Hi Pali, On Fri, 30 Dec 2022 at 09:44, Pali Rohár <p...@kernel.org> wrote: > > On Friday 30 December 2022 10:41:47 Tom Rini wrote: > > On Fri, Dec 30, 2022 at 04:24:43PM +0100, Pali Rohár wrote: > > > On Friday 30 December 2022 10:21:04 Tom Rini wrote: > > > > > In this case it would be better to build u-boot-dts.bin only by binman > > > > > (for all platforms) instead of cat-ing rules in Makefile. > > > > > > > > This would also be an easier path forward perhaps for making sure that > > > > the dtb is always 8 byte aligned? > > > > > > Well, no. With DTB the problem is that it is not put to the correct > > > offset as can be specified in linker script. So moving this code from > > > Makefile to binman also moves this problem to another location. > > > 8 byte alignment is just subset of the "correct offset" problem. > > > > Right, the high level answer is binman is intended to be the tool to > > assemble binaries, and has to deal with "make sure binary X is at offset > > Y, which also has a linker symbol for run-time references". > > Ok, if this tool has access to ELF/linker symbols (or will have in > future in case it does not have yet) then this problem could be solved > here.
It does have this access and already updates symbols in some cases. See [1]. I am a little nervous about a complete move to binman in this area even for simple things like u-boot.bin, since it would set off yet another migration. But perhaps most boards don't actually use u-boot.bin anyway? Part of me thinks we should solve this in the .lds files, since otherwise we are blurring the line between building and packaging. Regards, SImon [1] https://u-boot.readthedocs.io/en/latest/develop/package/binman.html#access-to-binman-entry-offsets-at-run-time-symbols
