[yocto] Can't boot to initramfs

Thu Jul 11 03:39:05 PDT 2019

Hello Zoran, Jupiter and list

The configuration you sent seems to be correct.

As I already said initramfs seems overly complicated in yocto. the most
important thing to note is that 2 kernel images are created, one is
called bzImage (in my case) an the other bzImage-initramfs. However only
the bzImage is written into the rootfs so you have to exchange them
manually. (in /boot/bzImage). I did not find a way of including the
bundled kernel right away.

What you can do is to build core-image-minimal-initramfs and delete the
symbolic link "bzImage", then recreate it and let it point to
bzImage-initramfs. However this is rather a hack than a solution.

An other mistake I made was to use IMAGE_INSTALL_append which is
ignored. Use PACKAGE_INSTALL_append.

Also I found that "break" does not work as a kernel parameter. Use
"shell" oder "debug-shell" instead. If you want to try to boot into
initramfs you can remove all parameters to the booting line so you
either end up in a kernel panic (initramfs doesn't work) or in the
rescue shell (initramfs works).

In the end I actually managed to get a working shell, but I often ended
up in initramfs telling me "dropping to shell..." but then freezing.

I don't have access to my files right now but I can tell you more on my
setup tomorrow. In case the solution is not included above.

Best regards
Moritz

On 11/07/2019 09:24, Zoran Stojsavljevic wrote:
> Hello Moritz,
>
> I need here some help from you. I'll try to reconstruct the parts of
> the local.conf you are using, so I (and Jupiter) can understand what
> should we do to also bundle kernel image with initramfs, to end up in
> Dracut/rescue shell.
>
> Here is what I anticipate after reading several YOCTO @ threads:
>
> IMAGE_FSTYPES_append = " cpio.gz"
> INITRAMFS_IMAGE = "core-image-minimal-initramfs"
> INITRAMFS_IMAGE_BUNDLE = "1"
> # debug: adds debug boot parameters like 'shell' and 'debug', see
> # meta/recipes-core/initrdscripts/initramfs-framework/debug for details
> # Could be removed in more minimal product image
> PACKAGE_INSTALL += "initramfs-module-debug"
>
> Could you, please, review these lines and fix, if something is not correct?
>
> I what I understood, this does the magic, but you could not stop in
> initramfs shell? Still, this problem is not solved?
>
>> I was following this site: https://wiki.debian.org/InitramfsDebug
>> Rescue shell (also known as initramfs shell)
>> Read man initramfs-tools to learn about the break=something kernel
>> parameter (where valid arguments for something are: top, modules,
>> premount, mount, mountroot, bottom, init), which starts a debug shell.
>> You can try, for example, break=premount. You can edit /boot/grub/grub.cfg
>> to add this to the end of the kernel line, or you can do it interactively
>> from the grub boot menu: "e" to edit, and "b" to boot once you've edited
>> the kernel line.
> Now, as my understanding is, you solved this problem actually adding
> to grub.cfg in command kernel line break=premount, and was able to
> stop in rescue shell?! Am I correct here?
>
> Thank you,
> Zoran
> _______
>
>
> On Mon, Jul 1, 2019 at 4:33 PM Moritz Porst <moritz.porst at gmx.de> wrote:
>> Hello,
>> I think I found the issue. ( see below )
>>
>> On 01.07.19 15:57, Zoran Stojsavljevic wrote:
>>
>> Hello Moritz,
>>
>> Too hot here, in Belgrade... Where I am resting for the Time being
>> (actually, this message given to my invisible spying security "angels"
>> on this list)...  :-) Projected +38C degrees today. Too hot for this
>> too old Siberian untouchable bobcat!
>>
>> Luckily it's a rather cold day in germany, thanks that you still take the time to answer !
>>
>> I started from the core-image-minimal to have a small image and
>> extended it with the features I need, which is e.g. a graphical system.
>> The console=[...] part in the kernel command line is probably a
>> remnant but my image boots into the GUI. Is this a problem ?
>>
>> Nope, it is not. If you need to do it correctly, you should use
>> bitbake -k core-image-sato build command (my best educated guess).
>> Then, I do not feel comfortable seeing in your kernel command line
>> serial interface, do you agree?
>>
