On 15/07/2020 16.04, Markus Armbruster wrote:
> Watch this:
>
> $ qemu-system-aarch64 -M ast2600-evb -S -display none -qmp stdio
> {"QMP": {"version": {"qemu": {"micro": 50, "minor": 0, "major": 5},
> "package": "v5.0.0-2464-g3a9163af4e"}, "capabilities": ["oob"]}}
> {"execute": "qmp_capabilities"}
> {"return": {}}
> {"execute": "device-list-properties", "arguments": {"typename":
> "msf2-soc"}}
> Unsupported NIC model: ftgmac100
> armbru@dusky:~/work/images$ echo $?
> 1
>
> This is what breaks "make check SPEED=slow".
>
> Root cause is m2sxxx_soc_initfn()'s messing with nd_table[] via
> qemu_check_nic_model(). That's wrong.
>
> We fixed the exact same bug for device "allwinner-a10" in commit
> 8aabc5437b "hw/arm/allwinner-a10: Do not use nd_table in instance_init
> function". Fix this instance the same way: move the offending code to
> m2sxxx_soc_realize(), where it's less wrong, and add a FIXME comment.
>
> Fixes: 05b7374a58cd18aa3516e33513808896d0ac9b7b
> Signed-off-by: Markus Armbruster <arm...@redhat.com>
> ---
> hw/arm/msf2-soc.c | 9 +++++----
> 1 file changed, 5 insertions(+), 4 deletions(-)
>
> diff --git a/hw/arm/msf2-soc.c b/hw/arm/msf2-soc.c
> index 16bb7c9916..33ea7df342 100644
> --- a/hw/arm/msf2-soc.c
> +++ b/hw/arm/msf2-soc.c
> @@ -82,10 +82,6 @@ static void m2sxxx_soc_initfn(Object *obj)
> }
>
> object_initialize_child(obj, "emac", &s->emac, TYPE_MSS_EMAC);
> - if (nd_table[0].used) {
> - qemu_check_nic_model(&nd_table[0], TYPE_MSS_EMAC);
> - qdev_set_nic_properties(DEVICE(&s->emac), &nd_table[0]);
> - }
> }
>
> static void m2sxxx_soc_realize(DeviceState *dev_soc, Error **errp)
> @@ -187,6 +183,11 @@ static void m2sxxx_soc_realize(DeviceState *dev_soc,
> Error **errp)
> g_free(bus_name);
> }
>
> + /* FIXME use qdev NIC properties instead of nd_table[] */
> + if (nd_table[0].used) {
> + qemu_check_nic_model(&nd_table[0], TYPE_MSS_EMAC);
> + qdev_set_nic_properties(DEVICE(&s->emac), &nd_table[0]);
> + }
> dev = DEVICE(&s->emac);
> object_property_set_link(OBJECT(&s->emac), "ahb-bus",
> OBJECT(get_system_memory()), &error_abort);
As long as nobody comes up with a proper clean up within the next days
that moves the nd_table[] access to msf2-som.c, I think this is an
acceptable work-around for 5.1.
Reviewed-by: Thomas Huth <th...@redhat.com>
