Manual U-Boot integration

tutorial

In order to support rootfs rollback, Mender requires integration with U-Boot. Normally, the Mender Yocto layer meta-mender integrates U-Boot automatically.

If you need to integrate with U-Boot manually, this page explains how to do so using your existing Yocto Project build environment.

How to integrate with U-Boot

Mender provides special integration points for your existing boot code to hook into so that Mender can be a part of the normal boot process. These integration points, along with certain required U-Boot features and configuration options, allow Mender to do safe, automatic updates of the device, with rollback support.

This normally requires patching of stock U-Boot versions, which is what the automatic process does. But this section explains the necessary steps to do it manually.

If you are not sure whether your board is using automatic patching, you can check it by executing this command:

bitbake -e u-boot | grep '^MENDER_UBOOT_AUTO_CONFIGURE='

The variable should be either 0 or 1, depending on whether automatic patching is enabled or not. Note that the string u-boot may be different if you are using a U-Boot fork with a different name.

A good starting point for manual patching, is to take the patch produced by the automatic patcher and use that as a basis, since it will often be close to complete, if not fully.

To extract the patch, execute the following bitbake command:

bitbake -c save_mender_auto_configured_patch u-boot

The command will tell you where the resulting patch is found. As in the previous command, the string u-boot may be different if you are using a U-Boot fork with a different name.

Disabling automatic patching

After acquiring the basis patch above, first thing you will need to do is to disable the automatic patching. This is done by adding a u-boot_%.bbappend file to your layer (or a different name if you board is using a fork of U-Boot), and inside it add:

MENDER_UBOOT_AUTO_CONFIGURE = "0"

U-Boot features

Mender require enabling certain U-Boot features to function correctly. The features are enabled in the board support headers in U-Boot, under include/configs.

  1. CONFIG_BOOTCOUNT_LIMIT: This is required for rollback support to work. For example:

    #define CONFIG_BOOTCOUNT_LIMIT
  2. CONFIG_BOOTCOUNT_ENV: This will store the boot counter in the U-Boot environment. This means that other CONFIG_BOOTCOUNT_ features should be turned off.

  3. CONFIG_ENV_IS_IN_MMC: This will store the U-Boot environment file on the memory card, before the first partition start. See MENDER_UBOOT_ENV_STORAGE_DEVICE_OFFSET for more information. Other CONFIG_ENV_IS_IN_ features should be turned off.

  4. FAT_ENV_INTERFACE, FAT_ENV_DEVICE, FAT_ENV_PART and FAT_ENV_DEVICE_AND_PART should be removed from the configuration if they are present. They will be automatically defined by Mender.

Integration points

These are the current integration points:

  1. mender_setup: This is an environment script that should be run at the very beginning of the boot process. It will not perform any boot steps, however it may modify, and potentially save, the U-Boot environment.

    For example, if your current bootcmd looks like this:

    bootcmd=run mmcboot

    you need to change it to:

    bootcmd=run mender_setup; run mmcboot
  2. mender_uboot_root: This is an environment variable that contains the description of the device currently set to boot. Whenever a U-Boot command is issued that needs to access the current boot partition, reference this variable.

    For example, if you have a script, loadimage, that loads the kernel from the filesystem, using mmc as the device and a ${bootpart} variable reference as the partition to load from:

    loadimage=load mmc ${bootpart} ${loadaddr} ${bootdir}/${bootfile}

    To target the correct active partition, change parameters to:

    loadimage=load ${mender_uboot_root} ${loadaddr} ${bootdir}/${bootfile}

    Note that ${mender_uboot_root} includes mmc in the string value; it is a complete description of the device and partition to load from.

  3. mender_kernel_root: This is essentially the same as the previous variable, but is a string tailored to the Linux kernel instead of U-Boot. This should be used as the root argument to the kernel. For example, if your current bootargs looks like:

    bootargs=console=${console},${baudrate} root=${mmcroot}

    change the root parameter to ${mender_kernel_root}:

    bootargs=console=${console},${baudrate} root=${mender_kernel_root}

Optional integration points

This section describes integration steps that are not strictly necessary for basic Mender functionality, but will improve functionality under certain conditions.

