hverkuil: am I missing something or we have a bug in vb2 in handling of v4l2_plane::length == 0?
hverkuil: __prepare_dmabuf() will call __fill_vb2_buffer(), which will call __verify_length(), which will fail if plane length is 0
hverkuil: however, in __prepare_dmabuf(), several lines below, there is code that handles the case when length is 0
hverkuil: https://elixir.bootlin.com/linux/latest/source/drivers/media/common/videobuf2/videobuf2-core.c#L1127
tfiga: do you mean that b->m.planes[plane].length can be zero but b->m.planes[plane].bytesused != 0?
svarbanov: yes
technically bytesused can be 0 too
it doesn't matter in this case
svarbanov: ah, it matters, because it's for OUTPUT planes
tfiga: it is for OUTPUT planes __only__
but then, for OUTPUT planes it's unlikely to have 0 bytesused
tfiga: bytesused can be filled by drivers only, right?
no, bytesused for OUTPUT buffers is filled by the application
the driver doesn't know how much data the application wrote to the buffer
tfiga: right, I always think about capture buffers :)
I always use uppercase when using V4L2 buffer type
so effectively, CAPTURE buffers are output buffers
the spec also allows the userspace to set bytesused to 0
and then it's assumed to be the same as length indeed
svarbanov: also the spec defines length as "Size in bytes of the plane (not its payload). This is set by the driver based on the calls to ioctl VIDIOC_REQBUFS and/or ioctl VIDIOC_CREATE_BUFS."
it doesn't say that the application needs to fill it...
so failing if it's 0 sounds like a bug
tfiga: but what is a plane with size = 0 :)
it is more like dmabuf workaround when the size is not set by userspace
svarbanov: why would the userspace need to set the size?
it doesn't set it for MMAP buffers either
because the size is known from the DMA-buf
and the code from my link seems to attempt to handle it
but it cannot happen, because __verify_length() fails before
in any case, we have to fix it by either specifying that the application is required to set it for USERPTR and DMA-buf or by allowing it to be 0
sailus, Hrm ... is your webserver/git server down? I can't clone : git://git.retiisi.org.uk/~sailus/raw2rgbpnm.git
sailus, Aha - never mind - seems I might have had an old url ? <git://salottisipuli.retiisi.org.uk/~sailus/raw2rgbpnm.git> worked.
kbingham: Yeah, it's a different virtual machine now...
hverkuil: tfiga: just sent a series with some pixelformat helpers. if it's what you had in mind, we can add the missing bits.
mchehab: Do you have plans to write a summit report from the last week's meeting?
I'd be happy to do that, perhaps apart from the CEC or Rc-core topics which I know little about.
(And remember little as well.)
sailus: yes, we should write a summit report
if you have time for doing it, that would be great!
mchehab: Ack.
I expect to post a draft (w/o CEC/RC-core) during the next week.
Time to leave.
Have a nice week-end!
have a nice weekend
kbingham: I just realised I had a problem with the firewall configuration.
The right hostname is indeed git.retiisi.org.uk; it seems you're the first to notice in some year or so.
Or... could it be others are using ipv6? ;)