Re: [PATCH] media: uvcvideo: Read support

[Hugues FRUCHET <hugues.fruchet@st.com>]

Hi Kieran, Laurent,

On 3/4/19 4:44 PM, Laurent Pinchart wrote:
> Hello,
> 
> On Mon, Mar 04, 2019 at 03:36:32PM +0000, Kieran Bingham wrote:
>> On 04/03/2019 12:35, Hugues Fruchet wrote:
>>> Add support of read() call from userspace by implementing
>>> uvc_v4l2_read() with vb2_read() helper.
>>
>> Just thinking out loud,
>>
>> This opens up UVC devices to read raw full frame images through this
>> interface as well.
>>
>> Due to the UVC protocol, there is /already/ a full memcpy to get these
>> images out of the URB packets, so using a read() interface would be
>> another full frame copy.
>>
>> I can perhaps see the usecase for reading compressed data through this
>> interface - but full frames don't seem appropriate. (not impossible of
>> course, just is it reasonable?)
>>
>> If this is to be enabled, should it be enabled for compressed formats
>> only? or would that complicate matters?
> 
> I've repeatedly refused read() support in uvcvideo for this reason, and
> also because read() doesn't carry framing information very well. It's
> just not a good API for capturing video frames from a webcam, and so far
> I haven't heard a compeling reason why it should be enabled. I thus
> haven't changed my mind :-)
> 

For sure read() is not optimal, but is very common and pretty simple to 
use from userspace as explained in the cover letter with some examples.

Moreover, read() is enabled by many (all ?) camera interfaces:

drivers/media/platform/pxa_camera.c:	.read		= vb2_fop_read,
drivers/media/platform/atmel/atmel-isi.c:	.read	= vb2_fop_read,
drivers/media/platform/rcar-vin/rcar-v4l2.c:	.read	= vb2_fop_read,
drivers/media/platform/stm32/stm32-dcmi.c:	.read	= vb2_fop_read,
drivers/media/platform/vivid/vivid-core.c:	.read   = vb2_fop_read,
drivers/media/platform/vimc/vimc-capture.c:	.read   = vb2_fop_read,
drivers/media/platform/marvell-ccic/mcam-core.c:.read   = vb2_fop_read,
drivers/media/platform/qcom/camss/camss-video.c:.read	= vb2_fop_read,

on my setup, this leads to have /dev/video0 (mapped on DCMI camera 
interface) supporting JPEG streaming through read() while /dev/video1 
(mapped on Hercule USB camera) fails with "invalid argument" error...


