Re: [PATCH 50/55] s390x/event-facility: Simplify creation of SCLP event devices

[Markus Armbruster]

Paolo Bonzini <pbonz...@redhat.com> writes:

> On 26/05/20 08:27, Markus Armbruster wrote:
>>> It cannot happen, because a device must be unparented before it's
>>> destroyed and unparenting takes care of unrealizing the device.
>> 
>> I can't see where unparenting takes care of unrealizing.  Can you help
>> me?
>
> Hidden in plain sight:
>
> static void device_unparent(Object *obj)
> {
>     DeviceState *dev = DEVICE(obj);
>     BusState *bus;
>
>     if (dev->realized) {
>         object_property_set_bool(obj, false, "realized", NULL);
>     }
>     ...
> }
>
> and the call stack is object_unparent -> object_property_del_child ->
> object_finalize_child_property (via prop->release) -> class->unparent.

Aha.

My attempt to trigger automatic unrealize of a child device after the
parent device's realize failed was unsuccessful.  I tried with
pci-serial-2x, hacked up to make it fail as if the second child's
realize failed, and hacked up some more to make it rely on automatic
unrealize.  No dice:

    $ qemu-system-x86_64 -S -nodefaults -display none -monitor stdio
    QEMU 5.0.50 monitor - type 'help' for more information
    (qemu) device_add pci-serial-2x
    ### serial_realize
    Error: mock error
    (qemu) q

Even though it doesn't really matter here, as David pointed out, it's
something I'd like to understand.


diff --git a/hw/char/serial-pci-multi.c b/hw/char/serial-pci-multi.c
index 5f9ccfcc93..433b5caefc 100644
--- a/hw/char/serial-pci-multi.c
+++ b/hw/char/serial-pci-multi.c
@@ -56,7 +56,7 @@ static void multi_serial_pci_exit(PCIDevice *dev)
 
     for (i = 0; i < pci->ports; i++) {
         s = pci->state + i;
-        object_property_set_bool(OBJECT(s), false, "realized", &error_abort);
+//        object_property_set_bool(OBJECT(s), false, "realized", &error_abort);
         memory_region_del_subregion(&pci->iobar, &s->io);
         g_free(pci->name[i]);
     }
@@ -106,6 +106,9 @@ static void multi_serial_pci_realize(PCIDevice *dev, Error 
**errp)
 
     for (i = 0; i < nports; i++) {
         s = pci->state + i;
+        if (i)
+            error_setg(errp, "mock error");
+        else
         object_property_set_bool(OBJECT(s), true, "realized", &err);
         if (err != NULL) {
             error_propagate(errp, err);
diff --git a/hw/char/serial.c b/hw/char/serial.c
index 7d74694587..55b0bbd8b0 100644
--- a/hw/char/serial.c
+++ b/hw/char/serial.c
@@ -938,6 +938,8 @@ static void serial_realize(DeviceState *dev, Error **errp)
 {
     SerialState *s = SERIAL(dev);
 
+    printf("### %s\n", __func__);
+
     s->modem_status_poll = timer_new_ns(QEMU_CLOCK_VIRTUAL, (QEMUTimerCB *) 
serial_update_msl, s);
 
     s->fifo_timeout_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, (QEMUTimerCB *) 
fifo_timeout_int, s);
@@ -954,6 +956,8 @@ static void serial_unrealize(DeviceState *dev)
 {
     SerialState *s = SERIAL(dev);
 
+    printf("### %s\n", __func__);
+
     qemu_chr_fe_deinit(&s->chr, false);
 
     timer_del(s->modem_status_poll);