Hi Jacopo, On Fri, Aug 11, 2023 at 11:43:26AM +0200, Jacopo Mondi wrote: > Hi Sakari > > On Fri, Aug 11, 2023 at 09:11:39AM +0000, Sakari Ailus wrote: > > Hi Jacopo, > > > > On Fri, Aug 11, 2023 at 08:31:16AM +0200, Jacopo Mondi wrote: > > > > > +V4L2_META_FMT_GENERIC_CSI2_10 > > > > > +----------------------------- > > > > > + > > > > > +V4L2_META_FMT_GENERIC_CSI2_10 contains packed 8-bit generic metadata, 10 bits > > > > > +for each 8 bits of data. Every four bytes of metadata is followed by a single > > > > > +byte of padding. The way the data is stored follows the CSI-2 specification. > > > > > + > > > > > +This format is also used on CSI-2 on 20 bits per sample format that packs two > > > > > +bytes of metadata into one sample. > > > > > + > > > > > +This format is little endian. > > > > > + > > > > > +**Byte Order Of V4L2_META_FMT_GENERIC_CSI2_10.** > > > > > +Each cell is one byte. "M" denotes a byte of metadata and "p" a byte of padding. > > > > > > > > I think you should document whether the padding is always 0 or can be any value. > > > > Perhaps 'X' is a better 'name' for the padding byte in the latter case. > > > > > > > > > > Did I get this right that this format is supposed to work as the RAW10 > > > CSI-2 packed image format, where 4 bytes contain the higher 8 bits of > > > the 10 bits sample and the 5th byte every 4 contains the lower 2 bits of > > > the previous 4 sample ? > > > > > > If that's the case, is 'padding' the correct term here ? > > > > What else would you call it? It'll be zeros that exist just due to the bit > > depth used and as such not interesting at all. > > Ah, they will be 0s! They won't contain the 2 lower bits of the > previous 4 bytes then ? > > I guess this is due to the fact the number of valid bits in each > metadata sample is always 8, regardless of the sample size on the > media bus ? Correct. This is also documented for the generic metadata formats. > > The DMA engine will re-pack them to CSI2_10 (or CSI2_12 etc) just to match > the image format sizes then ? Generally yes. The receiver hardware / DMA is of course free to strip out the padding but in practice that doesn't seem to be done. > > In this case, padding is indeed correct, sorry for the noise! No problem. -- Regards, Sakari Ailus