>> Yes that is true.
>>
>> YOCTO maintainers, any additional
>> advices?
>>
>> My bootloader is currently grub, the EFI is AM.
>>
>> Nope. Eeeeeeeeek. Wrong. Currently, your boot-loader is UEFI AMI. Your
>> OS (Linux probably, U name it) boot-loader is GRUB2. Let us keep it
>> contained, sane and sober.
>>
>> Sorry but I can't find this info in the EFI.
>>
>> Could you, please try the following command being root: dmesg | grep
>> MCU. or grep mcu, or grep CPUID or grep cpuid?? Please, post results
>> to this list (to me).
>>
>> No luck unfortunately (used grep -i)
>>
>> Could you, also, send to YOCTO list/me attached file:
>> /boot/microcode.cpio so I can somehow (?) inspect it? ;-)
>>
>> I send it to you directly, so I don't spam the list with a large attachment.
>>
>> 2GB, single channel.
>>
>> All Cool. E3825 by HW/silicon design could/does NOT support multiple
>> memory channels. ONLY single... But even YOCTO primes (INTEL ones from
>> this list) are not gonna tell this to you. Not 'cause they are nasty.
>> They are NOT aware/they are ignorant (with the purpose)! ;-))
>>
>> No, the boot does not stop. Even if I do issue "break=premount" I
>> end up in my graphical interface with the rootfs mounted. This is the
>> last message in the log: EXT4-fs (sdb2): re-mounted. Opts: (null)
>> The rootfs partition is always 2, so either /dev/sda2 or /dev/sdb2
>> (when booting from stick)
>>
>> Yes, from https://pastebin.com/ya7iCtq7, so it is remounted rw, not
>> read only. So, it seems that you have passed dracut phase. and mounted
>> SD or flash rootfs. So, initramfs is NOT your true problem, is it???
>>
>> The thing is that the boot works but I want an initramfs that can be used for updating (in case the rootfs is broken). However I need to be able to intercept the boot process there because otherwise I can't deploy an update mechanism, that's what I was trying.
>>
>> Zoran
>> _______
>>
>>
>> No, the boot does not stop. Even if I do issue "break=premount" I end up
>> in my graphical interface with the rootfs mounted. This is  the last message
>> in the log: EXT4-fs (sdb2): re-mounted. Opts: (null) The rootfs partition is
>> always 2, so either /dev/sda2 or /dev/sdb2 (when booting from stick)
>>
>>
>> So for the issue...
>> I expected yocto to put the bundled bzImage onto my rootfs. This was not the case. My image directory contains 2x bzImage, one bundled and one unbundled. Apparently yocto puts the >un<bundled image onto my /boot partition and uses it for boot. So of course I couldn't access initramfs in this case. Now I get to the initramfs statement "dropping to shell" if I intentionally boot with wrong rootfs.
>> Still I don't get the interactive shell.
>> On the github ostroproject site I found this:
>>
>> # debug: adds debug boot parameters like 'shell' and 'debug', see
>> #        meta/recipes-core/initrdscripts/initramfs-framework/debug for details
>> # Could be removed in more minimal product image.
>> PACKAGE_INSTALL += "initramfs-module-debug"
>>
>> including the module-debug still does not enable me to get an interactive shell.
>> I was following this site: https://wiki.debian.org/InitramfsDebug
>> I am aware that yocto is no debian but I expected that kernel parameters (like 'break') would be independent of the distribution.
>>
>> Lastly I do not really need the interactive shell, it is enough if I can deploy a custom init script in the initramfs. Still I think that getting an initramfs shell should be as simple as stating the name of the initramfs image and setting the "INITRAMFS_DO_BUNDLE" variable.
>>
>> Best regards
>> Moritz
>>
>>
>> On Mon, Jul 1, 2019 at 11:20 AM Moritz Porst <moritz.porst at gmx.de> wrote:
>>
>> Hello Zoran,
>> thanks for your answer
>>
>> On 28.06.19 14:26, Zoran Stojsavljevic wrote:
>>
>> INITRAMFS_IMAGE = "core-image-minimal-initramfs"
>> INITRAMFS_IMAGE_BUNDLE = "1"
>>
>> ...
>>
>> You can find the /var/log/dmesg here: https://pastebin.com/ya7iCtq7I
>>
>> Some hints...
>>
>> [1] Kernel command line: BOOT_IMAGE=/bzImage root=PARTUUID=71d1d94a-83e8-4895-98eb-35309f58119f
>> break=premount quiet rootwait rootwait rootfstype=ext4 console=ttyS0,115200 console=tty0
>>
>> input: Video Bus as /devices/LNXSYSTM:00/LNXSYBUS:00/PNP0A08:00/LNXVIDEO:00/input/input7
