In commit 8461bfdca9c we added the TYPE_AW_I2C_SUN6I, which is a
minor variant of the TYPE_AW_I2C device. However, we didn't quite
get the class hierarchy right. We made the new TYPE_AW_I2C_SUN6I a
subclass of TYPE_SYS_BUS_DEVICE, which means that you can't validly
use a pointer to this object via the AW_I2C() cast macro, which
insists on having something that is an instance of TYPE_AW_I2C or
some subclass of that type.
This only causes a problem if QOM cast macro debugging is enabled;
that is supposed to be on by default, but a mistake in the meson
conversion in commit c55cf6ab03f4c meant that it ended up disabled by
default, and we didn't catch this bug.
Fix the problem by arranging the classes in the same way we do for
TYPE_PL011 and TYPE_PL011_LUMINARY in hw/char/pl011.c -- make the
variant class be a subclass of the "normal" version of the device.
This was reported in
https://gitlab.com/qemu-project/qemu/-/issues/1586 but this fix alone
isn't sufficient, as there is a separate cast-related issue in the
CXL code in pci_expander_bridge.c.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
---
hw/i2c/allwinner-i2c.c | 4 +---
1 file changed, 1 insertion(+), 3 deletions(-)
diff --git a/hw/i2c/allwinner-i2c.c b/hw/i2c/allwinner-i2c.c
index f24c3ac6f0c..9e8efa1d63f 100644
--- a/hw/i2c/allwinner-i2c.c
+++ b/hw/i2c/allwinner-i2c.c
@@ -466,10 +466,8 @@ static void allwinner_i2c_sun6i_init(Object *obj)
static const TypeInfo allwinner_i2c_sun6i_type_info = {
.name = TYPE_AW_I2C_SUN6I,
- .parent = TYPE_SYS_BUS_DEVICE,
- .instance_size = sizeof(AWI2CState),
+ .parent = TYPE_AW_I2C,
.instance_init = allwinner_i2c_sun6i_init,
- .class_init = allwinner_i2c_class_init,
};
static void allwinner_i2c_register_types(void)