Kernel 6.9 fails to boot with "can't boot dump kernel from OF"

[Dave Vasilevsky]

Could you verify that this change actually fixes the problem? This would

save me some work.

Verified with PPC qemu. I fail to boot with "You can’t boot a kdump kernel
from OF" with CONFIG_CRASH_DUMP on, and I boot fine with it disabled.

Details:
* Tested against the Debian 6.10.4-1 kernel as unpacked with `dget
https://deb.debian.org/debian/pool/main/l/linux/linux_6.10.4-1.dsc`
* Built with the GCC 14.2.0 powerpc toolchain from kernel.org
* Using the exact kernel config from a stock sid powerpc system:
https://gist.github.com/vasi/be13936e9de9b5a6db8a71897c731aef. Tested
building both without any change, and with the change to CONFIG_CRASH_DUMP.
* Built with `PATH=... ARCH=powerpc CROSS_COMPILE=powerpc-linux-gnu- make
bindeb-pkg`
* Tested with a Debian sid powerpc VM in qemu 6.2.

Do you need me to try this kernel against real hardware? I can dig an old
iBook out of the closet to see if it works.

In terms of where this is coming from, I did some investigation:
* The value of CONFIG_CRASH_DUMP does change from n to y between 6.1 and
6.10. So it is indeed a problem with the config, not the kernel crashing
under the same config as before.
* The value of CONFIG_PROC_VMCORE also changes between 6.1 and 6.10, and it
depends on CRASH_DUMP. But PROC_VMCORE seems to be enabled in Debian for
years, so it's not immediately obvious why it's not in my config:
https://salsa.debian.org/kernel-team/linux/-/commit/432666d44950ff603d69537f5ea7c5843fac04d8
* Looking at upstream, CRASH_DUMP changed to `default y` in 6.9! There's
our problem:
https://github.com/torvalds/linux/commit/75bc255a7444801d64c7a7bd09e3f452f86b3585

* Easiest would be to simply change this downstream in Debian, since Open
Firmware is by far the most common way of starting Debian powerpc. I don't
know whether upstream should change or not--is there any real constituency
that runs PPC Linux without OF, and is also interested in crash dumps?
Maybe some embedded folks? Up to you whether you'd like us to try to change
this upstream as well, I'm happy to chip in if so.

Hope this helps,
Dave

[John Paul Adrian Glaubitz]

Hi Dave,

Could you verify that this change actually fixes the problem? This would save me some work.

Verified with PPC qemu. I fail to boot with "You can’t boot a kdump kernel from OF" with
CONFIG_CRASH_DUMP on, and I boot fine with it disabled.

Thanks for the confirmation.

* Tested against the Debian 6.10.4-1 kernel as unpacked with `dget https://deb.debian.org/debian/pool/main/l/linux/linux_6.10.4-1.dsc`
* Built with the GCC 14.2.0 powerpc toolchain from kernel.org
* Using the exact kernel config from a stock sid powerpc system: https://gist.github.com/vasi/be13936e9de9b5a6db8a71897c731aef. Tested building both without any change, and with the change to CONFIG_CRASH_DUMP.
* Built with `PATH=... ARCH=powerpc CROSS_COMPILE=powerpc-linux-gnu- make bindeb-pkg`
* Tested with a Debian sid powerpc VM in qemu 6.2.
Do you need me to try this kernel against real hardware? I can dig an old iBook out of the closet to see if it works.

No, that should not be necessary. Thanks for the testing.

* The value of CONFIG_CRASH_DUMP does change from n to y between 6.1 and 6.10. So it is indeed
a problem with the config, not the kernel crashing under the same config as before.
* The value of CONFIG_PROC_VMCORE also changes between 6.1 and 6.10, and it depends on CRASH_DUMP.
But PROC_VMCORE seems to be enabled in Debian for years, so it's not immediately obvious why
it's not in my config: https://salsa.debian.org/kernel-team/linux/-/commit/432666d44950ff603d69537f5ea7c5843fac04d8
https://github.com/torvalds/linux/commit/75bc255a7444801d64c7a7bd09e3f452f86b3585 
* Easiest would be to simply change this downstream in Debian, since Open Firmware is by far
the most common way of starting Debian powerpc. I don't know whether upstream should change
or not--is there any real constituency that runs PPC Linux without OF, and is also interested
in crash dumps? Maybe some embedded folks?

Thanks a lot for investigating the cause of this and the testing, of course. You saved me a lot
of investigating and debugging myself!

Up to you whether you'd like us to try to change this upstream as well, I'm happy to chip in if so.

Feel free to report this on the Linux PowerPC upstream kernel mailing list. It might be a good idea
to add a conflict for CONFIG_CRASH_DUMP upstream when the 32-bit Power Macintosh option is selected.

Adrian