We'd prefer to throttle the decoder if we run out of buffers, to keep a bound
on memory usage. Also, for OMAP4 it is a requirement of the decoder to not
alternate between memory alloced by the display driver and malloc'd userspace
memory.
It seems to cause strange occasional high latencies (almost 200ms) when dequeuing buffers from _buffer_alloc(). It is simpler and seems to work much better to dqbuf from the same thread that is queuing the next buffer.
This also does the following changes:
(1) pull the bufferpool code out into gstv4l2bufferpool.c, and make a
bit more generic so it can be used both for v4l2src and v4l2sink
(2) move some of the device probing/configuration/caps stuff into
gstv4l2object.c so it does not have to be duplicated between
v4l2src and v4l2sink
Fixes bug #590280.
The v4l2 driver for USB webcams on OpenSolaris does not support select()
calls. Detect when select() fails, and skip polling the device afterward,
which restores the pre 0.10.14 behaviour on OpenSolaris.
Signed-off-by: Jan Schmidt <thaytan@noraisin.net>
Use GstPoll to wait for the fd of the video device to become readable before
trying to capture a frame. This speeds up stopping v4l2src a lot as it no
longer has to wait for the next frame, especially when capturing with low
framerates or when the video device just never generates a frame (which seems a
common issue for uvcvideo devices)
Fixes bug #563574.
sort_by_frame_size is declared static and only used inside
an ifdef, so use the same ifdef to define the function. Fixes#572185
Signed-off-by: David Schleef <ds@schleef.org>
Original commit message from CVS:
* sys/v4l2/v4l2src_calls.c:
(gst_v4l2src_probe_caps_for_format_and_size), (sort_by_frame_size),
(gst_v4l2src_probe_caps_for_format):
Turns out we don't always get the frame sizes in a predefined
order from lowest to highest resolution, so let's just sort the
list by frame size once we've queried the possible resolutions
rather than assume any particular order. Fixes probed caps for
the camera in my HP2133 mini notebook and makes v4l2src default
to a decent size.
Original commit message from CVS:
* sys/v4l2/v4l2src_calls.c: (gst_v4l2src_format_get_rank):
Guard more uncommon formats with ifdefs so that we can compile on older
versions.
Original commit message from CVS:
* sys/v4l2/gstv4l2object.c:
* sys/v4l2/gstv4l2src.c:
* sys/v4l2/gstv4l2src.h:
* sys/v4l2/v4l2_calls.c:
* sys/v4l2/v4l2src_calls.c:
Fix memory leaks. Small code cleanups : No need for empty _init(). No
need to memset instance structures. Some more FIXME's.
Original commit message from CVS:
* sys/v4l2/gstv4l2src.c: (gst_v4l2src_v4l2fourcc_to_structure),
(gst_v4l2_get_caps_info):
Add S910 and PWC formats with a low priority.
* sys/v4l2/v4l2src_calls.c: (gst_v4l2src_format_get_rank),
(gst_v4l2src_probe_caps_for_format):
Add more debugging.
Original commit message from CVS:
* sys/v4l2/gstv4l2src.c: (gst_v4l2src_query):
Properly set the maximum latency value, in the same way it is done in
v4lsrc.
* sys/v4l2/v4l2src_calls.c:
Simplify fraction equality check, no need to use GValues for this.
Original commit message from CVS:
* sys/v4l2/gstv4l2src.c: (gst_v4l2src_query):
Add warning messages stating exactly why the latency query failed.
* sys/v4l2/v4l2src_calls.c: (gst_v4l2src_set_capture):
In some cases, the negotiated framerate might be the default one which
is already set internally. But we still need to mark it down in fps_n
and fps_d so that the latency query can happen properly.
Original commit message from CVS:
* sys/v4l2/v4l2src_calls.c: (gst_v4l2src_probe_caps_for_format):
When outputting a pad template range for the size, include a framerate
range too, to avoid 'not a real subset of template caps' errors.
