Re: [PATCH] media: platform: pxa_camera: convert to vb2
From: Robert Jarzmik
Date: Fri Mar 11 2016 - 10:40:27 EST
Hans Verkuil <hverkuil@xxxxxxxxx> writes:
> On 03/11/2016 02:41 PM, Robert Jarzmik wrote:
>> Hans Verkuil <hverkuil@xxxxxxxxx> writes:
> One area where I would like to see some helper functions is with respect to
> format/media bus processing. I played with this a little bit but it is surprisingly
> hard to do. A lot of devices have all sorts of weird and wonderful exceptions
> that make this quite problematic.
I'm also worried about the initial probing, where the subdevice, be that an I2C
sensor or something else has to be available, ie. the v4l2_async_notifier and
>> Ah, that's a special case we need to discuss.
>> I've written in the commit message a chapter about a "special port of this
>> code". This is it.
>> This usecase is when a user does the following :
>> - set format to 1280x1024, RGB565
>> - REQBUF for MMAP buffers
>> - QBUF, capture, DQBUF
>> - then set format to 640x480, RGB565
>> => here the new format fits in the previously allocated video buffer
>> - QBUF
>> => the test in pxa_vb2_prepare() detects this, and calls pxa_buffer_init()
>> Now if this usecase is impossible, then I'll do as you say to simplify the code
>> : use icd->sizeimage, remove the code in pxa_vb2_prepare(), etc ...
> Does this actually work with soc-camera? As far as I can see soc-camera returns
> -EBUSY in soc_camera_s_fmt_vid_cap() if you attempt to change the format while
It's not "while streaming" in the described usecase, it's after streaming is
finished actually. I should have added in the third dash VIDIOC_STREAMON before
"capture" and VIDIOC_STREAMOFF after DQBUF. I think it's working, even if I had
not tried recently. I certainly don't care that much about the usecase, and I
won't feel sad dropping it :)
> We theorized about this use-case, but nobody actually implemented it.
> As far as I can see this use-case isn't supported today, so I would certainly not
> implement it for this vb2 conversion.
Fair enough, simpler is better, I'll remove it for v2.
>>> If the latter, then just remove it. If you don't have the memory to allocate
>>> the buffers, then reqbufs will just return ENOMEM. I never saw a reason for
>>> such checks.
>> Okay, that was to be consistent with former driver behavior. This was from the
>> beginning in this driver. If Guennadi doesn't care, then I'll remove that, as he
>> is the original author of this limitation.
> I've removed it from other drivers in the past, nobody complained :-)
Good, let's do that here then.