Re: [PATCH v2 2/2] media: staging: rkisp1: replace the call to v4l2_async_notifier_parse_fwnode_endpoints_by_port

[Helen Koike <helen.koike@collabora.com>]

Hi Sakari,

On 3/13/20 6:18 AM, Sakari Ailus wrote:
> Hi Dafna,
> 
> Thanks for the patch.
> 
> On Thu, Mar 12, 2020 at 04:46:04PM +0100, Dafna Hirschfeld wrote:
>> don't call 'v4l2_async_notifier_parse_fwnode_endpoints_by_port'
>> in order to register async subdevices. Instead call
>> 'v4l2_fwnode_endpoint_parse' to parse the remote endpoints
>> and then register each async subdev with
>> 'v4l2_async_notifier_add_fwnode_remote_subdev'
>>
>> Also remove the relevant item in the TODO file
>>
>> Signed-off-by: Dafna Hirschfeld <dafna.hirschfeld@collabora.com>
>> ---
>>  drivers/staging/media/rkisp1/TODO         |  3 -
>>  drivers/staging/media/rkisp1/rkisp1-dev.c | 94 +++++++++++++----------
>>  2 files changed, 55 insertions(+), 42 deletions(-)
>>
>> diff --git a/drivers/staging/media/rkisp1/TODO b/drivers/staging/media/rkisp1/TODO
>> index 0aa9877dd64a..1aa3bb9fd6cb 100644
>> --- a/drivers/staging/media/rkisp1/TODO
>> +++ b/drivers/staging/media/rkisp1/TODO
>> @@ -1,6 +1,3 @@
>> -* Don't use v4l2_async_notifier_parse_fwnode_endpoints_by_port().
>> -e.g. isp_parse_of_endpoints in drivers/media/platform/omap3isp/isp.c
>> -cio2_parse_firmware in drivers/media/pci/intel/ipu3/ipu3-cio2.c.
>>  * Fix pad format size for statistics and parameters entities.
>>  * Use threaded interrupt for rkisp1_stats_isr(), remove work queue.
>>  * Fix checkpatch errors.
>> diff --git a/drivers/staging/media/rkisp1/rkisp1-dev.c b/drivers/staging/media/rkisp1/rkisp1-dev.c
>> index d2186856bb24..1035a39f3e49 100644
>> --- a/drivers/staging/media/rkisp1/rkisp1-dev.c
>> +++ b/drivers/staging/media/rkisp1/rkisp1-dev.c
>> @@ -233,35 +233,6 @@ static int rkisp1_subdev_notifier_complete(struct v4l2_async_notifier *notifier)
>>  	return 0;
>>  }
>>  
>> -static int rkisp1_fwnode_parse(struct device *dev,
>> -			       struct v4l2_fwnode_endpoint *vep,
>> -			       struct v4l2_async_subdev *asd)
>> -{
>> -	struct rkisp1_sensor_async *s_asd =
>> -			container_of(asd, struct rkisp1_sensor_async, asd);
>> -
>> -	if (vep->bus_type != V4L2_MBUS_CSI2_DPHY) {
>> -		dev_err(dev, "Only CSI2 bus type is currently supported\n");
>> -		return -EINVAL;
>> -	}
>> -
>> -	if (vep->base.port != 0) {
>> -		dev_err(dev, "The ISP has only port 0\n");
>> -		return -EINVAL;
>> -	}
>> -
>> -	s_asd->mbus_type = vep->bus_type;
>> -	s_asd->lanes = vep->bus.mipi_csi2.num_data_lanes;
>> -
>> -	/* Parallel bus is currently not supported */
>> -	s_asd->parallel_bus_flags = 0;
>> -
>> -	if (s_asd->lanes < 1 || s_asd->lanes > 4)
>> -		return -EINVAL;
>> -
>> -	return 0;
>> -}
>> -
>>  static const struct v4l2_async_notifier_operations rkisp1_subdev_notifier_ops = {
>>  	.bound = rkisp1_subdev_notifier_bound,
>>  	.unbind = rkisp1_subdev_notifier_unbind,
>> @@ -271,23 +242,68 @@ static const struct v4l2_async_notifier_operations rkisp1_subdev_notifier_ops =
>>  static int rkisp1_subdev_notifier(struct rkisp1_device *rkisp1)
>>  {
>>  	struct v4l2_async_notifier *ntf = &rkisp1->notifier;
>> -	struct device *dev = rkisp1->dev;
>> +	int next_id = 0;
>>  	int ret;
>>  
>>  	v4l2_async_notifier_init(ntf);
>>  
>> -	ret = v4l2_async_notifier_parse_fwnode_endpoints_by_port(dev, ntf,
>> -					sizeof(struct rkisp1_sensor_async),
>> -					0, rkisp1_fwnode_parse);
>> -	if (ret)
>> -		return ret;
>> +	while (1) {
> 
> I might loop over each port here instead.

