On Mon, Sep 13, 2021 at 06:47:46AM -0400, Michael S. Tsirkin wrote:
> A recent change checking of_device_is_compatible on probe broke some
> powerpc/pseries setups. Apparently there virtio devices do not have a
> "compatible" property - they are matched by PCI vendor/device ids.
>
> Let's just skip of_node setup but proceed with initialization like we
> did previously.
>
> Fixes: 694a1116b405 ("virtio: Bind virtio device to device-tree node")
> Reported-by: Alexey Kardashevskiy <aik@xxxxxxxxx>
> Cc: Arnd Bergmann <arnd@xxxxxxxx>
> Cc: Viresh Kumar <viresh.kumar@xxxxxxxxxx>
> Signed-off-by: Michael S. Tsirkin <mst@xxxxxxxxxx>
> ---
This patch also fixes a similar problem seen with sparc64.
Tested-by: Guenter Roeck <linux@xxxxxxxxxxxx>
Guenter
>
> Arnd could you help review this pls? Viresh is on vacation.
>
> drivers/virtio/virtio.c | 7 ++++++-
> 1 file changed, 6 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/virtio/virtio.c b/drivers/virtio/virtio.c
> index c46cc1fbc7ae..19a70a2361b4 100644
> --- a/drivers/virtio/virtio.c
> +++ b/drivers/virtio/virtio.c
> @@ -347,8 +347,13 @@ static int virtio_device_of_init(struct virtio_device *dev)
> ret = snprintf(compat, sizeof(compat), "virtio,device%x", dev->id.device);
> BUG_ON(ret >= sizeof(compat));
>
> + /*
> + * On powerpc/pseries virtio devices are PCI devices so PCI
> + * vendor/device ids play the role of the "compatible" property.
> + * Simply don't init of_node in this case.
> + */
> if (!of_device_is_compatible(np, compat)) {
> - ret = -EINVAL;
> + ret = 0;
> goto out;
> }
>
> --
> MST
>
