On Fri, Oct 02, 2026 at 08:31:37PM -0300, Jason Gunthorpe wrote: > - Non-EFI boot, particularly kexec
At least the kexec issue seems like it has a good solution, Ard created a way for kexec to use the EFI entry point, and we've shown now that this series plus Ard's work can kexec directly into a UKI and DRTM into the successor kernel. Pretty amazing. There is a thread here: https://lore.kernel.org/r/[email protected] Based on this, I think a clean picture for x86 is coming into focus now. I recommend that the x86 side create a "EFI helper", this would be a dedicated project and include something similar to the EFI package with GUID and the trampolines. As a dedicated github with as its own upstream it doesn't require any help from the grub community. The general boot flow would have this helper provide functions to the Linux stub, like has already been sketched out with the EFI package guid scheme. The helper would implement something very similar to ARM's DRTM spec, with the same kind of DLME layout and so on. This will give us a very strong cross-arch alignment and help create a basis for any furture industry work on DRTM. I feel like this is not a hard job with some AI assistance. If this is still unlcear I may be able to help show it with more details. That doesn't change anything about this series. I think anyone who can implement a sane protocol natively should just do so in the built in EFI stub. The external helper is more because x86 has so many tricky platform details, needs more ACPI parsing, external binary blob firmware and so on. Regards, Jason