ISP has a single port (please, see my comment below).

> 
>> +		struct v4l2_fwnode_endpoint vep = {
>> +			.bus_type = V4L2_MBUS_CSI2_DPHY
>> +		};
>> +		struct rkisp1_sensor_async *rk_asd = NULL;
>> +		struct fwnode_handle *ep;
>>  
>> -	if (list_empty(&ntf->asd_list))
>> -		return -ENODEV;
>> +		ep = fwnode_graph_get_endpoint_by_id(dev_fwnode(rkisp1->dev),
>> +			0, next_id, FWNODE_GRAPH_ENDPOINT_NEXT);
> 
> The port number is always zero, whereas the endpoint id changes on each
> iteration. Is that intended?

Yes, so ISP has a single connection port (a single MIPI-DPHY bus), but hardware can plug more then one
sensor in this port (but only one can be active at a time).

At least this is how I understand how the modeling should be.
And this is how we modeled the device tree bindings:
https://git.linuxtv.org/media_tree.git/tree/drivers/staging/media/rkisp1/Documentation/devicetree/bindings/media/rockchip-isp1.yaml#n139

Make sense?

Thanks for reviewing this,
Helen

> 
>>  
>> -	ntf->ops = &rkisp1_subdev_notifier_ops;
>> +		if (!ep)
>> +			break;
>> +
>> +		ret = v4l2_fwnode_endpoint_parse(ep, &vep);
>> +		if (ret)
>> +			goto err_parse;
>> +
>> +		rk_asd = kzalloc(sizeof(*rk_asd), GFP_KERNEL);
>> +		if (!rk_asd) {
>> +			ret = -ENOMEM;
>> +			goto err_parse;
>> +		}
>> +
>> +		rk_asd->lanes = vep.bus.mipi_csi2.num_data_lanes;
>> +		rk_asd->mbus_type = vep.bus_type;
>> +
>> +		/* Parallel bus is currently not supported */
>> +		rk_asd->parallel_bus_flags = 0;
>> +		ret = v4l2_async_notifier_add_fwnode_remote_subdev(ntf, ep,
>> +								   &rk_asd->asd);
>> +		if (ret)
>> +			goto err_parse;
>> +
>> +		dev_dbg(rkisp1->dev, "registered ep id %d with %d lanes\n",
>> +			vep.base.id, rk_asd->lanes);
>> +
>> +		next_id = vep.base.id + 1;
>> +
>> +		fwnode_handle_put(ep);
>>  
>> -	return v4l2_async_notifier_register(&rkisp1->v4l2_dev, ntf);
>> +		continue;
>> +err_parse:
>> +		fwnode_handle_put(ep);
>> +		kfree(rk_asd);
>> +		v4l2_async_notifier_cleanup(ntf);
>> +		return ret;
>> +	}
>> +
>> +	if (next_id == 0)
>> +		dev_warn(rkisp1->dev, "no remote subdevice found\n");
> 
> I guess the driver will be loaded if the module is around and the device
> exists. If the board has no cameras, is that something on which a warning
> should be produced? I'd perhaps use dev_dbg(), if I'd print this at all.
> 
>> +	ntf->ops = &rkisp1_subdev_notifier_ops;
>> +	ret = v4l2_async_notifier_register(&rkisp1->v4l2_dev, ntf);
>> +	if (ret) {
>> +		v4l2_async_notifier_cleanup(ntf);
>> +		return ret;
>> +	}
>> +	return 0;
>>  }
>>  
>>  /* ----------------------------------------------------------------------------
>