> -----Original Message----- > From: Hans Verkuil [mailto:hverkuil@xxxxxxxxx] > Sent: Wednesday, March 04, 2009 4:45 PM > To: Aguirre Rodriguez, Sergio Alberto > Cc: sakari.ailus@xxxxxxxxxxxxxxxxxxxxxxxxxx; DongSoo(Nathaniel) Kim; > Hiremath, Vaibhav; Toivonen Tuukka.O (Nokia-D/Oulu); linux- > omap@xxxxxxxxxxxxxxx; Nagalla, Hari; linux-media@xxxxxxxxxxxxxxx > Subject: Re: [REVIEW PATCH 11/14] OMAP34XXCAM: Add driver > > On Wednesday 04 March 2009 22:46:07 Aguirre Rodriguez, Sergio Alberto > wrote: > > > -----Original Message----- > > > From: Hans Verkuil [mailto:hverkuil@xxxxxxxxx] > > > Sent: Wednesday, March 04, 2009 3:06 PM > > > To: sakari.ailus@xxxxxxxxxxxxxxxxxxxxxxxxxx > > > Cc: DongSoo(Nathaniel) Kim; Hiremath, Vaibhav; Toivonen Tuukka.O > > > (Nokia- D/Oulu); Aguirre Rodriguez, Sergio Alberto; > > > linux-omap@xxxxxxxxxxxxxxx; Nagalla, Hari; linux-media@xxxxxxxxxxxxxxx > > > Subject: Re: [REVIEW PATCH 11/14] OMAP34XXCAM: Add driver > > > > > > On Wednesday 04 March 2009 20:22:04 Sakari Ailus wrote: > > > > Hans Verkuil wrote: > > > > > On Wednesday 04 March 2009 01:42:13 DongSoo(Nathaniel) Kim wrote: > > > > >> Thank you for your kind explanation Hans. > > > > >> > > > > >> Problem is omap3 camera subsystem is making device node for every > > > > >> int device attached to it. > > > > > > > > > > That's wrong. Multiple devices should only be created if they can > > > > > all be used at the same time. Otherwise there should be just one > > > > > device that uses S_INPUT et al to select between the inputs. > > > > > > > > There might be situations where multiple device nodes would be > > > > beneficial even if they cannot be used simultaneously in all cases. > > > > > > > > Currently the omap34xxcam camera driver creates one device per > > > > camera. A camera in this case contains an isp (or camera controller), > > > > image sensor, lens and flash. The properties like maximum frame rate > > > > or resolution of a camera are usually (almost) completely defined by > > > > those of the sensor, lens and flash. This affects also cropping > > > > capabilities. > > > > > > > > Several programs can access video devices simultaneously. What > > > > happens if another program switches the input when the first one > > > > doesn't expect it? The original user won't notice the change, > instead > > > > of getting -EBUSY when trying to open the other video device. > > > > > > It is actually quite common to be able to switch inputs using one > > > program (e.g. v4l2-ctl) while another program also has the video node > > > open. This will typically be used for debugging or experimenting. > > > Depending on the hardware, switching inputs while capturing is in > > > progress may or may not be > > > allowed (the driver might just return -EBUSY in that case). > > > > > > In addition the application can also call VIDIOC_S_PRIORITY to protect > > > it against outside interference, provided the driver supports this > > > ioctl. > > > > > > As an aside: many applications don't use VIDIOC_S_PRIORITY since > > > whether a driver implements it is hit-and-miss. As part of the new > v4l2 > > > framework I have a struct v4l2_fh planned that will integrate support > > > of this ioctl in the framework, thus making it generic for all drivers. > > > But this won't be available any time soon. > > > > As what I understand, we have 2 possible situations for multiple opens > > here: > > > > Situation 1 > > - Instance1: Select sensor 1, and Do queue/dequeue of buffers. > > - Instance2: If sensor 1 is currently selected, Begin loop requesting > > internally collected OMAP3ISP statistics (with V4L2 private based > IOCTLs) > > for performing user-side Auto-exposure, Auto White Balance, Auto Focus > > algorithms. And Adjust gains (with S_CTRL) accordingly on sensor as a > > result. > > Question: if you have multiple sensors, and sensor 1 is selected, can you > still get statistics from sensor 2? Or is all this still limited to the > selected sensor? Just want to make sure I understand it all. Everytime a RAW10 Bayer formatted frame is received on the OMAP3ISP, statistics are being generated and when done, stored in buffers allocated internally on the driver (last 5 frames only), marking every stats buffer with the frame number associated. When requesting stats, the user needs to specify the frame number (since last streamon call) to the which he needs the statistics. Receives in return the associated buffer. So, going back to answering directly your question, the stats are related to the frame_number since last stream on call, making it impossible to request from one specific sensor. > > > Situation 2 > > - Instance1: Select sensor1 as input. Begin streaming. > > - Instance2: Select sensor2 as input. Attempt to begin streaming. > > > > So, if I understood right, on Situation 2, if you attempt to do a > S_INPUT > > to sensor2 while capturing from sensor1, it should return a -EBUSY, > > right? > > That is probably what the driver should do, yes. The V4L2 spec leaves it > up > to the driver whether you can switch inputs while a capture is in progress. > In this case I think it is perfectly reasonably for the driver to > return -EBUSY. Ok, makes sense. > > > I mean, the app should consciously make sure the input (sensor) is > > the correct one before performing any adjustments. > > Can you elaborate? I don't follow this, I'm afraid. Normally an app already knows what are the capabilities/limitations for /dev/video0, and that wont change in our current approach, right? (Because video0 node always points to the same sensor) But with your suggested approach, if a switching happens, the capabilities could vary between sensors, and the app should need to re request capabilities everytime it attempts to adjust something. So this is the _consciousness_ I'm talking about.. (Constantly making sure of sensor capabilities before sending S_CTRL calls) Did I made myself clear? > > > I think our current approach avoids the user to be constantly checking > if > > the input is still the same before updating gains... > > If I understand it correctly, the normal approach would be one application > that might have a separate thread to do the gain adjustments, etc. Or do > it > all in the same thread, for example by doing the adjustments after every X > frames captured. Doing this in two separate applications seems awkward to > me. Usually the streaming thread and the adjustment thread need to > communicate with one another or be directed from some the main application. Hmm... Ok, got your point. I guess we can leave all responsibility of correct settings<->sensor association to the user app threads then. > > > I'm not clear if this single-node idea is really helping the user to > have > > a simpler usage in Situation 1, which I feel will become pretty common > on > > this driver... > > The spec is pretty clear about this, and existing v4l2 applications also > expect this behavior. Also, suppose you have two video nodes, what happens > when you want to use the inactive node? How can you tell it is inactive? Sorry, I think I'm not very clear on your intention... You mean that how can an app know if streaming is ongoing in a device node? If that's the case, then I think we'll need to have an IOCTL for knowing if the node is streaming or not. Please clarify. > > Things might change, however, if statistics gathering can proceed on an > inactive input. In that case I need to look into it more closely. > > Regards, > > Hans > > -- > Hans Verkuil - video4linux developer - sponsored by TANDBERG -- 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