  1. mender_altbootcmd: This integration point is only needed if your setup is already making use of U-Boot's altbootcmd functionality, otherwise you can skip this step.

    If using altbootcmd, you first need to disable Mender's built-in altbootcmd. To this end, add MENDER_NO_DEFAULT_ALTBOOTCMD to the board configuration header in U-Boot (inside include/configs).

    Then, add a call to run mender_altbootcmd at the beginning of altbootcmd. Like mender_setup, this will not perform any boot steps, but it may modify and potentially save the environment. Afterwards, the mender_uboot_root and mender_kernel_root variables will refer to the correct partitions, taking into account the potential rollback that may happen after calling altbootcmd. After executing the desired alternate boot steps, you can either call bootcmd to perform a normal boot using the new partitions, or you can perform a different type of boot sequence and refer to the Mender variables directly.

  2. mender_try_to_recover: We recommend adding a call to this boot script right after the normal disk based boot command for the board. Note that it should execute before other boot methods that are not considered a "normal" boot sequence for the board, such as network boots. The call will facilitate rollback in the event that a boot fails after an update, without reverting to alternative boot methods such as a network boot. For example, change:

    bootcmd=run mmcboot; run networkboot

    into:

    bootcmd=run mmcboot; run mender_try_to_recover; run networkboot

    If there is no update in progress, the script will do nothing and hence alternative boot methods will continue working.

    Note that if this integration point is not used, rollback will still work, but it may not activate until after attempting a network boot or rebooting the device by other means.

  3. mender_uboot_boot, mender_uboot_if, mender_uboot_dev: These variables are not required by Mender, but are available in the board boot code if you need access to:

    • the boot partition string (in U-Boot format, for example mmc 0:1): mender_uboot_boot

    • the storage device interface (in U-Boot format, for example mmc): mender_uboot_if

    • the storage device index (in U-Boot format, for example 0): mender_uboot_dev

Boot configuration

There are also a few other details that need to be in place for Mender to work.

Location of kernel

In a Mender based configuration, the rootfs partition stores the kernel - not the boot partition. This enables Mender to update the kernel on full rootfs updates. Usually, in a boot partition, the kernel is stored in the root, but on a rootfs partition it is usually stored in /boot. Therefore, paths that refer to the location of the kernel need to be updated to point to this location. This is usually the case for the device tree and initrd files as well, if the kernel has those. For instance:

uimage=uImage
fdt_file=uImage.dtb

should instead be:

uimage=boot/uImage
fdt_file=boot/uImage.dtb

Kernel loading method

Because the kernel and associated files loads from a rootfs partition, in the majority of cases it will be an ext4 or ext3 partition. If the existing boot code for the board uses the fatload command to load the kernel and/or any associated files, you need to change the command if the rootfs is not a FAT partition. We recommend replacing it with load, since it will work in both cases. You can also use ext2load or ext4load if desired.

Size of boot environment file

In the bitbake recipe for u-boot, BOOTENV_SIZE should match the value configured for CONFIG_ENV_SIZE in the board specific C header for U-Boot (inside u-boot/include/configs). The exact value is board specific, but it is necessary that the two values are the same.

For example, in u-boot/include/configs/myboard.h:

#define CONFIG_ENV_SIZE 0x20000

and in recipes-bsp/u-boot/u-boot_%.bbappend:

BOOTENV_SIZE = "0x20000"

Practical example

For a real life example of a patch used to integrate with Mender, check out the patch for BeagleBone on github.com. This patch displays all of the steps required to patch the U-Boot boot code for BeagleBone, including most of the steps described on this page. Note that the patch is not kept up to date anymore after the automatic patching capability was added, so it serves only as an example.

We welcome contributions to improve this documentation. To submit a change, use the Edit link at the top of the page or email us at .