u-boot fails to load FDT from linux-firmware on qemu builders

[Colin Watson]

I tried to do a routine upgrade and reflash of the Launchpad riscv64 builders, which pulled in linux-image-5.11.0-1017-generic 5.11.0-1017.18~20.04.1. The resulting image failed to boot with the following (complete) console output:

    Device 0: QEMU VirtIO Block Device
                Type: Hard Disk
                Capacity: 71680.0 MB = 70.0 GB (146800640 x 512)
    ... is now current device
    Scanning virtio 0:1...
    Found /boot/extlinux/extlinux.conf
    Retrieving file: /boot/extlinux/extlinux.conf
    1391 bytes read in 1 ms (1.3 MiB/s)
    U-Boot menu
    1: Ubuntu 20.04.3 LTS 5.11.0-1017-generic
    2: Ubuntu 20.04.3 LTS 5.11.0-1017-generic (rescue target)
    3: Ubuntu 20.04.3 LTS 5.8.0-29-generic
    4: Ubuntu 20.04.3 LTS 5.8.0-29-generic (rescue target)
    Enter choice: 1: Ubuntu 20.04.3 LTS 5.11.0-1017-generic
    Retrieving file: /boot/initrd.img-5.11.0-1017-generic
    189611112 bytes read in 509 ms (355.3 MiB/s)
    Retrieving file: /boot/vmlinuz-5.11.0-1017-generic
    25258496 bytes read in 70 ms (344.1 MiB/s)
    append: root=UUID=b797e480-278b-4efc-ae1a-bb1634c8cbcf rw noquiet root=/dev/vda1
    Unhandled exception: Load access fault
    EPC: 00000000fffb642c RA: 00000000fff73bc4 TVAL: 0000000000000000
    EPC: 000000008024a42c RA: 0000000080207bc4 reloc adjusted

    ### ERROR ### Please RESET the board ###

For the time being I've set `U_BOOT_DEFAULT="l1"` in `/etc/default/u-boot` to force it back to the 5.8 kernel, although that's obviously not very robust. Let me know if there's anything else useful I can provide. The full image is kind of massive (70G, though sparse so with only 13G of actual data), so I hope you can reproduce it with something more stripped-down.

I've attached the libvirt domain definition.

[Colin Watson (cjwatson)]

Hm, I actually get similar results when I revert to 5.8.0-29:

    Device 0: QEMU VirtIO Block Device
                Type: Hard Disk
                Capacity: 71680.0 MB = 70.0 GB (146800640 x 512)
    ... is now current device
    Scanning virtio 0:1...
    Found /boot/extlinux/extlinux.conf
    Retrieving file: /boot/extlinux/extlinux.conf
    1391 bytes read in 2 ms (678.7 KiB/s)
    U-Boot menu
    1: Ubuntu 20.04.3 LTS 5.11.0-1017-generic
    2: Ubuntu 20.04.3 LTS 5.11.0-1017-generic (rescue target)
    3: Ubuntu 20.04.3 LTS 5.8.0-29-generic
    4: Ubuntu 20.04.3 LTS 5.8.0-29-generic (rescue target)
    Enter choice: 3: Ubuntu 20.04.3 LTS 5.8.0-29-generic
    Retrieving file: /boot/initrd.img-5.8.0-29-generic
    181049573 bytes read in 461 ms (374.5 MiB/s)
    Retrieving file: /boot/vmlinuz-5.8.0-29-generic
    20059580 bytes read in 80 ms (239.1 MiB/s)
    append: root=UUID=b797e480-278b-4efc-ae1a-bb1634c8cbcf rw noquiet root=/dev/vda1
    Unhandled exception: Load access fault
    EPC: 00000000fffb642c RA: 00000000fff73bc4 TVAL: 0000000000000000
    EPC: 000000008024a42c RA: 0000000080207bc4 reloc adjusted

    ### ERROR ### Please RESET the board ###

I'll investigate further.

[Colin Watson (cjwatson)]

OK, this is actually due to the changes between u-boot-menu 4.0.1ubuntu0.1 and 4.0.2ubuntu5~20.04.1. The relevant difference is that the menu entry now has `fdtdir /lib/firmware/5.8.0-29-generic/device-tree/` (or similar) when it previously didn't load a device tree at all.

I successfully worked around this by adding the following to `/etc/default/u-boot`:

    # XXX cjwatson 2021-08-24: Revert to old (nonexistent) default, since using
    # the device tree in /lib/firmware/ results in an image that fails to boot.
    # See https://bugs.launchpad.net/bugs/1940984.
    U_BOOT_FDT_DIR="/usr/lib/linux-image-"

[Heinrich Schuchardt (xypron)]

The bug report indicates a "Load access fault" in U-Boot. This should be fixed.

The bug report does not contain the necessary information for analysis.

Please, provide the U-Boot version that you have been using.
Is the problem reproducible with the current U-Boot package?
Please, provide the output of U-Boot's command printenv.

Marking the bug report as incomplete.

[Launchpad Janitor (janitor)]

[Expired for u-boot-menu (Ubuntu) because there has been no activity for 60 days.]