Hi Jacopo,

(CC'ing Sakari)

On Tuesday, 19 December 2017 13:57:42 EET jacopo mondi wrote:
> On Mon, Dec 11, 2017 at 06:15:23PM +0200, Laurent Pinchart wrote:
> > Hi Jacopo,
> > 
> > Thank you for the patch.
> > 
> > [snip]
> > 
> >> +static int ceu_sensor_bound(struct v4l2_async_notifier *notifier,
> >> +                      struct v4l2_subdev *v4l2_sd,
> >> +                      struct v4l2_async_subdev *asd)
> >> +{
> >> +  struct v4l2_device *v4l2_dev = notifier->v4l2_dev;
> >> +  struct ceu_device *ceudev = v4l2_to_ceu(v4l2_dev);
> >> +  struct ceu_subdev *ceu_sd = to_ceu_subdev(asd);
> >> +
> >> +  if (video_is_registered(&ceudev->vdev)) {
> >> +          v4l2_err(&ceudev->v4l2_dev,
> >> +                   "Video device registered before this sub-device.\n");
> >> +          return -EBUSY;
> > 
> > Can this happen ?
> > 
> >> +  }
> >> +
> >> +  /* Assign subdevices in the order they appear */
> >> +  ceu_sd->v4l2_sd = v4l2_sd;
> >> +  ceudev->num_sd++;
> >> +
> >> +  return 0;
> >> +}
> >> +
> > > +static int ceu_sensor_complete(struct v4l2_async_notifier *notifier)
> > > +{
> > > + struct v4l2_device *v4l2_dev = notifier->v4l2_dev;
> > > + struct ceu_device *ceudev = v4l2_to_ceu(v4l2_dev);
> > > + struct video_device *vdev = &ceudev->vdev;
> > > + struct vb2_queue *q = &ceudev->vb2_vq;
> > > + struct v4l2_subdev *v4l2_sd;
> > > + int ret;
> > > +
> > > + /* Initialize vb2 queue */
> > > + q->type                 = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;
> > > + q->io_modes             = VB2_MMAP | VB2_USERPTR;
> > 
> > No dmabuf ?
> > 
> > > + q->drv_priv             = ceudev;
> > > + q->ops                  = &ceu_videobuf_ops;
> > > + q->mem_ops              = &vb2_dma_contig_memops;
> > > + q->buf_struct_size      = sizeof(struct ceu_buffer);
> > > + q->timestamp_flags      = V4L2_BUF_FLAG_TIMESTAMP_MONOTONIC;
> > > + q->lock                 = &ceudev->mlock;
> > > + q->dev                  = ceudev->v4l2_dev.dev;
> > 
> > [snip]
> > 
> > > +static int ceu_probe(struct platform_device *pdev)
> > > +{
> > > + struct device *dev = &pdev->dev;
> > > + struct ceu_device *ceudev;
> > > + struct resource *res;
> > > + void __iomem *base;
> > > + unsigned int irq;
> > > + int num_sd;
> > > + int ret;
> > > +
> > > + ceudev = kzalloc(sizeof(*ceudev), GFP_KERNEL);
> > 
> > The memory is freed in ceu_vdev_release() as expected, but that will only
> > work if the video device is registered. If the subdevs are never bound,
> > the ceudev memory will be leaked if you unbind the CEU device from its
> > driver. In my opinion this calls for registering the video device at
> > probe time (although Hans disagrees).
> > 
> > > + if (!ceudev)
> > > +         return -ENOMEM;
> > > +
> > > + platform_set_drvdata(pdev, ceudev);
> > > + dev_set_drvdata(dev, ceudev);
> > 
> > You don't need the second line, platform_set_drvdata() is a wrapper around
> > dev_set_drvdata().
> > 
> > > + ceudev->dev = dev;
> > > +
> > > + INIT_LIST_HEAD(&ceudev->capture);
> > > + spin_lock_init(&ceudev->lock);
> > > + mutex_init(&ceudev->mlock);
> > > +
> > > + res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> > > + if (IS_ERR(res))
> > > +         return PTR_ERR(res);
> > 
> > No need for error handling here, devm_ioremap_resource() will check the
> > res
> > pointer.
> > 
> > > + base = devm_ioremap_resource(dev, res);
> > 
> > You can assign ceudev->base directly and remove the base local variable.
> > 
> > > + if (IS_ERR(base))
> > > +         return PTR_ERR(base);
> > > + ceudev->base = base;
> > > +
> > > + ret = platform_get_irq(pdev, 0);
> > > + if (ret < 0) {
> > > +         dev_err(dev, "failed to get irq: %d\n", ret);
> > > +         return ret;
> > > + }
> > > + irq = ret;
> > > +
> > > + ret = devm_request_irq(dev, irq, ceu_irq,
> > > +                        0, dev_name(dev), ceudev);
> > > + if (ret) {
> > > +         dev_err(&pdev->dev, "Unable to register CEU interrupt.\n");
> > > +         return ret;
> > > + }
> > > +
> > > + pm_suspend_ignore_children(dev, true);
> > > + pm_runtime_enable(dev);
> > > +
> > > + ret = v4l2_device_register(dev, &ceudev->v4l2_dev);
> > > + if (ret)
> > > +         goto error_pm_disable;
> > > +
> > > + if (IS_ENABLED(CONFIG_OF) && dev->of_node) {
> > > +         num_sd = ceu_parse_dt(ceudev);
> > > + } else if (dev->platform_data) {
> > > +         num_sd = ceu_parse_platform_data(ceudev, dev->platform_data);
> > > + } else {
> > > +         dev_err(dev, "CEU platform data not set and no OF support\n");
> > > +         ret = -EINVAL;
> > > +         goto error_v4l2_unregister;
> > > + }
> > > +
> > > + if (num_sd < 0) {
> > > +         ret = num_sd;
> > > +         goto error_v4l2_unregister;
> > > + } else if (num_sd == 0)
> > > +         return 0;
> > 
> > You need braces around the second statement too.
> 
> Ok, actually parse_dt() and parse_platform_data() behaves differently.
> The former returns error if no subdevices are connected to CEU, the
> latter returns 0. That's wrong.
> 
> I wonder what's the correct behavior here. Other mainline drivers I
> looked into (pxa_camera and atmel-isc) behaves differently from each
> other, so I guess this is up to each platform to decide.

No, what it means is that we've failed to standardize it, not that it 
shouldn't be standardized :-)

> Also, the CEU can accept one single input (and I made it clear
> in DT bindings documentation saying it accepts a single endpoint,
> while I'm parsing all the available ones in driver, I will fix this)
> but as it happens on Migo-R, there could be HW hacks to share the input
> lines between multiple subdevices. Should I accept it from dts as well?
> 
> So:
> 1) Should we fail to probe if no subdevices are connected?

While the CEU itself would be fully functional without a subdev, in practice 
it would be of no use. I would thus fail probing.

> 2) Should we accept more than 1 subdevice from dts as it happens right
> now for platform data?

We need to support multiple connected devices, as some of the boards require 
that. What I'm not sure about is whether the multiplexer on the Migo-R board 
should be modeled as a subdevice. We could in theory connect multiple sensors 
to the CEU input signals without any multiplexer as long as all but one are in 
reset with their outputs in a high impedance state. As that wouldn' require a 
multiplexer we would need to support multiple endpoints in the CEU port. We 
could then support Migo-R the same way, making the multiplexer transparent.

Sakari, what would you do here ?

-- 
Regards,

Laurent Pinchart

Reply via email to