Hi Petr, On 2026-10-10T14:24:46, Petr Hodina via B4 Relay <[email protected]> wrote:
> Entry_fit's @fdt-SEQ generator copies the node's "compression" property > into the FIT. However, it embeds the raw .dtb. A FIT can therefore claim > "compression = lzma" while holding uncompressed data and SPL will fail > to decompress it during boot. Thanks for looking at this. It's a real gap, but I'm not keen on the approach. Binman normally keeps two separate properties: 'compression' is FIT metadata copied into the image node, while 'compress' tells binman to compress the data. The Rockchip U-Boot image already works this way: the image node has 'compression' and the u-boot-nodtb subnode has 'compress'. This series makes the metadata property also act as a binman instruction, but only for generated fdt nodes, which is inconsistent with the rest of the fit etype. It also turns any value other than lzma into a hard error for existing users of @fdt-SEQ. Please can you add a binman-level property instead, either 'compress' or 'fit,compress' to match the other fit,* properties, and set 'compression' in the output from it? What do you think? BTW there is a typo in the cover letter: 'onl' should be 'only'. Regards, Simon
