Ciao Mauro,

Thanks for the resource! That clarifies quite a bit. - u-boot.bin is renamed as the *kernel7l.img*. In this way *start.elf* loads u-boot instead of the kernel - The actual kernel "*kernel7l.img*" is used to create *uImage *using the tool: *mkimage*.

Now I have only one doubt. Why does u-boot need to convert the kernel image in order to load it (with *bootm*)? Could not *bootz *be used to directly load *kernel7l.img*? as this tutorial describes: https://www.vdoo.com/blog/setting-up-u-boot-to-harden-the-boot-process

BTW I followed the steps from that tutorial, but I didn't manage to boot the kernel in that way. So I guess the extra-step *mkimg -> bootm* is necessary. I am just trying to figure out why.

Best,

Anthony Arrascue

On Wed, Jul 7, 2021 at 2:15 PM Mauro Salvini m.salvini@koansoftware.com wrote:

On 07/07/21 10:05, Anthony Arrascue wrote:

...

Hello everyone,

Hello

...

I am learning about u-boot and how this works in the Raspberry Pi system. This is what I understood so far:

• The first stage from the RPI cannot or should be modified. In this

stage

...

the BootROM simply loads the bootloader from the flash EEPROM.

• In the second stage the EEPROM boot loader finds and loads start.elf,

whose task is to load the kernel. It first reads* config.txt* which contains a kernel parameter. This is where u-boot is "injected". *kernel=u-boot.bin*

Then u-boot can then in turn load the actual kernel. For a CM4 this would be "*kernel7l.img*".

I would be super satisfied with this knowledge, but in practice I have (possibly) seen other ways of integrating u-boot. I am here referring,

for

...

example, to Yocto recipes for the CM4 (meta-raspberrypi / u-boot). The

boot

...

directory of such an image contains: *boot.scr*, *uboot.env*, and

*uImage*.

...

The readable part of boot.scr specifies that uImage will be loaded:

*value bootargs /chosen bootargsfatload mmc 0:1 ${kernel_addr_r} uImageif test ! -e mmc 0:1 uboot.env; then saveenv; fi;bootm ${kernel_addr_r} - ${fdt_addr}*

I'm not a RPI expert, but this is what know:

...

config.txt does not contain a kernel parameter.

It must not contain kernel params: it is used by start.elf, and configurations are relative to the things that start.elf does. Kernel parameters are contained in loaded devicetree, and read from it by u-boot in u-boot.scr with

fdt addr ${fdt_addr} && fdt get value bootargs /chosen bootargs

command.

...

*So here my questions:*

1. start.elf cannot be modified since it is proprietary. How can it load

u-boot if there is no kernel parameter in config.txt pointing to a binary u-boot? boot.scr is supposed to run before u-boot.bin. Who reads and executes boot.scr?

start.elf loads the binary named kernel7l.img. You need to rename u-boot.bin to kernel7.img, so start.elf loads u-boot instead of the kernel. You can read this post [1] from Andrei Gherzan, one of the meta-raspberrypi authors.

u.boot scr is executed by u-boot, so the execution order is

start.elf -> u-boot.bin -> u.boot.scr

...

2. is uImage just another name for u-boot.bin?

No, it is the kernel image that u-boot loads throug u-boot.scr:

fatload mmc 0:1 ${kernel_addr_r} uImage

...

3. In this scheme how does u-boot know it should load "*kernel7l.img*"?

See above

...

4. Are there other ways of integrating u-boot in a RaspberryPi? Is there

any documentation which describes these different integration schemes?

AFAIK only the one described in [1] and done in meta-raspberrypi

...

Thank you very much for your help!

[1] https://andrei.gherzan.ro/linux/uboot-on-rpi/

-- Mauro Salvini | KOAN sas | Bergamo - Italia embedded software engineering Phone: +39 035-255235 http://KoanSoftware.com