Original commit message from CVS:
Patch by: Daniel Drake <dsd at gentoo dot org>
* sys/v4l2/v4l2src_calls.c: (gst_v4l2src_set_capture),
(gst_v4l2src_get_nearest_size):
Try progressive video if interlaced fails. Fixes bug #541956
and the usage of v4l2src on OLPC.
Original commit message from CVS:
Patch by: William M. Brack <wbrack at mmm dot com dot hk>
* sys/v4l2/v4l2src_calls.c: (gst_v4l2src_probe_caps_for_format):
Fix potential caps leak.
If we can't get the framerate with an ioctl, try to get it with the
current norm. Fixes#520092.
Original commit message from CVS:
Patch by: William M. Brack <wbrack at mmm dot com dot hk>
* sys/v4l2/v4l2src_calls.c:
(gst_v4l2src_probe_caps_for_format_and_size):
If we fail to get the frame intervals, simply don't touch the framerates
on the template caps instead of discarding the format. See #520092.
Original commit message from CVS:
Based on patch by: William M. Brack <wbrack at mmm com hk>
* sys/v4l2/v4l2src_calls.c: (fractions_are_equal),
(gst_v4l2src_set_capture):
Check whether the device supports setting the framerate before
trying to set it and then posting a warning or error if it doesn't
work (#516649, #520092). Also compare fractions more correctly.
Original commit message from CVS:
Patch by: William M. Brack <wbrack at mmm com hk>
* sys/v4l2/v4l2src_calls.c:
(gst_v4l2src_probe_caps_for_format_and_size),
(gst_v4l2src_probe_caps_for_format):
Make sure the probed frame sizes are reversed in the resulting
caps also when using V4L2_FRMSIZE_STEPWISE (so they end up
highest resolution first); also remove unused variable.
(Partly fixes#520092)
Original commit message from CVS:
* sys/v4l2/v4l2src_calls.c: (gst_v4l2_buffer_finalize),
(gst_v4l2_buffer_class_init), (gst_v4l2_buffer_pool_finalize),
(gst_v4l2_buffer_pool_class_init):
Chain up the finalize functions. Fixes bug #515984.
Original commit message from CVS:
* sys/v4l2/v4l2_calls.c: (gst_v4l2_fill_lists):
Treat ENOTTY (driver does not implement ioctl) the same as
EINVAL since it implies there are no available standards.
* sys/v4l2/v4l2src_calls.c: (gst_v4l2src_probe_caps_for_format),
(gst_v4l2src_get_nearest_size):
Replace gst_v4l2src_get_size_limits with 2 calls to new function
gst_v4l2src_get_nearest_size, and get it to use VIDIOC_S_FMT to
probe if the driver does not support VIDIOC_TRY_FMT for whatever
reason, and if we aren't yet actively capturing.
* sys/v4l2/v4l2src_calls.h:
Remove replaced function declaration.
Original commit message from CVS:
* configure.ac:
Detect video4linux headers on Solaris too.
* sys/v4l2/gstv4l2colorbalance.h:
* sys/v4l2/gstv4l2object.h:
* sys/v4l2/v4l2_calls.c:
* sys/v4l2/v4l2src_calls.c: (gst_v4l2_buffer_finalize),
(gst_v4l2_buffer_new):
Make v4l2 build on Solaris.
Patch by: Brian Cameron <brian.cameron at sun dot com>
Fixes: #510505
Original commit message from CVS:
* sys/v4l2/v4l2_calls.c: (gst_v4l2_fill_lists):
* sys/v4l2/v4l2src_calls.c: (gst_v4l2src_fill_format_list):
Init some structs to zero before we pass them to ioctl, which
avoids valgrind warnings. Also fix a small memory leak.
Original commit message from CVS:
Always copy buffers by default (handle safer with bugged drivers) and added a property to make it possible to use mmap effectively (no copy if possible) when application wants to. Fixes: #480557.
