difficult. Instead of skipping registration of the "instanceid" property, always register the property but validate its value against the instance id required by the class.

---
Note: due to the lack of concrete vmbus-dev subclasses in the QEMU tree, this patch couldn't be tested.

Quoted 8 lines
Quoted 4 lines
Quoted 1 line

Will test it tomorrow since I have a VMBus device implementation.

Thanks!

Tested the patch with a hv-balloon device and is seems to work okay, so:

I see this patch wasn't picked up - it still makes sense and applies cleanly to the current git, so I think it should be picked up.

Thanks,
Maciej

#5 lore raw
Quoted 19 lines · Maciej S. Szmigiero
Quoted 18 lines · Maciej S. Szmigiero
Quoted 16 lines · Eduardo Habkost
Quoted 6 lines · Maciej S. Szmigiero
Quoted 5 lines · Eduardo Habkost

Having properties registered conditionally makes QOM type introspection difficult. Instead of skipping registration of the "instanceid" property, always register the property but validate its value against the instance id required by the class.

Quoted 6 lines
Quoted 4 lines

---
Note: due to the lack of concrete vmbus-dev subclasses in the QEMU tree, this patch couldn't be tested.

Will test it tomorrow since I have a VMBus device implementation.

Thanks!

Quoted 6 lines
Quoted 3 lines

Tested the patch with a hv-balloon device and is seems to work okay, so:

I see this patch wasn't picked up - it still makes sense and applies cleanly to the current git, so I think it should be picked up.

I'm queueing it for 6.1, thanks for the reminder!

Signature · 1 line
Eduardo