>> fbcon: inteldrmfb (fb0) is primary device
>> Console: switching to colour frame buffer device 128x48
>> i915 0000:00:02.0: fb0: inteldrmfb frame buffer device
>> snd_hda_intel 0000:00:1b.0: bound 0000:00:02.0 (ops i915_audio_component_bind_ops [i915])
>>
>> Hmmmmm... You are using console and serial, but full i915 GFX kernel driver is still included in the build???
>>
>> I started from the core-image-minimal to have a small image and extended it with the features I need, which is e.g. a graphical system. The console=[...] part in the kernel command line is probably a remnant but my image boots into the GUI. Is this a problem ?
>>
>>
>> [2] efi: EFI v2.31 by American Megatrends
>>
>> Using AMI BIOS as boot loader FW... OK?! Am I correct?
>>
>> My bootloader is currently grub, the EFI is AM.
>>
>>
>> [3] smpboot: CPU0: Intel(R) Atom(TM) CPU  E3825  @ 1.33GHz (family: 0x6, model: 0x37, stepping: 0x9)
>>
>> This is CPUID ID 0x30679, which uses MCU... Which MicroCodeUnit? M0130679xxx (info from AMI BIOS)?
>>
>> Sorry but I can't find this info in the EFI
>>
>>
>> [4] Using INTEL ATOM BYT E3825 dual core (sans Hyper-threading), implies that you are using
>> 4GB (e820 messages) as single channel (one memory module DDR3 as 4GB)! Am I correct (important)?
>>
>> 2GB, single channel.
>>
>>
>> [5] Dracut phase?!
>>
>> To my understanding the initramfs should be embedded in /boot/bzImage.
>> However since I use an intel platform I also have a /boot/microcode.cpio
>> which gets loaded via "initrd /microcode.cpio". Removing this line in
>> grub does not enable me to get an initramfs prompt either (again, using
>> break as option).
>>
>> You are obviously stopping in boot phase called dracut. Please, try to mount by hand
>> /dev/sda(whatever)... You should use fdisk -l command, or do ls -al /dev | grep sda to
>> dig out which partition you need to mount to /tmp dir to see rootfstype=ext4 (HDD/SSD)
>>
>> No, the boot does not stop. Even if I do issue "break=premount" I end up in my graphical interface with the rootfs mounted. This is the last message in the log:
>> EXT4-fs (sdb2): re-mounted. Opts: (null)
>> The rootfs partition is always 2, so either /dev/sda2 or /dev/sdb2 (when booting from stick)
>>
>> _______
>>
>> Just thinking loud... .. .
>>
>> Hope this helps (has very little to do with YOCTO build system, BTW) . ;-)
>>
>> Zoran
>> _______
>>
>>
>> On Fri, Jun 28, 2019 at 11:22 AM Moritz Porst <moritz.porst at gmx.de> wrote:
>>
>> Hello,
>> I currently try to deploy a single rootfs update mechanism for my
>> embedded device. I can't boot to initramfs using either "break" or
>> "break=premount" (without quotes...). I tried this in systemd-boot and
>> grub-efi (always efi boot) but the boot process just continues normally.
>> If I insert at the same point e.g. "quiet" this argument is recognised.
>> I boot the .wic image with a separate boot partition from a USB stick.
>> in local.conf I have set:
>> INITRAMFS_IMAGE = "core-image-minimal-initramfs"
>> INITRAMFS_IMAGE_BUNDLE = "1"
>>
>> In order to reduce complexity I now use the standard
>> core-image-minimal-initramfs without .bbappend. I can confirm (from
>> seeing the task) that bitbake bundled the kernel with the initramfs.
>>
>> You can find the /var/log/dmesg here: https://pastebin.com/ya7iCtq7
>>
>> To my understanding the initramfs should be embedded in /boot/bzImage.
>> However since I use an intel platform I also have a /boot/microcode.cpio
>> which gets loaded via "initrd /microcode.cpio". Removing this line in
>> grub does not enable me to get an initramfs prompt either (again, using
>> break as option).
>>
>> Did I forget some configuration or do I have to put the break statement
>> at a very specific position within the "linux ..." boot command ? Do you
>> know which bitbake variables to check ? (both set in local.conf do not
>> get overwritten, already checked this). I got the thud branch checked
>> out in all my meta-layers except for meta-qt which is currently on
>> master branch.
>>
>> Help is much appreciated !
>>
>> Best regards
>> Moritz
>>
>> --
>> _______________________________________________
>> yocto mailing list
>> yocto at yoctoproject.org
>> https://lists.yoctoproject.org/listinfo/yocto