On 06.01.26 16:34, Mark Kettenis wrote: >> Date: Tue, 6 Jan 2026 14:47:30 +0100 >> From: Jan Kiszka <[email protected]> > > Hi Jan, > >> Hi, >> >> it seems that some kernel configs can cause problems with the >> CONFIG_EFI_GRUB_ARM32_WORKAROUND. Namely, 6.1-cip and 5.10-cip with its >> current defconfigs [1] no longer boot on the BeagleBone Black (other >> boards still need validation), both with the more recommendable >> systemd-boot and grub. I do not have the motivation to debug the reason >> in the kernel / its config, given that everything works fine once the >> firmware behaves according to the UEFI spec. >> >> Meanwhile, grub reached 2.12 now which is far away from the old, broken >> version of 2019. So, should we rather make this workaround opt-in in 2026? > > This isn't just about GRUB though. There is also the issue with the > the non-architected caches that are typically found on systems that > use Cortex-A9 cores such as i.MX6.
i.MX6 is a good hint to cross-check with some related board. But what does that mean exactly? If we turn off the workaround, not calling cleanup_before_linux() anymore, we will then leave caches for i.MX6, e.g., enabled, and that would also be against the spec? * According to the UEFI spec caches that can be managed via CP15 * operations should be enabled. Caches requiring platform information * to manage should be disabled. This should not happen in * ExitBootServices() but before invoking any UEFI binary is invoked. > > The OpenBSD bootloader used to have the same problem as those older > versions of GRUB. I fixed that back in 2019, and there shouldn't be > any users left with a broken version. > That's good. Thanks, Jan -- Siemens AG, Foundational Technologies Linux Expert Center
