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
