To use proper uboot wrapper initramfs

[adam900710]

Hi,

I found alarm kernel pretty big (20+M for aarch64) as it includes all needed fs/mmc drivers.
I guess the main reason is, uboot won't really load plain initramfs.

Uboot needs uboot wrapped initramfs (mkimage -T ramdisk) to proper pass it to kernel.
So under most case, uboot won't pass initramfs proper to kernel, thus we rely kernel sololy to boot whatever the board.

This is completely against the normal idea of initramfs. With initramfs, we should only compile what we need into initramfs for mounting the rootfs.

I have crafted uboot-rockpi4 and uboot-rockpro64, which included hooks for pacman to generate uboot wrapperd initramfs (initramfs-linux.img.uboot), and include uboot-tools as dep for mkimage.
https://github.com/adam900710/PKGBUILDs ... 8345684fd7

I guess we should follow the same principle to make all uboot-* packages to properly generate the wrapped initramfs, and then change kernel config to make everything compiled into modules other than into the kernel.

I tried to send out an issue to the PKGBUILDs repo, but the author has disabled issues.
Hopes this is the proper place to ask for such modification.

Thanks.

[summers]

Thought in your other thread you were asking how to make and modify the initramfs under arch, but here you see to have a different problem.

First off, the initramfs is loaded by the kernel, specifically one of the parameters that uboot passes to the kernel is the address of the initramfs in memory. This has been in uboot since the year dot, way before I started using it anyway, maybe 10 years ago.

So part of what is confusing you is how the initramfs gets into memory. This has changed in various versions of uboot. A long time ago, yes it had to be wrapped up in an image, and the header of that image told uboot how to load it into memory. All recent uboot though have been able to read the file systems (with common formats, e.g. ext4), take a file and load it to an area of memory. This is whats done these days with initramfs, the file is loaded into memory and that address passed to the kernel.

You also seem to question why the default arm arch kernel has so much enabled by default, and isn't loaded as a module. Thats a question for the arch arm maintainers, its set in the PKGBUILD in particular in the .config file. Can't say I've ever had a problem with it.

Just check your git set up - you havn't passed the addresses to the kernel in the modern way. Try looking at a modern boot.txt and understand how it works:

https://archlinuxarm.org/packages/armv7h/uboot-beaglebone/files/boot.txt

$this->bbcode_second_pass_quote('', 'T')hought in your other thread you were asking how to make and modify the initramfs under arch, but here you see to have a different problem.

First off, the initramfs is loaded by the kernel, specifically one of the parameters that uboot passes to the kernel is the address of the initramfs in memory. This has been in uboot since the year dot, way before I started using it anyway, maybe 10 years ago.

[adam900710]

BTW, you can check if your board really loaded the initramfs properly.

For my setup, using the uboot wrapped initramfs, the dmesg looks like:
$this->bbcode_second_pass_code('', '
Kernel command line: console=ttyS2,1500000 root=PARTUUID=4d82c01e-01 rw rootwait
...
Trying to unpack rootfs image as initramfs...
Freeing initrd memory: 7204K
...
')

There is no initrd passed from cmdline, but kernel still get the proper initramfs, and freed its space.
I guess there is something like the X86 kernel boot protocol header, to pass the real initramfs address.
From my previous search, some device-tree like structure is used to pass the initramfs location and size, thus the initramfs must be handled by U-boot.

If not properly extracted, after that "Trying to unpack rootfs" line, it will output a decompression failure line.
Exactly what I got using the so-called modern u-boot wrapped script (firstly copied from rock64 boot.txt)

The current boot sequence works completely because XFS/EXT4 is compiled into the kernel.
If it's not, any initramfs problem will just prevent anyone from proper booting.

So the two problems are in fact related.

[summers]

So what you are saying is on the latest uboot that the commands:
$this->bbcode_second_pass_code('', 'if load ${devtype} ${devnum}:${bootpart} ${kernel_addr_r} /boot/Image; then
echo fdt: ${fdtfile}
if load ${devtype} ${devnum}:${bootpart} ${fdt_addr_r} /boot/dtbs/${fdtfile}; then
if load ${devtype} ${devnum}:${bootpart} ${ramdisk_addr_r} /boot/initramfs-linux.img; then
booti ${kernel_addr_r} ${ramdisk_addr_r}:${filesize} ${fdt_addr_r};
else
booti ${kernel_addr_r} - ${fdt_addr_r};
fi;
fi;
fi')
Do not load the initramfs? Exactly what version of uboot are you using, and what commands did you use to compile it?

Passing the initramfs as the second address of the kernel line is the standard, I can't see uboot ever breaking this its so standard; so you must have a broken uboot, and very badly broken at that.

When I get home, I'll give what log messages are on my latest uboot, which is IIRC a 2018 version I needed for my PocketBeagle.

[adam900710]

$this->bbcode_second_pass_quote('', '
')Do not load the initramfs? Exactly what version of uboot are you using, and what commands did you use to compile it?

[moonman]

CONFIG_SUPPORT_RAW_INITRD needs to be enabled.
Also the syntax is a little different as you need to pass the size as well <initrd addres>:<initrd size> (which you did not in your example above).

[summers]

As moonman said. Also note that the load command sets ${filesize} - this isn't well documented, you need to read the source code to see this. So as initramfs needs a size, make sure its the last thing loaded, and so ${filesize} isn't overwritten by subsequent load commands.

The latest defconfig can be seen here https://gitlab.denx.de/u-boot/u-boot/blob/master/configs%2Frock-pi-4-rk3399_defconfig and indeed doesn't set raw intramfs.

And this README, the main uboot one, gives the config option described by Kevin. https://gitlab.denx.de/u-boot/u-boot/blob/master/README

[adam900710]

$this->bbcode_second_pass_quote('', 'C')ONFIG_SUPPORT_RAW_INITRD needs to be enabled.