| Message ID | 20250326075129.1705736-3-paul.elder@ideasonboard.com |
|---|---|
| State | New |
| Headers | show |
| Series |
|
| Related | show |
Le mercredi 26 mars 2025 à 16:51 +0900, Paul Elder a écrit : > Fill in the offset field of the frame metadata so that it the > application can read it. > > Signed-off-by: Paul Elder <paul.elder@ideasonboard.com> > --- > src/libcamera/v4l2_videodevice.cpp | 5 ++++- > 1 file changed, 4 insertions(+), 1 deletion(-) > > diff --git a/src/libcamera/v4l2_videodevice.cpp > b/src/libcamera/v4l2_videodevice.cpp > index f5b3fa09d9a0..5d163f7fd853 100644 > --- a/src/libcamera/v4l2_videodevice.cpp > +++ b/src/libcamera/v4l2_videodevice.cpp > @@ -1957,10 +1957,13 @@ FrameBuffer *V4L2VideoDevice::dequeueBuffer() > * The number of planes in the frame buffer and in > the > * V4L2 buffer is guaranteed to be equal at this > point. > */ > - for (unsigned int i = 0; i < numV4l2Planes; ++i) > + for (unsigned int i = 0; i < numV4l2Planes; ++i) { > metadata.planes()[i].bytesused = > planes[i].bytesused; > + metadata.planes()[i].offset = > planes[i].data_offset; In V4L2, data_offset is included in the total bytesused. Once you hit a non-zero offset, you will endup with byteused possibly bigger then FrameBuffer.length, which to my reading contradicts the doc. https://www.linuxtv.org/downloads/v4l-dvb-apis/userspace-api/v4l/buffer.html?highlight=data_offset#struct-v4l2-plane > + } > } else { > metadata.planes()[0].bytesused = buf.bytesused; > + metadata.planes()[0].offset = buf.m.offset; Should be: + metadata.planes()[0].offset = 0; There is no support for data_offset in the original v4l2_buffer API. This offset match planes[i].mem_offset and is only useful for CPU access using mmap() and the video node FD. I think some introspection is needed, since this offset was expected to be in the framebuffer, which apparently needs to be constant. Once these issue resovled, do you mind fixing GStreamer code to use the right one, it reads like this: GstMemory *mem = gst_fd_allocator_alloc(allocator, plane.fd.get(), plane.offset + plane.length, GST_FD_MEMORY_FLAG_DONT_CLOSE); gst_memory_resize(mem, plane.offset, plane.length); Plane.offset is never set (always zero atm), but if its constant, its the wrong one, this code should use the metadata once (the variable one). You will also notice the semantic difference, we assume length does not include the offset, this might need adjustment. Nicolas > } > > return buffer;
diff --git a/src/libcamera/v4l2_videodevice.cpp b/src/libcamera/v4l2_videodevice.cpp index f5b3fa09d9a0..5d163f7fd853 100644 --- a/src/libcamera/v4l2_videodevice.cpp +++ b/src/libcamera/v4l2_videodevice.cpp @@ -1957,10 +1957,13 @@ FrameBuffer *V4L2VideoDevice::dequeueBuffer() * The number of planes in the frame buffer and in the * V4L2 buffer is guaranteed to be equal at this point. */ - for (unsigned int i = 0; i < numV4l2Planes; ++i) + for (unsigned int i = 0; i < numV4l2Planes; ++i) { metadata.planes()[i].bytesused = planes[i].bytesused; + metadata.planes()[i].offset = planes[i].data_offset; + } } else { metadata.planes()[0].bytesused = buf.bytesused; + metadata.planes()[0].offset = buf.m.offset; } return buffer;
Fill in the offset field of the frame metadata so that it the application can read it. Signed-off-by: Paul Elder <paul.elder@ideasonboard.com> --- src/libcamera/v4l2_videodevice.cpp | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)