On Fri, Jun 12, 2020 at 8:42 PM Stanimir Varbanov <stanimir.varbanov@xxxxxxxxxx> wrote: > > Hi Tomasz, > > On 6/12/20 9:33 PM, Tomasz Figa wrote: > > Hi Stanimir, > > > > On Fri, Jun 12, 2020 at 5:20 PM Stanimir Varbanov > > <stanimir.varbanov@xxxxxxxxxx> wrote: > >> > >> I failed to found previous discussion on the subject, maybe it was on IRC. > >> > >> I stuck with fixing of issues which depends on this reqbuf(count=0). > >> Currently I'm counting the buffers in vb2::buf_init and vb2::buf_cleanup > >> but I don't want to rely on that implicit way anymore. > >> > >> If someone of you can remember let me know so that I can prepare an RFC. > >> > >> I can think of two options: > >> - change vb2::queue_setup so it could be called with num_buffers=0 > >> - add new vb2::queue_release op > >> > >> More options? > > > > Just to make sure we're on the same page: Are you looking for a way to > > handle some operation only when all buffers on the queue are being > > freed? > > Exactly :) > > Presently I'm counting capture and output buffers in vb2::buf_init and > vb2::buf_cleanup and release/reinit internal driver resources on the > last .buf_cleanup. And this is a workaround of the problem which I want > to solve properly. Got it. I suppose queue_release() could be a good idea to keep things symmetrical, but for now, wouldn't it be still possible to hook directly into .vidioc_reqbufs and do whatever necessary when count == 0? Best regards, Tomasz
