On Thu, 2021-01-14 at 18:11 +0200, Sakari Ailus wrote:

On Thu, Jan 14, 2021 at 11:46:11AM -0300, Ezequiel Garcia wrote:

quoted

On Thu, 2021-01-14 at 15:47 +0200, Sakari Ailus wrote:

quoted

Hi Laurent, Ezequiel,

On Thu, Jan 14, 2021 at 03:59:10AM +0200, Laurent Pinchart wrote:

quoted

quoted

diff --git a/drivers/media/platform/rockchip/rkisp1/rkisp1-dev.c b/drivers/media/platform/rockchip/rkisp1/rkisp1-dev.c
index 68da1eed753d..235dcf0c4122 100644
--- a/drivers/media/platform/rockchip/rkisp1/rkisp1-dev.c
+++ b/drivers/media/platform/rockchip/rkisp1/rkisp1-dev.c

@@ -252,6 +252,7 @@ static int rkisp1_subdev_notifier(struct rkisp1_device *rkisp1)

                        .bus_type = V4L2_MBUS_CSI2_DPHY
                };
                struct rkisp1_sensor_async *rk_asd = NULL;
+               struct v4l2_async_subdev *asd;
                struct fwnode_handle *ep;
 
                ep = fwnode_graph_get_endpoint_by_id(dev_fwnode(rkisp1->dev),

@@ -264,21 +265,16 @@ static int rkisp1_subdev_notifier(struct rkisp1_device *rkisp1)

                if (ret)
                        goto err_parse;
 
-               rk_asd = kzalloc(sizeof(*rk_asd), GFP_KERNEL);
-               if (!rk_asd) {
-                       ret = -ENOMEM;
+               asd = v4l2_async_notifier_add_fwnode_remote_subdev(ntf, ep,
+                                                       sizeof(*rk_asd));
+               if (IS_ERR(asd))

The problem with registering the sub-device already here is that the driver
can proceed to use the information in the async sub-device object which is
initialised below.

Note that this interface is not really registering sub-devices.

Not directly, but this will happen as a by-product of registering the async
sub-device and other functions that will be called. All this takes place
synchronously, meaming that by the time this function returns, the
character devices that are the user space interface have already been
created.

That's not the case, as I've explained before, v4l2_async_notifier_add_fwnode_remote_subdev
is _not_ about registering any actual v4l2 subdevice/char device.

It's just about adding the v4l2 async subdevice descriptor
to a given (unregistered) notifier.

Thanks,
Ezequiel