Hi Ahmad, On Mon, 4 May 2026 at 02:32, Ahmad Fatoum <[email protected]> wrote: > > Hello, > > On 4/29/26 4:23 PM, Simon Glass wrote: > > On Tue, 28 Apr 2026 at 08:06, Tom Rini <[email protected]> wrote: > >> On Tue, Apr 28, 2026 at 04:26:31PM +0530, Beleswar Prasad Padhi wrote: > >>>> I guess the real question here is, does the spec allow this to be > >>>> omitted like that? > >>> > >>> > >>> Could you please point me to the spec? I could not find it. > >> > >> Sure: https://github.com/open-source-firmware/flat-image-tree/ > > > > It is mandatory for kernel and firmware images. Instead of allowing it > > to be optional for 'kernel', we created 'kernel_noload' some years > > back. > > > > Which image type are you using in your case? > > FWIW, barebox has always interpreted lack of the load property to mean > that it should just compute a suitable address itself (i.e. free memory > aligned as Linux expects on that platform - same as kernel_noload). > > Writing that into the spec also allows for Beleswar's use case here, > although it doesn't give the guarantee that the address will always be > the image address. > > I think that giving a guarantee about the image address is wrong: > > - If the code is not relocatable, then its load address can equally well > be written into the FIT image node > > - If the code is relocatable, then forcing the load address to be the > image data addresses will make non-external FIT unfeasible for most > users, because of code having the wrong alignment and for external FIT, > it will mean careful placement of the FIT itself is required, so we > don't end up with misbehavior at runtime. > > I thus suggest, the spec leaves it to bootloaders to select appropriate > load addresses if its missing and if the bootloader can't, it should > refuse boot (current U-Boot behavior). Future U-Boot behavior will > likely need to be adapted, so this loading at image data address is only > done for these specific firmware images and not in general as otherwise > it would preclude any other use in future.
This seems reasonable to me. So load becomes optional for all image types, not just kernel_noload/fpga. I drafted something here: https://github.com/open-source-firmware/flat-image-tree/pull/50 Regards, Simon
