kHi Xavier,
On Tue, 26 Jul 2022 at 13:08, Xavier Drudis Ferran xdrudis@tinet.cat wrote:
El Tue, Jul 26, 2022 at 11:08:21AM +0200, Quentin Schulz deia:
You could also have a fit,align = <8>; property instead of hardcoding it.
I'm not sure the fit entry in binman heeds this property, may have overlooked something. I'd have to try it.
[..]
This is unfortunately not possible since binman parallelizes the build of images in the binman DT node. This means there is no guarantee the u-boot.itb file is generated before this section is worked on by binman. Or maybe I misunderstood the docs.
But good progress, I guess this phandle thing "just" needs to be fixed to have something nice (both for this patch series and mine).
I'm sending another patch below "fixing" the phandle issue, but it's a dirty hack without too much thought given. It could still fail if threads try to read data from the image of another thread before it's ready or something. Only lightly tested with binman -T 0, not parallel. It seems to run mkimage -B 8 -E -t -F ./itb.fit.fit SIX times each time I run binman. Not sure why.
But it boots.
I wonder if we shouldn't just run binman several times from make instead of once at the end, have make run it once for each file we want to create, so that we can reuse u-boot.itb for both the MMC and the SPI image. We could have one .dts for each image, so when make want u-boot.itb it runs binman -a of-list="rk3399-rock-pi-4b.dts rockchip-itb-u-boot.dts"
We must not move to multiple invocations of binman. It is supposed to handle everything at once.
But for the case you have here, we could perhaps add a feature to specify an ordering for images and to explicitly allow one image to include another?
[..]
Regards, Simon