Re: V4L2_MEMORY_USERPTR support in videobuf-core

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi Pavel,

On Tuesday 27 October 2009 17:17:59 Pawel Osciak wrote:
> Hi Mauro,
> thank you for your reply.
> 
> On Tuesday, October 27, 2009 1:36 PM
> 
> Mauro Carvalho Chehab [mailto:mchehab@xxxxxxxxxxxxx] wrote:
> > > could anybody confirm that there is no full/working support for USERPTR
> > > in current videobuf-core? That is the conclusion I came up with after a
> > > more thorough investigation.
> >>
> > > I am currently working to fix that, and will hopefully be posting
> > > patches in the coming days/weeks. Is there any other development effort
> > > underway related to this problem?
> >
> > (...)
> > The last time I tested the support for userptr at videobuf-core, it were
> > working on x86 plataforms. On that time, I used vivi with videobuf-dma-sg
> > for such tests (it were before its conversion to use videobuf-vmalloc).
> > As support for userptr on videobuf-vmalloc is missing, vivi can't be used
> > for such tests anymore (a good contribution would be to add userptr
> > support on videobuf-vmalloc).
> 
> I might be missing something, but for me the path looks as follows
> (sources: kernel, LWN articles, V4L2 API Specification):
> 
> 1. open, query, format, other stuff, unimportant
> 2. VIDEOBUF_REQBUFS - pass type and set memory to V4L2_MEMORY_USERPTR only.
> 3. VIDEOBUF_QUERYBUFS - only for memory-mapped I/O, so not called.
> 4. VIDEOBUF_QBUF - pass type, memory, userptr and length fields only.
> 
> As the API Specification states in section 3.3:
> "No buffers are allocated beforehands, consequently they are not indexed
>  and cannot be queried like mapped buffers with the VIDIOC_QUERYBUF ioctl."
> 
> But when one calls QBUF, videobuf_qbuf() uses b->index for all types of
> memory. I have found no mention in the API Specs about passing/returning
> indexes in USERPTR, quite the contrary, they actually state that indexes
> are not used in that mode for neither REQBUFS nor QBUF at all.

I'm not sure about the history of this issue, but it's a specification problem 
in my opinion. Most (if not all) drivers that support USERPTR require a number 
of buffers to be specified when calling VIDEOBUF_REQBUFS and use indexes like 
they would for MMAP buffers.

> Even if that was the case though, would an application be supposed to
> arbitrarily choose what index to pass? If so, how would it know what range
> is valid? And even if it would, the next check:
> (buf->state != VIDEOBUF_NEEDS_INIT && buf_state != VIDEOBUF_IDLE) would
>  most probably fail anyway.
> 
> How to enqueue and handle multiple userptr buffers?

You should set the number of buffers are VIDIOC_REQBUFS time and handle 
QBUF/DQBUF like you would for MMAP buffers.

-- 
Regards,

Laurent Pinchart
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[Index of Archives]     [Linux Input]     [Video for Linux]     [Gstreamer Embedded]     [Mplayer Users]     [Linux USB Devel]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [Yosemite Backpacking]
  Powered by Linux