On 05/06/2025 13.56, Thomas Huth wrote:
On 05/06/2025 10.34, Nina Schoetterl-Glausch wrote:
On Tue, 2025-06-03 at 15:56 +0200, Shalini Chellathurai Saroja wrote:
Add Control-Program Identification (CPI) data to the QEMU Object
Model (QOM), along with the timestamp in which the data was received
as shown below.
[...]
Signed-off-by: Shalini Chellathurai Saroja <shal...@linux.ibm.com>
---
hw/s390x/sclpcpi.c | 47 +++++++++++++++++++++++++
include/hw/s390x/event-facility.h | 5 +++
qapi/machine.json | 58 +++++++++++++++++++++++++++++++
3 files changed, 110 insertions(+)
diff --git a/hw/s390x/sclpcpi.c b/hw/s390x/sclpcpi.c
index 935fa87acd..ec711e2291 100644
--- a/hw/s390x/sclpcpi.c
+++ b/hw/s390x/sclpcpi.c
@@ -15,7 +15,9 @@
*/
[...]
+static void cpi_init(Object *obj)
+{
+ SCLPEventCPI *e = SCLP_EVENT_CPI(obj);
+
+ object_property_add_str(obj, "system_type", get_system_type, NULL);
+ object_property_add_str(obj, "system_name", get_system_name, NULL);
+ object_property_add_str(obj, "sysplex_name", get_sysplex_name, NULL);
+ object_property_add_uint64_ptr(obj, "system_level", &(e->system_level),
+ OBJ_PROP_FLAG_READ);
+ object_property_add_uint64_ptr(obj, "timestamp", &(e->timestamp),
+ OBJ_PROP_FLAG_READ);
+}
I think it would be cleaner if those were class properties.
You could use object_class_property_add_str in cpi_class_init,
but I think it'd be nice to use DEFINE_PROP_(STR|UINT64) and
device_class_set_props.
For "normal" properties I'd say "yes" ... but in this case, this would also
allow the user to set the properties from the host side - which would be a
little bit weird? So I think it might be cleaner to keep it this way here
without the "setter" functions? WDYT?
By the way, Shalini, could you please add a description of the properties
via object_property_set_description() here, please?
Thanks,
Thomas
PS: FWIW, Markus recently also provided a nice summery about the different
property types here:
https://lore.kernel.org/qemu-devel/87o6vfzdig....@pond.sub.org/
