Hans Verkuil <hverkuil@xxxxxxxxx> wrote: >I've been working on control events and while doing that I encountered >a >V4L2 API problem. > >Currently calling select() without REQBUFS having been called first >will >assume that you want to read (or write for output devices) and it will >start the DMA accordingly if supported by the driver. > >This is very nice for applications that want to use read (usually for >MPEG >encoders), since they don't need to do an intial read() to kickstart >the >DMA. It is also pretty much what you expect to happen. > >Unfortunately, this clashes with applications that just select on >exceptions >(like a control panel type application that just wants to get events >when >controls change value). Now a select() on an exception will cause the >driver >to start the DMA. Obviously not what you want. > >Ideally the driver's poll() implementation should check whether the >user >wanted to wait for input, output or exceptions. Unfortunately, that >information >is not in general passed to the driver (see the code in fs/select.c). > >I am not certain what to do. One option is that poll no longer can kick >off >a DMA action. Instead apps need to call read() or write() at least >once. In >that case a read/write count of 0 should be allowed: that way you can >start >the DMA without actually needing to read or write data. > >So apps can do a read(fd, buf, 0), then use poll afterwards to wait for >data >to arrive. > >I'm not sure whether this is OK or not. > >Another alternative would be to try and change the poll() op so that >this >information is actually passed to the driver. That's a rather painful >alternative, though. > >Regards, > > Hans >-- >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 Off the cuff: When the stream is not started, just have the poll() method return ready for reading and/or writing. For blocking fd's it is technically true, depending on how one defines "ready". If the fd is opened nonblocking, the first read() of DQBUF could return -EWOULDBLOCK, and schedule the stream to be started with a one-off work item. Regards, Andy -- 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
