On Mon, Aug 17, 2026 at 10:54:51PM +0200, Linus Walleij wrote: > On Mon, Aug 17, 2026 at 6:45 PM Stephan Gerhold > <[email protected]> wrote: > > > Hm okay so you're basically dropping support for the internal eMMC in > > this patch? Can we support both? BOOTSTD should be able to scan multiple > > storages for boot targets. > > Yeah just laziness on my part, I'll fix it up. > > > I'm also not really sure about selecting Janice as "default" device tree > > here. Do you expect users to change this setting when they want to build > > stemmy for another board? > > I have a script that loops over the Samsung device trees, > and recompiles U-Boot for each of them. > https://dflund.se/~triad/krad/makefiles/ux500-uboot.mak > > > I wonder if we could continue using a single generic binary for all > > boards using CONFIG_MULTI_DTB_FIT. We should have "fw_mach" in > > board/ste/stemmy/stemmy.c that should tell us which device we are > > running on, which seems to be mostly(?) unique: > > > > https://github.com/stericsson-mainline/android_kernel/blob/9508895090960012f916561b2ede5eaf8124fabd/arch/arm/tools/mach-types#L3956-L3965 > > > > I'm not sure about codina-tmo, you would need to check which number the > > bootloader passes. > > > > If these numbers are unique, it should be easy to implement > > board_fit_config_name_match(). > > I see the appeal of the ambition to have a "one size fits (pun!) all" > U-Boot. > > But as it is, some of the DTS attributes on the external SD card are > different between the platforms, so we would need to modify the > U-Boot internal Stemmy device tree for those, copying over yet > properties from the device trees and into the boardfile. > > For example Janice and Codina uses a regulator for > vqmmc-supply, and those using the "sd_level_translator" > for vqmmc-supply actually use different GPIO pins on say > Gavini compared to Codina or Golden. > > The pin direction settings on the MMC host vary. (Some of > these were really tricky to get right.) > > So I think no, if we want external SD card boot we need > per-device compiled U-Boot binaries with the unique device > trees. >
I wasn't suggesting to use the same DT for all devices. What I suggested is to bundle all the upstream DTBs for the individual devices into a single U-Boot binary (CONFIG_MULTI_DTB_FIT) and pick the correct one at runtime. :) That should be equivalent to your setup, just more convenient for the user. The only open question for me is if we can reliably detect codina-tmo based on the mach-type ID, I can't find that in the kernel sources I have. Thanks, Stephan