>>> Signed-off-by: Hugues Fruchet <hugues.fruchet@st.com>
>>> ---
>>>   drivers/media/usb/uvc/uvc_queue.c | 15 ++++++++++++++-
>>>   drivers/media/usb/uvc/uvc_v4l2.c  | 11 ++++++++---
>>>   drivers/media/usb/uvc/uvcvideo.h  |  2 ++
>>>   3 files changed, 24 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/drivers/media/usb/uvc/uvc_queue.c b/drivers/media/usb/uvc/uvc_queue.c
>>> index 682698e..0c8a0a8 100644
>>> --- a/drivers/media/usb/uvc/uvc_queue.c
>>> +++ b/drivers/media/usb/uvc/uvc_queue.c
>>> @@ -227,7 +227,7 @@ int uvc_queue_init(struct uvc_video_queue *queue, enum v4l2_buf_type type,
>>>   	int ret;
>>>   
>>>   	queue->queue.type = type;
>>> -	queue->queue.io_modes = VB2_MMAP | VB2_USERPTR;
>>> +	queue->queue.io_modes = VB2_MMAP | VB2_USERPTR | VB2_READ;
>>>   	queue->queue.drv_priv = queue;
>>>   	queue->queue.buf_struct_size = sizeof(struct uvc_buffer);
>>>   	queue->queue.mem_ops = &vb2_vmalloc_memops;
>>> @@ -361,6 +361,19 @@ int uvc_queue_streamoff(struct uvc_video_queue *queue, enum v4l2_buf_type type)
>>>   	return ret;
>>>   }
>>>   
>>> +ssize_t uvc_queue_read(struct uvc_video_queue *queue, struct file *file,
>>> +		       char __user *buf, size_t count, loff_t *ppos)
>>> +{
>>> +	ssize_t ret;
>>> +
>>> +	mutex_lock(&queue->mutex);
>>> +	ret = vb2_read(&queue->queue, buf, count, ppos,
>>> +		       file->f_flags & O_NONBLOCK);
>>> +	mutex_unlock(&queue->mutex);
>>> +
>>> +	return ret;
>>> +}
>>> +
>>>   int uvc_queue_mmap(struct uvc_video_queue *queue, struct vm_area_struct *vma)
>>>   {
>>>   	return vb2_mmap(&queue->queue, vma);
>>> diff --git a/drivers/media/usb/uvc/uvc_v4l2.c b/drivers/media/usb/uvc/uvc_v4l2.c
>>> index 84be596..3866832 100644
>>> --- a/drivers/media/usb/uvc/uvc_v4l2.c
>>> +++ b/drivers/media/usb/uvc/uvc_v4l2.c
>>> @@ -594,7 +594,8 @@ static int uvc_ioctl_querycap(struct file *file, void *fh,
>>>   	strscpy(cap->driver, "uvcvideo", sizeof(cap->driver));
>>>   	strscpy(cap->card, vdev->name, sizeof(cap->card));
>>>   	usb_make_path(stream->dev->udev, cap->bus_info, sizeof(cap->bus_info));
>>> -	cap->capabilities = V4L2_CAP_DEVICE_CAPS | V4L2_CAP_STREAMING
>>> +	cap->capabilities = V4L2_CAP_DEVICE_CAPS | V4L2_CAP_STREAMING |
>>> +			    V4L2_CAP_READWRITE
>>>   			  | chain->caps;
>>>   
>>>   	return 0;
>>> @@ -1434,8 +1435,12 @@ static long uvc_v4l2_compat_ioctl32(struct file *file,
>>>   static ssize_t uvc_v4l2_read(struct file *file, char __user *data,
>>>   		    size_t count, loff_t *ppos)
>>>   {
>>> -	uvc_trace(UVC_TRACE_CALLS, "uvc_v4l2_read: not implemented.\n");
>>> -	return -EINVAL;
>>> +	struct uvc_fh *handle = file->private_data;
>>> +	struct uvc_streaming *stream = handle->stream;
>>> +
>>> +	uvc_trace(UVC_TRACE_CALLS, "uvc_v4l2_read\n");
>>> +
>>> +	return uvc_queue_read(&stream->queue, file, data, count, ppos);
>>>   }
>>>   
>>>   static int uvc_v4l2_mmap(struct file *file, struct vm_area_struct *vma)
>>> diff --git a/drivers/media/usb/uvc/uvcvideo.h b/drivers/media/usb/uvc/uvcvideo.h
>>> index c7c1baa..5d0515c 100644
>>> --- a/drivers/media/usb/uvc/uvcvideo.h
>>> +++ b/drivers/media/usb/uvc/uvcvideo.h
>>> @@ -766,6 +766,8 @@ struct uvc_buffer *uvc_queue_next_buffer(struct uvc_video_queue *queue,
>>>   					 struct uvc_buffer *buf);
>>>   struct uvc_buffer *uvc_queue_get_current_buffer(struct uvc_video_queue *queue);
>>>   void uvc_queue_buffer_release(struct uvc_buffer *buf);
>>> +ssize_t uvc_queue_read(struct uvc_video_queue *queue, struct file *file,
>>> +		       char __user *buf, size_t count, loff_t *ppos);
>>>   int uvc_queue_mmap(struct uvc_video_queue *queue,
>>>   		   struct vm_area_struct *vma);
>>>   __poll_t uvc_queue_poll(struct uvc_video_queue *queue, struct file *file,
> 

Best regards,
Hugues.