Currently, VEA (Video Encode Accelerator) gives Client::RequireBitstreamBuffers input video frame's coded_size to allocate input buffer [1]. However, after the CL: "media/gpu: Use VideoFrameLayout for format on input buffer" [2], input VideoFrameLayout's coded_size is assigned by negotiated v4l2_format's (pix_mp.width, pix_mp.height) (see [3] V4L2Device::V4L2FormatToVideoFrameLayout). The device driver's response may not inlcude per v4l2_plane's padding, which is given by pix_mp.plane_fmt[i].sizeimage. Because of this, using input's VideoFrameLayout's coded_size to RequireBitstreamBuffers() causes insufficient input buffer allocation bug [4]. 

Tentative solution of  crbug.com/918088  is to give RequireBitstreamBuffers the input size calculated by V4L2Device::CodedSizeFromV4L2Format(format). But the right solution is to make Client::RequireBitstreamBuffers [1] using input's VideoFrameLayout to allocate input buffer instead of just using gfx::Size.
 
[1] https://cs.chromium.org/chromium/src/media/video/video_encode_accelerator.h?sq=package:chromium&g=0&l=172
[2] https://crrev.com/c/1361743
[3] https://cs.chromium.org/chromium/src/media/gpu/v4l2/v4l2_device.cc?type=cs&q=V4L2FormatToVideoFrameLayout&sq=package:chromium&g=0&l=1249
[4]  https://crbug.com/918088