Status of multiboot for amd64

cesura

Hi all, long time lurker but first time poster.

I'm interested in running a stripped-down NetBSD kernel in QEMU. I've been able to get a working setup with an i386 build, but the status of amd64 multiboot is a bit more unclear to me.

Following the same steps for a 64-bit build, QEMU (via the -kernel option) returns:

Error loading uncompressed kernel without PVH ELF Note

This leads me to believe that something might be off with the multiboot header. Indeed a search of the mailing lists has shown me a related issue: http://mail-index.netbsd.org/port-i386/2020/11/13/msg004008.html

If I look at the source tree, I see the MULTIBOOT option is available for amd64:
https://github.com/NetBSD/src/blob/trunk/sys/arch/amd64/conf/ALL#L51

Additionally, a man page for multiboot(8) on amd64 exists: https://man.netbsd.org/NetBSD-9.3/amd64/multiboot.8

This email seems to be the biggest clue, suggesting that the kernel (and therefore multiboot section) is loaded at a memory address much higher than what is suggested by the spec (if it's linked at all?): https://mail-index.netbsd.org/tech-toolchain/2019/10/24/msg003655.html

If someone more familiar with NetBSD internals could confirm/deny if multiboot on amd64 is even supported, I would greatly appreciate it!

iMil

I brought this subject to the NetBSD crowd a couple of times and it seems there's a status-quo about it.
Like you, I want to build small virtual environments as I explained here https://imil.net/blog/posts/2020/fakecracker-netbsd-as-a-function-based-microvm/ but can only do it with i386 kernels / userland.

iMil

If this topic is still of interest, it is now possible to boot NetBSD current using qemu's -kernel parameter using PVH mode, the work is available here https://github.com/NetBSDfr/NetBSD-src/tree/mmio_cmdline and should be reviewed "soon" for further inclusion in NetBSD's official source tree.

Void

iMil
Hi, first time poster too. I am currently working on a xen+netbsd setup virtualized by qemu inside a debian:bookworm container (that's also inside wsl2). As I am fairly new to the BSD landscape (and virtualization in general) I decided before doing anything with Xen to try booting from the kernel.

I tried with a kernel built from the newly released NetBSD 10.0 tarballs, hoping that changes were already brought into the official source tree but got the same "ELF NOTE" error.

I tried cloning the repo you linked and building the kernel from it but i receive an error while compiling the toolchain regarding a missing definition for gzip from the linker ( if I understood correctly the error). Am i missing something ?

Void

Void
Update ragarding this. The problem while building the kernel from the mmio_cmdline cloned repository is probably linked to wsl2. I built it inside ubuntu and I managed to boot into it with option -kernel. Same thing doesn't work for XEN3_DOM0 configuration, that seems to retain the elf note problem