Original commit message from CVS:
* sys/v4l2/v4l2src_calls.c: (gst_v4l2src_probe_caps_for_format):
If VIDIOC_ENUM_FRAMESIZES is defined (= recent kernel), but the
corresponding ioctl() call fails even though the driver claims to
support this format, just fall back to the pre-2.6.19 kernel
routine that creates caps with suitable height and width ranges
(see #448278).
Original commit message from CVS:
* sys/v4l2/v4l2src_calls.c: (gst_v4l2src_grab_frame):
Fixes "v4l2src ! queue ! xvimagesink". The queue ask for buffer too
early. It is temporary until we find something better.
Original commit message from CVS:
* sys/v4l2/v4l2src_calls.c: (gst_v4l2_buffer_finalize),
(gst_v4l2src_grab_frame):
Add some more debug info. Generate an error when we run out of buffers
for some reason. See #480557.
Original commit message from CVS:
* sys/v4l2/gstv4l2src.c:
* sys/v4l2/v4l2src_calls.c:
When probing the formats and sizes a camera supports, make
sure the best ones (highest resolution, prefered format)
end up at the beginning of the probed caps and the less
desirable ones at the end. This is important because the
order within the caps matters for things like fixation and
negotiation, ie. what format is chosen in the end.
With recent kernels, the current probing code will end up
querying the supported sizes from lowest resolution to
highest resolution, adding them to the probed caps in that
order, resulting to v4l2src fixating to the lowest possible
resolution if downstream does not express a size preference.
Also make up a somewhat random ranking of prefered output
formats for the same reason. Fixes#485828.
Original commit message from CVS:
* sys/v4l2/v4l2src_calls.c:
(gst_v4l2src_probe_caps_for_format_and_size):
Fix framerate detection code some more.
Handle the case where there is a weird step in the stepwise framerates.
Don't overwrite the min interval with the framerate, use a temp variable
instead.
Use max in the Continuous framerate intervals instead of step, which is
1 according to the docs. Fixes#475424.
Original commit message from CVS:
* sys/v4l2/v4l2src_calls.c:
(gst_v4l2src_probe_caps_for_format_and_size):
Don't overwrite our GValue with 0 but instead use the previously
computed value. Fixes#471823 some more.
Original commit message from CVS:
* sys/v4l2/gstv4l2src.c:
* sys/v4l2/gstv4l2src.h:
* sys/v4l2/v4l2src_calls.c:
Implement LATENCY queries in the crudest way possible so I don't
have to use sync=false any longer when testing with videosinks.
Original commit message from CVS:
* sys/v4l2/v4l2src_calls.c:
(gst_v4l2src_probe_caps_for_format_and_size):
Add some more debugging in the framerate function.
Iterate stepwise framerate up to and _including_ the max and if nothing
was added to the list, add a dummy 0/1 to 100/1 framerate so that we
don't end up with an empty list.
Original commit message from CVS:
* sys/v4l2/gstv4l2src.c: (gst_v4l2_get_caps_info),
(gst_v4l2src_set_caps), (gst_v4l2src_get_mmap):
Restructure the setcaps function so that we can also compute the
expected GStreamer output size of the video frames.
Set frame_byte_size correctly so that read-based devices have a chance
of working correctly.
When grabbing a frame, discard frames that are not of the expected size.
Some cameras don't output the right framesize for the first buffer.
Try only a couple of times to get a valid frame, else error out.
* sys/v4l2/v4l2_calls.c: (gst_v4l2_get_capabilities),
(gst_v4l2_fill_lists), (gst_v4l2_get_input):
Add some more debug info when scanning the device.
* sys/v4l2/v4l2src_calls.c: (gst_v4l2_buffer_new),
(gst_v4l2_buffer_pool_new), (gst_v4l2_buffer_pool_activate),
(gst_v4l2src_fill_format_list), (gst_v4l2src_grab_frame),
(gst_v4l2src_set_capture), (gst_v4l2src_capture_init):
Add some more debug info when dequeing a frame.
